Mechanistic modeling reveals a configurable AP-1 network governing cell-state heterogeneity and adaptive plasticity in melanoma cells.

AP-1 transcription factors have been implicated in cellular plasticity, differentiation-state heterogeneity, and phenotype switching, enabling adaptation to anti-cancer therapies. Although AP-1 states, defined by the combinatorial expression of AP-1 proteins, are heterogeneous within cell populations, only a subset of possible states is observed. How these states are constrained, why their distributions vary across cell populations, and what drives their phenotypically consequential transitions remain unclear. We develop a mechanistic model of the AP-1 network, capturing dimerization-dependent, co-regulated, and competitive interactions. Calibrated to single-cell protein measurements across diverse melanoma populations and combined with statistical learning, the model reveals parameters explaining population-specific AP-1 state distributions. These parameters correlate with MAPK signaling across cell populations. The model predicts and experiments validate adaptive AP-1 reconfiguration following MAPK inhibition, driving a dedifferentiated, therapy-resistant state that is attenuated through model-guided perturbations. These findings establish AP-1 as a configurable network and provide a framework for modulating AP-1-driven cell-state plasticity.