Specific induced responses to different predator species in anuran larvae


Van Buskirk, Institute of Zoology, University of Zürich, CH-8057 Zürich, Switzerland. Tel.: +41 1635 4983; fax: +41 1635 6821; e-mail: jvb@zool.unizh.ch


Models of defence against multiple enemies predict that specialized responses to each enemy should evolve only under restrictive conditions. Nevertheless, tadpoles of Rana temporaria can differentiate among several predator species. Small tadpoles used a refuge when Notonecta backswimmers were in the pond, but showed a weaker hiding response to adult Triturus alpestris newts and no response to aeshnid dragonfly larvae (Aeshna and Anax). All predators caused a decline in feeding and swimming activity. Large tadpoles reserved the strongest behavioural response for dragonflies, while Triturus caused no response. The shift during development suggests that tadpoles distinguished among predators, rather than exhibiting a graded dosage response to a single cue associated with predation. Information on habitat distributions of predators suggests that they are regularly encountered, which would facilitate evolution of adaptive behavioural responses. Morphological responses to all predators were similar, perhaps because similar morphologies defend against all four predators. The evolutionary maintenance of specialized responses to multiple predators may be possible because adaptive responses do not conflict and the predators themselves do not interact strongly.


Two kinds of theory highlight potential impediments to the evolution of specialized defences against natural enemies such as herbivores or predators. One set of models, exploring conditions under which phenotypic plasticity can evolve, states that optimal reaction norms are difficult to achieve if some environments are encountered infrequently (Via & Lande, 1985; van Tienderen, 1991). If different species of enemy can be thought of as representing distinct environments for their victims, then these models suggest that specific responses to each enemy cannot evolve if there are many different enemies and each is encountered only rarely. A similar prediction emerges from a second group of models exploring reciprocal coevolutionary responses of consumers and their victims. In this case, specific responses to multiple enemies are unlikely if the enemies have interactive effects on the fitness of victims, either because the enemies interact directly with one another or because an adaptive response to one enemy renders the prey more vulnerable to another (Hougen-Eitzman & Rausher, 1994). These models also show that genetic correlations among defences against multiple enemies will further hinder the evolution of specific defences, because selection imposed on resistance to one enemy will cause correlated responses in resistance to other enemies (Rausher, 1992). The conditions specified by both kinds of theory may be common in nature, because many natural enemies are encountered infrequently and fitness effects of multiple enemies are often correlated (Strauss, 1991; Sih et al., 1998). The conclusion therefore is that specific responses to different enemies should be difficult to evolve and maintain.

Contrary to this expectation, many plants and animals show different defences against different kinds of natural enemies, involving changes in morphology, behaviour, and chemistry. The freshwater snail Physa heterostropha, for example, possesses an elongated shell when it occurs with a crayfish predator, but a more rotund shell when it occurs with sunfish, and both responses improve resistance to attacks from the predator species that induce them (DeWitt, 1998; DeWitt et al., 2000). Many other examples exist (Agrawal & Karban, 1999; Tollrian & Dodson, 1999).

The widespread occurrence of specific responses to multiple predators suggests two possible explanations. On the one hand, the conditions specified by theory may often be fulfilled: enemies may be encountered with high frequency and may only rarely interact in their fitness effects on victims. Alternatively, the existence of specific responses may not imply that victims distinguish among enemies and produce adaptive responses to each one. Different predator species might produce different intensities of a single cue that prey use to detect enemies, and different levels of response may simply emerge from a graded dosage response to the cue. Distinguishing among these possibilities will help evaluate how often victims evolve specialized responses to multiple enemies, in spite of the possible impediments highlighted by theory.

Here I address several questions about distinct defensive responses that anuran larvae exhibit to different predators. First, can the differences be interpreted as arizing from a dosage response to a single cue, leading to a generalized antipredator response that is expressed in different degrees with different predators? If not, can the differences be related to the prey capture behaviour of the predators, indicating that the responses are appropriate for avoiding or escaping each predator, as expected if specialized defences have evolved? Finally, I ask whether tadpoles commonly encounter the different predators in natural ponds, as required by models for the evolution of adaptive plasticity in multiple environments.


Measuring phenotypic responses to predators

I reared tadpoles of the common frog (Rana temporaria) in artificial ponds containing either no predators or one of four predator species confined within cages, and sampled the behaviour and morphology of tadpoles on 2–3 occasions during the larval period. This approach allowed me to measure phenotypes under fairly natural conditions, while avoiding the confounding problem of tadpole mortality from predation, when predators are free.

The ponds were 1.4 m2 fibreglass tanks, filled to a depth of 40 cm (550 L volume) and placed outdoors in a field on the campus of the University of Zürich, Switzerland. To establish self-sustaining and seminatural pond communities within the tanks I added 0.5 kg leaf litter to each, along with standard quantities of algae and zooplankton collected from a nearby pond. The tanks were filled on 7 March 1998, and they supported flourishing populations of periphyton and zooplankton by the time tadpoles were added a month later. Each tank was provided with four floating cages, constructed of plastic tubing 12 cm in diameter (1 L volume) and covered at both ends with fibreglass window screen. Each cage either housed a predator or was left empty, depending on the treatment.

The experimental design included five treatments: empty cages without predators, or all cages containing a single individual of one of four species of predator. The predator species, collected from nearby ponds, were larval Aeshna cyanea (dragonfly; Order Odonata: Family Aeshnidae), larval Anax imperator (dragonfly; Odonata: Aeshnidae), adult Notonecta glauca (backswimmer; Hemiptera: Notonectidae), and adult Triturus alpestris (newt; Caudata: Salamandridae). These species were selected because they are known to feed upon R. temporaria, and because they employ distinct methods of prey capture. The odonates catch prey with a protrusive labium; they move slowly along the substrate, frequently stopping to watch for moving prey. Aeshna is somewhat more active than Anax, and more clearly restricted to ponds without fish. Notonecta is a highly visual predator of the open water column and water surface; it typically swoops down from above to capture prey with its legs. Adult newts move along the bottom and through vegetation in search of prey, which they capture by suddenly striking while opening the mouth to create suction. All four predators are highly sensitive to movements of potential prey.

The predators were fed with tadpoles of R. temporaria every other day throughout the experiment (300 mg wet mass per predator); all predators received aliquots of food drawn haphazardly, so any difference in response to treatments cannot be attributed to differences in predator diets. Tadpoles were probably unable to see or sense movements from predators within the cages, but they could detect chemical cues produced by the act of predation (Hews, 1988; Stauffer & Semlitsch, 1993).

Eighty hatchling R. temporaria were added to each tank on 4 April (4 days after hatching, Gosner stage 23–24, mean mass 16.2 mg); 15 tadpoles were removed from each tank in late April for use in other experiments. The experiment ended on 18–19 May, when tadpoles were at stage 37–39, and all survivors were counted and weighed. The initial density of 57/m2 was very low in comparison with hatchling densities in natural ponds, but survival was high and the final average density of 44/m2 was higher than 81% of natural R. temporaria populations at the same time of the year (J. Van Buskirk, unpublished).

I measured behaviour during two time periods, once when the tadpoles were small (14–16 April, mass ∼40 mg) and once again late in development (11–16 May, mass 400–500 mg). On both occasions I recorded two kinds of response designed to estimate the fraction of tadpoles that was hiding in the leaf litter and the activity budget of tadpoles that were not hiding. The fraction of individuals hiding was estimated by counting the number visible either in the water column or on the walls or bottom of the tank. Counts were repeated three times within each sampling period, and the mean number visible in each tank was divided by the estimated number alive on that date (calculated from the number remaining alive at the end of the experiment and assuming a constant mortality risk through time) to yield the proportion of individuals not hiding. Activity budgets were sampled during the same two time periods by observing in each tank six focal tadpoles for 45 s each, recording the proportion of time spent active (feeding or swimming).

I measured the size and shape of a sample of eight tadpoles drawn from each tank on three occasions (23 April, 1–3 May, and 18–19 May). Tadpoles were photographed individually in lateral and ventral views within a small glass chamber. I later projected the photographic images onto a computer monitor and digitized the lengths of five traits known to exhibit predator-induced plasticity (Smith & Van Buskirk, 1995; McCollum & Van Buskirk, 1996). These traits were the length, width, and depth of the body, and the length and maximum depth of the tailfin. The repeatabilities of these measurements, estimated from replicate photographs digitized three times each by three different people, were between 0.979 and 0.990.

Analyses of body and tail shape focused on traits that were corrected for variation in size. I first performed nonlinear regressions of each trait against body size and the square of body size; residuals from these regressions were normally distributed and showed no trends when plotted against size. Regressions were highly significant (R2 between 0.935 and 0.991), and both linear and nonlinear terms were significant in all cases. Body size was defined as the sum of distances in three-dimensional space among all pairs of 21 landmarks distributed across the body and tail (centroid size; Bookstein, 1991). A complete map of landmarks is available from the author. Repeatabilities of the five size-corrected measures of shape range from 0.553 to 0.919.

Sampling exposure to predators in nature

I made quantitative surveys of predator densities in ∼35 natural ponds between 5 and 18 May 1997–2000, during the middle of the larval period of R. temporaria. Detailed methods are presented elsewhere (Van Buskirk & Schmidt, 2000). Briefly, I estimated the densities of relatively common predators based on 20–35 samples collected in each pond with a hollow pipe (35 cm diameter), and noted the presence of rarer taxa by dipnetting for 10–20 min per pond. All predators were identified and measured to the nearest mm. Here I focus only on the presence or absence of the four predator taxa used in the plasticity experiment, within the subset of ponds occupied by R. temporaria tadpoles. Because I am interested in exposure to predators that represent an appreciable mortality threat to medium-sized tadpoles, I restrict analysis to predator individuals that were at least 15 mm long.


Behavioural responses to multiple predators

Rana temporaria tadpoles showed different behavioural responses to the four predator species, although predator diets were the same throughout the experiment. When they were small (age 14–16 days, Fig. 1A), tadpoles responded to all predators with a ∼70% decrease in the amount of time spent moving, but a greater proportion was hiding in the presence of Notonecta than with the two odonates. Most tadpoles became more active and spent less time hiding as they grew larger (age 41–46 days, Fig. 1B), but they exhibited large decreases in both visibility and activity at this stage when in the presence of odonates. Large tadpoles showed an intermediate response to Notonecta, and no response to adult Triturus.

Figure 1.

 Behaviour of R. temporaria tadpoles in the presence of four predator species and in the absence of predators, measured at two stages during the larval period. Activity is the sum of feeding and swimming. Age refers to the number of days since hatching; the experiment began when the tadpoles were 4 days old (on 4 April). Bars show ±1 SE of the mean.

Multivariate repeated measures analysis (Littell et al., 1991: 266) of the two behavioural responses revealed strong effects of time, predator treatment, and their interaction (Table 1). A planned contrast among the four treatments with predators demonstrated that tadpoles differed in their behaviour depending on the predator species. Univariate tests suggested that the proportion of tadpoles hiding contributed disproportionately to the difference among predator treatments. Activity showed less variation among predators, mostly late in development (Table 1, Fig. 1). The significant Time-by-Treatment interaction in the multivariate analysis is important here, because it indicates that behavioural responses to the different predator species changed as tadpoles grew larger. Small tadpoles responded most strongly to Notonecta, whereas large tadpoles reserved their strongest response for dragonflies.

Table 1.   Multivariate (A) and univariate (B) repeated measures analyses of behavioural responses to predators by R. temporaria tadpoles. The two response variables (proportion of individuals hiding and proportion active) were measured on 14–16 April and 11–16 May. Multivariate effects for Time and Time-by-Predator were tested in a model that did not include Block, due to insufficient degrees of freedom. The predator effect is tested over its interaction with block, and the contrast tests for significant variation among the four predator treatments. Thumbnail image of

Morphological responses to multiple predators

All predators induced R. temporaria to develop relatively short bodies and deep tail fins (Figs 2 & 3). All tadpoles had similar shapes in the earliest sample, and the nonpredator treatment became more dissimilar as tadpoles grew larger. There were no obvious differences in morphological response to the different predator species (Figs 2 & 3).

Figure 2.

 Body shape of R. temporaria tadpoles in the presence of four predator species and in the absence of predators, measured on three occasions during the larval period. Shape variables are residuals after regression against body size. The age of the tadpoles is the number of days since hatching. The data for one morphological response, relative body width, are not shown in this figure but are included in statistical analyses (Table 2). Bars depict ±1 SE of the mean.

Figure 3.

 Tail fin shape of R. temporaria tadpoles in the presence of four predator species and in the absence of predators, measured on three occasions during the larval period. Shape variables are residuals after regression against body size; bars show ±1 SE.

There were significant effects of Time, Treatment, and Time-by-Treatment in multivariate repeated measures analysis of five measures of tadpole body and tail shape (Table 2). The Time effect resulted mostly from an increase in relative body length during development in most treatments, whereas the treatment effect derived from large differences between nonpredator and predator treatments in both relative body length and tail fin depth (Figs 2 & 3). The Time-by-Treatment interaction developed because the predator-free treatment became more dissimilar from all others, and not because responses to the different predators changed during development. Multivariate and univariate contrasts confirmed that the four predator species caused similar morphological responses (Table 2).

Table 2.   Multivariate (A) and univariate (B) repeated measures analyses of morphological responses to predators by R. temporaria tadpoles. The response vector is defined by five variables (body length, depth, and width, tail length, and tail depth, all corrected for body size) each measured on 22 April, 1 May, and 18–19 May. The contrast tests for variation among the four predator treatments, and the predator treatment effect is tested over its interaction with block. Thumbnail image of

Exposure to predators in nature

The field sampling data show that natural populations of R. temporaria are exposed to the four predator species at high frequency (Table 3). Over the four years, 63% of ponds with tadpoles also contained adult T. alpestris, 51% had adult Notonecta, 44% had Aeshna, and 25% had Anax. In each year I found a small number of ponds with R. temporaria that had no predators, and many ponds that had just one kind of predator, but most ponds (56%) contained two or more species and one pond had all four predators together.

Table 3.   Co-occurrence of R. temporaria and four kinds of predators in ponds near Zürich, Switzerland. The table lists the number of ponds with R. temporaria tadpoles that also had the given species composition of predators when sampled during mid-May of 4 years. Only predators >14 mm in body length are included. Thumbnail image of


This study demonstrates that anuran larvae produce significantly different defensive responses to different species of predator, confirming previous results showing that tadpoles accurately ascertain predation risk and can respond to multiple predators (Hews, 1988; Semlitsch & Reyer, 1992). The data extend earlier results by illustrating that tadpoles make a qualitative distinction among predators, rather than simply showing a graded response to different concentrations of a chemical substance. Further, the results indicate that different aspects of the phenotype differ in the specificity of response: tadpoles showed different behavioural responses to different predators, and their morphological responses to the same set of predators were not different. There are suggestions that both behavioural and morphological responses to these predators are adaptive.

Prey may show different responses to different predator species without actually distinguishing among them. If different predators produce different quantities of a cue that is associated with predation risk, and if prey have a graded dosage response to that cue, then the prey will exhibit different levels of response when exposed to different predators. Predators that produce more cue will cause greater responses. This is true regardless of whether the cue is visual, chemical, or tactile, and whether the cue is produced by the victim or the predator itself. Earlier studies in which tadpoles exhibited differing degrees of response to different predators are all consistent with a graded dosage response to a single cue signalling risk (Semlitsch & Reyer, 1992; Kiesecker et al., 1996; Lefcort, 1996; Manteifel & Zhushev, 1998; Relyea, 2001).

My finding that behavioural responses to the different predators changed in rank as tadpoles grew larger is not consistent with the existence of a single graded cue of any kind. Different responses recorded during any single observation period might reflect a dosage response, but a clear developmental shift in response contradicts such an interpretation, assuming that there is no dramatic change in the type or quantity of cue produced by the predators. A dramatic shift in the cue during this experiment seems unlikely, because predators did not grow appreciably larger (Notonecta and Triturus were adults, while the aeshnids moulted only from instar F-1 to F-0), and I fed all predators with equivalent diets throughout. My results must therefore stem from changes in specific response to the different predators.

A shifting behavioural response during development implies that predators are perceived differently by tadpoles. Consider the effects of predators late in the larval period. The strong response to aeshnid dragonflies at this stage might be explained on the basis of differences in the quantity of a cue released by tadpoles undergoing predation. Larval odonates consume prey piecemeal, holding them with the labial palps while using the mandibles to tear off pieces. Inevitably, fluids from the prey’s body or skin leak into the water. In contrast, adult newts capture prey by engulfing them whole, providing little opportunity for the prey to release a chemical directly into the water. Predation by Notonecta probably causes intermediate quantities of fluid to be released from the prey, because Notonecta inserts its beak into the victim, which is held outside the predator’s body. Thus, a chemical cue following a simple dosage response model predicts that tadpoles should respond most strongly to odonates, which are likely to spill the greatest quantity of their prey’s blood. This is what I observed in the late sample.

However, this interpretation cannot be reconciled with results from the early behavioural sample. Here, the relative responses to aeshnids and Notonecta were opposite to that in the late sample, which contradicts the notion that dragonflies produce a greater quantity of cue. Thus, differences among predators did not arise as a consequence of a graded dosage effect, unless the shape of the dosage response curve itself changed dramatically during development. I conclude that at least some predators produce qualitatively different cues that enable tadpoles to distinguish among them. This conclusion remains tentative until more specific information is available on the identity of the cues used by tadpoles to detect predators (Kats & Dill, 1998).

The possible adaptive basis of different responses to particular predators is uncertain without information on how individual fitness varies with phenotype. There is good evidence that the morphological response improves survival in the presence of predators, and is maintained by natural selection imposed by predators (Van Buskirk & Relyea, 1998; Van Buskirk & McCollum, 2000a). In this study, the similar changes in body and tail shape caused by all four predators implies that morphology is more constrained than behaviour, or that escaping different predators does not require distinct phenotypes. The uniform morphological response did not arise from a perceptive limitation, because tadpoles were clearly capable of detecting differences in the predator environment. Instead, functional considerations support the idea that all four predators select similar changes in tail shape. Predator-induced morphology in anuran larvae increases survival by facilitating escape once an attack is underway. The advantages of a large tail fin, for improving swimming performance or distracting the predator (Van Buskirk & McCollum, 2000a,b), should operate when a tadpole is challenged by any type of predator. This explanation makes the testable prediction that odonates, newts, and Notonecta all impose natural selection favouring individuals that have similar tail and body shape.

In contrast with morphological responses, behavioural responses to predators function at an earlier stage in the predation sequence, by reducing the probability that prey will encounter predators. Because the predators employed here hunt in different ways and show different habitat distributions, it is plausible that the adaptive behavioural responses to each are different. For example, adult Notonecta are visual predators of the open water, so hiding under leaf litter may be an effective defence. Adult Triturus are dangerous while tadpoles are small, but they are gape limited and therefore present little risk for prey that are larger than the width of their head (Zaret, 1980). The pattern of behavioural response to newts, which was strong when tadpoles were small but absent as they grew large, may be adaptive for gape limited predators. Odonates such as Aeshna and Anax hunt in and on top of the vegetation, and are dangerous even for large tadpoles. For these species the adaptive response may be to minimize activity for the entire larval period, as observed here. It can be argued, then, that the different behavioural responses to the four predators, and the similar morphological responses, are adaptive.

If the distinct behavioural responses turn out to be specialized adaptations to predators, then these predators should occur at high frequency, according to models for the evolution of plastic reaction norms across multiple environments (van Tienderen, 1991). The field sampling data support this expectation: all four predators are locally common, and R. temporaria tadpoles co-occur with them at high frequency. Local populations may not have the opportunity to specialize on a single predator species because different predators often occur together in combination, and there is likely to be some change in predator species composition between years (Jefferies, 1994; Schneider, 1997; Van Buskirk & Relyea, 1998). Many tadpoles are exposed to multiple predators within their lifetime, or their offspring face different predators than they do.

Theory suggests that the maintenance of specialized responses to multiple enemies under these conditions requires that the predators act independently (Rausher, 1992; Hougen-Eitzman & Rausher, 1994; Iwao & Rausher, 1997). This could arise in two ways. First, the appropriate responses to different predators may be identical, and the predators act additively on prey phenotypes and vulnerability. From the perspective of the prey, the multiple predators act as a single enemy. Alternatively, the adaptive responses to multiple predators may differ, but they are genetically uncorrelated and the presence of one predator does not strongly influence vulnerability to others. This second scenario is frequently violated in other multiple predator systems (Sih et al., 1998), but many such studies are purposefully targeted at systems in which adaptive responses to two predators are expected to be in direct conflict (e.g. DeWitt et al., 2000). The situation in which appropriate responses to multiple enemies are either congruent or independent may be more common. For the four predators included here, it is unlikely that behavioural or morphological responses to one predator would greatly increase vulnerability to one of the others. In this case, then, the evolution of distinct responses to each predator may be fairly straightforward.


I am grateful to Gerda Saxer for help with the experiment, to Yvonne Willi for help in measuring tadpoles, and to Jukka Jokela, Derek Roff, and Benedikt Schmidt for constructive comments on the manuscript. Thanks also to the Swiss Nationalfonds (31–40476.94, 31–50525.97) for financial support.