A perturb-seq map of a differentiation hub reveals synergistic vulnerabilities in KMT2A-rearranged AML

Abstract

Abstract Introduction: KMT2A-rearranged (KMT2A-r) acute myeloid leukemia (AML) is characterized by epigenetic dependencies, yet it often remains clinically resistant to therapies targeting individual regulators. This suggests that resistance is not attributable to any single regulator but is an emergent property of a compensatory epigenetic network. A systematic map of this network's architecture is essential for developing rational and durable combination therapies. Methods: We performed a high-throughput Perturb-seq screen targeting 16 key epigenetic regulators in the KMT2A-r AML cell line MOLM-13. Using computational modeling, we identified co-regulated transcriptional modules and inferred functional interactions. Key findings were validated through pharmacological inhibition assays (in vitro), bulk RNA-sequencing, and clinical correlation analyses using two large, independent patient cohorts (TCGA, n=151; Beat AML, n=672). Results: Our analysis uncovered a core transcriptional module, which we termed the “Myeloid Program”, that was silenced by a compensatory epigenetic repression circuit. This circuit is maintained by a synergistic hub where KAT6A, Menin, and DOT1L converge to maintain leukemic identity. While individual perturbations of these hub components only partially derepressed the Myeloid Program, their simultaneous pharmacological inhibition collapsed the circuit's buffering capacity, leading to robust Myeloid Program reactivation and potent synergistic anti-leukemic activity (Menin/KAT6A ZIP score: 20.19; DOT1L/KAT6A ZIP score: 25.94). Bulk RNA-seq analysis confirmed that dual Menin/KAT6A inhibition induced a unique transcriptional state, synergistically and robustly upregulating the Myeloid Program (NES = 1.77, p-adj = 2.13e-08) more effectively than either single agent. Clinically, high Myeloid Program activity served as a powerful independent prognostic biomarker for improved overall survival (TCGA HR = 0.91, p = 0.0017; Beat AML HR = 0.29, p = 0.046). Our model also predicts potential drug interaction antagonisms. For example, we showed that disruption of an antagonistic regulator, the PRC1.1 component PCGF1, confers strong resistance to DOT1L inhibition (IC50 shift >50-fold). Finally, we established the Myeloid Program as a predictive biomarker whose low activity score identifies a vulnerable, undifferentiated state with unique sensitivity to PI3K/AKT/mTOR pathway inhibitors (e.g., Selumetinib, r = -0.37, p = 2.2e-16; MK-2206, r = -0.31, p = 1.2e-11; Rapamycin (r = -0.34, p = 8.44e-14). Consequently, combining a differentiation-inducing Menin inhibitor with agents targeting this pathway (e.g., MK-2206, ZIP: 21.738, Rapamycin, ZIP: 16.224) was highly synergistic. Conclusion: Our work establishes circuit-level epigenetic compensation as a core mechanism of resistance in KMT2A-r AML. We identify the Myeloid Program as a central regulatory node, a dual-purpose biomarker, and a key therapeutic objective. These findings provide a data-driven rationale for two distinct precision strategies: 1) collapsing the compensatory repressive circuit with synergistic co-inhibition of its core components, or 2) exploiting the emergent vulnerability of the undifferentiated state by combining differentiation agents with PI3K/mTOR pathway inhibitors.