Non-transitivity – commonly illustrated by the rock-paper-scissors game – is purported to be common in evolution despite a lack of examples of non-transitive interactions arising along a single line of descent. We identify a non-transitive evolutionary sequence in the context of yeast experimental evolution in which a 1,000-generation evolved clone loses in direct competition with its ancestor. We show that non-transitivity arises due to the combined effects of adaptation mediated by the evolving nuclear genome combined with the stepwise deterioration of an intracellular virus. We show that multilevel selection is widespread: nearly half of all populations fix adaptive mutations in both the nuclear and vira... More
Non-transitivity – commonly illustrated by the rock-paper-scissors game – is purported to be common in evolution despite a lack of examples of non-transitive interactions arising along a single line of descent. We identify a non-transitive evolutionary sequence in the context of yeast experimental evolution in which a 1,000-generation evolved clone loses in direct competition with its ancestor. We show that non-transitivity arises due to the combined effects of adaptation mediated by the evolving nuclear genome combined with the stepwise deterioration of an intracellular virus. We show that multilevel selection is widespread: nearly half of all populations fix adaptive mutations in both the nuclear and viral genomes, and clonal interference and genetic hitchhiking occur at both levels. Surprisingly, we find no evidence that viral mutations increase the fitness of their host. Instead, the evolutionary success of evolved viral variants results from their selective advantage over viral competitors within the context of individual cells. Overall, our results show that widespread multilevel selection is capable of producing complex evolutionary dynamics – including non-transitivity – under simple laboratory conditions.
A common misconception is that evolution is a linear “march of progress,” where each genotype along a line of decent is more fit than all those that came before (1). Rejecting this misconception implies that evolution is non-transitive and evolutionary succession will, on occasion, produce organisms that are less fit compared to a distant ancestor. In ecology, nontransitive interactions are well-documented in response to resource (2) or interference competition (3). Early studies in experimental evolution have suggested that non-transitive interactions can arise (4); however, little is known regarding how specific events along a single evolutionary line of descent can lead to non-transitivity. Here we determine the sequence of events leading to the evolution of non-transitivity in a single yeast population during a 1,000-generation evolution experiment. We show that non-transitivity arises through multilevel selection involving both the yeast nuclear genome and the population of a vertically-transmitted virus. By expanding our study of host-virus genome evolution to over 100 additional yeast populations, we find that multilevel selection, and thus the potential for the evolution of non-transitive interactions, is widespread.
Previously we evolved ~600 haploid populations of yeast asexually for 1,000 generations in rich glucose medium (5). We characterized extensively the nuclear basis of adaptation for a subset of these populations through whole-genome whole-population time-course sequencing (6) and/or fitness quantification of individual mutations (7). For one population (BYS1-D08) we isolated individual clones from Generation 0 (Early), Generation 335 (Intermediate), and Generation 1,000 (Late) and performed pairwise competition experiments at multiple starting frequencies (Fig. 1A, detailed methods are available in the supplementary materials). We find that the Intermediate clone is 3.8% more fit relative to the Early clone and that the Late clone is 1.2% more fit relative to the Intermediate clone (Fig 1B). The expectation, assuming additivity, is that the Late clone will be more fit than the Early clone, by roughly 5.0%. Surprisingly, we find that the Late clone is less fit than expected, to the extent that it often loses in pairwise competition with the Early clone (Fig 1B). This fitness disadvantage can be overcome if the Late clone is started at high frequency relative to the early clone (Fig. S1), thus creating a bi-stable system where, above a certain frequency, the Late clone wins but below this frequency it loses.