The Big Picture of Multi-Organelles’ Interactome in Live Cells

The study of organelle interactome at system level requires the simultaneous observation of the many subcellular compartments, beyond diffraction limit, and fast to track their dynamics. Conventional multicolour approaches rely on specific fluorescence labelling, in which the number of resolvable colours is far less than the types of cellular organelles.

Here, I will present a new paradigm to address the challenges in spatial resolution, speed, and throughput for imaging of multiple organelles. Instead of using multiple fluorescent dyes for specific labelling, we use a universal lipid-specific dye to stain all the membrane-associated organelles. Then, we use spinning-disk microscopes with extended resolutions for high spatiotemporal, three-dimensional image acquisition in live cells and tissues. Due to the chromatic polarity sensitivity, the difference between the organelles can be separated through spectral ratiometric imaging.

With deep convolutional neuronal networks for segmentation, we can predict 15 subcellular structures using one laser excitation and two detection channels. It not only bypasses the limitations of multi-colour imaging with single-dye labelling but also accelerates the imaging speed by more than one order-of-magnitude, with proportionally reduced phototoxicity. We show transfer learning can predict both 3D and 2D datasets from different microscopes, different cell types, and even complex system of living tissues. It enables us to resolve the 3D anatomic structure of live cells at different mitotic phases and to track down the fast dynamic interactions.