Variable RNA sampling biases mediate concordance of single-cell and nucleus sequencing across cell types

Abstract

AbstractThe capacity for nuclear RNA measurements to recapitulate results from whole cells is essential to the utility of single-nucleus RNA-seq. Early studies argued that nuclear samples could yield comparable results to single-cell RNA-seq if intronic reads from pre-mRNAs were included in the analysis of both assays. While pre-mRNA sampling has since been acknowledged to be subject to sampling bias related to gene length, the impact of this phenomenon across cell types has been largely ignored. Here, we describe the contrasting effects of mRNA and pre-mRNA sampling on the concordance of gene expression estimates between cells and nuclei. We also address the generalizability of a recently published normalization method intended to maximize assay similarity by removing gene length bias from pre-mRNA sampling. Comparing nuclei to cells among cell types of the cortex, we show that pre-mRNA (intron) abundances are much more similar than mRNA (exon) abundances. When comparing overall gene expression, the magnitude of gene length bias reflects the relative enrichment of pre-mRNAs in nuclei, which varies considerably among cell types of the cortex. This variability leads to unreliable performance of the normalization method, which emphasizes mRNA measurements by downweighting pre-mRNA measurements according to gene length. As a potential alternative, we demonstrate adaptation of an existing method for removing systematic bias from gene set enrichment analysis results. Broadly, our analysis provides a mechanistic explanation for variation in assay similarity across cell types and argues for the application ofpost hocnormalization approaches as an avenue to improved biological interpretation.