Novel Natural Ligand Chimeric Antigen Receptors (CARs) Targeting CD70 with a Redesigned CD27 Binding Interface Provide Enhanced Potency Against Acute Myeloid Leukemia

Introduction: Chimeric antigen receptor (CAR) T cell therapies have brought tremendous progress in the treatment of B lymphoid and plasma cell malignancies, but development for acute myeloid leukemia (AML) has been difficult. CD70 is an attractive target in AML due to differential expression on malignant versus normal hematopoiesis, an important role in AML biology, and a good safety profile. CARs based on the natural ligand CD27 or antibody fragments including optimized hinges have previously been developed. Here, we utilized a rational approach integrating state-of-the-art computational protein design to generate a novel set of CD27-based CARs with optimized binding interface, and a unique combination of hinge, transmembrane (TM) and signaling endo-domains, resulting in novel CARs with enhanced potency against AML compared to other CD70 CARs (PMID:21304103, PMID:35452603). Methods: Four different CD27-based CAR assemblies were evaluated with variations in hinge (H), TM and endodomains and encoded into a third-generation self-inactivating lentiviral vector. Protein design calculations using physics and deep learning-based algorithms were applied to optimize the CD27-CD70 binding interface. Thirty-two variants with binding interface modulation were introduced in a parental CAR construct. Human T cells were activated, transduced and expanded with a good manufacturing practice-compatible protocol. Transduction efficiencies and T cell phenotypes were analyzed by flow cytometry. Cytotoxicity and T cell expansion were investigated in a sequential killing assay, co-culturing CAR-T cells with CD70+ MOLM-13.GFP-ffLuc AML cells. IFNγ and IL2 production was analyzed by Meso Scale Discovery. Off-target screening was performed by Retrogenix Cell Microarray Technology. In vivo anti-leukemic activity was assessed in NSG mice engrafted with MOLM-13.GFP-ffLuc cells by bioluminescent imaging (BLI) or with a CD70low patient-derived xenograft (PDX) by bleeding. Results: To circumvent natural cleavage of CD27, we evaluated the impact of different combinations of the CD27 ligand binding domain with CD8αH, CD8αTM or CD27TM and CD27 or 41BB with CD3ζ endodomains on CAR-T cell potency. Among the four CARs tested (CD27ζ (PMID:21304103), CD8αH-CD27ζ, CD8αH&TM-41BBζ (PMID:35452603), CD8αH&TM-CD27ζ), the new CD8αH-CD27ζ with CD27TM domain was most potent in the sequential killing assay (p<0.05, n=3 donors, CD8αH-CD27ζ vs CD8αH&TM-41BBζ or vs CD8αH&TM-CD27ζ), and cytokine secretion was significantly higher (IFNγ: p<0.001, IL2: p<0.001). The CD8αH-CD27ζ CAR outcompeted the other CARs with significantly improved leukemia control in vivo and enhanced overall survival of mice at limiting CAR-T cell doses (MOLM-13.GFP-ffLuc, leukemia growth BLI: p<0.01, survival: p<0.05, n=5 mice/group); PDX: complete leukemia clearance, n=4 mice/group). We computationally redesigned the CD27-CD70 binding interface to enhance potency of the CD8αH-CD27ζ CAR. Thirty-two affinity modulated variants consisting of single, double or triple point mutations in CD27 were selected and CAR-T cells produced. The 32 variant CAR-T cells were screened in the sequential killing assay and by cytokine secretion. Two specific amino acid positions within CD27 were of particular interest providing enhanced cytotoxicity and IL2 production. The top 5 constructs were retained for in vivo studies. In the MOLM-13.GFP-ffLuc NSG model, we identified two variants with significantly enhanced anti-leukemic activity compared to the parental CD8αH-CD27ζ CAR (p<0.05, n=4-8 mice/group). Leukemia re-challenge confirmed persistence of functional CAR-T cells in mice and their capacity to clear AML long term (BLI p<0.05 and p<0.001, for the 2 variants, respectively). Tissues of surviving mice showed AML clearance from blood, bone marrow and spleen, with persisting CAR-T cells. Off-target screening of the most potent CAR involved over 6500 human proteins and confirmed maintained specificity with no off-target hits by the lead CAR candidate. Conclusions: We have designed novel CARs targeting CD70 on AML and demonstrate enhanced potency compared to other CD70 CARs, both in vitro and in vivo in mice. Our lead candidate thereby maintains target antigen specificity. Integration of computational protein design in CAR development can help with the identification of more potent candidates for clinical translation.