Coevolutionary arms races and the Red Queen hypothesis

Intuition Beginner

An arms race in biology is a cycle of retaliation between two species. A predator evolves sharper claws, so prey evolve thicker armor; the prey's armor then selects for an even sharper claw, which selects for even thicker armor, and so on. Each adaptation is valuable only relative to the other side's current state. The same loop runs between hosts and parasites, plants and herbivores, and between rivals. This is reciprocal selection: each species is part of the other's selective environment, and neither ever reaches a fixed best form.

Parasites and pathogens usually evolve faster than their hosts because they are smaller, far more numerous, and reproduce in days rather than years. A host that stands still is quickly overtaken. Leigh Van Valen captured this in 1973 with the Red Queen hypothesis, named for the character in Lewis Carroll's Through the Looking-Glass who tells Alice, "it takes all the running you can do, to keep in the same place." Species must keep evolving merely to avoid falling behind their ever-adapting enemies. This constant pressure is a leading explanation for why most organisms reproduce sexually, shuffling genes to present parasites with a moving target.

Arms races are not always hostile. In mutualistic arms races, two partners each evolve to serve the other better: cleaner fish and their client fish, or plant roots and the soil fungi that trade minerals for sugars. And arms races are not uniform across the map. John Thompson's geographic mosaic theory shows that reciprocal selection is fierce in some populations and absent in others, with gene flow stitching the patchwork together. The takeaway: evolution's target moves because the other species moves with it.

Visual Beginner

Arms races come in antagonistic and mutualistic forms, and each leaves a distinctive signature of trait change.

Arms-race type	Interaction	Signature example
Predator-prey	Antagonistic	Cheetah-gazelle speed
Host-parasite (Red Queen)	Antagonistic	Newt tetrodotoxin vs. snake resistance
Plant-herbivore	Antagonistic	Milkweed toxins vs. monarch butterflies
Competitive	Antagonistic	Finch beak divergence
Mutualistic escalation	Mutualistic	Cleaner wrasse and client reef fish
Mutualistic escalation	Mutualistic	Mycorrhizal fungi and plant roots

The Red Queen makes allele frequencies cycle rather than settle:

Red Queen allele-frequency cycles (matching-alleles model)

  frequency
     1.0 |  *           *           *
         |   *         * *         * *
 host A  |    *       *   *       *   *
         |  p  *     *     *     *
     0.5 |------*---*-------*---*--------  0.5
         |        *   *       *   *
 parasite|         * *         * *         q
     0.0 |          *           *
         +-------------------------------- generations
             host flees <- parasite chases <- host flees

When a host allele peaks (p high), the matching parasite surges (q rises), which drives the host allele down, which then starves the parasite allele, letting the host allele rebound. Neither side wins; both just keep running.

Worked example Beginner

The Red Queen is hardest to see but easiest to feel where sex is concerned, because sex looks like a bad deal. Consider two populations of freshwater snails that differ only in how they reproduce. Asexual females clone themselves, each producing 2 daughters. Sexual females each produce 2 young, but only half are daughters (the rest are sons), so each sexual female replaces herself with 1 daughter.

Start with 100 females of each kind.

Step 1 (generation 1). Asexual: 100 times 2 = 200 females. Sexual: 100 times 1 = 100 females.

Step 2 (generation 2). Asexual: 200 times 2 = 400. Sexual: 100 times 1 = 100.

Step 3 (generation 3). Asexual: 400 times 2 = 800. Sexual: 100 times 1 = 100.

After three generations the asexuals outnumber sexuals 8 to 1, with no deaths at all. This is the twofold cost of sex: an asexual lineage doubles its share every generation.

What this tells us: for sex to persist, something powerful must erase this cost. The Red Queen answer is that coevolving parasites adapt to whichever clone is common, hammering the abundant asexuals until rare sexual genotypes — which parasites have not yet cracked — win out instead. Sex pays its twofold toll because it buys a moving genetic target.

Check your understanding Beginner

Formal definition Intermediate+

A coevolutionary arms race is reciprocal, directional evolutionary change between two or more interacting lineages, in which an adaptation in one lineage selects for a counter-adaptation in the other, which selects for a further adaptation in the first, with no terminal equilibrium reached. If $W_A$ denotes the mean fitness of lineage $A$ and ${\mathbf{t}}_B$ the trait distribution of its interactor $B$, the arms-race condition is that a change in ${\mathbf{t}}_B$ which raises $W_B$ lowers $W_A$, provoking a compensatory shift in ${\mathbf{t}}_A$ that in turn lowers $W_B$ again — each "improvement" devalues the other side.

The Red Queen hypothesis ^{[Van Valen 1973]} is the claim that, over long times, the biotic environment is a more important driver of continual evolution than the abiotic environment, precisely because the biotic environment is itself evolving. Van Valen grounded this in the law of constant extinction: within a taxon, the per-lineage probability of extinction is roughly independent of how long the lineage has already existed. A constant hazard is what one expects if each lineage races a continually renewed field of competitors, predators, and pathogens, rather than a fixed physical backdrop that lineages eventually become matched to.

Three genetic architectures organise arms races. In the matching-alleles model, exploitation requires an exact allele match at all relevant loci, producing symmetric negative frequency-dependent cycles. In the gene-for-gene model ^{[Flor 1956]}, a dominant host resistance allele $R$ defeats a pathogen carrying the corresponding avirulence allele $Avr$, so the dynamics are directional and typically settle to a resistance-virulence equilibrium rather than neutral cycles. In quantitative-trait models, arms races are driven by polygenic features (body size, running speed, toxin titre) under stabilising selection around a moving optimum set by the interactor.

Counterexamples to common slips

• Adaptation to a static partner is not an arms race. If only one lineage evolves while the other is fixed, the interaction is sequential adaptation, not reciprocal coevolution. Janzen's criteria of specificity, reciprocity, and simultaneity must all hold.
• A single selective sweep is not an arms race. An arms race requires sustained reciprocal change; a one-off adaptation to a new host is ordinary adaptation unless the host responds in turn.
• Cost-free escalation does not exist. Every counter-adaptation carries a metabolic, performance, or developmental cost. Arms races are bounded by these trade-offs, which is why they stall, settle into cycles, or terminate in standoffs rather than running forever.

Core model Intermediate+

The signature quantitative object of Red Queen theory is the matching-alleles model (MAM), in which infection requires the parasite's allele to match the host's allele at a resistance locus ^{[Flor 1956]}. Take one host locus with allele $A$ at frequency $p$ (allele $a$ at $1-p$) and one parasite locus with allele $B$ at frequency $q$ (allele $b$ at $1-q$), where $B$ infects hosts carrying $A$ and $b$ infects hosts carrying $a$. Selection on the host drives the matched allele down when its parasite is common; selection on the parasite drives the matched allele up when its host is common. In continuous time with a single selection coefficient $s$:

$$\dot{p}=sp(1-p)(1-2q),\dot{q}=-sq(1-q)(1-2p).$$

Reading the equations. When $q$ is large (parasite allele $B$ common), the factor $(1-2q)$ is negative, so $\dot{p}<0$: the host allele that $B$ exploits declines. When $p$ is large (host allele $A$ common), $(1-2p)<0$, so $\dot{q}>0$: the parasite allele that exploits $A$ rises. The host flees and the parasite pursues. The sole interior equilibrium is $p=q=1/2$.

Negative frequency-dependent selection. Linearising about $(1/2,1/2)$ with $u=p-1/2$, $v=q-1/2$ gives $\dot{u}=-(s/2)v$ and $\dot{v}=(s/2)u$, whence $\ddot{u}=-(s^2/4)u$: allele frequencies oscillate as a harmonic oscillator of angular frequency $s/2$. A host allele that is common this generation is, by the next, the parasite's prime target and so becomes rare; rarity then grants refuge, and the allele rebounds. Polymorphism is maintained indefinitely, because there is no fixed optimum to converge on.

Empirical signature. Time-shift experiments on the Daphnia longispina-Pasteuria ramosa system ^{[Lively 1987]} recover exactly this oscillation: parasites adapt to the host clones that were common in previous generations, so that past-common clones now suffer the heaviest infection, and currently rare clones gain the advantage. Negative frequency-dependent selection is the model's central prediction, and it is the prediction that field data confirm.

Bridge. The matching-alleles oscillator is the foundational reason the Red Queen has real dynamical teeth: reciprocal selection converts a stable optimum into a limit geometry of cycles, and this is exactly the structure that builds toward 18.10.01 antigenic diversity and appears again in 19.13.01 as the engine behind the maintenance of sex. The central insight — that fitness is frequency-dependent because the opposing species tracks you — generalises from a single locus to the genome-wide arms races of the geographic mosaic treated below, and the bridge is between single-population genetics and the spatially structured coevolution that dominates real landscapes.

Exercises Intermediate+

Advanced results Master

Thompson's geographic mosaic theory. The matching-alleles oscillator describes a single panmictic population, but real species are spread across heterogeneous landscapes. Thompson's geographic mosaic theory of coevolution ^{[Thompson 2005]} decomposes the spatial problem into three coupled processes. First, selection mosaics: the strength and even the sign of reciprocal selection vary across populations because community context varies — a given parasite may be present in some lakes and absent in others, or a third interactor may reverse the selective pressure. Second, coevolutionary hotspots and coldspots: only some populations harbour strong reciprocal selection (hotspots), while in others one or both interactors are missing, or selection is unidirectional (coldspots). Third, trait remixing: gene flow, genetic drift, and local extinction-colonisation dynamics continually shuffle coevolved traits among populations, so that a trait favoured in one hotspot may be exported into a coldspot where it is neutral or deleterious. The mosaic predicts that local maladaptation should be common rather than exceptional, and empirical surveys of the Greya-Lithophragma and Daphnia-Pasteuria systems confirm it: across a species' range, parasites are often better adapted to sympatric hosts in some populations and poorly adapted in others.

Diffuse coevolution. Most pairwise models are an idealisation; in nature a species typically interacts with a guild of partners simultaneously. Diffuse (guild) coevolution is the reciprocal evolution driven by the aggregate selective pressure of many interactors. A plant's defensive chemistry is shaped not by a single herbivore but by the whole herbivore community, and each herbivore's counter-adaptation is shaped by the combined chemistry of many host plants. Diffuse coevolution weakens the tight one-to-one matching of pairwise models: selection coefficients are averaged over partners, pairwise cycles are damped, and the system's behaviour is governed by network-level statistics — connectance, nestedness, and modularity — rather than by a single interaction's dynamics.

The Red King effect in mutualisms. Bergstrom and Lachmann ^{[Bergstrom 2003]} showed that the pacing logic of the Red Queen inverts when the interaction is mutualistic rather than antagonistic. In a coordination game, both partners gain from aligning their traits, so the partner that evolves more slowly cannot profitably be abandoned; the faster-evolving partner is compelled to adjust toward it, and the slower partner therefore fixes the equilibrium terms. The Red King effect explains why slowly evolving mutualists — long-lived trees, for instance — can impose stable, favourable interaction conditions on their fast, short-lived microbial partners. It also warns that the "fast parasite wins" intuition from antagonistic arms races does not transfer to mutualism, where speed confers compliance rather than dominance.

Mutualistic escalation. Arms races can escalate cooperation rather than weaponry. Cleaner fish and their client reef fish coevolve increasingly fine-tuned signals and responses: cleaners evolve colour patterns and ritualised "dance" displays that advertise their service and suppress client flight, while clients evolve the ability to recognise honest cleaners and to punish or avoid cheaters. Mycorrhizal fungi and plant roots coevolve ever more elaborate exchange interfaces, with plants allocating carbon to the most generous fungal partners and fungi delivering more phosphorus to the best-supplied roots. These are arms races in the strict sense — reciprocal, directional, no terminal equilibrium — yet the escalation is in the machinery of cooperation, policed by partner choice and sanctions that prevent either side from drifting toward exploitation.

Arms-race limits: the life-dinner principle and trade-offs. Arms races do not escalate without bound. The life-dinner principle captures an asymmetry that brakes many predator-prey races: the prey runs for its life while the predator runs only for its dinner, so selection on the prey is far stronger than selection on the predator, and the race reaches a standoff at the prey's escape threshold. More generally, every counter-adaptation carries a cost — tetrodotoxin-resistant snake sodium channels impair locomotion, thicker shells slow growth, exaggerated ornaments burden their bearer. When the marginal benefit of further escalation falls below its marginal cost, the race stalls in a cycle, a standoff, or a shift to a new axis of attack. These trade-offs are why measured arms races in the wild show bounded rather than runaway trait values.

Escalation in the fossil record. Vermeij's escalation hypothesis ^{[Vermeij 1987]} generalises the arms race to geological time: the long-term history of life shows a directional increase in the frequency and intensity of predation, driving corresponding increases in defensive elaboration. Drilling predation by gastropods on bivalves rose steadily through the Cenozoic; the proportion of heavily armoured gastropods increased while thin-shelled forms declined; repair scars from failed attacks became more common. Unlike a pairwise oscillation, escalation is a community-level, directional trend sustained over tens of millions of years, and it identifies predation-driven coevolution as a dominant macroevolutionary force.

The Red Queen and the maintenance of sex. The most consequential application of arms-race theory is to the twofold cost of sex. Hamilton ^{[Hamilton 1980]} argued that parasites, evolving fast and tracking common host genotypes, supply exactly the negative frequency-dependent selection needed to favour rare sexual recombinants over abundant asexual clones. Field tests on the New Zealand snail Potamopyrgus antipodarum ^{[Lively 1987]} confirm the prediction: sexual individuals are more frequent in populations with high trematode prevalence, and the most common asexual clones are disproportionately infected. The Red Queen thereby links the population genetics of matching-alleles oscillators to one of the broadest patterns in eukaryotic life — the ubiquity of sex.

Synthesis. The Red Queen is the foundational reason that adaptation in coevolving systems is geometric rather than asymptotic: the matching-alleles oscillator is a centre, not a sink, and this is exactly why lineages never settle to a fixed optimum. The central insight — that each species' optimum moves because its antagonist moves — generalises from a single locus to the geographic mosaic, where the bridge is between deterministic local cycles and the stochastic patchwork of hotspots and coldspots stitched together by gene flow. Putting these together, the pattern recurs across scales: allele cycles within populations, population mismatches across landscapes, diffuse coevolution across guilds, and community-wide escalation across geological time, and it appears again in 18.10.01 as the pathogen-driven polymorphism of the major histocompatibility complex.

Full proof set Master

Proposition (neutral cycles of the symmetric matching-alleles arms race). In the continuous-time matching-alleles model $\dot{p}=sp(1-p)(1-2q)$, $\dot{q}=-sq(1-q)(1-2p)$ with $s>0$, every trajectory beginning in the interior $(0,1{)}^2$ is periodic, and the interior equilibrium $(p,q)=(\frac{1}{2},\frac{1}{2})$ is a centre.

Proof. The boundary edges $p=0$, $p=1$, $q=0$, $q=1$ are invariant, since each vanishes the corresponding time-derivative; an interior initial condition therefore never reaches the boundary in finite time. The sole interior fixed point solves $\dot{p}=\dot{q}=0$ with $p,q\in (0,1)$, namely $(1/2,1/2)$. Compute the divergence of the vector field $\mathbf{F}=(\dot{p},\dot{q})$:

$$\nabla \cdot \mathbf{F}=\frac{\partial \dot{p}}{\partial p}+\frac{\partial \dot{q}}{\partial q}=s(1-2p)(1-2q)-s(1-2q)(1-2p)=0,$$

so the flow is area-preserving (Liouville). A source or sink would contract or expand area, which is impossible; a limit cycle with non-unit Poincaré return map would likewise violate area preservation. Linearising at $(1/2,1/2)$ yields the Jacobian with eigenvalues $\pm is/2$, a centre. With the divergence-free flow excluding spirals and the interior containing no further fixed point, the Poincare-Bendixson theorem forces the $\omega$-limit of every interior orbit to be a periodic orbit. All interior trajectories are therefore closed. $\square$

Corollary (asymmetric selection breaks neutrality). If host and parasite selection coefficients differ, $s_H\ne s_P$, the divergence becomes $(1-2p)(1-2q)(s_H-s_P)$, which changes sign across the four quadrants of the unit square. The flow is no longer area-preserving: orbits acquire a per-cycle drift, the centre becomes a weak spiral, and sustained oscillations decay or amplify until bent back by the nonlinear boundary of allele-frequency space. This is why real arms races — with unequal selection, finite population size, and recurrent mutation — display transient cycles rather than the perpetual neutral ringing of the symmetric idealisation.

Connections Master

• Coevolution 19.13.01. This unit is the deep dive into the arms-race and Red Queen slice of the broader coevolution framework introduced there. The matching-alleles oscillator proved here supplies the dynamical machinery that the overview unit invoked qualitatively; the geographic mosaic and diffuse-coevolution results refine the overview's community-level claims into testable, spatially explicit predictions.

• Natural selection 19.03.01. Arms races are frequency-dependent natural selection applied to the case where the selective environment is itself evolving. The selection coefficients $s_H$ and $s_P$ in the matching-alleles model are exactly the selection coefficients of 19.03.01, now coupled across two species, and negative frequency-dependent selection is the mechanism that maintains the polymorphisms discussed in the selection unit.

• Immunology 18.10.01. The extraordinary polymorphism of the major histocompatibility complex is a direct molecular manifestation of Red Queen dynamics: pathogens evolve to evade the most common MHC alleles, which favours hosts carrying rare alleles, maintaining diversity through the same negative frequency-dependent mechanism derived here. Adaptive immunity's somatic recombination is the within-generation analogue of the between-generation recombination that the Red Queen invokes to justify sex.

• Macroevolution 19.08.01. Vermeij's escalation hypothesis reframes arms races as a directional macroevolutionary trend: the long-term rise in predation intensity drives the geological increase in defensive elaboration documented across the Cenozoic fossil record. Escape-and-radiate dynamics, in which a key innovation briefly frees a lineage from its antagonists and fuels a radiation, connect microevolutionary arms races to the macroevolutionary diversification patterns of 19.08.01.

• Community ecology 19.10.01. Diffuse coevolution and the nested, modular structure of plant-pollinator and host-parasite networks place arms races inside the broader architecture of ecological communities studied in 19.10.01. The network statistics — connectance, nestedness, modularity — that govern community stability are themselves products of coevolutionary history, so the community-level patterns of 19.10.01 cannot be fully understood without the coevolutionary dynamics formalised here.

Historical & philosophical context Master

Leigh Van Valen introduced the Red Queen hypothesis in 1973 ^{[Van Valen 1973]} as the interpretation of a striking empirical regularity he had uncovered in the fossil record: the probability of extinction of a lineage within a taxon is approximately independent of the lineage's duration. He named this the law of constant extinction and argued that it is most naturally explained if the effective environment of any lineage is the set of other lineages with which it coevolves, since that environment renews itself continually rather than holding still to be adapted to. The literary allusion to Lewis Carroll's Red Queen gave the hypothesis its enduring name and captured its core claim — that continued evolutionary change is the price of holding one's fitness — in a single image. Van Valen's proposal initially met resistance because the law of constant extinction rested on then-controversial compilations of fossil ranges, but subsequent macroevolutionary analyses broadly upheld the regularity.

The hypothesis was transformed from a macroevolutionary observation into a population-genetic mechanism by William Hamilton, who in 1980 ^{[Hamilton 1980]} argued that host-parasite arms races supply the negative frequency-dependent selection needed to overcome the twofold cost of sex, making the Red Queen the leading explanation for the ubiquity of sexual reproduction. John Thompson's geographic mosaic theory ^{[Thompson 2005]} then generalised the framework spatially, showing that reciprocal selection is patchily distributed across species' ranges and that gene flow among patches predicts widespread local maladaptation. Bergstrom and Lachmann's Red King effect ^{[Bergstrom 2003]} completed the pacing logic by showing that mutualisms invert the Red Queen's "fastest wins" rule: in coordinated interactions the slowest partner sets the terms. Together these developments turned arms-race theory from a vivid metaphor into a quantitative, spatially structured, and empirically tested body of evolutionary biology.

Bibliography Master

1. Van Valen, L. "A new evolutionary law." Evolutionary Theory 1 (1973) 1-30.

2. Hamilton, W. D. "Sex versus non-sex versus parasite." Oikos 35 (1980) 282-290.

3. Flor, H. H. "The complementary genic systems in flax and flax rust." Advances in Genetics 8 (1956) 29-54.

4. Thompson, J. N. The Geographic Mosaic of Coevolution (University of Chicago Press, 2005).

5. Bergstrom, C. T. & Lachmann, M. "The Red King effect: when the slowest runner wins the coevolutionary race." Proceedings of the National Academy of Sciences 100 (2003) 593-598.

6. Lively, C. M. "Evidence from a New Zealand snail for the maintenance of sex by parasitism." Nature 328 (1987) 519-521.

7. Vermeij, G. J. Evolution and Escalation: An Ecological History of Life (Princeton University Press, 1987).

8. Futuyma, D. J. & Kirkpatrick, M. Evolution, 4th ed. (Sinauer, 2017), Ch. 18.

9. Bergstrom, C. T. & Dugatkin, L. A. Evolution, 2nd ed. (W. W. Norton, 2016).

10. Campbell, N. A. & Reece, J. B. Biology, 12th ed. (Pearson, 2020), Ch. 23 and 54.

11. Janzen, D. H. "When is it coevolution?" Evolution 34 (1980) 611-612.

12. Ehrlich, P. R. & Raven, P. H. "Butterflies and plants: a study in coevolution." Evolution 18 (1964) 586-608.

13. Dawkins, R. & Krebs, J. R. "Arms races between and within species." Proceedings of the Royal Society of London B 205 (1979) 489-511.