Metacell-based differential expression analysis identifies cell type specific temporal gene response programs in COVID-19 patient PBMCs

Abstract

BackgroundBy resolving cellular heterogeneity in a biological sample, single cell RNA sequencing (scRNA-seq) can detect gene expression and its dynamics in different cell types. Its application to time-series samples can thus identify temporal genetic programs active in different cell types, for example, immune cells responses to viral infection. However, current scRNA-seq analysis need improvement. Two issues are related to data generation. One is that the number of genes detected in each cell is relatively low especially when currently popular dropseq-based technology is used for analyzing thousands of cells or more. The other is the lack of sufficient replicates (often 1-2) due to high cost of library preparation and sequencing. The third issue lies in the data analysis --usage of individual cells as independent sampling data points leads to inflated statistics. MethodsTo address these issues, we explore a new data analysis framework, specifically whether "metacells" that are carefully constructed to maintain cellular heterogeneity within individual cell types (or clusters) can be used as "replicates" for statistical methods requiring multiple replicates. Toward this, we applied SEACells to a time-series scRNA-seq dataset from peripheral blood mononuclear cells (PBMCs) after SARS-Cov-2 infection to construct metacells, which were then used in maSigPro for quadratic regression to find significantly differentially expressed genes (DEGs) over time, followed by clustering analysis of the expression velocity trends. ResultsWe found that metacells generated using the SEACells algorithm retained greater between-cell variance and produced more biologically meaningful results compared to metacells generated from random cells. Quadratic regression revealed significant DEGs through time that have been previously annotated in the SARS-CoV2 infection response pathway. It also identified significant genes that have not been annotated in this pathway, which were compared to baseline expression and showed unique expression patterns through time. ConclusionsThe results demonstrated that this strategy could overcome the limitation of 1-2 replicates, as it correctly identified the known ISG15 interferon response program in almost all PBMC cell types. Its application further led to the uncovering of additional and more cell type-specific gene expression programs that potentially modulate different levels of host response after infection.