The Neutral Substitution Rate

A substitution is different from a mutation. A mutation is a single new variant arising in one individual's genome; a substitution is the fixation of that variant — its eventual spread to 100% frequency, replacing the ancestral allele throughout the whole population (or species). Mutations happen constantly and are mostly lost; substitutions are rare events, each representing a completed sweep through the entire gene pool, whether driven by selection or by chance. The neutral substitution rate, denoted k, is the number of substitutions fixed per site per generation (or per year) at sites where mutations are selectively neutral — that is, they do not change fitness. Kimura's central result, derived in his 1968 and 1969 papers and elaborated with Tomoko Ohta, is:

k = μ where μ is the neutral mutation rate per site per generation. Notice what is absent from the right-hand side: no N (population size), no N_e (Effective Population Size), no measure of genetic diversity. This independence from population size is the single most important and most counterintuitive feature of the result, and it is what makes molecular clocks conceptually possible — a species with a population of 10,000 and one with a population of 10 billion are predicted to fix neutral substitutions at the same rate per generation, provided their mutation rates are equal. It is essential to keep two timescales straight. Measured per generation, k = μ exactly. Measured per year, the rate depends on generation time, since species with shorter generations undergo more mutation-and-fixation cycles per year — this is the basis of the well-documented "generation-time effect" in molecular clock calibration, a real complication addressed later.

Step 1: New neutral mutations enter the population

Consider a diploid population of effective size N_e, with 2N_e gene copies at a given locus. In each generation, the population as a whole introduces a total of

2N_e × μ new mutations at that locus, where μ is the neutral mutation rate per gene copy per generation. Each new mutant starts as a single copy among 2N_e copies, at initial frequency p = 1/(2N_e).

Step 2: Each new mutation has a small, calculable chance of fixation

Because the mutation is neutral, its fate is governed purely by genetic drift — the random sampling of gametes from generation to generation. A cornerstone result from diffusion theory (developed by Kimura using the Kolmogorov backward equation) gives the fixation probability of a neutral allele starting at frequency p in a randomly mating population:

u(p) = p This says, remarkably, that a neutral allele's chance of eventually becoming fixed is simply equal to its current frequency — no more, no less. This makes sense from a symmetry argument: under pure drift, every one of the 2N_e gene copies present today is equally likely to be the ancestor of all future copies (they are exchangeable). Since a new mutation occupies exactly 1 of the 2N_e copies, its fixation probability is

u = 1/(2N_e)

Step 3: Multiply mutations entering × probability each one fixes

The rate of substitution — new fixed differences accumulating per generation — is the product of how many new neutral mutations arise and each one's chance of eventually sweeping to fixation:

k = (2N_e × μ) × (1/(2N_e)) = μ The 2N_e terms cancel exactly. A tenfold larger population produces ten times as many candidate mutations, but each faces ten times worse odds of fixation, since it must out-drift ten times as many competing lineages. The two effects are not approximately equal — they are exactly reciprocal, because both stem from the same quantity, 2N_e, appearing in the mutation-supply term and the drift term.

Why fixation probability equals initial frequency

This deserves its own explanation because it is the mechanistic heart of the result. Under the Neutral Theory of Molecular Evolution, neutral alleles do not differ in reproductive expectation. Track the population's genealogy backward through time: eventually all present-day gene copies coalesce to a single common ancestral copy some generations in the past (a certainty in any finite population absent continual new variation). Since no copy is favored over any other, each of the 2N_e present-day copies is equally likely to be the one whose descendants ultimately take over the whole population. A new mutation, occupying 1 copy, therefore has probability 1/(2N_e) of being that lucky lineage — exactly u(p) = p with p = 1/(2N_e).

Where selection breaks the cancellation

Contrast this with a mutation under selection with selective advantage s. Kimura's diffusion approximation gives, for small s and N_e,

u ≈ 2s / (1 − e^(−4N_e s)) For a beneficial mutation, this fixation probability depends on N_e s, the product of population size and selection coefficient — so population size does not cancel out for selected mutations, and larger populations do fix advantageous substitutions faster (and are much more efficient at eliminating deleterious ones). Neutrality is what forces the special cancellation; it is a very particular, almost fragile, coincidence between the mutation-supply term and the drift term that only holds when s = 0.

Kimura's 1968 argument from protein and DNA divergence data. The neutral theory itself was proposed in response to an empirical puzzle rather than derived purely from theory. In the mid-1960s, protein sequencing (Zuckerkandl and Pauling's work on hemoglobin and cytochrome c across mammals, birds, and fish) revealed that amino acid substitutions accumulated at a rate that looked roughly constant per unit time across very different lineages — the observation that gave rise to the informal idea of a "molecular clock" even before its theoretical basis was understood. Kimura, in his 1968 Nature paper "Evolutionary Rate at the Molecular Level," pointed out that classical selectionist population genetics predicted this constancy would be extraordinary: if most substitutions were adaptive, the rate should depend on the rate of environmental change and the intensity of selection, both wildly variable across lineages with vastly different population sizes, generation times, and ecologies. He calculated that the observed substitution rate in mammalian genomes — roughly one amino acid change per polypeptide per few million years across many genes — implied, if each substitution required a selective "cost" of gene substitution (per Haldane's earlier cost-of-selection argument), an impossibly high genetic load; population sizes could not sustain simultaneous positive selection at the rate required by observed sequence divergence across the whole genome. The neutral theory resolved this: if most substitutions are neutral, k = μ requires no selective death toll at all, because there is no selection acting.

Direct test: comparing synonymous and nonsynonymous rates. The clearest confirmation of k = μ comes from comparing substitution rates at sites under different degrees of functional constraint within the same genes. Because μ is presumed roughly constant across nearby sites (mutation is a chemical/DNA-replication process largely blind to which base changes affect protein function), the neutral prediction is that less functionally constrained sites should show substitution rates closer to the raw mutation rate, while functionally important sites (removed by purifying selection) should show suppressed rates. Kimura and later researchers (notably studies compiling data across many mammalian genes by Wen-Hsiung Li, Masatoshi Nei, and others in the 1980s) found exactly this gradient:

Pseudogenes — duplicated gene copies that have lost function through inactivating mutations — are especially informative, because once a gene is a pseudogene essentially all new mutations within it are neutral (there is no protein product to disrupt). Li and colleagues (1981) showed that processed pseudogenes accumulate substitutions at rates statistically indistinguishable from the estimated germline mutation rate, and comparable to synonymous-site rates in the pseudogene's functional paralog — a direct, quantitative confirmation that k ≈ μ when selection is truly absent.

Population-size independence tested across taxa. A second, harder test of the theory is the population-size prediction itself: species with enormously different N_e (say, a large-N_e marine invertebrate versus a small-N_e large mammal) should show similar neutral substitution rates per generation if their mutation rates are similar, despite orders-of-magnitude differences in effective population size and standing genetic diversity. This is one of the theory's most striking successes and simultaneously one of its trickiest quantitative tests, because genetic diversity (π, related to N_e μ) does scale (weakly) with N_e across species, as neutral theory also predicts, while the substitution rate itself does not track N_e in comparative genomic studies — consistent with the k = μ cancellation. This decoupling of diversity (which does depend on N_e) from divergence rate (which does not) is a signature, quietly confirmed by decades of comparative sequencing, that most polymorphism and most substitution is governed by mutation and drift rather than adaptive sweeps tied to census size.

The Lewontin "paradox of variation" as an indirect corroboration. Richard Lewontin noted in 1974 that genetic diversity varies far less across species than the enormous range of census population sizes would predict under simple neutral expectations (π ≈ 4N_e μ). This "paradox" is now understood to arise partly because N_e itself is strongly reduced below census size by linked selection (background selection and hitchhiking) in large populations — but the underlying neutral machinery (u = p, k = μ) that this article derives remains the correct baseline model against which such departures are measured and quantified.

Example 1 — Comparing two lineages with very different population sizes. Suppose two rodent lineages diverged from a common ancestor. Lineage A (an island population) has effective population size N_e = 2,000; lineage B (a mainland population of the same species complex) has N_e = 500,000. Assume both share the same neutral mutation rate at synonymous sites, μ = 2.2 × 10⁻⁹ per site per generation, and the same generation time of 1 year. Per-generation substitution rate for each lineage, since both are governed by k = μ: k_A = k_B = μ = 2.2 × 10⁻⁹ substitutions per site per generation Despite Lineage B having 250 times more individuals (and therefore 250 times more new neutral mutations entering the population each generation — 2 × 500,000 × μ versus 2 × 2,000 × μ), each individual mutation in lineage B is 250 times less likely to fix (1/1,000,000 versus 1/4,000). The two lineages are predicted to accumulate neutral substitutions at statistically indistinguishable rates over, say, 5 million generations: Expected divergence per site (each lineage from the ancestor) ≈ μt = 2.2 × 10⁻⁹ × 5 × 10⁶ ≈ 0.011, i.e., about 1.1% sequence divergence at synonymous sites. This is exactly the pattern molecular clock studies exploit: a small, historically bottlenecked island population is not expected to show a depressed or elevated substitution rate merely because of its smaller N_e — a common misconception addressed below. (Its diversity, π ≈ 4N_e μ, would be far lower than the mainland population's, but its long-term substitution rate would not.)

Example 2 — Estimating divergence time from synonymous-site substitutions. Two primate species share a coding gene. Sequencing reveals that at 4-fold degenerate synonymous sites (assumed neutral), the two lineages differ at 148 out of 2,000 compared sites (a raw proportion of differences, p_diff = 0.074). Assume a per-generation neutral mutation rate μ = 1.0 × 10⁻⁸ and a generation time of 20 years (so μ per year = 5.0 × 10⁻¹⁰ per site per year). Step 1: Correct for multiple hits at the same site using the Jukes-Cantor formula, since raw proportion differences underestimate true substitutions when p_diff is not small: d = −(3/4) ln[1 − (4/3)p_diff] = −(3/4) ln[1 − (4/3)(0.074)] = −(3/4) ln(0.9013) ≈ −(3/4)(−0.1040) ≈ 0.078 This d is the estimated number of substitutions per site accumulated along both lineages combined since their common ancestor. Step 2: Since substitutions accumulate independently along both branches from the ancestor, divergence time T (in years) satisfies d = 2 × k_year × T, where k_year = μ_year (neutral prediction) 0.078 = 2 × (5.0 × 10⁻¹⁰) × T T = 0.078 / (1.0 × 10⁻⁹) = 7.8 × 10⁷ years ≈ 78 million years This is the basic calculation underlying molecular-clock divergence dating (foreshadowing Molecular Clocks in full): it depends entirely on trusting that synonymous substitutions are effectively neutral, so that k = μ lets sequence divergence be converted directly into elapsed time, once μ is calibrated independently (e.g., from fossil-dated splits or direct germline mutation-rate sequencing studies in parent-offspring trios).

Motoo Kimura developed the mathematical population genetics that makes k = μ rigorous, building on the diffusion-equation methods pioneered earlier by Sewall Wright and Ronald Fisher but turning them toward a new empirical target: the flood of protein and DNA sequence data becoming available in the 1960s. Kimura worked largely at the National Institute of Genetics in Mishima, Japan, and was known for combining formidable analytical skill in stochastic processes with a willingness to challenge the dominant "selectionist" consensus of his era, which held that essentially all observed genetic variation and molecular change reflected the guiding hand of natural selection. Kimura's insight was as much about scientific courage as mathematics: proposing that the vast majority of molecular evolutionary change is selectively neutral, driven by mutation and random genetic drift rather than adaptation, ran directly against the dominant view of evolution as a process centrally defined by Darwinian selection. He was careful to distinguish his claim — that most molecular substitutions are neutral — from any denial that natural selection shapes morphology, physiology, and adaptive traits, a distinction sometimes lost by critics and popularizers alike.

Dating species and gene divergences. Molecular clock methods, calibrated using k ≈ μ at neutral or nearly-neutral sites, remain a primary tool for estimating when lineages split — used across systematics, biogeography, and human origins research, and refined with relaxed-clock Bayesian methods that model rate variation across branches (foreshadowing the Nearly Neutral Theory refinements needed when the strict neutral assumption is only approximate).

Detecting selection in genomes. The neutral expectation k = μ (and its polymorphism analogue, the neutral theory's prediction for the ratio of polymorphism to divergence) is the null hypothesis against which genome scans for selection are tested. Tests like the McDonald-Kreitman test compare the ratio of nonsynonymous to synonymous substitutions between species against the same ratio for polymorphisms within species; deviations from the neutral expectation reveal past adaptive substitutions, positive selection, or purifying selection intensities — directly built on the k = μ baseline.

Forensic and conservation genetics. Estimating mutation rates and expected divergence times from sequence data underlies wildlife forensics (dating poaching events or population splits), conservation genomics (estimating effective population sizes and historical demography from patterns of neutral diversity), and studies of endangered-species population history.

Medical genomics and mutation-rate estimation. Direct sequencing of parent-offspring trios now measures human germline mutation rates directly (roughly 1.0–1.2 × 10⁻⁸ per site per generation), and these estimates are cross-checked against long-term substitution rates inferred from human-chimpanzee divergence — a direct empirical test of k = μ within our own lineage, important for calibrating disease-mutation rate estimates and for understanding baseline rates of de novo mutation relevant to genetic disease.

Viral evolution and epidemiology. Rapidly evolving RNA viruses (influenza, HIV, SARS-CoV-2) are analyzed with molecular clock models rooted in the same neutral-substitution logic, though here selection at antigenic sites is strong and must be modeled explicitly — illustrating exactly where the strict k = μ approximation breaks down and richer nearly-neutral or selection-explicit models are required.

Ask the tutor: Why does the fixation probability of a neutral allele equal its initial frequency, and how does that fact alone force k = μ?