Analysis of Single Cells on a Pseudotime Scale along Postnatal Pancreatic Beta Cell Development

Abstract

Single-cell RNA-seq generates gene expression profiles of individual cells and has furthered our understanding of the developmental and cellular hierarchy within complex tissues. One computational challenge in analyzing single-cell data sets is reconstructing the progression of individual cells with respect to the gradual transition of their transcriptomes. While a number of single-cell ordering tools have been proposed, many of these require knowledge of progression markers or time delineators. Here, we adapted an algorithm previously developed for temporally ordering bulk microarray samples [1] to reconstruct the developmental trajectory of pancreatic beta-cells postnatally. To accomplish this, we applied a multi-step pipeline to analyze single-cell RNA-seq data sets from isolated beta-cells at five different time points between birth and post-weaning. Specifically, we i) ordered cells along a linear trajectory (the Pseudotime Scale) by applying one-dimensional principal component analysis to the normalized data matrix; ii) identified annotated and de-novo gene sets significantly regulated along the trajectory; iii) built a network of top-regulated genes using protein interaction repositories; and iv) scored genes for their network connectivity to transcription factors [2]. A systematic comparison showed that our approach was more accurate in correctly ordering cells for our data set than previously reported methods and allowed for direct comparisons with external data sets. Importantly, our analysis revealed never before seen changes in beta-cell metabolism and in levels of mitochondrial reactive oxygen species. We demonstrated experimentally a role for these changes in the regulation of postnatal beta-cell proliferation. Our pipeline identified maturation-related changes in gene expression not captured when evaluating bulk gene expression data across the developmental time course. The proposed methodology has a broad applicability beyond the context here described and could be used to examine the trajectory of other single cell types along a continuous course of cell state changes.