PURPOSE Targeting nonspecific, tumor-associated antigens (TAA) with chimeric antigen receptors (CAR) requires specific attention to restrict possible detrimental on-target/off-tumor effects. A reduced affinity may direct CAR-engineered T (CAR-T) cells to tumor cells expressing high TAA levels while sparing low expressing normal tissues. However, decreasing the affinity of the CAR-target binding may compromise the overall antitumor effects. Here, we demonstrate the prime importance of the type of intracellular signaling on the function of low-affinity CAR-T cells. EXPERIMENTAL DESIGN We used a series of single-chain variable fragments (scFv) with five different affinities targeting the same epitope of the multiple myeloma-associated CD38 antigen. The scFvs were incorporated in three different CAR costimulation designs and we evaluated the antitumor functionality and off-tumor toxicity of the generated CAR-T cells in vitro and in vivo. RESULTS We show that the inferior cytotoxicity and cytokine secretion mediated by CD38 CARs of very low-affinity (Kd {\textless} 1.9 × 10-6 mol/L) bearing a 4-1BB intracellular domain can be significantly improved when a CD28 costimulatory domain is used. Additional 4-1BB signaling mediated by the coexpression of 4-1BBL provided the CD28-based CD38 CAR-T cells with superior proliferative capacity, preservation of a central memory phenotype, and significantly improved in vivo antitumor function, while preserving their ability to discriminate target antigen density. CONCLUSIONS A combinatorial costimulatory design allows the use of very low-affinity binding domains (Kd {\textless} 1 mumol/L) for the construction of safe but also optimally effective CAR-T cells. Thus, very-low-affinity scFvs empowered by selected costimulatory elements can enhance the clinical potential of TAA-targeting CARs.

Development of targeted cancer therapy requires a thorough understanding of mechanisms of tumorigenesis as well as mechanisms of action of therapeutics. This is challenging because by the time patients are diagnosed with cancer, early events of tumorigenesis have already taken place. Similarly, development of cancer immunotherapies is hampered by a lack of appropriate small animal models with autologous human tumor and immune system. In this article, we report the development of a mouse model of human acute myeloid leukemia (AML) with autologous immune system for studying early events of human leukemogenesis and testing the efficacy of immunotherapeutics. To develop such a model, human hematopoietic stem/progenitor cells (HSPC) are transduced with lentiviruses expressing a mutated form of nucleophosmin (NPM1), referred to as NPM1c. Following engraftment into immunodeficient mice, transduced HSPCs give rise to human myeloid leukemia, whereas untransduced HSPCs give rise to human immune cells in the same mice. The de novo AML, with CD123+ leukemic stem or initiating cells (LSC), resembles NPM1c+ AML from patients. Transcriptional analysis of LSC and leukemic cells confirms similarity of the de novo leukemia generated in mice with patient leukemia and suggests Myc as a co-operating factor in NPM1c-driven leukemogenesis. We show that a bispecific conjugate that binds both CD3 and CD123 eliminates CD123+ LSCs in a T cell-dependent manner both in vivo and in vitro. These results demonstrate the utility of the NPM1c+ AML model with an autologous immune system for studying early events of human leukemogenesis and for evaluating efficacy and mechanism of immunotherapeutics.

Myeloid leukemia in Down syndrome (ML-DS) clonally evolves from transient abnormal myelopoiesis (TAM), a preleukemic condition in DS newborns. To define mechanisms of leukemic transformation, we combined exome and targeted resequencing of 111 TAM and 141 ML-DS samples with functional analyses. TAM requires trisomy 21 and truncating mutations in GATA1; additional TAM variants are usually not pathogenic. By contrast, in ML-DS, clonal and subclonal variants are functionally required. We identified a recurrent and oncogenic hotspot gain-of-function mutation in myeloid cytokine receptor CSF2RB. By a multiplex CRISPR/Cas9 screen in an in vivo murine TAM model, we tested loss-of-function of 22 recurrently mutated ML-DS genes. Loss of 18 different genes produced leukemias that phenotypically, genetically, and transcriptionally mirrored ML-DS.