The Neutral Theory of Molecular Evolution

The neutral theory of molecular evolution, formulated by Motoo Kimura in 1968 and independently supported by Jack King and Thomas Jukes in 1969, makes a precise and falsifiable claim: at the level of DNA and protein sequence, the great majority of mutations that become fixed within a species — and thus the great majority of the differences we observe between species — are neutral or nearly neutral with respect to fitness. Their fate is decided by Genetic Drift, the random sampling of gametes from generation to generation, not by natural selection. This is a claim about substitutions (mutations that have spread to fixation) and about standing variation (polymorphism within populations), not a claim about phenotypic evolution in general. Kimura never denied that natural selection shapes adaptations — eyes, wings, immune recognition, beak shape. His claim was narrower and, at the time, more radical: beneath the visible phenotype, the sea of molecular variation segregating in populations and diverging between species is mostly selectively irrelevant. A mutation is "neutral" if it has no effect on an organism's fitness — commonly formalized using the population-genetic parameter s, the selection coefficient, and N, the Effective Population Size. A mutation behaves as effectively neutral when the strength of selection acting on it is small relative to the power of drift, conventionally stated as the condition |Nes| < 1 (or |4Nes| < 1 in diploids), where drift overwhelms selection's ability to reliably increase or purge the allele. Synonymous codon changes (which do not alter the amino acid, due to redundancy in the genetic code), mutations in noncoding DNA far from any regulatory element, and many amino acid changes that do not affect protein folding or function all fall into this category. Kimura's theory should not be confused with two things it is often wrongly equated with. First, it is not "genetic drift is the only evolutionary force" — deleterious mutations are still purged by purifying selection, and beneficial mutations still sweep to fixation by positive selection; the neutral theory simply claims these two classes constitute a minority of fixed molecular changes, with the neutral middle ground dominating the count of substitutions. Second, it is not equivalent to non-Darwinian evolution in the colloquial sense of "chance rules everything" — it is a rigorously quantitative population-genetic model that makes numerically precise, testable predictions, several of which have been confirmed and some of which required later refinement (the nearly-neutral theory).

Step 1 — Where neutral mutations come from. Mutation is a continuous stochastic process operating on DNA: replication errors, deamination, transposon activity, and repair failures generate new variants at a genome-wide rate on the order of 10⁻⁸ to 10⁻⁹ per base pair per generation in most eukaryotes. Because the genetic code is degenerate (61 sense codons specify only 20 amino acids), a large fraction of mutations at third codon positions are synonymous — they change the codon but not the encoded amino acid. Other mutations fall in introns, intergenic spacer DNA, or regions with no known regulatory role. These mutations arise with no reference to their eventual fitness effect; mutation itself is not directed by selective need — a principle independently established by the Luria-Delbrück Fluctuation Test.

Step 2 — Fixation probability under drift alone. Once a new mutation appears, it exists as a single copy in a diploid population of N individuals — an initial frequency of 1/(2N). If the mutation is truly neutral, its trajectory through subsequent generations is a random walk driven purely by which gametes happen to be sampled to form the next generation. A cornerstone result of population genetics, derivable from a simple identity-by-descent argument, is:

Probability of eventual fixation of a new neutral mutation = 1/(2N) This holds regardless of the mutation's molecular identity — it is a pure consequence of finite population sampling.

Step 3 — Substitution rate equals mutation rate. This is the theory's most elegant and consequential result. In a diploid population of size N, the number of new neutral mutations arising each generation is: 2N individuals × μ (neutral mutation rate per gamete per generation) = 2Nμ new mutations per generation Each has fixation probability 1/(2N). The expected number of neutral substitutions fixed per generation is therefore:

k = 2Nμ × 1/(2N) = μ The population size N cancels exactly. The rate of neutral substitution equals the neutral mutation rate, independent of population size, independent of demographic history, independent of everything except μ. This is the mathematical seed of the molecular clock hypothesis: if μ is roughly constant across lineages (per generation, or per unit absolute time if generation times and mutation rates trade off in a compensating way), substitutions should accumulate at an approximately constant rate through time.

Step 4 — Purifying selection removes the rest. Not all new mutations are neutral. Most amino-acid-changing (nonsynonymous) mutations are mildly to strongly deleterious because proteins are the product of millions of years of fine-tuned folding, binding, and catalysis; a random change is far more likely to disrupt than to improve function. Purifying (negative) selection eliminates these before they can drift to appreciable frequency, keeping the "effective" neutral mutation rate at nonsynonymous sites much lower than at synonymous sites. A small residual class of nonsynonymous mutations is advantageous and swept rapidly by positive selection — but Kimura argued these are numerically rare compared to the neutral substitutions accumulating everywhere else in the genome.

Step 5 — Neutral polymorphism and the drift-mutation balance. Within a population at any instant, neutral genetic diversity reflects a balance between new mutation input and loss by drift. Kimura and Crow's diffusion approximation gives expected heterozygosity at a neutral locus as:

H = 4Neμ / (1 + 4Neμ) where Ne is effective population size. This predicts that species with larger effective population sizes should carry more neutral polymorphism — a testable, and largely confirmed, relationship, though the correlation is weaker than a naive Ne-only model predicts (the "paradox of variation," discussed below).

1. King and Jukes (1969), "Non-Darwinian Evolution." Working independently of Kimura but arriving at the same conclusion, Jack King and Thomas Jukes compiled amino acid substitution data across mammalian hemoglobin, cytochrome c, and other proteins. They noted several patterns inconsistent with pervasive positive selection: substitutions occurred preferentially at amino acid sites where the change was chemically conservative (e.g., one hydrophobic residue for another), the substitution rate correlated with the number of nucleotide changes required rather than with any measure of adaptive significance, and rates were far too high and too uniform across radically different lineages (with wildly different population sizes and ecologies) to be plausibly driven by selection acting the same way in every case. Their title deliberately provoked — "non-Darwinian evolution" did not mean anti-Darwinian in the sense of denying selection's role in adaptation, but signaled that molecular change need not be adaptive at all.

2. Kimura's mathematical argument from mutation load (1968). Kimura's original paper was as much theoretical as empirical. He calculated that if every one of the roughly 10⁷–10⁸ nucleotide substitutions separating mammalian lineages (extrapolated from measured protein evolution rates) had been driven by positive selection, the "cost of selection" — the number of selective deaths or reproductive failures required per gene substitution, a concept developed by J.B.S. Haldane — would be catastrophically large, far exceeding what any real population's reproductive excess could sustain over the observed timescales. Haldane's earlier calculation set an upper bound of roughly one selected substitution per 300 generations, given typical reproductive excess in vertebrates. Given the observed rate of protein evolution (about one amino acid substitution per polypeptide of average length every few million years, translating to a genome-wide substitution rate that would require substitutions far more frequently than Haldane's cost model could support if all were selected), Kimura concluded that most substitutions could not have been individually driven by selection — they had to be neutral, whose fixation costs the population nothing beyond ordinary drift.

3. The molecular clock observations (Zuckerkandl and Pauling, 1962–1965). Even before Kimura's theory, Emile Zuckerkandl and Linus Pauling had noticed, comparing hemoglobin and cytochrome c sequences across vertebrates, that amino acid substitutions accumulated at an approximately constant rate per unit time across lineages as different as fish, birds, and mammals — the observation that became known as the molecular clock hypothesis. This was puzzling under a purely selectionist view, since wildly different species under wildly different selective regimes would not be expected to substitute amino acids at the same rate by chance coincidence of adaptive pressures. Kimura's theory explained the clock naturally: since neutral substitution rate equals mutation rate (k = μ), and mutation rate per generation is roughly similar across many lineages (or compensates predictably with generation time), a roughly constant molecular clock is the expected signature of neutral evolution — not a mysterious coincidence requiring a selective explanation.

4. Direct sequence comparison — synonymous versus nonsynonymous rates. The most direct and repeatedly replicated test compares the rate of synonymous substitutions per synonymous site (dS) to the rate of nonsynonymous substitutions per nonsynonymous site (dN) within the same gene, between the same pair of species. Because synonymous changes do not alter the protein, they are largely shielded from selection on protein function (though not entirely — codon usage bias shows some synonymous sites are weakly selected too). Nonsynonymous changes are exposed to whatever selection acts on the protein. Across essentially every protein-coding gene ever surveyed — from Kimura and Ohta's early comparisons in the 1970s through modern genome-wide studies — dS exceeds dN, typically by a factor of 3 to 10 or more for genes under typical purifying selection, exactly as predicted if most nonsynonymous mutations are deleterious and purged while synonymous ones drift freely. The ratio ω = dN/dS became a standard diagnostic: ω ≈ 1 indicates neutrality, ω < 1 indicates purifying selection, and ω > 1 (rare, but real in genes such as MHC and some reproductive proteins) indicates positive selection — itself a direct descendant of the neutral theory's logic, since you need a neutral null model to detect the signature of adaptation at all.

5. The paradox of variation and Lewontin's puzzle. Kimura's diffusion formula H = 4Neμ/(1+4Neμ) predicts that species with enormously different effective population sizes (spanning many orders of magnitude, e.g., bacteria versus elephants) should show correspondingly enormous differences in neutral genetic diversity. Richard Lewontin's 1974 survey of allozyme variation, and subsequent genomic surveys, found that heterozygosity varies far less across species than Ne does — species with population sizes differing by six orders of magnitude often show diversity differing by less than one order of magnitude. This "paradox of variation" was not a refutation of neutral theory's logic but a stimulus for refining it: linked selection (genetic hitchhiking and background selection reducing diversity at linked neutral sites, discussed under the coalescent framework) compresses the effective population size actually experienced by a neutral locus, decoupling census Ne from the Ne relevant to the neutral prediction.

Example 1 — calculating expected neutral heterozygosity and comparing to observation. Consider two species: a marine invertebrate with effective population size Ne = 10⁶ and a large mammal with Ne = 10⁴. Assume both have a neutral mutation rate μ = 1 × 10⁻⁸ per site per generation. For the invertebrate: 4Neμ = 4 × 10⁶ × 10⁻⁸ = 0.04, so H = 0.04/1.04 ≈ 0.038. For the mammal: 4Neμ = 4 × 10⁴ × 10⁻⁸ = 0.0004, so H = 0.0004/1.0004 ≈ 0.0004. Naively, the neutral theory predicts the invertebrate should be roughly 100-fold more heterozygous than the mammal, tracking the 100-fold difference in Ne. Empirically, marine invertebrates do tend to show higher neutral diversity than large mammals, confirming the qualitative direction of the prediction — but the ratio observed (often only 5- to 20-fold) is far smaller than the 100-fold predicted from census/effective population size estimates alone. This gap is the paradox of variation: it points to additional reducers of effective diversity — particularly linked selection continually removing diversity at neutral sites near selected loci (hitchhiking during selective sweeps, and background selection removing neutral variants linked to deleterious mutations) — that shrink the "genetic" Ne below the demographic Ne, especially in genomically compact or strongly selected genomes.

Example 2 — using dN/dS to detect selection at a candidate gene. Suppose you sequence an immune-recognition gene in two primate species and find, over an alignment of 900 codons (2,700 nucleotides): 45 synonymous differences out of approximately 675 synonymous sites, and 60 nonsynonymous differences out of approximately 2,025 nonsynonymous sites (a simplified count ignoring multiple hits, for illustration). dS = 45/675 = 0.0667 substitutions per synonymous site dN = 60/2025 = 0.0296 substitutions per nonsynonymous site ω = dN/dS = 0.0296/0.0667 ≈ 0.44 Here ω < 1, indicating this gene, overall, is under purifying selection — most amino acid positions are functionally constrained, exactly as expected for most genes under the neutral theory's framework, where synonymous sites drift freely while nonsynonymous sites are filtered by selection. Now suppose a second gene, a viral surface receptor under an evolutionary arms race with a pathogen, shows dS = 0.03 and dN = 0.05, giving ω ≈ 1.67. Because ω exceeds 1, more amino-acid-altering mutations have been fixed than expected under strict neutrality — a signature that positive selection has been actively favoring protein-changing mutations, most plausibly to escape immune recognition or maintain binding specificity against a rapidly evolving pathogen. This contrast is precisely how the neutral theory functions in practice: not as a claim that all evolution is neutral, but as the necessary null model against which departures — signatures of selection — are statistically identified.

Motoo Kimura (1924–1994) was a Japanese population geneticist trained originally in botany and cytology at Kyoto University, who later crossed into theoretical population genetics after encountering the work of Sewall Wright — one of the founders of the mathematical treatment of genetic drift. Kimura pursued graduate study in the United States under James F. Crow at the University of Wisconsin, where he sharpened the diffusion-equation methods (borrowed from statistical physics) that would let him model allele frequency change under drift, mutation, and selection simultaneously. He returned to Japan to work at the National Institute of Genetics in Mishima, where he spent the rest of his career largely as a theoretician, publishing prolifically on the mathematical population genetics of drift, and rarely running a wet-lab experiment himself — his primary "experiments" were on paper, using accumulating protein sequence data collected by others (notably Margaret Dayhoff's protein sequence atlases) as the empirical test of his equations. Kimura proposed the neutral theory in a short 1968 paper in Nature, "Evolutionary Rate at the Molecular Level," motivated initially by a back-of-envelope calculation of the implausibly high cost that pervasive positive selection would impose on a population trying to fix the observed number of amino acid substitutions across vertebrate lineages. The theory met fierce resistance from prominent selectionists, especially in the early 1970s, generating one of the most productive controversies in twentieth-century evolutionary biology — pushing both camps to develop sharper statistical tests (dN/dS analysis, tests of neutrality such as the McDonald-Kreitman test and Tajima's D) that ultimately became standard tools regardless of which side of the debate a researcher favored.

Modern genome-wide scans for selection in humans, crops, and pathogens all start from a neutral null model: candidate regions are flagged as "under selection" only when they deviate from neutral-theory expectations (unusually low diversity, skewed site-frequency spectra, elevated dN/dS), making Kimura's framework the statistical backbone of essentially all molecular evolution and population genomics research, including scans for genes underlying human local adaptation (lactase persistence, high-altitude adaptation) and crop domestication traits.

Molecular clock dating, used throughout phylogenetics and epidemiology to time species divergences and viral outbreak origins (including real-time SARS-CoV-2 lineage dating during the COVID-19 pandemic), rests directly on the neutral prediction that substitution rate tracks mutation rate; "relaxed clock" Bayesian methods used in modern software explicitly model departures from strict neutral-clock constancy using nearly-neutral rate variation.

Conservation genetics uses neutral markers (microsatellites, synonymous SNPs, mitochondrial control-region sequence) as a proxy for genome-wide diversity and effective population size in endangered species, because neutral loci are not confounded by locus-specific selection, making inferences about population history and inbreeding risk more reliable.

Cancer genomics applies dN/dS-style logic to somatic mutations in tumor sequencing: genes showing an excess of nonsynonymous mutations relative to the neutral expectation are flagged as cancer drivers under positive selection within the tumor, while the bulk of passenger mutations are treated, correctly, as neutral hitchhikers — a direct clinical descendant of Kimura's original logic, now used to prioritize genes for targeted cancer therapy development.

Pharmacogenomics and conservation of essential genes in drug target discovery uses strong purifying selection signatures (very low dN/dS) to identify functionally indispensable genes/protein domains, informing which regions are least likely to tolerate drug-resistance mutations without a fitness cost to the pathogen.

Ask the tutor: How does the McDonald-Kreitman test use polymorphism and divergence data together to test for selection without assuming a strict molecular clock?