Novel single-cell genomic approaches for deciphering cellular heterogeneity

Abstract

Single-cell genomics has reshaped our understanding of developmental biology by uncovering intricate molecular states at unprecedented scale. However, continued development of experimental and computational approaches is needed to fully realize its potential. In this dissertation, I will introduce three distinct projects, each centered around adapting computational tools or developing novel experimental approaches to decipher cellular heterogeneity. In the first project, we adapted latent Dirichlet allocation to single-cell combinatorial indexed Hi-C (sci-Hi-C) intra-chromosomal contact maps to decompose the data into chromatin topics. Our approach enabled co-embedding and clustering of sci-Hi-C data derived from five different cell lines (GM12878, H1Esc, HFF, IMR90, and HAP1) and identification of cell type-specific topics of chromatin interactions. In the second project, we developed inexpensive spike-in controls for single-cell combinatorial indexed RNA-seq (sci-RNA-seq) experiments using a set of single-stranded hash oligonucleotides (“hash ladder”). To normalize for technical variation introduced within individual cells, we calculated a cell-specific size factor that is derived from the hash ladder. We applied the ladder to study the effects of various chemical perturbations, including RNA pol II elongation, histone deacetylation, and activation of the glucocorticoid receptor. In the third project, we vastly improved Visual Cell Sorting (VCS), which is an automated imaging workflow that enables binning and sorting of cells by visual phenotypes, by making it compatible with cell fixation and three-level sci-RNA-seq (sci-RNA-seq3). We applied VCS to sort over one million E15 F1 B6xCAST mouse embryo derived nuclei based on nucleolar and nuclear speckle size, and we profiled the sorted nuclei with sci-RNA-seq3. We revealed differences in these nuclear compartment sizes within and across cell types and identified both expected and unexpected correlations with proliferation and differentiation status. We identified 42 genes that positively correlated with relative nucleolar size, reflecting the activation of gene expression programs relating to ribosomal biogenesis and proteostasis stress response. Finally, we demonstrated that these genes can be used to quantify relative nucleolar size across mouse, human, and zebrafish developmental atlases.