Gene expression is an intrinsically stochastic process, which often results in substantial expression noise, i.e. cell-to-cell protein abundance variability in clonal cell populations. Interestingly, some genes appear systematically noisier than other, rising the question of what are the molecular determinants that can modulate expression noise in a gene-specific manner. Previous experimental studies and computational models have highlighted that transcription happens in bursts, and that promoter architecture has a strong influence on such dynamics, and therefore on expression noise. In contrast, little is known about the impact of post-transcriptional regulatory processes, and a quantitative understanding of how distinct cis-acting feature contribute to expression noise is lacking.
I will present an in-depth data analysis of molecular determinants of expression noise in yeast. We integrated single-cell measurements of expression noise together with multiple genome-scale experimental data sets that describe attributes of expression and of regulatory control, ranging from mRNA synthesis to protein degradation. Using Partial Least Square Path modeling, we (i) measured the influence of individual cis-acting features on noise while accounting for the indirect contribution of other features, and (ii) quantified the impact of distinct stages of regulation on noise. This analysis showed that determinants of transcript stability, translation efficiency and protein stability contribute to expression noise in comparable proportions to that of transcription initiation. Further, we show that these molecular features have an additive impact on expression noise, suggesting that their presence or absence in a single gene might yield predictive amounts of expression noise. These results thus indicate that it might be possible to genetically engineer expression noise at any level of the central dogma.