Genetic Drift and Random Change

Genetic drift is the change in allele frequency across generations that results purely from chance sampling in a finite population, rather than from differences in fitness, mutation, or migration. Every generation, only a subset of possible gametes actually unites to form the next generation's zygotes. Even if every genotype has identical survival and reproductive success, the specific alleles that get sampled into the next generation's gene pool will deviate from the parental frequencies by chance, exactly as a small sample of coin flips deviates from a 50:50 expectation more than a large sample does. Because this deviation is random with respect to allele identity, drift is classified as a nonadaptive evolutionary force: it can increase or decrease the frequency of a beneficial, neutral, or harmful allele with equal disregard for its phenotypic consequences. Formally, drift is the variance in allele frequency generated by binomial (or multinomial, for multiple alleles) sampling of `2N` gametes from a population of `N` diploid individuals to form the next generation. This is the essential content of the Wright-Fisher model, developed independently by Sewall Wright and Ronald Fisher in the 1930s. If an allele has frequency `p` in generation `t`, the expected frequency in generation `t+1` is still `p` — drift has no directional bias — but the variance around that expectation is: ``` Var(p') = p(1 − p) / (2N) ``` This single equation is the mathematical heart of the topic. It shows that the expected magnitude of random frequency change scales as `1/N`: doubling the population halves the variance, and a population of 10 individuals will show roughly 100 times the sampling variance of a population of 1,000. Drift is not "a small effect that happens sometimes" — it is a universal, ever-present statistical consequence of finite reproduction, whose magnitude depends entirely on `N`. Two consequences follow inescapably from repeated sampling. First, allele frequencies perform a random walk bounded at 0 and 1, and because 0 and 1 are absorbing boundaries (once an allele is lost or fixed, sampling cannot reintroduce it without new mutation or migration), every allele is eventually driven to fixation or loss if drift acts alone for long enough. Second, because heterozygosity depends on having intermediate allele frequencies, drift systematically erodes genetic variation within a population over time, even as it can, when many small populations are considered together, increase differentiation between them. This tension — drift removes diversity within a population while creating divergence among populations — is central to understanding phenomena from island endemism to the genetic structure of endangered species.

Step 1 — Finite gamete sampling. In any real population, only `2N` gametes (for a diploid population of `N` individuals) unite to found the next generation, out of a vastly larger pool of gametes actually produced. Which specific gametes succeed is influenced by chance encounters, chance fertilization, chance survival to reproductive age — factors uncorrelated with the genotype at a neutral locus. The next generation's allele frequency is therefore a binomial sample from the current frequency, not an exact copy of it.

Step 2 — The Wright-Fisher idealization. To make this tractable, population geneticists model reproduction as if all `N` individuals contribute equally and independently to an infinite gamete pool with allele frequency `p`, from which `2N` gametes are drawn at random (with replacement) to build the next generation. Under this idealization, the number of copies of an allele in the next generation follows a binomial distribution with parameters `2N` and `p`, giving mean `2Np` (frequency unchanged in expectation) and variance `2Np(1−p)`, which for allele frequency translates to the `p(1−p)/(2N)` formula above.

Step 3 — Iteration and the random walk. Because each generation's frequency is a fresh binomial draw centered on the previous generation's realized frequency (not the original starting frequency), the frequency wanders unpredictably up or down across generations — a random walk with no restoring force pulling it back to its starting value. Some replicate populations drift up, some down, purely by chance, even though they all started identical.

Step 4 — Absorption at the boundaries. Once an allele frequency hits `p = 0`, it is lost and cannot return (barring new mutation or migration). Once it hits `p = 1`, it is fixed, and the alternative allele is gone. These are "absorbing states" of the random walk. Critically, the probability that a given neutral allele eventually becomes fixed equals its current frequency — this is one of the most elegant results in population genetics. A new mutation, present as a single copy in a diploid population of `N` individuals, has initial frequency `1/(2N)`, so its probability of ultimate fixation by drift alone is: ``` P(fixation) = 1 / (2N) ``` This means most new neutral mutations are lost, but a tiny fraction — inversely proportional to population size — eventually sweep to fixation by chance alone, with no selective advantage whatsoever.

Step 5 — Timescale. The expected time to fixation or loss also depends on `N`. For a neutral allele starting at frequency `p`, the expected number of generations until fixation (conditional on eventual fixation) is approximately `4N` generations (in units scaled by effective population size), meaning drift in large populations is not just weaker in magnitude per generation — it also takes far longer to run to completion, so standing genetic variation persists longer in big populations.

Step 6 — Decline of heterozygosity. Averaged across many independent replicate populations (or many independent loci), expected heterozygosity declines geometrically each generation: ``` Ht = H0 × (1 − 1/(2N))^t ``` For large `N`, `(1 − 1/(2N))` is close to 1 and heterozygosity erodes slowly; for small `N`, it erodes quickly. This equation is the quantitative backbone of why small, isolated, or bottlenecked populations lose genetic diversity so much faster than large ones — a principle central to The Founder Effect and Bottlenecks and to Effective Population Size, which corrects the idealized `N` for real-world complications like unequal sex ratios and variance in reproductive success.

Peter Buri's Drosophila bottle experiment (1956) is the textbook demonstration that drift behaves exactly as the Wright-Fisher model predicts. Buri set up 107 replicate laboratory populations of Drosophila melanogaster, each maintained at a constant N = 16 breeding adults per generation, all starting from an identical heterozygous frequency of `p = 0.5` at the bw (brown eye color) locus, using the neutral (or nearly neutral) alleles `bw` and `bw⁷⁵`, which are distinguishable by eye color and do not detectably affect fitness under lab conditions. Every generation, Buri randomly picked 8 males and 8 females from each bottle's offspring to found the next generation — a direct laboratory realization of finite binomial sampling. He tracked all 107 populations for 19 generations and recorded the allele frequency distribution across populations at each generation. The results matched drift theory strikingly well:

• The variance in allele frequency among the 107 populations increased generation by generation, exactly as predicted by `Var(p) = p(1-p)/(2N)` accumulated over time.
• The distribution of frequencies across populations spread out and became increasingly bimodal, piling up near `p = 0` and `p = 1` — the signature of a random walk approaching its absorbing boundaries.
• By generation 19, a substantial fraction of populations had already reached fixation for one allele or the other — some 30 populations fixed for `bw⁷⁵` and 28 for `bw`, with the alleles fixing in each direction in roughly equal numbers, exactly as expected since both started at `p = 0.5` and neither was selectively favored.
• The mean allele frequency across all populations stayed close to 0.5 throughout, confirming that drift has no directional bias — it changes frequencies randomly, not toward any particular value.

Buri's experiment was powerful because it isolated drift from selection and migration by design: identical starting genotype frequencies, controlled small population size, no migration between bottles, and a locus believed close to selectively neutral. The only variable differing among replicate populations was the outcome of finite random sampling each generation — and that alone produced divergence, variance growth, and eventual fixation, exactly matching Wright's mathematics derived decades earlier from theory alone. A complementary line of evidence comes from the study of conserved molecular markers and the neutral theory of molecular evolution, developed by Motoo Kimura in the 1960s, who used the mathematics of drift (extending Wright's framework) to argue that the vast majority of molecular substitutions observed between species are neutral or nearly neutral, fixed by drift rather than positive selection — a claim supported by the observation that synonymous (silent) DNA substitutions, which do not change protein sequence and are largely invisible to selection, accumulate at remarkably steady, clock-like rates consistent with drift-driven fixation governed by mutation rate and population size rather than by adaptive pressure. This is direct evidence that drift is not confined to the laboratory bottle but is a major driver of real genomic change over evolutionary time.

Example 1 — Fixation probability and expected time in a bottlenecked bird population. A population of an endangered bird crashes to N = 25 breeding pairs (so 50 diploid individuals) after habitat loss. A previously rare but selectively neutral allele `A` exists at frequency `p = 0.04` in the surviving population (roughly 4 copies among 100 gene copies, since `2N = 100`). Because the allele is neutral, drift theory gives: ``` P(eventual fixation of A) = p = 0.04 P(eventual loss of A) = 1 − p = 0.96 ``` So there is only a 4% chance this allele ever becomes universal in the population, and a 96% chance it disappears entirely — purely from sampling, with no assumption that `A` is harmful. Compare this to a healthy population of `N = 2,500` pairs (2× larger by two orders of magnitude) with the same allele frequency: the fixation probability is identical (`P = p = 0.04`, since fixation probability of a neutral allele depends only on its frequency, not directly on `N`), but the variance per generation is 50-fold smaller (`1/(2×2500)` vs `1/(2×25)`), and the expected time to reach fixation or loss scales with `N`, roughly 4N generations — about 100 generations in the bottlenecked population versus 10,000 generations in the large one. The bottlenecked population's fate is sealed almost immediately in evolutionary time; the large population's rare allele persists as standing variation for a much longer window, giving natural selection or environmental change far more opportunity to act on it before drift resolves the outcome. This is precisely the logic behind conservation concern for bottlenecked populations: not that alleles are "doomed" in a different statistical sense, but that whatever happens, happens fast, and diversity is lost quickly.

Example 2 — When does drift overpower selection? The Nₑs threshold. A mildly deleterious mutation reduces relative fitness by `s = 0.001` (a 0.1% selective disadvantage) — the kind of small-effect mutation ubiquitous in real genomes. Population genetic theory (Wright, later formalized by Kimura) shows that the fate of such an allele depends on the compound parameter `Nes` (effective population size times selection coefficient):

• If `|Nes| ≫ 1`, selection dominates and reliably removes the deleterious allele.
• If `|Nes| ≪ 1`, drift dominates and the allele behaves as if effectively neutral, regardless of its true fitness cost — it can drift to fixation despite being harmful.

Consider two species with effective population sizes `Ne = 10,000` and `Ne = 100`: ``` Species 1: Ne·s = 10,000 × 0.001 = 10 → |Nes| ≫ 1 → selection dominates, allele purged Species 2: Ne·s = 100 × 0.001 = 0.1 → |Nes| ≪ 1 → drift dominates, allele behaves as neutral ``` The identical mutation, with the identical fitness cost, is efficiently eliminated in the large population but can drift to fixation in the small one. This single calculation explains a major pattern in comparative genomics: species with historically small effective population sizes (such as many endangered mammals, or organisms with strong population bottlenecks) tend to accumulate more slightly deleterious substitutions in their genomes than species with huge effective population sizes (many insects, marine invertebrates, and bacteria), because in small populations selection's grip weakens relative to the randomizing pull of drift. This quantitative threshold is the essence of what will be developed further as Drift and Selection Interplay.

Sewall Wright (1889–1988) was an American geneticist trained originally in animal breeding and physiological genetics at Harvard under William Castle, where he worked on coat color inheritance in guinea pigs — hands-on breeding work that gave him an intuitive feel for how small, isolated populations behave differently from large panmictic ones. Wright's genius was mathematical: he built the theoretical machinery — path analysis, the F-statistics still used today to measure population structure, and the diffusion equations describing allele frequency change under drift, mutation, migration, and selection simultaneously — largely from first principles, working somewhat apart from the mainstream biometrical tradition dominated by Fisher in Britain. Wright's "shifting balance theory" (now largely superseded in its strongest form but historically influential) proposed that small, semi-isolated subpopulations experiencing strong drift could explore fitness landscapes in ways large panmictic populations could not, occasionally stumbling onto new adaptive peaks that selection alone could not reach — placing drift not merely as noise but as a potential creative partner to selection, a controversial claim still debated in modern evolutionary theory.

Genetic drift is not a historical curiosity confined to fruit-fly bottles; it is a working, quantitative tool across modern biology. In conservation genetics, drift calculations directly inform species recovery plans: population geneticists estimate the effective population size of endangered species (condors, cheetahs, black-footed ferrets) precisely because small Ne predicts rapid loss of heterozygosity, inbreeding depression, and reduced capacity for future adaptation — the same mathematics from Buri's bottles scaled up to real, irreplaceable populations. In human medical genetics, drift explains why certain recessive disease alleles (such as those causing Tay-Sachs disease or various metabolic disorders) reach unusually high frequencies in populations that passed through historical founder events or bottlenecks, even though the alleles are harmful in homozygous form — a direct application of the Nes threshold showing that a population history dominated by drift can override selection against deleterious variants. In molecular evolution and genomics, the neutral theory built on drift underlies the "molecular clock" used to date evolutionary divergences and to calibrate phylogenies, since silent or nearly neutral substitutions accumulate at rates set largely by mutation and drift rather than by adaptive pressure. In agriculture and captive breeding programs, breeders explicitly manage effective population size to slow the drift-driven erosion of genetic diversity in livestock breeds and seed banks, since diversity lost to drift cannot be recovered without reintroducing variation from outside sources. Across all these applications, the same core insight from Wright's equations and Buri's bottles recurs: chance sampling in finite populations is not noise to be dismissed but a measurable, predictable, and often consequential force shaping the genetic future of every real population on Earth.

Ask the tutor: How does effective population size get calculated when a real population has unequal sex ratios or fluctuating size across generations?