Cell morphology and mechanosensing can be decoupled in fibrous microenvironments and identified using artificial neural networks

• Published in: Scientific reports Vol. 11; no. 1; p. 5950
• Main Authors: Bonnevie, Edward D., Ashinsky, Beth G., Dekky, Bassil, Volk, Susan W., Smith, Harvey E., Mauck, Robert L.
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 15-03-2021; Nature Publishing Group; Nature Portfolio
• Subjects: 631/80/2373; 631/80/79/2066; Algorithms; Animals; Cell culture; Cell Line, Tumor; Cell morphology; Cell Nucleus; Cell Physiological Phenomena; Cells - cytology; Cells - pathology; Cellular Microenvironment; Contractility; Cytology; Fibroblasts; Heterogeneity; Humanities and Social Sciences; Humans; Invasiveness; Learning algorithms; Localization; Machine learning; Mechanotransduction, Cellular; Mice; Microenvironments; Models, Biological; Morphology; multidisciplinary; Neural networks; Neural Networks, Computer; Prediction models; Science; Science (multidisciplinary); Tumor Microenvironment; Yes-associated protein
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-021-85276-5

Cells interpret cues from and interact with fibrous microenvironments through the body based on the mechanics and organization of these environments and the phenotypic state of the cell. This in turn regulates mechanoactive pathways, such as the localization of mechanosensitive factors. Here, we leverage the microscale heterogeneity inherent to engineered fiber microenvironments to produce a large morphologic data set, across multiple cells types, while simultaneously measuring mechanobiological response (YAP/TAZ nuclear localization) at the single cell level. This dataset describing a large dynamic range of cell morphologies and responses was coupled with a machine learning approach to predict the mechanobiological state of individual cells from multiple lineages. We also noted that certain cells (e.g., invasive cancer cells) or biochemical perturbations (e.g., modulating contractility) can limit the predictability of cells in a universal context. Leveraging this finding, we developed further models that incorporate biochemical cues for single cell prediction or identify individual cells that do not follow the established rules. The models developed here provide a tool for connecting cell morphology and signaling, incorporating biochemical cues in predictive models, and identifying aberrant cell behavior at the single cell level.