Dynamic nanoscale reconstruction of endocytosis with high-throughput superresolution microscopy and machine-learning

Clathrin-mediated endocytosis is an essential cellular process for the uptake of molecules from the environment. During endocytosis, more than 50 different proteins in many copies self-assemble into a complex machinery that invaginates the plasma membrane and forms an endocytic vesicle. Currently, no technology can measure the precise locations of the proteins during endocytosis, which limits its mechanistic understanding.

Here we will overcome this technological gap by developing an approach to reconstruct the structural organization of the endocytic machinery in yeast with nanometer spatial and sub-second temporal resolution from hundreds of thousands of superresolution snapshots imaged in fixed cells. Using high throughput single-molecule localization microscopes, we will image every protein of interest together with a reference structure that will be used to sort the snapshots according to their progression along the endocytic timeline, align them in space and average them to obtain multi-color ‘movies’ of the endocytic process.

To analyze the huge amount of data, we will develop new machine learning methods based on deep learning and simulation-based inference to i) automatically detect structures of interest and ii) to rapidly extract parameters from measurements to calculate dynamic structural averages. We will then monitor how this dynamic organization changes upon mutations of specific proteins, to gain mechanistic understanding of their function.

This project will result in detailed insights into the structure and dynamics of endocytosis and in a machine learning-based workflow that can reconstruct the time-resolved structural organization of various cellular protein machines.