Neutral Theory

"The theory does not deny the role of natural selection in determining the course of adaptive evolution, but it assumes that only a minute fraction of DNA changes in evolution are adaptive in nature" — Motoo Kimura, The Neutral Theory of Molecular Evolution, 1983¹

"we have reached this peculiar situation where the analysis of amino acid substitution rates has both led to a clear rejection of a major claim of the neutral theory—that most observable changes are neutral—and at the same time indisputably demonstrated the importance of drift in protein evolution." — Nicolas Galtier, Genome Biology and Evolution, 2024²

Summary

The neutral theory of molecular evolution says that most of the DNA differences between species were not chosen by natural selection.  They are mutations that made no difference to survival, and they spread through a population by chance, the process called genetic drift.  Motoo Kimura proposed it in 1968, and Jack King and Thomas Jukes independently the next year.^{3 4}  This article asks whether it is true, and what follows for evolutionary biology and for the young-earth and design arguments that use it.  The formulas behind it are in Population Genetics;  Kimura's answer to Haldane's cost of selection is in Haldane's Dilemma.

1. The theory is a claim about most changes, not all of them. Kimura's own statement is that "only a minute fraction of DNA changes in evolution are adaptive," and that the theory "does not deny the role of natural selection in determining the course of adaptive evolution."¹ Its modern defenders add that most mutations that do matter are harmful, and selection removes them.⁵
2. The founding argument was wrong, and both sides now say so. Kimura's 1968 case was that DNA changes too fast for selection to pay for it. He got that rate by spreading the protein rate over all four billion letters of a mammal genome. Andrew Kern and Matthew Hahn, who oppose the theory, write that this overstates the protein-coding sites "by two orders of magnitude,"⁶ and Masatoshi Nei, who defended it, made the same correction in 2005.⁷
3. Across the whole human genome, the mainstream's own numbers agree with the theory. About 15 million single-letter changes separate the human line from the human–chimpanzee ancestor.⁸ The best recent estimate of the adaptive protein changes among them is about 4,000, with a range of 2,900 to 5,100.⁹ That is about 0.03% of the total, and the most generous figures in the literature raise it to about 0.4%. This site's Functional DNA article says over 99.98% of mammal differences spread by chance. The mainstream count gives about 99.97% at its central estimate and 99.6% at its most generous, so that page's figure is right in substance and slightly high in its last digit.⁸
4. Inside the proteins of species with large populations, the theory's central claim has failed. About half the amino-acid changes between fruit-fly species look adaptive,¹⁰ so do more than half in enteric bacteria¹¹ and wild mice,¹² and across 44 animal species the median is 0.65.¹³ Humans sit at 0 to about 25%,^{14 15} and most plants near zero.¹⁶ Nicolas Galtier's 2024 review calls this "a clear rejection of a major claim of the neutral theory," and in the same sentence says drift's importance in protein evolution has been "indisputably demonstrated."²
5. The nearly neutral version passes a test it set itself. Tomoko Ohta predicted in 1973 that slightly harmful mutations would fix more often in small populations.¹⁷ The adaptive share does track population size, mainly because large populations weed out more of the slightly harmful changes.¹⁸ Relative to silent change, protein change runs at about 0.20 on the human and chimpanzee branches against 0.13 on the mouse and rat branches, and the genome authors put the excess down to relaxed constraint, not adaptation.¹⁹
6. Lewontin's paradox is real and unsolved, and it is not a junk-DNA question. Under strict neutrality, genetic diversity should rise with population size. Across 172 animal taxa, population sizes span about twelve orders of magnitude and diversity about three.²⁰ Selection at linked sites removes some diversity, about 17% in humans,²¹ but by its defenders' own arithmetic not nearly enough.²² The leading explanation is that large populations are still far from equilibrium after past growth.²² A young earth is the extreme case of that, so the paradox is no evidence against one, nor for one.
7. The 2018 challenge is not a junk-DNA argument. Kern and Hahn say outright that the large nonfunctional fraction of eukaryotic genomes is not in dispute.⁶ Their case rests on proteins and on linked selection, and the reply from seven senior population geneticists defends the theory on the same ground.⁵
8. But the genome-wide form of the theory does lean on junk. The defenders answer the fruit-fly figures by pointing out that they concern "only the small fraction of the genome that codes for proteins."⁵ Ohta built her theory on the premise that "most genomic DNA of higher organisms can freely accumulate base substitutions."²³ Kimura's showcase for the theory's top speed was pseudogenes, which he regarded "as indeed dead."¹ Today the premise is measured, not assumed: estimates of the human genome under purifying selection run from 3.5% to about 26%.^{24 25 26 27} If far more of it had a function that depends on its exact letters, the 15 million human-branch changes could not mostly be neutral, and no evolutionary account could easily explain them, as the mainstream's Dan Graur concedes.²⁸
9. Young-earth writers mostly use the theory rather than refute it. John Sanford's genetic entropy argument needs Ohta's slightly harmful mutations to drift to fixation, and Robert Carter calls it an idea that "leverages the beliefs and language of the neutralist position."²⁹ That premise is measured: a quarter to a third of new amino-acid mutations in humans are too weakly harmful for selection to see,³⁰ and Kimura himself wrote that such mutations "accumulate continuously in every species," though he judged the loss tiny and easily offset.³¹ How much of the genome lies in that zone is the real dispute. Carter dates Adam and Eve with the theory's rate result,³² and Walter ReMine concedes that "the vast majority of substitutions would be neutral."³³ Their claim that the theory requires junk DNA is right about its genome-wide form and wrong as a statement of the theory, which allows function without sequence-specific constraint.¹ On a young earth, neutral substitution at the measured rate changes about three sites in a million per lineage,⁸ so differences between created kinds are not fixed mutations at all; within kinds, drift sorts created variety fast enough only in small populations.

Verdict.  Most DNA differences between mammal species are neutral or nearly so, by the mainstream's own count, and that conclusion rests mainly on the measured finding that most of the genome is not under constraint.  Inside the proteins of species with large populations the original theory has failed.  For the design argument, the theory is a real concession that selection did not build most differences, but it becomes evidence against evolution only if a large, sequence-specific function is shown in the DNA that differs, and the data within living humans do not yet show one.  For the genetic-entropy argument the theory supplies the premise rather than the refutation:  Kimura conceded that effectively neutral mutations are slightly harmful and accumulate, and the open question is how much of the genome they touch.

Part 1: What the neutral theory claims

Kimura's 1968 argument

Motoo Kimura was a Japanese population geneticist who had spent the 1950s and 1960s working out the mathematics of chance in small populations.  In February 1968 he published a three-page letter in Nature whose subtitle states the whole idea:  "Calculating the rate of evolution in terms of nucleotide substitutions seems to give a value so high that many of the mutations involved must be neutral ones."³

His reasoning ran in three steps.  First, he averaged the rates at which three proteins had changed across mammals:  haemoglobin, cytochrome c and triosephosphate dehydrogenase.  Second, he spread that rate over the whole genome, which he took to be four billion DNA letters, and found that a mammal lineage swaps one letter about every two years.  Third, he compared that with J. B. S. Haldane's 1957 estimate that selection can afford to replace one gene about every 300 generations.  The gap was several hundredfold, so he concluded that the replacements could not all have been paid for by selection:  "the very high rate of nucleotide substitution which I have calculated can only be reconciled with the limit set by the substitutional load by assuming that most mutations produced by nucleotide replacement are almost neutral in natural selection."³

The same letter contains the theory's most useful result.  For a mutation that makes no difference to fitness, the chance that it eventually spreads to the whole population equals its starting frequency.  The number of new neutral mutations each generation is proportional to population size, and each one's chance of spreading is inversely proportional to it, so population size cancels out.  In Kimura's words, "new alleles may be produced at the same rate per individual as they are substituted in the population in evolution."³  The rate at which neutral changes accumulate between species equals the mutation rate.  That is the basis of the molecular clock, which has its own article.  The derivation is in Population Genetics.

King and Jukes, 1969

A year later two American biochemists, Jack Lester King and Thomas Jukes, reached the same conclusion from the chemistry of proteins and published it in Science as "Non-Darwinian Evolution."  Their summary line is the one most quoted:  "Natural selection is the editor, rather than the composer, of the genetic message."⁴  They made two observations that matter for this article.  At the cytochrome c gene, "at least 90 percent of all substitutional mutations ... are harmful and are rejected by natural selection,"⁴ so the theory never said that most mutations are harmless, only that most of the ones that survive to become differences are.  And they argued that "not much more than 1 percent of mammalian DNA codes for proteins,"⁴ so most of the genome had no protein to protect.

The theory in its mature form

Kimura's 1983 book, The Neutral Theory of Molecular Evolution, states the claim with its limits:

The neutral theory asserts that the great majority of evolutionary changes at the molecular level, as revealed by comparative studies of protein and DNA sequences, are caused not by Darwinian selection but by random drift of selectively neutral or nearly neutral mutants.  The theory does not deny the role of natural selection in determining the course of adaptive evolution, but it assumes that only a minute fraction of DNA changes in evolution are adaptive in nature, while the great majority of phenotypically silent molecular substitutions exert no significant influence on survival and reproduction and drift randomly through the species.¹

Two things follow from that paragraph.  The theory is about molecules, not about eyes or wings;  Kimura left the shaping of bodies to selection.  And "neutral" does not mean "unconstrained."  Kimura wrote that "the existence of selective constraint itself does not contradict the neutral theory."¹  A gene can be under heavy constraint, with most of its possible mutations rejected, and still have most of its accepted changes be neutral ones.

The book's main evidence was a pattern Kimura stated as a rule:  "Functionally less important molecules or parts of molecules evolve (in terms of mutant substitutions) faster than more important ones."¹  The surface of haemoglobin changes faster than its oxygen pocket.  The middle piece of proinsulin, which is cut away before insulin works, changes several times faster than insulin itself.  Synonymous sites, where a letter change does not alter the amino acid, change faster still.  And the fastest of all were pseudogenes, broken copies of genes, which Kimura regarded "as indeed dead."¹  Kimura argued that a selection-driven view gives no reason to expect this order, while the neutral view requires it:  the less a part matters, the larger the share of its mutations that do no harm, and so the more of them drift through.¹  The data matched that order.

Ohta's nearly neutral theory

Tomoko Ohta, Kimura's colleague at the National Institute of Genetics in Japan, extended the theory in 1973 to mutations that are slightly harmful:  "The concept of neutral mutant substitution in the population by random genetic drift can be extended to include random fixation of very slightly deleterious mutations which have more chance of being selected against than of being selected for.  If this class of mutant substitution is important, we can predict that the evolution is rapid in small populations or at the time of speciation."¹⁷

The reason is that selection's grip depends on population size.  A mutation that lowers fitness by one part in a hundred thousand is weeded out efficiently in a population of a million, but in a population of ten thousand it behaves almost as if it were neutral and can drift to fixation.  The threshold, the "drift barrier," is roughly one over twice the effective population size;  Population Genetics explains why it is a soft shoulder rather than a wall.  Ohta's 1992 review put her thesis this way:  "I propose that a substantial fraction of mutant substitutions at the molecular level are caused by random fixation of very slightly deleterious mutations."²³  She divided the genome in two.  "The fundamental idea is that most genomic DNA of higher organisms can freely accumulate base substitutions, i.e. most new mutations are selectively neutral.  On the other hand, amino acid substitutions are more likely to be influenced by natural selection, i.e. many of them may be regarded as nearly neutral, or very slightly deleterious."²³  That two-tier picture comes back in Part 5.

The modern statement

The fullest recent statement is a 2019 reply by seven senior population geneticists, Jeffrey Jensen, Bret Payseur, Wolfgang Stephan, Charles Aquadro, Michael Lynch, Deborah Charlesworth and Brian Charlesworth.  They define the theory as the claim that "most de novo mutations are either sufficiently deleterious in their effects on fitness that they have little chance of becoming fixed in the population, or are under such weak selection that they may become fixed as a result of genetic drift."⁵  Their five-point "modern version" opens with two claims:  "A large fraction of the genome of organisms studied to date is subject to mutations that are effectively neutral with respect to their fitness effects, and hence evolve under genetic drift,"⁵ and "The great majority of newly arising mutations that do affect fitness (i.e., non-neutral mutations) are deleterious, and the predominant mode of natural selection is purifying in nature, removing these deleterious mutations from populations."⁵

So the theory makes five checkable predictions:

1. Neutral changes accumulate between species at the mutation rate, whatever the population size.
2. Parts of the genome under less constraint change faster.
3. Within a species, neutral genetic diversity should scale with population size times mutation rate.
4. Polymorphism within species and divergence between species are two stages of one process, so the two should be in proportion.
5. On the nearly neutral version, slightly harmful changes accumulate faster in species with small populations.

Prediction 1 is the molecular clock, and it holds only roughly.  If neutral substitutions arrived at random, like the clicks of a Geiger counter, the variance in their number across lineages would equal the mean, an index of dispersion of 1.  Trevor Bedford, Ilan Wapinski and Daniel Hartl measured mean indexes of 2.1 across 3,081 yeast proteins, 4.2 across 7,174 fruit-fly proteins and 6.5 across 8,065 mammal proteins.³⁴  The clock is least regular where populations are smallest, and the authors add that this "does not necessarily imply positive selection or adaptive evolution;  overdispersion is compatible with nearly neutral evolution occurring in a heterogeneous fashion."³⁴  Rates measured over a few generations also run faster than rates measured over millions of years.  One of the mainstream's proposed causes is the slightly harmful class of mutations described in Part 3, counted in families before selection removes them.  Molecular Clocks works through both departures, and through the young-earth clocks of Robert Carter and Nathaniel Jeanson that rest on the short-term rate.  Prediction 4 is what the McDonald–Kreitman test checks, in Part 3.  Prediction 3 is Lewontin's paradox, in Part 4.

The founding argument has not survived, on either side's account

Kimura's 1968 calculation needed every one of four billion letters to be a possible target of selection.  Kern and Hahn, the theory's leading critics, write that "Kimura overestimates the number of protein-coding sites in the genome by two orders of magnitude (he uses 4×10⁹ bp, whereas the real number is closer to 3×10⁷ bp in the largest genomes)," and that fixing this number alone removes the conflict with Haldane.⁶  Masatoshi Nei, who defended both Haldane's cost and the neutral theory, made the same correction in 2005:  counted per gene, the substitution rate falls far below Haldane's limit, and counted per amino acid it is "about two times higher," not hundreds of times.⁷  Haldane's Dilemma has the full history.  The theory's second original pillar, that proteins change at a steady rate in every lineage, Kern and Hahn call "an illusion" once more proteins were sequenced.⁶  Whether the theory stands today depends on the later evidence, which is the subject of the rest of this article.

Part 2: The 2018 challenge and the reply

Kern and Hahn: "overwhelmingly rejected"

For the theory's fiftieth anniversary, Molecular Biology and Evolution asked Andrew Kern of Rutgers University and Matthew Hahn of Indiana University how it had fared.  Their answer, in the paper's second sentence, was "In a word, poorly."⁶  Their abstract:  "We argue that the neutral theory was supported by unreliable theoretical and empirical evidence from the beginning, and that in light of modern, genome-scale data, we can firmly reject its universality."⁶

They began by stating what they were not disputing.  The neutral theory, they wrote, "is not simply a statement about the presence of neutral mutations, nor about the large fraction of eukaryotic genomes that are nonfunctional—neither of these assertions would be contested by current competing hypotheses."⁶  What they rejected was a stronger thesis:  "that differences between species are due to neutral substitutions (not adaptive evolution), and that polymorphisms within species are not only neutral but also have dynamics dominated by mutation-drift equilibrium."⁶

Their case has four parts.

1. The original evidence failed. Kimura's cost argument used the wrong number of sites, and steady protein rates turned out to be uneven, as Part 1 describes.
2. Many amino-acid changes look adaptive. McDonald–Kreitman studies in two fruit-fly species found "∼50% of all amino acid substitutions have been fixed by positive selection," and similar results followed in other species.⁶ They concede the human exception: "In humans, though the overall fraction of amino acid substitutions fixed by positive selection is estimated to be zero," while genes that deal with pathogens show strong signs of adaptation.⁶
3. Selection reaches almost everywhere through linkage. Across 40 genomes, regions where chromosomes recombine less carry less diversity. Under strict neutrality recombination should not matter, so they read the pattern as the footprint of selection on nearby sites: "these results imply that almost no loci are free from the effects of selection, in any organism."⁶ Because "An equilibrium between mutation and drift is a central tenet of the neutral theory," they conclude that "current data appear to be fundamentally incompatible with the neutral theory."⁶
4. Adaptation is not limited by mutation. It draws on standing variation, on genes borrowed from other species, and on many genes of small effect at once, none of which leaves the classic signature of a single sweep.

Their verdict:  "we believe that the neutral theory has been overwhelmingly rejected."⁶

Jensen and six colleagues: "abundantly clear" that it is correct

The reply came the following January, in Evolution, from seven population geneticists, among them Michael Lynch and Brian and Deborah Charlesworth, two of the three authors who first described background selection in 1993.³⁵  They called Kern and Hahn's position "probably a minority view" and wrote that "it is now abundantly clear that the foundational ideas presented five decades ago by Kimura and Ohta are indeed correct."⁵

Their answers, point by point:

1. The fruit-fly numbers concern a small part of the genome. "These inferred frequencies of adaptive substitutions mostly concern only the small fraction of the genome that codes for proteins (e.g., <2% of the human genome; see Lander et al. 2001)."⁵ Focusing on fast-evolving genes and then rejecting neutrality is, they write, circular.⁵
2. Most linked selection is purifying, not adaptive. "In an explicit comparison between models of widespread purifying selection on weakly deleterious alleles versus recurrent positive selection on beneficial alleles, Lohmueller et al. (2011) found a much better fit of the former to the observed pattern in humans."⁵ Background selection, the loss of neutral variation that happens to sit beside harmful mutations being weeded out, explains the recombination pattern without adaptation.
3. Linkage does not change the rate of neutral change between species. Neither background selection nor sweeps "affect the probability of fixation of neutral mutations ..., which determines the rate of neutral sequence evolution."⁵ Both lower the effective population size of the affected region, which makes more mutations behave as neutral, not fewer.
4. Without a neutral baseline, stories multiply. "In the absence of an appropriate null model accounting for these processes that are common to the genome as a whole, inappropriate adaptive story-telling will be likely to proliferate."⁵

Where the dispute stands

The two papers mostly disagree about what the theory says.  Kern and Hahn test the strong form:  between-species differences are neutral, and diversity sits at mutation–drift equilibrium.  Jensen and colleagues defend a weaker form:  much of the genome evolves by drift, and most selection removes harmful mutations.  Both can be true at once.  The fruit-fly protein figures falsify the strong form for fruit-fly proteins.  The noncoding bulk of a mammal genome fits the weaker form.

Nicolas Galtier's 2024 review of fifty years of the debate gives the cleanest verdict from inside the field.  The theory's two founding observations, abundant polymorphism and roughly steady protein evolution, "are still very much alive and important aspects of current molecular evolutionary thinking."²  But on proteins, the field has "both led to a clear rejection of a major claim of the neutral theory—that most observable changes are neutral—and at the same time indisputably demonstrated the importance of drift in protein evolution."²  His diagnosis of the founders:  "Enthused by their discovery, they slightly overstated the prevalence of neutrality."²  The question is now one of proportion:  "The issue, however, is no longer phrased in terms of corroborating versus challenging the neutral theory, but rather in terms of quantifying the relative contribution of neutral and selective processes to the observed patterns."²

None of the three papers argues that most of the genome is functional.  That matters for how creationists have read the 2018 exchange, and it is taken up in Part 5.

Part 3: How much of the change is adaptive? The measurements

The McDonald–Kreitman test

In 1991 John McDonald and Martin Kreitman found a way to count adaptive changes directly.  Their reasoning:  "If the observed substitutions are neutral, the ratio of replacement to synonymous fixed differences between species should be the same as the ratio of replacement to synonymous polymorphisms within species."³⁶  Silent changes serve as the neutral yardstick.  If protein-changing differences between species outnumber what the variation within species predicts, the excess is the share that selection drove to fixation.  That share is called alpha (α).  At the Adh gene of fruit flies they found "more fixed replacement differences between species than expected."³⁶

The test has known biases, and they matter for the numbers below.  Slightly harmful protein changes linger as rare variants inside a species but seldom fix, so they inflate the within-species count and push α down, sometimes below zero.  Population growth or shrinkage pushes it either way.  Modern methods filter out rare variants or model the distribution of fitness effects (taken from Beneficial and Harmful Mutations as an input here), and different methods give different answers on the same data.¹⁵

The numbers by taxon

Estimates across species run "from nearly zero ... up to roughly 90%."²  The plain conclusion is that in most animals with large populations, about half or more of the protein changes between related species were driven by selection.  Galtier, after measuring 44 species pairs, wrote that "Observing a high α therefore seems to be the rule, and a low α the exception, making humans and apes a peculiar, intriguing case."¹⁸  For proteins in those species, the neutral theory's founding claim is false.

Humans are the exception, and the count is in the thousands

Human estimates run from zero to about a quarter, depending on method.  Kern and Hahn summarise the human figure as "estimated to be zero," citing Adam Boyko and colleagues among others,⁶ but Boyko's own abstract gives 10–20%.¹⁴  Ying Zhen and colleagues show why the methods disagree:  the plain test gives −0.41 for human against chimpanzee, and the figure rises to 0.25 once the sizes of the outgroup and ancestral populations are modelled.¹⁵

The best recent count is from Lawrence Uricchio, Dmitri Petrov and David Enard.  They counted 29,925 protein-changing differences on the human branch since the split from chimpanzees and estimated α at 0.135.⁹  That is about 4,040 adaptive protein changes, 2,900 to 5,100 on their interval.  Out of about 15 million single-letter changes on the human branch, it is 0.027%;  at Boyko's top figure of 20% it is still only 0.04%.⁸  Haldane's Dilemma compares these counts with the reproductive budget for paying for them.

So for humans, both halves of the theory hold:  most protein changes look neutral or nearly so, and the genome as a whole is overwhelmingly so.  The Haldane's Dilemma article found that the popular replies to young-earth critics concede fewer adaptive changes than this literature now estimates, so the literature's count is the one to use, on either side.

Population size is the pattern, and it is the nearly neutral prediction

The adaptive share is not random across species.  It tracks population size.  Galtier found α "positively correlated to species effective population size," but traced most of the correlation to the other side of the ratio:  it "appears to be primarily driven by a decreased rate of nearly-neutral amino-acid substitution because of more efficient purifying selection in large populations."¹⁸  Within one species of mouse, the subspecies with the largest population has α near 40%, the others about 12% and 13%, and "the higher estimate of a for M. m. castaneus appears to reflect not only more adaptive fixations but also more effective purifying selection."³⁹  Plants mostly show little adaptation, and Gossmann and colleagues suggest it "may be because many plant species have modest effective population sizes."¹⁶

This is what Ohta predicted in 1973.  Slightly harmful changes fix more readily in small populations, so small populations have more non-adaptive protein changes and a lower α.  The chimpanzee genome paper shows the same effect directly.  The protein-changing rate relative to the silent rate is about 0.20 on the human and chimpanzee branches against about 0.13 on the mouse and rat branches, so about 35% of the hominid amino-acid changes are in excess of the rodent pattern.  The authors conclude that relaxed constraint, not extra adaptation, "seems to be the principal explanation."¹⁹  That is why Galtier could write that the same data rejected a major claim of the neutral theory and "indisputably demonstrated the importance of drift."²  The strict theory failed;  the nearly neutral theory passed a test it set in advance.

The slightly harmful class, measured

The nearly neutral theory needs a large class of mutations whose harm is too small for selection to see.  That class has been measured, and it is real.  Adam Eyre-Walker and Peter Keightley's 2007 review starts from a definition:  "it seems unlikely that any mutation is truly neutral in the sense that it has no effect on fitness."  A mutation counts as "effectively neutral" when the product of the effective population size and its selection coefficient, N_es, "is much less than 1," so that its fate "is largely determined by random genetic drift."⁴¹  For hominids, with effective sizes of 10,000 to 30,000, they estimated that fewer than 30% of amino-acid-changing mutations are effectively neutral, which "implies that at least 70% are deleterious."⁴¹  Fits to human variation agree.  Adam Boyko and colleagues put 27–29% of new amino-acid mutations below an effect of 0.01% and 30–42% between 0.01% and 1%.¹⁴  Bernard Kim, Christian Huber and Kirk Lohmueller, with samples of over 1,000 Europeans, found 34–37% below 0.01% and about a quarter below 0.001%, and concluded that "nearly neutral forces play a larger role in human evolution than previously thought."³⁰  Beneficial and Harmful Mutations compares these fits and their limits:  each depends on a model of population history, and most have no beneficial class, so slightly helpful mutations are counted as neutral.

Kimura saw what this class implies.  In the 1979 paper that set out his distribution of mutation effects, he wrote:  "Under the present model, effectively neutral, but, in fact, very slightly deleterious mutants accumulate continuously in every species."  He put the loss at about one part in ten million of fitness per generation, from 10,000 protein-coding genes, and called the threat to a species "a moot point," because "this will easily be taken care of by adaptive gene substitutions that must occur from time to time (say once every few hundred generations)."³¹  In his 1983 book he called the accumulation "essentially equivalent to the deterioration of the environment," which adaptive substitutions must answer "to save the species from extinction."¹  Alexey Kondrashov extended the arithmetic to the whole genome in 1995.  Mutations whose harm lies between about one over the genome size and one over four times the effective size are "neutral enough to accumulate almost freely, but are still deleterious enough to make an impact at the level of the whole genome."  If a tenth of all sites were in that range, an average person would carry about 100 lethal equivalents.  Unlike Kimura, Kondrashov wrote that "This paradox cannot be resolved by invoking beneficial mutations or environmental fluctuations," and he weighed other ways out, including soft selection and interactions that make many small harms count for more together than apart.⁴²

This is the premise of the young-earth genetic-entropy argument, stated by the founder of the theory and by a leading mainstream theorist in their own words.  It does not settle the size of the effect.  Kimura's own figure of one part in ten million per generation compounds to about 0.002% over 240 generations and about 2% over 200,000.⁴³  The dispute is over how many sites outside the protein-coding genes sit in the zone, and the constraint measurements of Part 5 cannot answer it.  A site whose harm sits at the upper edge of Kondrashov's range still changes at 58% of the neutral rate, and one ten times weaker at 95%, so a conservation scan, which looks for sites that change far more slowly than neutral ones, sees little of the range.⁴³  Christian Huber, Bernard Kim and Kirk Lohmueller tested a standard conservation score by simulation in 2020.  Once sites gain and lose function over evolutionary time, as most human noncoding constraint seems to, "There is no improvement in statistical power with increasing tree size for very weak selection," and "Commonly used GERP score thresholds miss over half of the noncoding sites in the human genome where mutations are under purifying selection."^{44 44}  Genetic Entropy prices that question and the mainstream's answers to it.

dN/dS: most protein change is rejected

The ratio of protein-changing to silent differences, written dN/dS or Ka/Ks, measures how much of the mutational supply survives.  For human against chimpanzee it is 0.23, which the genome authors read as "77% of amino acid alterations in hominid genes are sufficiently deleterious as to be eliminated by natural selection."¹⁹  Only 585 of 13,454 genes, 4.4%, showed a ratio above 1, the classic sign of adaptation, and the authors' simulations expected at least 263 such genes "simply by chance."¹⁹  Between human and mouse the median ratio across 12,845 genes is 0.115.⁴⁵  A scan of about 16,500 human genes against five other mammals found significant evidence of positive selection in 400, with the authors warning that the counts reflect "differences in power more than differences in the prevalence of selection."⁴⁶  This is King and Jukes's pattern at genome scale:  most changes to working proteins are thrown out, and most of the ones kept look neutral or slightly harmful.

Noncoding DNA

Adaptive change is not confined to proteins.  In fruit flies, Peter Andolfatto estimated that about 20% of intron and intergenic divergence and 60% of UTR divergence was driven by positive selection, and concluded "that a large fraction of the non-translated genome is functionally important."³⁸  Fruit-fly genomes are compact, with far less noncoding DNA per gene than ours.  In wild mice, α in conserved noncoding elements is lower than in proteins, 0.19 against 0.32–0.38, but those elements are so much larger a target that they hold at least four times as many adaptive changes as the proteins do.⁴⁰  In humans, Eyre-Walker and Keightley found "little evidence for adaptive substitutions" in protein-coding and noncoding data alike, while warning that the true protein figure "could be as high as 40%."³⁷  Noncoding adaptation is real in species with large populations and compact genomes, and poorly measured in ours.

Part 4: Lewontin's paradox and selection at linked sites

The paradox

If most variation within a species is neutral, the amount of it should be set by a balance between new mutations and their loss by drift.  For neutral DNA the expected diversity is about four times the effective population size times the mutation rate.  Mutation rates differ only about tenfold across animals,²⁰ while population sizes differ by many orders of magnitude.  So an insect species numbering in the billions should be vastly more diverse than an elephant species.

Richard Lewontin, a Harvard geneticist whose own protein surveys in the 1960s had helped show how variable natural populations are, pointed out in 1974 that this does not happen.  The heterozygosities then known lay "between 0.056 and 0.184," which on the neutral formula meant "we are required to believe that higher organisms including man, mouse, Drosophila and the horseshoe crab all have population sizes within a factor of 4 of each other."⁴⁷  His conclusion:  "Despite the unchallengeable agreement between the heterozygosity observed in any given case and the predictions of neutral theory, the application of neutral theory to every case of heterozygosity turns out to lead to an absurd result."⁴⁷  This is Lewontin's paradox of variation.

DNA sequencing has widened the range without solving the problem.  Across 167 species, Ellen Leffler and colleagues found diversity "ranges from 0.01% per base pair in Lynx lynx to 8.01% in Ciona savignyi, a span of only 800-fold."⁴⁸  Vince Buffalo's 2021 compilation of 172 animal taxa puts diversity across about three orders of magnitude against population sizes that vary over about twelve.²⁰

The paradox began as an argument against the neutral theory.  John Maynard Smith and John Haigh, who first described genetic hitchhiking in 1974, wrote that "the uniformity of H between species is powerful evidence against the view that the observed polymorphisms are in the main selectively neutral," and called their own hitchhiking model "a last ditch attempt to save the neutral mutation theory."⁴⁹

First explanation: selection at linked sites

When selection acts on one site, it changes the fate of neutral variants sitting next to it on the chromosome.  A favourable mutation that sweeps to fixation drags its neighbours with it and wipes out their variation;  this is hitchhiking.  A harmful mutation that selection removes takes its neighbours with it too;  this is background selection.  Both effects are strongest where recombination is rare, so they predict lower diversity in low-recombination regions.  That correlation is found in fruit flies⁵⁰ and in most of the 40 genomes Russell Corbett-Detig and colleagues examined, where the effect was larger in species with larger populations.  They wrote that "natural selection may provide an explanation for this longstanding paradox of population genetics."⁵¹

How much diversity does linked selection remove?

The largest figure has a weakness its authors state:  the fitted harmful-mutation rates behind it "far exceed direct estimates of the total mutation rate (by 4-9-fold in exons and UTRs".⁵⁵  And even taken at face value it does not solve the paradox.  Brian Charlesworth and Jeffrey Jensen, defenders of the neutral theory, worked through it in 2022:  an 80% reduction means the neutral coalescence time is "about 5-fold larger than the value indicated by current diversity levels;  this is far smaller than the factor of 100 or more suggested by the probable current population size of D. melanogaster ..., and so is insufficient to resolve LP for this species."²²  Buffalo's verdict across all 172 taxa is the same:  "even assuming strong and abundant selection, models of linked selection are unlikely to explain the observed relationship between diversity and census sizes across species."²⁰

In humans, linked selection reaches almost every variant.  Florian Pouyet and colleagues found that background selection and a recombination-linked repair bias "together affect as much as 95% of the variants of our genome."⁵⁶  That is linkage to a small constrained fraction, not a sign that 95% of sites are themselves functional.

Second explanation: populations far from equilibrium

The neutral formula gives the diversity a population reaches once mutation and drift have balanced.  Reaching that balance takes about four times the effective size in generations, which for a large population is an enormous time.  Charlesworth and Jensen calculate that for a population that grew to an effective size of a billion, "the time for π to approach ¾ would be of the order of 10⁹ generations."²²  Populations that expanded long ago may still be far below their equilibrium diversity, and "This scenario thus seems an excellent candidate for resolving LP."²²  Buffalo reached the same place from the other side:  since linked selection fails, "this supports the hypothesis the diversity across species are shaped primarily by past demographic fluctuations."²⁰  Life history fits this reading.  Across 76 animal species, Jonathan Romiguier and colleagues found diversity "accurately predicted by key species traits related to parental investment," not by geographic range.⁵⁷

What the paradox does to the neutral theory

It falsifies the strict version Kern and Hahn attacked, in which diversity sits at mutation–drift equilibrium.  Nobody defends that version now, and Galtier notes that the demographic explanations are "compatible with the neutral theory."²  It does not touch the theory's claim about divergence between species, because linked selection does not change how often a neutral mutation fixes.⁵  The irony is that a datum first raised against neutralism is now mostly explained by neutral processes plus a modest share of selection, most of it purifying.

What it means for a young earth

A young earth is the most extreme non-equilibrium history there is.  At the measured human rate, reaching mutation–drift balance takes 40,000 to 80,000 generations, 1 to 2.3 million years;⁸ for a large insect population it takes far more generations.  On a young earth no species is anywhere near equilibrium, so diversity would not be expected to track population size.  The paradox is therefore not evidence against a young earth, and the mainstream's leading explanation for it is the same kind of explanation a young-earth model would give.

It is not evidence for a young earth either.  An expansion five million years ago does the same job in Charlesworth and Jensen's arithmetic.²²  And a young-earth model still has to say where the existing diversity came from.  Mutation at the measured rate over about 240 human generations adds only about three new differences per million sites per lineage,⁸ while living humans differ at about one site in a thousand, the figure behind the familiar human effective size of 10,000 to 20,000 explained in Population Genetics.  The young-earth answer is created diversity, which is argued in Adam, Eve, and Human Genetic Diversity.

Part 5: Does the theory depend on junk DNA?

What the 2018 critics did not say

It is easy to read the 2018 exchange as "the neutral theory failed because the genome turned out to be functional."  That is not what Kern and Hahn argued.  They set junk DNA aside at the start:  the theory "is not simply a statement ... about the large fraction of eukaryotic genomes that are nonfunctional—neither of these assertions would be contested by current competing hypotheses."⁶  The strongest critics of the theory accept that most of a large genome does nothing.  Their case is about adaptive change inside proteins and about selection reaching neutral sites through linkage.  Nobody in the professional dispute argued that the genome is mostly functional, and an article that cites Kern and Hahn for that is citing them for the opposite of what they wrote.

Where the theory leans on a nonfunctional genome

Three of the theory's supports do lean on most of the genome being free of constraint.

1. The genome-wide claim. Faced with fruit-fly proteins where half the changes look adaptive, the theory's defenders answered that those figures "mostly concern only the small fraction of the genome that codes for proteins."⁵ The claim that most genomic differences are neutral is carried by the 98% that is not protein-coding. If that 98% were under selection, the answer would fail.
2. Ohta's explanation of the generation-time effect. DNA divergence in general runs faster in short-lived animals, while protein divergence does not. Ohta explained this with two tiers: "most genomic DNA of higher organisms can freely accumulate base substitutions," while amino-acid changes are often slightly harmful.²³ The first tier is a no-constraint premise for most of the genome.
3. Kimura's top speed. The fastest-changing DNA Kimura knew was in pseudogenes, where "base substitutions at the first and the second codon positions in the pseudogenes occur just as frequently as those at the third position."¹ That is what a dead gene should look like, and he used it as the theory's best case. But he answered the suggestion that pseudogenes might have "an unknown but important function" by opinion: "In my opinion, it is much more appropriate to regard pseudogenes as indeed dead."¹ This site's Pseudogenes article finds that a few dozen of about 14,700 human pseudogenes have a measured function, and most have never been tested one by one.

Where it does not

The protein evidence does not lean on junk at all.  The human-chimpanzee ratio of protein-changing to silent changes is 0.23, which the chimpanzee genome authors read as "77% of amino acid alterations in hominid genes are sufficiently deleterious as to be eliminated by natural selection."¹⁹  The McDonald–Kreitman estimates in Part 3, the hominid–murid contrast and the pattern that less important parts of a protein change faster are all measured inside genes whose function no one disputes.  Whatever the noncoding genome turns out to do, these results stand.

How much is measured to be under constraint

The mainstream estimates of how much of the human genome is held in place by purifying selection are these:

Every one of these is far below a majority.  The cross-species estimates assume common descent and treat ancient transposon copies as the neutral yardstick, so a design proponent can fairly say they measure shared conservation, not function.  This site's Functional DNA article makes that objection.  But the human-only part of the Kellis estimate, 4–11%, rests on variation within living humans and assumes no common ancestry.

Selection does touch almost the whole genome indirectly.  Florian Pouyet and colleagues found that background selection and a recombination-linked repair bias "together affect as much as 95% of the variants of our genome."⁵⁶  That is the reach of linkage to a small constrained fraction, not a finding that 95% of sites are functional.

What changes if much more of the genome is functional

There are two very different ways the genome could be "more functional."

Function that does not depend on the exact letters.  Dan Graur, one of the most outspoken defenders of junk DNA, allows a category he calls "indifferent DNA":  "DNA sites that are functional, but show no evidence of selection against point mutations," such as "spacers and flanking elements whose presence is required but whose sequence is not im[portant]."²⁸  If most of the extra function is of this kind, the neutral theory is untouched.  A letter change in a spacer is neutral even though the spacer does a job.  Much of the biochemical activity the ENCODE project reported could fall here, and nothing in the neutral theory forbids it.

Function that depends on the exact letters.  This is the claim that matters.  The Functional DNA article estimates that at least 30% of human DNA is sequence-specific.  This site's GWAS article reran that arithmetic and got 2% to 30% depending on inputs, and 6% to 10% with heritability inputs;  but take 30% as the test case.  If the 15 million human-line changes fell at random, about 4.5 million of them landed in DNA whose letters matter.⁸  They cannot mostly be neutral, by the definition of "sequence-specific."  Each one must then be one of three things.

1. Adaptive. Then selection paid for millions of substitutions. Haldane's Dilemma puts the budget at hundreds to a few thousand on the human line. No one on either side proposes millions.
2. Slightly harmful, carried by drift. Then each had an effect below the drift barrier, about 2.5 × 10⁻⁵ for a human effective size of 20,000.⁸ Function that weak is "sequence-specific" in only the faintest sense, and millions of such fixations are a fitness decline, which is the Genetic Entropy argument. This is close to John Sanford's own picture: "While most sites are probably nearly-neutral, very few, if any, should be absolutely neutral."⁵⁹
3. Never fixed by any population process. This is the young-earth answer: the differences between human and chimpanzee were created.

Or the changes did not fall at random, and the 30% shows a deficit of differences, which is exactly what a conservation measurement looks for.  The measurements find a deficit in about 3.5% to 15% of the genome across mammals, and 4% to 11% more within humans.  So the data do not show a sequence-specific 30% that evolution would have to explain.  These scans see function strong enough for selection to slow change.  They cannot rule out the very slightly harmful class of item 2, which changes at more than half the neutral rate (Part 3).  If such a fraction were shown, the professional literature agrees on what would follow.  Graur wrote that "there exists a misconception among functional genomicists that the evolutionary process can produce a genome that is mostly functional."²⁸

A larger sequence-specific fraction would also raise the number of harmful new mutations per newborn.  At the pedigree mutation rate a child carries about 77 new point mutations, so 3.5% sequence-specific DNA means about 2.7 harmful ones a child, and 30% means about 23.⁸  That is the genetic-load argument, which Genetic Entropy prices against the mainstream's answers, and it is not re-argued here.  And more constrained DNA would mean more background selection, so lower neutral diversity everywhere:  that pushes the data toward Kern and Hahn's side of the dispute, not toward a functional genome that is also neutral.

The junction, argued both ways

For the design side.  The junction is real, and the mainstream says so.  The genome-wide form of the neutral theory is not independent evidence against widespread function, because two of its supports, Ohta's free-accumulating bulk and Kimura's dead pseudogenes, assumed the answer.  If sequence-specific function were shown in a large part of the genome, the neutral account of the 15 million human-line differences would fail, and so would every other evolutionary account of them, as Graur concedes.

For the mainstream.  The no-constraint premise is no longer only an assumption.  It is measured, by conservation across mammals and by variation within humans, and the measurements agree on a minority.  Pseudogenes passed the neutral theory's own test:  all three codon positions change at the same rate, as they should in dead DNA, while in working genes the third position changes fastest.  The theory's sharpest critics concede the nonfunctional bulk.  And the theory is not the reason the bulk is thought nonfunctional;  it is one of several lines that agree.

What would settle it.  A sequence-specific function shows up as missing variation in living humans, whatever one believes about ancestry.  The test a design proponent needs to pass is a within-human depletion of variation across far more than the 4–11% now measured, in DNA outside known genes and regulatory elements.  The large population sequencing projects now under way can run that test.

Part 6: What follows for both sides

What young-earth and design writers have said

Young-earth writers have said a good deal about the neutral theory, and most of it uses the theory rather than attacks it.

1. They have Kimura's motive right. Robert Carter of Creation Ministries International wrote that "Kimura saw Haldane's Dilemma as a serious problem and listed it as his main reason for proposing the Neutral Theory of evolution."⁶⁰ John Sanford and colleagues: "Kimura agreed with Haldane's assessment of the problem, and this motivated him to develop his neutral model of evolution."⁶¹ Walter ReMine: "These arguments were never 'for' the neutral theory. They are arguments against selective evolution, severely limiting its speed."⁶² Kimura's 1968 paper bears this out, as Part 1 shows. What they leave out is that the cost argument rested on the four-billion-site denominator both sides now reject.
2. They link it to junk DNA. Carter in 2009: "Junk DNA is a necessary mathematical extrapolation. It was invented to solve a theoretical evolutionary dilemma," and "If 'Haldane's Dilemma' was correct, the majority of DNA must be non-functional."⁶³ Jeffrey Tomkins and Jerry Bergman say new findings of function "present serious obstacles to the evolutionary necessity of a large fraction of the genome being 'junk'."⁶⁴ Sanford, in Genetic Entropy, says Kimura's theory "led him to believe that most mutations are unselectable, and therefore most genetic sequences must be irrelevant,"⁵⁹ and that "As the functional genome expands, the likelihood of neutral mutations shrinks."⁵⁹
3. ReMine concedes the central point. He writes that his book "has an entire chapter on neutral evolution, and argues that the vast majority of substitutions would be neutral, due to genetic drift, not selection."³³ His dispute is about whether a neutral substitution has a cost, not about whether most substitutions are neutral.
4. Genetic entropy is built on the nearly neutral theory. Sanford writes that "most sites are probably nearly-neutral" while doubting that any are "absolutely neutral."⁵⁹ Carter describes genetic entropy as "an idea which leverages the beliefs and language of the neutralist position to make a strong case against it."²⁹ The argument needs Ohta to be right: slightly harmful mutations must drift to fixation in populations as small as ours. Part 3 shows that this is the part of the theory the data support best, and that Kimura himself wrote that effectively neutral mutants are "in fact, very slightly deleterious" and "accumulate continuously in every species."³¹ Sanford overstates how the mainstream has treated them. He writes that in the standard account "everything in the 'near-neutral box' is redefined as being completely neutral, and is thereby dismissed."⁵⁹ Kimura did not dismiss the box. He priced it, at one part in ten million of fitness per generation for the protein-coding genes, and answered it with occasional adaptive substitutions. Kondrashov priced the genome-wide case in 1995 and rejected that answer.⁴² So the question between Sanford and the mainstream is not whether the class exists, but how much of the genome lies in it and what offsets it.
5. Some young-earth arguments need the neutral rate result. Carter uses the equality of mutation rate and substitution rate to date the human ancestors with pedigree mutation rates: "since most mutations are selectively neutral, genealogical mutation rates should be directly applicable to estimates of the timing of Y Chromosome Adam and Mitochondrial Eve."³² Elsewhere he writes that "The attempted rescue mechanism of neutral evolution theory, and the associated belief in 'junk DNA', failed as an escape mechanism."⁶⁵ Both cannot be fully right unless he says which part of the theory failed; he does not.
6. Nobody answered 2018 or Lewontin. No creationist reply to Kern and Hahn, and no creationist treatment of Lewontin's paradox, was found in searches of the main young-earth and design sites.

Is "neutral" the same as "junk"?

Partly.  The creationist step from neutral theory to junk DNA is right about where the theory's genome-wide claim gets its weight, as Part 5 shows.  It is wrong as a statement of what the theory requires.  Kimura wrote that "the existence of selective constraint itself does not contradict the neutral theory,"¹ and his 1979 model built constraint into the theory explicitly.³¹  A site can have a function and still accept most of its changes neutrally, and a stretch of DNA can do a job without its exact letters mattering, as Graur's "indifferent DNA" shows.  What the theory cannot absorb is a large share of the genome whose exact letters matter strongly enough for selection to see.  That is a narrower claim, and it is the one a design argument has to prove.

What follows for the mainstream

The theory is the mainstream's own statement that natural selection did not build most of the DNA differences between humans and chimpanzees.  On the best current count, selection fixed about 4,000 protein changes on the human branch, with an unmeasured number in regulatory DNA, out of about 15 million.  Popular accounts that present the human–chimpanzee difference count as a record of adaptation are describing something the field's own theory denies.

It also means the theory cannot be used as independent evidence that the genome is mostly junk.  Its genome-wide success was partly assumed from the start, in Ohta's freely changing bulk and Kimura's dead pseudogenes.  The independent evidence is the constraint measurements, and that is where the argument belongs.

What follows for the young-earth side

The concession is real.  The figure on this site's Functional DNA page, that over 99.98% of mammal differences spread by chance, is close to the mainstream count:  about 99.97% of human-branch changes were not adaptive at Uricchio's central estimate, and 99.6% on the most generous figures.⁸  A young-earth writer can cite the neutral theory for the claim that selection did not build most DNA differences, and should cite Uricchio's 4,000, not Kimura's 1968 arithmetic.

The inference needs care.  "A highly neutral evolutionary process could create nothing better," as that page argues, applies to the DNA that changed:  the 4% that differs between humans and chimpanzees.  On common descent the shared 96% was inherited, not built in six million years, and the neutral theory says nothing about how it arose.  The mainstream also proposes a neutral route to complexity, "constructive neutral evolution," a "unidirectional evolutionary ratchet leading to complexity";⁶⁶ it is a scenario, not a measured rate, but it means the claim that neutral processes build nothing is contested.

Between created kinds, the neutral theory does not apply on a young earth.  At the measured rate, neutral substitution changes about three sites in a million per lineage in 6,000 years.⁸  So none of the differences between created kinds can be fixed mutations, neutral or otherwise.  Young-earth models say they were created, and Nathaniel Jeanson predicts that "the majority — if not the vast majority — of nuclear DNA differences among species within families (for vertebrates) ... are functional participants in the biology of each species, not functionally neutral."⁶⁷  That is a clear, testable claim.  The McDonald–Kreitman estimates for close relatives give a mixed answer:  in fruit flies about half of protein changes and a fifth of intron changes look adaptive, while between mouse and rat, which share a family, the adaptive share is 19% to 38% in proteins and conserved elements and the rest of the genome is treated as neutral.⁴⁰  Neither is a direct test, because "adaptive" and "functional" are not the same thing.  The test the prediction calls for, measuring selection on the within-family differences themselves, has not been run.

Within kinds, drift has to work fast.  Jeanson and Jason Lisle propose that created variety was sorted after the Flood, and that "genetic drift and other processes likely have diluted this concentrated heterozygosity to lower levels."⁶⁸  Drift on that timetable is the neutral theory's own mechanism, and it has a speed limit.  A neutral allele starting at half the population takes on average 4N_e ln 2, about 2.8 N_e generations, to fix, when it does.⁶⁹  In 4,500 years that allows an effective size of up to about 1,600 for an animal that breeds every year, and about 160 for one that breeds every ten years.⁸  Small post-Flood populations meet that limit;  large ones do not, unless selection rather than drift did the sorting.  The Creation Model of Rapid Speciation article takes up how the models handle it.

Objections and responses

"Kern and Hahn showed the neutral theory is dead, so this article is out of date."

They argued that its universality is rejected, and on proteins in species with large populations they are right, as Part 3 shows.  But they concede the human figure for proteins is near zero,⁶ and they set junk DNA aside as uncontested.⁶  Seven senior population geneticists replied that their view is "probably a minority view,"⁵ and the 2024 review from inside the field treats the question as one of proportions, not of life or death.²  For the human genome as a whole, nothing in the 2018 paper moves the conclusion.

"Selection reaches 95% of human variants, so nothing is neutral."

That figure measures linkage to selected sites, not selection on the sites themselves.⁵⁶  Background selection lowers diversity near constrained DNA but does not change the rate at which neutral mutations fix between species.⁵  And Murphy and colleagues found that in humans sweeps had "little or no effect," so most of that reach is the removal of harmful mutations, which is the neutral theory's own second clause.²¹

"The 4,000 adaptive changes is an underestimate."

It may be.  Eyre-Walker and Keightley warn that the human protein figure "could be as high as 40%,"³⁷ and adaptive change in human regulatory DNA has hardly been measured.  Take the generous case:  40% of the 29,925 human-branch protein changes, plus four times as many in noncoding DNA, the ratio Halligan and colleagues found in wild mice.  That is about 60,000 adaptive changes, 0.4% of the human branch.⁸  The conclusion that most of the 15 million changes were not adaptive survives any figure the literature supports.

"Conservation only measures shared DNA under common descent. Under common design it means nothing."

For the cross-species estimates, that is a fair point, and this site's Functional DNA article makes it.  But constraint can also be measured within one species, from which variants are missing in living people, and that needs no assumption about ancestry.  The human-only estimate is 4% to 11% beyond protein-coding DNA.²⁵  A design argument that relies on a much larger sequence-specific fraction has to show it in that kind of data.

"The neutral theory proves evolution cannot build function, so function refutes evolution."

The theory says selection did not fix most of the differences between species.  It says nothing about the shared DNA, which on common descent was inherited, not built in the last six million years.  The mainstream also proposes that neutral processes can build complexity, though this "constructive neutral evolution" is a scenario rather than a measured rate.⁶⁶  The argument has force only if a large share of the DNA that differs turns out to be sequence-specific, which is the test in Part 5.

"Kimura himself said most of the genome must be irrelevant."

Sanford reads him that way.⁵⁹  Kimura's own words are more careful:  "the existence of selective constraint itself does not contradict the neutral theory."¹  The fair version of the charge is narrower and correct:  Kimura's showcase evidence, the fast-changing pseudogenes, rested on his opinion that they were "indeed dead."¹

"Effectively neutral mutations are harmless by definition, so genetic entropy misreads the theory."

Not by the theory's own definition.  Eyre-Walker and Keightley call neutrality "operational rather than functional":  it depends on whether selection can act in a given population, "not solely on the effect of the mutation on fitness."⁴¹  Kimura wrote that effectively neutral mutants are "in fact, very slightly deleterious" and "accumulate continuously in every species."³¹  The theory and genetic entropy agree that the class exists and accumulates.  They disagree about its size and about what offsets it:  Kimura put the loss at one part in ten million per generation for protein-coding genes and trusted adaptive substitutions to cover it, while Kondrashov's genome-wide version is a paradox he could not resolve with beneficial mutations.⁴²  Genetic Entropy prices that dispute.

"Synonymous mutations are mostly harmful (the 2022 yeast study), so the neutral yardstick is broken."

The yeast result has not held in human cells, and in apes about a fifth to two fifths of synonymous sites show weak constraint.  Synonymous Mutations and Codon Usage works through it and finds the bias in dN/dS is modest, a factor of 1.26 to 1.57, and in the direction that makes protein change look less constrained, not more.  It does not overturn the numbers here.

"The neutral theory can fit anything by adjusting population size, so it explains nothing."

Lewontin made a version of this complaint in 1974.  But the theory made risky predictions, and some of them failed:  protein evolution in flies is not mostly neutral, and diversity does not track population size at equilibrium.  Others passed:  pseudogenes change equally at all three codon positions,¹ and slightly harmful changes are commoner in small populations, as Ohta predicted.¹⁹  A theory that can fail and sometimes does is testable.

"A young earth cannot be squared with any of this."

Part 4 and Part 6 price the two places it bears.  Lewontin's paradox is what a recent, non-equilibrium history predicts, so it is no evidence against a young earth.  But drift sorts created variety slowly in large populations:  a neutral allele at half frequency needs about 2.8 N_e generations to fix,⁶⁹ which fits only small populations in 4,500 years.

Issues and unanswered questions

1. Adaptive change in human noncoding DNA is barely measured. Every human α in this article is for proteins. Regulatory DNA is most of the constrained genome, and the mouse data suggest it holds most adaptive changes. The human-branch count could rise several-fold without changing the conclusion that most change was not adaptive.
2. The methods disagree. For human against chimpanzee, the same data give −0.41 by the plain McDonald–Kreitman test and 0.25 under a fuller model.¹⁵ Gossmann and colleagues in 2012 found the adaptive rate rising with population size; Galtier in 2016 found it flat. These are not settled.
3. Lewontin's paradox is open. Linked selection falls short;²⁰ non-equilibrium history is the leading candidate but has not been shown species by species. No young-earth writer was found to have addressed it.
4. Jeanson's prediction is untested. Whether most nuclear differences within a created family are "functional participants" or neutral could be tested by measuring selection on those differences in, for example, mouse and rat or cat species. No one has done it.
5. Drift's speed inside young-earth models. The fixation-time arithmetic in Part 6 assumes a constant effective size. Real post-Flood populations grew from a few founders; a model with growth and selection would give a better answer, and none has been published.
6. ReMine's claim that neutral substitutions carry a cost is a definitional dispute with Brookfield and others that this article does not settle. A simulation counting the lost reproduction per neutral fixation would settle it.
7. A citation defect in the 2018 paper. Kern and Hahn cite Boyko and colleagues for a human figure of zero; Boyko's abstract gives 10–20%.¹⁴ Li and colleagues 2008, their other source, was not checked.
8. Carter's two positions on the neutral theory, that it failed as an escape from Haldane's dilemma and that its rate result dates Adam and Eve, need reconciling by their author.
9. How much of the genome is very slightly harmful when changed is unmeasured. Conservation scans cannot see harm near the drift barrier, because such sites still change at more than half the neutral rate (Part 3). So the 3.5% to 26% constraint figures bound sequence-specific function that selection can act on, not the class the genetic-entropy argument needs. No current method bounds that class directly.

Verdict

Most DNA differences fixed between mammal species are neutral or nearly so, and the mainstream's own count puts the adaptive share on the human branch at a few hundredths of a percent;  even the most generous reading leaves it under half a percent.  That conclusion rests mainly on measurements showing that most of the genome is not under constraint, so the theory's genome-wide form does depend on most of the genome lacking sequence-specific function, as young-earth critics say.  Inside proteins of species with large populations, the theory's original claim has failed, and the 2018 challenge is right about that while conceding the rest.  For the design argument, the neutral theory is a real concession that selection did not build most differences, but it becomes evidence against evolution only if a large, sequence-specific function is shown in the DNA that differs, and the within-human data do not yet show one.  For the genetic-entropy argument, the theory supplies the premise:  its founder wrote that effectively neutral mutations are "in fact, very slightly deleterious" and accumulate in every species, and the dispute is over how many sites they touch and what offsets the loss.³¹

Sources