Interaction between mutation type and gene pleiotropy drives parallel evolution in the laboratory

Abstract

What causes evolution to be repeatable is a fundamental question in evolutionary biology. Pleiotropy, i.e. the effect of an allele on multiple traits, is thought to enhance repeatability by constraining the number of available beneficial mutations. Additionally, pleiotropy may promote repeatability by allowing large fitness benefits of single mutations via adaptive combinations of phenotypic effects. Yet, this latter evolutionary potential may be reaped solely by specific types of mutations able to realize optimal combinations of phenotypic effects while avoiding the costs of pleiotropy. Here, we address the interaction of gene pleiotropy and mutation type on evolutionary repeatability in a meta-analysis of experimental evolution studies with Escherichia coli. We hypothesize that single nucleotide polymorphisms (SNPs) are principally able to yield large fitness benefits by targeting highly pleiotropic genes, whereas indels and structural variants (SVs) provide smaller benefits and are restricted to genes with lower pleiotropy. By using gene connectivity as proxy for pleiotropy, we show that non-disruptive SNPs in highly pleiotropic genes yield the largest fitness benefits, since they contribute more to parallel evolution, especially in large populations, than inactivating SNPs, indels and SVs. Our findings underscore the importance of considering genetic architecture together with mutation type for understanding evolutionary repeatability. This article is part of the theme issue 'Interdisciplinary approaches to predicting evolutionary biology'.

Keywords: Escherichia coli; Fisher's geometric model; experimental evolution; pleiotropy; repeatability.

Figure 1.

Expected role of gene pleiotropy and mutation type in determining mutational fitness effects under FGM. Schematic overview of FGM in two-dimensional trait space with genotype (G) and fitness optimum (O). The fitness of each genotype (shown in colour with the black circle sector showing the fitness isocline for G) is determined by the combination of two phenotypes or molecular interactions described by a simplified protein–protein interaction network (top right). Mutations in gene B (with maximum effect m_B) and gene C (with maximum effect m_C) affect only one phenotype, while mutations in gene A can theoretically affect both phenotypes and may realize phenotype combinations across the area bounded by the maximum phenotypic effects of alleles in B and C (indicated by the black square). Some beneficial mutations in gene A (with effect m_A) could optimally combine both single-trait effects and reach higher fitness than mutations in genes B and C alone. Since single nucleotide polymorphisms (SNPs; small circles) generate more phenotypically diverse alleles than indels or structural variants (SVs; large squares), SNPs sample phenotypic space with greater resolution and are therefore more likely to produce this largest possible benefit. This argument also holds for mutations with larger phenotypic effects that may overshoot the optimum (not shown here), where the greater phenotypic diversity of SNPs allows some to more closely approach the optimum than other mutation classes. (Online version in colour.)

Figure 2.

Gene pleiotropy, measured by the level of connectivity, of mutation targets in experimental evolution studies with E. coli. (a) Connectivity of all E. coli genes versus genes targeted by mutations during adaptation. Connectivity estimates were obtained from the STRING interaction database (means ± 95% confidence interval). Statistical significance was determined by a Wilcoxon's test. (b) Boxplots of average per-experiment connectivity of target genes with one or multiple independent mutations. The p-value was determined by a post hoc contrast test in a mixed effects model with experiment as random factor. (c,d) Boxplots of the average per-experiment connectivity of genes targeted by different mutation types (c) and by gain-of-function (GoF) or LoF SNPs (d). p-values are based on post hoc contrast tests (with Tukey's p-value adjustment) in linear mixed models (LMMs) with experiment as a random factor. (Online version in colour.)

Figure 3.

Evolutionary repeatability is driven by large-benefit SNPs, particularly in large populations. Boxplots (a) and (b) show the repeatability at the gene level of different mutation types (a) and GoF and LoF SNPs (b). p-values are based on post hoc constrast tests (with Tukey's p-value adjustment) in LMMs with experiment as a random factor. Only significant differences are indicated. (c) Per-experiment evolutionary repeatability at the gene level depends positively on effective population size, suggesting it is driven by selection rather than mutation bias. Labels show the selective conditions of the different evolution experiments. The Kendall rank correlation coefficient (τ) with associated p-value is shown in the top left corner. (d) Dependence of evolutionary repeatability at the gene level of different mutation types on effective population size. The Kendall rank correlation coefficient (τ) with associated p-value are shown at the top of each subplot. (e) Per-experiment comparison of SNP and indels/SV frequencies for genes belonging to different functional classes alongside the average connectivity of each class. Asterisks indicate signficant differences between the connectivity of each functional class and global connectivity of all E. coli genes (indicated by the red vertical line). Statistical significance was determined by a Wilcoxon's test with Bonferroni correction. Abbreviations: expr. mach., expression machinery; TFs, transcription factors; bact. moti., bacterial motility. (Online version in colour.)