High-throughput, unbiased single-molecule displacement mapping with deep learning reveals spatiotemporal heterogeneities in intracellular diffusivity

Abstract

Single-molecule displacement/diffusivity mapping (SMdM) has revolutionized the study of molecular motion in live cells by providing high spatial resolution insights. However, its utility is restricted by measurement biases and low throughput, making it challenging to capture temporal dynamics of intracellular diffusivity. Here we report a high-throughput single-molecule diffusivity microscopy approach (Hi-SMdM) that rapidly and unbiasedly maps nanoscale heterogeneities in molecular motion inside mammalian cells, enabling time-resolved imaging of local diffusivity dynamics. Hi-SMdM employs a self-supervised deep-learning denoising framework with a general noise model to effectively restore the scarce signals of fast-moving single molecules, while eliminating artifact motions without relying on spatial redundancies or temporal correlations. It then provides unbiased estimations of nanoscale diffusion coefficient and improve both throughput and temporal resolution by up to an order of magnitude. We demonstrate the versatility of Hi-SMdM through time-resolved mapping of the spatially heterogeneous diffusivity of free proteins in the cytoplasm. Hi-SMdM also unveils the temporal dynamics of intracellular diffusion under hypotonic conditions in live cells. Additionally, Hi-SMdM tracks the spatiotemporal dynamics of intraorganellar diffusivity during rapid rearrangements of the ER network and functional activities of mitochondria. Overall, Hi-SMdM provides exceptional opportunities for high-throughput and unbiased single-molecule diffusivity mapping with excellent spatiotemporal resolution.