A modular γδ TCR-T platform combining KRAS pMHC targeting with re-dosable mRNA engager redirection

Abstract

Solid tumors often evade TCR-engineered αβ T cells when antigen expression varies or when the restricting Human Leukocyte Antigen (HLA) allele is lost. γδ T cells, in contrast, detect cellular dysregulation through non-peptide/Major Histocompatibility Complex (MHC) cues, including phosphoantigens and stress ligands, and can be developed as allogeneic therapies. Although intratumoral γδ T cell signatures are associated with improved outcome across cancers, γδ recognition itself is broad and still selected within the thymus just as αβ T cell receptors (TCRs) are. It does not, however, anchor specificity to a defined driver-mutation pMHC epitope. We therefore asked whether a high-affinity, co-receptor–independent αβ TCR could graft oncogenic-driver specificity onto γδ T cells while leaving the endogenous γδ TCR intact. We knocked the KRASG12V/HLA-A*11:01 TCR A11v into primary human γδ T cells. Engineered cells co-expressed the transgenic αβ TCR and the endogenous γδ TCR and lysed KRASG12V/HLA-A*11:01⁺ tumor cells in vitro and in vivo. To cover potential resistance through loss of HLA-A*11:01, we delivered an mRNA lipid nanoparticle (LNP) encoding a secreted mesothelin×CD3 (M5) bispecific T cell engager (TCE). LNP-M5 produced circulating TCE that redirected γδ A11v T cells and polyclonal bystander T cells to kill mesothelin⁺ targets, accompanied by development of higher γδ A11v T cell counts in vivo. In humanized mice bearing mixed HLA-A*11:01⁺ and HLA-A*11:01 − KRASG12V tumors, γδ A11v T cells produced transient control, whereas adding LNP-M5 yielded complete responses and prolonged survival. Thus, this two-part therapy couples invariant driver targeting to tunable redirection and addresses loss of the restricting HLA allele, a central escape route for TCR-based therapy. It provides an off-the-shelf reagent to enable KRAS-anchored treatment with the ability to redeliver the reagent.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13045-026-01836-0.

Introduction

Engineered T cell therapies can produce durable remissions in patients with hematologic malignancies [1, 2]. Solid tumors impose different constraints. Trafficking and persistence are limited, antigen expression is heterogeneous, and the tumor microenvironment restricts effector cell infiltration and suppresses effector function. Efforts to compensate by increasing potency can narrow the safety margin and can cause severe on-target, off-tumor toxicity [3–7]. These constraints frame a central problem for solid tumor cell therapy. Tumor recognition must be precise, and immune engagement must remain controllable in tissue.

Targets that tumors can discard will be discarded [8, 9]. Driver mutations are harder to evade because they are clonal and required to sustain tumor fitness. Most driver proteins are intracellular and reach the cell surface only as peptide–major histocompatibility complex (pMHC) ligands. This makes TCR-based recognition the principal route to targeting drivers. RAS mutations occur in approximately 19% of human cancers [10], and KRAS accounts for most RAS-mutant cases, making it a dominant oncogenic driver.

Mutant KRAS is now a validated immunologic target accessible by specific small molecule inhibitors [11] as well as cellular therapies. Adoptive transfer of KRAS-mutant–reactive T cells can mediate regression in patients with metastatic disease [12]. For KRAS^G12V presented by HLA-A*11:01, multiple groups, including ours, have isolated and characterized mutation-specific TCRs with antitumor activity in engineered T cells [13–16]. At the same time, the biology of pMHC targeting sets clear vulnerabilities. Tumors can escape by losing the restricting HLA allele, a common evolutionary event under immune pressure [17]. They can also escape by disrupting antigen presentation, including loss of β2-microglobulin [18]. A second limitation is cellular. In αβ T cells, transgenic TCR chains can mispair with endogenous chains and create new specificities, motivating designs that detect or prevent mispairing [19–24].

γδ T cells offer a distinct platform for TCR gene therapy. They can sense cellular dysregulation through non-peptide cues, including phosphoantigen-linked pathways mediated by BTN2A1 and BTN3A1 butyrophilins [25–27]. In pan-cancer analyses, intratumoral γδ T cell signatures emerge as a favorable prognostic immune population [28]. Clinical studies of allogeneic Vγ9Vδ2 products report an encouraging safety profile [29]. Importantly, γδ T cells lack endogenous αβ TCR chains, avoiding αβ/αβ mispairing while preserving a native γδ TCR. This native recognition is broad and stress-linked, and it is not anchored to a single clonal driver mutation. Most circulating γδ T cells also lack CD4 and CD8 co-receptors [30], which limits the activity of many αβ TCRs in this cellular context.

Prior work has shown that transfer of αβ TCRs or CAR constructs [31] into γδ T cells can generate functional antigen-specific effectors without mixed TCR dimers in vivo [32–34]. However, these studies also revealed that signaling and antitumor activity through the introduced αβ TCR are markedly reduced in the absence of CD4 or CD8 co-receptors, which most peripheral γδ T cells lack [32–34]. This co-receptor constraint suggests a direct solution: a high-affinity, co-receptor–independent αβ TCR should impose driver-mutation specificity on γδ T cells while preserving the endogenous γδ TCR.

Even with driver anchoring, solid tumors can escape through heterogeneity and loss of the restricting HLA context. We hypothesized that a second, tunable targeting element delivered as lipid nanoparticle (LNP)-formulated mRNA would extend antigen coverage while preserving control over the cellular product. LNP-mRNA provides transient, non-integrating expression that decays within days, can be administered in multiple cycles without anti-vector immunity, and allows dose and schedule to be tuned independently of the engineered cells. Preclinical studies support these properties, as LNP-mRNA delivery of CD3-engaging bispecifics generates circulating functional TCE and antitumor activity [35–37], and clinical experience with transient mRNA-engineered mesothelin CAR T cells demonstrates the feasibility of such reversible dosing strategies in patients [38, 39].

Here, we combine these concepts. We engineer primary human γδ T cells with a co-receptor–independent KRAS^G12V/HLA-A*11:01 αβ TCR (γδ A11v) and pair this product with an mRNA-LNP module encoding a secreted mesothelin×CD3 (M5) bispecific T cell engager (TCE). The TCR anchors recognition to a clonal driver mutation. The mRNA module provides dose-controlled redirection to a surface antigen and recruits both the engineered γδ product and polyclonal bystander T cells to mesothelin-expressing targets. We test whether this two-component design widens tumor coverage while sustaining specific immune engagement tunable in vivo.

Materials and methods

Experimental model and subject details

Cell lines and culture conditions

COR-L23 (ACC 882), YAPC (ACC 382), Caki-2 (HTB-47), K562 (CCL-243), and HBL-1 (T8204) were cultured in complete RPMI (R10; RPMI 1640 supplemented with 10% fetal bovine serum, 1% penicillin-streptomycin, 1% HEPES, and 1% GlutaMAX) at 37 °C in 5% CO₂. HEK293T (ATCC CRL-3216) and ASPC-1 (ATCC CRL-1682) were cultured in complete DMEM (D10; DMEM supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin) at 37 °C in 5% CO₂.

Unless otherwise indicated, tumor cell lines were engineered to express luciferase and GFP for tracking in vitro and in vivo (click beetle luciferase variants are specified in the relevant experiments). Cell lines were obtained from ATCC or DSMZ, authenticated by the University of Arizona Genetics Core (2019), and routinely screened for Mycoplasma (MycoAlert; Lonza).

Human cancer cell lines engineered to express HLA-A*11:01 (YAPC-A11 and COR-L23-A11) or KRAS^G12V (Caki-2^G12V) were generated as previously described (stable lentiviral transduction for HLA-A*11:01 and genome engineering for KRAS^G12V) and cultured in R10 medium under standard conditions [14, 40]. Artificial antigen-presenting cells (aAPCs) were generated by lentiviral transduction of parental K562 cells with constructs encoding human CD86, 4-1BBL, IL-15Rα, IL-15, and membrane-bound IL-7 (IL-7 fused to the IgG4 hinge and Fc regions and the CD4 transmembrane domain). Cells were single-cell sorted using a FACSAria (BD) into 96-well plates containing R10 supplemented with Normocin (100 µg/ml). After 3 weeks, individual clones were screened for transgene surface expression and T-cell stimulatory activity. Clones with stable high expression and stimulatory capacity were expanded, γ-irradiated (100 Gy), and cryopreserved for subsequent use as previously described [41].

Primary human T cells

Peripheral blood mononuclear cells (PBMCs) were obtained from healthy donors by leukapheresis through the Human Immunology Core (RRID: SCR_022380) at the University of Pennsylvania under Institutional Review Board–approved protocols. All donors provided written informed consent in accordance with the Declaration of Helsinki, International Conference on Harmonization Good Clinical Practice guidelines, and the United States Common Rule.

Mice

Male and female NOD/SCID/IL-2Rγ-null (NSG) mice (8–12 weeks) were obtained from the University of Pennsylvania Stem Cell and Xenograft Core (RRID: SCR_010035). All animal studies were approved by the University of Pennsylvania Institutional Animal Care and Use Committee and performed in a specific pathogen–free facility on a 12-hour light/dark cycle with ad libitum access to sterilized food and acidified water.

Methods details

Design and molecular cloning of lentiviral and AAV constructs

mKRAS-specific TCR sequences were derived from constructs previously generated at the University of Pennsylvania Center for Cellular Immunotherapies (as reported elsewhere) [40]. To enable selective disruption of endogenous TCR loci during genome editing, the TRAC and TRBC regions of the mKRAS TCR were codon-modified to render them resistant to Cas9 ribonucleoproteins targeting endogenous TRAC and TRBC1/TRBC2 sequences. Codon-modified TCR constructs (GenScript) were cloned into the third-generation self-inactivating lentiviral transfer vector pTRPE. High-titer lentivirus was produced as previously described [42].

For targeted knock-in, an EF1α-driven mKRAS TCR cassette was cloned into an rAAV6 donor backbone (GenScript) and flanked by 400 bp homology arms spanning the CCR5 guide RNA cut site to enable homology-directed repair. rAAV6 was produced by triple plasmid transfection of a packaging cell line using the rAAV6 transfer plasmid, rep/cap plasmid, and adenoviral helper plasmid (Charles River).

T cell culture and expansion

PBMCs were isolated by Ficoll-Paque density gradient centrifugation (Cytiva) and cryopreserved (4 × 10⁷ cells per vial).

For αβ T-cell cultures, thawed PBMCs were enriched by negative selection (Pan T Cell Isolation Kit; Miltenyi). Cells were stimulated with CD3/CD28 Dynabeads (3:1 bead: cell; Thermo Fisher Scientific) and cultured for 3 days in OpTmizer CTS medium (Gibco) supplemented with 5% human AB serum and recombinant human IL-2 (100 U/mL; PeproTech) (“T cell medium”).

For γδ T-cell cultures, thawed PBMCs were cultured in T cell medium supplemented with zoledronic acid (STEMCELL Technologies) to expand γδ T cells. On day 5, residual αβ T cells were depleted using biotinylated anti-TCRαβ antibody followed by anti-biotin magnetic microbeads (Miltenyi). γδ T cells were genome engineered as described below. Twenty-four hours after editing, γδ T cells were expanded with irradiated K562 aAPCs. Medium was refreshed every 2 days. Cells were harvested and cryopreserved on day 13.

CRISPR-Cas9-mediated T cell genome engineering

Primary αβ T-cell engineering was performed as previously described [40]. Briefly, activated CD4 and CD8 T cells were lentivirally transduced with the mKRAS-specific TCR at 24 h after activation and subsequently electroporated with CRISPR-Cas9 ribonucleoproteins to disrupt endogenous TCRα and TCRβ expression.

For γδ T-cell engineering, we selected CCR5 as the integration locus based on its established clinical precedent as a target for therapeutic gene editing in human T cells and the documented tolerance of CCR5 disruption in human lymphocytes [23, 43, 44]. CCR5-targeted editing in γδ T cells was performed using established protocols [45]. At 120 h after activation, Cas9 ribonucleoproteins were assembled with a CCR5-targeting crRNA (Integrated DNA Technologies; 5′-CCTGCCTCCGCTCTACTCACTGG-3′) and tracrRNA. Cells were washed, resuspended in P3 buffer (Lonza), and electroporated using a Lonza 4D-Nucleofector X Unit. For targeted integration, cells were transduced with rAAV6 encoding the CCR5 knock-in donor (Charles River).

Flow cytometry

Cells were stained with LIVE/DEAD Fixable Aqua (Thermo Fisher Scientific) in PBS for 20 min at room temperature in the dark, washed, and stained with fluorophore-conjugated antibodies in FACS buffer (PBS + 2% FBS) for 30 min at 4 °C.

Transgenic TCR expression was assessed using PE/Cy5 anti-human αβ TCR and FITC anti-human TCR Vβ3.1. Additional antibodies included: BV570 CD45RO; BV650 or APC-H7 CD3; FITC Vδ1; PerCP-eFluor710 γδ TCR; BV421 Vδ9; BV711 Vδ2; PE/Cy7 or PE-CF594 CCR7; APC CCR5; PE-CF594 CD25; BV605 CD69; Alexa Fluor 700 HLA-DR; BV421 perforin; BV421 PD-1; PE/Cy7 LAG-3; PE TIM-3; BV785 CD19; PE/Cy7 HLA-ABC; biotinylated HLA-A*11:01 followed by APC streptavidin; APC mesothelin; and V500 CD14. Data were acquired on an LSR II Fortessa (BD Biosciences) and analyzed in FlowJo v10.

Generation of M5 TCE mRNA and LNP formulation

The M5 TCE mRNA encodes a tandem scFv consisting of an anti-mesothelin scFv linked to an OKT3 anti-CD3 scFv, with a C-terminal 6×His tag. The construct was designed with a synthetic 5′ UTR and Kozak sequence to support translation. DNA templates were synthesized (TriLink) and transcribed in vitro using T7 RNA polymerase with CleanCap M6 (TriLink) to generate N¹-methylpseudouridine (m¹Ψ)-containing capped mRNA. Purification was optimized to reduce dsRNA byproducts and enrich full-length transcripts (magnetic bead cleanup, cellulose-based purification, and oligo(dT) affinity purification). RNA integrity was assessed by Bioanalyzer (Agilent 2100), and concentration was measured by NanoDrop. CleanCap M6 firefly luciferase and EGFP mRNAs (TriLink; m¹Ψ-containing) served as controls.

mRNA was encapsulated into lipid nanoparticles using the Cytiva RNA Delivery LNP Kit on the NanoAssemblr™ Ignite™ microfluidic mixing system with NxGen™ cartridges (Cytiva), according to the manufacturer’s protocol. The kit provides a pre-optimized ionizable lipid mix together with formulation, dilution, and cryopreservation buffers; the composition contains a proprietary ionizable amino lipid alongside helper lipids, cholesterol, and a PEGylated stabilizer. As reported by the manufacturer for this kit, LNPs prepared under standard conditions exhibit a hydrodynamic size of ~ 60–80 nm with PDI < 0.2 and encapsulation efficiency > 90% (Cytiva, RNA Delivery LNP Kit Technical Brochure, CY45446-16Aug24-BR, 2024). mRNA encapsulation efficiency in our preparations was measured by the Quant-iT™ RiboGreen™ RNA assay (Invitrogen). LNPs were stored at − 65 °C to − 85 °C.

Lipofection with LNP-M5

In vitro LNP delivery was performed under serum-free conditions supplemented with recombinant human ApoE (1 µg/mL) to reconstitute the ApoE-mediated biomolecular corona that drives LDL receptor–dependent LNP internalization in vivo [46, 47], consistent with the manufacturer’s recommended protocol (Cytiva, RNA Delivery LNP Kit Technical Brochure, 2024). Adherent cells (HEK293T and tumor lines) were plated 24 h before treatment. Primary T cells were activated with CD3/CD28 Dynabeads (Thermo Fisher Scientific, Waltham, Massachusetts) for 3 days before LNP exposure. LNPs were added directly to cells at the indicated mRNA doses and incubated for 48 h at 37 °C in 5% CO₂. Supernatants were collected and clarified (500 × g, 10 min) and concentrated when indicated (Amicon Ultra; Merck). For primary T cells, residual ApoE and LNPs were removed by washing before downstream assays. Control LNPs were included as specified.

Expression and purification of recombinant M5 TCE

M5 TCE was expressed in CHO-S cells (GenScript) by transient transfection of a pTRPE-based expression construct. Conditioned supernatants were clarified and purified by immobilized metal affinity chromatography (GenScript) via the His tag, then buffer exchanged into PBS (pH 7.2). Purity was assessed by SDS-PAGE and SEC-HPLC. Endotoxin was measured by limulus amebocyte lysate assay. Final material was > 90% pure by SDS-PAGE and 91% by SEC-HPLC, with endotoxin 0.117 EU mg⁻¹, and stored at 4 °C.

SDS-PAGE, western blot, and SEC-HPLC

For SDS-PAGE, samples were mixed with Laemmli buffer with or without reducing agent, heated (95 °C, 5 min), and resolved by electrophoresis. Gels were stained with Coomassie Brilliant Blue or used for immunoblotting.

For Western blotting, proteins were transferred to PVDF using the iBlot 2 system (Thermo Fisher Scientific), blocked in TBST + 5% milk, and probed with anti-His primary antibody (GenScript) followed by HRP-conjugated secondary antibody. Signal was developed with ECL (Pierce) and imaged on an Odyssey XF (LI-COR).

SEC-HPLC was performed on a TSKgel G3000SWXL column using Na₂SO₄ in 0.18 mol/L phosphate buffer (pH 6.7 ± 0.3) as mobile phase, with detection at 280 nm.

Quantification of M5 TCE by capture ELISA

MaxiSorp plates (ThermoFisher Scientific, Waltham) were coated overnight at 4 °C with biotinylated human mesothelin (Fc, AviTag; ACROBiosystems; 5 µg/mL), washed (PBS + 0.05% Tween-20), and blocked (3% BSA in PBS, 2 h, room temperature). Recombinant M5 TCE was used to generate a standard curve. Supernatants or mouse plasma/serum were diluted in PBS + 0.2% BSA and incubated for 2 h at room temperature. Bound TCE was detected with HRP-conjugated anti-6×His antibody (clone GG11-8F3.5.1; 1:625; Miltenyi) for 1 h. Signal was developed with SuperSignal ELISA Femto substrate (Pierce; 5 min) and read on a SpectraMax M3 (Avantor) Pharmacokinetics of circulating M5 TCE.

NSG mice received 3 µg LNP-formulated luciferase control mRNA or M5 TCE mRNA by intravenous injection. Blood was collected by facial vein puncture at indicated time points into lithium-heparin microvettes (Sarstedt), centrifuged (4,000 rpm, 5 min, room temperature), and plasma was stored at -65 °C to -85 °C until ELISA.

In vitro cytotoxicity assay

For luciferase-based killing assays, engineered or non-engineered T cells were cocultured with luciferase-expressing targets at indicated effector-to-target (E: T) ratios. For 24-hour assays, targets included COR-L23, YAPC, and Caki-2. For TCE-dependent assays, purified TCE, TCE-containing supernatants, or LNP-encoded mRNA were added as specified, and freshly isolated PBMC-derived T cells were cocultured with luciferase-expressing targets (COR-L23, ASPC-1, K562, and HBL-1) for 48 h.

D-luciferin (final 20 µg/mL; Promega) was added, cultures were incubated (10 min, 37 °C), and luminescence was measured on a SpectraMax M3. Percent specific lysis was calculated as:

For flow cytometry-based killing against K562-MSLN, T cells were cocultured with 5 × 10⁴ target cells at indicated E: T ratios. After 24 h, cells were stained with CD3 PE-Cy5.5, Mesothelin APC, and LIVE/DEAD Aqua in PBS + 0.5% FBS. Counting beads (123count eBeads; ThermoFisher Scientific, Waltham) were added before acquisition. Residual live mesothelin⁺ targets were quantified by flow cytometry.

Cytokine secretion analysis

Cytokines were measured in coculture supernatants using the LEGENDplex human CD8/NK 13-plex panel (BioLegend) per the manufacturer’s protocol.

Cell line-derived mouse xenograft model

NSG mice were injected intravenously with 4 × 10⁵ luciferase-expressing COR-L23-A*11:01 tumor cells. Five days later, mice received 1 × 10⁷ engineered or non-engineered T cells by intravenous infusion. Tumor burden was monitored weekly by bioluminescence imaging. D-luciferin (150 mg/kg; Promega) was injected intraperitoneally (200 µL PBS), and mice were imaged 10 min later using an IVIS Lumina II (Caliper) in groups of up to five animals per session, with each animal physically separated by opaque dividers placed on the imaging stage to prevent optical bleed between adjacent mice. When an individual mouse exhibited bioluminescence intensity that could compromise quantification of neighboring animals, the animal was removed from the cohort and re-imaged separately. Signal was quantified as total photon flux (photons/s) using fixed-dimension regions of interest (ROIs) drawn around each animal in Living Image v4.8 (Caliper), with automated background subtraction applied uniformly across sessions. All photon flux values are reported as per-mouse measurements. Peripheral blood was collected by facial vein puncture to quantify circulating T cells by flow cytometry using 123count eBeads. Body weight and clinical status were monitored throughout. Mice meeting euthanasia criteria due to tumor burden or xenogeneic graft-versus-host disease were euthanized per institutional guidelines.

PBMC-humanized NSG tumor models

NSG mice were humanized by intraperitoneal injection of 1 × 10⁷ human PBMCs obtained from a healthy donor. Six days later, peripheral blood was collected, red blood cells were lysed, and human immune engraftment was assessed by flow cytometry (human CD45 and CD3). Engraftment performance, donor identifier, and per-experiment peripheral blood lymphocyte composition (human CD45⁺, CD3⁺, CD4⁺, CD8⁺, and CD19⁺ populations) for each PBMC-humanized in vivo experiment described in Figs. 5g–i and 6 are summarized in Supplementary Table 2. Independent donors were used across the two PBMC-humanized experiments; donor identities are indicated in Supplementary Tables 2 and referenced in the corresponding figure legends. Mice failing engraftment criteria (< 5% human CD3⁺ within CD45⁺ cells, as defined for each experiment) were excluded.

Fig. 5.

Tunable in vivo immune redirection via LNP-encoded M5 TCE. a, Dose–response curve of M5 TCE–mediated cytotoxicity against COR-L23-A11 and K562 target cells, measured as specific lysis after 24 h. The EC50 value for COR-L23-A11 is indicated. b, Antigen-specific cytotoxicity of T cells against HBL-1, ASPC-1 and COR-L23 target cells at an E: T ratio of 5:1, using supernatants collected from 293T cells transfected with increasing doses of LNP-M5, measured as specific lysis after 24 h (two-way ANOVA with Dunnett’s multiple-comparisons test vs. mock within each cell line; n = 6 independent donors). c, Cytokine secretion in supernatants collected 24 h after co-culture, measured using the LEGENDplex™ human CD8 13-plex panel (two-tailed paired t-test; n = 3 independent donors). d, Experimental schematic of in vivo LNP administration, serial blood sampling and longitudinal bioluminescence imaging (BLI). NSG mice received intravenous injection of LNP-M5 [3 µg mRNA] or LNP-Control [3 µg mRNA] (fLuc payload), followed by serial blood collection and BLI at the indicated time points. e, Longitudinal quantification of total photon flux (photons/s) following administration of LNP-M5 or LNP-Control (fLuc payload) (two-way RM ANOVA with Sidak’s multiple-comparisons test; n = 4 mice per group). f, Plasma concentration of M5 TCE measured by ELISA at the indicated time points following intravenous administration of LNP-M5 or LNP-Control (fLuc payload) (two-way RM ANOVA with Sidak’s multiple-comparisons test; n = 4 mice per group). g, Experimental schematic of the humanized PBMC-NSG xenograft model. Mice were implanted subcutaneously with COR-L23-A11 tumor cells (4 × 10⁵), followed by adoptive transfer of non‑transduced T cells (NT-T) (5 × 10⁶) and intravenous administration of LNP-M5 [3 µg mRNA] or LNP-Control [3 µg mRNA] (fLuc payload) at the indicated time points. h, Longitudinal quantification of total photon flux (photons/s) showing tumor growth kinetics for mice treated with NT T cells, LNP-Control (fLuc payload), or LNP-M5 (two-way mixed-effects model (REML) with Tukey’s multiple-comparisons test; n = 4–5 mice per group). i, Individual tumor burden values at day 41 for each treatment group (one-way ANOVA with Tukey’s multiple-comparison test; n = 4–5 mice per group). Data are shown as mean ± SD. *P < 0.05; **P < 0.01; ***P < 0.001; ns, not significant. Schematics [d and g] created with BioRender.com

Fig. 6.

Complementary immune redirection using γδ A11v T cells and LNP-encoded M5 TCE. a, Experimental schematic of the humanized mixed xenograft model. NSG mice were implanted intravenously with a 1:1 mixture of COR-L23-A11 (HLA-A*11:01⁺) and COR-L23-WT (HLA-A*11:01⁻) tumor cells (2 × 10⁵ each), followed at day 5 by adoptive transfer of 1 × 10⁷ γδ Mock T cells, 1 × 10⁷ γδ A11v T cells plus LNP-Control [3 µg mRNA] (GFP payload) and 1 × 10⁷ γδ A11v T cells plus LNP-M5 [3 µg mRNA]. LNPs were administered intravenously at the indicated time points. b, Longitudinal quantification of total photon flux (photons/s) showing tumor growth kinetics for each treatment group (two-way mixed-effects model (REML) with Tukey’s multiple-comparisons test; n = 5 mice per group). c, Individual tumor growth trajectories for mice receiving γδ Mock T cells, γδ A11v T cells plus LNP-Control (GFP payload), or γδ A11v T cells plus LNP-M5. d, Total tumor burden at days 19 and 33 for individual mice in each treatment group (one-way ANOVA with Tukey’s multiple-comparisons test; n = 5 mice per group). e, Absolute counts of circulating γδ A11v T cells in peripheral blood at day 21 (D21) (Kruskal–Wallis with Dunn’s multiple-comparisons test; n = 5 mice per group). f, Proportions of CD4⁺ and CD8⁺ αβ T cells within the circulating human T cell compartment on day 21 (D21). g, Kaplan–Meier survival analysis of mice treated with γδ Mock T cells, γδ A11v T cells plus LNP-Control (GFP payload), or γδ A11v T cells plus LNP-M5. Statistical significance was assessed by log-rank (Mantel–Cox) test. Data are shown as mean ± SD. *P < 0.05; **P < 0.01; ***P < 0.001; ns, not significant. Schematics [a] created with BioRender.com

Tumors were established by intravenous or subcutaneous inoculation of 4 × 10⁵ luciferase-expressing tumor cells (route and timing specified in the relevant figure schematics). Mice were randomized after tumor establishment by baseline bioluminescence. Treatments were administered by intravenous bolus injections. For LNP studies, mice received 3 µg LNP per dose at weekly intervals for four doses unless otherwise indicated.

Statistical analysis

Two-group comparisons used two-tailed unpaired or paired Student’s t-tests for parametric data or Mann-Whitney U tests for non-parametric data. Comparisons among ≥ 3 groups used one- or two-way ANOVA with appropriate multiple-comparison correction. Survival was analyzed by log-rank (Mantel–Cox) test. Analyses were performed in GraphPad Prism v9.1.0. P < 0.05 was considered significant.

Results

A11v TCR drives CD8‑independent antitumor activity in vivo

A11v is a KRAS^G12V/HLA‑A*11:01‑specific TCR selected for high sensitivity and reduced CD8 dependence in vitro [14, 40]. Here we test whether A11v is sufficient to drive antitumor activity in vivo in the absence of CD8 co‑receptor support, establishing a rationale for deploying this receptor in CD4⁻/CD8⁻ γδ T cell platforms.

Primary human CD4⁺ and CD8⁺ T cells were engineered to express A11v by lentiviral transduction with concurrent CRISPR-Cas9 disruption of endogenous TRAC/TRBC to reduce mispairing and competition for CD3 [20, 48]. Mock CD4⁺ and CD8⁺ T cells were nucleofected under identical conditions without TRAC/TRBC targeting and used as controls (Supplementary Fig. 1a). Flow cytometry confirmed uniform surface CD3 and A11v Vβ staining in both CD4⁺ and CD8⁺ A11v-engineered T cells (96% in each subset) (Supplementary Fig. 1b-c). Across effector-to-target (E: T) ratios (1:10 to 10:1), A11v‑engineered CD4⁺ and CD8⁺ T cells lysed HLA‑A*11:01⁺ KRAS^G12V tumor cells, with substantial killing at E: T ≥ 1:1, whereas mock controls were inactive (Supplementary Fig. 1d).

We then tested A11v function in vivo using isolated CD4⁺ (CD4 A11v) and CD8⁺ (CD8 A11v) products. NSG mice were injected intravenously with 4 × 10⁵ COR‑L23‑A11 human large cell lung carcinoma cells expressing click beetle red (CBR) luciferase (Fig. 1a). Five days later, mice received a single infusion of 1 × 10⁷ CD8 A11v, CD4 A11v, or a 1:1 mixture (CD4/8 A11v). Controls received a matched CD4/8 mock product (Fig. 1a). Tumor burden was followed by serial bioluminescence imaging (Fig. 1b; Supplementary Fig. 1e). By day fourteen, A11v‑treated cohorts showed marked tumor reduction versus controls, with the mixed CD4/8 product producing the deepest regressions (Fig. 1c). Either subset alone delayed progression, whereas the CD4/8 mixture sustained control longer (Fig. 1c) At day 100, complete clearance occurred in 7/17 mice (41.2%) treated with CD4/8 A11v, compared with 1/17 (5.9%) for CD8 A11v and 5/18 (27.8%) for CD4 A11v (Fig. 1c; Supplementary Fig. 1f). All A11v groups improved survival versus mock (P < 0.001), with the CD4/8 cohort showing the greatest benefit (76% alive at day 100) (Fig. 1d). Together, these data demonstrate that A11v can drive MHC class I-restricted antitumor activity in vivo, motivating its deployment in CD4⁻/CD8⁻ cellular platforms.

Fig. 1.

A11v TCR drives CD8-independent antitumor activity in vivo. a, Experimental schematic of the xenograft model. NSG mice were implanted intravenously with COR-L23-A11 tumor cells (4 × 10⁵), followed at day 5 by adoptive transfer of 1 × 10⁷ mock-transduced CD4/CD8 T cells (black), CD4 A11v T cells (light blue), CD8 A11v T cells (orange), or combined CD4/CD8 A11v T cells (green). Tumor burden was monitored longitudinally by bioluminescence imaging. b, Longitudinal quantification of total photon flux (photons/s) showing tumor growth kinetics for each treatment group, shown for the sub-cohort of animals imaged on the first schedule (Supplementary Fig. 1e shows the second sub-cohort, imaged 1–2 days later) (two-way mixed-effects model (REML) with Dunnett’s multiple comparisons test comparing all treatment groups to mock CD4/CD8 T cells group, n = 10). c, Individual tumor growth trajectories for mice receiving mock CD4/CD8 T cells, CD4 A11v T cells, CD8 A11v T cells, or combined CD4/CD8 A11v T cells. d, Kaplan–Meier survival analysis of mice treated with mock CD4/CD8 T cells, CD4 A11v T cells, CD8 A11v T cells, or combined CD4/CD8 A11v T cells. Statistical significance was assessed by log-rank (Mantel-Cox) test. Data are shown as mean ± SD. *P < 0.05; **P < 0.01; ***P < 0.001; ns, not significant. Schematics [a] created with BioRender.com

A11v knock‑in yields γδ T cells that co‑express transgenic αβ TCR and the endogenous γδ TCR

We next deployed A11v in primary human γδ T cells to test whether A11v function and expression are maintained in a γδ T cell background. We used targeted genome engineering coupled to feeder‑based expansion methods previously shown to support robust γδ T‑cell propagation [49, 50] (Fig. 2a). Activated γδ T cells were engineered using CRISPR-Cas9 targeting the CCR5 safe‑harbor locus with AAV6 donor delivery to enable site‑specific knock‑in of the A11v cassette [45, 51] (Supplementary Fig. 2a). The A11v αβ TCR was expressed from a constitutive EF1α promoter to enforce defined expression in the γδ background. Flow cytometry showed homogeneous A11v surface expression in 50% of engineered γδ T cells (γδ A11v) with limited donor‑to‑donor variation (Fig. 2b-c). αβ TCR knock-in cells retained endogenous γδ TCR expression while co‑expressing the introduced αβ TCR (Fig. 2b). The proportion of A11v-expressing cells was lower in γδ A11v T cells (approximately 50%) than in lentivirally transduced αβ A11v T cells (approximately 99%), a difference set by the distinct delivery methods rather than by competition between the introduced αβ TCR and the endogenous γδ TCR for CD3; per-cell A11v surface density, measured as Vβ3.1 mean fluorescence intensity, was comparable between the two products, and endogenous Vδ2 expression was retained (Supplementary Fig. 2d-e). Engineering was efficient, with > 90% concordance between knock‑in and associated knockout events (Supplementary Fig. 2c).

Fig. 2.

Targeted A11v knock-in generates γδ T cells that co-express the transgenic αβ TCR and the endogenous γδ TCR. a, Schematic of the genome-editing strategy for targeted insertion of the A11v αβ TCR into the CCR5 locus of γδ T cells (CCR5KI–TCR-A11v). Engineered γδ T cells co-express the introduced αβ TCR specific for mKRAS and the endogenous γδ TCR. b, Representative flow cytometry plots showing surface expression of the introduced A11v TCR, detected by αβ TCR and TCRVβ staining, in γδ Mock and γδ A11v T cells, gated on live CD3⁺ cells. c, Summary of TCRVβ editing efficiencies in γδ Mock and γδ A11v T cells (two-tailed unpaired t test with Welch’s correction; n = 10–14 independent donors). d, T cell expansion during manufacturing and CRISPR–AAV editing over 13 days of culture (D13/D0) for γδ Mock and γδ A11v T cells (two-tailed unpaired t test with Welch’s correction; n = 6–10 independent donors). e, Purity of γδ T cells in γδ Mock and γδ A11v products on day 13 of culture, shown as the percentage of γδ (dark blue) versus αβ T cells (white). f, Representative flow cytometric plots of CCR7 and CD45RO expression in γδ Mock and γδ A11v T cells on day 13 of culture. g, Memory T cell subset distribution on day 13 of culture, defined by CCR7 and CD45RO expression: stem cell–like memory (TSCM), CCR7⁺CD45RO⁻; central memory (TCM), CCR7⁺CD45RO⁺; effector memory (TEM), CCR7⁻CD45RO⁺; terminally differentiated effector memory (TEMRA), CCR7⁻CD45RO⁻. All In vitro experiments were conducted using TCR T cells manufactured from different healthy subjects. Data are shown as mean ± SD. *P < 0.05; **P < 0.01; ***P < 0.001; ns, not significant. Schematics in [a] were created with BioRender.com

To expand engineered γδ T cells, we used irradiated K562 artificial antigen‑presenting cells (aAPCs) expressing CD86, 4-1BBL, IL-15R, IL-15 and membrane‑bound IL‑7, an approach established for γδ T cell propagation [45, 49, 50]. This regimen supported sustained proliferation of γδ A11v cells, yielding 124‑fold expansion by day 13 while maintaining A11v expression and knock-in efficiency (Fig. 2d; Supplementary Fig. 2f). End‑of‑expansion phenotyping showed a highly enriched γδ T‑cell population (> 99%) with predominately memory phenotypes (41% central memory, 56% effector memory) (Fig. 2e-g; Supplementary Fig. 2b). These data demonstrate that targeted A11v insertion is compatible with γδ T cell expansion and yields a reproducible γδ A11v product that retains the endogenous γδ TCR.

γδ A11v T cells mediate strict HLA‑restricted, mutation‑specific killing in vitro and durable control in vivo

Expanded γδ A11v cells displayed an activated cytotoxic phenotype with increased expression of CD25, HLA‑DR, and CD38, and high intracellular perforin (Supplementary Fig. 3a). LAG‑3 and TIM‑3 were also increased (Supplementary Fig. 3b), consistent with activation-associated regulatory features of innate-like T cell populations [52].

We next tested A11v‑dependent killing against COR‑L23‑A11 tumor cells expressing endogenous KRAS^G12V. γδ A11v cells lysed targets in vitro within twenty-four hours and outperformed γδ mock controls (Fig. 3a). Across E: T ratios (1:10 to 10:1), lysis exceeded 50% at 1:1 (Fig. 3a). Killing was accompanied by increased secretion of TNFα and IFNγ and higher levels of granzyme B, perforin, and granulysin versus mock controls (Fig. 3b). No killing was observed against HLA‑A*11:01⁻ COR‑L23‑wildtype cells, confirming HLA‑restricted recognition (Supplementary Fig. 3c-d). In contrast, γδ mock T cells showed modest killing only at the highest E: T ratio and secreted lower levels of cytotoxic mediators (Fig. 3a-b; Supplementary Fig. 3c-d), consistent with endogenous γ9δ2 TCR activation and underscoring the added benefit of A11v TCR redirection for HLA-restricted tumor killing. To test specificity across tumor contexts, we used models in which HLA‑A*11:01 and KRAS^G12V were independently manipulated (Supplementary Fig. 3e). Caki‑2 cells (endogenous HLA‑A*11:01) were engineered to express KRAS^G12V, and YAPC cells (KRAS^G12V) were transduced to express HLA‑A*11:01(Supplementary Fig. 4). γδ A11v cells lysed targets only when both HLA‑A*11:01 and KRAS^G12V were present (Supplementary Fig. 3e), confirming strict, mutation‑specific, HLA‑restricted recognition.

Fig. 3.

γδ A11v T cells mediate mutation-specific killing in vitro and tumor control in vivo. a, Antigen-specific cytotoxicity of γδ Mock and γδ A11v T cells against COR-L23-A11 target cells at increasing E: T ratios, measured by specific lysis after 24 h (two-way RM ANOVA with Sidak’s multiple-comparisons test; n = 10 independent donors). b, Supernatant was collected at 24 h after co-culture, and cytokine levels were measured using the 13-plex LEGENDplex™ human CD8 panel (two-tailed paired t-test; n = 3 independent donors). c, Experimental schematic of the in vivo xenograft model. NSG mice were implanted intravenously with COR-L23-A11 tumor cells (4 × 10⁵), followed at day 5 by adoptive transfer of 1 × 10⁷ γδ A11v T cells or 1 × 10⁷ γδ Mock T cells. Tumor burden was monitored longitudinally by bioluminescence imaging. d, Longitudinal quantification of total photon flux (photons/s) showing tumor growth kinetics for mice treated with γδ Mock or γδ A11v T cells (two-way RM ANOVA with Sidak’s multiple-comparisons test; n = 6–7 mice per group). e, Individual tumor growth trajectories for mice receiving γδ Mock or γδ A11v T cells. f, Kaplan-Meier survival analysis of mice treated with γδ Mock or γδ A11v T cells. Statistical significance was assessed by log-rank (Mantel–Cox) test. All In vitro experiments were conducted using TCR T cells manufactured from different healthy subjects. Data are shown as mean ± SD. *P < 0.05; **P < 0.01; ***P < 0.001; ns, not significant. Schematics [c] created with BioRender.com

We next tested γδ A11v in vivo in an intravenous COR‑L23‑A11 metastatic lung cancer model (Fig. 3c). A single infusion of 1 × 10⁷ γδ A11v cells on day 5 rapidly suppressed tumor growth. By study end, 6/7 treated mice had markedly reduced tumor burden versus mock controls (Fig. 3d-e; Supplementary Fig. 5a-c), and total photon flux remained suppressed throughout follow‑up. All γδ mock‑treated mice reached endpoints within five weeks (Fig. 3f; Supplementary Fig. 5a-b). In contrast, 6/7 γδ A11v‑treated mice maintained tumor control through day 90 (Fig. 3f; Supplementary Fig. 5c), without overt toxicity on longitudinal monitoring (Supplementary Fig. 5d). Thus, γδ T cells can be endowed with strict KRAS^G12V/HLA‑A*11:01 specificity and durable in vivo activity. However, single‑epitope targeting leaves a predictable escape route in solid tumors: outgrowth of antigen‑mismatched or HLA‑mismatched variants.

LNP‑M5 drives secretion of a functional M5 TCE in vitro

To broaden tumor coverage without further modifying the engineered γδ A11v TCR construct which would add manufacturing complexity and impose a second permanent specificity that could not be modulated after infusion, we implemented a transient mRNA-based strategy to generate an M5 TCE. mRNA‑encoded bispecific antibodies can reach therapeutic serum levels after systemic delivery [36, 53]. We therefore asked whether an LNP encoding M5 could provide a re-dosable source of bispecific engager in vivo (Fig. 4a). LNP‑M5 encapsulates nucleoside‑modified mRNA encoding a secreted M5 TCE. Encapsulation efficiency and particle properties were comparable to control reporter LNPs (Supplementary Fig. 6a), and LNP‑M5 did not reduce viability at 8–48 h versus mock or control LNP (Supplementary Fig. 6b-c). To determine whether LNP-derived M5 TCE retained dual binding specificity, we performed flow cytometric binding assays using supernatants from LNP-M5 transfected cells. Supernatants containing LNP‑derived M5 bound CD3⁺ human T cells and mesothelin⁺ tumor cells, with no binding to mesothelin⁻ controls (Fig. 4b). HEK293T transfection with LNP‑M5 yielded secreted TCE detectable by ELISA (Fig. 4c) and immunoblot. Immunoblot showed a predominant 53‑kDa species consistent with monomeric M5 TCE (Fig. 4d; Supplementary Fig. 7a-c). Functionally, LNP‑derived M5 activated primary human T cells only in the presence of mesothelin⁺ targets (as assessed by CD25/CD69 upregulation; Fig. 4e). No activation occurred without mesothelin expression or in the absence of tumor targets, and control LNPs were inactive (Fig. 4e).

Fig. 4.

LNP-M5 drives secretion of a functional M5 TCE in vitro. a, Schematic of the LNP-M5 platform. Nucleoside-modified mRNA encoding the M5 TCE was formulated into lipid nanoparticles (60–80 nm), translated and secreted as a tandem scFv TCE composed of a mesothelin-binding scFv and an anti-CD3 (OKT3) scFv, enabling redirection of T cells toward mesothelin-expressing tumor cells. b, Flow cytometric analysis of M5 TCE binding to K562, K562-MSLN, COR-L23, ASPC-1 and HBL-1 tumor cell lines and to T cells, using supernatants collected from LNP-M5–transfected 293T cells. c, ELISA-based detection of secreted M5 TCE in culture supernatants from cells treated with LNP-M5, LNP-Control (GFP payload) or purified M5 protein (Pur-M5), using BSA or recombinant human mesothelin-Fc (hMSLN-Fc) as capture substrates (two-tailed unpaired t-test; n = 3 independent experiments). d, Western blot analysis of secreted M5 TCE following LNP-M5 treatment under reducing (R) and non-reducing (NR) conditions. The expected molecular weight of the M5 monomer is ~ 53 kDa, indicated by the annotated band. e, Flow cytometric quantification of T cell activation markers CD25 and CD69 following co-culture of T cells with K562, COR-L23 or ASPC-1 target cells in the absence of TCE (Mock) or in the presence of LNP-M5, LNP-Control (GFP payload) or purified M5 (Pur-M5) (two-way ANOVA with Dunnett’s multiple-comparisons test vs. mock within each cell line; n = 3 independent experiments). Data are shown as mean ± SD. *P < 0.05; **P < 0.01; ***P < 0.001; ns, not significant. Schematics [a] created with BioRender.com

In dose-response assays, M5 TCE acted in the low‑nanogram range (EC₅₀ 2.6–15.7 ng/ml; Fig. 5a; Supplementary Fig. 8). Near‑maximal activity was reached at ≥ 80 ng/ml, with peak lysis 80% (Fig. 5a). Supernatants from LNP‑M5‑transfected cells drove mesothelin‑dependent killing across tested doses (Fig. 5b) and induced cytokines consistent with T cell engagement (TNFα, IFNγ, IL‑17 A, and soluble FASL). No lysis or cytokine induction occurred against mesothelin⁻ targets or in controls (Fig. 5c).

Direct delivery of LNP‑M5 to tumor cells also enabled selective killing by primary T cells within forty-eight hours in vitro (Supplementary Fig. 9), showing that locally produced TCE is sufficient for redirection. When delivered directly to T cells, LNP‑M5 supported detectable TCE expression for up to seventy-two hours (Supplementary Fig. 10a-b) and enabled ratio‑dependent killing of ASPC‑1 targets cells (max 82%; Supplementary Fig. 10c), accompanied by cytokine release (Supplementary Fig. 10d). Control‑transfected or mock T cells were inactive (Supplementary Fig. 10c). Together, these assays show that LNP‑M5 produces a properly assembled, biologically active TCE that drives CD3‑dependent, mesothelin‑restricted killing in vitro.

Tunable in vivo immune redirection via LNP-encoded M5 TCE

We next tested whether systemic LNP‑M5 dosing yields circulating, functional TCE in vivo. As a reporter control for mRNA delivery and translation, mice were injected with 3 µg reporter mRNA-LNP, producing a strong liver‑restricted signal by whole‑body imaging from eight hours through one-hundred twenty hours (Fig. 5d-e; Supplementary Fig. 11a). After injection of 3 µg mRNA-LNP‑M5, M5 TCE was detected in serum by ELISA with a six-hour peak and sustained levels through seventy-two hours (10–42 ng/ml) (Fig. 5f). Serum from LNP‑M5‑treated mice activated primary human T cells only in the presence of mesothelin⁺ targets (CD25/CD69 induction on K562‑MSLN; Supplementary Fig. 11b). Serum collected at six to forty eight hours mediated selective lysis of K562‑MSLN targets with minimal lysis of mesothelin^‑ HBL‑1 cells (Supplementary Fig. 11c). PBS and control reporter‑LNP sera were inactive (Supplementary Fig. 11c).

We then tested antitumor activity in a PBMC‑humanized NSG model bearing subcutaneous COR‑L23 tumors (Fig. 5g). PBMCs were injected two days before tumor inoculation, and treatment began five days later. Mice received four weekly intravenous doses of 3 µg LNP‑M5-mRNA, 3 µg control reporter‑LNP-mRNA, or 5 × 10⁶ non‑transduced T cells (NT-T) (Fig. 5g). Serial imaging showed reduced tumor burden in LNP‑M5‑treated mice, whereas controls progressed (Fig. 5h). By day forty-one, tumor signal was significantly lower with LNP‑M5 (P < 0.01; Fig. 5i). LNP‑M5 did not cause acute weight loss (Supplementary Fig. 11d). Later weight changes were consistent with graft-versus-host disease (GvHD) in PBMC‑humanized mice. These data show that systemic LNP‑M5 can generate circulating TCE with in vivo antitumor activity, motivating combination with γδ A11v to address heterogeneous tumors.

Dual-targeting with γδ A11v and LNP‑M5 prevents outgrowth of HLA‑mismatched tumor variants

To model HLA‑restricted escape, we first asked whether M5 could redirect γδ effectors to kill mesothelin⁺ targets independently of HLA‑A*11:01. In vitro, γδ A11v T cells lysed COR‑L23‑A11 but not HLA‑A*11:01⁻ COR‑L23‑wildtype cells. Adding M5 restored killing of COR‑L23‑wildtype cells by both γδ T cells expressing A11v and γδ mock cells across E: T ratios, demonstrating CD3‑mediated redirection independently of HLA‑A*11:01 (Supplementary Fig. 12a).

We then built a PBMC‑humanized mixed xenograft model using COR‑L23‑A11 and COR‑L23‑wildtype to mimic loss of the restricting HLA allele while retaining mesothelin as a second target (Fig. 6a). NSG mice received PBMCs followed by intravenous engraftment with a 1:1 mixture of COR‑L23‑A11 and COR‑L23‑wildtype cells (2 × 10⁵ each). Mice then received 1 × 10⁷ γδ mock transfected T cells, γδ A11v plus control reporter‑LNP, or γδ A11v plus LNP‑M5 (Fig. 6a). LNPs were dosed intravenously at 3 µg mRNA payload weekly for four doses.

γδ A11v plus control LNP produced an initial tumor reduction, consistent with selective clearance of the HLA‑A*11:01⁺ fraction, but tumors later progressed, consistent with outgrowth of the HLA‑A*11:01⁻ tumor cells (Fig. 6b-c). In contrast, γδ A11v plus LNP‑M5 produced sustained tumor reduction, with a lower signal than both control groups from day twenty-six onward (Fig. 6b-c). Most mice in the combination cohort maintained durable control, whereas γδ mock‑treated tumors progressed continuously (Fig. 6c).

Both γδ A11v groups showed reduced total photon flux by day nineteen, but only the LNP‑M5 group sustained control; control‑LNP animals relapsed, separating by day thirty-three (Fig. 6d). Peripheral blood at day 21 showed higher circulating γδ A11v counts in the LNP‑M5 cohort (Fig. 6e), consistent with TCE‑mediated activation and proliferation in vivo. Phenotyping showed a more balanced CD4⁺/CD8⁺ distribution in the combination cohort, whereas controls were CD8‑skewed (60–70%) (Fig. 6f), consistent with broader PBMC T cell engagement under dual targeting. Combination treatment significantly prolonged survival versus both controls (Fig. 6g; Supplementary Fig. 12b), with complete responses observed in the combination group (Fig. 6g; Supplementary Fig. 12b). No acute treatment‑associated weight loss was detected; later changes were consistent with GvHD in PBMC‑humanized mice (Supplementary Fig. 12c). Consistent with a complementary contribution of the two modules, LNP-M5 administered alone was sufficient to produce sustained tumor control while LNP-Control was not (Fig. 5g–i and Supplementary Fig. 13). Together, these data demonstrate that adding LNP‑M5 prevents outgrowth of HLA‑A*11:01⁻ variants by redirecting both γδ A11v cells and PBMC‑derived polyclonal T cells to a second antigen (mesothelin). Tunable mRNA‑encoded redirection therefore complements mutation‑specific cellular therapy to maintain pressure in heterogeneous tumors.

Discussion

Solid-tumor responses to engineered T cells are limited by two dominant recurring constraints. Target expression is often heterogeneous, creating selective pressure for antigen-low or antigen-negative outgrowth [8]. In parallel, TCR-based recognition depends on intact antigen processing and presentation and on retention of the restricting HLA allele. Clinical and genomic studies demonstrate that allele-specific HLA loss and other defects in antigen presentation are common routes of immune evasion [17, 18]. These mechanisms motivate therapeutic designs that do not rely on a single, fixed receptor for targeting.

Here we tested an architecture that separates invariant recognition from adjustable redirection. The invariant element is an αβ TCR specific for KRAS^G12V presented by HLA-A*11:01. The adjustable element is an mRNA–lipid nanoparticle that drives transient in vivo production of a mesothelin×CD3 TCE. In our models, the KRAS module provided potent mutation- and HLA-restricted killing, while the TCE module supplied an orthogonal and re-dosable route to engage mesothelin⁺ tumor cells and recruit additional T cell effectors. This division is pragmatic: the driver mutation provides a stable molecular anchor, whereas antigen presentation and surface antigen expression vary across the tumor and over time.

The driver-focused premise has clinical precedent in KRAS mutant tumors. Adoptive transfer of KRAS-specific T cells has produced tumor regression in patients, and clinical experience directly illustrates the vulnerability of TCR-based approaches to loss of the restricting HLA context [12]. The present study extends this principle in two ways. First, it establishes in vivo activity of the KRAS^G12V/HLA-A*11:01 TCR across T cell subsets in a way that is consistent with functional co-receptor independence. Second, it shows that this specificity can be implemented in a γδ T cell product while retaining the endogenous γδ TCR.

The γδ platform is relevant because prior work has shown that αβ TCR transfer into γδ T cells can generate antigen-specific effectors while avoiding αβ TCR chain mispairing, a safety and potency concern in conventional TCR gene transfer [21, 32, 33]. A major practical barrier, however, is that most peripheral γδ T cells lack CD4/CD8 co-receptors, and many αβ TCRs require co-receptor support for full signaling. Earlier studies addressed this by adding CD4 or CD8 to engineered γδ cells [32]. In our system, co-receptor independence is the enabling feature. It allows an αβ TCR to function in a γδ cellular context without introducing additional genetic components, while preserving the native γδ receptor. This positions the platform for use with co-receptor–independent αβ TCRs, a class that increasingly includes affinity-engineered and naturally high-affinity neoantigen receptors identified through ongoing discovery efforts.

Retention of the γδ receptor is not merely an implementation detail. Human Vγ9Vδ2 T cells can respond to cellular dysregulation through butyrophilin-dependent phosphoantigen sensing and through recognition of stress-associated ligands [25, 26]. Pan-cancer analyses have identified γδ-associated immune signatures as favorable prognostic features across multiple malignancies, consistent with a non-redundant role for γδ biology in human tumors [28]. γδ T cells are also being evaluated as allogeneic products, with clinical experience supporting feasibility and suggesting an acceptable safety profile in some settings [29]. Within that context, our findings demonstrate that a mutation-specific αβ TCR can be installed in γδ T cells without loss of specificity, producing cells that co-express the transgenic αβ TCR and the endogenous γδ TCR and mediate robust KRAS^G12V/HLA-A*11:01-restricted tumor control in vivo.

The mixed-tumor experiments expose the main biological limitation of driver-anchored TCR targeting in solid tumors. When a fraction of the tumor lacked the restricting HLA allele, KRAS-directed γδ T cells produced only transient control despite uniform KRAS mutation status. This is not an argument against driver targeting. It is a demonstration that clonal genetics does not guarantee clonal presentation. Allele-specific HLA loss is observed in human tumors and is associated with immune selection [17], and antigen presentation pathway disruption has repeatedly emerged as a route of resistance to immunotherapies that rely on class I presentation [18]. Our model was designed to isolate this variable and thereby define what must be added to sustain control when the KRAS-HLA axis is not uniformly available.

We addressed this with a second targeting mechanism that is orthogonal to pMHC restriction and is delivered as a transient, dose-controlled input. Systemic LNP-M5 dosing produced circulating, biologically active TCE that supported mesothelin-dependent T cell activation and killing. This behavior is consistent with emerging literature showing that mRNA-LNP delivery can generate functional bispecific engagers in vivo and achieve pharmacologically useful exposure windows without permanent genetic modification of the cellular product [35]. The value of transient expression is not conceptual alone. Transient receptor expression has been used clinically to manage risk while interrogating potency for mesothelin-directed CAR T cells [38, 39]. Our approach applies the same control principle to a secreted TCE produced in vivo rather than to an engineered receptor expressed by the infused cells.

Functionally, combining the KRAS-directed γδ product with LNP-M5 changed therapeutic behavior in the mixed-HLA setting. The combination yielded sustained tumor control, including complete responses and prolonged survival, whereas KRAS-directed γδ T cells alone provided only transient benefit. Two features are central in interpreting this result. First, the TCE redirected not only the infused γδ A11v cells but also bystander polyclonal human T cells present in the humanized system, increasing effector participation without additional ex vivo engineering. Second, the combination increased circulating γδ A11v T cell counts, consistent with in vivo engagement and expansion under CD3 redirection. In aggregate, the data support a model in which the KRAS module provides high-specificity pressure when antigen presentation is intact, while the TCE module maintains pressure when the restricting HLA context is absent, using a target that is independent of class I presentation.

The broader implication is architectural rather than antigen-specific. A fixed, driver-mutation–anchored recognition element can be paired with a re-dosable redirection element that is exchanged or combined as the tumor composition and presentation change. The specific implementation here uses KRAS^G12V/HLA-A*11:01 as the invariant anchor and mesothelin as the redirection target, but the design does not require either. The same logic can be applied to additional KRAS epitopes and alleles that are already being populated by natural and engineered TCR discovery efforts [13, 14, 40]. It can also accommodate alternative engager targets as dictated by tumor type, normal tissue constraints, and antigen distribution. This consideration is particularly relevant for mesothelin. The xenograft models used here cannot directly address this issue, as the anti-mesothelin scFv in M5 does not cross-react with murine mesothelin. Therefore, safety assessment will require cross-reactive surrogate systems and, ultimately, careful evaluation in early-phase clinical studies. Beyond this cross-reactivity limitation, broader validation in patient-derived xenograft models concurrently expressing HLA-A*11:01, KRASG12V, and mesothelin will be important to assess generalizability across clinically diverse tumor contexts. This modular logic also has practical translational implications, as decoupled manufacturing of the cellular and mRNA-LNP components could enable an off-the-shelf γδ A11v cell product to be paired with adaptable LNP-M5 dosing. This is particularly relevant because T cell persistence in xenograft models often overestimates that observed clinically. In patients, adoptively transferred T cells frequently show limited persistence [39, 54–56], thereby constraining antitumor activity when efficacy depends on the durability of the infused cell product. By contrast, LNP-M5 provides TCE exposure independently of the transferred cells, allowing repeated dosing to sustain tumor redirection and therapeutic pressure even when cellular persistence is limited. In this way, the duration of therapeutic activity can be partially uncoupled from the persistence of the cell product. A clinical workflow built on repeated mRNA-LNP administration also raises additional considerations, including immune responses against the encoded TCE, anti-LNP or anti-PEG reactivity, innate immune activation, and cumulative toxicity. Reassuringly, multi-cycle LNP-M5 dosing in our studies did not produce acute or cumulative toxicity within the limits of the models used here. However, dedicated repeat-dose toxicology and immunogenicity studies in more appropriate preclinical systems will be required to define the long-term tolerability of this strategy before clinical evaluation.

In summary, this study establishes a two-part strategy in which a co-receptor-independent KRAS-specific αβ TCR installs driver-mutation specificity onto γδ T cells, and an mRNA–LNP-delivered TCE supplies transient, tunable redirection to broaden tumor coverage when the restricting HLA context is not uniformly present. By distributing recognition and redirection across separable components, the system preserves the precision of pMHC targeting while adding a controllable layer that can be re-administered and modified without re-engineering the transferred cells. This architecture supports a practical translational route in which a fixed, allogeneic γδ TCR-T product is paired with re-dosable TCE delivery to tune breadth and intensity in vivo.

Supplementary Information

Supplementary Material 1. ^{(27.4KB, xlsx)}

Supplementary Material 2. ^{(14.8MB, pdf)}

Supplementary Material 3. ^{(3.7MB, docx)}

Supplementary Material 4. ^{(28MB, pdf)}

Author contributions

A.R.F.: Conceptualization, resources, funding acquisition, investigation, methodology, writing-original draft, writing-review and editing. A.S.B.: Conceptualization, resources, investigation, methodology and editing. E.K.: Investigation, methodology. R.B.: Investigation, methodology. M.H.: Investigation, methodology. G.M.C.: Investigation, methodology. D.D.: Investigation, methodology. L.C.E.: Investigation, methodology. J.L.: Investigation, methodology. M.D.: Investigation, methodology. M.V.S.: Investigation, methodology. A.J.: Investigation, methodology. J.H.N.: Investigation, methodology. C.R.H.: Investigation, methodology. O.K.: Investigation, methodology. J.M.: Investigation, methodology. A.J.D.: Investigation, methodology. E.W.: Investigation, methodology. M.L.: Investigation, methodology. J.S.: Investigation, methodology. A.S.T.: Investigation, methodology. J.B.R.: Investigation, methodology. J.K.J.: Investigation, methodology. D.J.P.: Investigation, methodology. A.C.: Investigation, methodology. V.E.G.: Investigation, methodology. D.L.S.: Investigation, methodology, writing-review and editing. B.L.L.: Investigation, methodology, writing-review and editing. M.T.L.: Investigation, methodology, writing-review and editing. C.H.J.: Investigation, methodology, writing-review and editing. J.L.R.: Investigation, methodology, writing-review and editing. R.H.V.: Investigation, methodology, writing-review and editing. F.H.: Supervision, validation, investigation, methodology, writing-review and editing. J.A.F.: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, writing-original draft, writing-review and editing.

Funding

This study was supported by the following grants: 2025 AACR-Johnson & Johnson Interventional Oncology Stimulating Therapeutic Advances through Research Training (25-40-52-RAMI) Grants, (to A.R.F.). J.A.F. receives funding for Swim Across American Gene Editing Innovation Award. The study additionally received funding from NIH grants P01 CA214278 (to D.L.S., C.H.J., and J.A.F.), and the Bob Levis Funding Group (to B.L.L., C.H.J., and J.A.F.). Further support was provided by the National Science Foundation Engineering Research Center for Cell Manufacturing Technologies (to B.L.L. and J.A.F.), the Alliance for Cancer Gene Therapy Investigator Award in Cell and Gene Therapy for Cancer (to J.A.F.), and funding from the Parker Institute for Cancer Immunotherapy (to C.H.J.) and BlueWhale Bio (JLR). Additionally, funding was received from a U01 AG066100 via the Samuel Waxman Cancer Research Foundation and a Chambers Centurion gift (to C.H.J. and J.A.F.), and an Abramson Cancer Center P30 Core Grant P30 CA016520 (to J.A.F.).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

B.L.L.: Consultancy and advisory roles with Terumo, GSK, and Kite; co-founder and equity holder in Tmunity Therapeutics (acquired by Kite) and Capstan Therapeutics; advisory board memberships with Avectas, Capstan (Chair), Immuneel, Immusoft, In8bio, Ori Biotech, Oxford Biomedica, ThermoFisher Pharma Services, and UTC Therapeutics; member of the Alliance for Cancer Gene Therapy Board of Directors. C.H.J.: Receives royalties from Novartis and Kite paid to the University of Pennsylvania; scientific co-founder and equity holder in Capstan Therapeutics, Dispatch Biotherapeutics, and Bluewhale Bio; board member of AC Immune; scientific advisory roles with various companies, including BluesphereBio, Cabaletta, Carisma, Cartography, Cellares, Cellcarta, Celldex, Danaher, Decheng, ImmuneSensor, Kite, Poseida, Verismo, Viracta, and WIRB-Copernicus group. JLR received sponsored research to support these studies from Tmunity/Kite Pharma. JLR is a scientific founder of Tmunity Therapeutics which was acquired by Kite Pharma and he recently served as a paid Scientific Advisor. He also is a scientific founder BlueWhale Bio where he serves as an unpaid Scientific Advisor. He had (Tmunity) or has (BlueWhale) an equity interest in these companies. J.A.F.: Holds patents and intellectual property in T-cell-based cancer immunotherapy with royalties; receives funding from Tmunity Therapeutics and Danaher Corporation; consultancy with Retro Biosciences; scientific advisory board memberships with Cartography Bio, Shennon Biotechnologies Inc., CellFe Biotech, OverT Bio, Inc., and Tceleron Therapeutics, Inc. All other authors declare no competing interests.