Abstract
ABSTRACTEndocytosis is a fundamental cellular process for eukaryotic cells to transport molecules into the cell. To understand the molecular mechanisms behind the process, researchers have obtained abundant biochemical information about the protein dynamics involved in endocytosis via fluorescence microscopy and geometric information about membrane shapes via electron tomography. However, measuring the biophysical information, such as the osmotic pressure and the membrane tension, remains a problem due to the small dimension of the endocytic invagination. In this work, we combine Machine Learning and Helfrich model of the membrane, as well as the dataset of membrane shapes extracted from the electron tomography to infer biophysical information about endocytosis. Our results show that Machine Learning is able to find solutions that both match the experimental profile and fulfill the membrane shape equations. Furthermore, we show that at the early stage of endocytosis, the inferred membrane tension is negative, which implies strong compressive forces acting at the boundary of the endocytic invagination. This method provides a generic framework to extract membrane information from the super-resolution imaging.SIGNIFICANCEEndocytosis is a fundamental cellular process that has been extensively studied with the help of fluorescence microscopy and electron microscopy. A large amount of data has been accumulated about the protein dynamics and the membrane shapes. In this work, we combine the widely used Helfrich model and experimental data of membrane shapes to infer the physical information about endocytosis, including the membrane tension and the osmotic pressure. Our work not only proves Machine Learning as a power tool is able to solve the complicated membrane shape equations, but also provides novel biological insights about the initiation of endocytosis in yeast cells.
Publisher
Cold Spring Harbor Laboratory