Deep learning-based image quantification of epithelial cell shapes and its application to polycystic kidney disease

Cell shape is a fundamental determinant of tissue architecture and organ function. In epithelial tissues, cytoskeletal organization and tight junctions regulate cell geometry, shaping functional tissue units. Disruption of these mechanisms may cause diseases such as autosomal dominant polycystic kidney disease (ADPKD), in which cyst formation is characterized by abnormal regulation of epithelial cell shape. The mechanisms of cystogenesis remain incompletely understood, highlighting the need for robust, high-throughput methods to quantify the morphology of epithelial cells. Here, we present a fully automated, deep learning-based image analysis pipeline to quantify epithelial cell shape and tight junction morphology from immunofluorescence images. Our approach employs a U-Net convolutional neural network for accurate segmentation of fluorescence labeled tight junctions. We introduce novel algorithms to quantify overall cell shape and tight junction morphology, as well as to estimate cytoskeletal traction at shared cell borders. Our analysis pipeline objectively identifies subtle morphogenetic changes associated with disease-related mutations, applied to a genetically modified Madin-Darby Canine Kidney cell model of ADPKD. The method enables high-throughput, standardized analysis, reduces observer bias, and facilitates comparison across experiments. We further demonstrate the pipeline's generalizability by applying it to Drosophila egg chamber epithelia. Our results establish a robust and scalable framework for analyzing cell shape and mechanical interactions in epithelial tissues, with broad applications in phenotypic screening, disease modeling, and morphogenesis research.