Fig. 5

Map evolvability depends on selective grain, and it is maximal for Environment-Phenotype maps. In Figs. 3 and 4, simulations assumed that natural selection could act on the entire map. Since we define selective grain as the average number of parameter-phenotype points that be experienced by a single individual in each generation (and hence “seen” by natural selection, see “Methods”), this corresponds to very fine-grained selection. In this experiment, the assumption about high fine-grainedness has been relaxed. For each level of selective grain, the ability of natural selection to evolve a linear map with an arbitrary slope is recorded as the Euclidean-distance (ED)-based fitness after t = 10⁴ generations. Points correspond to individual replicates, and dashed lines to averages over the n = 30 replicates. Point colour represents map type. For each replicate, the target map is a linear function of arbitrary non-zero slope. These plots show that the ability to adapt to a target slope increases non-linearly with selective grain, and that maximal efficiency is achieved when selection is fine-grained (> 1), which corresponds to scenarios in which single individuals can experience more than one input per generation. Such high levels of selective grain are typically only attainable for Environment-Phenotype (EP) maps (see main text). In the GP and PP-maps, in contrast, such high levels of fine-grainedness (inputs/generation > 1; greyish shadowed areas) represent biologically unrealistic scenarios that can only be revealed by means of in silico experiments. However, understanding the evolutionary dynamics in these “unbiological” regions is fundamental to establish whether the evolutionary dynamics observed in real-world populations arise from differences in the selection grain between maps or from other confounding factor(s) (see main text for a “Discussion”). p = 64 individuals, GRN + Multilinear model