If mutation is the ultimate source of genetic novelty, can a bias of mutation inflow alter the evolutionary trajectory of a population?
On the surface, it would appear the answer should be yes. Intuitively, the path through genotype space should be influenced by the manner in which mutations are introduced into the population. For an easy to digest analogy, consider a rephrased thought experiment proposed by Stoltzfus and Yampolsky called “Climbing Mount Probable” [Stoltzfus, A., & Yampolsky, L. Y. (2009). Climbing mount probable: Mutation as a cause of nonrandomness in evolution. Journal of Heredity, 100(5), 637–647]. Beginning with the 80 year old fitness-landscape as a mountain analogy, we can envision a population of haploid organisms clustered about the face of a mountain with the organisms’ elevation representing their absolute fitnesses. Genetic novelty introduced through mutation during reproduction allows our hypothetical population to move across the surface of the mountain, exploring new areas. Provided the mountain isn’t too rugged, our population should climb its way toward the apex via selection.
However, part of the process of moving across the mountain is blind: the parent cannot choose where the offspring is placed on the mountain since mutation is a random process. Selection will move the population to greater finesses up the mountain, but the introduction of mutations is the result of a roll of dice: a mutant could be better, worse, or around the same fitness as its parent. In Stoltzfus and Yampolsky’s thought experiment, they propose that the introduction, in contrast to selection’s “acceptance” of a mutation, can be a driver of evolution. In their mountain-climbing analogy, they imagine a bias in the introduction of mutations that tends to move our hypothetical population toward the left. It would stand to reason that our mountaineers, given a relatively smooth landscape, would have a tendency to move up and to the left as offspring are produced.
Is there any evidence that such mutation bias exists in nature? If so, can this bias drive the direction of evolution, and can mutation bias itself evolve?
To begin, let’s look at evidence for a bias by looking at genetic composition. During the first few years after DNA was discovered to be the information storage mechanism behind heredity, scientists discovered that species had different ratios of Watson-Crick base pairs: the ratio of AT to GC differed. Indeed these differences can be quite extreme and sometimes ecologically consistent. Endosymbionts, for example in the gut-dwelling Buchnera bacteria of some termites, tend to have an extremely high AT to GC ratio. This may suggest that environmental pressures select for a particular mutation bias.
However, such a bias could also be caused by the need to encode for particular amino acids necessary in such environments. Further, there is an inherent direction in the mutation rates for nucleotides. Cytosine, for example, tends to mutate more often than the other nucleobases, often leading to a transition to thiamine. Still, it is not clear if the mutation biases drive differences in composition or if the observed differences in DNA content is simply the result of selection without having to invoke a particular mutation bias as a causal factor.
At a higher level, there is evidence that amino acid substitutions tend to be biased in a particular direction. In early proteomics, Margaret Dayhoff and her colleagues created a substitution matrix for what they termed “accepted” mutations. They found that a near-sync state of their Markov-style substitution matrix came pretty close to the observed frequency of amino acids in their sample set. Later studies showed that amino acid transitions tend to cluster into groups depending upon their physical properties. Again, these observations do not paint a clear role of mutation bias. There is no good way to distinguish if a mutation bias is moving a population in a particular direction in the fitness landscape or if selection is the single, overwhelming force.
One way to untangle the relationships among fitness, mutation bias, and selection is to rely on computational systems. The aforementioned Yampolsky and Stoltzfus, for example, have produced a few papers looking at the conditions in which a mutation bias can drive evolution both in simple 2-allele and NK models. For our work, we used Avida to explore whether or not an imposed mutation bias can alter the outcome of evolution, whether mutation bias can be an evolvable trait itself, and what possible impacts a mutation bias has on the long-term behavior of evolution.
(A) A substitution matrix for successful mutations in an Avida experiment and (B) it’s saturation state after 2000 applications.
Inspired by Dayhoff’s work, we created our own substitution matrix of “accepted” mutations in our digital organisms over the course of multiple experiments. We hypothesized that using this mutation bias (shown in the figure), we would observe faster and more robust evolution of selected phenotypes because we were biasing the introduction of mutations based upon observed successful mutations. Indeed, we measured both significantly faster evolution of our most-complex phenotype (the logic operation EQU, used in many Avida experiments) as well as that particular phenotype evolving more often across replicate populations.
These results are not terribly surprising. We should expect our imposed mutation bias to reproduce some of the effects of selection from previous experiments. However, what if a population could evolve to use mutation bias? If the case, it might allow a population to adapt its internal genetic environment to the pressures of its external world.
In order for a trait to be evolvable, it must be heritable, variable, and selectable. There are mechanisms in natural organisms that bias the introduction of particular types of mutation that meet at least two of these criteria. DNA replication machinery is variable in fidelity and heritable as are mechanisms for repair. For our experiment, we provided Avidians with these two properties by providing a set of heritable mutation biases they could mutate among. We found that replicate populations would evolve to use a particular mutation bias significantly more often than chance, indicating that mutation bias can be an evolvable trait.
Interestingly, we also found that mutation bias might be self re-enforcing. When we took Avidians evolved under one fixed mutation bias and placed them in an environment with a different fixed mutation bias (disallowing beneficial mutations), they did not perform as well as organisms that evolved under the second-stage mutation bias the entire time. In other words, genotypes seemed to have adapted to confer a type of robustness to deleterious mutations based upon their native mutational environment.
However, before we get our hopes up for the potential for mutation biases to be a driver of evolution, we did find some limitations. When we gave organisms the ability to directly evolve their bias outside of a predetermined small set, the sheer variability available caused no signal to emerge. This result may be the caused by (relatively) small population sizes or not enough evolutionary time elapsed before the analysis. But it could also indicate a more fundamental limitation: that mutation is simply too weak of a evolutionary driver to be a strongly evolvable trait. It appears the verdict on the strength of mutation to evolve populations in a particular direction remains supported in computational systems, but unclear in the natural world.
Do you think this could be the case in natural systems?