scReQTL: an approach to correlate SNVs to gene expression from individual scRNA-seq datasets

Abstract

Background Recently, pioneering expression quantitative trait loci (eQTL) studies on single cell RNA sequencing (scRNA-seq) data have revealed new and cell-specific regulatory single nucleotide variants (SNVs). Here, we present an alternative QTL-related approach applicable to transcribed SNV loci from scRNA-seq data: scReQTL. ScReQTL uses Variant Allele Fraction (VAFRNA) at expressed biallelic loci, and corelates it to gene expression from the corresponding cell. Results Our approach employs the advantage that, when estimated from multiple cells, VAFRNA can be used to assess effects of SNVs in a single sample or individual. In this setting scReQTL operates in the context of identical genotypes, where it is likely to capture RNA-mediated genetic interactions with cell-specific and transient effects. Applying scReQTL on scRNA-seq data generated on the 10 × Genomics Chromium platform using 26,640 mesenchymal cells derived from adipose tissue obtained from three healthy female donors, we identified 1272 unique scReQTLs. ScReQTLs common between individuals or cell types were consistent in terms of the directionality of the relationship and the effect size. Comparative assessment with eQTLs from bulk sequencing data showed that scReQTL analysis identifies a distinct set of SNV-gene correlations, that are substantially enriched in known gene-gene interactions and significant genome-wide association studies (GWAS) loci. Conclusion ScReQTL is relevant to the rapidly growing source of scRNA-seq data and can be applied to outline SNVs potentially contributing to cell type-specific and/or dynamic genetic interactions from an individual scRNA-seq dataset. Availability:https://github.com/HorvathLab/NGS/tree/master/scReQTL

Improved SNV discovery in barcode-stratified scRNA-seq alignments

“…In this study we find 20 significant cis-scReQTLs. This number is expected given the input size (up to 70 SNVs and up to 3000 cells per dataset), and, based on our previous studies is likely to be significantly higher in larger datasets [11,18].…”

“…To explore if cells bearing certain scSNVs have related gene expression features, we assessed the scSNV expression in the individual cells after graph-based cell-clustering. For this analysis we processed the scRNA-seq datasets as we have previously described [11,18]. Briefly, after alignment with STARsolo [12] and quality filtering, the gene-expression matrices were processed using Seurat [24] to normalize gene expression and correct for batch-and cell-cycle effects; the normalized gene expression values were then used to assign likely cell types using SingleR [25] (Methods).…”

“…For each scSNV called in more than 5 cells, we performed analysis for a correlation between the VAFRNA and the gene expression (cis-scReQTL) of the harbouring gene using scReQTL as previously described [11]. Briefly, the VAFRNA were correlated to the normalized gene expression values of the most variable genes using a linear regression model as implemented in Matrix eQTL [31].…”

“…Finally, we assessed if some of the scSNVs are correlated the expression of their harboring gene, for which we applied the linear regression model implemented in scReQTL [11]. For this analysis we used scSNVs detected in five and more cells (between 35 and 70 scSNVs per datasets).…”

“…These analyses can complement DNA-based SNV-studies and maximize the potential of scRNA-seq datasets. Importantly, SNVs from scRNA-seq studies can provide crucial information on the SNV functionality through studying the allele specific dynamics and its correlation to phenotype features, such as gene expression and splicing [10][11][12].…”