Time-dependent solutions for a stochastic model of gene expression with molecule production in the form of a compound Poisson process

TL;DR

This paper derives exact solutions for a stochastic gene expression model with time-dependent burst sizes and parameters, revealing how external oscillations and burst distributions influence gene expression noise and its biological implications.

Contribution

It provides the first analytical expressions for the time evolution of the cumulant-generating function in a general time-dependent gene expression model with arbitrary burst size distributions.

Findings

Gene response to periodic activation resembles an RC low-pass filter.

The nth cumulant depends on the nth moment of burst size distribution.

Different noise measures vary differently over time.

Abstract

We study a stochastic model of gene expression, in which protein production has a form of random bursts whose size distribution is arbitrary, whereas protein decay is a first-order reaction. We find exact analytical expressions for the time evolution of the cumulant-generating function for the most general case when both the burst size probability distribution and the model parameters depend on time in an arbitrary (e.g. oscillatory) manner, and for arbitrary initial conditions. We show that in the case of periodic external activation and constant protein degradation rate, the response of the gene is analogous to the RC low-pass filter, where slow oscillations of the external driving have a greater effect on gene expression than the fast ones. We also demonstrate that the $n$-th cumulant of the protein number distribution depends on the $n$-th moment of the burst size distribution. We use these results to show that different measures of noise (coefficient of variation, Fano factor, fractional change of variance) may vary in time in a different manner. Therefore, any biological hypothesis of evolutionary optimization based on the nonmonotonicity of a chosen measure of noise must justify why it assumes that biological evolution quantifies noise in that particular way. Finally, we show that not only for exponentially distributed burst sizes but also for a wider class of burst size distributions (e.g. Dirac delta and gamma) the control of gene expression level by burst frequency modulation gives rise to proportional scaling of variance of the protein number distribution to its mean, whereas the control by amplitude modulation implies proportionality of protein number variance to the mean squared.