As you can gather from my publications and my writings, my world view has become deeply affected by neutral theory. This is the idea that much of genomic evolution (at least in eukaryotes) is shaped by nearly neutral changes.
The shift from an adaptationalist point of view to that of a neutralist, is akin to the shift from Newtonian physics to General Relativity. Newtonian physics applies to certain circumstances and is a special case of motion that is more generally explained by relativity. Likewise, adaptationsm applies to certain circumstances and is a special case of evolution that is more generally explained by neutral theory.
Now I don’t want to delve too deeply into the nuts and bolts of neutral theory. If you want to read more on this topic, a good place to start would be, Non-Darwinian Molecular Biology, a perspective I wrote with my grad student, Nevraj Kejiou.
Although on the surface, neutral theory seems controversial, for most researchers studying molecular evolution, this debate has been mostly settled (dissenters do exist, however their views are in the minority). What has remained unclear, is how much neutral theory applies to phenotypes.
As a refresher, let’s look at one key aspect of neutral theory, the limits of selection. For natural selection to act on an allele, it must have an significant impact on the reproductive fitness of the organism. This metric is called the selection coefficient (s). If the mutant allele causes the barer to have (on average) 1% greater number of offspring, then the mutation has a selection coefficient of 0.01. If the mutant allele causes the barer to have (on average) 1% fewer offspring, then the mutation has a selection coefficient of -0.01. Through statistical and mathematical modeling, neutral theory predicts that the absolute value of s (so |s|) has to be greater than 1/Ne, where Ne is the effective population size, for natural selection to even see and act on the mutation. If |s| is smaller, the mutation is essentially neutral and its fixation rate is dictated by random drift.
Note that Ne is affected by many parameters. For example, whether individuals in the population mate randomly, or whether they tend to chose their mates locally. Ne is also affected by historical trends, for example, whether the population has gone through bottlenecks. Other parameters, such as recombination patterns, also affect Ne.
For humans the Ne is roughly 10,000. So in other words, alterations in the genome that have an associate |s| value that is less than 0.0001 are essentially neutral. The evolution of these mutations is thus determine not by selection but by drift.
How neutral theory applies to molecular evolution is very plain. Changes the genome that lead to little change in reproductive success will be subject to evolution by random drift. This includes slightly deleterious and slightly beneficial mutations. For natural selection to kick in mutations have to have significant effects on reproduction, either positively or negatively.
Viewing this from the vantage point of the mutation is also useful. The associated |s| must clear a barrier before it can be seen by selection. This is what Michael Lynch termed the drift barrier hypothesis.
If we are to apply neutral theory to phenotypes, in principle this should be simple. If there is a mutation that causes a trait in a clear cut manner, then everything we spoke about at the molecular level (at the level of the mutation) should also apply to the trait. The problem is, how individual changes in the genome affect phenotypic changes is not so simple. The causal link from one to the other is long. The effects can be minor. The effects may not be consistent. One genotypic change may affect multiple phenotypes. Despite these problems, we can plainly see that there is quite a bit of phenotypic diversity in the human population, and much of the diversity has some genetic basis. Moreover, for evolution by natural selection to even work, there has too be some link. If the links between genotype and phenotype are too diffuse, and too weak, then most mutations in functional regions of the genome would be only slightly deleterious and fail to clear the drift barrier. These would build up overtime, and over the generations our genomes would undergo a meltdown process. So the links between genotype and phenotype can’t all be “super-weak”.
So how do we link the two together? How does this causal link inform us about the role of selection and neutral theory in how phenotypes evolve?
This has been a big problem. But I can start to see a way forward. Recently, many researchers have been conducting large GWAS (gene wide association studies) to link variations in the genome with the propensity of an individual to have a certain trait. If the trait is associated with a reduction in fitness (and by this I really mean associated with having fewer offspring), neutral theory would predict that genomic alterations that have a strong effect should be rare. Why? When these come into existence, selection would limit how much time they could spread in the population, and they would be quickly eliminated. Neutral theory would also predict that the weaker the effect of the mutation has on the deleterious phenotype, the less natural selection plays a role, and as a result, these mutations can drift in the population for longer periods of time. As they drift, a certain percentage of them will reach significant levels. In short, if you were to graph the effect on phenotype of each mutant (y-axis) against the levels of these mutants as a percentage of the population, neutral theory predicts an asymptotic slope, where at some point the mutations will be so weak in their effects, that they would be effectively neutral and their levels should be all over the place. The upshot of all this, is that we could then work our way backwards and figure out how this phenotypic trait affects fitness - and thus relate it back to |s|. We could do this, because we could use neutral theory to calibrate |s| in terms of Ne.
This graph is my rough approximation of what theory predicts and the densities of the data points along the two axes are likely off, but the general idea is that data points should increase with lower frequencies, and with lower effects. The key point is the upper bound of the graph. This is the drift barrier. At some point the impact on the deleterious trait is nearly neutral (where s = -1/Ne), and those types of mutations should be present at percentages across the entire spectrum.
So, do we see this? Yes.
Here is a recent figure from a review on GWAS examining genomic variants linked to schizophrenia:
You’ll notice that if we compare reality to theory, we only see the upper bounds of graph that we previously predicted. This is expected, as our current ability to detect low frequency and low effect alleles is limited by the statistical power of these surveys. This lack of statistical power is what is causing the “Discovery gap” that is labeled in the above figure. But what is clear is that a genetic alteration that increases your odds ratio of being diagnosed with schizophrenia by 1% is effectively neutral. In other words, these mutants have an associated s of about -0.0001.
Now you might be wondering why there are data points that are near the 100% frequency level and slightly above the drift barrier. Some of these are likely the “wildtype” ancestral genotypes that are “less fit” than newer mutations that have recently been created. So an allele at 98% frequency, and an odds ratio of 1.5%, has a complementary allele present in 2% of the population with an odds ratio of -1.5%. For these alleles |s| > 1/Ne, which means that the more fit version (i.e. the new mutant) should spread in the population and eventually become fixed. In other words, positive selection.
So there are a few takeaways.
1) Neutral theory is consistent with data on phenotypes and frequencies that we can presently measure.
2) There are clear and tangible limits to natural selection - all those who simply view evolution through a panadaptational lens should take note.
As the number GWASs increase, and as they analyze more and more subjects, we will be able to better understand the selection coefficients of different phenotypes. We will also begin to understand which phenotypes are effectively neutral. Finally, we will untangle the thorny issue of how much of our phenotypic evolution is dictated by selection and how much is due to the random drift of nearly neutral alleles.