To explore how gene expression changes affect organismal fitness, we computationally analysed previously published experimental data (Keren et al., Cell 2016). These data recorded how yeast growth changes when 80 different genes were driven by 120 different promoters. Additionally, for these promoters the average and time-varying expression strength was already known (Sharon et al.,Genome Research 2014). Data were integrated to reconstruct continuous landscapes of yeast fitness as a function of average and time-varying expression of each genes using a mathematical technique called Gaussian smoothing. We used machine learning approaches to show that, while each landscape looks unique, all landscapes originate from the combination of two principal topologies: the effect of lowered fitness if the gene is expressed at too low copy numbers, or the effect of lowered fitness if the gene is expressed at too high copy numbers. Each gene has a unique combination of these two fitness topologies, depending on its cellular function, and thus a unique fitness landscape. Moreover, integrating over all analysed landscapes we could show that expression noise itself is detrimental to yeast fitness, thus highlighting the adverse effects of the fundamental randomness of all molecular motion on life. Finally, we used stochastic evolutionary simulations to explore possible scenarios for the evolution of gene expression parameters, revealing that for genes with certain landscape topologies, only a combination of transcriptional and post-transcriptional changes can quickly optimize gene expression parameters. This work was disseminated in Nature Communications (Schmiedel et al., Nature Communications 2019) and presented in a talk at the Evolutionary Systems Biology Conference in Cambridge, UK in 2018.
