Allele Frequency Dynamics
Why do brown eyes stay common while a new mutation for lactase persistence spreads across whole continents? This happens in just a few thousand years. Both are the same underlying quantity in motion: an allele frequency, changing generation after generation. Hardy-Weinberg Equilibrium tells you what happens when nothing pushes on a population's gene pool. Frequencies stay put. Real populations are never left alone. Mutation introduces new variants, selection favors some over others, drift shuffles frequencies by chance, and migration mixes gene pools together. Allele frequency dynamics is the study of how these four forces move populations away from equilibrium. They can act alone or in combination. Together they generate the genetic changes we call evolution. This is arguably the most important quantitative idea in evolutionary biology. It converts "evolution" from a vague historical narrative into something you can compute. Given a starting frequency, a selection coefficient, a mutation rate, or a population size, you can predict what happens next. An actual formula tells you what the population will look like in ten, a hundred, or a million generations. That predictive power is what lets biologists explain the spread of insecticide resistance in mosquitoes. It also explains the fixation of industrial-melanism alleles in peppered moths. It explains the loss of genetic diversity in a captive cheetah population. And it explains the sweep of a beneficial mutation through a bacterial culture. The big picture is deceptively simple: evolution, at the population-genetic level, is change in allele frequency across generations. Everything else — speciation, adaptation, extinction — is this basic process compounded over time and across loci. The four forces don't act in isolation in nature. They interact, sometimes reinforcing each other and sometimes canceling out. The outcome depends on their relative strengths and on population size. This article builds the mathematical and mechanistic scaffolding for each force. It walks through the classic experiments that demonstrated each one operating in real populations. It also works two quantitative examples so you can see the numbers move. It sets up three ideas explored more deeply elsewhere: The Selection Coefficient, Effective Population Size, and Mutation-Selection Balance.
- Allele frequency change across generations is the operational definition of evolution at the population level.
- Four forces drive change: mutation (creates variation, very slow alone), selection (deterministic, favors fitter alleles), drift (random, stronger in small populations), migration (homogenizes gene pools).
- Selection changes allele frequency predictably via Δp = p·q·s / (1 + ...) type equations; drift changes it unpredictably, with variance ∝ 1/(2N).
- The Illinois long-term corn selection experiment and Endler's guppy predation experiments are direct experimental proofs of selection changing frequencies.
- Buri's Drosophila bottle experiment is the classic direct demonstration of genetic drift.
- Small effective population size amplifies drift and can override weak selection — a key theme carried into Effective Population Size.
Evolution at the population level is simply the change in allele frequency from one generation to the next, driven by mutation, selection, drift, and migration acting on the variation Mendelian inheritance provides.
The core idea
An allele frequency is the proportion of one allele among all alleles at a locus in a population's gene pool — a number between 0 and 1 (or 0% and 100%). If a population of 500 diploid individuals has 1,000 alleles at a locus, and 300 of them are the A variant, then p = f(A) = 0.30 and q = f(a) = 0.70. Hardy-Weinberg Equilibrium describes the special, idealized case where p and q do not change: no mutation, no selection, infinite population size (no drift), no migration, and random mating. That equilibrium is a null hypothesis, not a description of nature — its real value is that any deviation from Hardy-Weinberg proportions or any change in p across generations tells you that one of its assumptions is being violated, and the four forces are exactly the four biological ways that violation happens.
Mutation is the ultimate source of all new alleles; without it, evolution would eventually run out of raw material. But mutation rates are low — typically 10⁻⁸ to 10⁻⁵ per locus per generation in eukaryotes — so mutation alone changes allele frequency extremely slowly. Selection is differential survival and reproduction caused by heritable differences in fitness; it is the only one of the four forces that predictably increases the frequency of alleles that improve reproductive success, generation after generation, in a consistent direction (adaptation, not "goal-directed" — selection has no foresight, it simply amplifies whatever currently reproduces better). Genetic drift is random sampling error in which alleles happen to be passed to the next generation, unrelated to fitness; it is powerful in small populations and can fix or eliminate alleles regardless of their effect on survival. Migration (gene flow) moves alleles physically between populations, tending to homogenize allele frequencies across subpopulations and counteracting local differentiation from selection or drift. Crucially, these forces are not mutually exclusive; a real population usually experiences all four simultaneously, and the observed change in allele frequency is the net result of their interaction — sometimes reinforcing, sometimes canceling.
Mechanism
Selection, mechanistically. Selection changes allele frequency by making genotypes contribute unequal numbers of offspring to the next generation. Define relative fitness w for each genotype (the fittest genotype scaled to w = 1), and let the fitness of the disfavored homozygote be reduced by a selection coefficient s (so its fitness is 1 − s). For a recessive deleterious allele a with genotype fitnesses w(AA) = 1, w(Aa) = 1, w(aa) = 1 − s, starting at frequency q, the change per generation is: Δq = −s·q²(1 − q) / (1 − s·q²) This single equation encodes an important, often-missed fact: selection against a recessive allele slows down dramatically as q gets small, because rare recessive alleles are mostly hidden in heterozygotes where selection cannot "see" them. Selection against a dominant allele, by contrast, removes it quickly because every copy is expressed. This asymmetry — introduced lightly here — is the seed of The Selection Coefficient and of Mutation-Selection Balance, where new deleterious mutations reach a stable equilibrium frequency because mutation keeps introducing what selection keeps removing.
Drift, mechanistically. Each generation, only a finite sample of parental gametes actually becomes the next generation's zygotes — a random sampling process directly analogous to drawing marbles from an urn. If a population has effective size N and allele frequency p, the sampling variance of p in the next generation is: Var(p′) = p(1 − p) / (2N) Smaller N means larger variance — bigger, more erratic jumps in frequency each generation, purely by chance. Over many generations, drift is expected eventually to fix (p = 1) or lose (p = 0) every allele in a finite population, with the probability of eventual fixation of a neutral allele equal to its current frequency (a result derived by Sewall Wright and central to the neutral theory of molecular evolution). This population-size dependence is the direct link forward to Effective Population Size.
Mutation, mechanistically. Mutation converts allele A to a at rate μ per generation (and possibly back at rate ν). Its per-generation effect on frequency is tiny: Δp = −μp + νq With μ ≈ 10⁻⁶, a single generation changes p by about one part in a million — mutation alone is a weak force, but over evolutionary timescales (and combined with selection, which can amplify a mutation once it exists) it is the ultimate wellspring of all genetic variation.
Migration, mechanistically. If a mainland population with allele frequency p_m sends migrants making up a fraction m of a recipient population's next generation (whose frequency is p), the recipient's new frequency shifts toward the source: Δp = m(p_m − p) Migration is powerful precisely because it can be a large per-generation force (m can be 0.01–0.5 in real metapopulations) — much stronger than mutation, and often on par with selection.
Interaction of forces. In real populations these curves overlay. A classic outcome is mutation–selection balance: recurrent deleterious mutation (arriving at rate μ) is exactly balanced by selection removing copies (at rate roughly s·q for a recessive allele under Hardy–Weinberg proportions), giving an equilibrium frequency q̂ ≈ √(μ/s) for a fully recessive lethal-ish allele — the entry point to Mutation-Selection Balance. Another outcome is that drift can overwhelm weak selection: a mutation with |s| < 1/(2N) behaves as if effectively neutral, because random sampling noise dominates the tiny deterministic push of selection — a threshold central to Effective Population Size and to Motoo Kimura's neutral theory.
The evidence
Buri (1956) — genetic drift caught in the act. Peter Buri set up 107 independent laboratory populations of Drosophila melanogaster, each started with exactly 16 flies heterozygous at the bw⁷⁵ eye-color locus (initial allele frequency p = 0.5 in every single population, with no selection acting on this nearly neutral marker). Each generation, he randomly picked 8 males and 8 females as parents for the next generation — deliberately keeping the effective population size tiny (N ≈ 16) — and tracked allele frequency for 19 generations. The result: the 107 populations, which all started identically, fanned out into wildly different frequencies purely by chance; the variance in p across populations increased almost exactly as predicted by the drift formula Var(p) = p(1−p)/(2N) per generation, and by generation 19, 30 populations had fixed the allele (p = 1) and 28 had lost it (p = 0) — with intermediate frequencies becoming progressively rarer, exactly the "emptying middle, filling edges" signature drift theory predicts. This is the textbook direct proof that random sampling alone, with no selective difference between alleles, can drive allele frequencies to fixation or loss.
Kettlewell (1953–1956) and later reanalysis — natural selection changing a frequency in real time. The peppered moth (Biston betularia) exists in a pale speckled typica form and a dark carbonaria form controlled largely by a single dominant allele. Bernard Kettlewell released marked moths of both forms into polluted, soot-blackened woodland near industrial Birmingham and into unpolluted woodland in Dorset, then recaptured survivors and also filmed birds actively picking moths off tree trunks. In polluted woodland, dark moths survived and were recaptured at higher rates than pale moths; in unpolluted woodland, the reverse held. Predation by birds was directly observed selecting against whichever morph was more conspicuous against the local bark. Independently, museum and trap records document carbonaria rising from under 2% of northern English populations around 1848 to over 90% by the 1900s, then declining again after the Clean Air Acts reduced soot pollution in the mid-20th century — a real allele frequency trajectory rising and falling in step with an independently measured environmental variable (pollution), with the underlying gene (a transposable-element insertion in the cortex gene) identified by genome sequencing in 2016. While some of Kettlewell's specific methods have been criticized and refined by later workers (van't Hof et al. and Cook & Saccheri reanalyzed and largely vindicated the bird-predation mechanism with better-controlled experiments), the core conclusion — visual predation drives frequency change in this gene — has held up.
The Illinois Long-Term Selection Experiment (begun 1896, still running). Researchers at the University of Illinois selected corn plants for high or low oil content and high or low protein content every generation from an open-pollinated starting population, with no molecular knowledge of the underlying alleles. After more than 100 generations, oil content diverged from an initial ~4.7% to over 20% in the high-oil line and below 1% in the low-oil line — direct, cumulative, multi-locus evidence that consistent directional selection systematically shifts allele frequencies (here, across many quantitative-trait loci) generation after generation, with no sign of the response plateauing even after a century, showing how much standing genetic variation a population can carry.
Endler's guppies (1980) — selection demonstrated by translocation, not just observation. John Endler moved guppies (Poecilia reticulata) from high-predation pools (where dull, cryptic males survive better) into predator-free upstream pools in Trinidad. Within roughly 15 generations (about two years), male coloration in the transplanted populations evolved to be significantly more colorful and conspicuous, matching populations naturally found without predators — a controlled, replicated field experiment showing selection pressure (removed predation) changing an allele-frequency-based phenotype in a predictable direction, with matched control populations left under predation showing no such shift.
Worked examples
Example 1 — a genetics cross meets selection over generations. Suppose a diploid population has a fully recessive allele a causing an enzyme deficiency, currently at frequency q = 0.20 (so p = 0.80). Genotype fitnesses are w(AA) = 1, w(Aa) = 1, w(aa) = 0.60 (a selection coefficient s = 0.40 against homozygous recessives, with no dominance effect on heterozygotes). Starting genotype frequencies under Hardy-Weinberg are:
| Genotype | Frequency before selection | Fitness w | Contribution (freq × w) |
|---|---|---|---|
| AA | p² = 0.64 | 1.00 | 0.640 |
| Aa | 2pq = 0.32 | 1.00 | 0.320 |
| aa | q² = 0.04 | 0.60 | 0.024 |
Mean population fitness w̄ = 0.640 + 0.320 + 0.024 = 0.984. New q after selection: q′ = [q² · w(aa) + p·q·w(Aa)] / w̄ = [0.024 + 0.5(0.32)]/0.984 = (0.024+0.160)/0.984 = 0.1870 So q drops from 0.200 to 0.187 in a single generation — a change of only Δq ≈ −0.013, even though the homozygote suffers a serious 40% fitness disadvantage. Iterating this recursion (or using Δq = −sq²(1−q)/(1−sq²) ≈ −0.40(0.04)(0.80)/(1−0.40·0.04) = −0.0128/0.984 ≈ −0.013, matching) shows why purging a recessive deleterious allele is slow once it becomes rare: at q = 0.20 most copies (2pq = 0.32 vs. q² = 0.04 — eight times as many) are safely hidden in heterozygotes, invisible to selection. Getting q from 0.20 down to 0.01 would take roughly 100+ generations under this s, not the handful you might naively expect from a 40% fitness cost — a direct, quantitative illustration of dominance masking selection, foreshadowing The Selection Coefficient and Mutation-Selection Balance.
Example 2 — drift vs. selection: does population size decide the outcome? A conservation-genetics scenario: two isolated populations of an endangered bird each carry a new mutant allele at frequency p = 0.05, conferring a mild fitness benefit of s = +0.01 (heterozygote advantage assumed additive, so Δp per generation from selection is approximately s·p·q = 0.01 × 0.05 × 0.95 ≈ 0.00048 — small but positive and consistent in direction). Population X is large and well-connected: effective size N_e = 5,000. Population Y is a small, isolated remnant: effective size N_e = 25. Compare the deterministic selective push to the standard deviation of drift each generation, using Var(p) = p(1−p)/(2N_e):
| Population | N_e | SD from drift per generation √[p(1−p)/2N_e] | Deterministic Δp from selection |
|---|---|---|---|
| X (large) | 5,000 | √(0.0475/10,000) ≈ 0.00218 | 0.00048 |
| Y (small) | 25 | √(0.0475/50) ≈ 0.0308 | 0.00048 |
In population X, the selective signal (0.00048) is smaller than one generation's drift noise (0.00218) but not wildly so — over many generations the consistent directional push of selection will accumulate and reliably win out, since drift has no consistent direction to add up in. In population Y, drift noise (0.0308) swamps the selective signal by more than 60-fold in a single generation — this mutation is, for all practical purposes, behaving as a neutral allele: whether it rises to fixation or vanishes is dominated by chance, not by its (real, positive) fitness effect. This is the concrete numerical content of the "nearly neutral" threshold |s| ≲ 1/(2N_e) (here 1/50 = 0.02, comfortably above s = 0.01) — precisely the calculation that motivates Effective Population Size as a quantity conservation biologists must estimate and manage, since a shrinking N_e doesn't just lose diversity randomly — it can neutralize the effectiveness of natural selection itself.
Common misconceptions & traps
| Misconception | Correction | ||
|---|---|---|---|
| "Dominant alleles increase in frequency because they're dominant." | Dominance affects phenotype expression, not allele frequency dynamics directly; a dominant allele's frequency changes only through selection, drift, mutation, or migration, exactly like a recessive one. | ||
| "A rare recessive disease allele will eventually disappear if it's harmful." | Selection against rare recessives is very weak because most copies are hidden in unaffected heterozygotes (Example 1); mutation-selection balance keeps such alleles at low but stable, nonzero frequency indefinitely. | ||
| "Genetic drift only matters in tiny populations, so it's usually negligible." | Drift's relative importance (compared to selection) scales with 1/N_e, so it matters enormously in small or bottlenecked populations, and even in large ones, drift dominates for alleles with | s | < 1/(2N_e). |
| "Selection acts on genotype frequencies to produce a 'better' species over time." | Selection has no foresight or goal; it merely differentially propagates whatever variants currently reproduce more, and can be reversed instantly if the environment changes (as in the peppered moth's post-pollution decline). | ||
| "Evolution requires selection; drift and mutation are minor side effects." | All four forces independently satisfy the definition of evolution (a change in allele frequency); molecular evolution at synonymous/neutral sites is dominated by drift and mutation, not selection (Kimura's neutral theory). | ||
| "Migration always increases diversity within every population." | Migration increases diversity within a recipient population if the migrant frequency differs from the resident one, but it always decreases differentiation between populations, potentially erasing local adaptation. |
The scientist behind it
The mathematical machinery of allele frequency dynamics was built in the 1920s–30s largely by three figures collectively called the architects of the "Modern Synthesis": Ronald Fisher, who derived the fundamental theorem of natural selection and formalized selection's deterministic force in infinite populations; J.B.S. Haldane, who calculated selection coefficients from real data (notably estimating s for the peppered moth's carbonaria allele decades before Kettlewell's fieldwork, from historical frequency records alone); and Sewall Wright, who developed the mathematics of genetic drift and effective population size, introducing the very concept that "the fate of an allele can depend on chance alone, not just fitness" — a claim controversial at the time because it seemed to challenge selection's primacy, but which became essential once Kimura extended it into the neutral theory of molecular evolution in the 1960s.
Origins · 1 minSewall Wright and how this was discovered
Where this matters today
Tracking and predicting allele frequency change is now central, applied science. In medicine, the rise of antibiotic-resistance alleles in bacterial populations under drug pressure, and antiviral-resistance mutations in HIV or influenza under treatment, are monitored using exactly these selection equations to predict how fast a resistant strain will dominate and to design multi-drug regimens that slow it. In agriculture, the same logic underlies breeding programs (directly descended from the Illinois corn experiment) and the surveillance of pesticide-resistance alleles in insect pests, where knowing the selection coefficient and population size lets growers time insecticide rotations to keep resistance alleles rare. In conservation biology, estimating effective population size in endangered species (California condors, cheetahs, Isle Royale wolves) tells managers whether drift is likely to fix harmful alleles or erode adaptive variation faster than selection can act — directly motivating the concept of Effective Population Size as a conservation-management target, not just a theoretical number. In public health genomics, genome-wide association studies track how disease-allele frequencies differ across human populations partly due to differing histories of selection, drift, and migration (e.g., the classic case of the CCR5-Δ32 HIV-resistance allele, whose frequency and geographic distribution reflect historical selection and drift acting on a single well-characterized locus). And in synthetic biology and gene-drive engineering, designers must explicitly model selection, drift, and migration together to predict — and try to control — whether an engineered allele (say, one that suppresses malaria-transmitting mosquitoes) will actually spread through a wild population or be lost to chance in the very first few generations.
Ask the tutor: How do biologists actually measure a real selection coefficient or effective population size from field or genomic data?
Is allele frequency change the same thing as evolution?
Yes, at the population-genetic level — this is essentially the operational definition of evolution used in population genetics: any change in allele (or genotype) frequency within a population across generations counts as evolution, whether caused by selection, drift, mutation, or migration. This definition deliberately doesn't require adaptation or "improvement"; a population that loses genetic diversity purely by chance through drift has evolved just as surely as one that's been reshaped by strong natural selection. Larger-scale patterns — speciation, long-term phenotypic trends, adaptive radiations — are built from this basic process repeated across many loci, many populations, and long timescales, but the fundamental unit of change is still a shifting frequency at a locus.
Why does selection act so slowly on rare recessive alleles?
Because natural selection can only "see" a genotype's phenotype, not the allele itself, and rare recessive alleles are overwhelmingly carried by heterozygotes who don't express the recessive phenotype. At q = 0.01, roughly 98% of all copies of the allele sit safely inside Aa heterozygotes (2pq ≈ 0.0198) while only about 0.01% of the population is aa (q² = 0.0001) and actually exposed to selection. This is why fully recessive genetic diseases (like many inborn errors of metabolism) persist at low but stable frequencies for very long periods even under strong selection against affected homozygotes — the mechanism formalized in Mutation-Selection Balance.
How is genetic drift different from natural selection, mechanistically?
Selection changes allele frequency because of a real, causal, heritable difference in survival or reproduction tied to the allele itself — it has a consistent direction each generation, even if the magnitude is small. Drift changes allele frequency purely through the random luck of which gametes happen to be sampled to form the next generation, with no consistent direction — it's exactly analogous to sampling error when you flip a coin only a few times. Mathematically, selection produces a deterministic Δp with a definite sign each generation, while drift produces a Δp with expected value zero but nonzero variance that scales as 1/(2N_e), meaning it matters far more in small populations.
Can genetic drift and natural selection push allele frequency in opposite directions at the same time?
Yes, and this conflict is central to real evolution in finite populations. A mildly deleterious allele can still drift to fixation in a small population even though selection is constantly pushing against it, if the population size is small enough that |s| < 1/(2N_e) — the selective push is simply too weak to overcome the random noise. This is thought to explain the accumulation of mildly deleterious mutations in small, isolated populations (a phenomenon sometimes called "drift load"), and it is a major reason conservation biologists worry about effective population size, not just census size, when managing endangered species.
What's the difference between allele frequency and genotype frequency, and why does it matter here?
Allele frequency (p, q) counts individual gene copies in the gene pool; genotype frequency (like p², 2pq, q² under Hardy-Weinberg) counts diploid combinations of those alleles in individuals. Selection acts directly on genotype fitness (because phenotypes, and thus survival/reproduction, are properties of whole organisms, not naked alleles), but the quantity we track generation-to-generation as "evolution" is allele frequency, since that's what actually gets passed on. The mathematical trick in all the selection equations is converting genotype-level fitness effects back into an allele-frequency-level Δp, which is exactly what the Δq formula in the Mechanism section does.
Why did the peppered moth's dark allele decline again after initially spreading?
Because the selective advantage of dark coloration was entirely dependent on a specific environment — soot-blackened tree bark — and that environment changed. After the UK's Clean Air Acts (starting 1956) reduced industrial soot pollution, lichens recolonized tree trunks and bark lightened again, reversing which morph was camouflaged and which was conspicuous to bird predators. The carbonaria allele's frequency fell back down over subsequent decades, tracking measured reductions in atmospheric pollution — direct evidence that selection coefficients are not fixed properties of an allele but depend on the current environment, and that selection can reverse direction as fast as the environment does.
How does migration interact with selection and drift in real populations?
Migration (gene flow) tends to work against both local selection and local drift by importing alleles at the frequency found elsewhere, pulling a recipient population's frequency toward the source population's frequency each generation (Δp = m(p_m − p)). This can prevent local adaptation if migration is strong relative to selection (swamping locally favorable alleles with maladapted ones from elsewhere), and it can also replenish genetic diversity lost to drift in a small, isolated population — which is exactly why conservation programs sometimes deliberately introduce migrants ("genetic rescue") into small, inbred populations to counteract drift-driven loss of variation.