A versatile phenotyping platform combining three-photon microscopy, optimized tissue clearing, and deep-learning...

Imaging the internal structures of plant organs at the subcellular level is essential for understanding key processes such as photosynthesis, growth, and stress responses. However, conventional optical microscopy struggles to visualize deep tissues owing to intense light scattering and spectral crosstalk from autofluorescent components-most prominently chlorophyll. To overcome these barriers, we present an integrated platform that combines deep-penetrating three-photon microscopy (3PM), tissue clearing, and a custom deep-learning network (SE-UNet) for intelligent signal separation. First, our sucrose vacuum-infused refractive-index optimization (Su-VIRO) method drastically reduces scattering. Combined with 3PM, it overcomes practical scattering barriers, enabling high-resolution volumetric imaging of chloroplasts throughout entire Arabidopsis leaves (>230 μm depth). Such imaging revealed a gradient in chloroplast size and distribution from the palisade to the spongy mesophyll. Second, to resolve crosstalk without additional hardware, we developed SE-UNet. Trained on a small labeled dataset, it computationally disentangles overlapping signals from single-wavelength acquisitions, enabling simultaneous four-channel visualization of chloroplasts, cell walls, mitochondria, reactive oxygen species, and nuclei within the same leaf volume. Spatial analysis revealed a non-random organization of organelles that is consistent with their known functional interactions, demonstrating the platform's capacity for spatial organelle analysis. We assessed the applicability of our platform to additional leaf types by performing dual-channel imaging of chloroplasts and the plasma membrane in rice. Our work establishes a versatile, deep-tissue, multi-parameter phenotyping platform for plant cell biology, providing support for future genotype-to-phenotype studies.