Differences in molecular sampling and data processing explain variation among single-cell and single-nucleus RNA-seq experiments

Abstract

The capacity for nuclear RNA measurements to recapitulate results from whole cells is essential to the utility of single-nucleus RNA-seq. While early studies argued that nuclear samples could yield comparable results to single-cell RNA-seq, both assays have since been appreciated to be subject to certain systematic biases. These include differences in the localization and capture mode of spliced and unspliced RNAs. However, variability in these biases across cell types has received limited direct attention. Here, we describe the contrasting effects of mRNA and pre-mRNA sampling on inter-assay compatibility among cell types of the cortex, their mechanisms, and their impact on the generalizability of a recently published normalization method. We show that baseline similarity is comparatively high in cortex cell types, and is moderated both by gene length bias in pre-mRNAs and by inherent differences in mRNA abundances, the balance of which varies considerably across cell types. These factors cause inconsistent performance of the normalization method, which emphasizes mRNA measurements by downweighting pre-mRNA measurements according to gene length. Critically, the high baseline similarity of single-cell and nucleus RNA-seq in cortex depends on the inclusion of pre-mRNA signal, but pre-mRNA abundances alone are not strongly predictive of mRNA abundances. This is important to problems such as bulk sample deconvolution. Broadly, our analysis provides a mechanistic explanation for variation in assay similarity across cell types and argues for the potential application of post hoc normalization approaches as an alternative route to improved biological interpretation.

scCensus: Off-target scRNA-seq reads reveal meaningful biology

“…We note, however, that we have adopted in this manuscript a rather conservative notion of spliced and unspliced status [Eldj’sarn Hjörleifsson et al, 2022, He et al, 2023], and other approaches, that consider purely exonic reads as arising from spliced RNA, will lead to different ratios. Existing studies have shown the prevalence of intronic reads [He et al, 2023, Pool et al, 2023, Chamberlin et al, 2022, 10x, 2021] and proposed algorithms that utilize unspliced reads in different ways [La Manno et al, 2018, 10x, 2022b, Gorin et al, 2023].…”

“…In general, throughout this article, we will refer to “unspliced” molecules in the understanding that they may be actively undergoing splicing and hence be partially spliced. Meanwhile, researchers have realized the underappreciated value of off-target reads and designed sophisticated algorithms to incorporate off-target reads in scRNA-seq data analysis [Chamberlin et al, 2022, 10x, 2022b, Pool et al, 2023, Chari et al, 2023, Gorin et al, 2023]. One example is single-cell RNA velocity, in which cellular splicing dynamics are inferred using spliced gene counts from reads compatible with spliced transcripts, and unspliced counts accounting for reads compatible with unspliced transripts [La Manno et al, 2018].…”

Understanding and evaluating ambiguity in single-cell and single-nucleus RNA-sequencing

“…Further, the conventions related to how UMIs arising from different splicing statuses should be handled when quantifying different assays are actively shifting as we learn more, as a community, about the nature of the data and the type of information contained within it. For example, recent work suggests potential benefits to including UMIs derived from unspliced molecules or other sources, even when quantifying what is expected to be spliced RNA (1–3). However, even these approaches are unlikely to be the final word about best practices for quantification.…”

“…The evaluation of the splicing status of molecules, as spliced (mature) or unspliced (nascent), is an important task in the analysis of single-cell and single-nucleus RNA-seq data. Popular droplet-based technologies, like the 10x Genomics Chromium platform, often sequence a mixture of spliced (mature) and unspliced (nascent) RNA sequences (1), and recent work has suggested that there may be benefits to accounting for UMIs arising from both spliced (mature) and unspliced (nascent) molecules (1)(2)(3) in subsequent analysis. Moreover, it may often be important to separate the counts of molecules of different splicing status, allocating the associated UMIs as having a spliced or unspliced (or sometimes ambiguous) origin.…”

“…In reality, ambiguous UMIs are likely to come from a combination, with unknown mixture proportions , of both spliced (mature) and unspliced (nascent) RNAs. Further, these proportions can vary from sample to sample, cell type to cell type, across genes (3), and, of course between different assays. Simply classifying these UMIs as ambiguous, while a conservative approach, vastly reduces the amount of usable data under existing downstream processing pipelines, further exacerbating the challenges posed by the sparse sampling of the most prevalent single-cell sequencing technologies.…”