Bidirectional coupling among EMT, AXL-RB1 signaling and lineage switch drives resistance to osimertinib and worse clinical outcomes in NSCLC

Abstract

Acquired resistance to osimertinib remains a major barrier in EGFR-mutant lung adenocarcinoma (LUAD), and in many patients cannot be explained by secondary targetable mutations. This pattern highlights a central role for non-genetic plasticity programs, including epithelial-mesenchymal transition (EMT), drug tolerance, immune evasion, and lineage switch. Here, we used a systems-level framework to define how these processes are coordinated. We constructed a minimal gene regulatory network integrating core EMT regulators with AXL, RB1, PD-L1, and NF-{kappa}B, and analysed its emergent behaviour using dynamical simulations. The network resolved into two mutually inhibitory, self-reinforcing "teams": an epithelial/sensitive team centred on RB1, miR-200, miR-34, p53, and E-cadherin, and a mesenchymal/resistant team centred on ZEB1, SNAIL, AXL, PD-L1, and NF-{kappa}B. Simulations predicted a strong coupling between EMT and osimertinib resistance, which was validated across bulk transcriptomic datasets from NSCLC cell lines, EGFR-mutant patient cohorts, and perturbation experiments. Inducing EMT increased RB1-loss programs, whereas osimertinib exposure induced AXL and EMT programs, supporting bidirectional regulation and reinforcement. Single-cell and spatial transcriptomic analyses further showed that EMT, AXL, PD-L1 activity, and reduced RB1 signaling co-occur within tumors. Clinically, activation of individual axes such as EMT, RB1 loss, or PD-L1 upregulation was associated with worse outcomes, while combined activation produced markedly poorer survival than any single axis alone. Extending the network to incorporate lineage regulators further linked a partial LUAD-to-LUSC shift with EMT, RB1 loss, and resistance. Together, these findings identify a network topology that coordinates multiple plasticity programs driving osimertinib resistance and suggest that disrupting this cooperative architecture may offer a therapeutic strategy in EGFR-mutant LUAD.