Mechanisms of disease-induced extinction




Parasites are important determinants of ecological dynamics. Despite the widespread perception that parasites (in the broad sense, including microbial pathogens) threaten species with extinction, the simplest deterministic models of parasite dynamics (i.e. of specialist parasites with density-dependent transmission) predict that parasites will always go extinct before their hosts. We review the primary theoretical mechanisms that allow disease-induced extinction and compare them with the empirical literature on parasitic threats to populations to assess the importance of different mechanisms in threatening natural populations. Small pre-epidemic population size and the presence of reservoirs are the most commonly cited factors for disease-induced extinction in empirical studies.


Parasites and disease are frequently cited as important drivers of population and community dynamics (Anderson & May 1992; McCallum & Dobson 1995; Levin et al. 1997; Hudson & Greenman 1998; Norman et al. 1999; Kohler & Hoiland 2001; Hudson et al. 2002; Shea & Chesson 2002; MacNeil et al. 2003). In conservation biology, disease is presented as a threat to population viability and a contributing factor to disease extinction (McCallum & Dobson 1995). However, the simplest disease models – which have formed the theoretical foundation for the field of disease ecology – suggest that disease alone cannot drive host populations extinct (Anderson & May 1992). More specifically, deterministic models of directly transmitted specialist parasites with density-dependent transmission predict that disease will always die out when the host population falls below a (non-zero) threshold density, before the host population can go extinct (Swinton et al. 1998; McCallum et al. 2001); stochastic models suggest that disease will often go extinct by so-called ‘fade-out’ even above this threshold (Bartlett 1960; Black 1966; Keeling & Grenfell 1997). The first part of this paper reviews the important exceptions to these simple conclusions – the qualitative mechanisms that drive disease-induced extinction in theoretical models. The second reviews the existing empirical literature on disease-induced extinction, and makes a first attempt to assess the relative importance of the different mechanisms in natural systems.

The literature search was performed in the ISI Web of Science, selecting any article containing the words [‘extinct*’ AND (‘disease’ OR ‘parasit*’ OR ‘pathogen’)] in the title, abstract or keywords. In total, 336 references were found. References to the coefficient of extinction of some substance (typically in the context of human physiology and medicine), articles that only vaguely mentioned disease as a possible threat for populations, and those referring to the extinction of parasites (rather than hosts) were dropped (260 in total). The remaining articles (76), which are the base of this review, were classified as either theoretical (33) or empirical (43) (some quantitative simulation models of specific host–parasite systems were included in the empirical rather than the theoretical section). We are aware that the so-called ‘gray literature’, not covered by the Web of Science, contains many references on conservation subjects, but we feel confident that our reference base is representative. We have not included cases of captive populations, since they are subjected to conditions very different from wild populations.

Theoretical mechanisms

The list of adjectives qualifying ‘disease models’ above – deterministic, density-dependent, specialist – suggests the mechanisms that allow diseases to drive their hosts extinct in models. Disease can drive populations temporarily or permanently to low numbers or densities, predisposing them to extinction by demographic stochasticity or Allee effects; diseases with frequency-dependent or spatial transmission can remain at high incidence even when populations have become globally rare; and diseases that can exploit other hosts (biotic reservoirs) or survive and grow in the environment (abiotic reservoirs) can remain at high incidence independent of population crashes in the focal host (Table 1).

Table 1.  Theoretical mechanisms that could produce disease-induced extinction
MechanismPredisposing factors
Stochasticity/small populationsSmall pre-epidemic populations (endangered/endemic species)
Small equilibrium density with disease present
Transient or periodic low population minima
Allee effects
Inbreeding effects
Fecundity-reducing parasites
Non-density dependent transmission/inhomogeneous mixingFrequency-dependent transmission
 Sexually-transmitted diseases
 Vector-borne diseases
 Social behaviour
Spatial dynamics/metapopulations
 Biotic: apparent competitors
 Abiotic: amplification in the environment

Small population sizes/stochasticity

While deterministic models allow host populations to fall to infinitesimal fractions of an individual without going extinct, stochastic models more realistically constrain the population to discrete values. With this change, it is easy to see how parasites could either drive their hosts directly to extinction, or drive them to such low densities that they can easily be driven extinct by demographic fluctuations. Once we recognize the importance of small population sizes, we can immediately see a variety of ecological scenarios that will enhance the risk of disease-induced extinction: for example, diseases attacking small (endemic or endangered) populations, or the presence of Allee effects (negative growth rate of populations at low densities, potentially caused by a variety of social, behavioural, or genetic factors). Lack of genetic variability, often caused by population bottlenecks, can reduce the ability of individuals to mount an immune response, increasing the probability that parasites will spread in the population. Low genetic variability may thus cause a positive feedback loop, where pathogens more easily invade small populations which, in turn, decreases their population size and genetic variability even more, in a process similar to Lynch et al.'s (1995)‘mutational meltdown’. We have not found any direct theoretical analysis of this effect, although we discuss possible empirical examples below (O'Brien et al. 1985; Sanjayan et al. 1996). Even if host populations naturally occur at high densities prior to an epidemic, parasites can transiently or permanently depress population densities. Permanent depression is most severe in the case of parasites that spread efficiently (i.e. that have a high R0) and have intermediate virulence (as measured by the rate of disease-induced host mortality), such that they can kill hosts but not so quickly that they limit their own reproduction (Anderson & May 1992). Transient depression of host populations, either during the troughs of host–parasite cycles or during the initial host population crash that follows the initial introduction of a parasite to a host population, is enhanced by long demographic (population regrowth) time scales relative to the time scale of disease spread (although this imbalance would also increase the chance of disease extinction); long incubation periods; and parasites that affect host fitness through sterility or decreased fecundity, especially if the pathogen does not cause significant mortality.

Frequency-dependent transmission/inhomogeneous mixing

The predictions of simple disease models also depend critically on the assumption of density-dependent transmission (McCallum et al. 2001), which is intimately linked with the assumption of spatially homogeneous mixing. The simplest departure from density-dependent transmission is frequency-dependent transmission, where the spread of disease is proportional not to the absolute number of susceptible or infectious hosts but rather to the frequency of infected individuals in the population; the clearest example of frequency-dependent transmission is in sexually transmitted diseases (Thrall et al. 1993; Antonovics et al. 1995), but frequency dependence can also occur because of territorial or social behaviour (Altizer et al. 2003). Since frequency-dependent transmission eliminates the host density threshold for parasite establishment, it also allows parasites to drive their hosts to extinction. Getz & Pickering (1983) show that frequency-transmitted parasites can coexist with their hosts if some density-dependent factor (other than the parasite) controls the host population, especially if density-dependence is stronger for the healthy class than for the infected class (Thrall et al. 1993).

A parasite transmitted by a vector also has a form of frequency-dependent transmission, especially for vectors that are efficient in finding their target species, maintaining high transmission rates even at low host populations (Boots & Sasaki 2003). In classical models of vector-borne disease, transmission is actually inversely proportional to host density, since vector populations remain constant and thus vectors become more concentrated on hosts as host populations decrease. In principle, this inverse density dependence could lead to a downward spiral as vector-borne disease increases in a dwindling host population; however, one has to assume in this case (unrealistically) that the vector population can support itself independently of the focal host population, but still maintain its preference for the focal host.

Thrall et al. (1993), using the same model for sexually or vector transmitted diseases, define the conditions that allow a frequency-transmitted disease to coexist with its host. Coexistence is impossible if infected individuals do not reproduce and density-dependent effects are equal for infected and uninfected hosts. One well-studied class of vector-transmitted diseases is pollinator-borne fungi, which also fall under the category of sexually-transmitted diseases, and typically sterilize their hosts. Ingvarsson & Lundberg (1993), using a deterministic model of the interaction of the castrating fungus Ustilago violacea and its host, the perennial herb Lychnis viscaria, show that the parasite can drive it to extinction as transmission efficiency approaches 100%. Alexander & Antonovics (1988) analyse the case of a vector-transmitted pathogen (U. violacea) of the plant Silene alba. They model transmission as dependent on the frequency of diseased plants, since the parasite is vector-transmitted (although they do not model the vector population explicitly). With a high transmission rate of the pathogen and low host recruitment, the predicted result is the local extinction of both host and pathogen populations.

Spatial dynamics

Models with explicit spatial structure in host and parasite densities also induce a sort of frequency-dependent transmission. Disease transmission in spatial models depends on local rather than global average population densities, and local densities of infected individuals often remain high even after global numbers have crashed, because infection is clustered within the population. Models of lattice-structured populations show that a host can be deterministically driven to extinction by a parasite (Sato et al. 1994; Knudsen & Schotzko 1999; Haraguchi & Sasaki 2000; Boots & Sasaki 2001). Boots & Sasaki (2001) demonstrate that, in this type of model, host extinction occurs only if the parasite significantly reduces host reproduction, and that parasites that have a relatively small effect on host death rate are more likely to cause host extinction. Models that consider space as continuous also show that spatial structure increases the probability of host extinction (Haydon et al. 2002). Models that describe space in terms of patches rather than continuous landscapes also allow for the extinction of host species if disease increases patch extinction rate above the colonization rate (of course, these models necessarily assume that disease can cause or increase the chances of local extinction within a patch). Metapopulation models of disease have been of great interest, both because many species of conservation concern live (naturally or through human intervention) in patchy landscapes, and because there has been some concern that conservation measures designed to increase population persistence by increasing connectivity among patches may also encourage the transmission of disease (Hess 1994). However, recent theoretical studies (Gog et al. 2002; McCallum & Dobson 2002) have suggested that increasing connectivity often favours species persistence, even if it encourages disease transmission as a side-effect. The difference between these recent studies and Hess’ approach is that the recent studies assume that infections have spilled over from more abundant alternative hosts (see below, section Specialist vs. generalist parasites: reservoirs). In that case (frequent for species of conservation concern), movement between patches will rarely have a negative impact, even for low probabilities of external infection.

Spatial models also make qualitatively different predictions about the evolution of host–pathogen interactions, as compared with models assuming homogeneous mixing. O'Keefe & Antonovics (2002) analyse the dynamics of parasites that reduce the fecundity of the host, but not its survival. According to a deterministic, non-spatial model, if host fecundity declines with increased pathogen transmission rate, then selection on the pathogen will cause host fecundity to decline toward zero, causing the extinction of the host and the pathogen. On the other hand, a spatially explicit stochastic model with the same assumptions predicts that virulence evolution does not necessarily lead to pathogen or host extinction, as it does in a well-mixed community. Boots & Sasaki (2003) also show that, in spatially explicit models, although disease-induced extinction may occur, it is unlikely to evolve. Moreover, they show that there must be a relationship between transmission and virulence for evolution to endemic coexistence of the host and the parasite. Haraguchi & Sasaki (2000) show that the evolutionary stable strategy for transmission and virulence in a spatial setting may be different than that predicted by a non-spatial or well-mixed model. The spatial structure leads to an intermediate transmission rate, even in the absence of an explicit tradeoff between transmission and virulence. The possibility of host or parasite extinction or coexistence depends on the host reproductive rate. Rand et al. (1995) also predict a substantially lower selection pressure in a spatial individual-based model, compared with its non-spatial analogue. All of these studies are exploring essentially the same mechanism: local structure leads to local extinction and thus to a form of group selection that drives the evolution of intermediate virulence.

Specialist vs. generalist parasites: reservoirs

It is clear that generalist parasites can overcome host density thresholds and drive a focal host species to extinction. The detailed community dynamics of multiple parasites sharing multiple hosts can be quite complex (Holt & Pickering 1985; Bowers & Turner 1997; Gatto & De Leo 1998; Bowers & Hodgkinson 2001; Holt et al. 2003). However, multi-host parasites can clearly lead to apparent competition, where a host species drives a competitor to extinction by being more tolerant or encouraging reproduction of a parasite that harms its competitor (Schmitz & Nudds 1994; Holt et al. 2003). Few theoretical explorations of parasite-mediated competition have appeared, probably because the mechanism is so simple, but analysing the details of particular empirical cases can be complicated (e.g. Schmitz & Nudds 1994). In the same general category as these biotic reservoirs are abiotic reservoirs, where a parasite can also survive and amplify itself in the environment as a saprophyte (Thrall et al. 1997). Rosáet al. (2003) show that the presence of an external reservoir may easily drive the host to extinction if the level of external infection is high.

Another potential mechanism of extinction is indirect or trophically mediated extinction; the extinction of one species by disease, or even a radical change in its population density, can have cascading effects through an ecological community (cf. Dunne et al. 2002). For example, in a reversal of the typical scenario of extinction through apparent competition, reduced density of one prey species could lead to predator switching, increased predator pressure, and hence decreased population density or even extinction of a second prey species that shared that predator. Another, even more direct, mechanism is the reduction of a species that may lead to extinction of a specialist consumer. Many potential scenarios exist – our own explorations (F. de Castro, B. Bolker, unpublished manuscript) have clearly just scratched the surface. In the next section, we will present a few possible examples from the existing empirical literature.

Empirical literature

We have surveyed the empirical literature on disease extinction to find cases where species have been threatened with or driven to extinction by disease. There are very few cases where disease has been implicated as the direct cause in the global extinction of a species (Pounds & Crump 1994; Daszak & Cunningham 1999). In fact, we have found no examples of the complete extinction of a species in the wild which can be attributed to a pathogen with certainty. However, in some cases (Table 2) disease has either threatened species or driven local populations extinct (Thorne & Williams 1988; Schwartz et al. 1995; Goltsman et al. 1996; Richards et al. 1999; Funayama et al. 2001; Muths et al. 2003).

Table 2.  Empirical examples of mechanisms of disease-induced extinction
Small populations/ stochasticityTree snail (Partula turgida)ExtinctionDaszak & Cunningham (1999)
Thylacine (Thylacinus cynocephalus)Possible extinctionGuiler (1961) (in McCallum & Dobson 1995)
Golden toad (Bufo periglenes)Probable extinctionPounds et al. (1997)
Black-footed ferret (Mustela nigripes)Probable extinctionThorne & Williams (1988)
Mednyi arctic fox (Alopex lagopus semenovi)Probable extinctionGoltsman et al. (1996)
African wild dog (Lycaon pictus)Population crashBurrows et al. (1994)
Boreal toad (Bufo boreas)Population crashMuths et al. (2003)
Noble crayfish (Astacus astacus)Population crashTaugbol et al. (1993)
Spanish Ibex (Capra pyrenaica hispanica)Population crashFandos (1991), Leon-Vizcaino et al. (1999)
Big-horn sheep (Ovis canadiensis)Population crash (model)Gross et al. (2000)
Florida torreya (Torreya taxifolia)Population crash and predicted extinction (model)Schwartz et al. (1995, 2000)
Reduced genetic variabilityPocket gopher (Thomomys bottae)Increased susceptibility to diseaseSanjayan et al. (1996)
Wolves on Isle Royale (Canis lupis)Increased susceptibility to diseasePeterson et al. (1998)
Cheetahs (Acinonyx jubatus)Increased susceptibility to diseaseO'Brien et al. (1985)
Non-density dependent transmissionKoala/ChlamydiaPossible population crash (model)Augustine (1998)
Eupatorium makinoi AsteraceaePopulation crashFunayama et al. (2001)
Inhomogeneous mixingRabbit/Rabbit haemorrhagic diseasePossible population crash (model)White et al. (2003)
Common flax (Linum marginale)Possible population crash (model)Thrall et al. (2003)
Biotic reservoirIiwi (Vestiaria coccinea)Population crashAtkinson et al. (1995)
Amakihi (Hemignathus virens)Population crashAtkinson et al. (2000)
White-tailed deer (Odocoileus virginianus)/ moose (Alces alces)Population reductionSchmitz & Nudds (1994)
Red squirrel (Sciurus vulgaris)Population reductionRushton et al. (2000)
Ethiopian wolf (Canis simensis)Population reduction (model)Haydon et al. (2002)
Abiotic reservoirTussock moth (Orgya antiqua)Population crashRichards et al. (1999)
Indirect extinctionEelgrass limpet (Lottia alveus)ExtinctionCarlton et al. (1991)
Eagle owl (Bubo bubo)Population crashMartinez & Zuberogoitia (2001)

Small population size

Of the mechanisms discussed in the first section, small populations are the mostly widely reported driver of disease-induced extinction in natural populations, although of course this could be a bias caused by the higher attention usually received by endangered species.

Total extinction of a species is cited only in a few cases, and the role of the pathogen is confirmed in even fewer. The snail Partula turgida became extinct in 1996 due to a microsporidian infection (Steinhausia sp., Daszak & Cunningham 1999), although it should be noted that the last individuals of P. turgida were kept in captivity, which makes it easier for an epizootic to affect all the remaining individuals. Another possible case of disease-induced extinction is the Thylacine (Thylacinus cynocephalus) (Guiler 1961, in McCallum & Dobson 1995). Although hard evidence is absent, the precipitous nature of the Thylacine's decline from 1906 to 1910, its simultaneous occurrence throughout Tasmania and some anecdotal reports of animals dying of a ‘distemper-like’ disease, may suggest that the final extinction (after a long period of over-hunting) could have been caused by a pathogen. Finally, the golden toad (Bufo periglenes), an amphibian endemic to Costa Rica with a very limited distribution, underwent a population crash in 1987 (Pounds et al. 1997). The species is currently classified by the IUCN as critically endangered or extinct (IUCN 1996). Pounds & Crump (1994) suggest the synergistic interaction of climate variations with a pathogen as the possible cause of the decline. Lips (1999) suggests that a fungal infection shared by many other species in Panama, whose populations are also declining, could have been the cause of the apparent extinction of the golden toad.

In other cases, extinction of a species by a disease has been avoided (or delayed) only by human intervention. Without that intervention, the characteristics of the disease made the final extinction probable. One well known example is the black-footed ferret (Mustela nigripes). The last known wild colony was extirpated by an epizootic of canine distemper in 1985 (Thorne & Williams 1988). Sixteen individuals, captured as an emergency measure during the epidemic, were devoted to a programme of breeding in captivity. Without those captive animals, the species would have probably been driven globally extinct by the disease. Also, a subspecies of arctic fox, Alopex lagopus semenovi, of Mednyi Island (Commander Islands, Russia), was reduced by some 85–90%, to around 90 animals, as a result of mange introduced by dogs in the 1970s (Goltsman et al. 1996). The population is currently under treatment with antiparasitic drugs, but its viability is still uncertain. The extremely high cub mortality (up to 96%), the social habit of the species (potentially leading to frequency-dependent transmission) and the low pre-epizootic population (around 600 individuals) all contributed to make probable the final extinction of the subspecies.

In many cases, although disease did not cause the total extinction of the host, it did produce a mortality high enough to threaten the species or to cause the disappearance of local populations. Such examples are more frequent, but it should be noted that any pathogen, almost by definition, will cause some decrease in the host population density; it is difficult to draw a clear line separating parasites with the potential to cause extinction. (Gog et al. 2002 and McCallum & Dobson 2002 both give lists of pathogens with impacts on host populations; we have discussed here only those with the potential to lead to local or global extinction.) Burrows et al. (1994, 1995) suggest that the local extinction of African wild dog (Lycaon pictus) in the Serengeti was caused by a viral disease enhanced by stress due to intervention (vaccination, collaring, etc.), although Ginsberg et al. (1995) argue that the population disappearance may have been caused by other factors. Taugbol et al. (1993) report the disappearance of the noble crayfish (Astacus astacus) from some lakes and long stretches of several rivers in Norway due to a plague of a fungal pathogen (Aphanomyces astaci). The pathogen's characteristics (species-specific, directly transmitted, short-lived spores) make complete eradication of the host improbable. Global declines of several amphibian populations provide well-documented examples of population extirpation by a pathogen. The evidence that a fungus (Batrachochytrium dendrobatidis) is responsible for some of the declines is strong (see for instance Collins & Storfer 2003 and Daszak et al. 2003). In particular Muths et al. (2003) report the near-disappearance of several populations of boreal toad (Bufo boreas) in Colorado (USA) apparently due to this same pathogen, and populations of yellow-legged frogs (Rana muscosa) are apparently vanishing from the Californian Sierra Nevada, again apparently from chytrid fungus (C. Briggs, pers. comm.; Bradford 1991; Fellers et al. 2001). Some theoretical analyses of specific cases also predict the extinction of populations or species. For instance Gross et al. (2000), simulating populations of big-horn sheep (Ovis canadiensis) with an individual-based model, found a probability of extinction exceeding 20% over a 200 year period with multiple infections.

As noted above, finite-population effects can also be important in populations that are initially large, but are rapidly reduced in density and numbers by disease. For instance, shortly after 1987, the population of Spanish Ibex (Capra pyrenaica hispanica) in southern Spain suffered a 95% reduction due to a sarcoptic mange epizootic, from about 9,600 individuals to less than 200 (Fandos 1991), including the total disappearance of a local group of 35 individuals in 6 months (Leon-Vizcaino et al. 1999).

We have found only a single recent reference to a plant species threatened with extinction by a disease. The Florida torreya (Torreya taxifolia) suffered a severe population reduction during the 1950s, probably as a result of a fungal disease (Schwartz et al. 1995). The present population is composed of small, sexually immature trees, not producing seeds. All the models applied to its population dynamics predict the extinction of the species (Schwartz et al. 2000). Two other plant pathogens have also caused dramatic and widespread declines in their hosts: the chestnut blight (Endothia parasitica) (Griffin 2000) and the Dutch elm disease (Ophiostoma ulmi) (Gibbs 1978; Brasier 1987; Hubbes 1999).

Another aspect of the risk associated with small population size is lack of genetic variability. Extreme population bottlenecks can reduce the genetic variability of a species and reduce its effective immune response, making it more vulnerable to disease. There are several examples of high mortality attributed to this cause: Cheetahs (Acinonyx jubatus) and feline infectious peritonitis (O'Brien et al. 1985); pocket gopher (Thomomys bottae) (Sanjayan et al. 1996); and wolves (Canis lupis) at Isle Royale and canine parvovirus (Peterson et al. 1998).

Small host populations are the most commonly cited reason for disease threatening host population viability. Although only one reference conclusively attributes the extinction of a whole species (Partula turgida) to disease, there are two cases (the Thylacine and the golden toad) where disease is a plausible cause or cofactor, and others (the black-footed ferret and the Mednyi arctic fox) where extinction was probably avoided only because of human intervention. A large variety of other examples cite disease as a threat, but do not convincingly show that disease alone will drive extinction. Except in the case of the golden toad, which was locally abundant before the population crashed, none of the references cites the disease as the cause of the initial decline from abundance to scarcity, but rather as a threat to already small populations.

Frequency-dependent transmission/inhomogeneous mixing

Frequency-dependent transmission

As discussed above, simple models of host–parasite dynamics with frequency-dependent transmission (broadly including STDs and vector-borne diseases) show that a parasite can easily drive its host to extinction (Getz & Pickering 1983). Despite this theoretical generalization, we have found no empirical evidence of a population or species eliminated by frequency-dependent parasites. Some models applied to specific situations suggest that the extinction of the host is possible (Alexander & Antonovics 1988; Ingvarsson & Lundberg 1993), but apparently extinction has never been reported for a real population. A model of Koala (Phascolarctos cinereus) and the sexually transmitted disease Chlamydia psittaci shows that extinction of the host is possible if the transmission of the disease increases and, at the same time, Koalas become less resistant to the disease or if Koalas’ intrinsic population growth rate drops below 0.1 (Augustine 1998). However, stable coexistence of the Koala–Chlamydia system is predicted for a wide range of population and disease parameters, and the general consensus seems to be that Chlamydia has not led to local extinction of koala populations (Melzer et al. 2000). In general, there is little empirical evidence that diseases that are transmitted in a frequency-dependent way are particularly threatening.

Spatial dynamics

Despite the abundant theoretical literature on spatial host–pathogen interactions, we have found little empirical evidence on the subject. Thrall et al. (2003) use an experimental metapopulation approach to study the host–pathogen system Linum marginale–Melampsora lini, although they focus in the local extinction and recolonization of the pathogen, not the host. They show that more isolated patches tended to exhibit lower levels of disease during epidemic peaks than patches that were close together. White et al. (2003), studying rabbit haemorrhagic disease, describe how the different use of space of surface-dwelling and warren-based rabbits creates a hierarchical contact and transmission structure within the rabbit population that affects the population-level outcome of the infection. Although the rabbit populations studied did not disappear, this study is an example of how spatial structure can affect infection dynamics. The lack of empirical studies on this subject may be due to the difficulty of managing controlled experiments of disease dynamics together with spatial dynamics. Clearly, more efforts are needed in this area to verify theoretical predictions.

Specialist vs. generalist parasites: reservoirs

The effect of reservoirs in host–parasite dynamics and its influence on the relationship between hosts is straightforward and theoretically well understood, which is reflected in the relatively numerous empirical examples of this type of interaction. The case of avian malaria (Plasmodium relictum) and its effect on Hawaiian avifauna has been well described. The Iiwi (Vestiaria coccinea), a declining native species from Hawaii, is much more susceptible to avian malaria than the introduced Nutmeg Mannikin (Lonchura punctulata) (Atkinson et al. 1995). (One could argue that the vector transmission of avian malaria also affects its population impact, but the simple thought experiment of removing non-native species – which would harm native species by intensifying transmission if malaria's vector transmission is regulating its impact – suggests that reservoirs are more important.) The presence of the Mannikins acting as reservoirs for avian malaria threatens the remaining high elevation populations of endangered native birds. Another species, the Hawaii Amakihi (Hemignathus virens) is also restricted in its distribution by its higher susceptibility to malaria (Atkinson et al. 2000). The case of the interaction between the white-tailed deer (Odocoileus virginianus) and the moose (Alces alces) is also well known: the white-tailed deer can competitively exclude moose through its higher resistance to the parasitic meningeal helminth Parelaphostrongylus tenuis (Anderson 1974; Lankester 1987; Schmitz & Nudds 1994). Other examples include the parapoxvirus-mediated competitive exclusion of red squirrels by gray squirrels (Rushton et al. 2000); the spillover effects of rabies and canine distemper from domestic dogs to the threatened Ethiopian wolf (C. simensis) (Haydon et al. 2002); and the spillover of rinderpest to wild ruminants from domestic cattle (Rossiter et al. 1998; Kock et al. 1999).

In some conditions, a reservoir can also allow the coexistence of host and parasite. Funayama et al. (2001) describe how a reservoir species (Lonicera japonica) for a virus affecting the plant Eupatorium makinoi would enable the virus to persist after the local extinction of the host, allowing their long-term coexistence, assuming that recolonization by the host is possible from other populations.

Abiotic reservoirs can also drive extinction, although theory suggests that either the pathogen must be able to grow (and not merely persist) in the environment, or that an interaction with small-population effects is required. We have found only one reference to the role of an abiotic reservoir in the disappearance of a population (Richards et al. 1999). Early-instar larvae of the polyphagous vapourer moth (Orgyia antiqua), parasitized by a nucleopolyhedrovirus, disperse from tree canopy to the understorey, with subsequent relocation back to the tree as late-instar larvae. Virus spores are capable of persisting outside the host in a viable state for long periods of time if shielded from sunlight. Widespread epizootics provoked by larval foraging behaviour seemingly wiped out the population (meticulous searches of egg masses in the study area were unsuccessful in the 2 years following the epizootics).

Trophic and competitive interactions can also allow diseases to cause the extinction of non-focal species indirectly, even when their host population persists. If the disease causes strong enough reduction of the host population, some species dependent on the host may become extinct. The eelgrass limpet (Lottia alveus) has apparently gone extinct as a consequence of the decline of the eelgrass Zostera marina (Carlton et al. 1991), whose population suffered a catastrophic decline of in the early 1930s in the North Atlantic Ocean, attributed to the ‘wasting disease’ of Zostera, which is caused by the slime mould Labyrinthula zosterae (Muehlstein et al. 1991). Another example is the severe reduction of the Eagle Owl population (Bubo bubo) in Alicante (Spain) after an outbreak of rabbit haemorrhagic disease decimated the population of its main prey in the area (Martinez & Zuberogoitia 2001). Even if not causing extinctions, the effects of diseases in keystone species can be very widespread; for example, the consequences of rabbit population declines due to introduced myxomatosis (Sumption & Flowerdew 1985).


We have reviewed the basic ways in which simple epidemiological models can be extended to allow for disease-induced extinction of host populations. Despite a great deal of theoretical exploration, we have found that some of these mechanisms have few or no confirmed empirical examples (non-homogeneous mixing, spatial structure, social behaviour, sexually transmitted diseases). There is little documentation of the various mechanisms or predisposing factors other than small population sizes and reservoirs. Spatial and frequency-dependent mechanisms are interesting theoretically, but as yet there is little empirical evidence that they are important in natural systems. In some sense the paucity of data is not surprising, however, since we are looking at the intersection of two areas (extinction and disease ecology) which are both by their nature cryptic and difficult to study in natural communities.

Another factor to consider is that empirical studies typically include several of the mechanisms mentioned in this review: small populations and reservoirs, vector-borne and spatial dynamics, etc. while theoretical studies tend to focus in only one mechanism. In studying each empirical case, careful consideration is needed to include all possible factors implicated, to obtain a clear understanding of the process.


NSF Integrated Research Challenges in Environmental Biology grant IBN 9977063 supported this work. We would like to thank the comments of the members of our laboratory and four anonymous referees that greatly improved the manuscript.