A TCR-like CAR Promotes Sensitive Antigen Recognition and Controlled T-cell Expansion Upon mRNA Vaccination

Chimeric antigen receptor (CAR) T cells are efficacious in patients with B-cell malignancies, while their activity is limited in patients with solid tumors. We developed a novel heterodimeric TCR-like CAR (TCAR) designed to achieve optimal chain pairing and integration into the T-cell CD3 signaling complex. The TCAR mediated high antigen sensitivity and potent antigen-specific T-cell effector functions in short-term in vitro assays. Both persistence and functionality of TCAR T cells were augmented by provision of costimulatory signals, which improved proliferation in vitro and in vivo. Combination with a nanoparticulate RNA vaccine, developed for in vivo expansion of CAR T cells, promoted tightly controlled expansion, survival, and antitumor efficacy of TCAR T cells in vivo. Significance: A novel TCAR is tightly controlled by RNA vaccine–mediated costimulation and may provide an alternative to second-generation CARs for the treatment of solid tumors.


Introduction
Adoptive cell therapy (ACT) with autologous T lymphocytes genetically engineered to express chimeric antigen receptors (CAR) is an effective treatment for patients with B-cell malignancies. However, this technology has proven less effective in treatment of solid tumors. Three of the main challenges for CAR T-cell therapies in solid tumors are antigen escape, limited CAR T-cell persistence and CAR-mediated on-target or off-target toxicities. Hence, there is an unmet need for a CAR T-cell therapy that is effective in this context. Similar to T-cell receptors (TCR) which bind to antigen peptide-loaded MHC molecules, CARs mediate strong interaction of modified T cells with surface nonclassical immunologic synapses (11). In comparison with endogenous TCRs, CAR-28ζ have been shown to have lower sensitivity in recognition of their target and reduced signaling capacity (12), which leads to the hypothesis that signaling through the CD3 complex could avoid antigen low or negative tumor relapse in the clinics (13). In this context, previous studies on TCR function found that the full set and quality of the 10 immunoreceptor tyrosine-based activation motifs (ITAM) in the TCR/CD3 complex is important for TCR signaling and robust and scalable T-cell function (14). This might also explain some of the observed shortcomings of some second-generation CAR scaffolds, which rely exclusively on the three ITAM motifs of CD3ζ . In addition to the ITAM quality and multiplicity, the formation of presynaptic TCR clusters has also been associated with TCR sensitivity and function (15,16).
We hypothesized that in particular for a CAR treatment of solid tumors, the lack of cognate antigen encounter in the periphery and low antigen expression in the tumor lesion might favor receptors with physiologic signaling and higher antigen sensitivity. To test this hypothesis, and using our previous experience in TCR design (17), we developed a TCR-like CAR (TCAR) designed to more closely mimic the dimeric TCR structure and physiologic signaling pathway. Similar to previous designs, the scFv domains were fused to Cα and Cβ. In contrast to other CAR designs, however, we arranged identical variable domains in tandem on either chain (VH-VH-Cα/VL-VL-Cβ) to enhance dimerization of the construct via the variable domains (18,19) and optimize integration into a synthetic TCAR/CD3 complex. We postulated that enhanced dimerization may allow us to use weakly interacting human instead of strongly interacting murine TCR C-domains.
For proof-of-concept studies, we engineered the TCAR to target two wellcharacterized and safe tumor-associated antigens from the Claudin family of tight junction proteins, Claudin 6 (CLDN6) and Claudin 18.2 (CLDN18.2). Both have previously been deemed safe for CAR-based therapies due to their cancer tissue specific expression (20,21). Antibody and CAR therapies against both targets are currently in clinical development, and the antigen-specific variable domains selected for TCAR development are based on those existing designs. The CLDN6-specific variable domains are highly related to those of IMAB027, a clinical-stage IgG1 mAb [(22); NCT02054351, (23)] and used in a second-generation CAR that is currently being evaluated in patients with CLDN6-positive relapsed or refractory advanced solid tumors (NCT04503278). The CLDN18.2-specific variable domains originate from the CLDN18.2-specific antibody Zolbetuximab (previously IMAB362). This antibody is in advanced clinical development for the treatment of gastric and pancreatic cancers (24), while CLDN18.2 is also being targeted by secondgeneration CAR-based therapy in clinical trials (NCT04404595, NCT04966143, NCT03890198, NCT03874897).
As the TCAR was designed to recruit CD3, and signal in a TCR-like fashion, it was similarly expected to undergo only limited antigen-dependent proliferation against tumor cells, which are naturally lacking in costimulatory molecules. We therefore provided costimulatory signals by applying our recently described CAR-T cell Amplifying RNA-lipoplex vaccine (referred to as CARVac) technology in vivo. CARVac is based on the systemic administration of RNA lipoplexes (RNA-LPX) encoding the full-length CAR target antigen, which has previously been shown to provide costimulus to CAR-engineered T cells (20,25).
We first optimized the TCAR design, tested chain pairing and recruitment of Tcell endogenous CD3 by different formats in vitro with RNA-transfected cells.
The lead candidate was then developed further and retrovirally transduced into T cells, to characterize the receptor in long-term in vitro experiments and also test proliferation, expansion, and antitumor activity in mice in vivo in combination with the CARVac technology. This work furthers our understanding of CAR design and function and provides a new tool for the development of effective therapies to treat solid tumors.

TCAR Design
The goal of this study was to identify a CAR design, which uses the T-cell endogenous CD3 complexes for T-cell signaling and function. Two wellestablished tumor-associated target antigens of the Claudin family (CLDN6 and CLDN18.2) were used as CAR target antigens. Variable domains of IMAB206-C46S (WO20121560018) or IMAB362 (WO2013174404A1) were applied for the generation of CLDN6-or CLDN18.2-specific CARs, respectively. The CLDN6specific variable domains are highly related to IMAB027, a therapeutic antibody against CLDN6 (22), while Zolbetuximab [previously IMAB362 (24)] was taken for CLDN18.2-specific receptor generation. FMC63 was used for the generation of CD19-specific CARs. scFv fragments were cloned by fusion of variable domains in indicated orientation connected via glycine-serine rich linkers. Several different TCR-like CARs (referred to as TCAR) were generated by fusing antibody variable domains, and chimerized to the constant domains of the mouse or human TCRα or TCRβ chain via a short gagq containing sequence. An artificial disulfide bond between Cα and Cβ was introduced at position hTRAC T47C (Uniprot: P01848) or mTRAC T48C (UniProt: A0A075B662) and hTRBC S57C (UniProt: P01850) or mTRBC [UniProt: A0A075B5J4 (26)]. For the incorporation into retroviral backbones, both the CLDN6-and CLDN18.2-TCAR chains were linked via self-processing 2A-elements (27) on a single construct [VL-VL-Cβ-P2A-VH-VH-Cα (28)] and were subcloned into the self-inactivating retroviral vector pES12.6 with the orientation VL-VL-Cβ-P2A-VH-VH-Cα (29). CAR-BBζ and CAR-28ζ constructs were used as second-generation reference constructs (20,30). The CAR-BBζ construct was designed as described previously (20). For a CAR-28ζ , a previously published scaffold was used [Uniprot: P10747, hCD28 with mutations P208A, P211A, P212A (30)]. To perturb antigen binding of CLDN6-specific CAR constructs, site-specific mutations of CDR3 regions in VH domains were performed (XL QuikChange site-directed mutagenesis, Agilent). The CDR3 region of the heavy chain variable domain was identified using the IMGT database (31). According to IMGT database, nomenclature mutations were introduced in VH at position F113A.
A murine CLDN6-specific TCR was used (WO20121560018) as a TCR control. For Myc-tagged immunoreceptors, two Myc-tags were fused in tandem N-terminally to the receptors' variable domains.
Codon-optimized receptor and antigen sequences were synthesized by MWG Eurofins and subcloned into vectors for RNA production or retroviral vectors. For in vivo bioluminescence imaging experiments, the receptors were combined with 2A-linked firefly luciferase and enhanced GFP (eGFP) reporter genes.

RNA Production and Retroviral Supernatants
The vectors pST1-T7-GG-hAg-MCS-2hBg-A30L70 or A120 were used as in vitro transcribed (IVT) RNA production templates (human α-globulin, two 3 human β-globulin untranslated region (UTR) and a poly(A) tail containing 30 + 70 adenosine residues separated by a 10 nucleotide linker or a continuous 120 adenosine sequence). The 5 and 3 UTRs and poly(A) tails have been previously optimized for stability and protein translation (20,32). IVT RNA production and simultaneous capping with β-S-anti-reverse cap analog were performed as previously described from linearized DNA templates (33).
GALV and MLV-E pseudotyped retroviral supernatants for transduction of human and murine T cells were produced by transfection of 293Vec-Galv and Platinum-E cell lines with the receptor encoding self-inactivating retroviral vector pES12.6 (29).

Animals
NOD.Cg-Prkdc scid Ilrg tmWjl /SzJ (NSG) breeding pairs were purchased from Jackson laboratory. Age and sex mixed siblings were used for the xenograft experiment. Female BALB/c or C57BL/6BrdCrHsd-Tyr c mice (Envigo and Janvier) were purchased, while C57BL/6-Thy1.1 and BALB/c-Thy1.1 donor mice (Jackson Lab) were bred in the animal facility of BioNTech SE, Germany. All animal experiments were performed under specific pathogen-free conditions and according to German animal ethics experimentation regulations.
Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll-Paque PLUS (Amersham Biosciences) density gradient centrifugation from buffy coats from healthy donors. For immature dendritic cell generation (referred to as DC), CD14 + cells were enriched using magnetic-activated cell sorting (MACS) technology (Miltenyi Biotec). CD14 + cells were differentiated toward DCs by culture media consisting of RPMI1640 GlutaMax supplemented with 5% (v/v) pooled human serum (PHS, OneLambda), 1 mmol/L sodium pyruvate, nonessential amino acids and 1,000 IU/mL rhGMCSF and rhIL4 (both Miltenyi Biotec) over 5-7 days, reported previously (20). CD14 + cell-depleted PBMCs were subjected to MACS-mediated cell separation or frozen for proliferation assays with receptor-transfected T cells. Frozen PBMCs were thawed one day prior to coculture establishment, the respective T-cell population was isolated using MACS technology and electroporated with RNA. Each experiment using DCs was performed in the autologous setting.

Flow Cytometry
The following antibodies and fluorescent dyes were used for flow cy-

Retroviral Transduction of Human and Mouse T Cells
Human T cells were enriched and activated from CD14 + -depleted PBMCs on day 1 with Dynabeads Human T-Expander CD3/CD28 (Gibco; 3 beads/cell) in X-VIVO 15 media supplemented with 5% PHS in the presence of 450 IU/mL rhIL7 and 50 IU/mL rhIL15 (Miltenyi Biotec) for 3 days. On day 4, activated T cells were debeaded and incubated on GALV-pseudotyped gamma retrovirus precoated RetroNectin plates at a concentration of 10 6 cells/mL in IL7/IL15supplemented culture media. On day 6, cells were seeded for expansion at 0.5 × 10 6 cells/mL in fresh media containing cytokines. Cells were either directly used on day 8 to assess receptor surface expression, T-cell phenotype and in vitro/ in vivo effector functions or cryopreserved.

AACRJournals.org
Cancer Res Commun; 2(8) August 2022 1 mmol/L sodium pyruvate, 10 mmol/L HEPES, 50 μmol/L β-Mercaptoethanol, 50 IU/mL Penicillin, and 50 μg/mL Streptomycin 450 IU/mL rhIL7 and 50 IU/mL rhIL15. A total of 24 hours after bead activation, cells were gently spun down (1 hour, 37°C, 300 × g) and incubated on MLVE-pseudotyped gammaretroviral vector precoated-RetroNectin-plates. After additional overnight cultivation, spin-down transduction was repeated on freshly viral particles coated plates. On day 4, mouse T cells were debeaded and expanded in fresh media containing rhIL7 and rhIL15. Adoptive cell transfer was performed on day 7. To avoid the high competition of TCAR with endogenous TCRs for CD3 in mouse T cells, mouse T cells were transduced with TCAR carrying mouse C-domains constructs.

In Vitro Assays
Endogenously antigen-expressing or RNA-transfected tumor cells or autologous DCs were used for functional testing of receptor-transfected CD8 + or transduced CD4 + /CD8 + T cells. For IFNγ secretion and cytotoxicity assays, CD8 + T cells were activated for 3 days using plate-bound anti-hCD3ε antibody (clone OKT3; BioXcell) in the presence of 50 or 250 IU/mL rhIL2 (Novartis; Supplementary Fig. S1C) and rested for an additional 3 days. For proliferation assays, nonactivated CAR RNA-electroporated CD8 + or CAR-transduced CD4 + /CD8 + T cells were stained with 0.8 μmol/L carboxyfluorescein diacetate succinimidyl ester (CFSE) or 1 μmol/L CellTrace Violet (Invitrogen). Dilution of the fluorescent dye over daughter generations was measured by flow cytometry as a surrogate for proliferation after 5 days of coculture. IFNγ secretion assays were performed using the kit ELISA Ready-SET-Go (Invitrogen).
Absorbance was detected using a Sunrise Microplate Reader (Tecan).
Luciferase-based cytotoxicity assays were performed using preactivated electroporated CD8 + or transduced T cells as described previously (36). Target cells were coelectroporated with indicated amounts of antigen and 5 μg of luciferase encoding IVT RNA. The next day, transfected target cells were seeded into 96-well Nunc white plates (Thermo Fisher Scientific) and cocultured with preactivated and receptor modified T cells. After 3 hours, Triton-X100 (0.2% v/v) was added to positive control wells and HEPES (Gibco) buffered Moonlight d-luciferin (BD Biosciences) at a concentration of 12 mg/mL to all wells.
Luminescence of cocultures was measured after the indicated timepoints using an Infinite M200 plate reader (Tecan). For impedance-based cytotoxicity assays, the xCELLigence (OMNI life science) system was used. Attachment of adherent tumor cells to 96-well E-plates (Agilent) was measured as an increase in impedance or cell index (CI) units over 20 hours. CI impedance measurements were performed according to the instructions of the supplier. A total of 2 × 10 4 target cells were seeded per well in 96-well E-plates. After 20 hours, effector cells were added and cytotoxicity was determined by decrease in CI with the formula

Statistical Analysis
Descriptive statistics of flow cytometry data was performed using FlowJo v10.6.2. Inferential statistical analysis of datasets was performed using Graph-Pad Prism 6 or 9. All results are represented with mean + SD of technical replicates or + SEM of biological replicates. Unpaired t test was used to compare for statistical significance between slope of linear regressions. Two-way ANOVA tests were performed in case of comparison of experimental data of two or more groups with either one or two independent variables. Multiple comparison testing was done using the Sidak test. Analysis of mice with tumors <1,500 mm 3 was performed with Mantel-Cox. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.

Data and Materials Availability
The CLDN6 and CLDN18.2-specific antibodies IMAB027 and IMAB362 and their respective idiotype specific antibodies are courtesy of Astellas Pharma GmbH. Material requests should be addressed to Astellas Pharma GmbH. The data generated in this study are available upon request from the corresponding author.

A TCAR Format Mediates Improved Chain Pairing and CD3 Recruitment
We initially set out to develop a TCAR by testing four alternative formats (TCAR prototype, TCAR1, TCAR2, and TCAR3; Fig. 1A  To test the function of each of the TCAR formats, they were transfected into human T cells and cultured together with CLDN6-transfected autologous DCs (CLDN6 + DC), and IFNγ secretion was measured as an indicator of antigen recognition. The TCAR prototype, by design, was cotransfected with a mouse Cα domain to achieve heterodimerization and integration into the CD3 signaling complex. All four TCAR formats with mouse constant domains were found to be robustly expressed on the T-cell surface and showed antigen-specific recognition of target cells (Fig. 1B). The TCAR prototype was the least effective of the four formats at inducing IFNγ secretion, and also did not induce T-cell proliferation against CLDN6 + DCs ( Supplementary Fig. S1), potentially due to low receptor-mediated signaling. TCAR2 and particularly TCAR3, both capable of binding the antigen between both chains, were superior to both the TCAR prototype and TCAR1 regarding surface expression and IFNγ secretion (Fig. 1B).
Mouse TCR constant domains were initially chosen for the construct because they stabilize TCR dimer formation (39), but they may induce human antimouse antibodies when used clinically (40). We therefore also tested the TCAR formats 1-3 with human C-domains. All CARs were expressed on the cell surface of T cells. While TCAR1 and 2 were characterized by low surface expression and antigen-specific IFNγ secretion (Fig. 1B, right), TCAR3 was the format with most robust expression and mediated prominent IFNγ secretion.
To assess the impact of chain interaction and valency on the activity of TCAR3, we generated and tested three sequence variants of this construct. Backmutation of the artificial disulfide-bond integrated between the human constant domains, which reduces interaction between them (26) consequently reduced TCAR3 function (Supplementary Fig. S2A). Importantly, truncation of the Nterminal VH and VL domains on both TCAR3 chains, reduced the interaction between the two receptor chains, which almost completely abolished its function ( Supplementary Fig. S2B). Mutation of the distal VH1 domain (mutVH1), which affects antigen-binding avidity, resulted in strongly impaired TCAR3 functionality, while mutations in the proximal VH2 domain (mutVH2) did not ( Supplementary Fig. S2C). Taken together, these results indicate that tightness of interaction between the receptor chains, rather than avidity, improves function of the TCAR3 and allows for human TCR constant domains in its scaffold.
Interaction between TCR constant domains strongly impacts assembly with the CD3 signaling complex and subsequent downstream signaling. We therefore evaluated CD3 recruitment by TCAR1-3 by assessing surface expression of the subunit CD3ε. We used the T-cell leukemia cell line Jurkat76 (J76) that lacks endogenous TCR expression and hence CD3 surface translocation (41). TCAR-positive J76 cells showed surface expression of the endogenous CD3 subunit, CD3ε (Fig. 1C, left). As expected, we observed CD3ε translocation for the TCAR prototype, but not for CAR-28ζ . The TCAR1-3 with mouse C-domains recruited CD3ε to the cell surface more efficiently than TCAR1-3 with human C-domains (Fig. 1C, middle and right). Among the TCAR formats with human C-domains, TCAR3 led to the highest CD3ε signals (Fig. 1C, right). Because CD3ε forms heterodimers with CD3δ and CD3γ, CD3ε surface expression served as proxy for full CD3 complex recruitment. Co-immunoprecipitation studies with J76 cells confirmed that TCAR3 associated with CD3ζ and CD3γ chains, while CAR-BBζ did not ( Supplementary Fig. S3). On the basis of these aggregate data, TCAR3 with human constant domains, from now on referred to as TCAR, was selected for further characterization studies in human T cells.

TCAR Mediates Highly Sensitive Antigen Recognition
Next, we characterized TCAR function in RNA-transfected human T cells in vitro. In cocultures with CLDN6 + DCs, the extent of T-cell responses was largely comparable with that of second-generation CLDN6-CARs and depended on the CLDN6 expression level on DCs ( Fig. 2A). CLDN6-TCAR T cells exerted robust IFNγ secretion, cytotoxicity, and proliferation. Notably, in IFNγ secretion assays, TCAR T cells recognized even ultralow antigen densities on CLDN6 + DCs ( Fig. 2A, left). We observed that in human T cells transfected with high but equimolar amounts of receptor-encoding RNA, surface expression of CLDN6-TCAR was lower than that of second-generation CARs ( Supplementary Fig. S4A). This disparity in surface expression increased with escalating amounts of transfected receptor-encoding RNA ( Supplementary  Fig. S4B).

Because TCAR surface expression is reliant on CD3 recruitment and CAR-BBζ
is not, this result suggested that the availability of CD3 subunits and competition for them with endogenous TCR molecules may be a limiting factor in surface expression of TCAR (42). We therefore assessed the relationship between receptor surface expression and IFNγ secretion upon recognition of CLDN6 + DCs. For any given receptor expression level, the IFNγ secretion response was higher with CLDN6-TCAR than CLDN6-CAR-BBζ T cells suggesting a higher sensitivity of CLDN6-TCAR T cells (Supplementary Fig.  S4C). We confirmed this assumption by testing cytotoxicity against the human SK-OV-3 ovarian cancer cell line, which expresses CLDN6 endogenously at extremely low levels ( Supplementary Fig. S4D and S4E).

Because differences in receptor expression induced strong variation in T-cell
function, for all following T-cell assays the percentage of CAR T cells was calibrated (unless otherwise indicated) to ensure comparability between receptor types.
For early proof-of-concept studies, we used transient expression of receptorencoding RNA in T cells. To validate TCAR function in long-term experiments better reflecting the in vivo setting, we retrovirally transduced T cells for stable antigen receptor expression. We used titrated retroviral supernatants and an multiplicity of infection of 1 for transductions to achieve a maximum of one copy per cell, which corresponds to approximately 30% of CAR-expressing T cells. Both CLDN6-and CLDN18.2-specific TCAR T cells generated by viral transduction showed marked surface expression and sensitive recognition and killing of CLDN6 + and CLDN18.2 + DCs ( Supplementary Fig. S5).

Costimulation Promotes TCAR Function and Persistence
Next, we conducted a series of experiments to assess the performance of TCAR T cells upon engagement with human tumor cell lines. CLDN6-TCAR T cells exerted potent and antigen-dependent IFNγ secretion and cytotoxicity against human CLDN6 + Colo-699-N lung cancer cells ( Supplementary Fig. S6A and S6B, left and middle). However, proliferation of these cells was low, particularly in comparison with proliferation of CLDN6-CAR-BBζ T cells ( Supplementary  Fig. S6B, right).
In a repetitive in vitro killing assay, CLDN6-or CLDN18.2-TCAR and CAR-BBζ T cells were challenged three times at 5-to 7-day intervals with tumor spheroids derived from CLDN6 + or CLDN18 + PA-1 cells (Fig. 2B). The repetitive spheroid assay was performed in the absence of exogenous cytokines   (Fig. 2C). However, CLDN18.2-TCAR T cells were not as efficient as CLDN18.2-CAR-BBζ T cells, which mediated complete tumor regression in all mice within 4 weeks (Fig. 2C, middle).
Frequency of adoptively transferred CLDN18.2-TCAR T cells in peripheral blood of these mice was found to be low 17 days after adoptive T-cell transfer (Fig. 2C, right). As persistence of transferred CAR T cells is known to be critical for their clinical effect (43)(44)(45), we looked for ways to increase the persistence of the TCAR T cells. BBζ T cells did not benefit further from this provision of costimulatory signals, presumably due to costimulation already provided by the CAR construct itself ( Fig. 3A and B).
We next verified the effect of costimulation in the repetitive in vitro killing assay. CLDN-TCAR T cells rechallenged with spheroids coexpressing costimulatory molecules cleared CLDN + spheroids more successfully than those challenged without costimulation (Supplementary Fig. S7B). Similarly, rechallenge with CLDN + DCs as a physiologically relevant source of costimulation, improved the efficiency of TCAR T cells at killing CLDN + PA-1 spheroids ( Supplementary  Fig. S7C).

In Vivo Salvage via CARVac RNA-LPX Vaccination Promotes TCAR Activity Against Solid Tumors
Having shown that for sustained functionality and proliferation, TCAR T cells benefit from costimulation along with target antigen engagement in vitro, we explored in vivo translation of this concept. We have previously reported a nanoparticulate RNA-LPX vaccine (25), which delivers the CAR antigen bodywide into lymphoid compartments for antigen-specific stimulation of the transferred CAR T cells. Presentation of the natively folded target antigen on resident DCs in the lymphoid tissue was accompanied with activation of the DCs and thus local augmentation of costimulatory signals (20). We evaluated the upregulation of costimulatory molecules after RNA-LPX treatment and demonstrated antigen-specific expansion of CLDN6-and CLDN18.2-CAR-

BBζ T cells in vivo. Improved engraftment of CAR T cells and regression of large tumors in mouse models was achieved at subtherapeutic CAR T-cell numbers (20).
We further asked whether this so-called CARVac approach could be used to provide costimulation and thus enhance the performance of TCAR T cells in vivo. First, we treated DCs with titrated amounts of CLDN6-encoding RNA-LPX. The resulting expression of CLDN6 on DCs induced stimulation, cytokine secretion, and proliferation of cocultured CLDN6-TCAR T cells in a dose-dependent manner (Supplementary Fig. S8).
Next, we evaluated the CARVac approach in vivo. Because of the suboptimal lymphoid compartments in NSG mice, we switched to syngeneic mouse models. We first confirmed the upregulation of costimulatory molecules on splenic DCs after RNA-LPX treatment ( Supplementary Fig. S9A). Thy1.2 + C57BL/6 mice underwent total body irradiation and were engrafted with 10 6 congenic (Thy1.1 + ) mouse T cells transduced with CLDN6-or CLDN18.2-TCAR (containing mouse C-domains) coexpressing luciferase and eGFP for in vivo bioluminescence imaging and flow cytometric analysis (VL-VL-mCβ-P2A-VH-VH-mCα-T2A-luc-eGFP). We used TCARs with murine C-domains as surrogates in murine T cells to circumvent the weak competition of TCARs carrying human C-domains with the endogenous mouse TCRs for CD3 recruitment. Starting 7 days after ACT, mice were vaccinated weekly for 21 days with 20 μg RNA-LPX encoding CLDN6 or CLDN18.2. Bioluminescence imaging revealed that a single vaccination of RNA-LPX induced rapid expansion of TCAR T cells (Fig. 3C, top; Supplementary Fig. S9B). This expansion of TCAR T cells peaked 3-4 days after vaccination and was followed by a subsequent contraction phase, mimicking the physiologic T-cell response. Consecutive RNA-LPX administrations in the following weeks induced similar kinetics. Importantly, no in vivo expansion of CLDN6-or CLDN18.2-specific TCAR T cells was observed when RNA-LPX encoding a control antigen was administered. Antigen-specific activation of CLDN6-or CLDN18.2-specific TCAR T cells showed comparable expansion rates. As expected, CLDN-specific CAR-BBζ -T cells expanded strongly upon RNA-LPX treatment due to their in-build costimulatory signaling moieties ( Supplementary Fig. S10). After initial expansion, mice were vaccinated again after a 28-and then 24-day treatment break to examine the expansion potential of TCAR T cells at later timepoints after treatment breaks and assess exhaustion of the T cells after repetitive stimulation. Indeed, a large increase in TCAR T-cell frequency was achieved, as indicated by luciferase signals reaching even higher levels than after initial expansion. Flow cytometric analysis of peripheral blood confirmed stable engraftment and persistence of TCAR T cells supported by RNA-LPX administration (Fig. 3C, middle). Accordingly, assessment of T-cell phenotype revealed that antigen-specific RNA-LPX vaccination induced an enrichment for effector memory-like TCAR T cells (Fig. 3C,   bottom).
To assess the impact on antitumor activity, BALB/c mice with large mouse CLDN18.2 + CT26 colon tumors (mean tumor volume 87 mm 3 ) were submitted to adoptive cell transfer with a subtherapeutic dose of mouse Thy1.1 + CLDN18.2-TCAR T cells followed by two injections (2 weeks apart) of CLDN18.2-LPX or control LPX (Fig. 4A, top row). While CLDN18.2-TCAR T cells alone had no effect on tumor growth, combining them with vaccination significantly delayed tumor growth and prolonged survival (Fig. 4A, middle row). Each of the vaccine doses was found to increase TCAR-expressing T cells in peripheral blood followed by a contraction of this adoptively transferred T-cell population (Fig. 4A, bottom left). No signs of toxicity were observed in any of the treatment groups by visual inspection and the body weight of all animals remained stable (Fig. 4A, bottom right).

Discussion
CAR designs have greatly evolved in the last three decades, which has resulted in the approval of CAR therapies for hematologic disorders. However, clinical success of these therapies for solid malignancies is so far limited. In this article, we show a novel TCAR design with potent effector functions and higher sensitivity than second-generation CARs. In contrast to the latter, TCAR does not contain costimulatory signaling moieties like 4-1BB or CD28. Consequently, we observed lower antigen-specific expansion of TCAR the question of how a TCAR scaffold would perform, and designed a TCAR that retargets T cells toward a surface antigen using the endogenous T-cell CD3 signaling complexes for physiologic TCR-like activation.
The TCAR contains an innovative tandem configuration of variable domains (VH1-VH2-Cα and VL1-VL2-Cβ) that sets it apart from the single-chain arrangements found in other candidates that we and others tested, and from previous second-generation CAR formats. In TCAR, the antigen can serve as a "noncovalent linker" across both receptor chains as reported for soluble monomeric and dimeric scFvs (47). This, in combination with the tandem configuration of V-domains, might result in higher α/β-chain pairing of the TCAR along the entire polypeptide chains, thus strengthening the TCAR/CD3complex. Truncation of the N-terminal VH1 and VL1 domains in the respective chains, a structure similar to the recently published STAR receptor (48), completely abolished function of a TCAR with human C-domains ( Supplementary   Fig. S2B). However, mutation of the proximal VH2 had only a minor effect. This demonstrated that TCAR function is less determined by its valency, than by its strong chain interaction. Notably, and in contrast to the STAR receptor format, this allows for incorporation of human TCR C-domains in the TCAR (Supplementary Fig. S2). In TCRs, an increase in interaction between αand β-chains is associated with a higher TCR/CD3 interaction (39), which also underlies the observed variations in CD3 recruitment by different TCARs (Fig. 1). However, it is well known that human TCR α/β-domains do not have adequate domain pairing forces in mouse T cells to recruit endogenous CD3 for surface expression (49). Hence, we circumvented the weak expression of TCARs in mouse T cells by taking TCARs with murine Cα/β domains as a surrogate for in vivo experiments. Although murine Cα/β TCARs exhibited improved expression and function in vitro (Fig. 1) and pronounced antitumor reactivity in mice, we do not propose using them in future clinical trials but suggest using human Cα/β TCARs. With the latter, lower immunogenicity would be expected and hence, long-term persistence of TCAR-T in vivo.
By incorporation into the natural CD3 complex, and without fusions to costimulatory moieties, the TCAR follows a different and more physiologic T-cell activation than previous CAR designs and induces strong and sensitized responses against its antigen in vitro and in vivo. Davenport and colleagues observed the formation of nonclassical immune synapses characterized by multifocal Lck clusters in CAR-T as opposed to the monocentric synapse formation in TCR-T cells (11). This in turn resulted in different T-cell signaling kinetics.
Recently, Gudipati and colleagues reported on reduced antigen sensitivity of CAR-T due to inefficient recruitment of the most proximal adaptor protein ZAP-70 in T-cell signaling (12). These alterations in CAR-T-cell signaling most likely result from a different signalosome formation (TCR/CD3 vs. CAR clustering) in the immune synapses upon antigen encounter eventually impacting their effector function.
One proposed factor underlying those observations in second-generation CARs could be the absence of CD3ε ITAMs in their design. The latter have been shown to influence Nck recruitment, actin dynamics, maturation, and function of the immune synapses (50)(51)(52). Incorporation of CD3ε together with CD3ζ into a CAR-28ζ has consequently been shown to improve persistence of CAR T cells in vivo (53). But even incorporation of CD3ε into a CAR scaffold might not solve issues like tonic signaling seen for some second-generation CARs. Here, we postulate that TCARs in contrast to second-generation CARs form aggregates with CD3 chains highly related to antigen-sensitive TCR/CD3 complexes.
The intrinsic sensitivity of the TCAR could be used for the treatment of antigenlow solid tumor entities or hematologic malignancies such as CD19 leukemia or lymphomas to prevent relapse of B cell-derived tumor cells (54), the latter even providing costimulation in trans. Although we have observed a major contribution of the distal VH1 to antigen binding in the antigen monospecific TCAR, a bispecific TCAR may further decrease the chance of antigen escape variants in hematologic or solid tumors.
Other approaches that exploit the endogenous CD3 complex for signaling have been published previously (55)(56)(57). T cell membrane-bound bispecific antibodies targeting CD3 or tumor antigens (TAC) or alternatively scFv-CD3 fusions (TRuC) have been used to cluster the endogenous TCR/CD3 complexes. But in contrast to the TCAR design, they could not benefit from endogenous TCR knockout strategies (42,(58)(59)(60). This intervention likely reduces alloreactivity for off-the-shelf approaches and increases TCAR function. However, TCR knockouts can also improve safety of a T-cell product by reducing off-target toxicities mediated by transactivation of the endogenous TCR (61). In addition, CARs with fusions of Fab fragments to the constant domains of the γδTCR also recruit the T-cell endogenous CD3 complexes and show function in hematologic tumor models (57). However, γδTCRs reveal a quaternary structure varying from αβTCRs and expose different surfaces to CD3 subunits (62). Consequently, γδTCR constant domains are considered to show lower recruitment of CD3δε, implying distinct signal routes and strengths (63). This might interfere with the meticulously regulated activation and negative feedback loops seen in αβ-T cells. Of note, all previously mentioned CD3 complex-dependent CAR scaffolds were argued to induce lower toxicity due to lower cytokine secretion. However, we observed a reduction in cytokine secretion for TCAR T cells only upon exposure to high amounts of target antigen, pointing toward a negative feedback mechanism of this novel receptor ( Fig. 2A and Supplementary Fig. S6, left). In addition to the previous constructs, a recent publication by Liu and colleagues (48) provided insight into the function of the STAR receptor, a CAR (VH-mCα + VL-mCβ) similar to the truncated version of TCAR (VH-hCα + VL-hCβ) presented in this article, but with murine C-domains. In line with our observations (Supplementary Fig. S2), the STAR receptor elicited strong effector functions, such as cytokine secretion. However, contradictory AACRJournals.org Cancer Res Commun; 2(8) August 2022 to our data and other published studies (64)(65)(66), this receptor did not require costimulation for proliferation induction against tumor cells and could even outperform second-generation CARs (48).
Our novel TCAR design represents an alternative receptor scaffold to induce potent T-cell activity against tumor cells also in the context of solid malignancies. MHC-independent and well-regulated specific T-cell activation, in combination with the strong proximal CD3 signaling, induced potent and antigen-sensitive effector functions, underlining the potential of this novel receptor scaffold. The fact that TCAR T cells did not require fusions of costimulatory signaling moieties for long-term persistence in vivo (Fig. 3), supports the notion that this novel receptor construct largely benefits from T-cell activation mediated by the T-cell endogenous signaling machinery. In NSG mouse experiments, we saw potent initial antitumor responses of TCAR T cells (Fig. 2). Compared with T cells carrying the CAR-BBζ with intrinsic costimulation, TCAR T cells showed lower proliferation when exposed to cells not providing costimulation. We argue that this poses a major safety advantage of TCARs over second-generation and similar published CAR approaches, because the lack of costimulus perturbs strong TCAR T-cell expansion and reactivity in healthy tissues which can have fatal consequences due to on-target/off-tumor reactivities (67). Alternatively to the tightly controlled provision of costimulation using CARVac, a recent publication presented the benefit of T cell-encoded CD80 and 4-1BBL on T-cell persistence in the context of an HLA-independent TCR [VH-Cβ + VL-Cα (68)].
The data provided by us and others, demonstrate that CARs recruiting the Tcell endogenous CD3 complexes have high antigen affinity, and benefit greatly from antigen expression and costimulation provided in trans via RNA-LPX treatment or even in cis via T cell-encoded costimulatory molecules. However, we observed a delayed rather than a complete resolution of tumors in vivo in immunocompetent mice. Because RNA-LPX strongly supports in vivo expansion of TCAR-T, we argue that their antitumor function could be enhanced by additional mRNA vaccinations. Besides the lack of antigen and costimulation for TCAR-T, the encounter of a strong immunosuppressive tumor microenvironment might also influence the T-cell antitumor function.
The more physiologic signaling induced by the TCAR compared with secondgeneration CARs may render TCAR-T susceptible to coinhibitory signals such as CTLA-4, PD-1, TIM-3, LAG-3 or cytokines like TGFβ and IL10 originating from tumor-associated macrophages, myeloid-derived suppressor cells, regulatory T cells, or even the tumor cells themselves. Genomic modifications of TCAR-T or combination with established therapies like checkpoint inhibitors could abrogate such inhibitory signals. They could, for example, preserve Tcf7positive T-cell populations characterized by higher capacity for self-renewal (69) and counteract activation-induced cell death (AICD) or exhaustion. The latter state has previously been characterized by upregulation of factors like TOX/TOX2 or members of the NR4A family (70,71). While coexpression of the TCAR together with the CD3 subunits could increase TCAR-dependent signaling (42) and avoid antigen escape, the previously described TCR knockout strategies combined with TCR locus-targeted TCAR integration could support a fine-tuned TCAR signaling to avoid AICD (60). However, fine tuning of TCAR function alone might not resolve the roadblocks coming from coinhibitory signaling. An increasing number of studies have tested CAR-Ts genetically modified to cope with suppressive signals such as PD-L1 or TGFβ in the TME (72)(73)(74). For TCRs, first promising patient data have already been published in this regard (75).
In the clinical setting, our novel TCAR design together with the CARVac strategy could in the future enable physicians to follow an expand-on-demand approach and safely fine-tune TCAR T-cell functions in the patient. The TCAR could become a modular tool to integrate T-cell activating signals together with exogenous TCRs, immunomodulators (76), chimeric costimulatory receptors (77), checkpoint inhibitors, or cytokines to systemically expand genome engineered TCAR-Ts. The T-cell endogenous CD3 subunits recruited by TCAR could help to integrate complex T-cell signaling events and incorporate this biologic in its physiologic T-cell niche.