A pathway-dependency framework for acquired resistance to targeted therapy in EGFR-mutant, KRAS G12C-mutant, and BRAF V600E-mutant NSCLC.

Targeted therapies have improved outcomes in EGFR-, KRAS G12C-, and BRAF V600E-driven non-small cell lung cancer (NSCLC), but acquired resistance is molecularly and spatially heterogeneous. Driver-specific algorithms and validated biomarkers underpin post-progression care. Determining pathway activity, causal dependency, and a tractable therapeutic vulnerability requires evidence beyond detecting an acquired alteration. We propose a lesion- and time-specific pathway-dependency framework centered on the RAS-RAF-MEK-ERK mitogen-activated protein kinase (MAPK) pathway and the PI3K-AKT-mTOR pathway. It distinguishes three provisional biological states: MAPK-dominant resistance, shared-input MAPK-PI3K reactivation, and a candidate PI3K-enriched/MAPK-low state. A clinical management branch encompasses histologic transformation, central nervous system-limited progression, oligoprogression, and diffuse polyclonal progression and may coexist with a biological assignment. Spatially discordant mechanisms support a mixed assignment, whereas insufficient evidence remains indeterminate. Biological assignment integrates contemporaneous lesion-level findings, clonality, histology, and exploratory pathway readouts; progression pattern guides the clinical branch. The framework complements genotype-based classification and may clarify when better-supported systemic, histology-directed, or local treatment should take precedence. Evidence remains uneven: some driver-specific interventions have established clinical evidence or clinically supported activity, whereas most downstream MAPK strategies and all approaches matched to the candidate PI3K-enriched/MAPK-low state remain investigational. Limited tissue availability, spatial heterogeneity, unstandardized assays, and combination toxicity constrain implementation. Prospective studies should determine whether a locked classifier adds predictive value beyond the initiating driver and acquired genomic alterations. Until prospective validation is available, the framework is best suited to mechanistic interpretation and trial design; routine treatment continues to rely on validated biomarkers and established clinical evidence.