Introduction

Arms races between predators and their prey are ubiquitous in food webs (Wade 2007), including food webs containing venom producers (Casewell et al. 2013). Research has typically focused on venom production for prey capture, but venom use may also coevolve as a defensive adaptation (Casewell et al. 2013; Jansa and Voss 2011). Indeed, the evolution of discretely different offensive and defensive venoms has been recently identified in Conus geographus (Dutertre et al. 2014), and scorpions secrete a chemically cheap “prevenom” for use in low threat agonistic interactions before resorting to a more chemically complex and costly “main” venom (Inceoglu et al. 2003). It is still unclear, however, how tritrophic interactions alter venom evolution. To understand how venomous animals balance arms races between both prey and predators, we must extend beyond pairwise interactions to reveal unintuitive processes and dynamics, such as reinforcement or weakening of coevolution in one arms race due to the dynamics of another (Abrams 1991; Dercole et al. 2010). Given that consumer species are typically embedded in food webs, and in light of accumulating evidence for the importance of tritrophic interactions in coevolution (Currie et al. 2003; Dietl and Kelley 2002), valuable new insights may be gained by evaluating arms races simultaneously in venomous predator-prey systems.

Because some venom toxins act upon biochemical processes that are highly conserved across all of Animalia (Fry et al. 2009), a single toxin can sometimes be produced to mediate both offensive and defensive interactions. However, if the targeted physiologies of prey and predators differ significantly, two separate toxins may be necessary. Relatively, unexplored evidence for different secretions in a range of venomous taxa may vindicate recent calls to re-evaluate the more widespread potential for discrete “offensive” and “defensive” secretions (Dutertre et al. 2014; Morgenstern and King 2013). However, the circumstances under which separate venoms should evolve remain unclear. Furthermore, the view that adaptive venom composition is under selection for a specific prey type and efficiency has not been universally accepted (Casewell et al. 2013; Mebs 2001; Sasa 1999).

In order to explain mixed evidence for both the specificity of venom towards prey and for economical venom use (Casewell et al. 2013), particularly in snakes, the “overkill” hypothesis posited that the adaptive value of snake venoms may exhibit a continuum, where selective advantages of venom may be evident in some taxa but absent in others (Sasa 1999). Then, where selection for diet is weak, neutral evolutionary processes give rise to the wide variability observed in the chemical composition, injection volumes, and potency of venoms seen in snakes (Mebs 2001). However, implicit in the overkill hypothesis is a low cost for venom use, which has been called into question in some cases by evidence of a strong positive selection acting on venom (e.g., Li et al. 2005). More recently, an alternative “venom optimization” hypothesis has been developed to explain selection for economy as a consequence of the cost of venom use (Morgenstern and King 2013), and the methodologies of some studies which have found evidence of large injection volumes has been controversial (Morgenstern and King 2013). Nonetheless, evidence of an “overkill” effect—large (and seemingly wasteful) injection volumes and extraordinarily high toxicity of venoms in snakes—remains (Morgenstern and King 2013). To some extent, injection volumes greatly in excess of median lethal doses may also be explained by bet-hedging strategies to guarantee a meal (Hayes 1992). In addition, there may be new insights when arms races are placed in the greater context of food webs. We consider one possible alternative for the occurrence of “overkill.”

The debate surrounding the overkill hypothesis has generally focused on pairwise interactions (between venomous animals and their prey). However, because venom can function as a multipurpose weapon for prey capture and defense, predatory pressure in tandem with prey interactions may also play a role in the apparent paradox of “wasteful” venom expenditure. Motivated by these persisting questions in venom ecology and evolution, we consider the effects of coevolution on venom production in a venomous consumer embedded in a tritrophic food chain. Studies of coevolution in tritrophic systems involving plants, herbivores, and predators of herbivores suggest that herbivores attempt to balance the separate pairwise arms races with their prey (plants) and predators to minimize the fitness costs of investing in offensive and defensive traits (Nersesian et al. 2011). Thus, a consumer may adopt intermediate trait values when arms races between their predators and prey must be balanced. We used a model to evaluate how venomous animals might balance these two general arms races with their predators and prey and to determine the circumstances under which it would be preferable to evolve a separate defensive venom, or invest more heavily in the venom used for prey capture. Sunagar and Moran (2015) recently proposed that venoms undergo alternating “speeds” of evolution, where rapid diversification of the venom arsenal accompanies the earlier stages of ecological specialization and is followed by a longer period of purification to ensure the sustainability of venom potency. However, diversification may also be precipitated by sudden shifts in ecology (Sunagar and Moran 2015). We explore this scenario with the invasion of a coevolutionary arms race between a venom user and its prey by a novel predator.

In order to describe how species interactions alter venom evolution, we must make some simplifying assumptions. First, while a cost associated with chemical warfare is generally accepted, the exact nature of this cost remains controversial (Smith et al. 2014; McCue 2006). Because the exact nature is still unclear, we do not explicitly describe the source of this cost but assume that the effectiveness of venom in facilitating prey capture is a quantitative trait that increases with investment by the venom user. This assumption is consistent with evidence a strong positive selection pressure acting on venom inferred from the secondary loss of venom toxicity associated with shifts in diet (Morgenstern and King 2013). The quantitative trait we describe could represent the quantity or concentration of venom or the diversity in a particular class of venoms, so long as increasing diversity comes at a cost but benefits the organism.

Previous work suggests that high rates of gene duplication provides the engine of evolution and rapid diversification in venom, generating novel compounds via neofunctionalization of duplicates and higher production of gene product via gene dosing, enabling venom users to overcome venom resistance in predators and prey (Casewell et al. 2013). Although some models of gene duplication assume that duplications are non-costly, experimental work suggests that the fitness costs of carrying copies may in fact be high (Adler et al. 2014). Costs associated with the venom trait could also arise from maintenance of the venom storage apparatus (McCue 2006). Additionally, investment in both offensive and defensive venoms may be particularly costly due to the need to maintain specialized structures and tissues to separately store or secrete two separate venoms, as in Conus geographus (Dutertre et al. 2014). Thus, we also consider a multiplicative cost associated with investing in multiple venoms (see further description in “Model Description”).

Second, we made simplifying assumptions about spatial structure, which may influence predator-prey interactions by giving rise to a geographic mosaic of coevolution (Thompson 2005). Instead, to focus on local-scale adaptation and isolate the effects of species interactions on venom evolution, we assumed that each species was represented by a single, panmictic population.

Third, we assume trait change to be proportional to additive genetic variance for each trait, which we treat as constant. This assumes negligible environmental variance in traits, and the assumption of constant trait variance does not hold true in the long term (without mutations to maintain genetic variation), under very strong selection (resulting in loss of genetic variation), and for small populations (which lack large initial genetic variation and are affected by genetic drift). Nonetheless, we believe that this approach provides a reasonable starting point to investigate the short-term (hundreds of generations) effects of change in ecological interactions (Abrams 2001). Further, the fitness peaks identified by the equilibrium scenarios should not be affected by this assumption. We also assume discrete time, logistic prey growth, for simplicity a type 1 functional response, for each the mesopredator and top predator, and assume that the top predator does not attack the bottom prey (i.e., distinct trophic levels). Relaxation of these assumptions may lead to oscillations and other non-equilibrium dynamics.

Finally, we only consider the case where a venomous consumer is attacked by a generalist apex predator, which has alternative prey that is not modeled explicitly. Further, for simplicity, we assume that this generalist predator has some level of venom resistance which does not evolve over the course of our simulations. Venom resistance in an apex predator may be static due to demographic constraints such as a lack of genetic variance for the trait (e.g., due to an invasion bottleneck), if the mesopredator is not a main component of its diet and venom is non-lethal (so that selection pressure for venom resistance is negligible) or if the trait for venom defense is linked with another trait under strong stabilizing selection pressure. Although there is evidence of a coevolving venom resistance in some predators of venom users (e.g., Voss 2013), this scenario is beyond the scope of our exploratory study and we leave this line of investigation for future work.

Model description

We use a three-trophic-level, discrete time Lotka-Volterra model of coevolutionary trait change to investigate the evolution of two traits (offensive venom and defensive venom) in a venomous consumer. The consumer feeds on a coevolving prey and is itself preyed upon by a generalist apex predator. We assume that because the apex predator has multiple food types, it is not tightly coupled with the mesopredator and does not evolve in response to changes in mesopredator venom.

Our model is an extension of the two-trophic-level model described by Northfield and Ives (2013), slightly altered to suit venomous animals. First, the venomous consumer has two traits that correspond with two types of venom: offensive and defensive venoms. The rate of successful attacks on the prey by the consumer increases with the value of the offensive trait, whereas the rate of successful attacks on the consumer by the apex predator decreases with the value of the defensive trait. Thus, the mesopredator venom trait used for prey capture and prey trait described here are each the inverse of the associated traits in the model presented by Northfield and Ives (2013), such that predation increases, rather than decreases with offensive venom production. Additionally, we consider a super-additive cost administered to consumer mortality associated with the production of multiple venoms.

The changes in population densities of the prey (R t ), venomous consumer (C t ), and apex predator (P t ) at time t are given by the following equations, described here in terms of functions in uppercase (Table 1) and constants in lowercase (Table 2):

Table 1 Description of functions (indicated by uppercase) used in model equations describing change in population densities and trait values for a three-trophic-level system involving a resource, a venomous consumer, and a generalist apex predator. Where subscripts are used, i may indicated the resource (R), venomous consumer (C), or apex predator (P) and j may indicate venom 1 (prey capture) or venom 2 (predator defense)
Full size table
Table 2 Description of constants (indicated by lowercase) used in model equations describing change in population densities and trait values for a three-trophic-level system involving a resource, a venomous consumer, and a generalist apex predator. Where subscripts are used, i may indicated the resource (R), venomous consumer (C), or apex predator (P) and j may indicate venom 1 (prey capture) or venom 2 (predator defense). The specified values for all simulations except where otherwise stated
Full size table
$$ {\mathrm{R}}_{\mathrm{t}+1}={\mathrm{R}}_{\mathrm{t}}{\mathrm{e}}^{{\mathrm{r}}_R-\frac{f_R}{V_{R1}}-\mathrm{Rt}-{\mathrm{Q}}_{\mathrm{C}}\left({V}_{\mathrm{C}1},{V}_{\mathrm{R}1}\right){\mathrm{C}}_{\mathrm{t}}} $$
(1)
$$ {\mathrm{C}}_{\mathrm{t}+1}={\mathrm{C}}_{\mathrm{t}}{\mathrm{e}}^{c_c{Q}_{\mathrm{C}}\left({V}_{\mathrm{C}1},{V}_{\mathrm{R}1}\right){\mathrm{R}}_{\mathrm{t}}-{\mathrm{M}}_{\mathrm{C}}\left({V}_{\mathrm{C}1},{V}_{\mathrm{C}2}\right)-{Q}_{\mathrm{P}}\left({V}_{\mathrm{C}1},{V}_{\mathrm{C}2}\right){\mathrm{P}}_{\mathrm{t}}} $$
(2)
$$ {\mathrm{P}}_{\mathrm{t}+1}={\mathrm{P}}_{\mathrm{t}}{\mathrm{e}}^{{\mathrm{r}}_{\mathrm{P}}\left(1-{\mathrm{P}}_{\mathrm{t}}/\mathrm{k}\right)+{\mathrm{c}}_{\mathrm{P}}{\mathrm{Q}}_{\mathrm{P}}\left({V}_{\mathrm{C}1},{V}_{\mathrm{C}2}\right){C}_{\mathrm{t}}-{M}_{\mathrm{P}}\left({V}_{\mathrm{P}1},{V}_{\mathrm{P}2}\right)} $$
(3)

Equation 1 describes population dynamics of the prey (resource) species, which exhibits simple population growth offset by mortality due to predation. The prey species may mitigate mortality due to consumer predation by investing in reducing its susceptibility to venom, V R1, which is balanced by a cost to reproduction and carrying capacity. Setting the carrying capacity equal to the growth rate (\( {r}_R-\frac{f_R}{V_{R1}} \)) allows a straightforward approach to model venom resistance costs (f R ) that simultaneously alter the equilibrium density and the rate at which it approaches that equilibrium. This might occur in scenarios where, for example, investing in the arms race includes a response that reduces venom susceptibility, resulting in slower development due to resources being allocated to defense and a lower carrying capacity due to more resources tied up in physiological mechanisms to reduce venom impact. We relax this assumption by setting the intrinsic growth rate separate from the carrying capacity in Appendix 2 in Online Supplementary Materials.

Equation 2 describes the population dynamics of the venomous consumer. Consumer fitness is balanced by three factors. First is predation on the prey species, where higher prey capture venom investment increases the consumption rate of prey. Second is mortality due to other density independent effects, which increases with investment in each venom due to resource allocation from other survivorship traits. Third is mortality due to predation by the apex predator, which declines with defensive venom investment and, depending on parameter values (see below), offensive venom investment.

A venom evolved for prey capture can also exhibit toxicity towards predators and other threats (Kuhn-Nentwig et al. 2011). Venom composition is typically considered to be driven by selection pressures for offense (Casewell et al. 2013), so we assume that the mesopredator can always produce some level of prey capture venom. Thus, we specify the effectiveness of the prey capture venom on predator deterrence s (component of Q P , Table 1), such that the attack rate is proportional to an exponential function of s, the mesopredator’s prey capture venom trait (V C1 ), and the top predator’s venom resistance (V P1 ), (\( {e}^{-{sV}_{\mathrm{C}1}{V}_{\mathrm{P}2}} \), Table 1). Further, we assume s ranges from 0 to 1. When s is set to 0, the prey capture venom exhibits no effectiveness for deterring predators, while setting s to 1 represents the case where the offensive venom has equal effectiveness against both prey and predators. Investing in each venom trait is associated with a cost administered to consumer mortality. An example of this for venom production would be increased metabolic costs associated with venom production (e.g., McCue 2006, Nisani et al. 2012), which may deplete resource reserves in times of adverse conditions. We also allow the cost administered to consumer mortality associated with the production of two venoms, f C1 and f C2 (i.e., mortality (M C ) increases with each, f C1 and f C2 (Table 1)), to be super-additive, such that investing in both venoms is particularly expensive. Thus, the total cost of venom production on mortality is equal to the cost of producing each venom and the multiplicative cost producing the two venom types together. This super-additive cost may represent additional costs associated with maintaining the physiological machinery or super-additive (multiplicative) metabolic costs associated with producing two venom types.

Equation 3 describes apex predator population dynamics. Because we modeled the apex predator as a generalist predator, its growth does not entirely depend on preying on the consumer species. Rather, it has an alternative growth rate r P (1-kP t ), with an associated intrinsic rate of increase (r P ) and carrying capacity (k) which are constants (Table 2). Further, since the apex predator’s fitness is not tightly linked with the defensive venom of the consumer, we assume that it does not coevolve with the consumer.

The changes in traits for the prey (i.e., prey susceptibility) and consumer (i.e., offensive and defensive venom investment) at each time step are determined by the population’s genetic variance and mean fitness, as well as the derivative of mean fitness with respect to the trait (Abrams 2001). If F i is the mean fitness for species i, and σ R1 , σ C1 , and σ C2 are additive genetic variance for the trait prey susceptibility V R1 , consumer offensive venom V C1 , and consumer defensive venom V C2 , respectively, then evolution of these traits is given by the following derivative equations:

$$ {V}_{\mathrm{R}{1}_{\mathrm{t}+1}}={V}_{\mathrm{R}{1}_{\mathrm{t}}}+\frac{1}{F_{\mathrm{R}}}\frac{\partial {F}_{\mathrm{R}}}{\partial {V}_{\mathrm{R}1}}{\sigma}_{\mathrm{R}}={V}_{\mathrm{R}{1}_t}+\left(\frac{f_{\mathrm{R}1}}{{V_{\mathrm{R}{1}_t}}^2}-{V}_{\mathrm{C}{1}_t}{a}_{\mathrm{C}}{C}_t{\mathrm{e}}^{-{V}_{\mathrm{C}{1}_t}{V}_{\mathrm{R}{1}_t}}\right){\sigma}_{\mathrm{R}} $$
(4)
$$ {V}_{\mathrm{C}{1}_{t+1}}={V}_{\mathrm{C}{1}_t}+\frac{1}{F_{\mathrm{C}}}\frac{\partial {F}_{\mathrm{C}}}{\partial {V}_{\mathrm{C}{1}_t}}{\sigma}_1 $$
$$ ={\mathrm{V}}_{\mathrm{C}{1}_{\mathrm{t}}}+\left({\mathrm{V}}_{{\mathrm{R}}_{\mathrm{t}}}{\mathrm{c}}_{\mathrm{C}}{\mathrm{a}}_{\mathrm{C}}{\mathrm{R}}_{\mathrm{t}}{\mathrm{e}}^{-{\mathrm{V}}_{\mathrm{C}{1}_{\mathrm{t}}}{\mathrm{V}}_{\mathrm{R}{1}_{\mathrm{t}}}}-{\mathrm{f}}_{\mathrm{C}1}\left(1+{\mathrm{f}}_{\mathrm{C}1+2}{\mathrm{V}}_{\mathrm{C}{2}_{\mathrm{t}}}\right)+\mathrm{s}{\mathrm{V}}_{\mathrm{P}{2}_{\mathrm{t}}}{\mathrm{Q}}_{\mathrm{P}}{\mathrm{P}}_{\mathrm{t}}\right){\upsigma}_{\mathrm{C}1} $$
(5)
$$ {V}_{\mathrm{C}{2}_{t+1}}={V}_{\mathrm{C}{2}_t}+\frac{1}{F_{\mathrm{C}}}\frac{\partial {F}_{\mathrm{C}}}{\partial {V}_{\mathrm{C}{2}_t}}{\sigma}_2={V}_{\mathrm{C}{2}_t}+\left(-{f}_{\mathrm{C}2}-{f}_{\mathrm{C}1+2}{V}_{\mathrm{C}{1}_t}+{V}_{\mathrm{P}{2}_t}{Q}_{\mathrm{P}}{P}_t\right){\sigma}_{\mathrm{C}2} $$
(6)

Analysis and results

To understand the effects of coevolution and the introduction of a third-trophic level on venom production, we performed two sets of simulations. In the first set of simulations, we evaluate the effect of prey coevolution on investment in a second, antipredator venom in the consumer (a mesopredator). In the second simulations, in addition to coevolution, we investigate the effect of additional costs associated with maintaining two venoms (f C1 + 2 ) and the effectiveness of the prey capture venom as a defense against the predator (s) on the final densities and traits of each species. We conducted all simulations in R (R Development Core Team 2014), and we provide an example of our code in the Online Supplementary Materials.

In both types of simulation, we tracked an ecological invasion by an apex predator, with and without prey coevolution taking place, and then followed the effects of venom evolution on population densities and trait values through time. These simulations began with the prey and consumer at eco-evolutionary equilibrium before the introduction of the generalist apex predator. In each simulation, we ran two general scenarios (with and without coevolution) and set parameters to ensure non-zero equilibriums in both trait values and population densities. For all simulations, constants were chosen in order to (a) be biologically plausible and (b) to highlight thresholds in eco-evolutionary dynamics in the emergence of a defensive venom. Extreme constant values would either lead to the extinction of one or more members of the tritrophic system (e.g., due to excessively high predation rates associated with large attack or conversion rates) or failure of the venom user to evolve a defensive venom (e.g., due to high venom costs), a scenario which was not of interest to our study. In particular for high venom costs associated with predator defense, the mesopredator either did not evolve venom for moderate top predator densities or went extinct because of the extreme costs and predation pressure. The sensitivity of the predator-invasion simulations to alternative initial conditions was tested by considering a wide range of starting population densities and trait values, but no alternative stable states were found. It was not possible to solve the system of equations analytically due to the complexity of the model.

Two scenarios were distinguished by whether or not coevolution occurred. In the first scenario, both the venomous consumer and prey species were assumed to have large additive genetic variances (σ ij  = 1, for species i with respect to venom j) following the invasion of the predator, promoting a high rate of evolution and thereby enabling the coevolutionary arms race to respond to this predatory pressure. In the second scenario, the prey was assumed to have no additive genetic variance in its venom resistance trait after reaching eco-evolutionary equilibrium. This provides a stark contrast to the coevolutionary scenario to better understand the effects of coevolution in the prey species on the trait values and population density of the venomous consumer. We tracked the changes in population densities and trait values as the simulations progressed and they reached equilibrium.

In scenario one, where the prey species can always evolve (σ R1  > 0), elevated mortality due to the invasion of a predator results in a small decline in the consumer density (Fig. 1a). Consumption on prey is reduced, permitting the prey species to become more susceptible to the consumer venom (Fig. 1c). The arms race between consumer and prey thus weakens, enabling the consumer to divert investment to its antipredator trait (Fig. 1c). The reduced arms race cost and reduced consumer pressure together allow prey densities to increase (Fig. 1a). The reduced defensive investment in the prey in turn allows the consumer species to invest less in offensive venom and invest more in its own defense (Fig. 1c).

Fig. 1
figure 1

Eco-evolutionary effects of an invading apex predator on the densities (a, b) and traits (c, d) for a three-trophic-level food web where the prey species either can (a, c) or cannot (b, d) coevolve. Values for the prey, consumer, and predator species are shown in blue, red, and black, respectively. Values for traits involved in the coevolutionary arms race (venom susceptibility and prey capture trait) between the prey and venomous consumer are solid. Values for the antipredator trait are dashed, and values for apex predator density are dotted. Prey mortality increases with prey trait values (susceptibility; solid blue line) and consumer traits associated with prey capture (i.e., offensive venom investment; solid red line)

Full size image

In the second scenario, the introduction of the apex predator does not diminish investment in prey capture venom by the consumer species (Fig. 1d) unless top-down selection is exceedingly strong, because the predator carrying capacity and predator attack and conversion rates are high enough to allow equilibrium predator density to approximately equal or exceed equilibrium consumer density (data not shown). Since there is no negative feedback loop mediated by prey trait change to diminish the trophic cascade that occurs when the apex predator is introduced, consumer density is lower than in the first (coevolutionary) scenario (Fig. 1b). Without the negative feedback loop weakening the consumer-prey arms race, low venom susceptibility is fixed in the prey species, and the consumer can no longer decrease investment in prey capture without overly compromising growth (Fig. 1d). When the apex predator carrying capacity is low, even when the cost of investing in two traits is low and there is low overlap of the prey capture and antipredator venom traits, the consumer does not shift investment from prey capture to defense. Instead, the predatory pressure induces a greater investment in prey capture venom, utilizing it as a dual-purpose weapon (Fig. 1d). Thus, the addition of an apex predator when prey coevolution is “switched off” gives rise to an “overkill” effect (Fig. 1d). That is, investment in the prey capture trait superficially appears to be excessive relative to the susceptibility of the prey to that trait.

We performed further analysis of equilibrium values, presented in full in Appendices 1 and 2 in Online Supplementary Materials. These analyses suggest that increases in the apex predator carrying capacity are generally correlated with increased investment in antipredator venom in the consumer, reduced investment in venom for prey capture, and increased prey susceptibility to the consumer’s venom (Appendix 1 Fig. A1D in Online Supplementary Materials). Furthermore, we found that increased prey carrying capacity leads to increased investment in both types of consumer venom and decreased prey susceptibility to consumer prey capture venom (Appendix 1 Fig. A1C in Online Supplementary Materials). Finally, the “overkill” effect is also found when prey intrinsic rate of increase is not affected by venom production and is low, which slows the rate of evolution in the bottom prey (Appendix 2 in Online Supplementary Materials).

To evaluate the effects of coevolution and the invasion of the apex predator on the evolution of a novel “antipredator” venom, in both scenarios, we also investigated the effect of additional costs associated with maintaining two venoms (ranging from f C1 + 2  = 0 to 0.003) and the effectiveness of the prey capture venom as a defense against the predator (ranging from s = 0 to 1). We find that evolution of a second, defensive venom is strongly influenced by the magnitude of f C1 + 2 and to a lesser extent by s. Both parameters are inversely proportional to defensive venom investment, so that investment is lower when either or both of f C1 + 2 and s are high. The values of f C1 + 2 and s also affected the arms race between the venomous consumer and its prey, such that low values resulted in higher investment in offensive venom and higher venom susceptibility in the prey (Fig. 2). Equilibrium prey and mesopredator population densities are more strongly influenced by venom specificity, such that a more general venom that affects both prey and predators (higher s) results in higher mesopredator and lower prey densities. Conversely, apex predator densities are lowest when the offensive venom is less general (lower s) and the cost of maintaining a second, specialized defensive venom is also low.

Fig. 2
figure 2

a–f Eco-evolutionary effects of an invading apex predator on the densities (a, b, c) and traits (d, e, f) of a three-trophic-level food web in which species interactions are modulated by the offensive and defensive venoms of a venomous consumer, and where the prey species can always coevolve. Here, we show how equilibrium values are modulated by the dual-venom fitness cost f C1 + 2 and venom specificity s. s has a more marked effect on prey and mesopredator densities, while f C1 + 2 exerts a stronger influence on apex predator densities. Investment in the antipredator defensive venom is more strongly determined by the magnitude of f C1 + 2 at lower levels of s and is greater when f C1 + 2 is lower. Investment is affected to a lesser extent by s, to which it is also inversely proportional. The outcome of the arms race between the venomous consumer and its prey is also affected by the values of f C1 + 2 and s. Low f C1 + 2 and low s both result in higher prey capture investment and higher susceptibility in the prey species. At higher levels of f C1 + 2 , prey venom susceptibility and offensive venom are more strongly determined by s

Full size image

Discussion

Our model analyses suggest that a coevolutionary interaction between a venomous consumer and its prey can alter the traits used in a consumer’s evolutionary response to a generalized apex predator. A coevolutionary arms race between a venomous consumer and its prey can facilitate a reciprocal “disarmament,” enabling the reallocation of investment towards the evolution of a novel antipredator venom. Furthermore, our results suggest that increased investment in offensive venom for use in predator defense can arise when prey capture venom is not very effective for predator deterrence and there is little genetic variation for venom susceptibility in the mesopredator’s prey population. In order to maximize the efficacy of the prey capture venom in such a context, investment in the prey capture venom can elevate beyond what appears necessary for prey capture alone in order to compensate for low efficacy against a predator. Consequently, when the pairwise consumer-prey interaction is viewed in isolation from the apex predator, the decoupled arms race may superficially suggest maladaptation by the venomous consumer due to overinvestment in venom potency relative to its prey, reminiscent of the “overkill” hypothesis (Mebs 2001).

This finding brings new insight to an interesting paradox in the use of venoms in defensive contexts. Currently, our understanding of the evolution and use of venoms in defensive contexts is limited, and it is generally held that venoms are primarily an offensive adaptation (Casewell et al. 2013). However, the “life-dinner” principle, proposed by Dawkins and Krebs (1979), points out the inherent asymmetry of predator-prey interactions. If a predator fails in a given interaction, it only loses a meal, but failure by the prey results in death or injury. Given the life-dinner principle, for venomous animals that use venom as their first line of defense, it may seem more intuitive that chemical warfare for defense should be the more essential investment for venom users, rather than investment in venom as a weapon for prey capture.

There is a range of evidence suggesting that venom may be more important for defense than previously thought (Casewell et al. 2013). Recent identification of separate offensive and defensive venoms in cone snails has suggested that venoms arising to deter predators are more common than previously thought (Dutertre et al. 2014). The use and evolution of defensive venom has been documented in other trophically venomous taxa, including scorpion prevenom and spitting cobras (Hayes et al. 2008; Inceoglu et al. 2003). Furthermore, histological evidence suggests that a wider range of venomous taxa may possess the necessary tissues for regionalized toxin production and the deployment of separate offensive and defensive venoms (Moran et al. 2013; Morgenstern and King 2013). In light of all this, and the putative ubiquity of prey coevolution in venomous animal systems, our findings may support recent calls to reconsider the traditional view that defense is generally not a significant selection pressure or function of venoms (Dutertre et al. 2014).

Our findings suggest that venom users engaged in arms races with prey species are more likely to evolve a distinct defensive venom in the presence of a non-coevolving apex predator, even at relatively low predator densities. We found that consumers are most likely to “over invest” in the venom normally used for prey capture in order to thwart predator attacks when prey venom resistance is inelastic in response to diminishing venom investment (e.g., due to genetic constraints). Evolution of venom susceptibility in the prey population can become constrained due to trade-offs with other competing selection pressures (Dietl and Kelley 2002). This is more likely to occur when the prey species must cope with a diverse community of predators, for example. In addition, when prey intrinsic rate of increase is not affected by venom production, we found that low prey intrinsic rates of increase also promote the “over-investment” in the prey capture venom (Appendix 2 in Online Supplementary Materials). This occurs because the prey population does not evolve increased susceptibility fast enough, and mesopredator is not able to invest less in prey capture venom. Abrams (1991) modeled a similar system, in which a specialist (but non-coevolving) predator was introduced to a coevolutionary consumer-prey arms race. When he assumed a negative correlation of prey capture and antipredator traits in the consumer (as we did in our modeling using fitness costs, such that investing in one trait requires a trade-off in the other), he also found that consumer response to prey was diminished (Abrams 1991). Our results also support earlier work proposing that additional interacting species will reduce the coevolutionary response in the members of an arms race (Vermeij 1982). Indeed, empirical studies of tritrophic chains in plant-herbivore systems have shown that antiherbivore defenses are stronger in patches where there is less predation on herbivores (Nersesian et al. 2011). The extent of a reciprocal “disarmament” between venomous consumer and its prey species is strongest when the costs of possessing multiple venoms are high and there is high functional overlap between these two traits (i.e., when the prey capture venom is effective in defensive contexts).

While venom resistance is less likely to coevolve in generalist predators of venom users due to a lower encounter rate (leading to a lower selection pressure) (Dietl and Kelley 2002), venom resistance is known in at least some predators of venomous animals including opossums (Jansa and Voss 2011), honey badgers, hedgehogs, mongooses, pigs (Drabeck et al. 2015), snake-eating snakes (Philpot and Stjernholm 1984), and grasshopper mice (Rowe et al. 2013). Since we only consider the scenario in which the apex predator does not coevolve for simplicity, our findings do not address the coevolving apex predator scenario and should be regarded as a preliminary understanding of the effects of coevolution in tritrophic chains involving venom users. Dercole et al. (2010) found that the trait dynamics of a coevolving prey, consumer, and apex predator are often chaotic, with the greatest irregularity arising in the apex predator trait. However, Dercole et al. (2010) only consider evolution at one locus and that coevolution occurs in a single trait for each member of the tritrophic chain. To our knowledge, no previous modeling studies have investigated coevolution between a consumer, its predator, and its prey, where consumer coevolution is mediated by separate prey capture (i.e., offensive venom) and antipredator traits (i.e., defensive venom). Thus, the effect of apex predator coevolution on venom evolution remains an open question for future work. Venomous consumers offer fascinating opportunities as model systems for the study of arms races, escalation, and coevolution. For example, pit vipers and opossums may be engaged in a coevolutionary arms race involving a unique, biochemically mediated process of role switching between consumer and prey (Voss 2013). Our results suggest that considering venom evolution in three trophic levels can produce novel insights that contribute to our understanding of patterns in venom evolution and the ecology of venom users which are not readily apparent from pairwise interactions.