Many parasitic organisms (e.g., virus, helminths, parasitoids) have evolved the capacity to alter phenotypic traits of their hosts, extending from color, morphology, and behavior, in order to either increase their probability of transmission and/or survival in a given host or insure that their propagules will be released in an appropriate habitat (see reviews by Moore 2002; Thomas et al. 2005; Poulin 2010; Hughes et al. 2012). For example, tropical ants parasitized with nematodes go perch and develop bright red abdomens (filled with nematode eggs) that resemble ripe fruits in the tropical rain forest canopy (Yanoviak et al. 2008). This drastic alteration of the ants' appearance increases their predation by frugivorous birds, which then pass the parasite eggs in their feces. The widespread protozoan parasite Toxoplasma gondii, which must be transmitted from a rodent (intermediate host) to a felid (definitive host), reverses the innate aversion of the rodent to cat odor into attraction, thereby increasing the probability of intermediate host predation (Berdoy et al. 2000). Some hairworm species parasitizing Orthoptera alter the behavior of their hosts in a way that forces them to jump into water, where the worms can emerge from the host and search for a sexual partner (Thomas et al. 2002a). Several parasitoids can even usurp host behavior after leaving it (Brodeur and Vet 1994), converting the latter into a bodyguard protecting developing pupae from approaching predators and hyperparasitoids (Grossman et al. 2008; Harvey et al. 2008; Maure et al. 2011).
Determining why and how host manipulation by parasites evolves is a fascinating but challenging question for evolutionary biologists (Lefèvre et al. 2009; Thomas et al. 2012). The extended phenotype perspective (Dawkins 1982) postulates that host behavioral alteration should be regarded as the expression of the parasite's genes in the host phenotype. Natural selection is indeed expected to favor the ability of parasites to induce behavioral, morphological, or physiological alterations in their hosts that are beneficial for themselves, even though they are detrimental to the host's fitness. This scenario, also called manipulation sensu stricto, is a decidedly parasite-oriented view, and is traditionally considered as the main process used by parasites to manipulate their host's behavior. Recent studies, however, acknowledged that different evolutionary routes can lead to host manipulation, notably processes involving compromises between host and parasite strategies rather than a complete parasite takeover (Thomas et al. 2012). For instance, Lefèvre et al. (2008) proposed that parasites could theoretically achieve transmission by triggering host compensatory responses when these responses match, at least partially, with the transmission route. In this view, genes of the parasite are selected for their pathological effects that induce a host compensatory response. As behavioral changes both mitigate the costs of infection for the host and meet the objectives of the parasite in terms of transmission, natural selection is likely to favor the evolution of such interaction (Lefèvre et al. 2009). In accordance with this hypothesis, the sexually transmitted ectoparasite mite Chrysomelobia labidomera reduces the survival of its leaf beetle host (Labidomera clivicollis), and in response infected males exhibit increased sexual behavior before dying (Abbot and Dill 2001). This compensatory response from the host clearly benefits the sexually transmitted parasite as enhanced inter- and intrasexual contacts (i.e., copulation and competition) provide more opportunities for transmission (Drummond et al. 1989; Abbot and Dill 2001). Although a few other examples support the idea that parasites could indeed exploit host compensatory responses instead of manipulating sensu stricto their host (see Lefèvre et al. 2009), it remains unclear if this manipulative strategy is widespread or not. Issues of manipulation sensu stricto versus interactive scenarios have much to gain from a theoretical approach.
To address the issue, we developed a simulation model that predicts the conditions under which parasites should benefit from using a strategy based on the exploitation of compensatory responses, either alone or in concomitance with a manipulation sensu stricto strategy. Because predation of the intermediate host by the definitive host is necessary for parasite transmission, parasites that manipulate the behavior of their host may benefit from increasing either the vulnerability of hosts to predation or their susceptibility to predation by suitable predators. In order to do that, parasites can use two nonmutually exclusive strategies: they can induce a compensatory response that reduces the negative effects of infection on host fecundity (thus providing a benefit to the host) but in turn renders their transmission more probable (thus providing a benefit to the parasite as well) and/or they can exert a certain manipulative effort to alter the behavior and/or appearance of infected hosts, making them more susceptible to predation by definitive hosts. More precisely, parasites that trigger host compensatory responses can, for instance, affect the energy requirement for reproduction, thereby causing hosts to increase their foraging activity so that they can acquire enough resources to reproduce, but also inevitably their vulnerability to predation. Conversely, parasites that use a manipulative strategy sensu stricto can modify the response of their hosts to predation risks by definitive hosts, for example, by turning their innate aversion into an imprudent attraction (Berdoy et al. 2000). Among other parameters, we expect the evolutionary stable strategy to depend critically on the probability that hosts possess in their repertoire a compensatory response that matches the transmission objectives of the parasites as well as on the benefits for the hosts of opposing manipulation or compensating. Therefore, because parasitized hosts may show different levels of tolerance/resistance to parasite-induced behavioral changes (Thomas et al. 2011), we also considered in our model two types of hosts that differ in their ability to suppress the manipulative efforts of the parasite (i.e., manipulatable and unmanipulatable hosts).
To analyze the dynamics of hosts and parasites over time, we run simulations over consecutive generations until they reach equilibrium states. For a given simulation, all parameters remained fixed over time, except the relative proportion of unmanipulatable and manipulatable hosts (denoted xt and (1 − xt), respectively) as well as the prevalence of hosts (i.e., the probability that a parasite finds a host) and parasites (i.e., the probability that a host becomes infected by a parasite) that were estimated at each time t from parameters pt and qt, respectively. For the first generation (i.e., at t = 0), we used the following default starting values: x0 = 0.5, p0 = 0.5, and q0 = 0.8.
To find the optimal parasite strategy at each time t, we assume a two-step decision process: first both the parasites and their host decide, respectively, whether or not to induce a compensatory response and, if relevant, whether or not to compensate, and second the parasites decide how much effort they invest in manipulation to maximize their expected fitness.
Should parasites exploit host compensatory responses?
The passive expected fitness of a parasite that does not use any strategy to increase its transmission to a definitive host is denoted by Wατ. It depends on (1) W: the basic reproductive success of parasites; (2) τ: the predation rate of hosts whose behavior is not altered (which is assumed to be equal to that of uninfected hosts); and (3) α: the proportion of predatory events that are attributable to a suitable predator (i.e., a predator in which a parasite can complete its life cycle).
The induction of a compensatory response by a parasite not only reduces its fitness by CCR (the cost of inducing a compensatory response) but it may also alter the behavior of its host in a way that increases either its rate of predation (by a factor τCR) or its susceptibility to predation by suitable predators (by a factor αCR), provided that the host decides to compensate.
We assume that parasites decide to induce a compensatory response only if their expected fitness when they do so and the hosts compensate (i.e., [Wα(τ + τCR) − CCR] or[W(α + αCR)τ − CCR] if the parasite-induced change in host behavior increases the rate of predation or the susceptibility to predation by suitable predators, respectively) is larger than their passive fitness. Therefore, we predict that parasites should induce a compensatory response only when the benefit of compensation (τCR or αCR) is larger than: CCR/Wα or than: CCR/Wτ, if the change in host behavior concerns the rate of predation or the susceptibility to predation by suitable predators, respectively. From these conditions, we can conclude that parasites should induce a compensatory response more frequently when the cost of doing so (CCR) is relatively small and in parasite species in which individuals produce a large number of offspring (W). Furthermore, when hosts possess in their repertoire a compensatory response that increases their predation risk (τCR), parasites would benefit from exploiting this response only if there is a high probability that hosts are consumed by a suitable predator (i.e., high α). Conversely, when the available compensatory response affects the susceptibility to predation by suitable predators (αCR), the likelihood that parasites induce a compensatory response should increase with the risk of predation incurred by hosts (τ).
Should hosts compensate?
If parasites induce a compensatory response, then hosts can also decide to compensate or not. We assume that uninfected hosts produce on average w offspring. Thus, as their mortality rate due to predation is τ, their mean breeding success equals (1 − τ)w.
When hosts are infected by a parasite, the fecundity of female hosts is reduced, and their expected breeding success then becomes σ(1 − τ)w. This detrimental consequence of infection, however, can be reduced through compensatory responses. In that case, female hosts suffer less fecundity reduction than those that do not compensate (σCR and σ respectively, with σCR>σ), but in turn they become more vulnerable to predation or more susceptible to suitable predators. The fitness of hosts that do compensate therefore is: σCR(1 – τ − τCR)w if the parasite-induced change in host behavior affects the rate of predation, or σCR(1 − τ)w if the parasite-induced change in host behavior affects the susceptibility to predation by suitable predators. Note that in that latter case, the mortality rate of infected hosts is not reduced and hosts, therefore, only receive benefits from compensating. As above, we assume that hosts compensate only if the benefits of compensation in terms of increased fecundity outweigh the costs. As host females that compensate have a higher fecundity compared to those that do not (i.e., σCR > σ), we predict that hosts should always compensate when the parasite-induced change increases the susceptibility to predation by suitable hosts (τCR). Indeed, in that case the induced compensatory response only increases the proportion of suitable versus unsuitable predators that do consume the hosts, but not the overall rate of predation. Conversely, when the change in host behavior increases the rate of predation, one would expect hosts to compensate only when the benefit of compensation in terms of increased predation risk (τCR) is smaller than: . From this condition, we can conclude that hosts should compensate with a higher probability when (1) the reduction in fecundity in infected host females is important (i.e., small values of σ); (2) the benefits of compensating are large (i.e., high values of σCR); and (3) the predation rate of uninfected hosts (τ) is small.
How much effort should parasites invest in manipulation?
The last step of the decision process consists in determining how much energy parasites invest in manipulation, in order to maximize their expected fitness. Although we allow parasites to adjust their manipulative effort to conditions, we are not interested in exploring how it should vary in relation to various factors (see Poulin 1994a). Instead the aim of our model is to predict the conditions that should favor the use of a strategy based on the exploitation of compensatory responses either alone or in concomitance with a manipulation sensu stricto strategy. This is the reason why the benefit and the cost of manipulation, that are both functions of the amount of manipulative effort invested (ME), were kept constant. More specifically, we fixed the cost of manipulation CME to ME2 and, as for the strategy based on the exploitation of compensatory responses, we assumed that the parasite-induced change in host behavior resulting from manipulation may affect either the rate of predation (with τME = 0.2ME) or the susceptibility to predation by suitable predators (with αME = 0.2ME). In the case of an unmanipulatable host the benefits of manipulation are also proportional to its efficiency in suppressing the parasite efforts (ε) and are then equal to (1 − ε)τME or (1−ε)αME depending on whether the manipulation increases the rate of predation or the susceptibility to predation by suitable predators. To determine the optimal manipulative effort invested, we estimated the average fitness expected by parasites for each value of ME, and then we retained the value ME* for which the fitness is maximal (see Fig. 1 for an example of the gain expected by the parasite for each of the four possible cases when the host is manipulatable). Once we have retained the value of ME*, we can estimate the fitness of parasites as well as that expected by unmanipulatable and manipulatable hosts for the generation t. Figure 2 gives an example of the gain expected by a manipulatable host in each of the four possible cases. In the case of an unmanipulatable host, their expected fitness is reduced by CR (the cost of host resistance incurred by unmanipulatable hosts to suppress parasite manipulative efforts). The gains expected by hosts and parasites in each possible situation are given in Tables S1 and S2.
The expected fitness of parasites and hosts at each time (generation) t depend on the proportion of hosts infected by a parasite (qt), on the probability that a parasite finds a host (pt), as well as on the relative proportion of unmanipulatable hosts (xt). For the first generation (i.e., at t = 0), we use the initial starting values of x0, p0, and q0 to determine the success of each strategy. Then for the subsequent generations, the values of these three parameters have to be estimated at the beginning of each generation before we can evaluate the average success expected by hosts and parasites. For sake of simplicity, we assume that pt and qt are determined solely by the population sizes of hosts and parasites (and not, for instance, by the density of individuals), and that population sizes are limited to KP and KH individuals, for parasites and hosts, respectively (with KP = KH = 500). Also, for convenience, we hypothesize that all individuals die at the end of each generation t and are replaced by their offspring at generation (t + 1).
In order to evaluate the prevalence of hosts and parasites, we consider that each host can be infected by only one parasite and that parasites detect potential hosts (i.e., uninfected hosts) with a probability λ that represents their searching efficiency.
Finally, the proportions of unmanipulatable and manipulatable hosts at generation (t + 1) are proportional to their relative success at generation t.