αEP: Complexity is not usually the product of selection

I mentioned in the last one this annoying tendency of too many pro-evolution people to cite "complexity" as a factor that supports the assertion of selection for a trait. Strangely, the intelligent design creationists also yell "Complexity!" at the drop of a hat, only it's to prove that evolution can't work.

They're both wrong.

I ran across a prime example of this recently in a post by John Wilkins (It's pick-on-John-Wilkins day! Hooray!)

Opponents, largely following Gould and Lewontin’s 1979 attack, tend to assert (often without consideration of the particular attempts to give adaptive explanations) that any and all adaptive hypotheses are cheap and to be avoided. This has the effect of basically eliminating natural selective accounts of anything. But we know that selection is the only process that results in complexity over any time, and the fact there are complex traits among organisms leads to the inevitable conclusion that we should be able to give selective explanations from time to time. I have argued before that we should think of adaptation as a viable hypothesis at all times; but being viable doesn’t make it true. The problem is not that EP or sociobiology makes adaptive hypotheses. They should. It is that they often make them without testing them.

I will concede the general point: a well-designed and testable adaptive hypothesis is a perfectly reasonable starting point to do science, and I agree that one big problem with evolutionary psychology is that most of their adaptive hypotheses are poorly done. But I've highlighted the chunk in the middle that's just absurd: complex traits are the product of selection? Come on, John, you know better than that. Even the creationists get this one right when they argue that there may not be adaptive paths that take you step by step to complex innovations, especially not paths where fitness doesn't increase incrementally at each step. Their problem is that they don't understand any other mechanisms at all well (and they don't understand selection that well, either), so they think it's an evolution-stopper — but you should know better.

This is the trap Michael Behe falls into, too. It's the assumption that you have to have an adaptive scenario for every step, and an inability to imagine non-adaptive solutions. I think if selection were always the rule, then we'd never have evolved beyond prokaryotes -- all that fancy stuff eukaryotes added just gets in the way of the one true business of evolution, reproduction.

So let's work through a hypothetical scenario of increasing complexity, and you try to see where selection is essential. And then I'll give some real world examples.

Imagine an enzyme E with with a fairly nonspecific, broad spectrum of binding for various chemical substrates. It can bind substrate A or substrate B with equal facility, and then carries out some modification of the substrate — let's say it hydrolizes some side group. We'll call this enzyme E_AB for its substrates.

Now imagine a fairly common mutational event, a gene duplication. An individual in this population acquires a second copy of its gene, so now it is E_AB-E_AB. Is this event dependent on selection? No, obviously not.

But now this mutation spreads through the population, so a significant fraction of the individuals carry it. Is that dependent on selection?

It could be, but most likely isn't. There are situations where you might gain an advantage from having a second copy of a gene — cases where the organism is dependent on the concentration of the gene products, and the availability of the enzyme is to some degree rate-limiting. Human salivary amylase, for instance, exhibits a pattern of increasing copy number in populations with lots of starch in their diets. But you can't assume that a gene duplication is beneficial a priori; you have to do the hard work of correlating gene number with an effect.

That's because in many other cases, the duplication is going to be neutral in effect, or even deleterious. It's going to be a tiny cost in replication, for example (but negligible, because selection doesn't see very small fitness costs and replication is a small part of most cell's energy budget), and many processes are sensitive to dosage, where increases in gene number can damage the equilibrium of the cell. If this weren't true, then Down syndrome children, who carry an extra copy of a whole chromosome, would be superhuman.

So let's say that this step of the increasing frequency of a gene duplication is usually going to be the result of genetic drift, but sometimes will have an adaptive benefit…but you have to demonstrate the latter, rather than assuming it.

So now you have a population with a large subset carrying the E_AB-E_AB duplication. What next? Mutations can occur in either of the copies.

The most likely result is destruction of the function of one of the copies: E_AB-E_AB becomes E_AB-X, where X is now a damaged, non-functional pseudogene. We're back to square one, except of course if you were arguing that the duplication had to incur a benefit, because now the pseudogene is nothing but cost — it incurs that neglible replication cost, but also it would be at a disadvantage relative to those individuals that carried your hypothetically advantageous duplication. But look: the human genome contains at least 12,000 pseudogenes (out of 20,000 genes), and some genes, like cytochrome C, may have 49 pseudogene copies. Most of these mutations are also going to be nearly neutral.

The interesting case, though, is where something changes in one of the copies short of actually destroying the gene. What if one loses it's broad specificity and now becomes able only to bind one of the substrates…for example, we now have an individual carrying E_A-E_AB. Will this variant expand through the population by selection?

Again, not likely. This individual is still processing A and B with a pair of enzymes, and any difference between it and others would be small (and probably invisible to selection, at least if it's in a messy slow breeding metazoan, for instance).

Similarly, there would be other mutations in the population: we could have mutants carrying E_AB-E_B, for instance, with a more specific second enzyme. We could have mutants that knock out functions in any order: X-E_AB, or E_B-E_AB. All kinds of variants arise by chance mutation, and spread to varying degrees through the population by drift. No selection need be involved, and you get all this complexity in the population arising entirely spontaneously. And a good part of the reason that this can occur is because selection plays no role.

Things get more complex in the absence of selection. Selection, as Wilkins well understands, is generally a conservative process that removes phenotypic variation from the population (with some exceptions). It is precisely the inability of selection to cull all variants that deviate from the successful, well-tested norm that allows novel complexity to emerge.

In our hypothetical population, for instance, we have all kinds of different patterns of the E gene: we have the ancestral type, E_AB, the duplicated form, E_AB-E_AB, forms carrying pseudogenes, X-E_AB and E_AB-X, and forms with subtle differences in the specificity of the enzymes, E_A-E_AB and E_AB-E_B. None of this is the product of selection. But now you can get recombination, and maybe some of those combinations will be visible to selection.

For instance, when X-E_AB breeds with E_AB-X, some of the progeny could be X-X, a complete knockout of the enzyme. That could be, but isn't necessarily, deleterious. You could get the particularly interesting combination E_A-E_B, where now what you've accomplished is the separation of two of the functions of the ancestral enzyme into two separate, more specific enzymes.

Is the condition E_A-E_Bmore fit than the ancestral state of E_AB? I would argue that usually, no, it's not. It's going to be a nearly neutral variant in that both individuals process the A and B molecules just fine. Maybe somewhere further down the road, two separate molecules to do these two functions opens the door to more flexibility and interesting new possibilities, but evolution and selection do not care about future potential, only with short term consequences.

The bottom line is that you cannot easily explain most increases in complexity with adaptationist rationales. You have to consider chance as far more important, and far more likely to produced elaborations.

Do these phenomena operate in the real world, rather than just hypothetical scenarios? Yes. For example, look at the globin genes. These are found in two clusters, an α cluster with seven genes (including two pseudogenes) and a β cluster with five genes. All the genes are related and similar and function — these are all proteins with heme and oxygen-carrying functions — where each gene has slightly different functions (one may have a higher affinity for oxygen than another) and patterns of expression (one may be expressed at a different stage of development than another, or in a different range of tissues), and we can see specific functional adaptations for each, but their origin has to be in a selectively neutral condition.

A fetal hemoglobin with a high affinity for oxygen is an adaptation when expressed in a fetus, but detrimental when expressed in a pregnant woman (it wouldn't share oxygen as readily with the fetus). The selective advantage could only arise after the different globins had evolved, and when differential gene expression mechanisms had been coupled to them to assure that they would only be expressed under appropriate conditions. The origin was not a selection event, but the refinement to specific roles probably was.

The classic example, though, is Joe Thornton's work on the step-by-step evolution of mineralocorticoid and glucocorticoid receptors from an ancestral receptor. You cannot understand it if you think selection is all that matters: the process was driven by an initial combination of chance mutation, drift, and subsequent selection. As Thornton explained in a criticism of Behe,

In his posts about our paper, Behe’s first error is to ignore the fact that adaptive combinations of mutations can and do evolve by pathways involving neutral intermediates. Behe says that if it takes more than one mutation to produce even a crude version of the new protein function, then selection cannot drive acquisition of the adaptive combination.

This does not mean, however, that the evolutionary path to the new function is blocked or that evolution runs into a “brick wall,” as Behe alleges. If the initial mutations have no negative effect on the ancestral function, they can arise and hang around in populations for substantial periods of time due to genetic drift, creating the background in which an additional mutation can then yield the new function and be subject to selection. This is precisely what we observed in our studies of the evolution of the glucocorticoid receptor (GR).

In our 2007 paper in Science, we showed that multiple mutations were indeed required for the GR to evolve its specificity for the hormone cortisol; some of the mutations that trigger the change in function are deleterious if introduced in isolation, but others are “permissive”: they have no apparent effect on the function of the protein, but once they are in place the protein can tolerate the other mutations that shift and then optimize the new function.

Even in something as specific as the physiological function of a biochemical pathway, adaptation isn't the complete answer, and evolution relies on neutral or nearly neutral precursor events to produce greater functional complexity.