Lattice light sheet activation structured illumination volumetric super-resolution live microscopy

Volumetric super-resolution (SR) live imaging at high-speed for long-time remains challenging, especially for fluorescent thick specimens. In this work, we develop Lattice light sheet Activation Structured Illumination Microscopy (LA-SIM), which integrates the unique advantages of superior optical sectioning and low photo-toxicity offered by light sheet configuration with full-version SR-SIM in both linear and nonlinear schemes together, by means of reversibly photo-switchable fluorescent protein (rsFP). Moreover, LA-SIM allows implementing rationalized deep-learning (rDL) denoising in a self-supervised manner, enabling the production of high-quality three-dimensional (3D) SR images at limited SNR levels, while achieving bona fide optical resolution improvement beyond 2-fold relative to diffraction-limit in both lateral and axial dimensions. We utilize the high-quality rDL LA-SIM data to train a large-scale transformer model --- SRFormer that can overcome the inferior properties of rsFPs and enable multi-color long-term volumetric SR imaging. We demonstrate that SRFormer LA-SIM can shed new light on subcellular dynamics from single-cell to multi-cellular specimens.