Introduction: Relapsed/refractory acute myeloid leukemia (AML) is a devastating disease with an urgent need for novel therapies. CD70 is a promising AML target antigen due to its high expression on AML tumor cells and absence on healthy hematopoietic stem and progenitor cells (HSPCs). One of the most potent CD70-targeting strategies is a “natural ligand” (NL) binder derived from the extracellular domain of CD27, the receptor for CD70. Our group has previously shown that NL binder CAR-Ts outperform scFv-based CD70 binders (Kasap et al, Blood, 2025) and that deep learning-based algorithms can generate NL variant CARs that outperform the wildtype (WT) CAR-T in preclinical contexts (Kang et al, ASGCT Annual Meeting, 2025). Here, we demonstrate that our lead NL variant CAR outperforms the WT CAR in both CAR-T and CAR-NK cell platforms in preclinical AML models. Furthermore, we characterize how this variant binder differs from the WT binder during the formation of the CAR-CD70 immune synapse, providing clues to the mechanism of its enhanced performance.

Methods: CAR constructs were delivered into primary T-cells using AAV and integrated at the TRAC locus via CRISPR knock-in. CAR constructs were delivered into primary NK-cells using a γ-retrovirus vector. CAR surface expression was measured by flow cytometry. In vitro cytotoxicity was measured by luminescence after co-culturing CAR-Ts or CAR-NKs with luciferase-expressing target cell lines for 18-24 hours. In vivo evaluation of CAR-Ts was performed in NSG mice, with tumor burden measured using bioluminescence imaging and CAR-T expansion by flow cytometry. Biochemical affinity of binder variants was measured using biolayer interferometry with immobilized CD70 trimeric ligand on a streptavidin biosensor. Fluorescent microscopy of CAR synapses was performed using fluorescent markers for tumor cells, CARs, actin microfilaments, and β-tubulin.

Results:

We expressed CD27-variant CARs in primary T-cells and NK cells and evaluated their cytotoxicity against the AML cell lines MOLM-14, NOMO-1, THP-1, MV411, and MOLM-13. We observed that all CD27 variant CAR-T and CAR-NKs demonstrated high surface expression and antigen specific cytotoxicity, and a subset of our CD27 variant CARs outperformed the WT CAR, particularly at lower effector-to-target ratios. In a “stress test” in vivo model using the MOLM-14 cell line, we observed our lead candidate variant NL CAR-T achieved complete tumor control, in contrast to the WT NL CAR-T, which only achieved temporary tumor reduction before relapsing. This improved survival for the NL variant CAR also correlated with observed greater peripheral blood expansion and persistence compared to the WT CAR.

We utilized biolayer interferometry to characterize the binding kinetics and affinity of our lead NL variant binder compared to WT CD27 and observed an overall 2-fold stronger affinity, driven primarily by a slower dissociation rate for the NL variant that may lead to improved ligand-receptor stabilization and signaling. This receptor complex stabilization was further visualized with polarization of actin and the microtubule organizing center (MTOC) at the CAR-T immune synapse using high resolution confocal microscopy after coculture with CD70-expressing tumor cells.

Conclusion:We report here the development and characterization of a deep-learning algorithm-identified mutation in a natural ligand-based CAR binder that displays improved binding to CD70 and improved anti-tumor efficacy in vitro and in vivo. These results highlight the potential for this methodology to improve natural ligand-based CAR-T and CAR-NKs and their potential for translation to the clinic for several CD70-expressing tumor indications. Future directions for this work include using proteomics and transcriptomics to probe differences in CAR signaling between our NL variant CAR and the WT CD27-based and testing these CARs in additional cancer contexts, including solid cancers such as renal cell carcinoma.

This content is only available as a PDF.
2025
Sign in via your Institution