Abstract
The extent to which adaptation can be predicted, particularly for traits with complex genetic bases, is unknown. Here, we leveraged a model complex trait, model species, and high-powered longitudinal sampling design to test the efficacy of genomic prediction of complex trait variation and evolution in ecologically relevant settings. We monitored genome-wide allele frequencies and pigmentation variation in genetically diverse populations of Drosophila melanogaster across seven generations of evolution in both field mesocosms exposed to natural environmental fluctuations, as well as mesocosms housed in a controlled, lab-based setting. At two time points throughout trait evolution, we conducted a high-powered, tail-based mapping of pigmentation, producing a well-resolved genotype-phenotype map that reaffirms canonical pigmentation genes and unveils novel loci. While we were able to use this map to correctly infer the direction of pigmentation evolution in both the field and lab mesocosms, the particular loci responding to selection, and thus architecture of adaptation itself, were largely unpredictable. We suggest this unpredictability to be a result of pleiotropic constraint, which was more pronounced in the field, relative to the lab-based, environment. Finally, we quantified a striking stability of the genotype-phenotype map across genetically diverged populations, demonstrating that shifting epistatic landscapes associated with the evolutionary process itself do not alter trait architecture and preclude phenotypic prediction, provided the mapping is sufficiently powered. In concert our results highlight both the promise and limitations of genomic prediction, and exemplify the challenges of applying lab-based studies of complex traits to their evolutionary dynamics in the wild.