Abstract
Understanding the predictability of evolutionary change is of major importance in biology. Parallel adaptation provides unique insights through replicated natural experiments, yet mechanisms governing the ample variation in genomic parallelism remain unknown. Here, we investigate them using multi-scale genomic analyses of parallel evolution across populations, species and genera. By resequencing genomes of seven independent alpine lineages from two Arabidopsis species, we found that the degree of gene reuse decreases with increasing divergence between lineages. This relationship is well predicted by decreasing frequency of allele reuse, suggesting that availability of preexisting genetic variation is the prime mechanism. A meta-analysis demonstrated that the relationship further continues within the Brassicaceae family. Thus, we found empirical support for a longstanding hypothesis that the genetic basis of adaptive evolution is more predictable in closely related lineages while it becomes more contingent over larger evolutionary distances.