CAR-Machine: CAR-T Cell Target Identification Platform for Pediatric Acute Leukemia and Implementation to Identify Novel Targets for Pediatric AML

Pediatric acute leukemias are heterogenous diseases that represent over 25% of new childhood cancer cases. While survival rates have increased rapidly over the past decades, relapsed/refractory (R/R) disease is associated with poor survival. R/R leukemias tend to be chemo-refractory and thus, alternative therapies such as immunotherapies are needed to improve outcome. Compared to adult cancers, pediatric leukemias have a lower mutational burden, meaning immune checkpoint inhibitors are not as effective. Chimeric antigen receptor (CAR) T-cell therapy, in which T-cells are engineered to target specific surface antigens on tumor cells, provides an innovative immunotherapeutic approach to targeting leukemias. CD19 targeting CAR-T therapy has been particularly successful in B-cell acute lymphoblastic leukemia (B-ALL), achieving an over 90% complete remission rate, and while CD19-directed CAR-T therapy leads to concurrent B-cell aplasia, this can be overcome with immunoglobulin infusions. In contrast, CAR-T target identification in other pediatric leukemias - acute myeloid leukemia (AML), t-cell lymphoid leukemia (T-ALL), and mixed phenotype acute leukemia (MPAL) - are more difficult due to shared expression of leukemia surface markers in healthy T-cell, myeloid, and/or hematopoietic stem cells. We developed CAR-Machine, a platform that performs systematic analysis of single cell RNA sequencing data to identify CAR-T target genes and pairs based on (1) over-expression in tumor cells, (2) coverage across tumor cells, (3) surface expression, and (4) restricted healthy vital tissue expression. The platform calculates a CAR score for each target based on the four parameters listed above to identify highly specific targets with low off-tumor toxicity. The tool allows for identification of traditional single-gene based CAR targets as well as advanced CAR-pairs based on OR-, AND-, and NOT-gated CAR strategies. To assess off-tumor toxicity, we compiled a large healthy tissue single-cell atlas that contains expression data for over 500,000 single cells from 10 different vital tissues including brain, heart, kidney, colon, lung, lymph node, skin, esophagus, liver, and bone marrow, with cell types ranging from hematopoietic stem cells to neurons. To identify CAR targets that will benefit patients of specific or multiple cytogenetic and subtype classifications, we assess coverage in TARGET bulk RNA-sequencing datasets. We utilized this platform to identify potential targets for pediatric AML using a dataset generated in our lab (Mumme et al, Nature Communications 2023) that contains almost 40,000 cells from 14 patient diagnosis samples and 4 young adult healthy donor samples as controls. The platform identified 12 single gene targets with high expression in malignant blast cells across AML patient cells with different cytogenetics alterations, low expression on non-myeloid hematopoietic cells, and minimal expression in cell types from vital tissues. The 12 optimal targets include 4 genes that have already been utilized in AML CAR-T therapies (including PRAME and CD64) while the remaining 8 are novel CAR-T therapy targets. The 8 novel AML CAR-T targets will be assessed experimentally to identify the most optimal targets for CAR-T cell therapies. In the future, the platform will be utilized to identify pairs of targets for AML gated-CAR therapies and applied to pediatric and adult hematological malignances including T-ALL, MPAL, and multiple myeloma. The CAR-Machine platform is being developed as a software package so that researchers can utilize it to explore CAR-T cell targets within their datasets. The package will assist in identifying most potent CAR-T cell targets with highest probability for clinical translation.