Target Densities in Malignant and Normal Cells Determine CAR T Cell Efficacy and Off-Target Hematotoxicity

Abstract

Target antigen density has been established as a key factor for CAR activation and there are ongoing investigations into the number of surface molecules required for target cell lysis. A more precise understanding of differential target expression in malignant vs normal cells is needed to select the optimal CAR design, maintaining efficacy while limiting toxicity on normal cells. In AML, the lack of a leukemia-specific surface target antigen requires fine-tuned combinatorial approaches that enable distinctive target elimination based on differential antigen co-expression patterns in AML vs normal cells. Here, we present our target phenotypic profiling of primary AML and normal hematopoietic cells, and link it with CAR efficacy and changes in normal hematopoietic cell populations. We characterized the AML surface target phenotype across a cytogenetically and molecularly heterogeneous AML patient cohort (n>30) with relapsed/refractory disease. Using spectral flow cytometry on bone marrow and peripheral blood from AML patients and healthy control subjects, we quantified differential target expression of several AML targets, including CD33, CD123, CLEC12A and the novel target ADGRE2. We estimated the numbers of surface molecules per cell on AML bulk, LSC, normal HSPC and differentiated lineages. This allowed us to create two-dimensional maps of combinatorial target densities in AML vs normal. In this analysis, the combination of ADGRE2+CLEC12A offered the best therapeutic window. We developed a novel combinatorial CAR+CCR platform, ADCLEC.syn1, consisting of an ADGRE2-targeting 28z1XX-CAR and a CLEC12A-targeting chimeric costimulatory receptor (CCR) providing additional 4-1BB costimulation. This CAR+CCR configuration triggers killing of cells with high CAR target density alone, whereas cells with low CAR target density are only killed if the CCR target is co-expressed (IF-BETTER gate). Both in vitro and in vivo, we used MOLM13 AML cell line variants that we engineered to match a variety of different ADGRE2 and CLEC12A target density combinations found in AML or normal hematopoietic cells. The observed target density thresholds for ADCLEC.syn1-mediated cell lysis demonstrated potential for sparing vital normal hematopoietic cells. In addition, we evaluated ADCLEC.syn1 in several molecularly distinct AML PDX models with clinically representative target phenotypes, achieving long-term AML remissions in all studied PDX models. Next, we assessed the impact of ADCLEC.syn1 on normal hematopoietic cells. In vitro, we found that soluble factors from antigen-activated CAR T cells led to a reduction of CD34+CD38- HSPC, suggesting a potential role of pro-inflammatory cytokines leading to off-target hematotoxicity. To gain further insight, we established a humanized mouse model that allowed to serially monitor normal hematopoiesis by quantification of HSPC and several differentiated populations in the setting of controlled ADCLEC.syn1 or CD19 CAR activation based on an engineered CD19-positive AML cell line. We found that at time of AML remission, both ADCLEC.syn1 and CD19 CAR-treated mice had detectable CD34+ HSPC, indicating preferential killing of AML cells and relative sparing of HSPC. However, both ADCLEC.syn1 and CD19 CAR led to in vivo reduction of CD34+ HSPC as compared to untreated humanized controls, suggesting an indirect effect of inflammation upon CAR T cell activation. The observed HSPC loss was significantly reduced by blocking pro-inflammatory cytokines, further supporting off-target rather than on-target mechanisms of HSPC reduction. Overall, we profiled antigen densities of several CAR target combinations in AML and normal hematopoietic cells, which informed the development of the ADCLEC.syn1 CAR platform. Pre-clinical in vivo efficacy models based on AML with clinically representative target densities demonstrated high anti-leukemic efficacy of ADCLEC.syn1 and established killing thresholds with potential for sparing vital normal hematopoietic cells. Our in vitro and in vivo hematotoxicity models identified target-independent mechanisms of HSPC reduction in the context of both ADCLEC.syn1 and CD19 CAR activation, a finding that may be related to the off-target hematotoxicity observed clinically in some recipients of CD19 and other CAR T cell products.