Deep neural net tracking of human pluripotent stem cells reveals intrinsic behaviors directing morphogenesis

Abstract

Lineage tracing is a powerful tool traditionally used in developmental biology to interrogate the evolutionary time course of tissue formation, but the dense, three-dimensional nature of tissue limits the ability to assemble individual traces into complete reconstructions of development. Human induced pluripotent stem cells (hiPSCs) enable recapitulation of various aspects of developmental processes, thereby providing an in vitro platform to assess the dynamic collective behaviors directing tissue morphogenesis. Here, we trained an ensemble of independent convolutional neural networks to identify individual hiPSCs imaged via time lapse microscopy in order to generate longitudinal measures of individual cell and dense cellular neighborhood properties simultaneously on timescales ranging from minutes to days. Our analysis reveals that while individual cell parameters are not strongly affected by extracellular microenvironmental conditions such as pluripotency maintenance regime or soluble morphogenic cues, regionally specific cell behaviors change in a manner predictive of organization dynamics. By generating complete multicellular reconstructions of hiPSC behavior, our cell tracking pipeline enables fine-grained understanding of developmental organization by elucidating the role of regional behavior stratification in early tissue formation.