Fig. 1
Summary of methodology. OS25 mESCs were cultured and characterized by single-cell RNA-seq using the Fluidigm C1 system, applying the SMARTer kit to obtain cDNA and the Nextera XT kit for Illumina library preparation. OS25 cells are grown in conditions that select for undifferentiated cells (high Oct4-expressing). Libraries from 96 cells were pooled and sequenced on four lanes of a HiSeq. After quality-control analysis of cells, 90 cells out of 96 remained for further analysis. We first unraveled contributions of components of gene expression variation using the scLVM method13. Removing cell cycle variation and technical noise allowed us to focus on stochastic gene expression. Gene expression variation can be quantified by CV2 or DM, which is a measure of noise independent of gene expression levels and gene length. To explore the transcriptional kinetics of OS25 ES cells, poisson-beta model16 was fitted to single-cell gene expression data, leading to estimates of burst frequency and size. Next, histone and RNAPII promoter modifications were obtained from Brookes et al.5 and integrated with single-cell RNA-seq to investigate relationship between stochastic gene expression and epigenetics. Active genes with no PRC marks are usually in the “on” state with high burst frequencies (k on), PRCr genes are mostly “off” and PRC-active genes switch between “on” and “off” states very frequently. Considering the allele-level possibilities, at active genes, both alleles would be in an actively transcribing state. For PRCa genes, both alleles would be in an actively transcribing state, or both alleles would be in a silent PRC-marked state, or only one allele is in PRC-marked state, which, subsequently, would result in noisier gene expression. For PRC-repressed genes, both alleles are expected to be in a silent PRC-marked state
