Inducible offences affect predator–prey interactions and life-history plasticity in both predators and prey

Authors


Summary

  1. Phenotypic plasticity can have strong impacts on predator–prey interactions. Although much work has examined the effects of inducible defences, less understood is how inducible offences in predators affect predator–prey interactions and predator and prey phenotypes.
  2. Here, we examine the impacts of an inducible offence on the interactions and life histories of a cohort of predatory Hynobius retardatus salamander larvae and their prey, Rana pirica tadpoles. We examined larval (duration, survival) and post-metamorphic (size) traits of both species after manipulating the presence/absence of tadpoles and salamanders with offensive (broadened gape width) or non-offensive phenotypes in pond enclosures.
  3. Offensive phenotype salamanders reduced tadpole survival and metamorph emergence by 58% compared to tadpole-only treatments, and by over 30% compared to non-offensive phenotypes. Average time to metamorphosis of frogs was delayed by 30% in the presence of salamanders, although this was independent of salamander phenotype. Thus, offensive phenotype salamanders reduced the number of tadpoles remaining in the pond over time by reducing tadpole survival, not by altering patterns of metamorph emergence. Offensive phenotypes also caused tadpoles to metamorphose 19% larger than no salamander treatments and 6% larger than non-offensive phenotype treatments. Pooled across salamander treatments, tadpoles caused salamanders to reach metamorphosis faster and larger. Moreover, in the presence of tadpoles, offensive phenotype salamanders metamorphosed 25% faster and 5% larger than non-offensive phenotype salamanders, but in their absence, neither their size nor larval period differed from non-offensive phenotype individuals.
  4. To our knowledge, this study is the first to demonstrate that inducible offences in predators can have strong impacts on predator and prey phenotypes across multiple life stages. Since early metamorphosis at a larger size has potential fitness advantages, the impacts of offensive phenotypes on frog and salamander life histories likely have significant consequences for individuals and populations. Furthermore, increased predation on tadpoles likely causes offensive phenotype individuals to have strong impacts on pond communities. Future studies should examine the fitness consequences of morphological and life-history plasticity across multiple life stages and should address the population and community level consequences of offensive phenotypes.

Introduction

Phenotypic plasticity in behavioural, morphological, physiological and life-history traits due to predator–prey interactions are well-studied examples of adaptive responses to environmental variation (Tollrian & Harvell 1999; Kopp & Tollrian 2003a; Benard 2004; Kishida, Mizuta & Nishimura 2006). The expression of phenotypes in response to predation risk is called inducible defences (Tollrian & Harvell 1999), while plastic traits that allow predators to more effectively consume prey are called inducible offences (Padilla 2001; Michimae & Wakahara 2002; Kopp & Tollrian 2003a).

Inducible defences have important consequences for predator–prey interactions and have been the focus of much research (Tollrian & Harvell 1999; Benard 2004; Miner et al. 2005; Kishida et al. 2010), while inducible offences have received considerably less attention despite having potentially equal importance (Lima 2002; Kopp & Tollrian 2003a; Pfennig, Rice & Martin 2006; Kishida, Trussell & Nishimura 2009; Martin & Pfennig 2009; Mougi, Kishida & Iwasa 2011). For example, inducible offences often result in trophic polymorphisms, in which different morphological, life-history and/or behavioural phenotypes within a population consume different resources, which can affect the evolution and ecology of predator–prey interactions (Smith & Skúlason 1996; Pfennig, Rice & Martin 2006; Pfennig et al. 2010; Bolnick et al. 2011). Moreover, inducible offences can increase the expression of inducible defences in prey, which can also affect predator–prey interactions, although this topic is just beginning to be explored (Kopp & Tollrian 2003b; Kishida, Mizuta & Nishimura 2006; Edgell & Rochette 2007; Mougi, Kishida & Iwasa 2011). Thus, it is necessary to evaluate how inducible offensive traits affect predator–prey interactions and plasticity in both predators and prey.

In organisms with complex life cycles, predator–prey interactions can drive plasticity in the size and timing of transitions across life stages (e.g. metamorphosis) for both predators and prey. For example, predators can change the density and phenotype of prey in one life stage, as well as the transition rate between ontogenetic niches, which can all have important consequences for prey across multiple life stages and habitats (Benard 2004; Vonesh 2005; McCoy, Barfield & Holt 2009; Orrock et al. 2010). Although most studies have focused on the responses of prey to predation risk, less understood is how predaceous phenotypes affect life-history plasticity in predators (but see Pfennig 1992; Michimae & Wakahara 2002). Because larval energy intake can be one of the main determinants of size and age at metamorphosis (Rose 2005), prey identity and/or availability may also drive life-history plasticity in predators (O'Laughlin & Harris 2000; Hammill & Beckerman 2010; Michimae 2011; Davenport & Chalcraft 2012; Stoks & Córdoba-Aguilar 2012), which can depend on the predaceous phenotype and the ability to exploit different prey items (Michimae & Wakahara 2002). For example, cannibalism associated with an offensive phenotype can cause accelerated larval development and reduce the size at metamorphosis (Michimae & Wakahara 2002; Michimae 2011). To our knowledge, no study has examined how induced offensive traits in predators can affect life-history plasticity in both predators and prey and their consequences for predator–prey interactions.

In this study, we examine the impacts of inducible offences on the survival and life histories of a cohort of predatory Hynobius retardatus (Dunn) salamander larvae and their prey, Rana pirica (Matsui) tadpoles. These salamander larvae exhibit highly effective morphological and behavioural offences in which they develop a broadened gape early in ontogeny and increase their activity when R. pirica frog tadpoles are abundant, allowing them to encounter and swallow tadpoles more easily (Michimae & Wakahara 2002; Kishida, Trussell & Nishimura 2009; Takatsu & Kishida 2013). We evaluate the effects of these inducible offences on the larval (duration, survival) and post-metamorphic (size) traits of H. retardatus and R. pirica when alone and when salamander larvae have offensive or non-offensive phenotypes.

Methods

Rana pirica frogs and H. retardatus salamanders lay their eggs in small transient forest ponds formed by melting snow during spring (April–May) in Hokkaido, Japan, and they often coexist at high densities (Kishida & Nishimura 2006). Salamander larvae are carnivorous and consume aquatic invertebrates in addition to R. pirica tadpoles and conspecifics, while tadpoles consume algae and detritus. During early spring, invertebrates are often at low abundances so R. pirica tadpoles can be the primary prey of salamander larvae, thus creating strong predator–prey interactions between both species.

Animal collection and phenotypic induction of salamander larvae

In mid-May 2012, ten clutches of R. pirica eggs and five clutches of H. retardatus eggs were collected from several natural ponds. Eggs of each species were separated and divided arbitrarily and placed in tanks (33·4 cm × 20 cm × 10 cm; 200–300 eggs per tank) filled with 2 L of aged tap water maintained indoors at 16 °C with a natural light–dark regime. Tadpoles began to hatch in late May, and salamander larvae hatched in early June, at which point 200 salamander hatchlings were individually placed into small plastic cases (84 mm × 57 mm × 44 mm high) filled with 100 mL of aged tap water on 9 June 2012. We then put 10 small tadpoles (c. 200 mg) into each of half of the cases (i.e. 100 cases) as induction agents to induce the offensive phenotype of salamander larvae. The other 100 salamander hatchlings were kept in the same cases but without tadpoles to maintain the non-offensive phenotype. Because the tadpoles used as induction agents were consumed, we added tadpoles every two days to maintain a homogeneous induction environment. In both induction treatments, we also added tubifex worms ad libitum every two days as food for the salamanders, and the water from all cases was exchanged every two days. Two weeks after starting the induction treatment, most salamanders had expressed the expected phenotypes, and we added five tadpoles into each case of the non-offensive treatment to habituate the salamander larvae to tadpoles. This allowed us to focus on the effects of the morphological differences between the salamander phenotypes (e.g. not behavioural differences due to a novel environment with tadpoles). After three days, we scanned all salamander larvae ventrally and measured their gape width and body length (i.e. heart vent). We used this morphological data to select offensive and non-offensive phenotype larvae with similar body sizes. Of the measured individuals, 24 offensive and 42 non-offensive salamander larvae were randomly selected and used in the field experiment described below. Because salamander larvae rarely exhibit offensive phenotypes later in ontogeny even if they consume tadpoles, this allows us to compare the impacts of the offensive and non-offensive phenotypes on tadpole and salamander populations in the field experiment.

Field experiment

The experiment was conducted in an outdoor artificial pond (12 × 15 m) in the Teshio experimental forest of Hokkaido University (45°01′77·65″N, 142°01′47·71″E). This pond has no canopy cover and a bottom composed of soil and small rocks at approximately 40 cm water depth. Wild individuals of both amphibian species deposit their eggs in this pond in mid-May, and their larvae coexist and interact until late summer (August–September). We placed 42 rectangular (60 cm × 60 cm × 60 cm high) enclosures with PVC framing covered by nylon mesh (1 mm openings) on all sides, in the pond on 15 June 2012. Small aquatic insects such as chironomid larvae can pass through the mesh to provide prey for salamander larvae, and periphyton grows on the mesh, providing resources for tadpoles.

We used a spatially blocked design in which one enclosure representing each of six treatments was each assigned to seven different blocks. The six treatments included: (i) 100 tadpoles only (tadpole treatment); (ii) two offensive salamander larvae only (offensive salamander treatment); (iii) two non-offensive salamander larvae only (non-offensive salamander treatment); (iv) 100 tadpoles and two offensive salamander larvae (tadpole and offensive salamander treatment); and (v) 100 tadpoles and two non-offensive salamander larvae (tadpole and non-offensive salamander treatment). The additional sixth treatment (tadpole and non-offensive salamander removal treatment) was similar to the fifth treatment, in which 100 tadpoles and two non-offensive salamander larvae were assigned at the beginning of the experiment. However, because we expected offensive phenotype salamanders to reach metamorphosis faster when in the presence of tadpoles, we manually removed salamander larvae from this additional treatment when salamanders from the tadpole and offensive salamander treatment of the same block metamorphosed (i.e. when we found one offensive salamander metamorphosed, we removed one randomly selected non-offensive salamander larva from this treatment in the same block). This treatment was necessary to confirm the effects of reduced timing to metamorphosis in salamanders on tadpole response variables. Experimental densities of both organisms were chosen to reflect observed natural conditions (O. Kishida, unpublished data). Tadpoles were randomly assigned to appropriate enclosures on 25 June 2012, and salamander larvae were randomly assigned to their respective treatments the following day, beginning the experiment.

Due to logistical constraints about accessing the field site and checking enclosures, we counted the number of metamorphs and surviving tadpoles and salamander larvae at days 11, 36, 50 and 60 (i.e. all surviving amphibians metamorphosed by day 60). We also collected metamorphs on days 15, 22, 23, 25, 30, 32, 39, 43, 46, 53 and 57 of the experiment. We assumed time at metamorphosis was midway between the day in which the individuals were collected and the previous time the enclosure was checked. Timing of metamorphosis of frogs is defined as the time at which the tail is completely absorbed, whereas the timing of metamorphosis of salamanders is defined as the time at which the tail fin and external gills are completely absorbed. The ventral side of metamorphosed frogs and salamanders was digitally scanned and measured for snout–vent length to determine size at metamorphosis. The number of dead tadpoles (Dt) during a given census time (t) was calculated as Dt = (Nt–1 – Nt – Mt) where Nt–1 is the number of surviving tadpoles at the previous census, Nt is the number of tadpoles surviving at the given census period and Mt is the number of tadpoles that had metamorphosed since the previous census.

Statistical analyses

All statistical analyses were conducted using r, version 2.11.0 (R Development Core Team, Vienna, Austria) or jmp, version 8 (SAS Institute, Tokyo, Japan). Our preliminary analyses found that removal of non-offensive salamanders in the ‘tadpole and non-offensive salamander removal treatment’ had no significant effects on tadpole response variables (see Appendix S1). Hence, the data from treatments (5) tadpole and non-offensive salamander and (6) tadpole and non-offensive salamander removal were pooled before conducting the following statistical analyses. We first used one-way anovas to examine differences in body length and gape width between the offensive and non-offensive salamander phenotypes at the beginning of the experiment. Average timing and size at metamorphosis of tadpoles and salamanders were compared among the three tadpole-containing and four salamander-containing treatments, respectively, using linear mixed models in which block was treated as a random effect, and response variables were natural log-transformed to meet assumptions of normality. The effects of salamander phenotype on salamander response variables were examined using a priori planned contrasts crossing salamander phenotype with tadpole presence/absence using two-way anovas. Similarly, we also examined the impact of salamander phenotype (i.e. tadpole-only treatments were excluded) on tadpole size and time to metamorphosis using anovas. At the end of the experiment, we examined the effects of treatment and time on the number of dead tadpoles, the number of total metamorphs, the proportion of tadpoles reaching metamorphosis during each time period and the number of tadpoles remaining in the pond using generalized linear mixed models (glmmPQL) with a Poisson (number of dead tadpoles and metamorphs) or quasibinomial (proportion) error distribution in which repeated sampling of tanks nested within spatial blocks were treated as random factors. This was done to account for non-independence of repeated tank measures and overdispersion (Pinheiro & Bates 2000). We also examined potential interactions between salamander phenotype and time on these response variables. Because all surviving tadpoles reached metamorphosis, we included the proportion (i.e. the number of metamorphs collected at each time point divided by the total number of metamorphs collected from each enclosure) of metamorphs for each metamorph sampling date, allowing us to examine whether salamanders and induced phenotypes affected patterns of emergence independent of their impacts on mortality. Post hoc mean comparisons were performed using Tukey's HSD.

Results

At the beginning of the experiment, salamander larvae body length did not differ between the offensive [13·72 ± 0·12 mm (mean ± SD), n = 32] and non-offensive induction treatments (13·59 ± 0·65, n = 56; F1,86 = 0·86, = 0·354). However, the gape width of larvae from the offensive induction treatment (9·84 ± 0·67) was nearly 1·5 times wider than those from non-offensive treatments (6·73 ± 0·29; F1,86 = 911·85, < 0·0001). Salamander mortality was low, and only one salamander larva died in each of three different treatments. Time to metamorphosis for salamanders was affected by treatment (F3,18 = 37·5, P < 0·0001), the presence of tadpoles (F1,18 = 93·3, < 0·0001) and its interaction with salamander phenotype (F1,18 = 15·5, = 0·001), while phenotype itself was marginal (F1,18 = 3·76, P = 0·068; Fig. 1a). Tadpoles reduced time to metamorphosis of salamanders by 35% compared to no tadpole treatments and offensive phenotype salamanders emerged 24·7% faster than the non-offensive phenotype when in the presence of tadpoles. In the absence of tadpoles, the opposite trend was observed and offensive phenotype salamanders took 9·7% longer to reach metamorphosis, although this was not statistically significant (Tukey's HSD: P = 0·49). Overall, we found that size at metamorphosis of salamanders differed among treatments (F3,18 = 26·01, < 0·0001). Additional tests showed that the presence of tadpoles (F1,18 = 68·56, < 0·0001) and salamander phenotype (F1,18 = 9·32, = 0·0068), but not their interaction (F1,18 = 0·15, = 0·70) affected size at metamorphosis of salamanders (Fig. 1b). Pooled across treatments, tadpoles increased the size of salamanders by 14·3% compared to no tadpole treatments and offensive phenotype salamanders were 5% larger than the non-offensive phenotype.

Figure 1.

Effects of predatory Hynobius retardatus salamander phenotypes on the (a) average time to metamorphosis, (b) average snout–vent length (SVL) at metamorphosis of salamanders. Open and filled circles represent absence and presence of Rana pirica tadpoles, respectively. Means not sharing the same letter are significantly different (Tukey's HSD; P ≤ 0·001). Error bars denote 95% confidence intervals.

The number of dead tadpoles was affected by treatment (F2,19 = 90·79, < 0·0001), time (F4,100 = 30·58, < 0·0001) and their interaction (F8,100 = 14·73, < 0·0001; Fig. 2a). When the tadpole-only treatment was excluded, salamander phenotype, time and their interaction also affected the number of dead tadpoles (Fig. 2a; Table 1). By the end of the experiment, tadpole mortality in the offensive salamander treatment (62·4 ± 0·4) was 1·6 times greater than the non-offensive treatment (38·2 ± 0·3) and 5·7 times as large as the tadpole-only treatment (10·9 ± 0·2; Fig. 2b). The number of metamorphosed frogs was affected by the same factors similarly, as all surviving tadpoles reached metamorphosis (treatment: F2,19 = 46·95, < 0·0001; time: F4,100 = 68·97, < 0·0001; interaction: F8,100 = 25·75, < 0·0001; Table 1). The proportion of metamorphosed frogs showed a different pattern in which treatment (F2,19 = 0·328, = 0·72) and time (F1,389 = 2·01, = 0·158) were not significant but their interaction was (F2,389 = 7·04, = 0·001; Fig. 2c). When tadpole-only treatments were excluded, however, no factors were significant (Table 1). The number of tadpoles remaining in pond enclosures was affected by treatment (F2,19 = 89·12, < 0·0001), time (F4,100 = 563·5, P < 0·0001) and their interaction (F8,100 = 18·65, < 0·0001; Fig. 2d). When the tadpole-only treatments were excluded, salamander phenotype, time and their interaction also affected the number of tadpoles remaining in the enclosures (Table 1). In particular, the average number of tadpoles remaining over time was less in offensive phenotype treatments as compared to the non-offensive and tadpole-only treatments (Fig. 2d). Average time to metamorphosis for frogs was also affected by treatment, as the presence of salamanders delayed metamorphosis by 30% compared to no salamander treatments, although there were no differences between salamander phenotypes (Fig. 2e; Table 1). Average size at metamorphosis of frogs was also affected by treatment and salamander phenotype (Fig. 2f; Table 1), as frogs from offensive phenotype treatments were 5·6% larger than frogs from the non-offensive treatments and 18·6% larger than frogs from the no salamander treatments.

Figure 2.

Effects of predatory Hynobius retardatus salamander phenotypes on the (a) the average number of dead tadpoles over time, (b) average total mortality, (c) proportion of metamorphosed frogs over time, (d) average number of tadpoles remaining in the pond over time, (e) average time to metamorphosis, (f) average snout–vent length (SVL) at metamorphosis of Rana pirica. No, non-offensive, and offensive labels represent the tadpole-only treatment, tadpole and non-offensive phenotype salamander treatment, and the tadpole and offensive phenotype salamander treatments, respectively. Means not sharing the same letter are significantly different (Tukey's HSD; P ≤ 0·001). Error bars denote 95% confidence intervals.

Table 1. Results of analyses examining the effects of salamander phenotype, time and their interaction on the mortality, total number of metamorphs, the proportion of emerging metamorphs, the number of tadpoles remaining in pond enclosures, as well as the effects of treatment and salamander phenotype on time to and size (SVL) at metamorphosis of Rana pirica
Response variableFixed effectd.f. F P
Tadpole mortalityPhenotype1, 1390·52<0·001
Time4, 7626·88<0·001
Phenotype*Time4, 7625·45<0·001
Number of metamorphsPhenotype1, 1323·54<0·001
Time4, 7669·83<0·001
Phenotype*Time4, 763·54<0·001
Proportion of metamorphsPhenotype1, 130·010·99
Time1, 2920·240·62
Phenotype*Time1, 2920·230·63
Number of tadpoles remainingPhenotype1, 13119·38<0·001
Time4, 76449·3<0·001
Phenotype*Time4, 762·950·025
Time to metamorphosisTreatment2, 1946·45<0·001
Phenotype1, 131·010·33
SVLTreatment2, 1942·9<0·001
Phenotype1, 1326·7<0·001

Discussion

General theory regarding phenotypic plasticity states that the evolution of inducible offences, similar to inducible defences, requires a functional trade-off between the costs and benefits of the induced trait (Tollrian & Harvell 1999; Kopp & Tollrian 2003a). If there are no benefits to the trait, then it should not be expressed, and if there are no costs, then it should be expressed permanently. Our study is in general agreement with this framework. Offensive phenotype induction allowed salamanders to consume more tadpoles, which increased their growth and development. This decreased their larval period and increased their size at metamorphosis as compared to non-offensive phenotypes. Overall, tadpoles decreased time to metamorphosis for salamanders by 35% (16 days), but this effect was stronger for offensive phenotype salamanders, as they reached metamorphosis nearly 25% faster (8·4 days) than non-offensive phenotypes. In the absence of tadpoles, however, offensive phenotype individuals took nearly 10% (4·25 days) longer than non-offensive phenotypes to reach metamorphosis, although this difference was not statistically significant.

Although our work does not demonstrate costs of offensive phenotypes per se, we show that in the absence of tadpoles, their expression did not increase the development or growth of salamanders, suggesting salamanders received no fitness benefit from offensive trait expression in the absence of the inducing prey agent (i.e. tadpoles). Previous work has shown similar results in which costs of offensive phenotypes occur under different conditions of prey availability and/or identity. For example, predatory ciliates increase their gape size in response to large prey items in order to more effectively consume them. In the presence of small prey, however, they have lower feeding rates and lower maximal population growth than non-offensive phenotypes (Kopp & Tollrian 2003a). Similarly, some species that exhibit resource polymorphisms suffer costs such as decreased growth and foraging efficiencies when feeding on alternative prey (i.e. prey that did not induce the offensive phenotype; Ehlinger 1990; Swanson et al. 2003; Parsons & Robinson 2007; Martin & Pfennig 2009). This suggests that the costs of offensive phenotype expression may increase when prey availability and/or identity varies over time.

In this study, we investigated the impacts of inducible predator phenotypes on the abundance and life-history traits of predator and prey. Variation in the size and timing of metamorphosis is a common feature of organisms with complex life histories (Benard 2004) and can be an important driver of population and community dynamics (e.g. Kishida et al. 2010). Salamander phenotype had strong impacts on tadpole survival and life-history traits. After accounting for background mortality, offensive phenotype salamanders reduced tadpole survival and the number of metamorphs by 88% compared to non-offensive phenotypes. The earlier metamorphosis of offensive phenotype salamanders could release tadpoles from predation; however, we saw no differences in these traits between the non-offensive salamander removal and non-removal treatments (see Appendix S1). Thus, the different impacts of salamander phenotypes on tadpole mortality and metamorphosis operated early in the experiment and were due to differences in predation rates. Salamanders also affected the timing of metamorphosis for R. pirica, in which the presence of salamanders increased their larval period by 30% (8 days) independent of phenotype. Offensive H. retardatus phenotypes also caused tadpoles to metamorphose 5% larger than when in the presence of the non-offensive phenotype, and nearly 20% larger than those from tadpole-only enclosures. Although most models predict earlier and smaller sizes at metamorphosis due to predation risks, we discuss several mechanisms to explain our results including the role of basal resources, plasticity and size refugia (Benard 2004).

Increased predation by offensive phenotype salamanders likely increased per capita resource availability and may have indirectly increased the growth of surviving tadpoles (i.e. thinning effects; Relyea 2002; Vonesh 2005). Similarly, the presence of salamanders causes R. pirica tadpoles to reduce their activity (Kishida, Trussell & Nishimura 2009), which could also increase resources, resulting in a net positive effect of predators on prey growth, despite the fact that early ontogenetic development is likely to be reduced (Peacor 2002). Once tadpoles reach a size threshold from predation, however, their foraging rates may increase in order to make up for reduced growth early in ontogeny, which would result in a longer time to metamorphosis but at a larger size (Van Buskirk & Schmidt 2000; Benard 2004), as was observed in our experiment. In addition, salamander larvae selectively consume smaller tadpoles due to their gape limitation (Takatsu & Kishida 2013), and R. pirica tadpoles express morphological defences in response to predatory salamanders (Kishida & Nishimura 2004), which increases their larval period (O. Kishida, unpublished data). It is likely several of these non-exclusive mechanisms are contributing to the effects of salamander phenotypes on R. pirica life-history traits.

Our study provides evidence that inducible offences modify the life histories of both predator and prey due to the strengthening of predator–prey interactions. Regardless of the specific mechanisms, the effects of offensive salamander phenotypes on frog and salamander life histories likely have important consequences for individuals and populations because smaller size at metamorphosis is associated with reduced adult growth and survivorship and reduced reproductive success in both frogs and salamanders (Smith 1987; Semlitsch, Scott & Pechmann 1988; Scott 1994; Altwegg & Reyer 2003). Similarly, later metamorphosing individuals can have reduced adult survival and growth (Altwegg & Reyer 2003). For organisms with complex life cycles, increased larval development can be highly adaptive in temporary pond environments by reducing mortality due to desiccation and larval predators (Semlitsch & Wilbur 1988; Pfennig 1990, 1992; Vonesh 2005). Like many species that breed in lentic habitats, the breeding ponds used by our focal species contain predatory dragonfly larvae and are highly ephemeral and can dry before tadpoles or salamanders reach metamorphosis, suggesting tadpoles may suffer increased mortality due to an increase in the larval period when in the presence of salamanders. However, this cost may be offset by the benefit of metamorphosing at a larger size. Likewise, larval salamanders can receive a pronounced fitness benefit from tadpoles and the induction of the offensive phenotype by reaching metamorphosis faster and at a larger size. Although our experiment cannot address the impacts of offensive H. retardatus phenotypes on frog and salamander population dynamics, they are likely important. In general, our work suggests that for organisms with complex life cycles, the impacts of offensive phenotypes on predator and prey populations will be mediated not just by changes in survival, but by changes in their life-history traits which will also depend on environmental conditions in both the larval and adult environments (Altwegg & Reyer 2003).

We suggest that spatial and temporal variation in the population densities of R. pirica and H. retardatus may be driven by variation in factors affecting the expression of offensive phenotypes of H. retardatus larvae (Michimae & Wakahara 2002; Michimae 2006; Kishida, Trussell & Nishimura 2009; Michimae et al. 2009). For example, H. retardatus salamander larvae exhibit offensive phenotypes more frequently when in the presence of small tadpoles than with large tadpoles (Michimae & Wakahara 2002), and less frequently when in the presence of predatory dragonfly larvae (Kishida, Trussell & Nishimura 2009; Kishida et al. 2011). Hence, the community and the size structure of predators and prey may affect predator–prey interactions by influencing the expression of offensive phenotypes. Moreover, our experiment also suggests that for organisms with complex life cycles, the expression of offensive phenotypes and the community context likely have important consequences for both aquatic and terrestrial communities by affecting the fitness of larval and adult individuals of both predators and prey. For example, offensive phenotype treatments had fewer tadpoles remaining in the pond enclosures over time, which may impact aquatic communities because larval anurans can affect primary production, nitrogen cycling and other consumers (Seale 1980; Whiles et al. 2006). In addition, salamanders and offensive phenotypes likely decrease the export of nutrients from ponds, and by reducing the number of adult anurans, may have additional consequences for terrestrial food webs (Whiles et al. 2006; McCoy, Barfield & Holt 2009). Thus, it is likely that the expression of offensive phenotypes in organisms with complex life cycles can impact populations and communities across multiple life stages and habitats.

Theoretical studies suggest that inducible offences can stabilize population and evolutionary dynamics by affecting the fitness of predators and prey, which is often mediated by the prey's expression of inducible defences (Kondoh 2003; Mougi & Kishida 2009; Mougi, Kishida & Iwasa 2011). Our study provides empirical support for such works as we demonstrate inducible offences, and their impacts on prey phenotypes (i.e. inducible defenses) can alter key fitness components (e.g. survival, size and timing of metamorphosis) of both predators and prey. Moreover, inducible offences initially resulted in stronger predator–prey interactions between salamanders and tadpoles as more tadpoles were killed in offensive phenotype treatments. Over time, however, the interaction strength between the species declined and none of our treatments resulted in the extinction of either species in any enclosure, suggesting their interactions are stable over ecological time. Inducible offences are prevalent among a variety of taxa (e.g. Smith & Palmer 1994; Padilla 2001; Kopp & Tollrian 2003a) and are often countered with inducible defenses in prey (reviewed in Kishida et al. 2010), suggesting that reciprocal phenotypic plasticity may be a common feature of predator–prey interactions that leads to their stability (Mougi & Kishida 2009; Mougi, Kishida & Iwasa 2011). Future work should examine the relationships among environmental conditions and the impacts of inducible offences on predator and prey fitness, in order to deepen our understanding of the role of phenotypic plasticity in evolutionary and ecological dynamics.

Acknowledgements

We thank Kunio Takatsu, Tomoko Sato, Emiko Taniguchi, Shiho Chiba and all of members of Teshio Experimental forest for their unwavering support for our research. We are also grateful to the late Toshiaki Oiwa for his tremendous help in preparing our experiment. We also thank D. Morris and three anonymous reviewers for comments that helped improve the manuscript. This work was supported by a Grant-in-Aid for a Research Fellow of the Japan Society for the Promotion of Science (no. 24370004, 23247004) to O. Kishida and (no. 70447069) to H. Michimae.

Ancillary