Description

Characterisation of isoform-expression patterns in single-cell RNA-sequencing data

Trung Nghia Vu¹, Quin F Wills², Krishna R Kalari³, Liewei Wang⁴, Yudi Pawitan¹, Mattias Rantalainen^1
1Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Sweden; ²Wellcome Trust Centre for Human Genetics, University of Oxford, UK; ³Department of Health Science Research, Mayo Clinic, USA; ⁴Molecular Pharmacology and Experimental Therapeutics, Mayo Clinic, USA

Transcriptional heterogeneity is common in seemingly homogenous populations of cells. Single-cell gene-level expression variability has been characterized by RNA-sequencing in multitudes of biological context to date, but few studies have focused on heterogeneity at isoform-level expression. Here we investigated the commitment of individual cells to expressing specific isoforms by studying the co-variability of expression levels of pairs of isoforms from the same genes in single-cell RNA-sequencing data (Fluidigm C1) from a triple-negative breast cancer cell line (N=384 cells). We propose a novel method “ISOform-Patterns (ISOP)”, based on mixture modeling, to characterize dependency between isoforms from the same genes. Based on this method we defined a set of principal patterns of isoform expression relationships and demonstrated that these patterns were present in multiple single-cell data sets. We also assessed to what extent expression patterns of isoform pairs were associated with intrinsic factors, such as different transcriptional start sites, the number of annotated isoforms of the gene and average gene expression level. We found that the frequencies of isoform patterns were different between isoforms with the same and different transcriptional starting site, while isoform patterns were not directly associated with average gene expression level. We also investigated the effect on the distribution of isoform expression patterns under a small-molecule perturbation. Applying ISOP for analysis of isoform expression patterns allowed us to discover isoform patterns associated with the perturbation, including novel findings not discovered through conventional differential expression analysis.