Genomic tree scans identify loci underlying adaptive peaks in Antirrhinum

Species can be considered to occupy different peaks in a fitness landscape, visualised as an undulating surface, with height corresponding to fitness and horizontal coordinates to genotype1. A key evolutionary problem is to understand how populations traverse fitness valleys to reach different peaks. Hypothetical solutions include shifting balance theory2, peak movements due to environmental changes1,3, or peak connectivity in high-dimensional fitness spaces4,5. However, testing hypotheses is not straightforward because identifying loci and epistatic interactions underlying fitness peaks remains challenging6. Here we analyse the genetic architecture of a fitness landscape by studying barrier loci that reduce gene flow between populations. We identify barrier loci in an unbiased way, by scanning the genome for regions with deep-rooted trees with a common topology. Applying this method to populations of Antirrhinum, we identify seven barrier loci, all of which control flower colour. We propose these loci underlie two peaks in a fitness landscape, the heights and positions of which vary depending on genotype-environment interactions, allowing them to have been accessed without traversing a fitness valley. Our findings provide a model for how reproductive barriers between species can arise through populations taking diverse paths in fitness space.