Because a study of this scope is not feasible using real organisms, Barve and Wagner began with a model of the 1,397-reaction network used by the bacterium E. coli. From this starting point, they sought to evolve the network by swapping out a reaction from the E. coli network and replacing it with a randomly selected reaction from the pool of known metabolic reactions. (Although science has not documented every metabolic reaction in nature, metabolism is relatively well understood and is easier to work with and more universal than other systems.)
They set up one requirement for this swap: The network must remain able to use glucose. This requirement served as a stand-in for natural selection, and it filtered out the dysfunctional swaps.
Barve and Wagner produced 500 new metabolic networks, each the result of 5,000 swaps. They then evaluated each one, asking whether it could metabolize any of 49 other carbon sources in addition to glucose. It turned out that 96 percent of the networks could employ multiple carbon sources. The average network could use almost five of them. In other words, one adaptation (viability on glucose) was accompanied by multiple potential exaptations.
The results weren’t limited to glucose-driven networks. Wagner and Barve repeated the experiment, selecting for the ability to use each of the other 49 carbon source molecules, and found that the majority of these randomly created networks could function on multiple carbon sources.
They also found that this flexibility couldn't be easily explained by so-called metabolic proximity between carbon sources. In other words, a network that could use glucose was not reliably predisposed to be able to use a molecule that could be easily made from glucose. “If that was the only explanation for the incidence of exaptation, that would not be interesting,” Wagner said. “It would be a necessary consequence of how biochemistry works.”
Instead, the complexity of the network appeared to determine its flexibility; the more reactions in a network, the greater its potential for exaptation. “A lot of what organisms do could actually be engineered in a much simpler way,” Wagner said. “This result suggests this complexity can have important byproducts, namely traits that are potentially beneficial.”
Barve and Wagner’s work adds to a growing number of examples of exaptation at the molecular level. Thornton, for example, has studied the evolution of hormones and their receptors, which fit together like lock and key. Under the right circumstances, he found, one half of a partnership can be co-opted to give rise to a new hormone-receptor system.
Thirty-one years ago, Gould and Vrba suggested that repetitive DNA sequences known as transposons, which originated from viruses, might serve no direct function at first, but may be used to great advantage later on. Since then, research has shown that transposons played an important role in the evolution of pregnancy. “They come from viruses, but they can be utilized for something they are not built for,” said Gunter Wägner, an evolutionary biologist at Yale University and Andreas Wagner’s former doctoral adviser. The two are not related.
Shifting the Balance
The metabolism study suggests that a healthy portion of novel traits get their start as exaptations. In fact, the ratio skews heavily that way; networks selected for one trait, viability on glucose, had, on average, nearly five non-adaptive traits they could potentially draw upon. Barve and Wagner argue that this should prompt a rethinking of assumptions about the origins of beneficial traits.
Wagner explained by providing a scenario: Imagine that a microbiologist isolates a new bacterium and finds that the bacterium is viable on a fairly common carbon source. “So reflexively this microbiologist would say, well, the bacterium is viable on that carbon source because that is an adaptation, it has helped the bacterium survive in the past,” Wagner said. “But our observations say that is not necessarily true. Maybe this is just one of the byproduct traits.”