Model and analyse fluctuations and bursts in gene expression

The intrinsic stochasticity in the dynamics of mRNA and protein expression has important consequences on gene regulation and on non-genetic cell-to-cell variability. Recently experimental work in prokaryotes and eukaryotes relying on single cell resolution time lapse imaging has enabled a quantitative analysis and modeling of the stochastic processes underlying observed fluctuations.

In this project we develop algorithms to deconvolve time traces from single mammalian fibroblast cells that exploit a novel shorted-lived bioluminescence reporter providing unprecedented temporal resolution. One aim of this project is to investigate the bursting nature of transcription in general, and how it may be involved in the control of circadian or ultradian gene expression. To this end we analyze the promoter of a classical circadian transcription factor, BMAL1, as well as non-circadian regulatory sequences. In this work, we focus on the analysis of such signals using stochastic models that describe the three main processes of gene expression: gene activation, transcription and translation. Previous studies have mostly focused on describing the variability across populations and identifying the different sources of noise. Instead, we aim to reconstruct the temporal sequence of gene activity, mRNA and protein states from individual time traces. For this we developed a 3-layered Hidden Markov Model (see figure) to describe gene activation, mRNA synthesis and protein translation. Deriving analytical approximations for the transition probabilities, we implemented decoding and estimation algorithms that enable us both to infer instantaneous gene activity status, mRNA, and protein copy number. Moreover the same method is used to learn the activation, synthesis and degradation rates defining the stochastic model, as well as to compute the uncertainty of the inferred trajectories.



We have tested the implementation by extensive simulations using the Gillespie algorithm. Importantly, we have successfully applied the method to parse real recordings indicating that rapid bursting at timescales of tens of minutes may be an intrinsic property of the transcription process in mammalian cells. It remains an open question as to how the bursting relates to fundamental processes such as transcription factor binding dynamics or chromatin remodeling. Our new algorithms will enable us to tackle these important issues in a principled and quantitative fashion.