Coevolution and the Red Queen

Coevolution is a change in a heritable trait of one population in response to a trait in another population, followed by an evolutionary response in that second population, driven specifically by the first. The key word is reciprocal: it is not enough that species A affects species B's fitness; B must affect A's fitness back, and each lineage's evolutionary trajectory must be shaped by the other's, generation after generation. This distinguishes true coevolution from simple sequential adaptation, where one species merely tracks a fixed feature of another without reciprocity (e.g., an insect evolving to detoxify a plant compound that itself is evolutionarily static). Coevolution is typically classified by the sign of the fitness interaction. In antagonistic coevolution — predators and prey, parasites and hosts, pathogens and immune systems — an improvement in one party (better toxin, better detection) directly reduces the fitness of the other, which then selects for counter-adaptation. This dynamic tends to produce evolutionary arms races: escalating, sometimes runaway, investment in offense and defense, analogous to a Cold War buildup where neither side gains a lasting advantage but both sides pay ever higher costs. In mutualistic coevolution — plants and pollinators, ants and acacias, corals and zooxanthellae, legumes and nitrogen-fixing rhizobia — improvements on each side increase the fitness of the partner, and coevolution can drive tight matching of morphology or biochemistry, since each partner's fitness improves by becoming a better match to the other's traits. A third category, diffuse coevolution, occurs when a trait is shaped not by one partner species but by an entire guild of interacting species simultaneously (a flower coevolving with a community of pollinators, none of which alone controls the outcome), which is probably closer to the norm in real communities than tight pairwise coevolution. Overlaying all of this is the Red Queen hypothesis, proposed by Leigh Van Valen in 1973 and named for the Red Queen's remark to Alice in Through the Looking-Glass: "it takes all the running you can do, to keep in the same place." The claim has two related but distinct forms. The macroevolutionary (Van Valen) version says that because a species' physical environment is relatively stable over geological time but its biotic environment (competitors, predators, parasites, prey) is constantly evolving, extinction risk does not decrease with a lineage's age or accumulated adaptation — species must continually adapt merely to hold their ecological position steady. The microevolutionary version, developed later mainly around the evolution of sex, says that fast-generation antagonists (especially parasites) evolve so quickly relative to their hosts that a host's genotype which is common and successful this generation becomes a preferred target next generation, producing negative frequency-dependent selection that favors rare genotype combinations — which sexual recombination is unusually good at generating.

Step 1 — Genetic variation exists on both sides. In any host-parasite or predator-prey pair, both populations carry standing genetic variation for the interacting traits: toxin dose in a newt, sodium-channel structure in a snake; a plant's chemical defense, an herbivore's detoxifying enzyme; a bacterium's surface receptor, a phage's tail-fiber protein that binds that receptor.

Step 2 — Selection acts reciprocally each generation. A newt with slightly more toxin survives predation better; a snake with slightly more resistant sodium channels survives poisoning better and can still exploit the food resource. Selection on the snake is driven by the newt's phenotype distribution, and selection on the newt is driven by the snake's. Neither selective pressure is fixed — it shifts as the other population's mean phenotype shifts.

Step 3 — Genotype matching creates frequency-dependence. In many host-parasite systems (this is the clearest in gene-for-gene systems in plant pathology and matching-allele systems in invertebrates), infection success depends on a specific match between host and parasite genotype — a parasite adapted to attack the common host genotype does very well until that genotype becomes rare, at which point a previously rare host genotype (against which the parasite is poorly adapted) becomes favored. This generates negative frequency-dependent selection: whichever genotype is common is disadvantaged, whichever is rare is favored. This is the same logic that maintains genetic polymorphism in self-incompatibility alleles and in some predator search-image systems.

Step 4 — Cycling, not equilibrium. Because parasites typically have shorter generation times and larger population sizes, they can track and adapt to the currently common host genotype faster than the host can adapt to a new stable strategy. The result is not convergence to a single optimal genotype on either side but oscillating cycles in genotype frequencies over time — host genotype A rises, parasites adapted to A rise in response and drive A down, host genotype B (previously rare) then rises, parasites track B, and so on. This is directly analogous to (and can be modeled with) the same negative-frequency-dependent logic that maintains balanced polymorphisms.

Step 5 — Escalation in antagonistic systems. Where the interaction is a quantitative arms race rather than discrete-genotype matching (toxin dose vs. channel resistance, running speed vs. running speed, camouflage vs. visual acuity), each round of reciprocal selection can push both parties' trait values higher, sometimes to biologically extreme levels — newt toxin loads and snake resistance thresholds in some Oregon populations now sit far beyond what would be needed to deter or survive an "average" ancestral opponent, precisely because each has been chasing the other's escalating trait for many generations.

Step 6 — Costs create limits. Escalation is not unbounded. Toxin production and resistance both carry physiological costs (resistant snakes are measurably slower — a real fitness trade-off against non-newt-eating predators and prey capture generally), so the arms race proceeds only as far as marginal benefit exceeds marginal cost, producing a moving but bounded equilibrium rather than infinite escalation.

Step 7 — Sex as a Red Queen weapon. In the sexual-reproduction context, coevolution with fast-evolving parasites is one of the leading explanations for why costly sexual reproduction (which halves a female's genetic contribution to offspring relative to asexual reproduction — the two-fold cost of sex) persists. Recombination shuffles host genotypes every generation, constantly generating novel allele combinations that a parasite population, adapted to the previous generation's common genotypes, has not yet encountered — sex effectively outruns parasite tracking. Asexual clonal lineages, by contrast, present a fixed target that parasites can adapt to and then exploit indefinitely, which is why some of the best empirical support for this idea comes from mixed sexual/asexual populations of the same species living side by side.

Van Valen's fossil-record test (1973). Leigh Van Valen assembled survivorship curves for thousands of fossil taxa across many higher groups — bivalves, foraminifera, mammals, and others — plotting the proportion of genera or families surviving against their age since origination. Under a "static environment" model, one might expect older, more established taxa to have progressively lower extinction probability as they accumulate adaptations to a relatively fixed physical world (an improving fit, like a machine being debugged). Instead, Van Valen found that for most groups, survivorship curves were closer to exponential — meaning the probability of extinction per unit time was roughly constant regardless of a taxon's age. A genus that had already survived 50 million years was, on this measure, not much safer from extinction than a newly arisen genus. This was startling: it implied that adaptation does not accumulate into a permanently improved, low-risk state. Van Valen's explanation was that each species' biotic environment (competitors, predators, parasites) is itself continually evolving, so any fitness advantage gained by adaptation is continually eroded by counter-adaptation in interacting lineages — extinction risk stays flat because everyone is running, and nobody gets ahead for long. This is the origin of the "Red Queen's Hypothesis" name and remains the primary macroevolutionary evidence, though later reanalyses have shown the pattern is not universal — some clades do show declining extinction risk with age, so the strong constant-rate claim is a generalization with real exceptions, not an iron law.

The newt–garter snake arms race (Brodie & Brodie, ongoing since the 1990s). Edmund Brodie Jr. and Edmund Brodie III conducted extensive field and laboratory work on rough-skinned newts (Taricha granulosa), which sequester extraordinarily high concentrations of tetrodotoxin (TTX) — the same neurotoxin found in pufferfish — in their skin, and their main predator, the common garter snake (Thamnophis sirtalis). They measured toxin levels in newts and, separately, resistance (measured as how much TTX a snake could tolerate before crawling performance was impaired) in snakes from dozens of populations across Oregon, Washington, and California. The key finding: newt toxicity and snake resistance are tightly geographically matched — populations with the most toxic newts have the most resistant snakes, and this covaries across sites in a way inconsistent with either trait evolving independently. In some populations, resistant snakes can tolerate TTX doses that would be lethal many times over to an unadapted vertebrate, and correspondingly toxic newts carry doses far in excess of what would be needed to kill a non-resistant predator — direct evidence of escalation beyond any single "sufficient" defense level, the signature of a runaway arms race. Later work identified the molecular mechanism: resistant snake populations carry amino acid substitutions in the voltage-gated sodium channel gene (the TTX binding site) that reduce toxin binding, and resistant snakes pay a measurable cost — reduced crawling speed even in the absence of toxin — confirming the trade-off predicted by the model.

Greya moths and Lithophragma woodland stars (Thompson & Pellmyr, and later work). John Thompson's long-term studies of pollinating seed-parasite moths (Greya) and their host plants demonstrated geographic mosaic coevolution: the intensity and even the sign of the interaction (mutualistic pollination vs. purely parasitic seed predation) varies across populations, showing that coevolution is not a single fixed outcome but a shifting patchwork shaped by local ecological context — a major refinement of simple pairwise arms-race models.

Darwin's orchid prediction and its confirmation. Charles Darwin, examining the Madagascar star orchid Angraecum sesquipedale with its extraordinary 27–43 cm nectar spur, predicted in 1862 that a moth with a proboscis long enough to reach the nectar at the spur's base must exist, even though no such moth was then known. This is a coevolutionary prediction: pollinator tongue length and flower spur length should coevolve, because a slightly longer spur favors moths with slightly longer tongues that can still reach the nectar reward (better pollination for the plant), and slightly longer tongues favor moths that can access nectar inaccessible to shorter-tongued competitors. In 1903, Xanthopan morganii praedicta — a hawkmoth subspecies with a proboscis of matching length — was described from Madagascar, and direct observation of the moth pollinating the orchid was finally filmed in 1992 (Arnold, Wasserthal), confirming both the coevolutionary mechanism and Darwin's specific prediction. Comparative work across Angraecum species and their moth pollinators shows a general correlation between spur length and moth tongue length, consistent with ongoing reciprocal selection rather than a one-off coincidence.

Experimental coevolution of bacteria and phage. Because bacteria and bacteriophage have short generation times, coevolution can be watched happening in real time in the lab. Paul Rainey, Angus Buckling, and colleagues propagated Pseudomonas fluorescens together with its lytic phage SBW25φ2 for hundreds of generations. They found that bacterial resistance to phage infection and phage infectivity both increased over time when the two coevolved together, but populations exposed only to a fixed, non-evolving phage stock evolved resistance to a lesser degree and plateaued — direct experimental confirmation that ongoing reciprocal evolution, not one-sided adaptation to a static enemy, drives the escalation. Similar experiments (e.g., with E. coli and phage T7 or λ) have shown local adaptation: when bacteria from a given time point are tested against phage from the same time point versus phage from the past or future, contemporary phage populations are generally most infective against contemporary hosts, and hosts are best defended against past parasites, exactly as predicted by a lagging coevolutionary chase.

Example 1 — quantitative arms race in numbers. Suppose a garter snake population's baseline (non-resistant) LD50 for TTX corresponds to a serum toxin threshold of 1 unit, and newts in that valley currently carry a mean skin toxin load of 3 units — enough to kill an unadapted snake three times over. A resistance allele arises that raises a snake's tolerance to 4 units, at a fitness cost from slower sprint speed equivalent to a 5% reduction in prey-capture success. If the resistance allele's survival benefit (surviving newt predation attempts, which would otherwise be fatal roughly 60% of the time at toxin load 3) exceeds its 5% cost, it spreads. As resistant snakes become common, newts with toxin loads below 4 units suffer no extra predation cost relative to more toxic conspecifics — so selection for still higher toxin (say, to 6 units) resumes only if some new, more toxin-sensitive-but-otherwise-fitter snake genotype is common, or if the marginal predation-deterrence benefit of extra toxin still exceeds its metabolic cost. Field data broadly fit this logic: across the Brodies' Oregon transects, the most toxic newt populations (loads sufficient to kill numerous mice per newt) co-occur specifically with the most resistant snake populations, and in a few "toxin mismatch" populations where geographic barriers have prevented full co-adaptation, snakes are far more vulnerable, exactly as the escalating-and-locally-tracking model predicts.

Example 2 — a Red Queen cross with asymmetric host resistance alleles. Consider a simplified matching-allele host-parasite system in an invertebrate (modeled loosely on Daphnia–microparasite systems studied by Decaestecker, Ebert, and colleagues). Suppose host resistance is controlled by a single locus with two alleles, R (resistant to parasite genotype 1, currently at frequency 0.75 in the parasite population) and S (resistant to parasite genotype 2, at frequency 0.25). At generation t, host allele R is common (0.8) because it has recently been favored; parasite genotype 1 (which specifically overcomes R) is rare (0.2) because it only recently arose by mutation/recombination and hasn't yet had time to spread. Selection each generation favors: (a) in the host, whichever allele the current parasite population is least equipped to defeat, and (b) in the parasite, whichever genotype best exploits the current most common host allele. Working forward: because R is common, any parasite lineage able to infect R-hosts (parasite genotype 1) has a large host pool to exploit and increases rapidly, say from 0.2 to 0.6 over several generations. As parasite genotype 1 becomes common, R-hosts start dying at higher rates, so allele S — now advantaged simply because it's rare and mismatched to the currently common parasite genotype — rises from 0.2 to 0.6. This in turn favors parasite genotype 2 (matching S), which rises, which favors R again — a genuine oscillation with no stable equilibrium frequency, exactly the qualitative cycling documented in the resurrected "Red Queen from the past" experiments where researchers revived dormant Daphnia eggs and their co-buried parasites from sediment layers decades apart and found that hosts were consistently best defended against parasites from their own time layer and worse defended against both past and future parasites — a direct experimental time-machine confirmation of ongoing reciprocal tracking.

Leigh Van Valen (1935–2010) was an American evolutionary biologist at the University of Chicago known for combining rigorous quantitative paleobiology with a willingness to propose bold, falsifiable hypotheses outside prevailing consensus. Frustrated that his 1973 manuscript proposing the Red Queen's Hypothesis was rejected by mainstream journals, he founded his own journal, Evolutionary Theory, specifically to publish it — a decision emblematic of his combative, iconoclastic style. He worked across an unusually wide range of topics, from mammalian paleontology and extinction statistics to the definition of species and the ecology of niche width, and he was notorious among colleagues for dense, aphoristic writing and a refusal to simplify ideas for the sake of easy publication.

Antibiotic and antiviral resistance is applied host-pathogen coevolution in real time: bacteria evolve resistance mechanisms against antibiotics (efflux pumps, target-site mutations), and clinical strategy increasingly treats this as an arms race requiring rotation of drug classes, combination therapy, and monitoring of resistance evolution rather than assuming any single drug will remain effective indefinitely. Cancer treatment now explicitly borrows Red Queen and evolutionary-arms-race framing: tumor cell populations under chemotherapy or targeted-therapy selection evolve resistant subclones, and "adaptive therapy" protocols deliberately withhold maximal dosing to preserve drug-sensitive cell populations that compete with and suppress resistant clones, slowing the coevolutionary escalation. Agricultural crop breeding against rapidly evolving plant pathogens (wheat rust, potato blight) uses gene-for-gene resistance breeding informed directly by Red Queen logic — a resistance gene deployed uniformly across millions of acres creates strong selection for pathogen genotypes that defeat it, so breeders rotate and stack resistance genes to slow pathogen tracking, echoing the same dynamic documented in Daphnia-parasite systems. Vaccine design against fast-evolving viruses like influenza and HIV must anticipate antigenic evolution — the virus's antigenic sites evolve specifically to evade host antibody recognition, a direct analogue of the matching-genotype cycling described above, which is why influenza vaccines are reformulated annually based on surveillance of circulating strains. Conservation biology uses coevolutionary thinking when reintroducing or translocating species: a host population isolated from its historical parasite community, or a prey population reintroduced without its historical predators, may have lost defenses through relaxed selection, making coevolutionary history essential to predicting reintroduction outcomes. Biological control programs deliberately exploit antagonistic coevolution, introducing natural enemies of invasive species, while monitoring for the risk that the target pest may evolve resistance to the control agent, another live application of arms-race theory.

Ask the tutor: How does the geographic mosaic theory of coevolution change the classic single-arms-race picture, and what evidence supports it?