Background: Bivalent promoters marked with both H3K27me3 and H3K4me3 histone modifications are characteristic of poised promoters in embryonic stem (ES) cells. The model of poised promoters postulates that bivalent chromatin in ES cells is resolved to monovalency upon differntiation. With the availability of single-cell RNA sequencing (scRNA-seq) data, subsequent switches in transcriptional state at bivalent promoters can be studied more closely.
Results: We develop an approach for capturing genes undergoing transcriptional switching by detecting 'bimodal' gene expression patterns from scRNA-seq data. We integrate the identification of bimodal genes in ES cell differentiation with analysis of chromatin state, and identify clear cell-state dependent patterns of bimodal, bivalent genes. We show that binarization of bimodal genes can be used to identify differentially expressed genes from fractional ON/OFF proportions. In time series data from differentiating cells, we build a pseudotime approximation and use a hidden Markov model to infer gene activity switching pseudotimes, which we use to infer a regulatory network. We identify pathways of switching during differentiation, novel details of those pathway, and transcription factor coordination with downstream targets.
Conclusions: Genes with expression levels too low to be informative in conventional scRNA analysis can be used to infer transcriptional switching networks that connect transcriptional activity to chromatin state. Since chromatin bivalency is a hallmark of gene promoters poised for activity, this approach provides an alternative that complements conventional scRNA-seq analysis while focusing on genes near the ON/OFF boundary of activity. This offers a novel and productive means of inferring regulatory networks from scRNA-seq data.
Keywords: Bimodality; Bivalency; Chromatin state; Embryonic stem cells; Genome regulatory network; Hidden Markov model; Pseudotime; scRNA-seq.