Conflict between noise and plasticity in yeast

Abstract

Gene expression responds to changes in conditions but also stochastically among individuals. In budding yeast, both expression responsiveness across conditions ("plasticity") and cell-to-cell variation ("noise") have been quantified for thousands of genes and found to correlate across genes. It has been argued therefore that noise and plasticity may be strongly coupled and mechanistically linked. This is consistent with some theoretical ideas, but a strong coupling between noise and plasticity also has the potential to introduce cost-benefit conflicts during evolution. For example, if high plasticity is beneficial (genes need to respond to the environment), but noise is detrimental (fluctuations are harmful), then strong coupling should be disfavored. Here, evidence is presented that cost-benefit conflicts do occur and that they constrain the evolution of gene expression and promoter usage. In contrast to recent assertions, coupling between noise and plasticity is not a general property, but one associated with particular mechanisms of transcription initiation. Further, promoter architectures associated with coupling are avoided when noise is most likely to be detrimental, and noise and plasticity are largely independent traits for core cellular components. In contrast, when genes are duplicated noise-plasticity coupling increases, consistent with reduced detrimental affects of expression variation. Noise-plasticity coupling is, therefore, an evolvable trait that may constrain the emergence of highly responsive gene expression and be selected against during evolution. Further, the global quantitative data in yeast suggest that one mechanism that relieves the constraints imposed by noise-plasticity coupling is gene duplication, providing an example of how duplication can facilitate escape from adaptive conflicts.

Figure 1. Noise–plasticity coupling relates to promoter architecture.

Noise–plasticity coupling for genes initiating from TATA-box promoters (A) and non-TATA promoters (B). Scaled noise and plasticity levels are shown for all genes with available data in yeast. In (C,D) the same comparison is made, but excluding all essential genes, haploinsufficient genes and genes required for growth (C–D). Correlation coefficients and P-values are shown inset.

Figure 2. Noise–plasticity coupling is associated with promoter chromatin structure.

Noise–plasticity coupling for genes with high proximal promoter nucleosome occupancy (A) and low proximal promoter nucleosome occupancy (B). Here only genes not required for growth or viability are considered. The comparison for all genes and for genes transcribed from non-TATA promoters is shown in Figure S1. Spearman correlation coefficients and P-values are shown inset.

Figure 3. Low noise–plasticity coupling for core cellular components.

The correlation between gene expression noise and gene expression plasticity is shown for essential genes (A), haploinsufficient genes (B), and genes required for growth (C). Noise and plasticity data are scaled between 0 and 1. Spearman rank correlation coefficients (rho) and P-values are shown for each gene class.

Figure 4. Gene duplicates are enriched amongst genes with the highest expression noise and plasticity, and they tend to gain TATA promoters.

The proportion of duplicates is shown for genes with different expression noise (A) and plasticity (B). Proportions shown for equally populated bins of genes. (C) Genes retained as duplicates following the whole genome duplication in yeast are more likely to have gained a TATA box in their promoters since the duplication event than those reverting to a single copy (P = 1.24×10⁻⁵, Fischer's exact test). Here, only genes inferred to have ancestral TATA-less promoters are considered, N = 470 (retained as duplicates) and N = 1994 (not retained as duplicates).