Base Edited HSPCs Are Shielded from Targeted CD33 Therapy but Preserve CD33 Expression

Targeted therapy for AML remains challenging as there is broad overlap between targetable surface proteins on transformed cells and healthy hematopoietic stem and progenitor cells (HSPC). CD33 is expressed in about 90% of patients with AML but also on healthy myeloid cells with strong expression on progenitor cells and only weak expression on mature neutrophils. Given the higher expression on leukemic blasts compared to their healthy counterparts, CD33 is an attractive target in AML. However, targeting CD33 e.g. with the antibody-drug conjugate (ADC) gemtuzumab ozogamicin often leads to prolonged cytopenia due to suppression of normal myelopoiesis. Transplantation of genetically engineered CD33 knock-out allogenic HSPCs protects reconstituting hematopoietic cells from targeted immunotherapy administered for disease control after transplantation. Early clinical trials explore this option but the long-term effect of loss-of-function of CD33 remains unknown. We recently demonstrated that single amino acid substitutions engineered into CD45, CD117 or CD123 protect HSPCs from targeted therapies against these antigens while preserving their function. Here, we describe the identification of CD33 shielding amino acid substitutions that can be engineered by base editing and preserve CD33 expression on HSPCs. We identified binding sites of two CD33 antibodies by alanine scanning of the entire extracellular domain. The most favorable amino acid substitutions at key residues were analyzed computationally and binding reduction was subsequently validated by flow cytometry. 23 individual amino acid substitutions at 3 positions were tested. Eleven retained similar expression, purity and biophysical properties as wildtype CD33. Those variants are shielded from antibody 1 even at high concentrations when assessed by biolayer interferometry. In addition, three variants are shielded from antibody 2 binding. We used the ABE8e_SpRY base editor to engineer HSPCs and tiled 21 sgRNAs centered around the intended edit. Five days after electroporation, HSPCs were evaluated for editing efficiency (Sanger sequencing) and binding to antibody 1 (flow cytometry). sgRNA4 showed high editing efficiency of about 80% as well as loss of binding to antibody 1 (Figure A) while binding to the control antibody that binds a different epitope was maintained. Importantly, Sanger sequencing did not show off-target editing of the closely related SIGLEC22P pseudogene. Edited (sgRNA4) and non-edited con