C&SP24: Spatio-temporal cell biology

A. Collaborating Investigators: Alan Horwitz,¹ Gregory Johnson,¹ and Robert Murphy²

B. Institutions: ¹Allen Institute for Cell Science and ²Carnegie Mellon University

C. Funding Status of Project: Data generation fully funded by $100 million gift from the Allen Foundation

 

D. Biomedical Research Problem:

The goal of the Allen Institute for Cell Science (AICS) is to develop predictive models of cell behavior. The initial goal is to assemble a map of the shapes and positions of all major molecular machines and signaling complexes in human induced Pluripotent Stem (hiPS) cells and capture how they change during their differentiation into cardiomyocytes. hiPS cells will be genome-edited to express multiple fluorescent-tagged markers, and 3D images and short movies will be collected for 1000s of cells at various time points after initiation of differentiation, using a large fluorescence microscopy pipeline. We will combine images with various combinations of markers into a comprehensive 3D model of the undifferentiated stem cell and its changes during differentiation. To achieve this goal, several computational tools must be developed:

 

1) Models of causal spatio-temporal relationships of subcellular structures. Understanding how cell organization depends on the location and structure of other cellular components is an unknown and critical to understanding cell behavior and organization. These representations will allow others to model and understand the mechanisms underlying the ordered changes in cellular organization during differentiation. These models will be constructed from single images and short time series, and identify cellular spatio-temporal relationships at the mesoscale.

 

2) Integrated multi-protein subcellular organization models from images where only a small subset of markers are visible. Generating images of many labeled proteins in single cells is not feasible due to the limited number of non perturbing XFP colors. Therefore, it is necessary to build models of subcellular organization by combining information from images containing different, but overlapping, subsets of labeled structures.

 

3) Population modeling. Modeling the origin of the variation among cells in a population and how it may be based on position in cell cycle, partial early differentiation, and unknown effects of neighboring cells is necessary to understand how large populations of cells develop and mature. To parse out these relationships, we will need samples of large numbers of "similar" cells in the context of their differentiation, as well as how neighboring cells are related to cell organization.

 

E. Methods and Procedures:

This project takes advantage of the new CellOrganizer capabilities that will be developed in TR&D4, especially Aim 1 on point process models and causal inference. It will also draw on the existing capabilities for learning point process models^84,85 and for constructing models for a full differentiation process from short movies of parts of that process that were previously developed in conjunction with completed C&SP11 and similar to those in our recent study.⁸⁶ The collaboration will involve both large-scale testing of CellOrganizer in the context of the AICS images and joint development and refinement of methods.