The Nearly Neutral Theory

Every new mutation carries a selection coefficient, s, describing its proportional effect on fitness relative to the existing allele (s > 0 beneficial, s < 0 deleterious, s = 0 neutral — concepts introduced under The Selection Coefficient). Kimura's classical neutral theory treated most molecular variation as falling into one of two extremes: true neutrality (s = 0) or strong, clearly visible deleterious effects removed almost immediately by purifying selection. Ohta pointed out that the overwhelming majority of new mutations affecting protein structure or gene regulation likely have s values that are small in absolute magnitude — neither exactly zero nor large enough to be reliably purged. These are the "nearly neutral" mutations. Whether such a mutation behaves, in practice, like a neutral variant or like a deleterious one is not determined by s in isolation. It is determined by how large s is relative to the strength of genetic drift, and drift's strength scales inversely with Effective Population Size, Ne. The relevant compound parameter is Nes (sometimes written 4Nes or 2Nes depending on convention for diploid/haploid organisms and dominance). When the absolute value of Nes is much less than 1, selection is too weak to reliably overcome the random sampling noise of drift, and the mutation's fixation probability approaches that of a strictly neutral allele — roughly 1/(2N) for a new mutation in a diploid population. When |Nes| is much greater than 1, selection efficiently governs the mutation's fate: beneficial variants sweep, deleterious variants are purged, largely independent of chance. This produces a continuum rather than a dichotomy. A mutation with s = −0.0001 is effectively neutral in a population of Ne = 1,000 (Nes = −0.1, drift wins) but strongly deleterious and reliably eliminated in a population of Ne = 10,000,000 (Nes = −1,000, selection wins). The same mutation, same molecule, same biochemical effect — but a completely different evolutionary outcome, purely because the population size differs. This is the theory's central and somewhat counterintuitive lesson: the strength of selection acting on a given mutation is not a fixed biochemical property; it is a joint property of the mutation and the population it occurs in.

Step 1: Selection coefficients form a continuous spectrum, not two classes. New mutations arising in coding and regulatory DNA produce a distribution of fitness effects (the DFE). Empirical DFEs, estimated from site-frequency spectra and mutation-accumulation experiments, are typically unimodal and skewed, with a mode near zero and a long tail of increasingly deleterious effects; strongly beneficial mutations are rare. Ohta's insight was to focus attention on the large mass of mutations with |s| in the range of roughly 10⁻⁶ to 10⁻². These are mutations too weak to be treated as strongly counterselected under simple purifying-selection models, yet not so weak as to be dismissed as biochemically inert.

Step 2: Fixation probability depends on Nes through diffusion theory. Using diffusion approximations to the Wright-Fisher model (the same machinery underlying classical population genetics), the fixation probability of a new mutation with selection coefficient s in a diploid population of effective size Ne is:

for a semi-dominant (additive) mutation, where 2Nes is the compound parameter that actually enters the formula. Two limits fall out cleanly:

• If Nes ≈ 0 (either s is tiny or Ne is small), Pfix ≈ 1/(2N) — identical to the neutral fixation probability. Chance dominates.
• If Nes is large and negative (deleterious, and either s is not-so-small or Ne is large), Pfix shrinks exponentially toward zero. Selection dominates.

The transition between these regimes is not a hard threshold; it is smooth, governed by the magnitude of Nes. Population geneticists commonly use |Nes| ≈ 1 as an informal boundary of the "nearly neutral zone" — not a discontinuity, but the region where drift and selection contribute comparably to the outcome.

Step 3: Effective population size is the lever, and it varies enormously across genomes and lineages. Ne is not the census population size; it reflects breeding structure, variance in reproductive success, historical bottlenecks, linked selection, and mode of inheritance. Mitochondrial and chloroplast genomes, inherited uniparentally and without recombination, have roughly a quarter of the Ne of nuclear autosomes. Self-fertilizing plants and asexual lineages have reduced effective recombination and thus reduced effective Ne due to processes like background selection and Hill-Robertson interference. Large-bodied, long-lived, low-fecundity species (whales, elephants, raptors) tend to have small long-term Ne; small-bodied, high-fecundity species (many insects, marine invertebrates, bacteria) tend to have enormous Ne. Because Nes is the product of Ne and s, any factor that shrinks Ne effectively shifts more of the mutational spectrum into the nearly-neutral zone — even without any change in the mutations' underlying biochemistry.

Step 4: The consequence is a testable, quantitative prediction for nonsynonymous substitution rates. Silent (synonymous) substitutions are, to a good approximation, unaffected by this logic — they alter mRNA sequence but not (mostly) protein sequence, so their fixation rate approximately equals the neutral mutation rate regardless of Ne (this is the baseline established for The Neutral Substitution Rate). Nonsynonymous (protein-changing) substitutions, by contrast, are precisely the class enriched for mildly deleterious effects. The nearly neutral theory therefore predicts:

because small-Ne lineages fix a larger fraction of the mildly deleterious nonsynonymous mutations that arise, while large-Ne lineages purge them more efficiently. This single prediction — an inverse relationship between Ne and dN/dS — has become one of the most tested and most consistently supported predictions in molecular evolution.

Step 5: Linked selection can shrink local Ne along the genome. A further mechanism widens or narrows the nearly neutral zone locally within a single genome. Regions of low recombination experience stronger Hill-Robertson interference and background selection, both of which reduce the local effective population size experienced by a linked neutral or nearly neutral site. This produces heterogeneity in dN/dS and diversity across the genome even within one species — low-recombination regions (near centromeres, on non-recombining sex chromosomes, in asexual genomes) behave, locally, like populations with smaller Ne, accumulating more slightly deleterious substitutions than freely-recombining regions of the same genome.

Ohta's founding argument (1973): rate constancy that didn't fit strict neutrality. Kimura and Jack Lewontin King, and independently King and Jukes, had shown that many proteins evolve at roughly clock-like rates across lineages, consistent with a neutral substitution rate approximating the mutation rate. But an anomaly troubled the strict neutral model: generation-time effects. If most amino-acid substitutions were strictly neutral, the substitution rate per year should scale with the mutation rate per generation divided by generation time — organisms with short generations (mice) should accumulate mutations, and thus substitutions, much faster per calendar year than long-generation organisms (humans, elephants). Yet comparative protein and DNA sequence data (hemoglobin, cytochrome c, and later genome-wide comparisons) showed the generation-time effect was much weaker than expected under strict neutrality, and was markedly weaker for nonsynonymous than for synonymous substitutions. Ohta (1973, Nature) proposed that this discrepancy arises because nonsynonymous substitutions are disproportionately composed of slightly deleterious mutations whose fixation is governed by Ne, not simply by the mutation rate — and Ne tends to be inversely related to generation time and body size across mammals, partially canceling the generation-time effect on the mutation-input side. This was not a hand-wave; it was a quantitative resolution of a specific, previously unexplained empirical pattern in the comparative sequence data available at the time.

The dN/dS–Ne comparative test (multiple independent studies, 1990s–2010s). The theory's sharpest quantitative prediction — that dN/dS should rise as Ne falls — has now been tested across an enormous range of taxa using genome-scale sequence comparison:

• Mammals versus other vertebrates and invertebrates. Large-scale genomic comparisons (e.g., analyses led by groups including Adam Eyre-Walker, Peter Keightley, and later large consortium genome papers) consistently find that large-bodied mammals with small long-term Ne (humans, great apes, cetaceans) show elevated dN/dS relative to small-bodied, large-Ne taxa such as mice, and especially relative to invertebrates like Drosophila, whose dN/dS is markedly lower.
• Mitochondrial versus nuclear genomes. Because mtDNA is haploid, uniparentally inherited, and non-recombining, its effective population size is roughly one-quarter that of an autosomal nuclear locus in the same organism (ignoring differences in mutation rate). Consistent with the theory, animal mitochondrial genomes typically show higher dN/dS than nuclear genes in the same lineage, and accumulate mildly deleterious variants that would be purged if they occurred in the higher-Ne nuclear genome.
• Self-fertilizing versus outcrossing plants. Selfing dramatically reduces effective recombination and effective population size (through increased linkage and stronger genetic drift within predominantly homozygous lineages). Comparative genomic studies of selfing plant species (e.g., Arabidopsis thaliana versus its outcrossing relative Arabidopsis lyrata, and Capsella rubella versus outcrossing Capsella grandiflora) find higher dN/dS and elevated proportions of derived deleterious alleles in the selfers, exactly as predicted.
• Island versus mainland and endangered-species genomics. Conservation genomic surveys of small, isolated, or bottlenecked populations (Channel Island foxes, Isle Royale wolves, Iberian lynx) repeatedly find elevated ratios of nonsynonymous to synonymous heterozygosity and higher relative mutational load, tracking historical reductions in Ne.

Falsifiable structure of the test. Crucially, the theory makes a directional, quantitative prediction (dN/dS should track 1/Ne), not merely "some mutations are nearly neutral." Studies estimate Ne independently — typically from silent-site nucleotide diversity θ = 4Neμ, using the neutral substitution/diversity framework — and then test whether dN/dS correlates with this independent Ne estimate across many species. The consistent negative correlation between independently estimated Ne and dN/dS, replicated across mammals, plants, and invertebrates by multiple independent research groups using different genomic datasets, constitutes the primary empirical support for the nearly neutral theory. This is comparative correlational evidence rather than a single controlled experiment (population genetics rarely allows Meselson–Stahl–style direct manipulation at this scale), but the convergence of independent lines — generation-time anomalies, mitochondrial/nuclear contrasts, selfer/outcrosser contrasts, and conservation genomic case studies — gives the theory unusually broad empirical backing for a population-genetic model.

Direct experimental support: mutation accumulation lines. A complementary experimental line of evidence comes from mutation-accumulation (MA) experiments, in which laboratory populations are deliberately bottlenecked to Ne of 1–10 individuals per generation for many generations (classically in Drosophila, Caenorhabditis elegans, and Saccharomyces cerevisiae, pioneered by researchers including Michael Lynch). Under such extreme drift, selection coefficients that would normally be purged in wild-sized populations become invisible to selection, and MA lines show measurable, cumulative declines in fitness — a direct experimental demonstration that shrinking Ne allows mildly deleterious mutations to fix, exactly as Nes theory predicts. Lynch's extensive MA work, alongside genomic surveys, explicitly frames genome architecture (intron size, gene density, repetitive DNA content) as partly a product of long-term differences in Ne acting on nearly-neutral insertions and deletions — organisms with small Ne (many eukaryotes) tolerate more "junk," while organisms with enormous Ne (many bacteria, some yeast) have more streamlined genomes because slightly deleterious non-functional DNA is efficiently purged.

Example 1 — A bacterial population genetics calculation. Suppose a bacterial species has an effective population size of Ne = 10⁸ (fairly typical for a free-living bacterium with a large, well-mixed population and no strong recent bottleneck). A new mutation in a regulatory region reduces fitness by s = −2 × 10⁻⁸ per generation — an extremely small effect, biochemically almost undetectable in the lab. Compute Nes:

Since |Nes| = 2, this mutation sits just past the boundary of the nearly neutral zone — selection is weak but non-negligible; the mutation is somewhat more likely to be lost than a strictly neutral variant, though not overwhelmingly so. Now compare the same mutation (same s) occurring in a small, bottlenecked bacterial population, say a chronic-infection population within a single host with Ne = 10⁴ (a scenario documented in some persistent infections, such as Pseudomonas aeruginosa in cystic fibrosis lungs):

Here |Nes| ≪ 1: the mutation is, for practical evolutionary purposes, neutral. It can drift to fixation within the host population with essentially the same probability as a truly neutral allele, ≈ 1/(2Ne). This is not a hypothetical curiosity — it is precisely the logic used to explain why bacterial populations undergoing chronic, bottlenecked infection (or serial single-colony transfer in evolution experiments) accumulate more slightly deleterious mutations, including loss-of-function changes in genes still under weak selection, than the same species does in a large, well-mixed free-living population. The compound parameter Nes, not s alone, decided the outcome.

Example 2 — Reading a comparative genomic dataset. A researcher compares four mammal species' genome-wide average dN/dS ratios (nonsynonymous divergence over synonymous divergence, a standard proxy for the relative efficacy of purifying selection) alongside independent estimates of long-term effective population size from synonymous-site nucleotide diversity (θS = 4Neμ):

(Figures are illustrative, order-of-magnitude representative of published comparative estimates, not exact published values.) Reading this table through the nearly neutral lens: as Ne falls roughly 200-fold from mouse to killer whale, dN/dS nearly triples. This is the qualitative signature the theory predicts — smaller Ne widens the nearly neutral zone, so a larger fraction of nonsynonymous mutations that arise are fixed rather than purged, inflating dN/dS even though the underlying distribution of fitness effects (the DFE) may be similar across these related mammals. A student should resist two wrong readings of this table. First, it does not mean killer whales are "evolving faster" in an adaptive sense — the higher dN/dS mostly reflects relaxed purifying selection (drift fixing mildly deleterious variants), not more frequent positive selection; distinguishing the two requires methods that separate a genuinely elevated rate of adaptive substitution from an elevated rate of deleterious fixation, a key motivation for later methods used in Detecting Selection in Sequences such as the McDonald–Kreitman test. Second, it does not mean mice are somehow "more perfect" — mice simply have a larger Ne, so selection more efficiently screens out the same class of mildly harmful mutations that drift permits to fix in whales.

Tomoko Ohta (b. 1933) is a Japanese population geneticist who worked for most of her career at the National Institute of Genetics in Mishima, Japan, closely alongside Motoo Kimura, the architect of the neutral theory. Trained originally in agricultural science and later in population genetics (including graduate work in the United States), Ohta brought a quantitative, model-building rigor to a question that Kimura's framework had left partly unresolved: why did the molecular clock show systematic irregularities, especially tied to generation time and taxon, that pure neutrality struggled to explain? Rather than abandoning the neutral framework, she extended it, publishing her foundational nearly neutral model in a short but consequential 1973 paper in Nature, and then developing its mathematical and empirical implications across several decades of subsequent work, including influential treatments of gene duplication, multigene family evolution, and the role of population size in genome architecture. Ohta's career is notable for having built one of the field's most durable theoretical frameworks from a relatively narrow starting anomaly — an inconsistency in generation-time scaling of substitution rates — and generalizing it into a broadly applicable principle connecting molecular evolution, population size, and genome structure. Her work has been recognized with major honors including the Crafoord Prize in Biosciences (2015, shared with Richard Lewontin), the Prize often considered biology's counterpart to fields lacking a Nobel category, along with membership in the U.S. National Academy of Sciences and Japan's Academy.

Conservation genomics. Endangered and historically bottlenecked populations — Iberian lynx, Isle Royale wolves, Channel Island foxes, mountain gorillas — show elevated ratios of nonsynonymous to synonymous variation and increased homozygosity for putatively deleterious alleles, exactly as nearly neutral theory predicts for reduced Ne. Conservation geneticists now use genomic estimates of "realized" and "masked" genetic load, grounded in this framework, to help decide whether genetic rescue (introducing migrants to restore Ne and heterozygosity) is warranted, and to forecast extinction risk from mutational meltdown — a process in which shrinking Ne and rising deleterious fixation reinforce each other in a downward spiral.

Human medical genetics. Human Ne is modest by mammalian standards (on the order of 10,000), which the theory predicts should leave humans carrying a comparatively high burden of mildly deleterious coding variants relative to species with larger Ne. Large human sequencing projects (gnomAD, UK Biobank, and earlier 1000 Genomes analyses) confirm exactly this: every human genome carries thousands of predicted-damaging missense variants, most rare and individually mildly deleterious, consistent with a population too small (in evolutionary terms) to purge them all efficiently. This framework helps interpret why rare disease variants persist at low but non-zero frequency and informs models of complex-disease genetic architecture.

Livestock, crop, and captive-breeding management. Domesticated and captive populations routinely experience severe Ne reductions from founder effects and selective breeding. Nearly neutral logic predicts, and genomic surveys confirm, elevated deleterious mutation load in many domesticated lineages relative to wild ancestors — a consideration actively used in breeding-program design to manage inbreeding depression and maintain genetic diversity in both agricultural and zoo populations.

Foundational tool for detecting selection. Perhaps the theory's broadest modern legacy is methodological: because it specifies exactly how drift and weak selection jointly shape dN/dS and the site-frequency spectrum, it supplies the null expectation against which genuine positive selection must be distinguished. Tests such as the McDonald–Kreitman test, extensions estimating the distribution of fitness effects (DFE-alpha and related methods), and genome scans for adaptive substitution all build on the nearly neutral baseline — making this theory the conceptual bridge into the modern toolkit covered under Detecting Selection in Sequences.

Ask the tutor: How does the nearly neutral theory's Nes framework connect to the McDonald–Kreitman test used for detecting positive selection in sequences?