Mutation as an Evolutionary Force
A single human gene, the one encoding the blood-clotting factor VIII, mutates often enough. Roughly 1 in 25,000 male births carries a brand-new case of hemophilia A. This arises not from inherited genes but from a fresh copy error in that generation's germline. Multiply this by the roughly 20,000 genes in the human genome. Every one of us turns out to carry on the order of 70 new mutations relative to our parents. This raises an old and deceptively simple question. If mutation is the ultimate source of all genetic variation, why is it usually described as a weak evolutionary force? The answer lies in a distinction that this article exists to sharpen. Mutation creates variation, but by itself it changes allele frequencies extraordinarily slowly. A mutation rate of one in a million per generation can only shift an allele frequency by about one-in-a-million per generation acting alone. Yet mutation is indispensable. Every advantageous allele that selection ever fixed first appeared as a mutation. Every neutral allele that drift ever carried to fixation also first appeared as a mutation. Evolutionary biology therefore treats mutation as a force with a peculiar dual character. It is essential as the fountainhead of variation. But it is usually negligible as a direct sculptor of allele frequencies within any single generation. This article works out that duality quantitatively. It builds directly on the idea of mutation as a source of variation. It also builds on the Hardy–Weinberg equilibrium as a baseline of "no change." Mutation is treated as one of the specific forces pushing populations away from that baseline. The other such forces are selection, drift, migration, and non-random mating. The central quantitative tool is the recursion equation for allele frequency change under recurrent mutation. The central conceptual payoff is mutation-selection balance. This is the steady-state frequency at which the constant creation of harmful alleles by mutation is exactly offset by their constant removal by selection. By the end, you should be able to write down the mutation recursion. You should be able to derive the mutation-selection balance formula for both recessive and dominant harmful alleles. You should be able to explain why mutation alone changes frequencies so slowly. You should also be able to use real numbers, from human genetic disease, from Drosophila, and from bacteria, to see these principles in action. The article also sets up why mutation's importance is usually indirect. It supplies the raw material that selection and drift then act upon. This is a theme picked up when we turn to how drift and selection jointly determine a variant's fate.
- Mutation rates in most organisms are low, typically 10⁻⁸ to 10⁻⁵ per gene per generation, so mutation alone changes allele frequency by a comparably tiny amount each generation.
- The one-way mutation recursion, q′ = q(1 − μ) + p·ν, shows that mutation pushes allele frequencies toward an equilibrium but does so at a rate proportional to μ (or ν), which is small.
- Mutation-selection balance explains why harmful alleles persist at low but non-zero, predictable frequencies: new mutations continually replace copies purged by selection.
- For a recessive lethal, the equilibrium frequency is q̂ = √(μ/s); for a dominant harmful allele, q̂ ≈ μ/s — a qualitatively different scaling that matters clinically.
- Classic experiments (Luria–Delbrück fluctuation test, mutation-accumulation lines in Drosophila and C. elegans) directly measured mutation rates and showed mutations arise spontaneously, not in response to need.
- Mutation is evolutionarily indispensable as the ultimate source of variation, but as a direct force changing allele frequencies within a generation, it is almost always the weakest of the evolutionary mechanisms.
Mutation supplies the variation that evolution runs on, but because mutation rates are so low, mutation by itself changes allele frequencies too slowly to explain evolutionary change — it must be amplified by selection or drift, and its own long-run signature shows up instead as a predictable equilibrium frequency of harmful alleles (mutation-selection balance).
The core idea
Mutation is the process that introduces new alleles into a population by altering DNA sequence — through nucleotide substitutions, insertions, deletions, or larger chromosomal changes — and it is the ultimate source of all genetic variation on which selection, drift, and migration act. But "mutation as a source of variation" is a distinct question from "mutation as a force that changes allele frequencies," and this article is about the second question. As a force, mutation is described by a simple recurrence: each generation, a certain fraction μ of one allele (say A) mutates irreversibly to another allele (a), and possibly a fraction ν mutates back (a → A). This converts mutation from a qualitative fact ("new variants appear") into a quantitative, population-genetic parameter that can be plugged into the same allele-frequency bookkeeping used for The Hardy-Weinberg Equilibrium as a Null Model. The key quantitative fact that organizes everything else in this article is that per-locus mutation rates are small. Direct measurements across bacteria, invertebrates, and vertebrates converge on rates of roughly 10⁻⁸ to 10⁻⁵ per gene per generation (somewhat higher per genome, because genomes contain many sites). Because the one-generation change in allele frequency due to mutation alone is proportional to μ, and μ is on the order of 10⁻⁵ or smaller, mutation moves allele frequencies by only a tiny amount each generation — far too slowly, acting alone, to account for observed rates of evolutionary change, local adaptation, or speciation. This is the sense in which mutation is a "weak" evolutionary force: not weak in importance, but weak in its direct, per-generation quantitative effect on allele frequencies. Where mutation's quantitative weakness becomes biologically important is at mutation-selection balance: a state in which recurrent mutation (constantly creating a deleterious allele at rate μ) is exactly counterbalanced by selection (constantly removing that allele at a rate depending on its fitness cost s and its dominance h). At this balance point, the deleterious allele frequency q̂ stops changing — not because mutation has stopped, and not because selection has stopped, but because the two opposing flows are equal. This is a genuine equilibrium in the Hardy-Weinberg sense (allele frequencies constant across generations) even though two forces are actively pushing on the population at all times. Mutation-selection balance is one of the most empirically well-supported quantitative predictions in population genetics, and it explains a fact of great medical relevance: why genetic diseases, especially severe recessive ones, never disappear from populations despite constant purifying selection against them.
Mechanism
Step 1 — Define the mutation rates. Let A be the "wild-type" allele and a be an alternative allele arising by mutation. Define μ as the forward mutation rate (A → a per generation) and ν as the back-mutation rate (a → A per generation). Both are probabilities per gene copy per generation, typically estimated from direct sequencing of parent-offspring trios, from reporter-gene reversion assays in microbes, or from long-term mutation-accumulation experiments.
Step 2 — Write the one-locus recursion. If p and q = 1 − p are the frequencies of A and a in the current generation, then after one round of mutation (before selection, drift, or anything else acts): q′ = q(1 − ν) + pμ This says: the new frequency of a equals the old copies of a that did not mutate back to A, plus new copies of a created by forward mutation from A. Equivalently, Δq = pμ − qν. Note the structure: this is a linear recursion with two opposing flows, exactly analogous to a chemical system reaching equilibrium between forward and reverse reactions.
Step 3 — Find the mutation-only equilibrium. Setting Δq = 0 and solving: q̂ = μ / (μ + ν) This is the frequency at which forward and backward mutation exactly balance, in the total absence of selection or drift. Because μ and ν are both tiny (order 10⁻⁵–10⁻⁸), the approach to this equilibrium is glacially slow: the number of generations needed to close much of the gap to q̂ scales as roughly 1/(μ + ν), meaning millions of generations for typical mutation rates. This is the formal demonstration of "mutation alone is a weak force" — the equilibrium exists, but reaching it unaided by other forces would take geological time.
Step 4 — Introduce selection against the mutant allele. In reality, most new mutations affecting fitness are deleterious, and selection removes them each generation. Let s be the selection coefficient (the proportional reduction in fitness of the a allele's phenotype) and h the dominance coefficient (h = 0 fully recessive, h = 1 fully dominant, 0 < h < 1 partial dominance). The three genotype fitnesses are:
| Genotype | AA | Aa | aa |
|---|---|---|---|
| Fitness | 1 | 1 − hs | 1 − s |
Selection removes copies of a each generation at a rate that depends on how often a appears in heterozygotes (exposed to selection with coefficient hs) versus homozygotes (exposed with coefficient s). Mutation resupplies a at rate μ·p each generation (ignoring the usually negligible back-mutation ν, since ν ≪ μ for most harmful alleles and s ≫ ν in importance).
Step 5 — Solve for mutation-selection balance. Setting the rate of loss by selection equal to the rate of gain by mutation, and solving the resulting quadratic (approximating p ≈ 1 since q is small):
- Recessive deleterious allele (h = 0): loss occurs only in aa homozygotes, at rate s·q². Balance requires μ ≈ s·q̂², giving
q̂ = √(μ/s)
- Dominant (or partially dominant) deleterious allele (h > 0): loss occurs mainly in heterozygotes, at rate ≈ hs·q (since heterozygotes are far more common than homozygotes when q is small). Balance requires μ ≈ hs·q̂, giving
q̂ ≈ μ / (hs) These two formulas differ in a way that matters enormously in practice: because μ/s is being square-rooted in the recessive case, q̂ for a recessive lethal (s = 1) with μ = 10⁻⁵ works out to about 0.003 — three orders of magnitude larger than μ itself. For a fully dominant lethal, q̂ ≈ μ/s = 10⁻⁵, essentially equal to the mutation rate, because every copy that appears is immediately exposed to selection and removed within a generation or two. Recessive deleterious alleles hide from selection in heterozygotes and so can build up to much higher equilibrium frequencies than dominant ones with the same mutation rate and same homozygous cost.
Step 6 — Recognize the direction of causality. It is essential not to reverse the logic: mutation-selection balance is not evidence that harmful alleles are "trying" to persist, nor that selection is "failing." It is a dynamic steady state, like a bathtub with the faucet (mutation) and drain (selection) both running — the water level (q̂) is constant not because nothing is happening, but because inflow equals outflow.
Step 7 — Compare mutation's rate of change to selection's. A second way to see mutation's weakness is to plot Δq per generation against q for mutation alone versus selection alone, holding μ and s at realistic values. Selection acting on even a modestly deleterious allele changes frequency by orders of magnitude more per generation than mutation does, except when the allele is already extremely rare (near q = 0), which is precisely the regime mutation-selection balance describes for strongly deleterious alleles.
The evidence
The claims above — that mutation rates are low, that mutations arise spontaneously rather than in response to selective need, and that mutation-selection balance quantitatively predicts observed disease frequencies — each rest on specific, classic experiments.
Luria and Delbrück's fluctuation test (1943). Before this experiment, it was debated whether bacterial resistance to a virus (bacteriophage) arose in response to exposure to the phage (a Lamarckian, directed-mutation view) or whether resistance mutations arose spontaneously, before and independent of exposure, and were simply revealed by phage selection. Salvador Luria and Max Delbrück grew many small parallel bacterial cultures from a common ancestor, then plated each culture on phage-laden agar and counted phage-resistant colonies. If resistance were induced by exposure to phage, every culture — having faced phage at the same moment — should show roughly the same, Poisson-distributed number of resistant colonies (mean ≈ variance). Instead, they observed huge variance between cultures — some plates had zero resistant colonies, others had hundreds — far exceeding Poisson expectation. This "jackpot" pattern is exactly what is predicted if resistance mutations occur spontaneously at random times during the pre-selection growth of each culture: a mutation arising early in a culture's growth produces a large clone of resistant descendants by the time of plating (a jackpot), while a mutation arising late produces very few. Luria and Delbrück fit this distribution mathematically and extracted a mutation rate for resistance, of order 10⁻⁸ per cell per generation — a founding empirical measurement of a bacterial mutation rate, and the definitive refutation of directed mutation. Luria and Delbrück shared the 1969 Nobel Prize in Physiology or Medicine (with Alfred Hershey) for this and related work.
The Lederbergs' replica plating (1952). Joshua and Esther Lederberg extended this logic with an elegant physical demonstration. They grew bacteria on a master plate without phage (or antibiotic), then pressed a velvet stamp onto the master plate and "replica plated" the identical colony pattern onto several selective plates. Resistant colonies appeared at the same spatial positions across independent replica plates — meaning the resistant cells were already present, at those specific locations, on the master plate before ever being exposed to the selective agent. This directly visualized what Luria and Delbrück had inferred statistically: resistance pre-exists selection.
Mutation-accumulation (MA) experiments. To measure mutation rates directly rather than infer them from selection, researchers maintain many replicate lines of an organism (bacteria, Drosophila melanogaster, Caenorhabditis elegans, Arabidopsis) at population sizes so small (often a single self-fertilized or sib-mated pair per line each generation) that selection is rendered nearly powerless against all but the most severely deleterious mutations — drift dominates, so nearly all new mutations, including mildly deleterious ones, are allowed to accumulate and can be counted by sequencing the lines after many generations relative to an ancestral control. Mutation-accumulation studies in C. elegans (Denver et al.) and Drosophila (Mukai; Keightley) directly measured genome-wide mutation rates and estimated the average fitness cost of new mutations, finding per-generation genomic mutation rates on the order of 10⁻⁸ to 10⁻⁷ per site, and confirming that most new mutations affecting fitness are mildly deleterious rather than beneficial or neutral. Modern whole-genome sequencing of human parent-offspring trios (starting with Roach et al. 2010 and refined by many subsequent studies) has directly counted de novo mutations, converging on an estimate of roughly 1.0–1.2 × 10⁻⁸ per nucleotide site per generation, or on the order of 70 new single-nucleotide mutations per diploid human genome per generation, with a pronounced paternal-age effect (older fathers transmit more new mutations, consistent with the greater number of germline cell divisions in spermatogenesis versus oogenesis).
Confirmation of mutation-selection balance. The predicted formulas have been tested against real disease frequencies. For achondroplasia (a fully dominant, near-fully-penetrant skeletal dysplasia), essentially all cases in unaffected families arise from new mutations at the FGFR3 gene, and the observed incidence (roughly 1 in 15,000–25,000 births) combined with the measured mutation rate (~1–5 × 10⁻⁵ per generation, elevated further by paternal age) fits the dominant-balance prediction q̂ ≈ μ/(hs) reasonably well given that affected individuals have historically had markedly reduced reproductive fitness. For severe recessive diseases such as cystic fibrosis in European-descended populations, the observed carrier frequency (heterozygote frequency around 1 in 25, giving q ≈ 0.02) is far higher than the recessive mutation-selection balance formula alone would predict from direct mutation-rate estimates — a discrepancy resolved by recognizing that heterozygote advantage (or population history/drift) contributes to maintaining the allele, illustrating that mutation-selection balance is a baseline model that real populations can depart from for identifiable reasons, not a rigid law.
Worked examples
Example 1 — A recessive lethal disease: Tay-Sachs. Classical infantile Tay-Sachs disease is caused by loss-of-function mutations in the HEXA gene; affected homozygotes historically died in early childhood (effectively s = 1, since homozygotes leave essentially no offspring), and the allele is recessive (h ≈ 0, since heterozygous carriers are clinically normal). Suppose the mutation rate to loss-of-function HEXA alleles is μ = 1.1 × 10⁻⁵ per generation (a realistic order-of-magnitude estimate for a gene of this size). At mutation-selection balance for a fully recessive lethal: q̂ = √(μ/s) = √(1.1 × 10⁻⁵ / 1) ≈ √(1.1 × 10⁻⁵) ≈ 0.0033 That predicts an allele frequency of about 0.33%, and a carrier (heterozygote) frequency of approximately 2pq ≈ 2q̂ ≈ 0.0066, or roughly 1 in 150 in a population at pure mutation-selection balance with no other forces. In most populations, observed Tay-Sachs carrier frequency is indeed on this general order (historically around 1 in 250–300 in general populations), though in Ashkenazi Jewish populations it rises to roughly 1 in 30 — a striking excess above the mutation-selection prediction that points to an additional force (proposed heterozygote advantage against another disease, or founder effects and genetic drift in a historically bottlenecked population) supplementing or replacing pure mutation-selection balance. This is exactly the kind of quantitative mismatch that tells biologists another evolutionary force must be involved, foreshadowing how Drift and Selection Interplay can produce departures from simple deterministic equilibria.
Example 2 — Comparing dominant versus recessive disease frequency for the same mutation rate and cost. Suppose two hypothetical diseases each have mutation rate μ = 2 × 10⁻⁶ per generation and impose the same fitness cost of s = 0.5 on affected homozygotes (aa), but disease X is fully recessive (h = 0) while disease Y is fully dominant (h = 1, so heterozygotes suffer the same 50% fitness reduction as homozygotes). For recessive disease X: q̂ = √(μ/s) = √(2×10⁻⁶ / 0.5) = √(4×10⁻⁶) = 2×10⁻³ For dominant disease Y: q̂ ≈ μ/(hs) = 2×10⁻⁶ / (1 × 0.5) = 4×10⁻⁶ The recessive disease reaches an equilibrium allele frequency roughly 500 times higher than the dominant disease, despite identical mutation rate and identical homozygous fitness cost. The biological reason: dominant deleterious alleles are "seen" and purged by selection every single generation they appear (even in single copy), while recessive deleterious alleles are invisible to selection whenever they occur in heterozygotes, and only a tiny fraction of copies (q̂² of them) ever end up homozygous and exposed to selection. This asymmetry is the single most important qualitative lesson of mutation-selection balance theory and explains why the overwhelming majority of known severe single-gene human diseases are recessive rather than dominant — dominant lethals simply cannot accumulate to appreciable frequency, no matter how often they mutate in.
Common misconceptions & traps
| Misconception | Why it's wrong |
|---|---|
| "Mutation is the main force driving allele-frequency change generation to generation." | Mutation rates (~10⁻⁵–10⁻⁸) are far too low for mutation alone to explain typical per-generation frequency shifts; selection and drift act far faster once variation exists. |
| "Mutations arise because organisms need them (directed mutation)." | The Luria–Delbrück fluctuation test and Lederberg replica-plating experiment showed mutations (e.g., phage resistance) exist before and independent of the selective agent that reveals them. |
| "If a harmful allele is in mutation-selection balance, selection has failed to remove it." | Selection is removing it every generation, at the same rate mutation resupplies it — a genuine dynamic steady state, not a failure of selection. |
| "Recessive and dominant deleterious alleles should reach similar equilibrium frequencies for the same mutation rate." | They differ by orders of magnitude: q̂ = √(μ/s) for recessive vs. q̂ ≈ μ/(hs) for dominant, because heterozygotes shield recessive alleles from selection. |
| "A high carrier frequency of a recessive disease always means a high mutation rate." | It can instead indicate heterozygote advantage, a population bottleneck, or drift — as with the elevated Tay-Sachs carrier frequency in some populations, which exceeds simple mutation-selection predictions. |
| "New mutations are usually beneficial, which is why mutation matters for adaptation." | Mutation-accumulation experiments show most fitness-affecting new mutations are mildly deleterious or neutral; beneficial mutations are rare, and mutation matters for adaptation mainly by supplying the raw material selection later filters. |
The scientist behind it
Origins · 1 minSalvador Luria and how this was discovered
Salvador Luria trained as a physician in Italy before turning to physics-influenced biology, fleeing fascist Italy in 1940 and settling in the United States, where he began working on bacteriophages — viruses that infect bacteria — as simple, fast-reproducing model systems for studying heredity. Working with Max Delbrück, a former physicist who had brought a rigorously quantitative sensibility to biology, Luria set out to resolve a genuinely contested question of the era: when a bacterial population is challenged with phage and a resistant subpopulation survives, did those bacteria become resistant in response to the phage (a view with echoes of Lamarckian inheritance of acquired traits), or were resistant mutants already present, purely by chance, before the phage ever arrived? Luria devised the fluctuation test after watching a slot machine at a faculty dance and realizing that the "jackpot" statistics of rare, randomly timed wins might be mathematically identical to the statistics of mutations arising at random times during bacterial growth — an anecdote that captures how the experiment's power came from recognizing a statistical signature (variance vastly exceeding the mean) rather than simply observing individual events. Delbrück supplied the mathematical distribution predicting exactly this signature under the spontaneous-mutation hypothesis, and their combined statistical and experimental result became one of the cleanest demonstrations in the history of genetics that mutation is a spontaneous, selection-independent process — variation arises first, and selection only afterward decides what survives.
Origins · 1 minhow this was discovered
Where this matters today
Mutation-selection balance is not an abstract exercise; it underlies practical work across medicine, agriculture, and conservation. In clinical genetics, the recessive versus dominant scaling (√(μ/s) versus μ/(hs)) explains why genetic counselors expect newly diagnosed dominant conditions such as achondroplasia or many cases of neurofibromatosis type 1 to frequently represent de novo mutations with unaffected parents, while recessive conditions such as cystic fibrosis or sickle-cell disease imply two unknowingly carrier parents — a distinction that directly shapes recurrence-risk counseling for families. In livestock and crop breeding, mutation-selection balance theory helps breeders understand why deleterious recessive alleles (causing, for example, inherited metabolic defects in cattle breeds derived from small founder populations) persist at low but stable frequencies despite generations of culling, motivating genomic carrier-screening programs analogous to human carrier testing. In conservation biology, small, bottlenecked populations (cheetahs, Tasmanian devils, island endemics) experience weakened selection relative to drift, allowing mutation-selection balance predictions to be violated in the direction of higher deleterious-allele frequencies than expected — a quantitative early-warning signal for inbreeding depression and mutational meltdown risk that conservation geneticists explicitly model when designing captive-breeding and translocation programs. In antimicrobial resistance and oncology, the Luria–Delbrück logic of spontaneous, pre-existing mutation directly underlies modern understanding of why bacteria evolve antibiotic resistance and tumors evolve chemotherapy resistance during treatment rather than because of it — resistant subclones typically pre-exist exposure, which is why combination therapies (attacking a population with multiple drugs simultaneously, so that no single pre-existing mutant can survive all of them) are designed explicitly around this principle. Finally, direct trio-sequencing measurement of human mutation rates now feeds into estimates of the timing of human evolutionary events (using the mutation rate as a molecular clock) and into clinical algorithms flagging elevated de novo mutation risk with advanced paternal age — a live research and public-health application of a rate first estimated, indirectly, in 1943.
Ask the tutor: How would mutation-selection balance change if the deleterious allele also had a small heterozygote fitness advantage, like sickle-cell trait against malaria?
Why is mutation called "weak" if it's the ultimate source of all genetic variation?
These are two different claims. As a source of variation, mutation is irreplaceable — no other process creates genuinely new alleles from scratch (recombination and migration only reshuffle or import existing variation). But as a force changing allele frequencies within a generation, mutation's effect is proportional to the mutation rate, typically 10⁻⁸–10⁻⁵ per locus per generation. That is a minuscule shift compared to what selection (which can change frequencies by percents per generation) or drift (which can fix or lose alleles within tens to hundreds of generations in small populations) can achieve. So mutation is foundational but slow-acting alone; its real evolutionary power is unlocked only once selection or drift act on the variation it has supplied.
What exactly is mutation-selection balance, in plain terms?
It's the steady-state allele frequency reached when a harmful allele's constant creation by new mutations is exactly matched by its constant removal by natural selection. Picture a bathtub with the faucet (mutation, rate μ) and drain (selection, rate depending on s and h) both running continuously — the water level (allele frequency) stops changing once inflow equals outflow, even though both processes are still actively happening. This produces a specific, calculable equilibrium frequency, q̂ = √(μ/s) for recessive alleles or q̂ ≈ μ/(hs) for dominant alleles, that can be tested against real disease-frequency data.
Why do recessive genetic diseases dramatically outnumber dominant ones in human genetics?
Because of how dominance changes the exposure of an allele to selection. A dominant deleterious allele is expressed, and therefore selected against, in every carrier, every generation — so it gets purged almost as fast as it's created, keeping its equilibrium frequency close to the mutation rate itself (μ/hs). A recessive deleterious allele is invisible to selection whenever it's in a heterozygote, since heterozygotes are phenotypically normal; only the small fraction of the time it occurs in homozygotes is it exposed. This "hiding" lets recessive alleles accumulate to equilibrium frequencies many times higher (scaling as the square root of μ/s rather than linearly), which is exactly why conditions like cystic fibrosis and Tay-Sachs disease persist at far higher carrier frequencies than any dominant lethal condition does.
Did the Luria–Delbrück experiment prove that mutations are random with respect to fitness, or only that they arise before selection?
Strictly, the fluctuation test demonstrated that resistance mutations are pre-existing, arising at random times during growth independent of whether phage was ever applied — it directly refuted the idea that phage exposure induces or directs resistance mutations. It's often summarized as showing mutations are "random," meaning their occurrence is not caused by, or aimed at, the environmental challenge that later selects them. It does not claim that all mutations are equally likely at all sites (mutation rates vary by sequence context) — only that the decision of which mutations occur is not influenced by the organism's current need for them.
Can mutation rates themselves evolve, and does that complicate mutation-selection balance?
Yes — mutation rate is itself a trait shaped by selection, generally favoring lower rates in stable environments (since most mutations affecting fitness are deleterious) but sometimes favoring elevated, localized mutagenesis (as in some stress-response or hypermutable bacterial systems, and in the diversity-generating machinery of vertebrate immune receptor genes). This means μ is not always a fixed constant across all time and all loci; mutation-selection balance calculations use a locus- and lineage-specific μ, best estimated empirically (e.g., via trio sequencing or mutation-accumulation lines) rather than assumed universally constant.
How does mutation-selection balance connect to the topics of drift and multi-locus evolution that come after it?
Mutation-selection balance as derived here assumes an effectively infinite population, so that selection and mutation are the only forces and deterministic recursions apply exactly. In real, finite populations, genetic drift adds random fluctuation around q̂, and in small populations drift can overwhelm weak selection entirely, allowing deleterious alleles to drift to higher frequency than mutation-selection balance predicts (relevant to Drift and Selection Interplay). Likewise, this article treats one locus in isolation, but real deleterious mutations interact with variation at other loci through linkage and epistasis, which is the subject taken up under Evolution at Multiple Loci. Mutation-selection balance is thus the correct starting baseline, extended by both stochasticity and multi-locus complexity.
If mutation changes allele frequency so slowly, why do we still call it "the ultimate source of variation" that fuels adaptation?
Because "fuels adaptation" refers to a different timescale and a different causal role than "changes frequency each generation." Mutation supplies the raw alleles — without it, no new variant would ever exist for selection to act on, no matter how strong selection was. Once a mutation exists, however, it is selection (favoring beneficial variants) and drift (randomly fixing or losing neutral or nearly-neutral ones) that do the heavy lifting of actually changing its frequency toward fixation or loss, often within tens to hundreds of generations for strongly selected alleles. So mutation's role is upstream and enabling; selection and drift's role is downstream and rate-determining. Both are needed, at different stages, for the process to be called evolution at all.