Whole tissue imaging of cellular boundaries at sub-micron resolutions for deep learning cell segmentation: Applications in the analysis of epithelial bending of ectoderm

Abstract

Background: To understand cellular morphology, biologists have relied on traditional optical microscopy of tissues combined with tissue clearing protocols to image structures deep within tissues. Unfortunately, these protocols often struggle to retain cell boundary markers, especially at high enough resolutions necessary for precise cell segmentation. This limitation affects the ability to study changes in cell shape during major developmental events. Results: We introduce a method that preserves cell boundary markers and matches the refractive index of tissues with water. This technique enables the use of high-magnification, long working distance water-dipping objectives that provide sub-micron resolution images. We subsequently segment individual cells using a trained neural network segmentation model. These segmented images facilitate quantification of cell properties of the entire three-dimensional tissue. As a demonstration, we examine mandibles of transgenic mice that express fluorescent proteins in their cell membranes and extend this technique to a non-model animal, the catshark, investigating its dental lamina and dermal denticles—invaginating and evaginating ectodermal structures, respectively. This technique provides insight into the mechanical environment that cells experience during developmental transitions. Conclusions: This pipeline, named MORPHOVIEW, provides a powerful tool to quantify in high throughput the 3D structures of cells and tissues during organ morphogenesis.