The evolution of gene expression QTL in Saccharomyces cerevisiae

Abstract

Understanding the evolutionary forces that influence patterns of gene expression variation will provide insights into the mechanisms of evolutionary change and the molecular basis of phenotypic diversity. To date, studies of gene expression evolution have primarily been made by analyzing how gene expression levels vary within and between species. However, the fundamental unit of heritable variation in transcript abundance is the underlying regulatory allele, and as a result it is necessary to understand gene expression evolution at the level of DNA sequence variation. Here we describe the evolutionary forces shaping patterns of genetic variation for 1206 cis-regulatory QTL identified in a cross between two divergent strains of Saccharomyces cerevisiae. We demonstrate that purifying selection against mildly deleterious alleles is the dominant force governing cis-regulatory evolution in S. cerevisiae and estimate the strength of selection. We also find that essential genes and genes with larger codon bias are subject to slightly stronger cis-regulatory constraint and that positive selection has played a role in the evolution of major trans-acting QTL.

Figure 1. Genomic distribution of regulatory QTL between BY and RM.

Location of the gene whose expression level is under inspection (vertical axis) versus marker location (horizontal axis) for 2368 trait marker pairs (points) with significant linkage at an experiment-wide permutation based FDR≤0.05. Identities of known major trans-acting QTL are listed above. Lower panels show the synonymous site substitution rate for the two chromosomal regions indicated by boxes. Breaks in the curves are due to the absence of synonymous sites in intergenic regions. As previously described, regions with low neutral substitution rates contain fewer cis-acting QTL .

Figure 2. Likelihood surface for δ and n.

Black contours show the 63%, 95%, and 99% confidence intervals for the joint value of δ and n (based on the χ² ₂ distribution with Δ log_e likelihood = 1, 3, and 5, respectively). Gray contours correspond to Δ log_e likelihood = 10, 15, 20, …, 100.

Figure 3. Rate of accumulation of cis-acting QTL.

Genes were divided into bins based on their locus-specific maximum likelihood estimate of θ×t_i. The rate of cis-acting QTL in each bin (points, 95% CIs shown in vertical lines) was estimated based on the observed number of genes with linkage and the estimated power and false positive rate of linkage analysis. The least squares fit of the purifying selection model to the points (solid line) results in estimates of δ and n of 0.31 and 181 which are somewhat larger but not significantly different from the estimates obtained under the likelihood based approach. The dashed line shows the least squares fit of the neutral model, yielding n = 71.

Figure 4. Illustration of common and rare derived alleles between BY and RM.

The three possible rooted tree topologies for the S. cerevisiae strains are shown with branch lengths approximately to scale. Hypothetical genotypes for a polymorphism between BY and RM are given below. Orange and blue points represent mutations between BY and RM that result in common and rare derived alleles, respectively. Note that repeat mutation leads to apparently common derived alleles as illustrated for the right most topology.

Figure 5. Ancestral selection graph simulation scheme.

The left panel shows a percolation diagram illustrating the underlying Moran model with neutral births realized by all individuals (δ arrows) and extra births realized only by fitter individuals (2, 3, and 4 arrows) (see , for detailed discussion). The right panel shows a realization of the reverse time simulation process for four sampled individuals representing the three S. cerevisiae strains and S. paradoxus. After mutations have been placed on the graph, branching events are resolved depending on the fitness of the two potential ancestors. Resolution of branching events produces a typical coalescent tree but introduces a bias towards advantageous alleles.

Figure 6. Strength of purifying selection against cis-acting regulatory changes.

Light gray shaded areas indicate that 95% CIs for the proportion of rare derived alleles (vertical axis) in synonymous sites and in the promoter and 3′ UTR. Ninety-five percent CIs for the expected proportion of rare derived alleles at selected sites (dark gray shading) and linked neutral sites (black shading) are shown as a function of the scaled fitness difference between selective classes (horizontal axis). The dashed line indicates the scaled purifying selection coefficient (2.1) that is most likely to have produced the observed allele frequency skew based on linear interpolation between 2N_es = 2.0 and 2.2. The rate of substitution at the selected site relative to the linked neutral site, denoted by ζ, is indicated along with the 95% CI.