The rule of declining adaptability in microbial evolution experiments

Abstract

One of the most recurrent observations after two decades of microbial evolution experiments regards the dynamics of fitness change. In a given environment, low-fitness genotypes are recurrently observed to adapt faster than their more fit counterparts. Since adaptation is the main macroscopic outcome of Darwinian evolution, studying its patterns of change could potentially provide insight into key issues of evolutionary theory, from fixation dynamics to the genetic architecture of organisms. Here, we re-analyze several published datasets from experimental evolution with microbes and show that, despite large differences in the origin of the data, a pattern of inverse dependence of adaptability with fitness clearly emerges. In quantitative terms, it is remarkable to observe little if any degree of idiosyncrasy across systems as diverse as virus, bacteria and yeast. The universality of this phenomenon suggests that its emergence might be understood from general principles, giving rise to the exciting prospect that evolution might be statistically predictable at the macroscopic level. We discuss these possibilities in the light of the various theories of adaptation that have been proposed and delineate future directions of research.

Keywords: Fisher's model; beneficial mutations; distribution of fitness effects; epistasis; finite-sites model.

Figure 1

Fitness increase as an inverse function of initial fitness. The datasets used here correspond to evolution experiments with S. cerevisiae (BY4741 strain, gray small circles; DBY15108 strain at 250 generations, magenta crosses; at 500 generations, red diagonal crosses), E. coli (mutator MG1655 strain, blue squares; wild-type REL606 strain, green circles) and microvirid bacteriophages (black triangles). Dashed lines shows best-fit log-log linear regression model to each dataset (correlation coefficients: gray, 0.33; magenta, 0.80; red, 0.87; blue, 0.67; green, 0.92; black, 0.87; F-test, all P < 10–15). Note that both axes are on a natural logarithmic scale.

Figure 2

Log-log-linear scaling is readily generated by many adaptation models. The plots show the average ± SD of 1000 computer simulations of asexual populations adapting to a novel environment. All models explored here share the property of giving rise to macroscopic epistasis, albeit through completely different mechanisms: (A) only the average effect of beneficial mutations diminishes with adaptation; (B) only mutation rate declines with adaptation; (C) a finite-sites model; and (D) Fisher's Geometrical Model. Parameter range was chosen to ensure clonal interference dynamics.