Structural Changes in Peptide-Scfv Bispecific Cars Impact T Cell Effector Function Against Acute Myeloid Leukemia

Background Immune escape is a significant roadblock for chimeric antigen receptor (CAR) T-cell therapies for leukemias. We have demonstrated that acute myeloid leukemia (AML) blasts express cell-surface GRP78, CD123, and B7H3. We aimed to design bispecific CARs that could bypass immune escape and evaluate them in preclinical models. Methods and Results Bispecific CARs relying on two signaling chain variants (scFvs) as antigen binding domains, often do not fold properly due to scFv/scFv interaction. We took advantage of a peptide that recognizes GRP78 and a CD123-specific scFv for our bispecific antigen binding domain and designed four bispecific CARs (78.123 CAR) with different linkers: short (G4S)3, long (mutIgG4), and 2 rigid (B2m, GPcPcPc) [G4S-, mutIgG4-, B2m-, GPc-CAR T cells]. All bispecific CAR T cell populations recognized and killed GRP78+/CD123+ tumor cells (KG1a, MOLM13) as evidenced by sustained tumor lysis and cytokine secretion compared to controls (p<0.01, N=4). GPc-CAR T cells did not recognize GRP78+/CD123- targets while the remaining constructs did. We evaluated the anti-AML activity of G4S-, mutIgG4-, and B2m-CAR T cells in vivo against MOLM13 (CD123+/GRP78+). Mono-specific CAR T cells served as control. A single CAR T cell infusion had potent anti-AML activity, resulting in significant survival advantage (p<0.001, N=5-15). Leukemia recurred in all mice treated with monospecific-CAR T cells and mutIgG4-CAR on average by day 35, while G4S- and B2m-CAR T cells induced complete responses in 7/15 and 6/15 treated mice, respectively. Extramedullary disease was the main relapse mechanism. Based on these studies, we selected the top performing bispecific CARs (G4S-CAR) for further evaluation. We studied mono- and 78.123 CAR T cells in GRP78+/CD123+ KG1a and KG1aCD123.KO xenograft models. In both models, there was a significant disease reduction and survival benefit in the 78.123-CAR T cell group compared to tumor only (p<0.01), demonstrating that (G4S)3-linker preserved antigen recognition of both GRP78 and CD123 in vivo. To evaluate the impact of antigen recognition domain structure on CAR function, we performed protein structure prediction using AlphaFold2 neural network. The mono specific CARs were notable for compact folding of the scFv in the CD123-CAR and a beta sheet configuration for the GRP78 peptide domain. Structural prediction of the bispecific constructs was congruent with our experimental findings. GPc-CAR's antigen recognition domain, unable to recognize GRP78 targets, formed a relaxed loop rendering the GRP78 domain inaccessible. The B2M-CAR and mutIgG4-CAR demonstrated altered configuration of the heavy and light chain of CD123 scFv disrupting the mutual alignment, potentially leading to decreased CD123 antigen binding or steric hindrance. Finally, G4S-CAR antigen recognition domain lost the beta sheet structure in the GRP78-specific domain, suggesting a less rigid binding domain, potentially more accessible for antigen binding. To test whether this bispecific approach could be extrapolated to other scFvs, we designed a CAR consisting of a GRP78-specific peptide and B7H3-scFv binding domain (78.B7H3), using a G4S linker between peptide-scFv. In this case we tested two different hinge/transmembrane domains (CD28 or CD8a H/TM). The CD28 hinge, like the 78.123 CAR was the only design that resulted in robust cytolytic activity when stimulated with either GRP78, B7H3, or both. When the 78.B7H3 CAR T cells were cocultured with THP-1 cells there was significant secretion of IFN and IL2 and target cell death in comparison to controls (p<0.01, N=4). To test the in vivo efficacy of the 78.123 and 78.B7H3 CAR T cells, mice were engrafted with 3x106 THP-1 cells expressing firefly luciferase. A single infusion of 3x106 GRP78-CAR, CD123-CAR, B7H3-CAR or bispecific 78.123- and 78.B7H3-CAR T cells had potent in vivo anti-AML activity (p<0.05, N=5/group). Conclusions We demonstrate feasibility of targeting two AML-antigens with a peptide-scFv bispecific CAR design. Bispecific CAR T cells prevent immune escape and have improved antitumor activity compared to their mono-specific counterparts when both antigens are expressed. This has been rarely observed for scFv-scFv bispecific CARs, in which normally one of the scFvs dominate. Thus, peptide-scFv CARs present a promising bispecific CAR design to prevent immune escape for a broad range of malignancies.