A Dynamic Protein Atlas of Human Cell Division

A new 4D computational model which integrates CRISPR, confocal microscopy, and machine learning allows researchers to study the dynamics of specific proteins which drive mitosis. The model is a framework which can be adapted to study other cellular functions. 

Mitosis is an essential biological function which relies on the intricate coordination of protein assembly at the right time. While live cell imaging can reveal protein distribution and dynamics, the new computational framework, dubbed the Mitotic Cell Atlas, integrates information and enables quantification of dynamic interactions between molecular aspects of the machinery which drives mitosis.

A new study illustrates how exploring and mining data from this atlas can be used to develop new mechanistic hypotheses about the function of proteins inside the cell. The findings are published today in Nature and the research was conducted as a collaborative effort between the European Molecular Biology Laboratory (EMBL), Germany, and the Research Institute of Molecular Pathology, Vienna.

Dr. Stephanie Alexander, research manager at EMBL in Heidelberg and one of the authors, explained the significance of this study:

“Our study delivers the experimental and computational tools to the research community that enable the generation of dynamic protein atlases. We demonstrate that with the example of mitosis and several proteins that we showed to be relevant for mitosis in previous gene silencing studies.

To develop the program, researchers generated a 4D model of the morphological changes which occur during mitosis in human cells. Images of fluorescently knocked-in mitotic proteins from HeLa cells were captured, and data was arranged using a ‘mitotic standard time’ which was based on changes in chromosome structure. Establishing this ‘standard time’ allowed the researchers to objectively map all cell images, relative to a constant time reference for averaging.

28 proteins were selected and tracked using 3D confocal microscopy, to see where in the cell they were located at different time points. Alexander explains the rationale behind selecting these proteins for the pilot data set:

“We basically started with those proteins that appeared most relevant for mitosis to us and that had been decently characterized before. This was mainly done because our approach needed to be validated and recover existing knowledge (which it did). Surprisingly however, our approach also recovers new aspects or details that had been not described before.”

The team developed a supervised machine learning approach to define subcellular structures. They also trained a regression model to assign protein amounts to different reference compartments, which allowed them to quantitatively compare fluxes in specific protein levels.

The authors envision that their computational framework can be adapted to other essential biological functions, such as cell migration or cell differentiation. ‘The concept of standardizing the spatio-temporal cellular context for analyzing dynamic protein distributions in order to understand cellular processes as presented here is generic and we envision its adaptation to other essential biological functions,’ the authors wrote.

The model provides a new and objective approach for understanding a range of cellular processes, as outlined in the discussion:

“Our model provides a standardized yet dynamic spatio-temporal reference system for the mitotic cell that can be used to integrate quantitative information on any number of protein distributions sampled in thousands of different single cell experiments.”

Alexander elaborates: “These tools allow us to explore potential protein interactions and functional networks. And as this atlas is dynamic, we can understand how information is transmitted during mitosis and how protein interactions change. This allows to phrase targeted hypotheses for detailed follow-up stories… To see it all coming together in the very end, similar to a big puzzle, was very exciting.”

To enable future studies, the experimental methods, the quantitative microscopy platform, and the code to create dynamic protein atlases are now openly available for others to use. Alexander explains how this atlas can be used to understand cellular pathways:

“Once the protein atlas is more complete, meaning data of more proteins are integrated, we can study protein networks and identify hubs where all information comes together, or also where alternative pathways exist. Again, as our atlas covers the mitotic progress, we also see whether a hub is central to the whole process or whether hubs change over time. So, we can use the mitotic cell atlas to identify the vulnerable points of the whole system.”

Find out more: http://www.mitocheck.org/mitotic_cell_atlas/

Reference:
Cai, Y., Hossain, M. J., Hériché, J., Politi, A. Z., Walther, N., Koch, B., . . . Ellenberg, J. (2018). Experimental and computational framework for a dynamic protein atlas of human cell division. Nature. doi:10.1038/s41586-018-0518-z