SCALE: modeling allele-specific gene expression by single-cell RNA sequencing

Abstract

Allele-specific expression is traditionally studied by bulk RNA sequencing, which measures average expression across cells. Single-cell RNA sequencing allows the comparison of expression distribution between the two alleles of a diploid organism and the characterization of allele-specific bursting. Here, we propose SCALE to analyze genome-wide allele-specific bursting, with adjustment of technical variability. SCALE detects genes exhibiting allelic differences in bursting parameters and genes whose alleles burst non-independently. We apply SCALE to mouse blastocyst and human fibroblast cells and find that cis control in gene expression overwhelmingly manifests as differences in burst frequency.

Keywords: Allele-specific expression; Expression stochasticity; Single-cell RNA sequencing; Technical variability; Transcriptional bursting; cis and trans transcriptional control.

Fig. 1

Allele-specific transcriptional bursting and gene categorization by single-cell ASE. a Transcription from DNA to RNA occurs in bursts, where genes switch between the “ON” and the “OFF” states. k _on, k _off, s, and d are activation, deactivation, transcription, and mRNA decay rate in the kinetic model, respectively. b Transcriptional bursting of the two alleles of a gene give rise to cells expressing neither, one, or both alleles of a gene, sampled as vertical snapshots along the time axis. Partially adapted from Reinius and Sandberg [6]. c Empirical Bayes framework that categorizes each gene as silent, monoallelic and biallelic (biallelic bursty, one-allele constitutive, and both-alleles constitutive) based on ASE data with single-cell resolution

Fig. 2

Overview of analysis pipeline of SCALE. SCALE takes as input allele-specific read counts at heterozygous loci and carries out three major steps: (i) gene classification using an empirical Bayes method, (ii) estimation of allele-specific transcriptional kinetics using a Poisson-Beta hierarchical model with adjustment of technical variability and cell size, (iii) testing of the two alleles of a gene to determine if they have different bursting kinetics and/or non-independent firing using a hypothesis testing framework

Fig. 3

Allele-specific transcriptional kinetics of 7486 genes from 122 mouse blastocyst cells. a Burst frequency of the two alleles has a correlation of 0.852; 425 genes show significant allelic differences in burst frequency after FDR control. b Burst size of the two alleles has a correlation of 0.746; two genes show significant allelic difference in burst size. X-chromosome genes as positive controls show significantly higher burst frequencies of the maternal alleles than those of the paternal alleles. The p values for allelic burst size difference (bottom right) are uniformly distributed as expected under the null, whereas those for allelic burst frequency difference (bottom left) have a spike below significance level after FDR control

Fig. 4

Examples of significant genes from hypothesis testing. a The two alleles of the gene have significantly different burst frequencies from the bootstrap-based testing. b The two alleles of the gene have significantly different burst sizes and burst frequencies. c The two alleles of the gene fire non-independently from the chi-square test of independence

Fig. 5

Testing of bursting kinetics by scRNA-seq and testing mean difference by bulk-tissue sequencing. a Genes that are significant from testing of shared burst frequency and allelic imbalance. *Also includes the two genes that are significant from testing of shared burst size. Change in burst frequency and burst size in the same direction leads to higher detection power of allelic imbalance; change in different directions leads to allelic imbalance testing being underpowered. b Gene Dhrs7, whose two alleles have bursting kinetics in different directions, and gene Gprc5a, whose two alleles have bursting kinetics in the same direction. Dhrs7 is significant from testing of differential allelic bursting kinetics; Gprc5a is significant from the testing of mean difference between the two alleles

Fig. 6

Allele-specific transcriptional kinetics of 2277 genes from 104 human fibroblast cells. a Burst frequency of the two alleles has a correlation of 0.859; 26 genes show significant allelic difference in burst frequency after FDR. b Burst size of the two alleles has a correlation of 0.692. One gene has significant allelic difference in burst size. The results are concordant with the findings from the mouse embryonic development study