CD2 costimulation bridges potent CAR-induced cytolysis and durable persistence

Abstract

Background

Current second-generation CAR T cell products rely on CD28 or 4-1BB costimulatory domains, additions that respectively favor rapid cytolysis or long-term persistence, but rarely both. Preclinical modeling and retrospective analysis have linked CD2–CD58 engagement to superior preclinical and clinical responses, yet the direct contribution of CD2 intracellular signaling remains undefined.

Methods

We replaced the costimulatory domain of anti-mesothelin (SS1) and anti-TnMUC1 (5E5) CARs with the human CD2 cytoplasmic tail (CD2z) and benchmarked them against 28z and BBz formats. Transient mRNA expression was used to profile proximal Ca²⁺ flux and degranulation free of tonic viral signals; durable functional assays employed lentiviral CARs. Cytokine release, genome-wide transcriptional programs, and anti-tumor activity were assessed in vitro and in NSG xenograft models.

Results

CD2z CAR T cells degranulated as efficiently as other z-containing CARs and generated a Ca²⁺ flux signal intermediate to 28z and BBz CARs. Lentiviral CD2z CARs released a Th1-skewed cytokine panel and matched 28z cytolysis despite a lower acute cytokine release. Transcriptomic analysis characterized CD2z cells in an early effector-memory state: glycolytic, mTORC1, and TNFa–NF-κB hallmarks were upregulated, whereas exhaustion-up signatures were selectively depleted vs 28z. In vivo, a single CD2z infusion induced deep and durable tumor regressions over the 60-day observation period in subcutaneous mesothelin-positive mesothelioma and orthotopic TnMUC1-positive pancreatic tumor models, achieving tumor control comparable to 28z and more rapid early tumor clearance than BBz, while supporting peripheral T cell persistence similar to BBz.

Conclusions

The CD2 cytoplasmic tail, in combination with CD3z, delivers balanced costimulation that couples brisk tumor debulking to T cell persistence. CD2z therefore may provide a simple, versatile alternative to canonical CD28 and 4-1BB modules for next-generation CAR T therapies.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• All commercially approved CAR T cell therapies use CD28 or 4-1BB costimulation, which uniquely contribute to the effector functions of the T cells, including enhanced potency (CD28) or persistence (4-1BB). Current CAR T cell therapies show limited durability in solid tumors due to intrinsic limitations after chronic antigen exposure, including exhaustion and poor persistence. CD2 is a T cell adhesion/costimulatory receptor with underexplored potential as a CAR endodomain.

WHAT THIS STUDY ADDS

• This study incorporated a CD2 costimulatory domain into CARs targeting two solid tumor antigens, mesothelin and TnMUC1. CD2-costimulated CAR T cells maintained cytotoxicity and cytokine production comparable to other CAR T cell products, but displayed reduced exhaustion and favorable memory-like states, and enhanced in vivo tumor control, which was associated with improved in vivo persistence.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• This study establishes the CD2 cytoplasmic domain as a signaling module that augments z-driven cytotoxicity and durable persistence in CAR T cells, shifting receptor design emphasis toward signaling quality and motivating head-to-head clinical evaluation of costimulatory domains.

Introduction

Chimeric antigen receptor (CAR) T-cell therapy hinges on pairing the CD3z activation motif with an intracellular costimulatory tail that sustains effector function without precipitating exhaustion.¹ Current products rely almost exclusively on CD28 or 4-1BB domains; CD28 drives rapid cytolysis but is prone to tonic signaling and early dysfunction, whereas 4-1BB confers persistence at the cost of slower tumor clearance. Discovering alternative costimulatory elements that balance immediate potency with long-term fitness therefore remains a key objective in CAR design.

The CD2–CD58 adhesion pair has emerged as a potent amplifier of CAR activity. Single-cell imaging of clinical CD19 CAR products showed that strong CD2–CD58 engagement accelerates synapse formation, enables serial killing, and correlates with durable remissions.² Conversely, genome-wide screens and relapse biopsies identified CD58 loss as a recurrent escape route in B-cell malignancies.^{3 4} Biophysical studies revealed why: conventional CARs fail to recruit accessory receptors, rendering them >100-fold fold less antigen-sensitive than native T-cell receptors; re-introducing CD2–CD58 contacts restores responsiveness and curbs exhaustion in vivo.⁵ CD58 ligation of endogenous CD2 markedly boosts IL-2 release from first-generation CAR T cells and remains an important cytokine driver even when a CD28 costimulatory domain is present,⁶ while enforced CD2 expression remodels the synapse and improves tumor control in murine models.⁷ Although these findings spotlight CD2, they intertwine the dual functions of CD2: enhanced adhesion and intracellular signaling, obscuring the specific contribution of the CD2 cytoplasmic domain. Transcriptomic profiling of chronically stimulated human CD8⁺ T cells further suggests that CD2 costimulation actively opposes exhaustion and synergizes with PD-1 blockade.⁸ Earlier still, Linette et al showed that cross-linking CD2 alone restored Ca²⁺ signaling in HIV-infected T cells whose CD3 pathway was non-responsive,⁹ implying that CD2-derived signals can override functional anergy.

Collectively, these observations call for a direct test of CD2 as an autonomous costimulatory module. Here, we graft the human CD2 cytoplasmic domain onto CARs and benchmark it against CD28z and 4-1BBz formats targeting two distinct tumor-associated antigens: mesothelin and the tumor-associated glycoform Tn-MUC1. Using both transient mRNA expression and lentiviral delivery, we profile proximal Ca²⁺ signals, cytokine output, transcriptomic state, and anti-tumor efficacy in stringent xenograft models. The data show that CD2 costimulation induces an early effector-memory phenotype with reduced exhaustion, rapid tumor debulking, and sustained peripheral persistence, positioning CD2 as a versatile alternative to canonical costimulatory domains.

Materials and methods

Cell lines

Jurkat E6-1, K562, SupT1, AsPC-1, and Hs766T cell lines were obtained from ATCC, authenticated by STR profiling (Idexx BioAnalytics), and confirmed mycoplasma-free with MycoAlert PLUS (Lonza). M108 mesothelioma cell line was derived from a malignant pleural effusion obtained from a mesothelioma patient.¹⁰ K562meso and luciferase-tagged derivatives of AsPC1, Hs766T, and M108 were generated by transduction with lentivirus encoding mesothelin¹⁰ and Click Beetle Green luciferase and GFP, respectively, and cell sorting.

CAR construct design

The coding sequences of all CARs used in this study are included in online supplemental table S1. Lentiviral CAR vectors encoding anti-mesothelin (SS1) and anti-Tn-MUC1 (5E5) were built on a human CD8a signal peptide, 45-aa CD8a hinge, CD8a transmembrane domain, and intracellular modules as indicated. For CD2z constructs, pTRPE-SS1-BBz and pTRPE-5E5-BBz backbones were digested with EcoRV and SalI, and a synthesized cassette (IDT) was ligated in-frame containing: CD8a transmembrane domain (human CD8a, nucleotides 544-618, GenBank BC025715), the 117aa intracellular domain of CD2 (human CD2, nucleotides 703-957, GenBank M14362), followed by the CD3z cytoplasmic domain fragment (human CD247, nucleotides 154-408, GenBank AF228312), a stop codon, and a terminal SalI site. An additional anti-TnMUC1 CAR with CD28 transmembrane and cytoplasmic domain (28z) was generated by swapping the SgrAi-SalI “28z” cassette from pTRPE-SS1-28z into the 5E5 vector (pTRPE-5E5-BBz backbone). All junctions were verified by Sanger sequencing. For in vitro transcription, CAR ORFs were subcloned into pGEM via XbaI/SalI, linearized, and used as templates for mRNA production as detailed below.

In vitro transcription and electroporation of mRNA

pGEM plasmids containing CAR constructs were linearized with SpeI. mRNA was transcribed, polyA-tailed, and capped using 1 mg of linearized plasmids and the T7 mScript Complete Standard mRNA Production System (CELLSCRIPT), according to manufacturer’s instructions. Purified mRNA (10 µg per 10⁷ T cells) was electroporated using the ECM830 Square Wave Electroporation System (BTX) in OptiMEM, and cells were rested overnight before further testing.

Lentiviral production and transduction

VSV-G pseudotyped, replication-incompetent, third-generation lentivirus was produced in HEK293T cells transfected with lentiviral CAR transfer plasmid, pTRP gag/pol, pTRP RSV-Rev, and pTRP VSVg using Lipofectamine 2000 (Invitrogen). Viral supernatants collected at 24 hours and 48 hours were filtered (0.45 µm) and ultracentrifuged (25 K rpm, 2.5 hours, 4°C). Functional titers were determined by titrated transduction of SupT1 cells and readout of expression by flow cytometry 72 hours after transduction.

Custom dynabead production

For T cell expansion experiments, Dynabeads M-450 Tosylactivated (ThermoFisher) were conjugated with OKT3 (Biolegend), anti-CD28 (clone 9.3, BioXCell), and anti-CD2 (clones T11.2 and T11.3, generously provided by Dr. Ellis Reinherz) antibodies at a total of 150 µg per 4×10⁸ beads according to the manufacturer’s instructions. For in vitro stimulation of CAR T cells, recombinant 150 mg of mesothelin-Fc fusion protein was coupled to 4×10⁸ Dynabeads M-450 Tosylactivated beads according to manufacturer’s instructions. In both methods, beads were washed once and resuspended in 1 mL of sterile Borate solution (0.1M Boric acid, pH 9.5). 150 µg of total protein in 1 mL of Borate solution was added and incubated overnight (16–24 hours) at 37°C with constant mixing. After magnet bead capture, the solution was decanted and the beads were washed three times with bead-wash solution (3% human albumin, 0.1% sodium azide, and 0.4% 0.5M EDTA in phosphate-buffered saline [PBS]) for 10 min each, and then another overnight wash in fresh bead-wash solution with continuous rocking. The coated beads were washed three times in R-10 medium (RPMI supplemented with 10% FCS, 100 U/mL penicillin, 100 µg/mL streptomycin sulfate, and 2 mM L-alanyl-L-glutamine) before use for in vitro stimulation. Bead-to-cell ratios were 1:1 for all protocols.

T cell culture, activation, and transduction

Cells were maintained at 1×10⁶ cells/mL in R-10 medium. For lentiviral transduction, T cells were activated with anti-human CD3/CD28 microbeads (Dynabeads, Gibco) at a 1:1 ratio in R-10 medium. 16 hours postactivation, lentivirus encoding for second generation CAR was added to the T cell culture at a MOI of 3. T cell media (R-10 supplemented with 30 IU/mL recombinant human IL-2) was doubled on day 3 of T cell activation. On day 5, beads were magnetically removed, T cells were counted and sized using a Multisizer 4e Coulter Counter (Beckman Coulter), and diluted to 0.75×10⁶ cells/mL with T cell media. T cells were counted and diluted to 0.75×10⁶ cells/mL with additional T cell media every 2 days. Expansion of T cells with custom Dynabeads was also debeaded, counted, and diluted to 0.75×10⁶ cells/mL according to the above protocol.

Flow cytometry

Surface CAR expression was determined through cytometric staining of T cells with biotinylated goat anti-mouse F(ab′)₂ (Jackson ImmunoResearch) and streptavidin-PE (BD Pharmigen). For the degranulation assay, 5 mL of CD107a-APC (BD Pharmigen) was added at the initiation of T cells and K562 meso co-culture, and GolgiStop Protein Inhibitor containing Monensin (BD Biosciences) was added after 1 hour. After 4 hours of co-culture, cells were washed, stained for viability with LIVE/DEAD Violet Viability/Vitality Kit (Invitrogen) and CD3-PE (Biolegend). For calcium flux assay, 1×10⁶ T cells were washed in Hanks’ Balanced Salt Solution (HBSS) containing 1 mM CaCl₂, 1 mM MgCl₂, and 1% heat-inactivated FBS, then loaded with 1 mM Indo-1 AM (Invitrogen) at 37°C for 30 min as previously described.¹¹ Cells were analyzed at 37°C on a BD LSRFortessa equipped with a 355 nm UV laser; Indo-1 emission was split using a 505 nm long-pass dichroic and detected with 395/10 nm (violet) and 510/20 nm (blue) band-pass filters. Indo-1 violet/blue ratios were calculated in real-time with dead cells excluded by FSC/SSC and low-violet fluorescence. After a 30 s baseline, cells were stimulated with 12 mg mesothelin-Fc and analyzed for an additional 700 s. Kinetic traces were analyzed in FlowJo V.10. For the proliferation assay, T cells were loaded with 5 mM carboxyfluorescein succinimidyl ester (CFSE) using the CellTrace CFSE Cell Proliferation Kit (Invitrogen) at 37°C for 5 min and washed with R10 medium. CFSE-labeled T cells were co-cultured with K562 meso cells at a 1:1 ratio for 5 days with media doubling on day 3. Cells were stained for viability and CD3-PE as above, and CSFE dilution was determined. Division index was determined using FlowJo v10. For transcriptional validation by flow cytometry, CAR T cells were co-cultured with antigen-positive tumor cells at an effector:target ratio of 1:1 and maintained for 6 days to match RNA-seq conditions. Cells were stained with antibodies to CD3, CD4, CD8, CD45RO, CCR7, PD-1, CTLA-4, LAG-3, TIM-3, and TOX. Memory subsets were defined as naïve (CD45RO⁻CCR7⁺), central memory (T_CM; CD45RO⁺CCR7⁺), and effector memory (T_EM; CD45RO⁺CCR7⁻). All assays except the calcium flux assay were evaluated on a LSRFortessa II, and all data were analyzed using FlowJo v10.

Functional assays

Cytokine assays: T cells were co-cultured with mesothelin-Fc-coated Dynabeads or K562meso, Hs766T, or Jurkat cells at a 1:1 ratio. Supernatants were collected 24 hours after co-culture and analyzed using the Cytokine 30-Plex Human Panel (ThermoFisher) on the Luminex system. Luciferase-based cytotoxicity: T cells were co-cultured with luciferase⁺ AsPC1, M108, K562meso, and Jurkat cells at a 1:1 ratio. Lysis of luciferase-expressing targets was quantified after 16 hours with the Luciferase Assay System (Promega). Flow-based cytotoxicity: K562 meso cells and T cells were co-cultured at the indicated effector-to-target (E:T) ratios for 16 hours. Cells were stained for viability and CD3-PE, and viable K562meso cells were quantified by flow cytometry to calculate specific lysis. Cytotoxicity after chronic antigen stimulation: For prolonged stimulation, CAR T cells were co-cultured with Jurkat targets at E:T=1:1 in R-10. Cultures were maintained for up to 2 weeks, with fresh targets added every 2–3 days (media refreshed at each addition). Viable lymphocytes were retained by gentle centrifugation (300×g, 5 min) and resuspension in fresh medium before adding targets. At baseline (no prior stimulation) and after 1 and 2 weeks of Jurkat co-culture, Hs766T targets (1×10⁴ cells/well) were seeded in E-Plate View 96 plates (Agilent) and allowed to adhere for 24 hours before addition of CAR T cells. Effector counts were measured and normalized using counting beads, then added at E:T=1:1. Impedance was recorded every 15 min for up to 120 hours on an RTCA eSight (Agilent). Percent specific lysis was calculated as: 100×(spontaneous lysis–sample lysis)/(maximum lysis–spontaneous lysis).

Xenograft mouse models

NSG mice were purchased from the Stem Cell and Xenograft Core at the University of Pennsylvania. The mice were housed under specific pathogen-free conditions in microisolator cages and given ad libitum access to autoclaved food and acidified water. Tumor models were established by injecting 1×10⁶ luciferase⁺ M108 cells subcutaneously on the right flank or 5×10⁵ luciferase⁺ Hs766T cells intraperitoneally. Tumor measurements were determined through serial bioluminescence imaging of tumor burden after intraperitoneal injection of 3 mg luciferin (Gold Biotechnology) and measurement on a Xenogen IVIS-200 Spectrum camera. Mice were randomized into treatment groups and injected with 5×10⁶ T cells, normalized to 40% CAR expression, in 100 mL PBS. Peripheral blood was obtained from retro-orbital bleeding on indicated days after T cell injection and stained for the presence of human CD45⁺, CD4⁺, and CD8⁺ T cells. After gating on the human CD45⁺ population, the CD4⁺ and CD8⁺ subsets were quantified using TruCount tubes (BD Biosciences).

Differential expression analysis

CAR T cell products were activated by mesothelin-Fc-coated Dynabeads at a 1:1 ratio. 6 days after activation, 1×10⁶ cells were frozen as dry pellets and submitted to BGI Genomics for RNA extraction, library preparation, and sequencing; pellets of rested CD2z CAR T cells were frozen in the absence of bead activation. Read processing was completed using UseGalaxy.org.¹² Reads were cleaned by removal of low-quality bases from 5’ and 3’ ends and Illumina adaptors, trimmed N’s, imposed a minimum length of 30 bases using Cutadapt 5.1. Cleaned reads were mapped to the human reference genome, GRCh38, using RNA STAR 2.7.11b, and counts were determined using featurecounts 2.1.1. RNA-seq transcript count data were normalized using DESeq2 2.11.40.8 and used to calculate differential expression between activated and resting CD2z CAR T cells as well as between activated CD2 z CARs and 28z or BBz CARs 6 days after activation. Genes were identified as significantly differentially expressed if their adjusted p value was less than 0.05 and their absolute log₂ fold change was greater than 1 (ie, non–log₂-transformed fold change greater than 2). Gene set enrichment analysis (GSEA) of hallmark and immunology-related gene sets from MSigDB H and GSE9650¹³ within C7 was performed using the GenePattern 2.0¹⁴ GSEAPreranked 7.4.0 module with 3 log₂ fold change–ranked lists representing the expression patterns of the three comparisons listed above. Normalized enrichment scores (NES) and false discovery rate (FDR) q-values were reported; plots in online supplemental figure S4D–F were generated using ggplot2 and display the top 10 positively and negatively enriched Hallmark terms for each comparison.

Statistics

Data were analyzed with Prism V.10 (GraphPad). Group means were compared by one-way or two-way analysis of variance with Tukey’s multiple comparisons test. A p< 0.05 was considered significant.

Results

CD2 costimulation enhances ex vivo T cell expansion

Early evidence demonstrated that CD2 engagement can bypass defective CD3 signaling in anergic HIV-1-infected CD4⁺ T cells by restoring proximal Ca²⁺ flux.⁹ Motivated by this, we evaluated whether adding CD2 ligation to bead-based stimulation could enhance ex vivo expansion and potentially overcome tumor-induced anergy. CD4⁺ T cells cultured for 8 days and CD8⁺ T cells cultured for 11–12 days underwent ~2.5–5 and ~4–6 population doublings, respectively, across three donors when stimulated with beads bearing anti-CD3 together with anti-CD28, anti-CD2, or both (figure 1A); expansion was comparable across these costimulation conditions, with no significant differences in population doublings within either subset. In contrast, beads lacking costimulation (anti-CD3 only) or TCR engagement (anti-CD2 only, or anti-CD28+anti-CD2) failed to drive division and produced only an increase in cell size (online supplemental figure S1), confirming that CD2 delivers a potent costimulatory signal yet cannot initiate mitogenesis on its own.

Figure 1. CD2 provides potent costimulation during T cell expansion and functions within anti-mesothelin SS1 CAR. (A) Cumulative population doublings of CD4⁺ T cells (day 8) and CD8⁺ T cells (day s11–12) after stimulation with Dynabeads (1:1 bead:cell) bearing anti-CD3 with the indicated co-stimulators (CD28 and/or CD2). Replicates: n= 3 biological donors, 2 technical replicates per condition averaged within donor; symbols are donors; bars=mean ±SD across donors. Statistics: repeated-measures two-way ANOVA with Tukey’s multiple comparisons test (within subset across bead conditions). Within each subset, expansion was comparable across these costimulation conditions (ns). Conditions lacking costimulation or TCR engagement (anti-CD3 only, anti-CD2 only, or anti-CD28 and anti-CD2 without anti-CD3) are shown in online supplemental figure S1. (B) Schematic of SS1 CAR variants sharing a human CD8a hinge and transmembrane region, followed by intracellular z-only (SS1z), 28z (SS1-28z), BBz (SS1-BBz), CD2z (SS1-CD2z) or signaling-deficient (SS1-Δz) tails. (C) Surface CAR expression 24 hours after electroporation (10 µg capped/tailed mRNA per 1×10⁷ T cells); CAR⁺ cells detected with biotin-anti-mouse F(ab′)₂ and streptavidin-PE. Replicates: n=2 biological donors, 2 technical replicates per condition. Symbols show individual wells; bars represent mean±SD across donors. Given the limited number of donors (n=2), data are presented descriptively without formal hypothesis testing. (D) Degranulation after 4 hours co-culture with mesothelin⁺ K562 (K562meso) targets at E:T=1:1 in the presence of CD107a-APC and monensin; readout is %CD3⁺ CD107a⁺. Replicates: n=2 biological donors, 2 technical replicates per condition. Symbols show individual wells; bars represent mean±SD across donors. Given the limited number of donors (n=2), data are presented descriptively without formal hypothesis testing. (E) Indo-1 ratiometric Ca²⁺ traces of mRNA-electroporated CAR T cells stimulated with soluble mesothelin-Fc added at 30 s at 37 °C (BD LSRFortessa); representative of n=3 independent experiments with distinct donors. ANOVA, analysis of variance; E:T, effector-to-target; MFI, mean fluorescence intensity.

CD2 costimulation augments anti-mesothelin CAR signaling and effector function

We next replaced the 4-1BB module in an anti-mesothelin (SS1) CAR with the CD2 cytoplasmic tail (SS1-CD2z; figure 1B) and compared it with z-only, CD28z, 4-1BBz, and signaling-deficient variants (Dz) previously characterized (online supplemental table S1).¹⁰ All constructs were expressed at similar frequencies and expression levels after mRNA electroporation (figure 1C), which obviates the CD3/CD28 preactivation required for viral transduction and cleanly isolates CAR-intrinsic signaling.¹⁵ Because mRNA expression is rapid, tunable, and short-lived (< 72 hours), it offers a side-by-side read-out of proximal signaling free from the tonic activity or clonal bias seen with viral systems; however, durable functional studies that follow this section use lentivirally transduced CAR T cells. Only z-containing CARs degranulated on exposure to mesothelin-expressing K562 cells (figure 1D). CD2z drove increased antigen-induced proliferation and a trend toward higher cytotoxicity versus z-only (online supplemental figure S2A–C), establishing CD2 as a viable alternative costimulatory motif for CAR design. on engagement with recombinant mesothelin-Fc, CD28z elicited the largest and most sustained Ca²⁺ flux, 4-1BBz showed little-to-no flux above baseline, and CD2z produced an intermediate peak with faster decay than CD28z (figure 1E). Taken together, these data indicated that CD2 costimulation enhances antigen-dependent effector functions of CAR T cells and prompts deeper cytokine and transcriptomic profiling to define its effector program.

Lentivirally transduced CAR T cells from three independent donors were stimulated with mesothelin-coated beads, and supernatants were analyzed 24 hours later (figure 2). CD4⁺ T cells expressing SS1-CD2z released a broad Th1-skewed cytokine panel, including IFNg, IL-2, and TNFa, at levels consistently higher than SS1-BBz and approaching those of the potent SS1-28z construct (figure 2A). In CD8⁺ T cells (figure 2B), CD2z-driven cytokine output more closely resembled BBz than 28z, indicating a subset-specific polyfunctionality. Functionally, SS1-CD2z cells mediated efficient lysis of mesothelin-positive AsPC-1 (pancreatic cancer), M108 (mesothelioma), and K562meso targets, trending toward superior killing relative to SS1-BBz and matching SS1-28z at a 1:1 E:T ratio (figure 2C).

Figure 2. CD2z CAR T cells couple broad cytokine output to potent cytotoxicity. (A, B) Multiplex cytokine analysis (Luminex) of 24 hours supernatants from lentiviral SS1 CAR T cells expanded for 10 d and restimulated with mesothelin-coated beads at E:T=1:1, shown separately for the CD4⁺ (A) and CD8⁺ (B) T cells. Replicates: n = 3 biological donors, 2 technical replicates per condition averaged within donor; bars show mean±SD across donors. Statistics: Kruskal-Wallis with Dunn’s multiple comparisons across CAR variants within each subset. (C) Cytotoxicity against AsPC1 (PDAC), M108 (mesothelioma), and K562meso targets at E:T=1:1, reported as % specific lysis and analyzed within each target cell type. Replicates: AsPC-1 and M108, n=3 biological donors; K562meso, n=2 biological donors; 2 technical replicates per condition, averaged within donor. Statistics: ordinary one-way ANOVA with Tukey’s multiple comparisons test (within target). ANOVA, analysis of variance; E:T, effector-to-target; GM-CSF, granulocyte-macrophage colony-stimulating factor; NTD, non-transduced.

CD2 costimulation programs an early effector-memory state with reduced exhaustion versus CD28 and 4-1BB

RNA-seq showed that CD2-costimulated CAR T cells activate a broad effector and metabolic program while dampening exhaustion (figure 3). Compared with resting cells, antigen-activated CD2z T cells upregulated cytotoxic mediators (GZMA, GZMK, GNLY) and glycolytic regulators (PGK1, PFKFB4), while classic costimulatory transcripts (ICOS, CD28) were downregulated (figure 3A). Consistent with a reduced exhaustion profile compared with 28z, CTLA4 was ~threefold fold lower in CD2z cells, and TOX2, a gene associated with chronic dysfunction, was 50% lower, whereas TOX and LAG3 were unchanged (figure 3B). Relative to BBz, CD2z cells were enriched for effector/early-memory programs with only modest elevation of exhaustion modules (figure 3C).

Figure 3. Transcriptomic impact of CD2 costimulation. (A) Volcano plot (DESeq2) of differentially expressed genes comparing CD2ζ CAR T cells activated 6 days vs resting (day 0). (B, C) Volcano plots comparing CD2ζ to 28ζ (B) or BBζ (C) CAR T cells on day 6. RNA-seq replicates: n=2 independent donors for each CAR variant. (D) Normalized enrichment scores (NES) for top Hallmark pathways (GSEAPreranked 7.4.0) in CD2ζ versus resting. (E, F) GSE9650 Immunologic Signature enrichment for CD2ζ versus 28ζ (E) and CD2ζ versus BBζ (F); bar length indicates NES, bubble size indicates −log₁₀(false discovery rate [FDR]), and filled circles denote FDR<0.05. (G) Phenotype of CAR⁺ CD8⁺ T cells after 6 days stimulation with K562meso targets, showing memory subsets by CCR7/CD45RO (Naïve, CM, EM, Effector). Replicates: n=2 biological donors, 2 technical replicates per condition. Symbols show individual wells; bars represent mean±SD across donors. Because of the limited number of donors (n=2), these data are presented descriptively without formal hypothesis testing. (H) Exhaustion score for CAR⁺ CD8⁺ T cells calculated as the percentage of cells co-expressing 0–5 markers among CTLA-4, PD-1, LAG-3, TIM-3, and TOX after 6 days stimulation with K562meso. Replicates: n=2 biological donors, 2 technical replicates averaged within donor; mean±SD across donors. Points indicate individual wells; symbols denote donors. Given the limited number of donors (n=2), we present descriptive statistics (mean±SD) without formal hypothesis testing. CM, central memory; EM, effector memory.

Hallmark gene-set enrichment of the activated-versus-resting contrast for CD2z revealed positive NES across Glycolysis, Hypoxia, mTORC1, Cholesterol Homeostasis, and TNFa–NF-kB pathways, which are metabolic and inflammatory circuits needed for durable effector function (figure 3D). Compared with 28z, CD2z significantly depleted the GSE9650 “exhausted-vs-memory CD8 T-cell UP” module¹³ (NES=−1.36, FDR=0.049), while memory/effector-skewed sets trended upward (figure 3E). Against BBz, CD2z showed strong enrichment of effector/early-memory signatures (NES > +2) with only modest increases in exhaustion modules (figure 3F). Collectively, these data place CD2z CAR T cells in an early effector-memory state that couples robust cytotoxic potential and metabolic fitness with lower exhaustion than 28z and greater activation than BBz.

To orthogonally validate these transcriptomic programs, we performed a 6-day stimulation assay followed by multiparameter flow cytometry. CD2z CAR T cells reproducibly exhibited a shift toward early-differentiated subsets (higher naïve frequencies) with lower expression of exhaustion markers (CTLA-4, PD-1, LAG-3, TIM-3, TOX) relative to 28z and BBz, mirroring the RNA-seq signatures (figure 3G,H).

Anti-mesothelin-CD2z CAR T cells control mesothelioma in vivo with rapid tumor regression and sustained peripheral expansion

Consistent with the in vitro signaling hierarchy, CD2z CAR T cells controlled established M108 mesothelioma xenografts over the 60-day observation window with tumor burden reductions comparable to the benchmark 28z construct and greater than BBz. 60 days after subcutaneous inoculation of 1×10⁶ luciferase-tagged M108 cells, NSG mice received a single dose of the indicated CAR T cells (figure 4A, left). Longitudinal bioluminescence imaging showed rapid tumor regression in the 28z and CD2z cohorts, whereas BBz, non-transduced (NTD) T cells, and PBS controls exhibited only transient stabilization or continued growth. At study end, log cell-kill values calculated for each animal (figure 4A, right) confirmed that CD2z achieved a median >3 log reduction in viable tumor burden, statistically superior to BBz, NTD, and vehicle (p=0.0195 and 0.0034, respectively) and indistinguishable from 28z (p>0.5). Representative images (figure 4B) illustrate marked reductions in bioluminescence signal in most CD2z- and 28z-treated mice over the 60-day study period, contrasted with persistent signal in the BBz group. Individual tumor-growth trajectories (online supplemental figure S3) mirrored the group means: all CD2z-treated mice experienced ≥2 log reductions in bioluminescence, and 6/8 mice reached background signal during the observation window, consistent with profound tumor debulking but not excluding the possibility of late relapse beyond 60 days.

Figure 4. CD2z CAR T cells mediate deep tumor regressions and sustained expansion in a mesothelioma xenograft model. (A) Left, longitudinal bioluminescence imaging (BLI; total flux, photons/second (p/s)) of subcutaneous M108-luc tumors in NSG mice following a single intravenous infusion of 5×10⁶ total CAR T cells (1:1 CD4:CD8) on day 0; groups: phosphate-buffered saline (PBS), non-transduced T cells (NTD), SS1-28z, SS1-BBz, SS1-CD2z. Curves show mean ±SD. Right, per-mouse log cell-kill at day 62 relative to baseline (day −1), computed as log₁₀(endpoint/baseline). Replicates: n=8 mice per group. Statistics (right): ordinary one-way ANOVA with Tukey multiple comparisons test. (B) Representative BLI images at days −1, 20, 40, 47 and 62 for each treatment group. (C) Absolute counts of human CD4⁺ and CD8⁺ T cells in peripheral blood (CD45⁺ gate) on day 21 measured by flow cytometry with TruCount beads; mean ±SD. Replicates: n=8 mice per group. Statistics: ordinary one-way ANOVA with Tukey post hoc test. ANOVA, analysis of variance.

Interestingly, peripheral blood monitoring on day 21 revealed higher frequencies of both CD4⁺ and CD8⁺ T cells in mice given CD2z or BBz products, whereas 28z and NTD cells showed reduced T cell persistence (figure 4C). This expansion without exhaustion mirrors the transcriptional data, suggesting that CD2 costimulation supports in vivo persistence comparable to 4-1BB yet retains the potent antitumor functionality of CD28. Collectively, these in vivo results position CD2z CARs as a favorable alternative to canonical BBz or 28z designs and combine strong tumor debulking with sustained peripheral expansion.

Anti-TnMUC1-CD2z CAR T cells exhibit enhanced tumor regression and persistence in a pancreatic cancer model

To extend CD2 costimulation to another clinically relevant antigen, we retargeted the platform to the tumor-associated glycoform Tn-MUC1. The 5E5 scFv-based CAR was cloned into two new lentiviral designs: 28z (5E5-28z) and CD2z (5E5-CD2z), and benchmarked against the previously characterized 5E5-BBz CAR (online supplemental table S1).¹⁶ All three CARs transduced normal-donor T cells at similar frequencies, although 5E5CD2z reproducibly displayed a lower mean-fluorescence intensity, indicating reduced surface density (figure 5A). When cocultured 1:1 with Tn-MUC1-positive Hs766T pancreatic cancer cells, CD2z CAR T cells produced IFNg, IL-2, TNFa, IL-6, and granulocyte-macrophage colony-stimulating factor at significantly lower levels than their 28z or BBz counterparts (online supplemental figure S4A). The same pattern held against Jurkat leukemia cells (online supplemental figure S4B), showing that, across tumor types, CD2 costimulation supports TnMUC1-specific responses, but does so with a more restrained acute cytokine burst. In overnight cytotoxicity assays against antigen-positive Jurkat cells (E:T=1:1), all three 5E5 CAR formats mediated robust lysis, with each outperforming NTD controls (p=0.0003 for each comparison); we did not detect significant differences among CD2z, 28z, and BBz in this setting (figure 5B). We also evaluated the cytotoxicity of 5E5 CAR T cells against Hs766T targets at baseline (week 0) and after 1 or 2 weeks of chronic stimulation with Jurkat targets. While all constructs exhibited the expected attenuation by week 2, inter-construct differences at the indicated timepoints were not significant despite the lower surface density and restrained cytokine burst (all p>0.05; figure 5C). This pattern supports the interpretation that CD2 costimulation programs for persistence rather than increased immediate killing, which underlies the enhanced control seen in vivo.

Figure 5. CD2z enhances the anti-tumor activity and persistence of TnMUC1-targeted CAR T cells. (A) Flow cytometric surface staining of lentivirally transduced T cells expressing 5E5-28z, 5E5-BBz, or 5E5-CD2z CARs at 10 days post-transduction; representative plots shown. (B) Cytotoxicity against Jurkat targets at E:T=1:1. Replicates: n=2 biological donors, 2 technical replicates per condition averaged within donor. Statistics: ordinary one-way ANOVA with Tukey’s multiple comparisons test across CAR variants. (C) Cytotoxicity against Hs766T targets at baseline (week 0) and after 1 or 2 weeks of chronic stimulation with Jurkat cells (Jurkat added every 2–3 days), E:T=1:1. Values are normalized to NTD week 0. Replicates: n=2 biological donors, 2 technical replicates per condition averaged within donor. Statistics: repeated measures two-way ANOVA with Tukey’s multiple comparisons test (factors: CAR variant and time). (D) Left, longitudinal BLI (total flux, photons/second (p/s)) of orthotopic Hs766T-luc pancreatic tumors in NSG mice treated with a single intravenous infusion of 5×10⁶ CAR T cells on day 28 postimplant; mean ±SD over time by group. Right, per-mouse log cell-kill at day 56 relative to baseline (day −1), computed as log₁₀(endpoint/baseline). Replicates: n=8 mice per group. Statistics (right): ordinary one-way ANOVA with Tukey’s test. (E) Representative BLI images at days −1, 20, 34, and 56 for each group. (F) Absolute counts of circulating human CD4⁺ and CD8⁺ T cells in peripheral blood (CD45⁺ gate) on day 42 after infusion measured by TruCount beads; mean ± SD. Replicates: n=8 mice per group. Statistics: ordinary one-way ANOVA with Tukey’ s post-hoc test. ANOVA, analysis of variance; E:T, effector-to-target; NTD, non-transduced.

Despite the more restrained cytokine release observed in vitro, the in vivo performance of TnMUC1-directed CD2z CAR T cells echoed the mesothelin CAR results. NSG mice bearing orthotopic, luciferase-expressing Hs766T tumors (engrafted for 28 days) received a single intravenous infusion of each CAR product. Longitudinal imaging showed the most rapid early tumor regression and lowest median bioluminescence signal by day 56 in the CD2z cohort, whereas tumors in the 28z and BBz groups either regressed more slowly or plateaued (figure 5D, left). Per-mouse log-cell-kill analysis at day 56 indicated greater tumor burden reduction with CD2z confirmed with BBz (p<0.001) and 28z (p<0.05) in this model (figure 5D, right). Representative images illustrate marked reductions in bioluminescence signal in most CD2z-treated animals over the 56-day observation period, contrasted with residual signal in the other cohorts (figure 5E), and individual trajectories in online supplemental figure S5 show sustained regression during the study window in every CD2z-treated mouse. By study end, bioluminescence fell below the background threshold in 2/8 CD2ζ-treated mice and 0/8 in both CD28ζ and 4-1BBζ cohorts; however, later relapse beyond 56 days cannot be excluded.

Peripheral blood profiling on day 21 revealed a marked expansion of CD4⁺ T cells and a trend toward higher CD8⁺ counts in CD2z-treated mice relative to 28z and BBz recipients (figure 5F), mirroring the persistence observed in the mesothelin model. Thus, even with a lower acute cytokine output, CD2 costimulation drives superior tumor debulking and in vivo expansion, reinforcing CD2 as an optimal costimulatory module across distinct antigens.

Discussion

This study establishes the CD2 cytoplasmic domain as a costimulatory module that augments z-driven cytotoxicity and durable persistence in CAR T cells. Replacing CD28z or 4-1BBz with CD2z, in otherwise identical anti-mesothelin and anti-TnMUC1 CARs, preserved ex-vivo expansion, triggered an intermediate Ca²⁺ burst sufficient for degranulation, generated a Th1-skewed cytokine profile and low-exhaustion transcriptome, and controlled large, established tumors over a 56–60 day observation window as effectively as CD28z while outperforming BBz in these models. These advantages endured even though peak cytokine release was lower than with 28z, showing that maximal early secretion is not required for long-term control.

Proximal signaling placed CD2z between 28z, whose tonic activity predisposes to exhaustion,¹ and BBz, which favors persistence at the cost of slower cytolysis. Notably, mutation of a Grb2-interacting residue within the CD28 costimulatory domain enhances CAR T cell persistence, extends to durable tumor control, and is associated with reduced Ca²⁺ flux,¹⁷ suggesting that this signaling interaction contributes to reduced persistence. Transcriptomics confirmed activation of glycolytic, mTORC1, and TNFa–NF-κB gene sets with selective depletion of exhaustion-up signatures, mirroring evidence that CD2 costimulation steers chronically stimulated human CD8⁺ T cells toward a reversible, non-exhausted state.⁸ That bias accords with clinical data showing that CAR T-cell products enriched for early-memory T cells expand more vigorously and persist long-term in complete responders.¹⁸

Unlike prior efforts that increased CD2–CD58 adhesion, we left CD58 unaltered, disentangling intracellular CD2 signals from accessory binding. Strong CD2–CD58 engagement correlates with durable remissions,² whereas tumors frequently escape by losing CD58.^{3 4} Restoring the interaction can raise antigen sensitivity >100-fold fold⁵ and enforcing CD2 expression remodels the synapse and improves tumor control.⁷ Our data show that the 117-aa CD2 intracellular tail, in combination with CD3z, suffices to improve CAR T cell functionality, with slight contradiction to the finding that CD2 cross-linking reinstated Ca²⁺ signaling in CD3-defective, anergic T cells⁹; in this case, CD3z signaling was required for activity. CD2z also improved control of TnMUC1-positive tumors and increased T cell persistence despite lower surface CAR density, suggesting that CD2 signaling can compensate for modest receptor expression.

The limitations of this study include reliance on NSG xenografts, which lack myeloid and stromal compartments that modulate CD58, and the small donor pool. In addition, our in vivo studies were conducted over a 56–60-day period and were not designed to capture very late relapse. Although CD2z CAR T cells mediated profound tumor burden reductions and frequent transitions to background-level bioluminescence within this observation window, these data do not exclude the possibility of late relapse or differential long-term efficacy compared with 28z and BBz CARs. Extended follow-up in additional models will be required to rigorously define long-term relapse kinetics and durability. Future studies in humanized or syngeneic models will refine our understanding of how CD2 signaling contributes to interactions with the tumor microenvironment. Additionally, we did not systematically map antigen-density thresholds for activation and killing; future studies using calibrated surface-site quantification or tunable antigen-expression systems will be valuable to define how antigen abundance modulates the activity of CD2z costimulated CAR T cells.

Taken together, the data presented here demonstrate that CD2-derived signals can endow CAR T cells with rapid cytotoxicity, in vivo persistence, and reduced exhaustion, traits that may translate into improved tumor control relative to canonical CD28 and 4-1BB domains, a hypothesis that will require head-to-head testing in extended preclinical and clinical studies.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.