Mapping the Dynamic Protein Network of Dividing Cells in Space and Time

Live cell imaging is a powerful tool for studying the distribution and dynamics of proteins. However, due to the difficulties in absolute quantification and standardization of data obtained from individual cells, it has not been used to map large sets of proteins that carry out dynamic cellular functions. Cell division is a good example of this challenge for an essential cellular function, as rapid changes in protein localization and protein interactions result in dramatic changes to subcellular structures and cellular morphology, which in turn influence the behavior of the enclosed proteins. Here, I report an integrated experimental and computational pipeline to map the dynamic protein network of dividing human cells in space and time. Using 3D live confocal microscopy, I imaged human cell lines that stably expressed fluorescently tagged mitotic proteins throughout mitosis. To obtain the absolute quantities of protein abundance with high subcellular resolution over time, the microscopy pipeline was calibrated by fluorescence correlation spectroscopy (FCS). Cell and chromosome volumes were segmented as references of cellular context for temporal and spatial alignment based on fluorescent landmarks. Together with my colleague Julius Hossain, we computationally generated a canonical model of mitotic progression for both kinetics (“mitotic standard time”) and morphology (“mitotic standard space”) by averaging and kinetically and geometrically parametrizing many registered dividing cells. The resulting model enabled us to subdivide the mitotic process into 20 characteristic kinetic steps and integrate our complete proof of concept dataset of 13 mitotic proteins imaged in over 300 dividing cells, represented as the 3D protein localization probability of each protein over time. To measure localization similarities between different proteins and make predictions about their dynamic interactions, the integrated data was then mined using supervised as well as unsupervised machine learning. The power of this approach was demonstrated by our ability to automatically identify the major subcellular localizations of all proteins in the dataset and quantify protein fluxes between subcellular compartments and structures. Due to the quantitative nature of our imaging data, we were able to estimate the abundance of each protein in mitotic structures and complexes such as kinetochores, centrosomes, and the midbody, and determine the order and kinetics of their formation and disassembly. The integrated computational and experimental method I present in my thesis is generic and scalable and makes many dynamic cellular processes amenable to dynamic protein network analysis even for large numbers of components. The pipeline provides a powerful instrument for analyzing large sets of quantitative live imaging data of fluorescently tagged proteins. It allows the systematic mapping and prediction of dynamic protein networks that drive complex cellular processes such as mitosis, thus promoting our understanding of the mechanisms by which many molecules together achieve spatio-temporal regulation