During mammalian development, embryonic stem cells undergo cell-fate decisions to commit to one of three germ layer lineages: ectoderm, mesoderm and endoderm. Although decades of research in developmental biology have revealed key genes that are functionally important for the commitment of cells to each of these lineages, obtaining a comprehensive understanding of a) the intermediate cell types that arise as cells undergo cell-fate decisions, b) the dynamics of gene expression that accompany cells along the respective lineage trajectories and c) how individual genes interact in a so-called gene regulatory network to functionally govern these early cell-fate decisions has remained challenging.
Although gene expression profiling is more robust and accessible than ever before thanks to dramatic advances in RNA sequencing, inferring and modeling the dynamics from inherently static and high-dimensional data (with tens of thousands of genes which show significant changes in expression level) has proven to be a difficult task: Cells of a differentiating population display high levels of cell-to-cell heterogeneity in their gene expression profile, and using conventional PCA-based approaches have not been successful in extracting gene expression dynamics from RNA-seq data.
Using a novel iterative framework developed in a companion manuscript, we have analyzed single-cell RNA-seq data obtained from mouse embryonic stem cells taken at various time points during early germ layer differentiation to identify cell types as well as the lineage relationships between these individual cell types. This allowed us to essentially map out a lineage tree in gene expression space that describes the continuous gene expression dynamics of early germ layer differentiation. Importantly, our results show that cell types occupy discrete locations in gene expression space and correspondingly, cells exhibit abrupt changes in gene expression profile as they transition from one cell type to another. Taking advantage of the finding that certain locations in gene expression space are more “stable” (i.e., far more cells are found in these locations than expected by chance) than others, we used a novel framework to build a probabilistic model of the underlying gene regulatory network. These models allowed us to make and experimentally test predictions as to how cells respond to various signals and gene expression changes. We found that not only are there cell state states, but that each of these cell states have qualitatively different responses to the same perturbations.
Our findings reveal that cell types found during early in vitro germ layer differentiation are discrete both transcriptionally as well as functionally. Using these findings, researchers can now map mouse cells undergoing early differentiation to specific cell types by measurement of a set of genes discovered ab initio from single cell gene expression data; enabling direct and non-invasive measurements of the dynamics of differentiation, and thus the investigation of many new and exciting frontiers in developmental biology regarding how the dynamics of differentiation are regulated.