The Neutral Theory of Evolution

Austin Leland Hughes (1949–2015) taught in the Department of Biological Sciences at the University of South Carolina. Hughes studied the evolution of altruistic behavior in human beings and adaptive molecular evolution, subjects to which he made significant contributions.¹

Originally proposed by Motoo Kimura, Jack King, and Thomas Jukes, the neutral theory of molecular evolution is inherently non-Darwinian.² Darwinism asserts that natural selection is the driving force of evolutionary change. It is the claim of the neutral theory, on the other hand, that the majority of evolutionary change is due to chance.

Each individual in a typical mammal population has two copies of its genome in almost every cell. The exact DNA sequences they contain may differ as the result of mutations, random copying errors in which one nucleotide letter is replaced by another. Other changes can also occur, such as the deletion or duplication of larger DNA segments. The result is genetic variation, and it is estimated that, for human beings, each child acquires 100 new mutations—50 in each genome copy—that were not present in its parents’ DNA.³ Genetic variation means no more than spelling differences in the DNA sequences carried by different individuals in a population. When a new DNA spelling is generated, an allele is born. This alternative form of the original gene may or may not lead to changes in the organism’s physical characteristics.

Evolution involves the substitution of one allele for another in a population. Having come about by chance, a new allele becomes increasingly common, and finally replaces the old allele. An evolutionary substitution has occurred.⁴

When the technology enabling the study of molecular polymorphisms—variations in the sequences of genes and proteins—first arose, a great deal more variability was discovered in natural populations than most evolutionary biologists had expected under natural selection.⁵ The neutral theory made the bold claim that these polymorphisms become prevalent through chance alone. It sees polymorphism and long-term evolutionary change as two aspects of the same phenomenon: random changes in the frequencies of alleles. While the neutral theory does not deny that natural selection may be important in adaptive evolutionary change, it does claim that natural selection accounts for a very small fraction of genetic evolution.

A dramatic consequence now follows. Most evolutionary change at the genetic level is not adaptive.

It is difficult to imagine random changes accomplishing so much. But random genetic drift is now widely recognized as one of the most important mechanisms of evolution. Together with J. B. S. Haldane and Ronald Fisher, Sewall Wright was one of the founders of mathematical population genetics. Genetic drift, Wright reasoned, is inevitable in populations having a finite size.⁶ This is because natural populations are subject to all the uncertainties of random sampling. If a fair coin is tossed three times, the probability that it will land on heads, tails, and then heads (H-T-H) is one in eight.⁷ This is the probability to which these coins will converge in the long run. However, in the short run, it is impossible to predict exactly how many times H-T-H will occur. The sequence might not appear at all, or else appear surprisingly often. In keeping with the laws of large numbers, the error—the difference between the proportion of H-T-H observed in our samples and the true probability—will shrink toward zero as our number of trials increases.

When organisms mate, only about half of maternal or paternal genetic material is given to its offspring. Unless a pair of mates has an incredibly large number of children, some of their alleles must be randomly lost. When one allele dies out, others will take its place. Quite apart from the issue of random sampling, different breeding pairs will leave different numbers of offspring. If one family has four children while a second has five, the second family’s alleles will have a higher frequency in the next generation.

Random events make possible many evolutionary substitutions. This is genetic drift, a force than can easily become more powerful than natural selection. Differences in differential reproduction caused by natural selection can be slight, easily dwarfed by random differences.

The neutral theory is not necessarily incompatible with the occurrence of evolutionary substitutions by Darwinian evolution. However, to the extent that evolutionary change is neutral, selection is rendered superfluous. In other words, the majority of evolution may be neutral, but natural selection can be invoked to explain some key, albeit rare, changes that are adaptive.⁸

The Neutral Theory

Austin Hughes viewed Motoo Kimura, the primary developer and advocate of the neutral theory, as a figure as important as Charles Darwin in evolutionary biology.⁹ Law-like change had been a familiar concept since, at least, the Stoics, but Kimura’s neutral theory, together with Werner Heisenberg’s uncertainty principle and Kurt Gödel’s incompleteness theorem, suggested that the universe is, at its core, non-deterministic.

Adaptive evolution, Hughes noted, was very often treated as if it were itself a null hypothesis, remarking that:

This was an extraordinary proposal—surely unique in the history of science. For it is the usual practice in science that the null hypothesis is the hypothesis of no effect, which is accepted unless we are able to show that the data deviate significantly from what would be expected under the null hypothesis.¹⁰

The neutral theory would make predictions that “go beyond a mere null model,” especially in the work of Kimura’s student and colleague Tomoko Ohta.¹¹

Hughes spent much of his career testing these predictions.

Kimura first proposed the neutral theory because he thought selection could not explain the speed of evolution observed in animal proteins. In his first paper on the subject, he compared the amino acid sequences of hemoglobin protein molecules from several species and estimated that one amino acid substitution had occurred in every 1.8 years of mammalian evolution. This was shockingly fast. Imagine a mutation conferring some trait so advantageous that it allows its carrier’s offspring to replace the entire population in under two years. For organisms such as human beings, such change is biologically unfeasible, in part because it would require profound differences in reproduction rates that exceed the physiological limits of our species.¹²

The real picture is even worse. Only a small fraction of an animal’s DNA actually encodes proteins—linear chains whose building blocks are small molecules known as amino acids. Hemoglobin is one example. Twenty amino acids are used by biological organisms. Some of these can be encoded by several different DNA triplets, known as codons, making the genetic code partially redundant. Only some DNA changes will result in a change to a protein. For this reason, DNA sequences are said to contain synonymous and non-synonymous sites.¹³ At non-synonymous DNA sites, a change in the nucleotide changes the encoded amino acid. For example, changing G to A in ATG will result in ATA, switching the amino acid from methionine to isoleucine.¹⁴ On the other hand, changes at synonymous sites will not result in an amino acid change. Switching G in CTG to any other nucleotide will not alter the amino acid—CTA, CTG, CTC, and CTT all encode leucine.¹⁵

Because synonymous mutations do not change an amino acid, the resulting protein variant usually remains invisible to natural selection.¹⁶ Synonymous changes are free to accumulate by genetic drift. On the other hand, since the proteins encoded by DNA usually work well, non-synonymous mutations tend to disrupt their function and lower the reproductive rate of their carriers. The result is negative or purifying selection, which acts to eliminate harmful mutations from a population. Such deleterious mutations are quite common.¹⁷ Although they are usually deleterious, non-synonymous mutations may sometimes be actually beneficial, meaning that the lucky organism contributes an increased number of offspring to future generations. This is positive, or Darwinian, selection. Positive selection may then drive the evolutionary substitution of a beneficial mutation until it reaches fixation. Because beneficial mutations are rare, positive selection is rare as well.

It is possible to detect selection by comparing evolutionary rates at non-synonymous and synonymous sites. About 75% of the sites in a typical protein-coding DNA sequence are non-synonymous.¹⁸ This means that about three times as many will change the amino acid as not. Simply tallying the numbers of each type of change is not a fair comparison. One must divide the actual number of changes that have occurred by the possible number of changes of each type.

Imagine two DNA sequences from different individuals. Let n_N be the average number of non-synonymous sites in the sequences, n_S the average number of synonymous sites in the sequences, and m_N and m_S the number of non-synonymous and synonymous mutational differences between them. In this case, the corrected number of non-synonymous and synonymous differences (d_N and d_S, respectively) are:¹⁹

$d_{\mathrm{N}}=\frac{m_{\mathrm{N}}}{n_{\mathrm{N}}}$

and

$d_{\mathrm{S}}=\frac{m_{\mathrm{S}}}{n_{\mathrm{S}}} .$

Consider an example: if there are 100 sites in each of two sequences, an average of 75 might be non-synonymous, leaving 25 that are synonymous. If we observe three non-synonymous differences and one synonymous difference between them, d_N = 3/75 = 0.04 and d_S = 1/25 = 0.04. Because d_N = d_S, the rates of evolution at non-synonymous and synonymous sites are in fact equal. In this case, the evidence suggests that evolution has been dominated by genetic drift, since selection did not change d_N—the rate of evolution at non-synonymous sites—relative to d_S. In other cases, purifying selection can lead to d_N < d_S by acting against non-synonymous mutations that disrupt protein structure. Finally, positive selection can lead to the opposite pattern of d_N > d_S by promoting multiple non-synonymous changes.

Hughes and his doctoral student Meredith Yeager recognized that, because Kimura’s estimate of the rate of evolutionary substitutions was based on amino acid sequences alone, his analysis missed synonymous DNA changes.²⁰ When they estimated evolutionary rates by comparing mouse and rat genomes, their results suggested a much faster rate of 8.14 substitutions per year, or one substitution every 44 days in protein-coding regions of DNA.²¹ Even allowing that the rate of evolution may be up to 1.4 times higher in rodents than in humans and monkeys, this is a rate far too swift to be explained by positive selection.²²

These facts are explained by the neutral theory.

Huge numbers of selectively neutral mutations, the neutral theory insists, coexist in a population at any given time. The billions of alleles in a human genome are linked in a specific order. When mutations occur, they are scattered throughout the DNA, but some will be quite close together. Now imagine that, by genetic drift, one allele reproduces enough to replace another in the population. When it does so, any mutations in its neighboring regions will tag along. This is a strategy that allows neutral changes to accumulate very rapidly, but it is one not generally available to beneficial mutations.

In a population of size N, there are 2N copies of DNA, with each organism possessing one copy from both parents. Let K be the rate at which evolutionary substitutions occur. Now suppose that a new neutral mutation arises. This new mutation will be present in just one copy in one individual, at a frequency of 1/(2N). Since all copies have an equal chance of undergoing substitution, 1/(2N) is also the new mutation’s probability of being the eventual winner—the one that fixes. If the number of mutations occurring in each DNA copy per generation is u, then a population consisting of 2N copies will acquire 2Nu new mutations in every generation.

This leads to a remarkable result for the rate of substitution, and a hallmark of the neutral theory:

$K=2Nu \times \frac{1}{2N}=u .$

The 2Ns cancel and we are left with the mutation rate alone. Thus, the rate of substitution under neutral evolution is equal to the mutation rate u, and does not depend whatsoever on the population size N.²³

Hughes described the situation as follows.²⁴ If u_T is the total number of mutations that occur each generation, and f₀ is the fraction (ranging from 0 to 100%) of those mutations that are neutral, then the rate K₀ at which neutral mutations complete an evolutionary substitution each generation is

$K_0=u_{\mathrm{T}}{f}_0 .$

Suppose that each individual experiences u_T = 100 mutations per generation, f₀ = 90% of which are neutral. In this case, 100 × 90% = 90 neutral mutations will occur every generation, and the same number will complete the process of substitution by reaching fixation.

Since mutations at non-synonymous sites are more likely to disrupt function, f₀ is closer to 0% at these sites. On the other hand, f₀ will be nearer to 100% at synonymous sites, where mutations are much more likely to be neutral.

This leads to an important prediction of the neutral theory: because f₀ is lower at non-synonymous sites, non-synonymous evolution should generally be slower than synonymous evolution.²⁵ This is another way of saying that d_N should be less than d_S, or that purifying selection is more common than positive selection.

Few contributed as much as Hughes in showing that, for almost all genes of all species, d_N is a great deal lower than d_S.²⁶ In a comparison of 42 genes shared by mice and rats, he calculated that d_N was about 1/3 the value of d_S.²⁷ In a 2003 study of the human genome, he demonstrated not only that d_N was less than d_S, but also that d_N decreases for sites having more extreme amino acid changes.²⁸ It is entirely possible for d_N > d_S, but this occurs only on rare, though important, occasions. Purifying selection also constrains the evolution of non-protein-coding DNA in many species²⁹ and even dominates most regions of human immune genes, otherwise known as excellent examples of positive selection.³⁰

Hughes tested various other predictions of the neutral theory as well. Using Fumio Tajima’s D statistic, he showed that purifying selection dominates the evolution of bacteria, verifying the neutral theory's prediction that slightly deleterious mutations will be widespread in large populations.³¹

The ubiquity of purifying selection also revealed something important about how functionally important gene regions behave over the course of evolution; they rarely change. In bioinformatics, the alignment between DNA and protein sequences of different species is routinely used to identify regions with high levels of similarity between species. Sequences have been preserved between species because they play important functional roles, and purifying selection has eliminated many mutations disrupting them.³² Researchers are able to predict gene function and infer protein structure based on sequence similarity alone. This stands in sharp contrast to the original Darwinian view, which held that most evolutionary change is driven by positive selection. Were that so, most evolutionary change would occur at functionally important sites in the genome—the only sites capable of incurring mutations that alter fitness—and such sites would instead be the least similar between species.

In 2013, the ENCODE (Encyclopedia of DNA Elements) Project published results suggesting that eighty per cent of the human genome serves some function. This was considered a rebuttal to the widely held view that a large part of the genome was junk, debris collected over the course of evolution. Hughes sided with his friend Dan Graur in rejecting this point of view. Their argument was simple. Only ten per cent of the human genome shows signs of purifying selection, as opposed to neutrality.³³

When purifying selection on a protein is relaxed, mutations accumulate by drift at all sites, leading to d_N = d_S. It may seem surprising that proteins can tolerate amino acid changes, but data show a strong correlation between the mutation rate and the rate of non-synonymous substitutions.³⁴ A great many of these changes, it would seem, roughly preserve functionality. Going further, if positive selection intervenes to favor multiple amino acid substitutions in a protein region, this can even lead to d_N > d_S. Although such selection is “a relative rarity,” Hughes wrote, it is “of great interest, precisely because it represents a departure from the norm.”³⁵

Natural Selection and Immune Genes

Hughes was the world’s expert on the role of positive selection in shaping the human leukocyte antigen (HLA) genes—the genes of the immune system.³⁶ Encoding the major histocompatibility complex (MHC) receptors on the surface of cells,³⁷ they have long been known as the most polymorphic genes in vertebrates, with variability of around eighty per cent.³⁸ It was not always clear why. Some researchers suggested that there might be a higher mutation rate in the HLA genes.³⁹ Struck by the facts of self-incompatibility genes in plants, other researchers thought that maternal antibodies against fetal MHC molecules in humans benefited offspring by promoting genetic diversity.⁴⁰ The tendency in mice to choose mates with different MHC receptors has seemed a pertinent fact to some scientists.⁴¹ Hughes had another answer.

In all cells that have a nucleus, MHC molecules bind to peptide fragments randomly sliced from a sample of the proteins present inside the cell. Once a fragment is MHC-bound, it is transported to the surface of the cell, where passing immune system cells recognize and bind the MHC-peptide complex.⁴² If the peptide-presenting cell happens to be infected with a pathogen such as a virus, the peptide fragment it presents may have originated from the foreign invader. In that case, the immune cell may, after binding the MHC-peptide complex, initiate the destruction of the infected host cell.

In 1974, Rolf Zinkernagel and Peter Doherty demonstrated that MHC molecules made from different HLA alleles differ in the peptide fragments they can bind.⁴³ They went on to suggest that the high levels of genetic variability observed at the HLA loci may result from an evolutionary phenomenon called heterozygote advantage.⁴⁴ Heterozygotes are organisms whose two copies of a gene differ in sequence; they possess two different alleles. Instead of driving a beneficial mutation to fixation, overdominant selection promotes variation. In sickle-cell anemia, heterozygotes are able to resist malaria. In the case of HLA, organisms having multiple HLA alleles bind a wider range of peptide fragments, making them better able to fight infections.

A few years earlier, Takeo Maruyama and Masatoshi Nei had published theoretical work predicting that heterozygote advantage should speed up the rate of amino acid substitution at the sites under selection.⁴⁵ Hughes and Nei then predicted that the non-synonymous substitution rate should be increased in codons for the MHC’s peptide-binding region (PBR).⁴⁶ This would lead to an increase in d_N at just those sites, but not in d_S (since synonymous mutations do not change the amino acid), resulting in d_N > d_S. Given the recently published structure of the MHC, Hughes and Nei knew right where to look, and they had twelve DNA sequences with which to do it.

The prediction of heterozygote advantage was spectacularly vindicated. Every single comparison between DNA sequences showed that d_N > d_S in the PBR codons. Outside PBR codons, almost all comparisons exhibited the opposite pattern, and purifying selection reigned. Together with Tatsuya Ota, Hughes and Nei showed that these non-synonymous changes are disproportionately concentrated in just those twenty-nine amino acids of the PBR that are most likely to be involved in peptide-binding. Changes altering the electric charge of the amino acid are the ones most favored.⁴⁷ The more sequence data that poured in, the clearer these trends became.⁴⁸

These results falsified some of the competing hypotheses. If the gene regions encoding the PBR truly experienced a higher mutation rate, an increase should have been observed in d_S as well as d_N. But it turned out that d_S was no higher in the PBR than the rest of the genome. Hughes also rejected the idea that animals choose mates with different MHC molecules, earlier research with mice notwithstanding. Heterozygote advantage is the best explanation for the uniquely high levels of variability in PBR.

Hughes later asked the same question from the point of view of the infecting pathogen. Because viruses and the hosts they infect are constantly competing to identify and evade each other, selection produces multiple non-synonymous changes in this case as well. Hughes and his collaborators focused especially on malaria, human immunodeficiency virus (HIV), and simian immunodeficiency virus (SIV).⁴⁹ More recently, he and others developed an approach to perform this sort of analysis on the latest viral genetic data.⁵⁰ They were building on earlier theoretical work by Nei and Takashi Gojobori. Although it is too early to tell how useful this approach will be for vaccine development, studies with influenza, HIV, SIV, and various monkey viruses have already yielded valuable insights into how viruses evolve within their hosts.⁵¹

The methods used by Hughes and Nei soon took on a life of their own. A host of researchers set out to discover Darwinism at the nucleotide level. They were apparently unaware that d_N > d_S was a prediction only in cases of heterozygote advantage.⁵² Their approach was utterly misguided.

Hughes and Nei knew what to look for. The inequality d_N > d_S does not imply heterozygote advantage, let alone other forms of positive selection.⁵³ It can occur just by chance.⁵⁴ In fact, d_N > d_S is caused by positive selection only in the rarest of cases. In a typical account of positive selection, a new beneficial mutation arises, increases in frequency, and goes on to fixation.⁵⁵ Before the beneficial mutation occurs, there is no signature of positive selection; after fixation, there remains no signature. All diversity is purged when an evolutionary substitution reaches completion. There’s only a brief window of time during which one might observe that d_N > d_S.

Dismayed at the “vast outpouring of pseudo-Darwinian hype,” Hughes did much to correct this trend in the literature.⁵⁶ In the provocatively-titled “Looking for Darwin in All the Wrong Places: The Misguided Quest for Positive Selection at the Nucleotide Level,”⁵⁷ Hughes argued not only that many of the supposed cases of positive selection were incorrect, but further that they identified the relaxation of purifying selection. If most mutations in protein-coding DNA are slightly deleterious, then purifying selection will depress their frequencies.⁵⁸ Purifying selection having been suspended, as occurs when populations become small or form a new species, slightly deleterious mutations increase in frequency or fix by chance alone, increasing d_N.⁵⁹

The importance of this point cannot be overstated.

When researchers identify DNA regions that they think are under positive selection, they often infer that these regions are important to the function of the organism. The gene for microcephalin in the human brain is an example.⁶⁰ Medical research might then proceed on these grounds. If these cases actually result from the relaxation of purifying selection, the opposite is true; these are unimportant regions where slightly deleterious mutations flourish. Functionally-important regions are those with a very low frequency of non-synonymous mutations. They have a low frequency because purifying selection is acting to conserve the original function.⁶¹ It was for this reason that Hughes described the vast majority of studies identifying positive selection as worthless.⁶²

His concerns have been vindicated. The Graur group has shown that misapplying d_N vs. d_S falsely identifies gene regions that have sequencing errors or misalignments.⁶³ Other studies have shown that these methods consistently miss those codons that are already known to be under positive selection.⁶⁴ One striking example involves a study of a protein used in vision, in which all functional differences were determined experimentally and mapped to twelve codons. None of these important codons were identified by methods used for detecting selection; experimentally-induced changes in the eight codons that were identified by these methods had no effect on the protein’s function.⁶⁵

Other studies typically operate by computing some measure of genetic variation and then identifying as positively selected those genes having the very highest values, say, the top 1%.⁶⁶ Hughes would frequently compare this approach to lining up all humans by height and declaring that the top one percent are Martians.⁶⁷ He concluded the matter thus:

The so-called “codon-based” methods of testing for positive selection are derived from an unwarranted generalization of the MHC case. … Contrary to a widespread impression, natural selection does not leave any unambiguous “signature” on the genome, certainly not one that is still detectable after tens or hundreds of millions of years. To biologists schooled in Neo-Darwinian thought processes, it is virtually axiomatic that any adaptive change must have been fixed as a result of natural selection. But it is important to remember that reality can be more complicated than simplistic textbook scenarios.⁶⁸

That being said, positive selection does sometimes fix adaptations. Hughes led the effort to chronicle well-established cases in Adaptive Evolution of Genes and Genomes.⁶⁹ He noted that most recent examples involve loss-of-function mutations. It is relatively easy to damage unneeded proteins.⁷⁰ The number of convincing cases has barely increased since the 1970s.⁷¹

Hughes also explored the role of natural selection in the evolution of altruism. Evolution and Human Kinship is his most mathematical work.⁷² It is also a philosophical work, and contains a discussion of the scientific method, adaptation, and the distinction between social anthropology and sociobiology.⁷³ The book celebrates evolutionary theory as an instrument capable of addressing the social sciences.

In the 1960s, William D. Hamilton argued that altruism could be efficiently explained in purely Darwinian terms if the altruist’s costs were outweighed by recipient’s benefits when discounted by a measure of genetic similarity. Hughes goes further than Hamilton in predicting that individuals will be more likely to forgive one another when their genetic relatedness and the likelihood of future reciprocity (based on reputation) are high; on the other hand, a scarcity of resources will prevent even close kin from helping one another.⁷⁴ Hughes also predicted that community leaders will tend to be bien branché.⁷⁵ Although none of these concepts is surprising, Hughes did formulate them mathematically.

The hypothesis of natural selection makes the tacit but critical assumption that mutations affecting a trait of interest actually exist in the population.⁷⁶ For example, unless an appropriate escape mutation actually occurs in an HIV virus, it cannot evade the host’s immune system. If the variation that causes selfless behavior never arises, of what use selection? Even in adaptive evolution, mutation is king.⁷⁷ Hughes pioneered several ideas about how such evolution might work.

Adaptive Evolution by Other Means

Hughes was skeptical that positive selection can successfully explain much of adaptive evolution. His incredulity ran deeper than many may realize. In an interview with Heredity about positive selection, he declared that “there really isn’t all that much evidence that it actually happens to the extent to which it would be needed to explain all of the adaptive traits of organisms.”⁷⁸

What were his reservations about positive selection? For one thing, important genes do not change much. But there was a deeper reason. In a 2004 study with Robert Friedman, Hughes used evolutionary tree-building techniques to determine the relationships between worms, insects, fish, and humans based on the presence or absence of related genes.⁷⁹ The results were surprising. Instead of suggesting the gradual addition of new genes in different lineages, the trees implied that the ancestor of all these animals had the ultimate gene set, which was then whittled away as different animals evolved.

The primary mode of evolution is gene loss, not gene gain.

This finding led to a new idea about adaptive evolution.⁸⁰ Hughes called the mechanism plasticity-relaxation-mutation (PRM). We know that purifying selection is ubiquitous, and that this selection is sometimes relaxed, allowing the fixation of slightly deleterious mutations. We also know that plasticity—the ability of organisms to change their phenotype without changing their genes—can play a role in evolution.⁸¹ When these concepts are combined, it is possible to account for the evolutionary substitution of adaptive traits without selection.

Imagine a species able to change its behavior to avoid two different predators. Any mutation disrupting the ability to avoid predator 1 or predator 2 would likely be eliminated by purifying selection, thus maintaining plasticity. Suppose predator 2 disappears from the environment. Purifying selection would no longer maintain the response it elicits. Mutations would accumulate in the gene encoding the unused response to predator 2, deactivating it, and drift might then fix these new adaptive mutations. The result would be the fixation of only one gene state: the ability to evade predator 1.

Hughes noted evidence for PRM in the evolutionary literature, and argued that this non-Darwinian mechanism may play a role even in the adaptive evolution of such celebrated examples as that of the Galápagos Islands finch beak.⁸² His work also emphasized an important distinction not always appreciated by evolutionary biologists; knowing that a trait is adaptive is not the same thing as knowing how it arose.⁸³

Hughes also produced important work on the role of gene duplication in evolution. Beginning in the late 1960s, Susumu Ohno proposed that the duplication of entire genes plays a central role in the evolution of new protein functions.⁸⁴ A gene duplicates, but having both copies is redundant. Purifying selection is therefore relaxed, allowing one of the copies to mutate and lose its function.⁸⁵ In some cases, a random set of mutations might occur which produces a new function.

Hughes might have described this as nonadaptive storytelling.⁸⁶ The Darwinian explanation had been able to account for seemingly designed aspects of biological organisms by appealing to material causes that went beyond blind chance.⁸⁷ Work with frogs uncovered convincing evidence that duplicated gene copies are, in fact, usually not freed from purifying selection.⁸⁸

This led Hughes to advocate a radically different perspective.⁸⁹ New functions might be present before and not after gene duplication. A single gene in ducks encodes both an eye protein (crystalline) and a glycolysis enzyme (enolase).⁹⁰ The gene is duplicated. Mutations might then disable one of the two functions in each gene copy. If the two functions are A and B, then function B might be disabled in gene copy 1, and function A in gene copy 2. There are now two genes with two different functions. Call this subfunctionalization.

Positive selection might then improve each function. Since the original gene had been forced to maintain two functions concurrently, it may not have been able to maximize either. Freed from bifunctionality, it can.⁹¹ Either positive selection or genetic drift could fix these changes.

A great deal of evidence supports this scenario. One example from colobine monkeys involves two genes that enable the digestion of bacterial genomes as a source of nitrogen.⁹² A recent duplication event seems to have been followed by a change of nine amino acids between the two genes. Experimental analyses suggest that these differences have allowed one copy to become specialized for digestion in the small intestine, while the other copy is specialized for the pancreas.⁹³

Some evolutionary biologists objected to Hughes’s radical views on adaptive evolution.⁹⁴ Surely the genome could not accumulate genetic information only by losing it; one cannot build up a savings account by constantly withdrawing.

It is my view that Hughes considered this an interesting question, but not essentially a scientific one. Science should be concerned with theories that are empirically adequate, that is, theories that can explain our observations and make accurate predictions.⁹⁵

His hypotheses for adaptive evolution did just that.