On measuring selection in experimental evolution

Abstract

Distributions of mutation fitness effects from evolution experiments are available in an increasing number of species, opening the way for a vast array of applications in evolutionary biology. However, comparison of estimated distributions among studies is hampered by inconsistencies in the definitions of fitness effects and selection coefficients. In particular, the use of ratios of Malthusian growth rates as 'relative fitnesses' leads to wrong inference of the strength of selection. Scaling Malthusian fitness by the generation time may help overcome this shortcoming, and allow accurate comparison of selection coefficients across species. For species reproducing by binary fission (neglecting cellular death), ln2 can be used as a correction factor, but in general, the growth rate and generation time of the wild-type should be provided in studies reporting distribution of mutation fitness effects. I also discuss how density and frequency dependence of population growth affect selection and its measurement in evolution experiments.

Figure 1.

Selection and demography in isolation versus competition. The population sizes of two genotypes grown in isolation (first row: dark grey, wild-type; light grey, mutant) or in competition (second row: dark grey area, wild-type; light grey area, mutant) are shown together with the ratio of genotypic frequencies in competition (third row: note the logarithmic scale on the y-axis) for three demographic scenarios. Scenario (a) leads to frequency- and density-independent selection. Scenario (b) illustrates frequency-independent but density-dependent selection, in the particular case where both genotypes have the same carrying capacity K. In this case, s tends to 0 as the population approaches the carrying capacity. In scenario (c), selection is density-independent but frequency-dependent. Recursions were made from the discrete-generation model in equation (A2) from the electronic supplementary material, appendix, with logistic population growth, using λ_w = 1.08, λ_m = 1.09, K_w = 100 000 in all scenarios, and K_m = K_w ln(λ_m)/ln(λ_w) and ι_w = ι_m = 1 in (a); K_m = K_w and ι_w = ι_m = 1 in (b); K_m = K_w ln(λ_m/λ_w), ι_w = 1 and ι_m = 0.98 in (c).