Single‑cell analysis of BAX responses reveals venetoclax sensitivity and enables development of BAX‑targeting combinations in AML

Abstract Venetoclax (Ven)–based regimens have improved outcomes in acute myeloid leukemia (AML), but 20–30% of patients experience primary resistance and many responders relapse within two years, underscoring the need for strategies to predict and overcome treatment failure. No clinically available functional assay predicts Ven benefit; genomic profiling lacks mechanistic insight, and BH3 profiling, which measures mitochondrial priming in permeabilized cells, cannot detect downstream apoptotic defects. We developed the Single-cell Analysis of BAX REsponses (SABRE) assay, an imaging flow cytometry platform quantifying conformational BAX activation (BAX⁶ᴬ⁷) and mitochondrial localization (TOM20) in intact AML cells treated ex vivo. Optimization identified TOM20 as the most reliable mitochondrial marker; basal BAX⁶ᴬ⁷⁺ labeling was 1.34%, BIM SAHB achieved 88.2% as the maximal output, and BAX/BAK-null cells failed to label. In an ongoing analysis of clinically annotated AML samples, Ven-induced BAX activation is markedly higher in responders than in non-responders, and BIM SAHB restores activation toward maximal levels in resistant cases, supporting predictive patient stratification at diagnosis. To address BAX defects revealed by SABRE, we evaluated BAFL, a first-in-class allosteric BAX Actuating Funnel Ligand that binds a distinct regulatory site, cooperates with BIM, and disrupts BAX/BCL-2 complexes. In MOLM13, OCI-AML2, and primary AML samples (n=4), BAFL + Ven reduced viability by 15–20% compared to Ven alone and increased BAX⁶ᴬ⁷⁺TOM20⁺ cells in Ven-refractory models, including a TP53-mutated case with combination activity despite adverse genetics. As part of a broader BAX-targeting toolbox, BTSA1, a direct trigger-site activator, reduced viability from 60% to 32% when combined with Ven in a PDX AML model harboring FLT3-ITD, DNMT3A, and NPM1 mutations. SABRE is clinically adaptable, low-input, high-throughput, and archivable, enabling both prediction of Ven response and mechanistic dissection of resistance in a single workflow, and ongoing prospective validation in newly diagnosed AML patients will determine its clinical utility in guiding functional precision medicine.