19.04.02 · eco-evo-bio / drift

Neutral theory: Kimura's neutral theory and synonymous versus nonsynonymous substitution rates

stub3 tiersLean: nonepending prereqs

Anchor (Master): Kimura, M. — The Neutral Theory of Molecular Evolution (1983)

Intuition Beginner

Motoo Kimura proposed that most genetic differences between species are not driven by natural selection but by random genetic drift acting on neutral mutations -- ones that neither help nor harm the organism. This "neutral theory" explains why DNA changes accumulate at a roughly constant rate (the molecular clock). Mutations that don't change the protein (synonymous) accumulate faster than those that do (nonsynonymous), because protein changes are more likely to be harmful.

Visual Beginner

When $d_{N} / d_{S} ≪ 1$ , purifying selection removes most amino-acid changes. When $d_{N} / d_{S} \approx 1$ , changes are neutral. When $d_{N} / d_{S} > 1$ , positive selection drives amino-acid change faster than the neutral rate.

Worked example Beginner

A gene has accumulated 12 synonymous substitutions over 10 million years across 300 synonymous sites, and 3 nonsynonymous substitutions across 700 nonsynonymous sites. Calculate $d_{S}$ , $d_{N}$ , and the $d_{N} / d_{S}$ ratio.

$d_{S} = 12/300 = 0.04$ substitutions per synonymous site.

$d_{N} = 3/700 \approx 0.0043$ substitutions per nonsynonymous site.

$d_{N} / d_{S} = 0.0043/0.04 \approx 0.11$ .

The ratio is well below 1, indicating strong purifying selection: most amino-acid-changing mutations are harmful and removed before they can fix.

Check your understanding Beginner

Formal definition Intermediate+

The neutral substitution rate

Kimura's (1968) key result: the rate of neutral substitution equals the mutation rate and is independent of population size. In a diploid population of effective size $N_{e}$ with per-generation per-gene-copy mutation rate $μ$ to new neutral alleles, the number of new neutral mutations per generation is $2 N_{e} μ$ . Each fixes with probability $1/ (2 N_{e})$ 19.04.01. The substitution rate is:

k = 2 N_{e} μ \times \frac{1}{2 N _{e}} = μ .

The $N_{e}$ cancels exactly. This is the molecular clock: the rate of neutral molecular evolution is set by the biochemical error rate of DNA replication, not by population size or ecology.

The molecular clock and rate heterogeneity

If $k = μ$ per generation, species with shorter generation times should accumulate substitutions faster per year (more generations per unit time). Observed data broadly support this for synonymous sites but show weaker generation-time effects for nonsynonymous sites -- consistent with the nearly neutral theory (see Master section).

The clock is not perfectly constant. Rate heterogeneity arises from:

Variation in mutation rate across lineages (different DNA repair efficiencies, metabolic rates).
Variation in generation time.
Changes in effective population size altering the fate of nearly neutral mutations.

Synonymous and nonsynonymous substitution rates

A synonymous (silent) mutation does not change the encoded amino acid (redundancy in the genetic code). A nonsynonymous mutation changes the amino acid. Let $d_{S}$ be the number of synonymous substitutions per synonymous site and $d_{N}$ the number of nonsynonymous substitutions per nonsynonymous site.

The $d_{N} / d_{S}$ ratio (also written $ω$ ) classifies selective regimes:

$d_{N} / d_{S}$	Interpretation
$≪ 1$	Purifying selection: most amino-acid changes are deleterious and removed
$\approx 1$	Neutral evolution: amino-acid changes fix at the same rate as silent changes
$> 1$	Positive selection: amino-acid changes are beneficial and fix faster than neutral rate

Computing $d_{S}$ and $d_{N}$ requires correcting for the fact that most nucleotide substitutions at a codon are nonsynonymous at the pathway level. Standard methods include:

Nei-Gojobori (1986): counts synonymous and nonsynonymous sites and differences with a correction for multiple hits (Jukes-Cantor or Kimura 2-parameter).
Yang-Nielsen (2000): uses a codon-based maximum-likelihood model (implemented in PAML's codeml) that estimates $d_{N} / d_{S}$ while accounting for transition-transversion bias and codon frequency.

The McDonald-Kreitman test

The McDonald-Kreitman (MK) test (1991) compares the ratio of synonymous to nonsynonymous variation within a species (polymorphism) to the same ratio between species (divergence). Under strict neutrality:

\frac{d _{N}}{d _{S}}_{divergence} = \frac{P _{N}}{P _{S}}_{polymorphism},

where $P_{N}$ and $P_{S}$ are counts of nonsynonymous and synonymous polymorphic sites. The test constructs a $2 \times 2$ contingency table:

	Synonymous	Nonsynonymous
Fixed (divergence)	$D_{S}$	$D_{N}$
Polymorphic	$P_{S}$	$P_{N}$

Under neutrality, the ratio $D_{N} / D_{S}$ equals $P_{N} / P_{S}$ . An excess of nonsynonymous divergence ( $D_{N} / D_{S} > P_{N} / P_{S}$ ) indicates positive selection driving amino-acid substitutions. An excess of nonsynonymous polymorphism ( $P_{N} / P_{S} > D_{N} / D_{S}$ ) indicates either balancing selection maintaining amino-acid variants or weakly deleterious mutations segregating at low frequency (nearly neutral effects).

The neutrality index $N I = (P_{N} / P_{S}) / (D_{N} / D_{S})$ summarises the result: $N I = 1$ under neutrality, $N I < 1$ suggests positive selection, $N I > 1$ suggests purifying selection or balancing selection.

Tajima's D

Tajima's D (1989) compares two estimators of $θ = 4 N_{e} μ$ 19.04.01:

$\hat{θ}_{π} = π$ : the average number of pairwise differences between sequences.
$\hat{θ}_{W} = S / a_{n}$ : Watterson's estimator based on the number of segregating sites $S$ , where $a_{n} = \sum_{i = 1}^{n - 1} 1/ i$ .

Under neutrality, both estimate the same $θ$ , so $D = (\hat{θ}_{π} - \hat{θ}_{W}) / SE$ has expected value 0.

Tajima's D	Interpretation
$D \approx 0$	Consistent with neutrality
$D ≪ 0$	Excess of rare variants: population expansion or purifying selection
$D ≫ 0$	Deficit of rare variants: population bottleneck or balancing selection

Tajima's D is sensitive to the site frequency spectrum. A significantly negative D indicates more singletons and low-frequency variants than expected under the standard neutral model (equilibrium, constant population size). A significantly positive D indicates fewer rare variants than expected, consistent with balancing selection maintaining intermediate-frequency alleles or a recent bottleneck that pruned rare variants.

Polymorphism versus divergence

Polymorphism (variation within a species) and divergence (fixed differences between species) reflect different time scales. Polymorphisms are transient: they arise by mutation and are eventually lost or fixed by drift. Divergence accumulates over the entire history since the common ancestor. Under neutrality, the ratio of polymorphism to divergence should be the same at synonymous and nonsynonymous sites (the MK prediction). Discrepancies between the two reveal the action of selection on different time scales.

Key theorem with proof Intermediate+

Theorem (Neutral substitution rate). In a diploid population of effective size $N_{e}$ with per-generation per-gene-copy neutral mutation rate $μ$ , the rate of neutral substitution is $k = μ$ , independent of $N_{e}$ .

Proof. The number of new neutral mutations arising per generation is $2 N_{e} μ$ : each of the $2 N_{e}$ gene copies mutates to a new neutral allele with probability $μ$ . Each new mutation starts at frequency $p_{0} = 1/ (2 N_{e})$ . By the martingale property of neutral drift 19.04.01, the fixation probability of a neutral allele equals its initial frequency:

P_{fix} = p_{0} = \frac{1}{2 N _{e}} .

The expected number of substitutions per generation is the product of the number of new mutations and the fixation probability:

k = 2 N_{e} μ \times \frac{1}{2 N _{e}} = μ .

The $N_{e}$ dependence cancels: larger populations produce more mutations but each has proportionally lower fixation probability. $□$

Corollary (Molecular clock). The rate of neutral substitution per year is $k_{year} = μ / g$ , where $g$ is the generation time in years. If $μ$ and $g$ are roughly constant across lineages, neutral substitutions accumulate at a constant rate per year -- the molecular clock.

Bridge. This result builds directly on the fixation probability derived in 19.04.01. The substitution rate $k = μ$ is the foundational prediction of neutral theory and the basis for all downstream tests (dN/dS, MK, Tajima's D). The coalescent 19.04.03 pending provides the genealogical framework for studying the polymorphism side of neutral theory -- the distribution of variants within a population rather than the rate of fixation between species.

Exercises Intermediate+

Exercise 3 (medium, numeric).

A comparison between human and chimpanzee orthologs at a gene reveals $D_{S} = 30$ synonymous fixed differences, $D_{N} = 10$ nonsynonymous fixed differences, $P_{S} = 15$ synonymous polymorphisms, and $P_{N} = 6$ nonsynonymous polymorphisms within humans. Perform the McDonald-Kreitman test. Is there evidence for selection?

Hint

Construct a $2 \times 2$ table and compute the $G$ -test or Fisher's exact test. Under neutrality, $D_{N} / D_{S} = P_{N} / P_{S}$ .

Answer

$2 \times 2$ table:

	Synonymous	Nonsynonymous
Fixed	30	10
Polymorphic	15	6

Expected under independence: $D_{N} / D_{S} = 10/30 = 0.333$ , while $P_{N} / P_{S} = 6/15 = 0.400$ . The neutrality index $N I = (6/15) / (10/30) = 0.4/0.333 = 1.2$ , close to 1.

$G$ -test: $G = 2 [D_{N} ln (D_{N} / E_{D N}) + D_{S} ln (D_{S} / E_{D S}) + P_{N} ln (P_{N} / E_{P N}) + P_{S} ln (P_{S} / E_{P S})]$ where expected values are $E_{D N} = 16 \times 40/61 = 10.49$ , $E_{D S} = 45 \times 40/61 = 29.51$ , $E_{P N} = 16 \times 21/61 = 5.51$ , $E_{P S} = 45 \times 21/61 = 15.49$ .

$G \approx 2 [10 ln (10/10.49) + 30 ln (30/29.51) + 6 ln (6/5.51) + 15 ln (15/15.49)] \approx 2 [- 0.48 + 0.49 + 0.52 - 0.48] \approx 0.10$ .

With 1 degree of freedom, $p > 0.05$ . No significant evidence for selection at this gene: the data are consistent with neutrality.

Exercise 4 (medium, numeric).

A sample of $n = 20$ sequences has 15 segregating sites and an average pairwise difference of $π = 3.2$ . Compute Watterson's estimator $\hat{θ}_{W}$ , Tajima's estimator $\hat{θ}_{π}$ , and Tajima's D (use $a_{20} = \sum_{i = 1}^{19} 1/ i \approx 3.548$ and $SE \approx V$ where $V \approx 1.05$ for these parameters).

Hint

$\hat{θ}_{W} = S / a_{n}$ and $\hat{θ}_{π} = π$ . Tajima's $D = (\hat{θ}_{π} - \hat{θ}_{W}) / SE$ .

Answer

$\hat{θ}_{W} = 15/3.548 \approx 4.23$ . $\hat{θ}_{π} = 3.2$ .

$D = (3.2 - 4.23) / 1.05 = - 1.03/1.025 \approx - 1.00$ .

Tajima's D of $- 1.0$ is negative but not significant at the 5% level (significance requires roughly $∣ D ∣ > 1.5$ -- $2.0$ for $n = 20$ ). The slight excess of rare variants (segregating sites relative to pairwise differences) is within neutral expectation. A population expansion could produce this pattern but the evidence is not strong.

Exercise 5 (hard, symbolic).

Show that the McDonald-Kreitman test reduces to testing whether the ratio $D_{N} / D_{S}$ differs from $P_{N} / P_{S}$ , and derive the neutrality index $N I = (P_{N} \cdot D_{S}) / (P_{S} \cdot D_{N})$ . Interpret the cases $N I < 1$ , $N I = 1$ , and $N I > 1$ .

Hint

Under neutrality, the probability that a mutation is nonsynonymous is the same for polymorphic and fixed variants. Set up the contingency table and express independence as a ratio equality.

Answer

Under neutrality, both polymorphic and fixed variants are random draws from the same underlying distribution of mutation types. The probability that a random variant is nonsynonymous is the same within and between species. This gives:

\frac{D _{N}}{D _{N} + D _{S}} = \frac{P _{N}}{P _{N} + P _{S}},

which simplifies to $D_{N} / D_{S} = P_{N} / P_{S}$ .

The neutrality index is the ratio of these two quantities:

N I = \frac{P _{N} / P _{S}}{D _{N} / D _{S}} = \frac{P _{N} \cdot D _{S}}{P _{S} \cdot D _{N}} .

$N I = 1$ : the nonsynonymous-to-synonymous ratio is the same within and between species, consistent with neutrality.
$N I < 1$ : the ratio of nonsynonymous to synonymous variants is lower within species than between species, meaning relatively more nonsynonymous substitutions have fixed. This indicates positive selection driving amino-acid change to fixation.
$N I > 1$ : the ratio is higher within species, meaning nonsynonymous variants are disproportionately polymorphic. This indicates purifying selection preventing amino-acid changes from fixing (many slightly deleterious mutations segregate transiently), or balancing selection maintaining amino-acid polymorphism.

Exercise 6 (hard, numeric).

A population of $N_{e} = 50, 000$ has neutral mutation rate $μ = 2 \times 1 0^{- 9}$ per site per generation and generation time $g = 25$ years. Two species diverged $T = 6$ million years ago. Predict the expected neutral divergence (substitutions per site) between the two species.

Hint

Neutral substitution rate per generation $k = μ$ . Per year: $k_{year} = μ / g$ . Total divergence accumulates along two lineages.

Answer

$k = μ = 2 \times 1 0^{- 9}$ per site per generation. Per year: $k_{year} = 2 \times 1 0^{- 9} /25 = 8 \times 1 0^{- 11}$ per site per year.

Total divergence $d = 2 \times k_{year} \times T = 2 \times 8 \times 1 0^{- 11} \times 6 \times 1 0^{6} = 9.6 \times 1 0^{- 4} = 0.096%$ per site.

For a gene of 1000 bp, this predicts about 0.96 substitutions between the two species -- roughly 1 difference per gene. This is consistent with the observed low divergence between humans and chimpanzees at neutrally evolving sites (about 1.2% genome-wide, much of which is neutral).

Nearly neutral theory and the generation-time hypothesis Master

Ohta (1973) recognised that mutations with small but nonzero fitness effects ( $∣ s ∣ \sim 1/ (2 N_{e})$ ) behave differently across populations of different size. In large populations ( $2 N_{e} ∣ s ∣ ≫ 1$ ), selection efficiently removes slightly deleterious mutations and fixes slightly beneficial ones. In small populations ( $2 N_{e} ∣ s ∣ ≪ 1$ ), the same mutations behave as effectively neutral, drifting to fixation with probability $1/ (2 N_{e})$ . This nearly neutral theory predicts:

A negative correlation between substitution rate at nonsynonymous sites and effective population size. Species with large $N_{e}$ (e.g., marine bacteria, Drosophila) show lower $d_{N} / d_{S}$ than species with small $N_{e}$ (e.g., primates, island endemics), because selection in large populations removes a broader class of weakly deleterious mutations.
A weaker generation-time effect at nonsynonymous sites than synonymous sites. Synonymous substitutions (predominantly neutral) should follow the strict molecular clock per generation, so per-year rate scales as $1/ g$ . Nonsynonymous substitutions are influenced by selection whose efficiency depends on $N_{e}$ , which is itself correlated with life-history traits including generation time. Long-lived species tend to have smaller $N_{e}$ , so slightly deleterious nonsynonymous mutations drift more often, partially offsetting the generation-time slowdown. This explains why the molecular clock at nonsynonymous sites is more "clock-like" per year than per generation across taxa with different life histories.
The rate of molecular evolution should exceed the rate of morphological evolution. Morphological evolution requires mutations with large phenotypic effects, which are more likely to be under selection. Molecular evolution proceeds by the accumulation of neutral and nearly neutral changes at the sequence level, most of which have no detectable phenotypic effect. The dissociation between molecular and morphological rates is a direct prediction of the neutral and nearly neutral frameworks: most molecular change is invisible to selection and accumates steadily by drift, while morphological change requires the occasional fixation of mutations with large selective effects.

Selection on synonymous sites: codon usage bias

Synonymous mutations are not always neutral. In organisms with large $N_{e}$ (bacteria, yeast, Drosophila), natural selection favours codons that match abundant tRNA species, increasing translational efficiency and accuracy. This codon usage bias produces $d_{S}$ below the neutral expectation at highly expressed genes, where selection for optimal codons is strongest. The strength of codon usage bias is quantified by the effective number of codons $N_{c}$ (ranging from 20, all codons equally used, to 61, maximum bias) or by the codon adaptation index (CAI).

The implication for $d_{N} / d_{S}$ analysis: synonymous sites are not always a valid neutral baseline in organisms with strong codon bias. Methods that assume $d_{S}$ reflects the neutral rate may underestimate the true neutral rate and thus overestimate $d_{N} / d_{S}$ , potentially masking signals of purifying selection. Correcting for codon bias requires either restricting analysis to genes without strong bias, using intronic or intergenic flanking regions as a neutral reference, or incorporating codon frequencies into the substitution model.

Resolution of the neutralist-selectionist debate

The original debate between neutralists (Kimura, King and Jukes) and selectionists (Gillespie, Lewontin) centred on whether molecular variation is predominantly neutral or maintained by selection. The resolution, now supported by genomic data, is nuanced:

Purifying selection is pervasive. The vast majority of new nonsynonymous mutations are deleterious and removed. This is not controversial; both sides agree on this.
Positive selection drives some amino-acid substitutions but a minority. Genome-wide estimates from MK-type analyses suggest roughly 10--30% of amino-acid substitutions between species are driven by positive selection, with the remainder being neutral or nearly neutral.
Synonymous and noncoding variation is predominantly neutral or nearly neutral, with exceptions in high- $N_{e}$ species where codon bias and regulatory selection operate.
Nearly neutral mutations with $∣ s ∣ \sim 1/ (2 N_{e})$ constitute a substantial fraction of segregating variation, particularly in species with moderate to small $N_{e}$ .

The debate is resolved not by one side winning but by recognising that the relative importance of drift and selection depends on the class of mutation (synonymous vs nonsynonymous vs regulatory), the effective population size, and the time scale of observation (polymorphism vs divergence).

Genomic signatures of selection and neutrality tests in practice Master

Selective sweeps

A hard selective sweep occurs when a strongly beneficial mutation arises and rapidly fixes, carrying linked neutral variation to fixation along with it (genetic hitchhiking). The signature is a region of reduced polymorphism and high linkage disequilibrium (LD) around the selected site. The width of the sweep footprint is approximately $s / ln (2 N_{e} s)$ in recombination units: strong selection on a new mutation with large $s$ in a region of low recombination produces a wide valley of diversity.

A soft sweep occurs when multiple beneficial mutations arise independently (either at the same site from recurrent mutation or at linked sites from standing variation) and contribute to adaptation. Soft sweeps leave a weaker diversity footprint than hard sweeps because multiple haplotypes carry the beneficial allele, preserving some variation. Detecting soft sweeps requires haplotype-based statistics (haplotype homozygosity, iHS, XP-EHH) rather than diversity-based statistics (Tajima's D).

Background selection (Charlesworth et al. 1993) is the removal of linked neutral variation by purifying selection against deleterious mutations. Like selective sweeps, background selection reduces diversity, but it does so continuously rather than in a single event. The effect is strongest in regions of low recombination and high functional density (e.g., near centromeres). Background selection reduces $N_{e}$ locally, which means neutral diversity is correlated with recombination rate -- a pattern observed in Drosophila, humans, and many other species. Distinguishing background selection from selective sweeps is a practical challenge: both reduce diversity in low-recombination regions, but background selection is a genome-wide steady-state process while sweeps produce localized, transient signals.

The HKA test

The Hudson-Kreitman-Aguade (HKA) test (1987) compares levels of polymorphism and divergence at two or more loci. Under neutrality, the ratio of polymorphism to divergence should be the same across loci (both are proportional to $θ = 4 N_{e} μ$ , and the ratio cancels $μ$ ). A locus with significantly reduced polymorphism relative to its divergence shows evidence of a selective sweep; a locus with excess polymorphism suggests balancing selection.

The test statistic is:

X^{2} = i = 1 \sum L \frac{( S _{i} - E [ S _{i} ] ) ^{2}}{E [ S _{i} ]} + \frac{( D _{i} - E [ D _{i} ] ) ^{2}}{E [ D _{i} ]},

summed over $L$ loci, where $S_{i}$ is the number of segregating sites at locus $i$ , $D_{i}$ is the divergence at locus $i$ , and expectations are computed under the neutral model with a single shared $N_{e}$ but locus-specific $μ_{i}$ . The test assumes free recombination within loci and no recombination between loci, approximations that can inflate Type I error.

Fay and Wu's H

Fay and Wu's H (2000) detects an excess of high-frequency-derived alleles, the signature of a recent selective sweep. It uses a third estimator of $θ$ :

\hat{θ}_{H} = i = 1 \sum S \frac{2 i ^{2}}{n ( n - 1 )},

where $i$ is the number of sampled chromosomes carrying the derived allele and $n$ is the sample size. Fay and Wu's $H = \hat{θ}_{π} - \hat{θ}_{H}$ .

Under neutrality, $H \approx 0$ . After a sweep, derived alleles that hitchhike with the beneficial mutation reach high frequency, making $\hat{θ}_{H}$ large and $H$ significantly negative. Fay and Wu's H complements Tajima's D: D is sensitive to an excess of rare variants (low-frequency alleles), while H is specifically sensitive to an excess of high-frequency-derived alleles. Both can be negative after a sweep, but H is more specific because it uses the ancestral/derived state information (requiring an outgroup to polarise mutations).

The HKH test

The HKH expansion combines the HKA framework with Fay and Wu's H to test for selection using both polymorphism-divergence and site-frequency-spectrum information simultaneously. It extends the HKA test to incorporate derived-allele-frequency information, increasing power to detect both sweeps and balancing selection.

Neutrality tests in practice: DnaSP and PAML

DnaSP (DNA Sequence Polymorphism) is the standard GUI-based software for computing neutrality statistics from multiple sequence alignments. It calculates Tajima's D, Fu and Li's D and F, Fay and Wu's H, haplotype diversity, recombination estimates, and MK tests from aligned sequence data. Input: FASTA alignment. Output: summary statistics with coalescent-based confidence intervals obtained by simulation under the standard neutral model.

PAML (Phylogenetic Analysis by Maximum Likelihood; Yang 2007) implements codon-based models for estimating $d_{N} / d_{S}$ along branches of a phylogeny. Key models include:

M0 (single ratio): one $d_{N} / d_{S}$ for all branches and sites.
M1a (neutral): two site classes ( $ω = 0$ and $ω = 1$ ).
M2a (selection): adds a third class with $ω > 1$ .
M7 (beta): $ω$ follows a beta distribution on [0,1].
M8 (beta + selection): adds a class with $ω > 1$ .

The likelihood ratio test comparing M1a vs M2a or M7 vs M8 detects sites under positive selection. Branch-site models test whether $d_{N} / d_{S} > 1$ at specific sites on specific branches (e.g., along a lineage of interest), controlling for variation across both sites and lineages.

Implications for phylogenetics

Neutral markers provide the ideal data for phylogenetic tree building because their substitution rate is constant ( $k = μ$ ) and their genealogy reflects the species tree without distortion from selection. Synonymous sites, introns, and intergenic regions are standard neutral markers. The molecular clock provides the time calibration that converts branch lengths from substitutions to years, requiring fossil or geological calibration points to fix the clock rate.

Deep phylogenies require markers with appropriate substitution rates: rapidly evolving markers (e.g., mitochondrial DNA, synonymous sites) saturate over long time periods (multiple substitutions at the same site obscure the true count), while slowly evolving markers (e.g., ribosomal RNA, conserved protein domains) retain signal over hundreds of millions of years. The choice of marker trades off clock-like regularity against rate of informative change, and the neutral theory provides the framework for making this choice: use the most neutral marker whose rate produces a usable number of changes over the relevant time scale.

Full proof set Master

Proposition 1 (Neutral substitution rate equals mutation rate). In a diploid Wright-Fisher population of effective size $N_{e}$ with per-generation per-gene-copy neutral mutation rate $μ$ , the rate of neutral substitution is $k = μ$ .

Proof. Each generation, $2 N_{e} μ$ new neutral mutations arise (one per gene copy with probability $μ$ ). Each new mutation is present as a single copy at frequency $p_{0} = 1/ (2 N_{e})$ . By the martingale property of neutral drift, $E [p_{\infty}] = p_{0}$ . At absorption, $p_{\infty} \in {0, 1}$ , so $P_{fix} = p_{0} = 1/ (2 N_{e})$ . The expected number of substitutions per generation:

k = 2 N_{e} μ \times \frac{1}{2 N _{e}} = μ .

The population-size dependence cancels exactly. $□$

Proposition 2 (Expected heterozygosity at a neutral locus). At equilibrium between neutral mutation and drift in a diploid population of effective size $N_{e}$ with per-generation mutation rate $μ$ , the expected heterozygosity is $H = 4 N_{e} μ / (1 + 4 N_{e} μ) = θ / (1 + θ)$ where $θ = 4 N_{e} μ$ .

Proof. Let $H_{t}$ be the expected heterozygosity at time $t$ . Two processes change $H$ :

Drift reduces heterozygosity by a factor $(1 - 1/ (2 N_{e}))$ per generation 19.04.01.
Mutation introduces new alleles, increasing heterozygosity. The probability that two alleles are identical by descent and neither mutated is $(1 - μ)^{2} \approx 1 - 2 μ$ .

The recursion is:

H_{t + 1} = (1 - 2 μ) (1 - \frac{1}{2 N _{e}}) H_{t} + 2 μ (1 - H_{t}) \cdot 1.

Approximating $(1 - 2 μ) (1 - 1/ (2 N_{e})) \approx 1 - 2 μ - 1/ (2 N_{e})$ :

H_{t + 1} = (1 - 2 μ - \frac{1}{2 N _{e}}) H_{t} + 2 μ .

At equilibrium ( $H_{t + 1} = H_{t} = H$ ):

H = (1 - 2 μ - \frac{1}{2 N _{e}}) H + 2 μ,

H (2 μ + \frac{1}{2 N _{e}}) = 2 μ,

H = \frac{2 μ}{2 μ + 1/ ( 2 N _{e} )} = \frac{4 N _{e} μ}{1 + 4 N _{e} μ} = \frac{θ}{1 + θ} .

When $θ ≪ 1$ : $H \approx θ$ . When $θ ≫ 1$ : $H \approx 1$ . $□$

Proposition 3 (dN/dS under purifying selection). Consider a locus where a fraction $f_{d}$ of nonsynonymous mutations are deleterious with selection coefficient $- s$ ( $s > 0$ , $2 N_{e} s ≫ 1$ ) and the remaining fraction $1 - f_{d}$ are neutral. The ratio $d_{N} / d_{S}$ is:

\frac{d _{N}}{d _{S}} = 1 - f_{d} .

Proof. Synonymous mutations are neutral by assumption, so $d_{S} = μ_{S}$ (the synonymous mutation rate). Nonsynonymous mutations arise at rate $μ_{N}$ (the nonsynonymous mutation rate). Of these, fraction $1 - f_{d}$ are neutral and fix at rate $μ_{N} (1 - f_{d}) \times 1/ (2 N_{e}) \times 2 N_{e} = μ_{N} (1 - f_{d})$ . Fraction $f_{d}$ are deleterious with selection coefficient $- s$ ; their fixation probability is $\approx (1 - e^{4 N_{e} s p_{0}}) / (1 - e^{4 N_{e} s}) \approx 0$ when $2 N_{e} s ≫ 1$ and $p_{0} = 1/ (2 N_{e})$ . Therefore $d_{N} \approx μ_{N} (1 - f_{d})$ .

The $d_{N} / d_{S}$ ratio is:

\frac{d _{N}}{d _{S}} = \frac{μ _{N} ( 1 - f _{d} )}{μ _{S}} .

If the per-site mutation rate is the same for synonymous and nonsynonymous sites, $μ_{N} / μ_{S} = 1$ (after correcting for the different numbers of synonymous and nonsynonymous sites), giving $d_{N} / d_{S} = 1 - f_{d}$ . $□$

When $f_{d} = 0$ (no deleterious mutations): $d_{N} / d_{S} = 1$ (neutral). When $f_{d} = 0.9$ : $d_{N} / d_{S} = 0.1$ (strong purifying selection). This gives a direct mapping from $d_{N} / d_{S}$ to the fraction of deleterious nonsynonymous mutations.

Connections Master

Genetic drift 19.04.01. Neutral theory is the principal application of genetic drift to molecular evolution. The substitution rate $k = μ$ follows from the fixation probability $1/ (2 N_{e})$ derived in the drift unit. The drift-selection threshold $N_{e} s \approx 1$ determines whether a mutation is effectively neutral (drift dominates) or subject to selection, and this threshold is what separates strictly neutral from nearly neutral mutations.
The coalescent 19.04.03 pending. The coalescent is the natural framework for studying neutral variation within a population. Under neutrality, the genealogy of a sample is a random coalescent tree independent of the mutational process. This separation of genealogy from mutation (the "coalescent independence") is what makes coalescent-based inference possible: the genealogy provides the null distribution against which tests like Tajima's D and the MK test detect deviations from neutrality.
Natural selection 19.03.01. The $d_{N} / d_{S}$ ratio and MK test operationalise the distinction between drift and selection at the molecular level. Purifying selection ( $d_{N} / d_{S} < 1$ ) is the dominant selective force at the molecular level -- it removes deleterious mutations -- while positive selection ( $d_{N} / d_{S} > 1$ ) is rarer but detectable by comparing divergence rates at functional vs neutral sites.
Phylogenetics 19.07.01. The molecular clock ( $k = μ$ ) provides the time calibration for molecular phylogenies: branch lengths in substitutions per site convert to time once the clock rate is known. Neutral markers (synonymous sites, introns, pseudogenes) are preferred for tree building because their substitution process is clock-like and their genealogy tracks the species tree without distortion from selection.
Hardy-Weinberg extensions 19.02.02 pending. The neutral theory's prediction of equilibrium heterozygosity $H = θ / (1 + θ)$ is the infinite-alleles analogue of the Hardy-Weinberg equilibrium, extended to include mutation and finite population size. Where Hardy-Weinberg assumes infinite population size and no mutation, the neutral equilibrium incorporates both.

Historical & philosophical context Master

The neutral theory emerged from two independent observations in the late 1960s. Zuckerkandl and Pauling (1962) noted that amino-acid substitutions in haemoglobin accumulated at roughly constant rates across mammalian lineages -- the molecular clock. Kimura (1968) and King and Jukes (1969) independently argued that the observed rate of molecular evolution was too high to be explained by positive selection without excessive genetic load (Haldane's substitutional load argument). Kimura's calculation was direct: if $\sim 10$ amino-acid substitutions per protein per billion years across $\sim 1 0^{6}$ proteins in a mammal, the total selective deaths per generation would exceed the population size. The only way to reconcile the data was to posit that most substitutions are neutral, fixed by drift without selective cost.

The neutralist-selectionist debate dominated molecular evolution for two decades. Lewontin and Krakauer (1973) attempted to use the distribution of $F_{S T}$ across loci to test neutrality, but their method was criticised for assuming independence of loci under selection. Gillespie (1991) argued that the molecular clock's observed variance (overdispersion relative to Poisson expectation) was inconsistent with strict neutrality and suggested a role for selection. The debate was never fully resolved in its original terms; instead, the development of the MK test (1991) and codon-based likelihood methods (Goldman and Yang 1994, Nielsen and Yang 1998) shifted the focus from "is molecular evolution neutral?" to "what fraction of substitutions at each class of site are neutral, deleterious, or adaptive?"

Ohta's nearly neutral theory (1973) was the most important refinement. By recognising that mutations with $∣ s ∣ \sim 1/ (2 N_{e})$ are "nearly neutral" -- subject to selection in large populations but effectively neutral in small ones -- Ohta explained why the molecular clock is more regular per year than per generation for nonsynonymous sites, why $d_{N} / d_{S}$ varies across taxa in correlation with $N_{e}$ , and why the rate of protein evolution is higher in lineages with smaller effective population sizes.

The philosophical significance of neutral theory is that it decoupled molecular change from adaptive explanation. Before Kimura, the default assumption was that molecular differences between species were adaptive. After Kimura, the null hypothesis for any molecular substitution is neutrality, and the burden of proof is on demonstrating positive selection. This inversion of the null -- from adaptationism to neutralism -- is one of the most consequential shifts in the philosophy of evolutionary biology, comparable to the shift from typological to population thinking earlier in the century.

Bibliography Master

@article{Kimura1968,
  author = {Kimura, Motoo},
  title = {Evolutionary rate at the molecular level},
  journal = {Nature},
  volume = {217},
  year = {1968},
  pages = {624--626},
}

@article{KingJukes1969,
  author = {King, Jack Lester and Jukes, Thomas Hughes},
  title = {Non-{D}arwinian evolution},
  journal = {Science},
  volume = {164},
  year = {1969},
  pages = {788--798},
}

@book{Kimura1983,
  author = {Kimura, Motoo},
  title = {The Neutral Theory of Molecular Evolution},
  publisher = {Cambridge University Press},
  year = {1983},
}

@article{Ohta1973,
  author = {Ohta, Tomoko},
  title = {Slightly deleterious mutant substitutions in evolution},
  journal = {Nature},
  volume = {246},
  year = {1973},
  pages = {96--98},
}

@article{McDonaldKreitman1991,
  author = {McDonald, John H. and Kreitman, Martin},
  title = {Adaptive protein evolution at the {Adh} locus in {D}rosophila},
  journal = {Nature},
  volume = {351},
  year = {1991},
  pages = {652--654},
}

@article{Tajima1989,
  author = {Tajima, Fumio},
  title = {Statistical method for testing the neutral mutation hypothesis by {DNA} polymorphism},
  journal = {Genetics},
  volume = {123},
  year = {1989},
  pages = {437--464},
}

@article{HudsonKreitmanAguade1987,
  author = {Hudson, Richard R. and Kreitman, Martin and Aguade, Montserrat},
  title = {A test of neutral molecular evolution based on nucleotide data},
  journal = {Genetics},
  volume = {116},
  year = {1987},
  pages = {153--159},
}

@article{FayWu2000,
  author = {Fay, Justin C. and Wu, Chung-I},
  title = {Hitchhiking under positive {D}arwinian selection},
  journal = {Genetics},
  volume = {155},
  year = {2000},
  pages = {1405--1413},
}

@article{NeiGojobori1986,
  author = {Nei, Masatoshi and Gojobori, Takashi},
  title = {Simple methods for estimating the numbers of synonymous and nonsynonymous nucleotide substitutions},
  journal = {Molecular Biology and Evolution},
  volume = {3},
  year = {1986},
  pages = {418--426},
}

@article{YangNielsen2000,
  author = {Yang, Ziheng and Nielsen, Rasmus},
  title = {Estimating synonymous and nonsynonymous substitution rates under realistic evolutionary models},
  journal = {Molecular Biology and Evolution},
  volume = {17},
  year = {2000},
  pages = {32--43},
}

@book{Yang2007,
  author = {Yang, Ziheng},
  title = {Computational Molecular Evolution},
  publisher = {Oxford University Press},
  year = {2007},
}

@article{Charlesworth1993,
  author = {Charlesworth, Brian and Morgan, M. T. and Charlesworth, Deborah},
  title = {The effect of deleterious mutations on neutral molecular variation},
  journal = {Genetics},
  volume = {134},
  year = {1993},
  pages = {1289--1303},
}

@book{Gillespie1991,
  author = {Gillespie, John H.},
  title = {The Causes of Molecular Evolution},
  publisher = {Oxford University Press},
  year = {1991},
}

@incollection{ZuckerkandlPauling1962,
  author = {Zuckerkandl, Emile and Pauling, Linus},
  title = {Molecular disease, evolution, and genic heterogeneity},
  booktitle = {Horizons in Biochemistry},
  editor = {Kasha, Michael and Pullman, Bernard},
  publisher = {Academic Press},
  year = {1962},
  pages = {189--225},
}

@book{HartlClark2007,
  author = {Hartl, Daniel L. and Clark, Andrew G.},
  title = {Principles of Population Genetics},
  edition = {4th},
  publisher = {Sinauer Associates},
  year = {2007},
}

@book{FutuymaKirkpatrick2017,
  author = {Futuyma, Douglas J. and Kirkpatrick, Mark},
  title = {Evolution},
  edition = {4th},
  publisher = {Sinauer Associates},
  year = {2017},
}