The evolutionary process is often conceived as “hill climbing.” To understand the metaphor, imagine that you find yourself trying to escape an oncoming flood. To accomplish this, you need to reach high ground. The simplest approach would be to start walking, or perhaps running, uphill from wherever you are. That strategy will get you to a point that is at a higher elevation than you started.
In evolution, mutations that increase fitness can be thought of as moving uphill, whereas mutations that decrease fitness can be thought of as moving downhill. Natural selection rejects mutations that move downhill while accepting mutations that move uphill; thus, the evolutionary process is able to move uphill towards greater fitness much as the would be flood escapee heads uphill.
The Hills Are Alive?
But this strategy has a problem. You are far more likely to happen to start on the slope of a small hill than a large mountain. If you start climbing, you will not get very far before you reach the top of the hill. Then the strategy will stop working, and you will be “stuck” at the top of the hill. Unless you were lucky enough to start on the slope of a large hill, you will perish in the oncoming flood.
In a situation where no small change leads to an increase in fitness, evolution will be unable to increase fitness. It will be stuck in a local optimum. As such, unless evolution is “lucky” enough to start on a hill leading to high fitness, it will not be able to produce high fitness. It will be unable to account for the highly fit complex living things we observe.
A human facing this problem would not actually be stuck. Upon reaching the top of the hill and realizing that it is not high enough, he would look around for higher land and head towards that. To do that, he would have to go downhill and then back up again. This is because a human can look ahead and make long-term plans. Evolution cannot do this, because it can only take into account what increases fitness right now.
However, some evolutionary theorists have sought to argue against this idea. A case in point is a 2022 paper in Nature Ecology and Evolution, by Sam F. Greenbury et al., “The structure of genotype-phenotype maps makes fitness landscapes navigable.” They express the intuition that evolutionary processes can get stuck:
A key prediction is that a population must typically traverse an unfavourable valley of lower fitness to move from one fitness peak to another.
However, the authors point out two factors that they argue prevent this from being a problem in practice. Firstly, while in the hill metaphor there is a limited range of possible directions to move, there is an immense array of possible mutations. A person attempting to escape a flood can head north, east, west, or south or a combination of those directions, resulting in only a few effective directions to go. But evolution can mutate anywhere along the massive genomes of living things. Intuitively, if there are far more options to try, evolution is far less likely to get stuck.
But the Paper Focuses on a Second Point
It is that in many cases a mutation will have no discernible effect. For example, the standard genetic code is “degenerate,” meaning multiple codons specify the same amino acid. A mutation changing from one codon meaning “serine” to another with the same meaning has no effect for most practical intents and purposes. Many other changes likewise have essentially no effect.
What does that mean for an evolutionary search process? If a mutation that improves fitness is a move uphill and a mutation that decreases fitness is a move downhill, then a mutation with no effect is effectively a move on flat ground. But if every living thing has many possible mutations that have no effect, it is never really stuck; it can always explore the plateau of equivalent genomes. Furthermore, if that plateau is large and expansive, it is likely that somewhere on that plateau there is a hill that can be climbed to find higher ground. There is then no need to temporarily descend in order to reach higher ground.
The paper tests these ideas on RNA folds. A strand of RNA tends to fold into a particular shape depending on its sequence. Different shapes serve different functions. However, often mutating many parts of the sequence does not change the shape. The authors use a program called ViennaRNA, which predicts the folded shape for a given RNA sequence. Using this tool, the authors conducted a number of experiments. They concluded:
Our findings show that certain structural properties of GP maps give rise to navigable fitness landscapes, and that the resulting accessible paths are indeed likely to be exploited in the course of biological evolution.
Essentially, the authors argued that there were paths from one shape to another, and these paths were actually utilized in a model evolutionary process.
The Conservation of Information
The fundamental problem facing evolution is finding rare solutions among numerous possibilities. For example, to evolve a specific protein fold, evolution must “find” a specific sequence of amino acids that produces that fold. While some simple folds may be easy to find, more complex folds are much rarer. But evolution has limited time; it must therefore somehow locate successful solutions, such as suitable protein folds, while only testing a minuscule fraction of those possible.
Imagine searching for buried treasure on an island when you only have time to search a small portion of it. In that case, you would be unlikely to find the treasure. The only way that you have a hope of finding the treasure is if you have outside information about where it is located, such as a map, or if there is some sort of structure to the island that allows you to quickly search large portions of it.
This is the basis of the “conservation of information” critique of evolution. In order for evolution to succeed, it would have to exploit some form of information or structure to find successful possibilities. However, it is unclear what source of information the evolutionary process is supposed to exploit. What sort of structure is supposed to exist in evolutionary search spaces that would allow evolution to succeed?
The paper by Greenbury et al. is putting forward a possible structural property that might answer this question. They argue that the kind of neutral mutations that allow traversing large areas of the search space are a very common property of realistic scenarios. If this is a structure that evolution can exploit, it would help explain the success of evolutionary search and answer the challenge posed by the conservation of information.
New Paper by Ewert and Axe
The journal BIO-Complexity recently published a new paper by Winston Ewert and Douglas D. Axe responding to Greenbury’s results, entitled: “RNA Sequence-to-Structure Mapping has Limited Evolutionary Benefit.”They argue, contrary to Greenbury, that neutral mutations allowing broad exploration do not greatly facilitate evolution. Rather, they leave the fundamental challenge posed by the conservation of information intact.
Please recall the scenario described earlier. There is an oncoming flood and you have headed up the nearest hill in an attempt to escape. However, that hill was relatively small and will not keep you safe from the flood. Greenbury’s argument is essentially that an evolutionary search would not actually be stuck on the top of the hill but can rather explore a plateau of equivalent solutions, eventually finding a better solution, thereby escaping the “hill.”
This analysis fundamentally misunderstands why local optima are a problem. The problem with a local optimum in evolutionary search is not that the search is stuck, but that it does not know where to search instead. If you climb a hill and find yourself at the top, you can look around to identify a higher hill and move toward it. Evolution cannot do that, as evolution does not engage in long-term planning.
The point of the hill‑climbing metaphor is that the hill’s upward slope guides the search for higher ground. The problem of local optima is that while that guidance works for short distances, it is a poor guide over longer distances. Pointing out that evolutionary processes can wander around a vast, featureless plateau instead of being stuck misses the point. Without some sort of guidance about where to search for better solutions, evolution will not succeed.
The new paper by Ewert and Axe shows this mathematically. Structures like those appealed to by Greenbury et al can help an evolutionary search avoid being literally stuck, but cannot facilitate finding higher fitness solutions beyond the success of simply searching randomly.
RNA Demonstration
Greenbury et al. appeared to demonstrate the success of the evolutionary process in navigating these structures. This is because they made the wrong comparison. They did show that an evolutionary process able to traverse through neutral changes performed better than one not allowed to adopt neutral mutations. They did not compare the performance of evolution to random search. When this is done, we observe that random search performance is better than either of their searches.
It turns out that when looking at small RNA sequences, most fold into a small number of folds. This means that it is very easy to evolve one of those folds. Simply assembling RNA sequences at random will quickly find one that exhibits that fold. Indeed, such a strategy would find these common folds much more quickly than Greenbury’s experiments did. Greenbury et al. selected their targets in a way that ensured they consistently chose a common fold. As such, the experiment did not demonstrate the success of the evolutionary process but the ease of the task.
A Serious Challenge to Evolution
This new paper shows that the conservation of information presents a serious challenge to evolutionary explanations. Evolution faces the challenge of identifying rare solutions within a vast set of possibilities. It needs some structure or guidance to find them. The problem of local optima shows that the fitness effects of individual mutations are an insufficient guide. Pointing out that some mutations have no effect does not change this. Rather, it simply underscores the fact that evolutionary theory has no account of how the evolutionary process could be guided to rare fit states.