When is gene expression noise advantageous?

The variability of gene expression levels, also known as gene expression noise, is an evolvable trait subject to selection. While gene expression noise is detrimental in constant environments where the expression level is under stabilizing selection, it may be beneficial in changing environments when the phenotype is far from the optimum. However, expression noise propagates along the gene network, making the evolution of connected genes interdependent. Here, we explore how their position in the gene network constrains the evolution of genes under selection using an in silico evolution experiment. We simulate the evolution of populations of model gene regulatory networks under directional and fluctuating selection on the gene expression level while allowing the basal expression level and expression noise level to mutate. We find that expression noise is only transiently favoured under directional selection, but high levels of noise can be maintained in a fluctuating selection regime. Furthermore, target genes, regulated by other genes, were more likely to increase their gene-specific expression noise than regulator genes. These findings suggest that both the mean and variance of gene expression levels respond to selection due to changing environments - and do so in a network-dependent manner. They further point at gene expression noise as a putative mechanism for populations to escape extinction when facing environmental changes.