Directional Selection

Natural selection acts on phenotypic variation, but its evolutionary consequences depend on how selection reshapes the distribution of a trait across generations. Biologists recognize three principal modes based on which part of the distribution is favored. Stabilizing selection favors the middle of the distribution and prunes both extremes, reducing variance without shifting the mean, think human birth weight, where infants of intermediate mass survive best. Disruptive selection favors both extremes over the middle, potentially splitting a population into two phenotypic clusters. Directional selection favors one extreme over the other (and over the mean), causing the entire distribution to slide toward that extreme generation after generation. Formally, directional selection exists whenever fitness is a monotonic function of trait value across the relevant range, larger is always better (or always worse), rather than fitness peaking at some intermediate optimum. This distinguishes it cleanly from stabilizing selection, where fitness is a hump-shaped function peaking at an intermediate value. The population's phenotypic mean, call it $\bar{z}$, is expected to increase (or decrease) each generation as long as directional selection persists and heritable variation remains. The key quantities that describe directional selection are the selection differential ($S$), the difference between the mean trait value of selected (reproducing) individuals and the mean of the whole population before selection, and the selection gradient or response to selection ($R$), the actual shift in the offspring generation's mean. These are connected by the trait's heritability ($h^2$), the proportion of phenotypic variance attributable to additive genetic effects, through the breeder's equation:

This single equation is the quantitative backbone of directional selection: it says that evolutionary response is strong when selection is strong (large $S$) and when the trait is highly heritable (large $h^2$), and it vanishes if either term is zero, no response occurs from selection on purely environmental variation, and no response occurs if there is no selection differential even for a highly heritable trait. Understanding this equation is essential background from Modeling Selection at a Single Locus and the general logic of Natural Selection and the Concept of Fitness, both of which underlie how $S$ and $h^2$ are themselves generated by underlying genetic and ecological processes.

Step 1: Variation exists. Any population shows continuous phenotypic variation in most quantitative traits, e beak depth, body size, coloration intensity, arising from segregating variation at many loci (polygenic architecture) plus environmental effects on development. This variation is the raw material selection acts upon; without it, there is nothing to select.

Step 2: A consistent fitness gradient across the trait range. Directional selection arises when survival or reproductive success increases (or decreases) monotonically with trait value, e a drought that makes only large, hard seeds available so that only birds with large, strong beaks can crack them and survive to breed. Crucially, the cause of this fitness gradient is ecological, predation efficiency, climatic tolerance, competitive ability, resistance to a toxin, and the selection itself has no foresight: individuals with better matching phenotypes simply leave more descendants, mechanically, without any "goal" of adaptation.

Step 3: Differential survival/reproduction changes the breeding population's trait mean. The individuals who survive to reproduce (or who mate more, sire more offspring) are a non-random, phenotypically skewed sample of the original population. Their mean trait value differs from the population mean before selection by the selection differential $S$.

Step 4: Heritable variation transmits the shift to offspring. Because much of the phenotypic variation has an additive genetic basis, offspring of the selected parents tend to resemble them more than they resemble the pre-selection population average. The realized shift in the offspring generation's mean, $R$, is a fraction $h^2$ of $S$. This is where quantitative genetics meets single-locus population genetics: at the level of individual loci contributing to the trait, alleles that increase the favored phenotype rise in frequency each generation, exactly as in a simple one-locus directional selection model with selection coefficients acting against the disfavored allele or genotype.

Step 5: Iteration and its limits. Generation after generation, provided selection pressure and heritable variation persist, the population mean creeps (or leaps) toward the favored extreme. This cannot continue indefinitely: as favorable alleles approach fixation, additive genetic variance for the trait declines (the "selection limit"), environmental or physiological constraints (allometric, developmental, biomechanical) may cap further change, and antagonistic pleiotropy or correlated costs (e a bigger beak that is harder to control, or a resistance allele with a fitness cost in the absence of the toxin) can create trade-offs that eventually halt or reverse the trend. This connects forward to Adaptation and Its Limits, where the constraints on sustained directional change become the central topic.

A single-locus view. At one biallelic locus contributing to the trait, suppose allele $A_1$ increases the favored phenotype relative to $A_2$, with relative fitnesses $w_{11} > w_{12} > w_{22}$ (a simple additive or partially dominant case). Standard one-locus selection theory (built in Modeling Selection at a Single Locus) gives the allele-frequency change per generation:

Because both bracketed terms are positive under directional selection, $\Delta p > 0$ every generation: allele $A_1$ rises monotonically toward fixation (barring mutation-selection balance, drift, or a change in the environment), never oscillating around an internal equilibrium the way it would under balancing selection-type overdominance. This is the molecular-population-genetic signature that distinguishes directional selection from stabilizing selection (which typically maintains a stable interior allele-frequency equilibrium via heterozygote or intermediate-phenotype advantage).

Hermon Bumpus's sparrows (1898). After a severe winter storm in Providence, Bumpus collected 136 house sparrows (Passer domesticus) that had been knocked down, some dead, some merely stunned. He measured nine morphological traits, total length, wingspan, weight, and skeletal dimensions, on both survivors (72 birds that revived) and non-survivors (64 that died). Bumpus found that survivors clustered more tightly around the population mean for several traits, and that birds with extremely long wing bones or extreme body size were disproportionately represented among the dead. His original interpretation emphasized stabilizing selection (culling of extremes), but later re-analyses of his own published data, including reanalysis by Bumpus's contemporary and later statisticians, showed a more nuanced picture: several traits (like body length and sternum length) showed clear directional components, extreme large body size was selected against, consistent with directional selection against one tail rather than symmetric culling of both. Bumpus's dataset remains a textbook case precisely because it was one of the first to apply rigorous morphometric measurement to a natural episode of selection in the wild, and it established the method still used today, measure many individuals, compare survivors versus non-survivors statistically, and infer the shape of the fitness function.

Kettlewell's peppered moths (1950s). Bernard Kettlewell tested whether differential bird predation explained the rapid spread of the melanic (dark) form of the peppered moth (Biston betularia) in industrially polluted England. He performed mark-release-recapture experiments in polluted Birmingham and unpolluted Dorset woodlands, releasing known numbers of light and dark moths and measuring recapture rates, and he directly filmed birds preying preferentially on moths that contrasted with the bark background. In polluted, soot-blackened woodlands, dark moths were recaptured at substantially higher rates than light moths (in his Birmingham data, dark moth recapture rates were roughly twice those of light moths), the reverse held in unpolluted Dorset. This demonstrated directional selection favoring melanism specifically where soot-darkened bark made light moths conspicuous, and predicted (correctly) that melanic frequency would decline again as pollution controls reduced soot after the UK Clean Air Act (1956), a decline that was subsequently documented over the following decades, providing a rare case of a reversal in the direction of selection tracking a reversal in the environment.

The Grants and Darwin's finches on Daphne Major (1973–present). Peter and Rosemary Grant, with collaborators, have measured essentially every medium ground finch (Geospiza fortis) on the tiny Galápagos island of Daphne Major continuously since 1973, banding, measuring beak depth, body size, and tracking survival and reproduction of known individuals across generations. During the severe drought of 1977, small soft seeds were depleted and only large, hard seeds (like those of Tribulus) remained; birds needed deep, strong beaks to crack them. Population size crashed by roughly 85%, and survivors had significantly deeper beaks and larger body size than the pre-drought population, mean beak depth increased by about 4-5% in a single generation, a textbook directional selection event with a measured selection differential and a measured, heritable response (beak depth is highly heritable in this species, h² estimates around 0.7-0.9 from parent-offspring regression). Subsequent wet years and a second finch species' arrival on the island produced episodes of selection in the opposite direction (favoring smaller beaks), demonstrating that directional selection is not a fixed trend but a response to a shifting adaptive landscape, and confirming the breeder's equation quantitatively in a wild population, a rare feat matched by few other field studies.

Example 1: Quantitative genetics of the 1977 Daphne Major drought. Before the drought, mean beak depth in the G. fortis population was approximately 9.4 mm with a phenotypic standard deviation of about 0.90 mm. Among the survivors of the drought (roughly 15% of the population, selectively those able to crack large hard seeds), mean beak depth was approximately 9.96 mm. Selection differential: $S = 9.96 - 9.40 = 0.56$ mm. Using an estimated heritability of $h^2 \approx 0.79$ (from parent-offspring regression in this population):

So the predicted mean beak depth in the offspring generation is approximately $9.40 + 0.44 = 9.84$ mm. The Grants' actual field measurements of the next generation's mean beak depth came in close to this predicted value, one of the most direct field confirmations that the breeder's equation, developed originally for livestock breeding, correctly predicts evolutionary response in a wild vertebrate population under a single bout of strong, natural directional selection.

Example 2: Single-locus dynamics of insecticide resistance. Consider a mosquito population exposed to a new insecticide. A single locus confers resistance: genotype $RR$ has relative fitness 1.0 in the presence of the insecticide, $Rr$ has fitness 0.9 (partial dominance), and $rr$ (susceptible homozygote) has fitness 0.2 because most susceptible individuals die on contact with the insecticide. Suppose the resistance allele $R$ starts rare, at frequency $p_0 = 0.01$, with $q_0 = 0.99$. Mean fitness in generation 0:

Change in allele frequency (using $\Delta p = \dfrac{pq[p(w_{RR}-w_{Rr}) + q(w_{Rr}-w_{rr})]}{\bar w}$):

So after just one generation of spraying, $p$ jumps from 0.01 to roughly 0.042, more than quadrupling, because the fitness disadvantage of susceptibility is severe (0.2 versus 1.0) even though resistance started rare. Because $\Delta p$ remains positive at every intermediate allele frequency under this fitness scheme (directional selection with no interior equilibrium), repeated iteration shows $R$ sweeping toward fixation within roughly 10-15 generations of continuous spraying, a timescale of a few years for a mosquito with several generations per year, matching the real-world speed at which insecticide resistance often becomes a major public-health problem.

Hermon Carey Bumpus (1862-1943) was an American zoologist working at Brown University in the 1890s, a period when Darwinian natural selection was widely accepted in principle but had almost no direct, quantitative field demonstration. Bumpus's genius was methodological rather than theoretical: rather than speculating about selection, he treated a single catastrophic weather event as a natural experiment, with a clean "before" (the living population) and a rigorously measured "after" (dead versus surviving birds), and analyzed it using the statistical tools of his era (means, standard deviations, correlation) that were then novel imports from Karl Pearson's biometrical school. His 1898 paper on the Providence sparrows is now considered one of the founding empirical documents of the field that would become population biology, predating the Modern Synthesis by three decades and predating rigorous quantitative-genetic theory (Fisher, Wright, Haldane) by twenty to thirty years.

Directional selection is not a historical curiosity; it is actively reshaping organisms of direct human concern in real time. Antibiotic and antimalarial resistance evolve by exactly the single-locus directional mechanism worked out above: a drug imposes a steep fitness disadvantage on susceptible genotypes, and resistance alleles sweep through bacterial or Plasmodium populations within years, a central reason why clinical guidelines now stress rotating drug classes and combination therapy to slow the sweep. Pesticide and herbicide resistance in insects and weeds follows the identical logic and has produced billion-dollar agricultural losses, driving the design of resistance-management strategies such as refuge requirements for Bt-crop-resistant insects, which deliberately weaken directional selection by preserving susceptible alleles in untreated refuges. Artificial selection in agriculture and animal breeding is directional selection deliberately engineered, breeders directly compute (or approximate) $S$ and $h^2$ to design breeding programs for milk yield, growth rate, or disease resistance, and modern genomic selection uses dense marker data to predict breeding values and increase the effective accuracy underlying $h^2$. Conservation biology must reckon with human-caused directional selection, e size-selective fishing that consistently removes the largest individuals has driven measurable declines in mean body size and age at maturity in commercially harvested fish stocks (Atlantic cod being a well-documented case), an unintended, human-imposed directional selection regime with serious implications for fisheries sustainability. Climate change biology increasingly documents directional shifts in phenology and body size (e advancing breeding dates, shrinking body size in some endotherms) attributed to directional selection tracking warming trends, work that draws directly on the Grant-style framework of measuring selection differentials in wild populations. Finally, directional selection is the conceptual gateway to Sexual Selection, where directional selection on ornaments and armaments (peacock tails, elk antlers) can run away well past naturally-selected optima, precisely because the "environment" doing the selecting is the mating preferences or competitive tactics of conspecifics rather than the external ecology.

Ask the tutor: How does directional selection on a polygenic trait actually change allele frequencies at each of the many underlying loci, and why doesn't the trait run out of genetic variance the way a single-locus trait would?