Computational Tools for Mapping Cellular Differentiation by Single-Cell Transcriptomics

Abstract

Proper function of adult tissues depends on a balanced process of cell proliferation, differentiation, and death. Understanding the mechanisms that govern these fate decisions is a central goal of stem cell biology. Classical methods typically rely on a small number of marker genes, making them susceptible to missing meaningful heterogeneity in progenitor populations. By measuring the expression of thousands of genes in tens of thousands of cells, single-cell RNA-sequencing (scRNA-seq) offers more comprehensive descriptions of complex differentiation processes. However, scRNA-seq data also present new analysis challenges and must be related to prior knowledge and existing functional assays in order to provide meaningful insights. This dissertation first discusses computational approaches to two common tasks in the analysis of any scRNA-seq dataset. Chapter 1 presents a general method for identifying cell multiplets, artifacts that arise from the the labeling of two or more cells with the same barcode, generating a hybrid transcriptome that can lead to misleading results in downstream analyses. Chapter 2 describes SPRING, an interactive data visualization tool that facilitates the intuitive exploration of scRNA-seq data, particularly in the context of differentiation. Next, the power of using scRNA-seq to map cellular differentiation is demonstrated in two adult stem cell tissues. In the work discussed in Chapter 3, scRNA-seq was applied to hematopoiesis, the production of blood cells in the bone marrow. Hematopoiesis is among the best-studied adult stem cell systems, facilitating follow-up and making it useful for validation of our approaches. However, questions remain about the structure of the hierarchy, how cells commit to a single lineage, and the stages of maturation following commitment. Using a single-cell snapshot of cells in the early stages of hematopoiesis, the data were used to reconstruct the differentiation hierarchy, identify the earliest stages of red blood production, and generate predictions that could be experimentally validated using traditional functional assays for cellular differentiation. Finally, the study in Chapter 4 demonstrates the generality of these methods by applying them to the poorly characterized bone marrow mesenchymal stromal cell (MSC) population. These cells give rise to fat, bone, and cartilage and also serve as the niche for hematopoietic progenitors. Similar approaches to those described in Chapter 3 enabled the computational mapping of the MSC hierarchy and suggested new regulators of MSC differentiation.