Efforts to quantify a Waddington-like landscape from the perspective of dynamical systems, in which the valleys of the landscape correspond to the attractors of the governing cell state dynamics, have largely focused on describing cell state dynamics in terms of molecular mechanisms such as gene regulatory networks. Here, we use morphological data to quantify stable cell states. We use the ability of human induced pluripotent stem cell-derived endothelial-like cells (hiPSC-ECs) to align, elongate, and migrate parallel or perpendicular to the direction of flow under low and high fluid shear stress (FSS), respectively, as a model of an inducible cell state change. We developed a segmentation-free unsupervised machine learning (ML) approach to extract features from the transmitted light images from timelapse videos of hiPSC-ECs expressing GFP-tagged VE-Cadherin under different FSS. We applied a dynamical systems framework to these features and constructed vector field representations of the cell state dynamics, from which we identified stable fixed points at varying levels of FSS. We found that the way these attractors change with FSS suggests a first-order phase transition. We confirmed that these stable states capture qualitative characteristics of the observed biological cell states by using the ML model to reconstruct representative images of the VE-Cadherin channel at these fixed points. We also integrated the cell-agnostic ML features with segmentation-based metrics, such as cell density and alignment, and individual tracked cell trajectories. This analysis supported our earlier biological interpretation of the stable states and showed that cells take varied paths toward the fixed points. Our work acts as a proof-of-concept framework for quantifying the non-equilibrium driving forces of cell state switching and response to perturbation from microscopy image-based data.
Angelini et al. (Sun,) studied this question.