Engineering advanced logic and distributed computing in human CAR immune cells

The study addresses how to engineer complex logic and distributed computation in human immune cells to improve specificity, safety, and control of cellular immunotherapies. Conventional CARs typically detect a single antigen and cannot integrate multiple inputs or coordinate across cell types, limiting discrimination between tumor and healthy tissues. Prior combinatorial CAR strategies (e.g., AND gates using separate CD3ζ and co-stimulatory signals) require carefully balanced, fixed designs with limited tunability. The authors build on the SUPRA CAR framework—comprising a universal receptor (zipCAR) and soluble antigen-binding adaptors (zipFvs) with orthogonal leucine zippers—to introduce tunable logic operations (AND, NOT) across multiple immune cell types, enable three-input logic within single cells, and implement distributed, multicellular circuits and communication channels. The goal is to program immune consortia with user-defined computation for applications from cancer to autoimmune disease.

The paper situates its work within CAR-T advancements and limitations. Conventional CAR-T therapies show efficacy in select cancers but face specificity issues due to shared antigen expression on healthy tissues. Combinatorial recognition (AND logic) has been explored using separate CD3ζ and co-stimulatory receptors, but performance depends on delicate balancing of signaling strengths and lacks post-engineering tunability (e.g., Kloss et al., Lanitis et al.). Inhibitory CARs (iCARs) using PD-1 or CTLA-4 have been reported for NOT-like logic (Fedorov et al.), yet broader co-inhibitory domains and cell-type compatibility remained underexplored. The SUPRA CAR (Cho et al., 2018) introduced a split, universal, programmable architecture enabling ON/OFF control, tuning, AND logic, and orthogonal control of T-cell subsets. The present work extends SUPRA to additional immune cell types, explores inhibitory domains (notably BTLA), and integrates logic within and across cells, addressing gaps in computation breadth, tunability, and multicellular coordination.

• SUPRA CAR design: zipCARs feature extracellular cognate leucine zippers fused to the human CD8α hinge, CD8α transmembrane, and intracellular signaling domains (co-stimulatory CD28 or 4-1BB; co-inhibitory PD-1, LAG3, TIM3, BTLA, CTLA-4; NK-activating 2B4, DAP12, NKG2D; CD3ζ). Expression under SFFV promoter; constructs include surface tags and fluorescent reporters or puromycin resistance for verification/selection.
• zipFv adaptors: scFvs (α-HER2, α-Axl, α-CD19, α-MESO, α-CD5) linked via 35-aa Gly/Ser linker to orthogonal leucine zippers (e.g., EE, SYN2, SYN5, SYN9). Cloned in pSecTag2A (CMV promoter, Igκ leader, C-terminal c-Myc and 6xHis). Produced by transient expression in Freestyle 293-F, Ni-NTA purification, dialysis to PBS, verified by SDS-PAGE and western blot, quantified by BCA.
• Cells: Primary human CD4+ and CD8+ T cells; naïve CD4+ differentiated to Th1/Th2; regulatory T cells (Tregs); γδ T cells; NK-92MI and primary NK cells; THP-1-derived macrophages. Target cells: engineered NALM6 leukemia cells and SKOV3-luc (Her2+). γδ T cells activated with zoledronic acid; NK cells isolated and expanded with feeders, IL-2, OKT3.
• Genetic delivery: Lentiviral transduction into immune cells (HEK293FT packaging with pHR, pMD2.G, pCMVR8.74, pAdv). Retronectin-mediated spinoculation. Puromycin selection where applicable. Dual-vector strategy for triple zipCAR expression (P2A-linked constructs plus separate vector with third zipCAR and puromycin resistance).
• Assays: Cytotoxicity via flow cytometry or luciferase readout; cytokine secretion (IFN-γ, IL-4) ELISAs; activation markers (CD69, CTLA-4); Treg suppression (CellTrace Violet dilution); macrophage phagocytosis (dual-positive gating) and polarization (HLA-DR, CCR7 for M1; CD206 for M2); intercellular communication using NFAT promoter to drive zipFv secretion in sender T cells and CD69 readout in receiver cells; in vivo xenograft model with NSG mice receiving CAR-NK-92MI and daily intraperitoneal zipFv dosing.
• Logic implementations: AND gates using separate zipCARs carrying CD3ζ and co-stimulatory (CD28 or 4-1BB) domains; NOT gates via co-inhibitory domains screening (BTLA selected); 3-input (A AND B) AND NOT C using three orthogonal zipCAR/zipFv pairs controlling CD3ζ, CD28, and BTLA; multicellular A AND NOT (B AND C) via Tconv activation and Treg AND suppression; kill switch via anti-V5-FITC antibody bridging V5-tagged zipCAR+ cells to α-FITC-CAR NK cells; intercellular AND via NFAT-driven zipFv secretion.
• Statistics: Student’s t-test (two-tailed) unless noted; p-values reported; experiments repeated at least twice with technical replicates; in vivo and γδ T-cell experiments had single biological replicate as noted.

• SUPRA CAR functions across seven immune cell types: CD8+ T cells (antigen-specific cytotoxicity), Th1 (IFN-γ secretion), Th2 (IL-4 secretion), Tregs (CD69 upregulation), γδ T cells (IFN-γ secretion), NK cells (cytolysis), and macrophages (zipFv-dependent phagocytosis).
• Orthogonal control of Th1/Th2 polarizes macrophages: α-Axl-EE activates Th1 (IFN-γ) driving M1 markers (HLA-DR, CCR7); α-Her2-SYN9 activates Th2 (IL-4) driving M2 marker CD206.
• Tunable AND logic:
  • In CD8+ T cells, combining FOS-CD3ζ and RR-CD28 (or 4-1BB) zipCARs yields synergistic target killing when both corresponding zipFvs are present; CD28-based AND gate exhibited broader operational concentration range than 4-1BB.
  • In Tregs, CD3ζ (SYN6) AND CD28 (SYN1) upregulated CTLA-4, demonstrating AND logic in suppressive cells.
• NOT logic via BTLA:
  • In CD4+ T cells, BTLA activation (via RR-BTLA zipCAR) significantly suppressed IFN-γ secretion triggered by FOS-CD28-CD3ζ (only in presence of the relevant antigen), establishing a NOT gate.
  • In NK-92MI cells, BTLA robustly reduced cytotoxicity driven by CD3ζ or 2B4 activation; at optimal zipFv doses, 2B4-based NOT gate suppressed killing by >40% upon BTLA activation.
  • In vivo xenograft (SKOV3-luc, Her2+): activating CD3ζ accelerated tumor killing by NK cells, while concurrent BTLA activation significantly increased tumor burden (reduced cytotoxicity), supporting safety-improving NOT behavior.
• Three-input logic in single cells: Primary CD8+ and CD4+ T cells expressing three orthogonal zipCARs (controlling CD3ζ, CD28, BTLA) achieved (A AND B) AND NOT C behavior. IFN-γ showed synergistic upregulation with dual activation (AND) and strong dose-dependent inhibition with BTLA (NOT). Reported significance in CD8+ T cells (e.g., p=0.028, p=0.036, p=0.037 for specific conditions).
• Distributed computing:
  • Treg-mediated suppression of Tconv: α-Her2-SYN9 induced ~40% proliferation of Tconv; activation of Tregs with α-Axl-EE significantly inhibited Tconv proliferation (intercellular NOT). An inducible A AND NOT (B AND C) logic was demonstrated where Treg AND gate (CD3ζ and CD28) strongly suppressed Tconv growth compared to single-domain activation.
  • Kill switch: Anti-V5-FITC antibody bridges V5-tagged zipCAR+ CD8+ T cells to α-FITC-CAR NK cells, specifically eliminating engineered CAR+ T cells and consequently reducing tumor cell killing, demonstrating an orthogonal, programmable safety switch.
  • Intercellular AND via secretion: NFAT-driven α-Axl-SYN2 zipFv secretion from sender CD4+ T cells upon α-Her2-SYN5 activation activated receiver zipCAR-expressing cells (CD69 upregulation). No activation occurred without the secretion module, evidencing a programmable communication channel.

The work demonstrates that a split, universal CAR architecture enables sophisticated, tunable logic in human immune cells at both single-cell and multicellular levels. By decoupling antigen recognition (zipFv) from signaling (zipCAR), signal strength can be modulated post-engineering via zipFv dosing, improving logic performance and specificity versus fixed CAR designs. The identification of BTLA as an effective co-inhibitory domain for NOT logic, particularly in NK cells, underscores the importance of cell-type-dependent pairing of activating and inhibitory domains. The 3-input logic circuits (AND combined with NOT) show that complex computation is achievable in primary human T cells. Multicellular circuits—including Treg-mediated suppression, a CAR+ cell-specific kill switch, and NFAT-driven zipFv secretion—illustrate distributed computation and synthetic communication within immune consortia. These advances address key challenges in CAR therapy—target specificity, safety, and control—and provide a platform for adaptable, multifunctional immunotherapies.

This study expands the SUPRA CAR platform to diverse immune cell types and introduces inhibitory signaling to achieve NOT and combined 3-input logic within single cells. It further establishes multicellular distributed computing through Treg-mediated suppression, an orthogonal kill switch, and a genetically encoded intercellular communication channel. Collectively, the results provide a foundation for programmable, multi-antigen, multi-cell-type immune consortia with user-defined functions, enhancing specificity and safety for cancer and beyond. Future work should characterize single-cell response heterogeneity, refine activating/inhibitory domain combinations for optimal NOT gates (especially in T cells), and explore additional orthogonal, inducible systems to expand the computational repertoire and therapeutic adaptability.

• Predominantly population-level measurements (bulk cytokines, cytotoxicity); single-cell functional heterogeneity and multiplex cytokine profiles were not deeply characterized.
• Logic performance depends on careful tuning of zipFv concentrations and signaling domain strengths; translating precise dosing in vivo may be challenging.
• Cell-type and signaling-domain compatibility is critical; BTLA NOT gate was more effective in NK cells than in CD8+ T cells for cytolysis suppression, indicating limited generalizability of specific designs.
• In vivo validation was limited to a xenograft model with NK-92MI cells and had limited biological replicates as noted; broader in vivo efficacy and safety across models and primary cells remain to be tested.
• Reliance on exogenous zipFv administration for control, though partially mitigated by the NFAT-driven secretion module, may complicate clinical delivery logistics.