AI enables nanotextural multiplexing of sub-cellular structures

With the rise of optical super-resolution microscopy (SRM), intracellular structures can be analyzed with nanometer resolution. Here, specific fluorescent markers enable multi-color imaging of several sub-cellular structures, by multiplexing the labels based on their emission wavelengths and complementary spectroscopic readouts. However, the best resolving SRM approaches, i.e. STED, SMLM and MINFLUX, rely on special dyes with delicate photophysical properties, which often limits the applicability of multicolour acquisitions, especially in complex biological contexts. To that end, we explore the analysis of nanoscopic textures inherent to intra-cellular organelles that can be utilized for computational multiplexing. Introducing textural demixing following Unet feature generation, trained on publicly available data, we demonstrate its potential to regressively extract at least three complex structures at a time from a single-channel grayscale SMLM image. We investigate the capacity of this context agnostic texture recognition and prove that AI is capable of identifying individual structures even when heavily overlaying each other. We farther demonstrate that AI trained on single organelles imaged by SMLM is able to multiplex single-color, multi-organelle SMLM data and show that our method is also applicable to MINFLUX data without the need for additional training. Texture-sensitive nanoscopy significantly broadens the scope of multi-color SRM by straight-forwardly enabling the multiplexed use of the best performing dyes in complex biological contexts.