Evolution of virulence

Virulence—the degree of harm a pathogen inflicts on its host—is one of the central variables in the evolutionary ecology of infectious disease. Why do some pathogens cause devastating illness and rapid death while others coexist with their hosts for decades in a state of chronic but tolerable infection? For much of the twentieth century, the prevailing assumption was that pathogens inevitably evolve toward avirulence, on the logic that a parasite that kills its host destroys its own habitat. This intuition, sometimes called the avirulence hypothesis, relied implicitly on group selection arguments and was challenged in the early 1980s by a new framework grounded in individual-level natural selection.^{1, 2} The modern theory of virulence evolution, developed primarily by Roy Anderson, Robert May, and Paul Ewald, holds that virulence is not destined to decline to zero but instead evolves toward a level determined by the ecological relationship between pathogen replication, transmission, and host survival.^{1, 3}

The evolution of virulence intersects with nearly every major theme in evolutionary biology: coevolution between hosts and parasites, arms race dynamics, horizontal gene transfer, antibiotic resistance, and the emergence of novel zoonotic diseases. Understanding why pathogens evolve to be more or less harmful is not merely an academic exercise; it has direct consequences for vaccination strategy, public health intervention, and the management of host-parasite relationships in agriculture and medicine. This article traces the theoretical foundations and empirical evidence for the evolution of virulence, from the classic trade-off hypothesis to the complications introduced by within-host dynamics, mixed infections, and emerging pathogens.

Defining virulence

In evolutionary biology, virulence is defined as the reduction in host fitness caused by infection with a parasite or pathogen. This definition is deliberately quantitative: virulence is measured not as a binary property—virulent or avirulent—but as a continuous variable reflecting the magnitude of host damage. In mathematical models, virulence is most commonly represented as the parasite-induced host mortality rate, designated α in the standard susceptible-infected (SI) framework.^{1, 5} This formalization allows virulence to be treated as an evolving trait subject to natural selection, with fitness consequences for both the pathogen and the host.

The choice of virulence measure matters greatly for bridging theory and experiment. Theorists overwhelmingly use host mortality as their metric because it translates directly into the epidemiological parameters of compartmental models. Empiricists, however, often measure virulence through sub-lethal proxies: weight loss, anemia, tissue damage, reductions in fecundity, or changes in behavior.¹³ These measures may correlate with mortality, but the relationship is not always linear or even monotonic. A pathogen that causes severe morbidity may not kill its host, and a pathogen that kills rapidly may cause little visible pathology before death. Bull and Lauring have argued that this disconnect between theoretical and empirical virulence measures is one of the most significant obstacles to testing evolutionary predictions about virulence in real biological systems.¹³

It is also important to distinguish virulence from pathogenicity, though the terms are sometimes used interchangeably in the medical literature. Pathogenicity refers to the qualitative ability of a microorganism to cause disease; virulence refers to the quantitative degree of damage inflicted. All virulent organisms are pathogenic, but not all pathogenic organisms are highly virulent. An organism with low virulence may still be pathogenic if it reliably causes mild illness. This distinction becomes critical when considering virulence as an evolving trait, because selection may shift the degree of harm without necessarily changing whether disease occurs at all.⁵

The trade-off hypothesis

The trade-off hypothesis, articulated by Anderson and May in 1982, replaced the avirulence hypothesis as the dominant framework for understanding virulence evolution. Its central claim is that virulence is not an incidental by-product of infection but is mechanistically linked to the pathogen's transmission rate, creating an evolutionary trade-off.¹ In the simplest formulation, a pathogen must replicate within its host to produce the infectious propagules necessary for transmission to new hosts. Higher within-host replication yields more propagules and thus a higher transmission rate (β), but it also damages host tissues more severely, increasing host mortality (α) and shortening the duration of infection. The pathogen's basic reproductive number, R₀, can be expressed as a function of these competing forces: R₀ = β(S) / (μ + α + γ), where S is the number of susceptible hosts, μ is the natural host death rate, and γ is the recovery rate.^{1, 5}

If transmission rate (β) increases with virulence (α) but so does the rate at which infected hosts are removed from the population (through death or recovery), then there exists an intermediate level of virulence that maximizes R₀. Pathogens that are too avirulent fail to produce enough transmission events; pathogens that are too virulent kill or incapacitate their hosts before transmission can occur. Natural selection favors the virulence level at which the marginal gain in transmission from increased replication exactly offsets the marginal cost of shortened infectious periods.^{1, 10} This optimum is not necessarily benign from the host's perspective; it may involve substantial morbidity and mortality.

Ewald extended this framework by emphasizing the role of transmission mode. In a landmark 1983 paper, he observed that among 45 directly transmitted diseases, only five had case fatality rates exceeding one percent, whereas among 18 vector-borne diseases, eight exceeded that threshold.² His explanation was that vector-borne pathogens do not depend on host mobility for transmission—a mosquito can feed on a bedridden patient just as easily as on a healthy one—and therefore face weaker selection against high virulence. Directly transmitted pathogens, by contrast, require their hosts to be mobile and socially active enough to contact new susceptibles, imposing a ceiling on how much damage they can inflict without compromising their own reproductive success.^{2, 3} Ewald later developed these ideas into a comprehensive framework in his 1994 book Evolution of Infectious Disease, which argued that transmission mode is the single most important predictor of virulence across pathogen taxa.³

The trade-off hypothesis was a significant conceptual advance because it grounded virulence evolution firmly in individual-level selection theory. Under the older avirulence hypothesis, reduced virulence was beneficial to pathogen lineages in the long run—a group selection argument. Under the trade-off hypothesis, intermediate virulence is favored because it maximizes the transmission success of individual pathogen genotypes in the short run.¹⁰ This shift placed virulence evolution squarely within the mainstream of Darwinian theory, treating virulence as a life-history trait analogous to clutch size or growth rate in free-living organisms.

Myxoma virus and Australian rabbits

The introduction of myxoma virus into the European rabbit (Oryctolagus cuniculus) population of Australia in 1950 produced what remains the most thoroughly documented natural experiment in virulence evolution. European rabbits, introduced to Australia in 1859, had multiplied into a devastating agricultural pest numbering in the hundreds of millions. The myxoma virus, native to South American Sylvilagus rabbits in which it causes only mild disease, was deliberately released near the Murray River in southeastern Australia as a biological control agent. In its new host, the virus proved extraordinarily lethal: the original Standard Laboratory Strain (SLS) killed more than 99 percent of infected European rabbits, typically within ten to fourteen days.^{8, 9}

Frank Fenner, recognizing the unprecedented opportunity for evolutionary study, established a long-term monitoring program that tracked the virulence of field isolates over subsequent decades. He and his colleagues developed a grading system for virulence, ranging from grade I (case fatality rate above 99 percent, as in the original SLS) through grade V (case fatality rate below 50 percent).^{8, 9} Within just two years of the initial release, and despite continued supplemental releases of the highly virulent SLS strain, field isolates showed a pronounced shift toward intermediate virulence. By the late 1950s, grade III strains—with case fatality rates between 70 and 95 percent—had become dominant in most Australian rabbit populations.⁸

Virulence grades of myxoma virus in Australian rabbits, 1950s–1960s^{8, 9}

The dominance of intermediate virulence was precisely what the trade-off hypothesis would later predict. Grade I viruses killed rabbits so quickly that the insect vectors (primarily mosquitoes and fleas in Australia) had limited time to feed on infected animals and transmit the virus onward. Grade V viruses, while allowing long infectious periods, produced low viral titres in skin lesions and thus generated fewer successful transmission events per unit time. The intermediate grade III strains struck the optimal balance: they maintained high enough viral loads for efficient vector-borne transmission while keeping their hosts alive long enough for multiple vector feeding opportunities.^{8, 10}

Crucially, the evolutionary story did not end with viral attenuation. Fenner also documented the reciprocal evolution of host resistance. Wild rabbit populations exposed to repeated myxomatosis epidemics evolved heritable genetic resistance, increasing their survival rates even when challenged with highly virulent strains.⁹ This coevolutionary response created a moving target for the virus: as rabbit resistance increased, the fitness advantage of intermediate virulence shifted, and the virus was under renewed selection pressure. By the 1990s, Kerr and colleagues documented a dramatic new escalation: many field isolates now induced an entirely novel disease phenotype characterized not by the classic myxomatosis skin tumors but by a rapid immunosuppressive syndrome resembling septic shock.¹⁵ Genomic analysis revealed that these viruses had accumulated mutations in multiple immunomodulatory genes, apparently as adaptations to overcome the enhanced immune defenses of resistant rabbit populations.²⁵ The myxoma-rabbit system thus illustrates not a simple trajectory toward reduced virulence but an ongoing arms race in which both partners continue to evolve.

Short-sighted evolution within hosts

The trade-off hypothesis treats virulence as a trait shaped by between-host selection: pathogens are selected to maximize their transmission to new hosts, and virulence evolves to the level that optimizes this population-level outcome. But pathogens do not experience selection only at the between-host level. Within each infected host, pathogen populations undergo rapid replication, mutation, and selection, and the traits favored by within-host competition may differ sharply from those favored by between-host transmission. This insight was formalized by Levin and Bull in 1994 under the concept of short-sighted evolution.⁴

Within a host, pathogen variants that replicate faster typically outcompete slower-growing variants for limited resources such as target cells, nutrients, or space within tissues. Faster replication is often associated with greater tissue damage—that is, higher virulence—but the fitness advantage it confers within the current host is immediate, whereas the fitness cost of killing the host (reduced future transmission) is diffuse and delayed. Selection within the host is therefore "short-sighted" in the sense that it favors mutations that increase local competitive advantage without regard for their consequences at the epidemiological level.⁴

Levin and Bull identified several diseases in which short-sighted evolution plausibly drives virulence above the between-host optimum. Bacterial meningitis provides a striking example: the bacterial pathogens responsible (Neisseria meningitidis, Streptococcus pneumoniae) are typically harmless commensals of the human nasopharynx, and their invasion of the central nervous system is an evolutionary dead end from which onward transmission is virtually impossible. The extreme virulence of meningitis is better explained as a by-product of within-host selection for variants that can exploit new tissue niches, even though doing so destroys the host and terminates transmission.⁴ Similarly, the devastation caused by poliovirus in a small fraction of infected individuals reflects the virus's rare penetration of the central nervous system—a tissue from which it cannot transmit—rather than any adaptation that enhances its epidemiological fitness.

HIV provides perhaps the most intensively studied example of within-host evolution. Over the course of a multi-year infection, HIV undergoes rapid mutation and selection within each patient, producing a diverse quasispecies. Variants that replicate most aggressively tend to dominate in late-stage infection, driving the CD4+ T-cell decline that characterizes AIDS. But these highly replicative, highly virulent variants are not necessarily the most transmissible between hosts; in fact, the transmitted viral population is typically a narrow bottleneck of founder variants, often drawn from the less diverse population of earlier infection stages.¹¹ The conflict between within-host and between-host selection in HIV creates a dynamic in which the virus may evolve toward higher virulence within individual patients even as between-host selection favors the intermediate set-point viral loads that maximize lifetime transmission potential.

HIV virulence and the set-point viral load

The set-point viral load—the quasi-stable level of HIV replication established after the acute phase of infection—serves as a natural measure of virulence that is both heritable between transmission pairs and strongly predictive of disease progression. Patients with high set-point viral loads progress to AIDS faster but may also be more infectious per contact; patients with low set-point viral loads survive longer but transmit less efficiently per unit time. This creates exactly the kind of trade-off the Anderson-May framework predicts.¹¹

Fraser and colleagues analyzed data from two large cohort studies—one in the Netherlands and one in Zambia—and found that the viral loads observed in practice are clustered around the value that would be predicted to maximize lifetime transmission potential. Specifically, they estimated that the optimal set-point viral load for transmission was approximately 4.5 log₁₀ copies per milliliter, close to the observed population means of 4.36 (Netherlands) and 4.74 (Zambia).¹¹ This concordance between observed and predicted optima provides some of the strongest quantitative evidence that a transmission-virulence trade-off operates in a real human pathogen.

Subsequent studies have examined whether HIV virulence has been changing over time in populations with long-established epidemics. Research from the Rakai district of Uganda documented a decline in set-point viral loads over several decades, consistent with a transmission-virulence trade-off in which virulent strains burn through susceptible hosts rapidly and leave behind a population structure that favors less virulent genotypes.¹² However, the interpretation of these trends is complicated by the widespread rollout of antiretroviral therapy (ART), which alters selection pressures by prolonging the lives of infected individuals regardless of their viral load. Whether the observed attenuation reflects genuine evolutionary adaptation or a statistical artifact of treatment coverage remains debated.^{12, 13}

Superinfection and competition between strains

The standard trade-off model assumes that each host harbors a single pathogen genotype, but in reality, hosts are frequently infected by multiple strains simultaneously. When a host already carrying one pathogen strain is invaded by a second, the resulting superinfection introduces within-host competition between genotypes with potentially different virulence levels. Nowak and May showed in 1994 that superinfection fundamentally alters the evolutionary dynamics of virulence, typically driving it above the single-infection optimum.⁷

The mechanism is straightforward. In a population where superinfection is common, a pathogen genotype that replicates faster than its competitors can displace them within co-infected hosts, gaining exclusive access to the host's resources and to onward transmission. This within-host competitive advantage favors higher replication rates and, by extension, higher virulence. The result is an arms race among pathogen strains, analogous to the tragedy of the commons: each strain benefits individually from faster replication, but the collective effect is greater host damage than any single strain would cause alone.⁷

Read and Taylor extended this analysis by reviewing the ecological consequences of genetically diverse infections across a range of parasitic systems. They found that mixed-genotype infections are the rule rather than the exception in many host-parasite systems, including malaria, schistosomiasis, and helminth infections, and that these mixed infections tend to be more virulent than single-genotype infections.¹⁴ In rodent malaria (Plasmodium chabaudi), experimental mixed infections consistently produce more severe anemia and higher parasite densities than infections with any single clone, because within-host competition selects for the most aggressively replicating variants.²⁴

The implications for public health are significant. Any intervention that reduces the diversity of co-infecting strains—such as treatment of the most virulent genotype or vaccination against specific serotypes—may inadvertently alter the competitive landscape within hosts and shift the evolutionary trajectory of virulence. Conversely, interventions that increase the opportunity for superinfection, such as crowded living conditions or inadequate sanitation, may select for more virulent pathogen populations.^{14, 26}

Transmission mode and virulence

One of the most robust predictions of virulence theory is that transmission mode shapes the evolved level of virulence. Ewald's original observation that vector-borne diseases tend to be more virulent than directly transmitted ones has been supported by comparative analyses across pathogen taxa, though the pattern admits numerous exceptions and complications.^{2, 3}

The logic extends naturally to the distinction between horizontal and vertical transmission. Vertically transmitted pathogens—those passed from parent to offspring through gametes, placental transfer, or birth canal exposure—have their fitness inextricably tied to the reproductive success of their hosts. A vertically transmitted pathogen that kills or sterilizes its host before reproduction eliminates its own lineage. Selection therefore strongly favors low virulence or even mutualism in obligately vertically transmitted symbionts.⁶ Many endosymbiotic bacteria, such as Wolbachia in insects and Buchnera in aphids, exemplify this trajectory, having evolved from parasitic ancestors into mutualists that provide essential nutrients or reproductive advantages to their hosts.

Lipsitch, Siller, and Nowak formally modeled the evolution of virulence in pathogens capable of both horizontal and vertical transmission and found that increasing the proportion of vertical transmission always lowers the evolutionarily stable level of virulence.⁶ Interestingly, their analysis also showed that increases in opportunities for horizontal transmission can, under certain parameter regimes, lower the ESS virulence as well—a counterintuitive result that arises because increased horizontal transmission can allow less virulent strains to outcompete more virulent ones by sustaining longer chains of infection.⁶

Water-borne and fomite-borne pathogens introduce further complexity. Ewald argued that pathogens transmitted through contaminated water, such as Vibrio cholerae, face little selection against high virulence because severely ill or dead hosts can continue to contaminate water supplies. Historically, cholera strains prevalent in regions with poor water sanitation have tended to be more virulent than those in regions with treated water, consistent with this prediction. Improvements in water quality, by severing the transmission link from immobilized hosts, should in principle select for reduced virulence over evolutionary time.³

Virulence factors and their genetic basis

At the molecular level, virulence is mediated by specific gene products—virulence factors—that enable pathogens to colonize hosts, evade immune defenses, acquire nutrients, and damage tissues. In bacteria, the genes encoding these factors are frequently organized into large chromosomal regions called pathogenicity islands (PAIs), which are acquired through horizontal gene transfer and represent some of the most dramatic examples of evolutionary innovation in microbial genomes.^{19, 20}

Pathogenicity islands are typically 10 to 200 kilobases in length, carry multiple virulence-associated genes, and are inserted adjacent to tRNA genes or other conserved sequences that serve as integration hotspots. They differ in GC content and codon usage from the core genome, betraying their foreign origin. Many PAIs are flanked by direct repeats or mobile genetic elements, indicating their capacity for excision, transfer, and reintegration—a modular architecture that allows virulence capabilities to be gained or lost in a single evolutionary step.¹⁹ The type III secretion systems of Salmonella, Yersinia, and Shigella, for example, are encoded on PAIs and allow these bacteria to inject effector proteins directly into host cells, subverting immune signaling, rearranging the cytoskeleton, and inducing apoptosis or invasion.²⁰

The modularity of virulence genetics has profound implications for virulence evolution. Because virulence factors can be acquired en bloc through phage transduction, conjugation, or natural transformation, a previously avirulent bacterium can become pathogenic in a single genetic event. The emergence of enterohemorrhagic Escherichia coli O157:H7, which causes severe bloody diarrhea and hemolytic uremic syndrome, was driven in part by the acquisition of Shiga toxin genes from a lambdoid bacteriophage—a clear example of virulence evolution through horizontal transfer rather than gradual point mutation.²⁰ Similarly, the pandemic spread of cholera has been shaped by the repeated acquisition and loss of the CTX phage, which carries the cholera toxin gene, and the Vibrio pathogenicity island, which encodes the toxin-coregulated pilus essential for intestinal colonization.

In viruses, virulence determinants are often distributed across the genome rather than concentrated in discrete islands, but the same evolutionary principles apply. Point mutations in the polymerase, surface glycoproteins, or immune evasion genes can dramatically alter virulence. In influenza A virus, the 1918 pandemic strain's extraordinary pathogenicity appears to have resulted from a constellation of mutations across multiple gene segments rather than a single virulence gene, illustrating how epistatic interactions between mutations can produce emergent virulence phenotypes.²²

Influenza, antigenic drift, and pandemic virulence

Influenza A virus provides a compelling case study in virulence evolution because its biology combines rapid mutation, antigenic drift, reassortment between co-infecting strains, and periodic zoonotic spillover from avian and swine reservoirs. The result is a pathogen whose virulence fluctuates on multiple timescales: incrementally between seasonal epidemics through antigenic drift, and catastrophically during pandemics when major antigenic shifts introduce novel surface proteins into immunologically naive human populations.²²

The 1918 influenza pandemic killed an estimated 50 to 100 million people worldwide, making it the deadliest single infectious disease event in recorded history. The extraordinary virulence of the 1918 H1N1 strain appears not to have been a product of gradual within-human adaptation but rather of a recent introduction of an avian-adapted virus into the human population, possibly through a series of adaptive mutations rather than reassortment with a pre-existing human strain.²² Once established, descendant lineages of the 1918 virus underwent attenuation over subsequent decades: the H1N1 lineage that circulated seasonally through the mid-twentieth century caused progressively milder disease as both the virus evolved and the human population accumulated partial immunity.

This attenuation pattern is consistent with the trade-off hypothesis. A directly transmitted respiratory pathogen like influenza depends on host mobility and social contact for transmission; extreme virulence that immobilizes patients may reduce transmission opportunities relative to strains that cause milder illness but keep hosts ambulatory and coughing in public spaces. However, the picture is complicated by antigenic drift, which continuously erodes population immunity and allows the virus to maintain substantial infection rates even without high per-contact virulence. Reassortment with avian or swine influenza strains can periodically introduce novel surface proteins against which the human population has no immunity, resetting the virulence landscape by enabling highly virulent genotypes to spread through immunologically defenseless populations.²²

Coevolution with host immune systems

Virulence evolution cannot be understood in isolation from the coevolutionary response of host immune systems. Hosts are not passive victims; they evolve defenses that reduce the fitness of their parasites, and parasites evolve counter-adaptations in an ongoing reciprocal dynamic described by the Red Queen hypothesis. The immune system is the primary arena of this coevolution, and its evolutionary elaboration is itself partly a response to the selective pressure imposed by virulent pathogens.²¹

Woolhouse and colleagues reviewed the evidence for host-pathogen coevolution and noted that while the theoretical expectation of reciprocal adaptation is well established, direct demonstration of coevolution in the sense of measurable allele frequency changes in both partners is remarkably difficult to achieve in practice.²¹ The strongest evidence comes from systems where both host and pathogen can be tracked over evolutionary time. In the myxoma-rabbit system, for instance, the parallel evolution of viral virulence strategies and rabbit immune resistance genes has been documented genomically: Australian rabbits evolved changes in genes involved in T-cell-mediated immunity, while the virus evolved novel immunosuppressive proteins to counteract these defenses.^{15, 25}

The major histocompatibility complex (MHC) provides the most celebrated example of pathogen-driven host evolution. MHC genes, which encode the cell-surface proteins responsible for presenting pathogen-derived peptides to T cells, are the most polymorphic loci in vertebrate genomes. This extreme diversity is maintained by balancing selection driven by pathogen pressure: hosts with rare MHC alleles can present pathogen peptides that the pathogen has not yet evolved to evade, conferring a frequency-dependent fitness advantage.²¹ Pathogens, in turn, evolve mutations in their antigenic epitopes that allow them to escape recognition by the most common host MHC alleles, driving a perpetual cycle of adaptation and counter-adaptation.

From the pathogen's perspective, the evolution of immune evasion mechanisms is itself a form of virulence evolution. Many virulence factors are not directly cytotoxic but instead function to subvert host immunity: bacterial capsules that prevent phagocytosis, viral proteins that block interferon signaling, helminth secretions that skew host immune responses toward ineffective pathways. The boundary between immune evasion and virulence is blurred because immune suppression often causes collateral tissue damage, and because the inflammatory response mounted against a pathogen frequently contributes more to disease pathology than the pathogen's direct effects.²¹

Emerging infectious diseases and virulence

When a pathogen spills over from its reservoir host into a new host species, the initial level of virulence in the new host is often poorly matched to any evolutionary optimum. Zoonotic pathogens may be highly virulent in their novel hosts simply because the host-pathogen interaction has had no evolutionary history, and the pathogen's replication machinery, calibrated to the physiology and immune defenses of the reservoir host, may cause inadvertent damage in the unfamiliar environment of the new host's tissues. This mismatch explains why many emerging infectious diseases—Ebola, Nipah, SARS, MERS—present with case fatality rates far higher than those typical of endemic human pathogens.¹²

Ebola virus (EBOV) illustrates the virulence consequences of recent zoonotic origin. In its suspected bat reservoir hosts, EBOV likely causes mild or asymptomatic infection, maintained by the bats' robust innate immune tolerance.¹² When the virus spills over into humans, it encounters an immune system that responds with massive inflammatory cytokine release, vascular damage, and disseminated intravascular coagulation—a pathological cascade that kills 50 to 90 percent of infected individuals depending on the viral species and outbreak context. This extreme virulence is not adaptive for the virus in any meaningful sense; it is a by-product of the mismatch between a bat-adapted pathogen and the human immune response. The high fatality rate, combined with the requirement for close contact with bodily fluids for transmission, tends to limit outbreak size and provides the virus with little opportunity for evolutionary adaptation to human-to-human transmission.

SARS-CoV-2, the virus responsible for COVID-19, offers a contrasting and more recent case study in virulence evolution during a pandemic. The ancestral Wuhan strain and early variants of concern (Alpha, Delta) displayed substantial virulence, with Delta in particular causing severe pneumonia and high mortality in unvaccinated populations. The emergence of the Omicron variant in late 2021, however, was associated with a marked reduction in pathogenicity. Suzuki and colleagues demonstrated that Omicron exhibited attenuated fusogenicity—its spike protein was less efficiently cleaved and less able to mediate cell-cell fusion compared to Delta and ancestral strains—and that it caused less severe disease in hamster models.²³

Whether Omicron's reduced virulence represents adaptive evolution toward the transmission-virulence trade-off optimum, a chance by-product of immune-evasion mutations in the spike protein, or some combination of both remains actively debated. Omicron accumulated an unprecedented number of mutations in the receptor-binding domain, many of which appear to have been selected for their ability to evade neutralizing antibodies rather than for any direct effect on virulence. The reduced pathogenicity may therefore be a pleiotropic side effect of antigenic evolution rather than a response to selection for lower virulence per se.²³ This distinction is important: if attenuation is an incidental consequence of immune escape, there is no guarantee that future variants will continue the trend toward mildness.

Virulence management and intervention

If virulence is an evolving trait shaped by ecological conditions, then in principle it should be possible to manipulate those conditions to steer pathogen evolution toward lower virulence—a concept known as virulence management. Ewald was among the first to advocate for this approach, arguing that improvements in sanitation, water quality, and vector control could reduce the transmission advantage of virulent strains and thereby select for milder pathogens over evolutionary time.³ The idea is intellectually appealing: rather than engaging in an endless arms race of drugs and vaccines against ever-evolving pathogens, public health interventions could harness evolution itself as a tool.

Ebert and Bull, however, challenged the feasibility of virulence management in a provocative 2003 review. They argued that the theoretical justification for virulence management rests on assumptions that may not hold in practice: that virulence and transmission are tightly linked, that the trade-off curve is known and stable, and that intervention can shift the optimum substantially enough to produce clinically meaningful reductions in virulence on relevant timescales.¹⁷ They pointed out that there is little empirical evidence for virulence reductions following public health interventions, and that the mechanism appears too weak for rapid selection of substantial changes in virulence. The debate between proponents and skeptics of virulence management remains unresolved, though both sides agree that more empirical data on the shape and stability of trade-off curves in real pathogens is urgently needed.^{17, 26}

Vaccination introduces its own evolutionary complications. Gandon and colleagues demonstrated theoretically that vaccines designed to reduce pathogen growth or toxicity within the host—so-called "imperfect" or "leaky" vaccines that do not prevent infection entirely—can paradoxically select for more virulent pathogen strains.¹⁶ The mechanism is that by reducing the fitness cost of virulence to the host (vaccinated hosts survive even with virulent infections), such vaccines relax selection against high virulence while still allowing transmission. The result is that pathogen populations evolving under widespread imperfect vaccination may reach higher virulence levels than they would in an unvaccinated population, making disease more severe for those who remain unvaccinated.¹⁶ This prediction has been partially supported by data from Marek's disease in poultry, where decades of vaccination with leaky vaccines have been associated with the emergence of increasingly virulent viral strains.

The relationship between antibiotic resistance and virulence adds another layer of complexity. Antibiotic resistance genes are sometimes carried on the same mobile genetic elements (plasmids, transposons, pathogenicity islands) as virulence factors, meaning that selection for resistance can co-select for virulence and vice versa.²⁰ In other cases, resistance mutations impose a fitness cost that reduces virulence in the absence of antibiotic pressure, but compensatory mutations can restore both fitness and virulence over time. The entanglement of resistance and virulence on shared genetic elements means that antibiotic use can have indirect and sometimes paradoxical effects on the virulence of circulating pathogen populations.

Empirical tests and ongoing challenges

Despite the elegance of virulence evolution theory, the empirical evidence remains uneven. A 2019 meta-analysis by Acevedo and colleagues evaluated the four core relationships predicted by the trade-off hypothesis across a broad range of host-pathogen systems: (1) that within-host replication increases with virulence, (2) that transmission rate increases with within-host replication, (3) that transmission rate increases with virulence but eventually decelerates, and (4) that recovery rate decreases with virulence.¹⁸ The meta-analysis found strong support for the first two relationships—replication is positively associated with both virulence and transmission—but the evidence for the crucial decelerating transmission-virulence curve, which generates the fitness peak at intermediate virulence, was equivocal due to high within-study variability.¹⁸

Empirical support for trade-off hypothesis predictions (meta-analysis)¹⁸

Cressler and colleagues reached a similar conclusion in their comprehensive review of theoretical predictions and empirical tests, noting that while the broad qualitative predictions of the trade-off hypothesis are well supported—virulence does tend to correlate positively with replication and transmission—the quantitative predictions about optimal virulence levels remain largely untested.¹² The problem is partly methodological: measuring transmission rates in natural populations is extremely difficult, and the shape of the trade-off curve (whether it is concave, convex, or linear) has enormous consequences for predicted outcomes but is almost never empirically determined.

Several alternative and complementary frameworks have been proposed. The coincidental virulence hypothesis argues that for many pathogens, virulence in a particular host is not under selection at all but is a by-product of adaptations to a different ecological context, such as survival in the external environment or replication in a reservoir host.¹³ The within-host evolution framework of Levin and Bull emphasizes that selection at multiple levels—within hosts and between hosts—can produce virulence outcomes that no single-level model can predict.⁴ Adaptive dynamics approaches have attempted to unify these perspectives by explicitly modeling the interaction between within-host and between-host selection as a multilevel evolutionary process.²⁶

The field continues to grapple with the fundamental challenge that virulence is not a single, stable trait but a context-dependent outcome of interactions between pathogen genotype, host genotype, host immune status, co-infecting microorganisms, environmental conditions, and medical interventions. Frank's 1996 review of models of parasite virulence identified this complexity as the field's central challenge and called for models that could accommodate the full range of ecological and genetic factors that shape virulence in nature.⁵ More than two decades later, that call remains relevant. The evolution of virulence is a field in which the theory has outpaced the data, and closing the gap will require long-term empirical studies in natural host-pathogen systems, experimental evolution in laboratory models, and continued refinement of mathematical frameworks that can bridge the scales from molecular pathogenesis to population epidemiology.