Deep-Learning Super-Resolution Microscopy Reveals Nanometer-Scale Intracellular Dynamics at the Millisecond Temporal Resolution

Abstract

Single-molecule localization microscopy (SMLM) can be used to resolve subcellular structures and achieve a tenfold improvement in spatial resolution compared to that obtained by conventional fluorescence microscopy. However, the separation of single-molecule fluorescence events in thousands of frames dramatically increases the image acquisition time and phototoxicity, impeding the observation of instantaneous intracellular dynamics. Based on deep learning networks, we develop a single-frame super-resolution microscopy (SFSRM) approach that reconstructs a super-resolution image from a single frame of a diffraction-limited image to support live-cell super-resolution imaging at a ∼20 nm spatial resolution and a temporal resolution of up to 10 ms over thousands of time points. We demonstrate that our SFSRM method enables the visualization of the dynamics of vesicle transport at a millisecond temporal resolution in the dense and vibrant microtubule network in live cells. Moreover, the well-trained network model can be used with different live-cell imaging systems, such as confocal and light-sheet microscopes, making super-resolution microscopy accessible to nonexperts.