Virtual labeling of mitochondria in living cells using correlative imaging and physics-guided deep learning

In silico labeling in cell biology: Potential and limitations

In silico labeling is the computational cross-modality image translation where the output modality is a subcellular marker that is not specifically encoded in the input image, for example, in silico localization of organelles from transmitted light images. In principle, in silico labeling has the potential to facilitate rapid live imaging of multiple organelles with reduced photobleaching and phototoxicity, a technology enabling a major leap toward understanding the cell as an integrated complex system. However, five years have passed since feasibility was attained, without any demonstration of using in silico labeling to uncover new biological insight. In here, we discuss the current state of in silico labeling, the limitations preventing it from becoming a practical tool, and how we can overcome these limitations to reach its full potential.

Fluorescence microscopy and correlative brightfield videos of mitochondria and vesicles in H9c2 cardiomyoblasts

This paper presents data acquired to study the dynamics and interactions of mitochondria and subcellular vesicles in living cardiomyoblasts. The study was motivated by the importance of mitochondrial quality control and turnover in cardiovascular health. Although fluorescence microscopy is an invaluable tool, it presents several limitations. Correlative fluorescence and brightfield images (label-free) were therefore acquired with the purpose of achieving virtual labelling via machine learning. In comparison with the fluorescence images of mitochondria, the brightfield images show vesicles and subcellular components, providing additional insights about sub-cellular components. A large part of the data contains correlative fluorescence images of lysosomes and/or endosomes over a duration of up to 400 timepoints (>30 min). The data can be reused for biological inferences about mitochondrial and vesicular morphology, dynamics, and interactions. Furthermore, virtual labelling of mitochondria or subcellular vesicles can be achieved using these datasets. Finally, the data can inspire new imaging experiments for cellular investigations or computational developments. The data is available through two large, open datasets on DataverseNO.

Organelle-specific phase contrast microscopy (OS-PCM) enables facile correlation study of organelles and proteins

Current methods for studying organelle and protein interactions and correlations depend on multiplex fluorescent labeling, which is experimentally complex and harmful to cells. Here we propose to solve this challenge via OS-PCM, where organelles are imaged and segmented without labels, and combined with standard fluorescence microscopy of protein distributions. In this work, we develop new neural networks to obtain unlabeled organelle, nucleus and membrane predictions from a single 2D image. Automated analysis is also implemented to obtain quantitative information regarding the spatial distribution and co-localization of both protein and organelle, as well as their relationship to the landmark structures of nucleus and membrane. Using mitochondria and DRP1 protein as a proof-of-concept, we conducted a correlation study where only DRP1 is labeled, with results consistent with prior reports utilizing multiplex labeling. Thus our work demonstrates that OS-PCM simplifies the correlation study of organelles and proteins.