I am grateful to my collaborators @jaescudero.bsky.social
and Lucia Garcia-Pastor, my colleagues at the Anrdeas Wagner group. I am grateful to the reviewers for their helpful feedback and the Science production team for choosing a vivid illustration of a rugged landscape for the cover

Interestingly, all high fitness peaks (27Asp and 27Glu) essentially share one enormous basin of attraction. This means that the evolving populations have access to multiple peaks, which makes it difficult to predict evolutionary outcomes.

It turned out that most genotypes have a large number of paths leading to high fitness peaks, and very few paths leading to low fitness peaks. Consequently, high peaks are evolutionary accessible to most genotypes (i.e. have large basins of attraction).

This result is robust against different population genetics assumptions

Given the presence of multiple low and high peaks, we considered this landscape as rugged and expected that adaptive evolution would be constrained. However, our simulations showed the opposite. High fitness peaks were reached in 77% of simulations via short adaptive paths

74 peaks have high fitness. All of them have negatively charged amino acids Asp or Glu at the conserved DHFR position 27. Otherwise, high fitness peaks have different amino acids at other amino acid positions

The resulting fitness landscape has >500 fitness peaks, most of which are low fitness peaks.

We took great care to validate our fitness measurements

We measured fitness at the presence of the antibiotic using a mass selection experiment and deep sequencing.