Natural Selection and the Concept of Fitness
A peppered moth lands on a soot-darkened tree trunk in industrial Manchester, 1850s. A bird flies past without seeing it. Its pale-winged sibling, sitting on the same trunk, is eaten within the hour. Nothing mysterious happened here — no moth "tried" to survive. No population "wanted" to adapt. Yet after a few dozen generations, nearly every moth in the region was dark. This chapter is about the quantitative machinery that turns differences in survival and reproduction into evolutionary change: fitness, and its formal partner, the selection coefficient. Natural selection is often described in a single vague sentence — "survival of the fittest." That phrase hides more than it reveals. Fittest at what? Measured how? Compared to whom? Evolutionary biology answers these questions with numbers. Fitness is not a mystical life-force or a measure of strength. It is a bookkeeping device that tracks, on average, how many copies of a genotype end up in the next generation. This is measured relative to other genotypes in the same population. Once fitness is defined this precisely, natural selection stops being a metaphor. It becomes an algebraic engine that predicts, generation by generation, how allele frequencies change. This matters because fitness is the single quantity that connects Allele and Genotype Frequencies to the outcome of evolution. A population may evolve toward one extreme (directional selection). It may converge on an intermediate optimum (stabilizing selection). Or it may split toward two extremes (disruptive selection). These topics are explored under Directional Selection and Stabilizing and Disruptive Selection. All of them reduce to differences in relative fitness among genotypes. Without a quantitative fitness framework, selection is just a story. With it, selection becomes a prediction machine. It can be tested against real data, from moth wing color to HIV drug resistance to lactase persistence in humans. This article builds the concept from first principles. It explains what fitness precisely means, and how selection coefficients quantify the strength of selection. It shows how these numbers translate into changes in allele frequency across generations. It also reviews classic and modern experiments. These experiments show that these numbers are not abstractions. They are measurable, predictive biological realities.
- Fitness (w) is the average relative contribution of a genotype's offspring to the next generation, usually normalized so the best genotype has w = 1.
- The selection coefficient (s) measures the fitness disadvantage of a genotype relative to the fittest one: s = 1 − w.
- Fitness combines survival and reproductive success — it is relative, not absolute, and it is genotype- and environment-specific, never a fixed property of a species.
- Deterministic single-locus models use fitness values to predict how allele frequencies shift each generation under The Hardy-Weinberg Equilibrium as a Null Model-violating conditions.
- Classic experiments — from Kettlewell's moth predation studies to Endler's guppy transplants and modern E. coli long-term evolution — have directly measured selection coefficients in nature and the lab.
- Fitness is not synonymous with strength, dominance, or population abundance; conflating these produces some of the most persistent misconceptions in evolutionary biology.
Natural selection is simply differential reproductive success among heritable variants, and fitness is the precise number — relative, not absolute — that measures how much more (or less) successful one variant is than another.
The core idea
Fitness, symbolized w, is the average number of offspring (more precisely, of gene copies transmitted to the next generation) produced by individuals of a given genotype, relative to other genotypes in the same population and environment. Absolute fitness (sometimes W) could be measured in raw offspring counts, but because population size and mean fertility fluctuate for reasons unrelated to genotype (drought, resource pulses, predator cycles), evolutionary biologists almost always work with relative fitness: each genotype's absolute fitness divided by the highest absolute fitness in the population, so the best-performing genotype gets w = 1 and everything else is scaled between 0 and 1. This rescaling is what makes fitness comparable across studies, environments, and species — it strips away the noise of absolute population growth and leaves only the differential success that drives allele frequency change. The selection coefficient, s, is defined as: ``` s = 1 − w ``` It measures how much a genotype is disfavored relative to the fittest genotype. If the fittest genotype has w = 1.00 and a competitor has w = 0.90, then s = 0.10 — a 10% relative disadvantage per generation. Selection coefficients can be vanishingly small (s = 0.001, essentially undetectable over a human lifetime but decisive over geological time) or dramatic (s = 1, meaning the genotype leaves no offspring at all — a lethal). Crucially, s is always defined relative to a reference genotype in a specified environment; there is no such thing as an absolute, context-free selection coefficient. Fitness is best understood as a composite of two multiplicative components that separate the classical stages of the life cycle: viability (probability of surviving to reproductive age) and fecundity/mating success (average number of offspring produced once reproductive). A genotype can have high viability but poor mating success (or vice versa), and total fitness is the product of both components across the entire life cycle, including offspring viability itself. This is why fitness is fundamentally a lifetime, per-generation measure — not simply "how many babies," but "how many babies survive to breed themselves," iterated across the life cycle.
Mechanism
Step 1 — Heritable variation exists. Selection requires a population in which individuals differ in some trait, and that variation has a genetic basis (fully or partially) so that it can be transmitted to offspring. Without heritability, differential survival changes nothing about the next generation's gene pool — this is precisely the assumption The Hardy-Weinberg Equilibrium as a Null Model uses as its baseline: no selection means genotype frequencies stay in fixed Hardy-Weinberg proportions, generation after generation.
Step 2 — Genotypes differ in fitness. Suppose a single locus has two alleles, A₁ and A₂, producing three genotypes: A₁A₁, A₁A₂, A₂A₂, with genotype frequencies p², 2pq, q² under Hardy-Weinberg (where p and q are the allele frequencies). Each genotype is assigned a relative fitness: ``` Genotype: A₁A₁ A₁A₂ A₂A₂ Fitness: w₁₁ w₁₂ w₂₂ ```
Step 3 — Selection acts within a generation. Individuals are born in Hardy-Weinberg proportions (p², 2pq, q²), then survive and reproduce differentially according to their fitness. After selection, the genotype frequencies among the breeding adults are no longer p², 2pq, q² — they are reweighted by each genotype's w: ``` freq(A₁A₁) after selection ∝ p² · w₁₁ freq(A₁A₂) after selection ∝ 2pq · w₁₂ freq(A₂A₂) after selection ∝ q² · w₂₂ ```
Step 4 — Allele frequencies are recomputed from surviving genotypes. The new frequency of allele A₁ in the next generation, p′, is obtained by counting A₁ alleles among the survivors and dividing by the total, then normalizing by the population's mean fitness, w̄ = p²w₁₁ + 2pq·w₁₂ + q²w₂₂: ``` p′ = [p²w₁₁ + pq·w₁₂] / w̄ ```
Step 5 — Iterate. This new p′ becomes the starting allele frequency for the next round of random mating, producing new Hardy-Weinberg genotype proportions, which are again subjected to selection. Repeated generation after generation, this recursive process is what actually produces long-term evolutionary change — a moth population shifting from 5% dark morphs to 95% dark morphs over 50 years is this exact algebra, run fifty times in a row.
Step 6 — The rate of change depends on s, on dominance, and on allele frequency itself. Selection is fastest when the disadvantaged allele is common (many individuals are exposed to selection) and when it is not fully recessive or fully dominant, because sheltering effects (heterozygotes carrying a hidden recessive allele) slow the purge of rare deleterious alleles and slow the rise of rare beneficial ones. This is why a beneficial dominant allele sweeps to fixation far faster than a beneficial recessive allele starting at the same low frequency — recessive rare alleles are almost always masked in heterozygotes and invisible to selection until they become common enough to appear in homozygotes.
The evidence
Bernard Kettlewell's peppered moth experiments (1953–1956, England) were among the first controlled field tests of natural selection acting on a visible trait, providing direct, quantitative evidence for differential survival driven by predation. Biston betularia exists in a pale, speckled "typica" form and a melanic "carbonaria" form controlled largely by a single dominant allele. Kettlewell released marked moths of both morphs into two woodlands: polluted Birmingham, where tree trunks were blackened by soot, and unpolluted Dorset, where trunks retained pale lichen. He then recaptured survivors and, in a further test, filmed birds (great tits, redstarts) preferentially picking off moths that stood out against the bark. In the polluted wood, dark moths survived and were recaptured at roughly twice the rate of pale moths; in the unpolluted wood, the pattern reversed. This gave a direct field measurement of differential viability — the raw material from which a selection coefficient is calculated — and matched the historical record: carbonaria rose from under 2% of the population in 1848 near Manchester to over 90% by the 1890s, a change explicable only if s for the pale allele in the sooty environment was substantial (subsequent estimates put s in the range of 0.1–0.5 per generation during the peak of industrial pollution). Later re-analyses (Majerus, Cook, and others in the 2000s) refined methodology and confirmed the core finding — bird predation, biased by camouflage against pollution-darkened bark, is the principal selective agent — while also showing the story is more nuanced (moths rest in more varied locations than Kettlewell assumed), a useful reminder that classic experiments are refined, not simply overturned, by follow-up work.
John Endler's guppy transplant experiments (1970s–1980s, Trinidad and controlled ponds) provided an even more rigorously quantified demonstration, because Endler could manipulate predation pressure directly and measure resulting changes in trait frequency and estimated selection coefficients across known numbers of generations. Wild guppies (Poecilia reticulata) from streams with a fierce predator (the cichlid Crenicichla) are drab and dull-spotted; guppies from streams with a weaker predator (the killifish Rivulus) or no predator are gaudier, with large, colorful spots favored by female mate choice. Endler transplanted guppies from high-predation pools to previously guppy-free, low-predation pools upstream. Within roughly 2 years (about 15 generations), transplanted populations evolved significantly larger, more numerous, and more colorful spots — the color pattern shifted toward the low-predation phenotype. He replicated this in outdoor artificial ponds with controlled predator regimes (no predator, Rivulus, or Crenicichla) and got the same directional result each time, allowing him to estimate selection differentials on spot number and size directly from generation-by-generation phenotype distributions. This experiment is prized because it demonstrates selection acting in real time, with replication, a manipulated independent variable (predation regime), and quantifiable selection coefficients — exactly the rigor that turns "survival of the fittest" from slogan into science.
Richard Lenski's Long-Term Evolution Experiment (LTEE, begun 1988, ongoing) with Escherichia coli extends this logic into the laboratory across tens of thousands of generations. Twelve populations founded from a single clone have been propagated in identical glucose-limited minimal medium, with 1% of each population transferred daily (roughly 6.6 generations per day). Because ancestral clones were frozen at intervals, researchers can directly measure fitness by competing evolved clones against the ancestor in head-to-head growth competitions, computing relative fitness as the ratio of realized growth (Malthusian) parameters. By generation 50,000, mean fitness relative to the ancestor had increased roughly 70% in most lines, with the rate of gain following a decelerating (power-law-like) trajectory — large early jumps, diminishing but still-detectable returns even after 30,000+ generations. One population (Ara-3) famously evolved the ability to use citrate as a carbon source aerobically around generation 31,500, a rare, historically contingent mutation whose fitness advantage (once available) was dramatic. The LTEE turns fitness and selection coefficients from theoretical constructs into directly measured, repeatable, publishable numbers — the same conceptual quantity as Kettlewell's recapture ratios, just measured with a spectrophotometer instead of a mist net.
Worked examples
Example 1 — A selection cross with an asymmetric lethal. Consider a single locus in a large, randomly mating population of an annual plant, with allele frequencies p(A₁) = 0.8 and q(A₂) = 0.2 entering the season. Suppose A₂A₂ homozygotes are outcompeted for light and rarely set seed, giving fitnesses: ``` w₁₁ = 1.00 (A₁A₁) w₁₂ = 0.90 (A₁A₂) w₂₂ = 0.40 (A₂A₂) ``` Selection coefficients: s₁₂ = 1 − 0.90 = 0.10 (mild heterozygote disadvantage — this locus shows partial dominance of the deleterious allele, not full recessivity); s₂₂ = 1 − 0.40 = 0.60 (severe homozygote disadvantage). Starting Hardy-Weinberg genotype frequencies: p² = 0.64, 2pq = 0.32, q² = 0.04. Mean fitness: ``` w̄ = (0.64)(1.00) + (0.32)(0.90) + (0.04)(0.40) = 0.640 + 0.288 + 0.016 = 0.944 ``` New frequency of A₁ after one generation of selection: ``` p′ = [p²w₁₁ + pq·w₁₂] / w̄ = [0.64(1.00) + (0.8)(0.2)(0.90)] / 0.944 = [0.640 + 0.144] / 0.944 = 0.784 / 0.944 = 0.830 ``` So p rises from 0.800 to 0.830 in a single generation — a gain of 3 percentage points. Because A₂ is only partially recessive (heterozygotes suffer some fitness cost, s₁₂ = 0.10), selection can "see" and remove A₂ alleles even when they are hidden in heterozygotes, so the allele frequency change is faster than it would be if A₂ were fully recessive (w₁₂ = 1.00) at the same starting frequency. Iterating this recursion generation after generation is exactly how a locus moves toward fixation of A₁ — the trajectory decelerates as q becomes rare, because rare alleles increasingly hide in heterozygotes.
Example 2 — Estimating a real-world selection coefficient from field data (Kettlewell-style calculation). Suppose a mark-release-recapture study releases 200 pale (typica) and 200 dark (carbonaria) moths onto sooty bark, and after 24 hours, recaptures 30 pale moths and 65 dark moths (recapture rate treated as a proxy for survival). Absolute "fitness" proxies: ``` W(pale) = 30/200 = 0.15 W(dark) = 65/200 = 0.325 ``` Relative fitness, normalizing to the higher value: ``` w(dark) = 0.325/0.325 = 1.00 w(pale) = 0.15/0.325 = 0.46 ``` Selection coefficient against the pale morph in this sooty habitat: ``` s(pale) = 1 − 0.46 = 0.54 ``` This is a strong selection coefficient — over half the relative reproductive potential of pale moths is lost each generation in this environment, consistent with the historical speed of the carbonaria rise (from ~2% to ~90% in roughly 50 generations near industrial Manchester). Note the essential caveat: this s applies only to this environment; the same calculation performed in unpolluted Dorset woodland would flip the sign, illustrating that fitness values are never fixed species properties — they are environment-specific, context-dependent numbers, recalculated for every habitat and every generation.
Common misconceptions & traps
| Misconception | Why it's wrong |
|---|---|
| "Fitness means physical strength or health" | Fitness is reproductive success, not vigor — a frail but highly fecund genotype can have higher fitness than a robust but sterile one. |
| "The fittest genotype is whichever one is most common" | Common genotypes can be common for historical or drift reasons; current fitness is about relative reproductive output now, not past abundance. |
| "Dominant alleles have higher fitness / are more common" | Dominance is a molecular relationship between alleles in a heterozygote; it says nothing about survival value. Recessive alleles can be common, and dominant alleles can be strongly deleterious (e.g., Huntington's allele in humans is dominant, yet historically had modest fitness costs pre-reproductive-age). |
| "Selection acts to benefit the species or the population" | Selection acts on differential reproduction of genotypes within a population; it has no foresight and cannot "plan" for long-term species survival — a trait can spread even if it drives the population toward decline. |
| "A high selection coefficient always means rapid fixation" | The rate of allele frequency change also depends on current allele frequency and dominance; a large s produces almost no change per generation if the allele is very rare and fully recessive, since it is nearly invisible to selection while hidden in heterozygotes. |
| "Fitness is a fixed number for a genotype" | Fitness values are specific to an environment, a population's genetic background (epistasis), and even to frequency-dependent contexts (e.g., rare-morph advantage); the same genotype can have different w in different settings. |
The scientist behind it
The mathematical formalization of fitness as a numerical, selection-coefficient-based quantity is largely credited to Ronald Aylmer Fisher, working alongside J.B.S. Haldane and Sewall Wright as one of the founders of population genetics and the modern evolutionary synthesis. Fisher, a statistician by training, sought to reconcile Mendelian genetics (discrete, particulate inheritance) with Darwinian gradualism (continuous, slow evolutionary change) — a reconciliation that seemed paradoxical to many biologists in the early twentieth century, who thought Mendelism implied evolution by sudden jumps. Fisher showed mathematically that selection acting on many small-effect Mendelian loci could produce smooth, continuous evolutionary change indistinguishable from classical Darwinian gradualism, and he formalized fitness, selection coefficients, and the change in mean fitness across generations in rigorous algebraic terms in his 1930 book The Genetical Theory of Natural Selection — the same work that introduced the Fundamental Theorem of Natural Selection, relating the rate of fitness increase in a population to the additive genetic variance in fitness.
Origins · 1 minRonald Aylmer Fisher and how this was discovered
Where this matters today
Quantitative fitness thinking is not a historical curiosity confined to moths and finches — it is the operational core of applied evolutionary biology across medicine, agriculture, and conservation. In oncology, tumor cell populations evolve under selection imposed by chemotherapy and targeted drugs; clinicians and researchers now model drug-resistant clones using the same relative-fitness/selection-coefficient framework, informing adaptive therapy strategies that deliberately allow drug-sensitive cells to persist (because they outcompete resistant clones when the drug is withdrawn) rather than driving tumors toward pure resistance through relentless maximal dosing. In infectious disease, selection coefficients quantify how fast antibiotic-resistant bacteria or antiviral-resistant viruses (like HIV under incomplete drug regimens, or influenza under vaccine-induced immune pressure) spread through a population, directly informing dosing schedules and combination-therapy design meant to keep resistance-allele selection coefficients low or negative. In agriculture, resistance management for pesticides and herbicides is fundamentally an exercise in manipulating selection coefficients — refuge strategies for Bt-crop pest resistance work by keeping a reservoir of susceptible insects so that resistance alleles remain rare and largely hidden in heterozygotes, slowing the spread predicted by exactly the recursive equations shown above. In conservation biology, fitness estimates from pedigree and genomic data help quantify inbreeding depression and predict extinction risk in small, isolated populations, while assisted gene flow programs (moving individuals between populations to introduce beneficial alleles) are designed and evaluated using selection-coefficient estimates for climate-adaptive traits, such as heat tolerance alleles in corals or drought tolerance in trees facing rapid climate change.
Ask the tutor: How does overdominance (heterozygote advantage), like the sickle-cell allele in malaria-endemic regions, keep two alleles stably balanced instead of one going to fixation?
What's the difference between absolute fitness and relative fitness?
Absolute fitness (W) is the raw average number of offspring (or gene copies) a genotype produces per generation — a count that depends on resource availability, population density, and other ecological factors independent of genotype. Relative fitness (w) rescales absolute fitness by dividing every genotype's W by the highest W in the population, so the best genotype becomes w = 1 and all others fall between 0 and 1. This rescaling matters because population-genetic models care about differences among genotypes, not absolute numbers — a population crashing from famine can still show clear selection if one genotype consistently outperforms another proportionally, even though every genotype's absolute offspring count has fallen.
Can a genotype have a selection coefficient greater than 1?
No — by definition s = 1 − w, and since relative fitness w is normalized to lie between 0 and 1 (with the best genotype at w = 1), s can range only from 0 (no disadvantage; equally fit as the best genotype) to 1 (a lethal or completely sterile genotype that contributes zero offspring). An s of exactly 1 corresponds to complete elimination from the breeding population in that generation, such as a fully lethal recessive homozygote in an organism with no compensating mechanisms. Selection coefficients cannot go negative in this "disadvantage relative to the best" formulation either, because by construction nothing outperforms the reference (w=1) genotype.
Does natural selection act on individuals, genotypes, or alleles?
Selection operates through the differential survival and reproduction of individual organisms, but because organisms don't perfectly replicate themselves (sexual reproduction reshuffles alleles every generation), the statistical, population-level consequence of that individual-level process is tracked using genotype and allele frequencies. An individual either survives and reproduces or it doesn't — fitness values (w₁₁, w₁₂, w₂₂) are averages computed across all individuals sharing a genotype. Evolutionary biologists routinely switch between these levels of description because they are mathematically connected: individual differential success, averaged within genotype classes, is exactly what produces the allele-frequency change predicted by single-locus selection models.
Is heterozygote fitness always intermediate between the two homozygotes?
No — this is a common but incorrect default assumption. When w₁₂ falls exactly between w₁₁ and w₂₂, this is called additive or codominant fitness. But heterozygotes can also have fitness equal to one homozygote (complete dominance/recessivity of fitness effects), fitness exceeding both homozygotes (heterozygote advantage or overdominance, as in the classic case of sickle-cell allele heterozygotes having higher malaria-endemic-region fitness than either homozygote), or fitness below both homozygotes (underdominance). Each of these dominance relationships produces qualitatively different evolutionary trajectories — overdominance, notably, can maintain both alleles indefinitely at a stable intermediate frequency rather than driving either to fixation.
How does fitness relate to the concepts of directional, stabilizing, and disruptive selection?
Those three classic modes describe how fitness varies across a range of phenotypic values, rather than across just two or three discrete genotypes. In directional selection, fitness increases monotonically toward one phenotypic extreme (e.g., ever-larger beak size), pulling the population's trait mean in that direction generation after generation. In stabilizing selection, fitness peaks at an intermediate phenotype and declines toward both extremes (e.g., human birth weight, where both very low and very high weights carry mortality risk), which reduces phenotypic variance without shifting the mean. In disruptive selection, fitness is lowest at the intermediate phenotype and highest at both extremes, potentially splitting a population into two morphs. All three are simply different fitness functions mapped onto phenotype, built from exactly the same w and s machinery introduced here.
Why doesn't natural selection produce "perfect" organisms if it relentlessly favors higher fitness?
Because selection can only act on the heritable variation currently present, it optimizes traits relative to the immediate environment and existing genetic architecture, not toward some absolute ideal. Constraints include pleiotropy (one gene affecting multiple traits, so improving one can worsen another), genetic linkage (beneficial and harmful alleles inherited together), finite mutational supply (the "right" mutation may simply never arise), trade-offs (e.g., a fitness gain in reproduction often costs somatic maintenance), and the fact that environments themselves keep changing, constantly resetting what counts as high fitness. Additionally, genetic drift in finite populations can fix mildly deleterious alleles or lose beneficial ones by chance, especially when selection coefficients are small relative to population size.
How do biologists actually measure selection coefficients in wild populations?
Common methods include mark-recapture studies (comparing survival rates of different phenotypes/genotypes, as Kettlewell did), longitudinal studies of marked individuals with known genotypes tracked for lifetime reproductive success, common-garden and reciprocal transplant experiments (as Endler did with guppies), and, increasingly, genomic methods that scan for signatures of selection in DNA sequences (reduced diversity around a selected locus, unusual allele frequency trajectories in long-term or ancient DNA datasets, or elevated rates of nonsynonymous substitution). Laboratory systems like Lenski's E. coli LTEE allow the most direct measurement: competing evolved and ancestral strains head-to-head and computing relative growth rates gives a nearly exact experimental estimate of relative fitness and its associated selection coefficient.