Characterization of transcript enrichment and detection bias in single-nuclei RNA-seq for mapping of distinct human adipocyte lineages

Abstract

Single-cell RNA-sequencing (scRNA-seq) enables molecular characterization of complex biological tissues at high resolution. The requirement of single-cell extraction, however, makes it challenging for profiling tissues such as adipose tissue where collection of intact single adipocytes is complicated by their fragile nature. For such tissues, single-nuclei extraction is often much more efficient and therefore single-nuclei RNA-sequencing (snRNA-seq) presents an alternative to scRNA-seq. However, nuclear transcripts represent only a fraction of the transcriptome in a single cell, with snRNA-seq marked with inherent transcript enrichment and detection biases. Therefore, snRNA-seq may be inadequate for mapping important transcriptional signatures in adipose tissue. In this study, we compare the transcriptomic landscape of single nuclei isolated from preadipocytes and mature adipocytes across human white and brown adipocyte lineages, with whole-cell transcriptome. We demonstrate that snRNA-seq is capable of identifying the broad cell types present in scRNA-seq at all states of adipogenesis. However, we also explore how and why the nuclear transcriptome is biased and limited, and how it can be advantageous. We robustly characterize the enrichment of nuclear-localized transcripts and adipogenic regulatory lncRNAs in snRNA-seq, while also providing a detailed understanding for the preferential detection of long genes upon using this technique. To remove such technical detection biases, we propose a normalization strategy for a more accurate comparison of nuclear and cellular data. Finally, we demonstrate successful integration of scRNA-seq and snRNA-seq datasets with existing bioinformatic tools. Overall, our results illustrate the applicability of snRNA-seq for characterization of cellular diversity in the adipose tissue.

Length biases in single-cell RNA sequencing of pre-mRNA

“…1 a , top ), which is consistent with previous studies of UMI-based protocols ( 11 ). On the other hand, unspliced mRNA counts strongly correlate with gene length ( 12 ) ( Fig. 1 a , bottom ).…”

Mapping the temporal transcriptional landscape of human white and brown adipogenesis using single-nuclei RNA-seq

“…For brown adipogenesis, preadipocytes showed heterogeneity at day 0 (Preadipocyte-1 and Preadipocyte-2, Fig. S1B), with the Preadipocyte-2 population expressing genes that correspond to adipogenic capacity (Gupta et al 2021). Independent analysis of brown nuclei harvested at later time-points (day-5 to day-20) showed similar preadipocyte heterogeneity along with capture and detection of mature adipocytes (Fig.…”

“…Preadipocytes from both lineages were isolated and then immortalized to allow for long-term in vitro cell-culture. Previously reported data demonstrated highly concordant molecular features between primary and immortalized preadipocytes (Gupta et al 2021), and in vitro differentiated adipocytes were shown to recover gene expression profiles and function of primary human neck BAT and WAT (Xue et al 2015).…”

“…Our use of snRNA-seq was critical to reduce bias in cell recovery, because mature adipocytes are typically incompletely recovered during single cell isolation (Deutsch et al 2020). Our previous work demonstrated that snRNA-seq was able to capture similar cellular diversity as scRNA-seq in adipose tissue (Gupta et al 2021). In the current work, we isolated intact nuclei from differentiating white and brown preadipocytes at 5 stages of adipogenesis (Fig 1A).…”

Interpretable generative deep learning: an illustration with single cell gene expression data

“…So far, the lower-dimensional latent representations learned by scVI have primarily been used to correct for batch effects, i.e., systematic differences between, e.g., well plates during the execution of an experiment, and to integrate biological data from different modalities. For example, Gupta et al (2021) investigated single-nuclei RNA-sequencing (snRNA-seq) for its ability to identify cell types at different states of adipogenesis. They used scVI for removing technical batch effects which facilitated integrating snRNA-seq and scRNA-seq data for inferring joint patterns between both data types.…”