Triple base-edited dual targeting CD70/33 CAR T cells to avoid cytokine-mediated resistance in Acute Myeloid Leukemia Free

Background: CAR-T cells have yet to demonstrate substantial efficacy for acute myeloid leukemia (AML) despite successes in lymphoid malignancies. A recent clinical trial using CD123 CAR T cells (CART-123) to target AML in patients whose disease has failed previous treatments showed limited efficacy (NCT03766126) and the authors found that myeloid-supporting cytokines, GM-CSF, IL3, and FLT3LG, secreted by CAR T cells during therapy, actually supported the growth of AML blasts and drove CAR T cell exhaustion (Bhagwat Nat. Med. 2024). We previously demonstrated that a CD70-targeting chimeric antigen receptor (CAR) that secretes a bispecific T-cell-engaging antibody molecule targeting CD33 (70³³) avoids CD70 and CD33 antigen escape in heterogeneous AML models (Silva*, Martin* Blood 2025). We used our scRNAseq data from patients with lymphoma treated with tisagenlecleucel to confirm production of these cytokines by CAR-T cells. To prevent the secretion of pro-AML survival cytokines from our CAR-T cells, we genetically engineered 70³³ CAR T cells deficient in GM-CSF, IL3, and FLT3LG. Given the deleterious effect of double-strand breaks and subsequent DNA damage response with the use of traditional CRISPR-Cas9 DNA editing, we pursued CRISPR cytosine base editing, which does not introduce double-stranded DNA breaks and preserves CAR-T cell potency during manufacturing (Engel PNAS 2025).

Methods: Donor T cells were thawed and activated (Day 0), transduced with lentivirus for 70³³ CAR (Day 1), and electroporated using MaxCyte (Day 2). CAR T cells were frozen on Day 14, and the efficiency of editing was determined using intracellular flow cytometry, western blot, ELISA, and DNA sequencing. AML lines were evaluated for specific cytokine dependencies based on DepMap (https://depmap.org/portal/). PCR amplified DNA extracts were sent to Plasmidsaurus for nanopore sequencing and analyzed using Crispresso2. Guides to introduce an early stop codon were designed using iSTOP (https://www.ciccialab-database.com/istop/#/) and guides designed to disrupt RNA splicing were designed using SpliceR (https://moriaritylab.shinyapps.io/splicer/). Previous single-cell RNA sequencing (scRNAseq) data of PBMC's from patients with LBCL treated with tisa-cel (Kymriah, n=13) or axi-cel (Yescarta, n=19) were reanalyzed for GM-CSF, IL3, and FLT3L expression in samples taken at baseline and after treatment (Day 7, Haradvala*, Leick*, Mauer*, Gohill* Nat. Med. 2022).

Results: scRNAseq data indicated high levels of CSF2 (GM-CSF) and IL-3 in CD8+ CAR T cells from Day 7 (D7) after treatment in patients treated with tisa-cel but not axi-cel. FLT3LG expression levels were more heterogeneous among post-infusion PBMC cell types (including CAR-T). At least three guides per cytokine were screened for editing efficiency. Peak individual editing efficiencies of 85%, 55%, and 75% were obtained for CSF2, IL-3, and FLT3LG, respectively,which were corroborated by flow cytometry and ELISA for individual protein levels. Triple edited 70³³ were 74%, 40%, and 68% edited forCSF2, IL-3, and FLT3LG, respectively. Triple-edited 70³³ CAR-T cells demonstrated enhanced anti-leukemia potency and proliferation in vitro relative to non-edited 70³³ control CAR-T cells. Additional in vitro as well as in vivo experiments are ongoing and will be presented at the annual meeting.

Conclusion: Use of multiplex base editing to generate cytokine triple-knockout CAR T cells is feasible and appears to enhance the efficacy of CAR-T cells targeting myeloid blasts. Efforts are ongoing to further characterize the relative potency of these triple-edited CAR-T cells.