Single cell morphology trajectory analysis reveals different cell states and transition paths in Epithelial-to-Mesenchymal transition

From physics perspective a cell is a nonequilibrium nonlinear dynamical system that evolves over time, and cell phenotype conversion corresponds to transitions between attractors. Quantitative and mechanistic understanding of a phenotypic conversion process is of fundamental significance, but conventional fixed-cell snapshot data miss some key dynamical information. We developed a deep-learning based live cell imaging and analysis procedure to trace single cell trajectories in high-dimensional morphology space. Using TGF-β induced Epithelial-to-Mesenchymal transition (EMT) in human HK2 cells as a model system, we experimentally depicted the quasi-potential landscape of EMT, identified three distinct morphology states that are correlated with different expression levels of EMT regulators and markers. We further reconstructed an extended state network that depicts coupling between cell cycle and EMT, and quantified the transition matrix from single cell data. In many aspects the single cell studies are analogous to the more established single molecule approaches in molecular biophysics, while the transition process under study is complicated by existence of dynamic and static disorders in the system due to coupling to other dynamical processes.