Would evolution repeat if it ran again?

Since 1988, twelve populations of the same bacterium have been grown in the same medium, transferred daily, with samples frozen every 500 generations. The frozen samples are a living fossil record that can be revived and replayed.

Much of what happened was parallel. All twelve grew larger cells, all twelve improved their growth rate, and the same genes were hit repeatedly across independent lines. At that level evolution is repeatable and to a degree predictable.

A 2015 analysis of the same populations addressed a simpler question: does adaptation run out. The expectation was that fitness would approach a ceiling as the easy improvements were used up.

Not yet. In all nine populations measured, fitness was still rising between forty and sixty thousand generations, more slowly than before, and the trajectories are better described by a curve with no ceiling than by one with one.

Those nine did not include the one that did something unusual, which was left out of the measurements. It is a result about ordinary adaptation, and the more general finding of the two.

The growth medium contains citrate, which this bacterium is characteristically unable to use in the presence of oxygen. After more than thirty thousand generations, one population began using it, and its population size jumped.

The key result came from replaying the tape. Clones revived from the frozen samples re-evolved the ability only if they had been taken at generation 20,000 or later: something earlier in that lineage’s history had made the step far more likely. Later work found more than one such change. The first step toward using citrate was slightly beneficial even in the ancestor, became harmful as other mutations accumulated, and was beneficial again by generation 29,000.

Its authors called it an explicit demonstration of historical contingency. The same environment, the same organism, and an outcome that depended on what had already happened by generation 20,000.

Not everyone reads it that way. Van Hofwegen, Hovde and Minnich selected directly for citrate use and got it within 12 to 100 generations, through the same class of mutation, and conclude that the rarity in the long experiment was an artefact of its conditions. Leon and colleagues answer that starving cells on citrate masks the dependence on genetic background that the long experiment exposes.

Convergence is common, perhaps ubiquitous: eyes, wings, echolocation and streamlined bodies have appeared repeatedly in unrelated lineages, which argues that selection finds the same solutions given the same problems. The argument survives examination and does not survive it unqualified. Convergence in similar environments often is selection at work, it can also happen for other reasons, and species under the same pressure sometimes answer it with adaptations that are not convergent at all.

Both observations are correct, which means the question is quantitative rather than binary. How much of what exists is inevitable and how much is an accident nobody can currently say, because the experiment that would tell us is the history of the planet.

Short-term evolution is forecast in practice. A fitness model for the influenza surface protein, inferred from the sequences of strains already circulating, predicts how common each strain’s descendants will be the following year, and was published as a principled basis for choosing what goes into the vaccine. Experts now choose vaccine strains using forecasts of this kind alongside laboratory measurements, and forecasts that add those measurements of how the virus escapes antibodies do better than sequences alone.

Those predictions are statistical and they are about the next step. They do not extend to what a lineage will look like in a million years, and nothing suggests they could.