Volume 82, Issue 5 p. 1031-1041
Standard Paper
Free Access

Foraging and vulnerability traits modify predator–prey body mass allometry: freshwater macroinvertebrates as a case study

Jan Klecka

Corresponding Author

Jan Klecka

Department of Ecosystems Biology, Faculty of Science, University of South Bohemia, Branišovská 31, 37005 České Budějovice, Czech Republic

Laboratory of Theoretical Ecology, Biology Centre of the Academy of Sciences of the Czech Republic, v.v.i., Institute of Entomology, Branišovská 31, 37005 České Budějovice, Czech Republic

Correspondence author. E-mail: [email protected]Search for more papers by this author
David S. Boukal

David S. Boukal

Department of Ecosystems Biology, Faculty of Science, University of South Bohemia, Branišovská 31, 37005 České Budějovice, Czech Republic

Laboratory of Theoretical Ecology, Biology Centre of the Academy of Sciences of the Czech Republic, v.v.i., Institute of Entomology, Branišovská 31, 37005 České Budějovice, Czech Republic

Search for more papers by this author
First published: 19 July 2013
Citations: 68

Summary

  1. Predation is often size selective, but the role of other traits of the prey and predators in their interactions is little known. This hinders our understanding of the causal links between trophic interactions and the structure of animal communities. Better knowledge of trophic traits underlying predator–prey interactions is also needed to improve models attempting to predict food web structure and dynamics from known species traits.
  2. We carried out laboratory experiments with common freshwater macroinvertebrate predators (diving beetles, dragonfly and damselfly larvae and water bugs) and their prey to assess how body size and traits related to foraging (microhabitat use, feeding mode and foraging mode) and to prey vulnerability (microhabitat use, activity and escape behaviour) affect predation strength.
  3. The underlying predator–prey body mass allometry characterizing mean prey size and total predation pressure was modified by feeding mode of the predators (suctorial or chewing). Suctorial predators fed upon larger prey and had ˜3 times higher mass-specific predation rate than chewing predators of the same size and may thus have stronger effect on prey abundance.
  4. Strength of individual trophic links, measured as mortality of the focal prey caused by the focal predator, was determined jointly by the predator and prey body mass and their foraging and vulnerability traits. In addition to the feeding mode, interactions between prey escape behaviour (slow or fast), prey activity (sedentary or active) and predator foraging mode (searching or ambush) strongly affected prey mortality. Searching predators was ineffective in capturing fast-escape prey in comparison with the remaining predator–prey combinations, while ambush predators caused higher mortality than searching predators and the difference was larger in active prey.
  5. Our results imply that the inclusion of the commonly available qualitative data on foraging traits of predators and vulnerability traits of prey could substantially increase biological realism of food web descriptions.

Introduction

Predation is one of the most important ecological interactions. Body size plays a central role among the many factors affecting the strength of predator–prey relationships (reviewed, e.g. in Barbosa & Castellanos 2005). Put simply, larger predators tend to eat larger prey (Elton 1927; Peters 1983; Warren & Lawton 1987; Cohen et al. 1993). Predator and prey body sizes are thus positively correlated across whole communities – the prey is on average c. 100 times smaller than the predator, although the ratio can vary greatly (Brose et al. 2006; Bersier & Kehrli 2008; Naisbit et al. 2011; Riede et al. 2011) and depends on whether interactions between species or individuals are considered (Woodward & Warren 2007; Gilljam et al. 2011). Size-selective predation is recognized as a major force structuring food webs (e.g. Cohen et al. 1993; Woodward et al. 2005; Woodward & Warren 2007; Ings et al. 2009), and predator–prey mass ratios critically affect interaction strengths and hence food web dynamics and stability (Jonsson & Ebenman 1998; Emmerson & Raffaelli 2004; Berlow et al. 2009). In some cases, such as marine pelagic food webs, body size is also one of the few routinely recorded traits, which has inspired the development of size spectrum modelling approaches based solely on body size (Andersen & Beyer 2006).

Although measured size spectra (Sheldon, Prakash & Sutcliffe 1972) provide useful information on the status of the entire ecosystem (Mulder & Elser 2009; Petchey & Belgrano 2010), studies of food web interactions (Hildrew, Raffaelli & Edmonds-Brown 2007; Petchey et al. 2008; Ings et al. 2009; Rall et al. 2011) repeatedly expressed the need to accommodate further traits for more realism. Which traits should that be? Both predators and their prey can be characterized by various phenotypic traits that describe their morphology, behaviour and habitat use. These traits can modify the strength of predator–prey interactions, for example, by making the predators better foragers in a certain habitat type or prey less vulnerable to predation through morphological or behavioural adaptations (Miner et al. 2005; Kishida et al. 2010; Toledo, Sazima & Haddad 2011). Hence, a narrow focus on body size may distort our understanding of the strengths of predator–prey interactions (Rall et al. 2011; Naisbit et al. 2012). Behavioural observations have shown that the activity and foraging behaviour of predators may affect prey choice in both aquatic and terrestrial systems (Allan, Flecker & McClintock 1987b; Downes 2002). Predators with different foraging modes can have profoundly uneven effects on prey population dynamics with important consequences for food web structure (Beckerman, Uriarte & Schmitz 1997; Křivan & Schmitz 2004; Schmitz, Křivan & Ovadia 2004). However, other traits have not been quantitatively evaluated so far.

Lack of empirical data also hampers our ability to validate food web models. Most currently available data focus on predator–prey size relationships and have been used in different food web models (Beckerman, Warren & Petchey 2006; Petchey et al. 2008; Rohr et al. 2010; Williams, Anandanadesan & Purves 2010; Williams & Purves 2011). Additional abstract trophic traits characterizing the foraging and vulnerability properties of each species which determine the presence or strength of trophic links have been included in recent food web models (Rossberg et al. 2006; Rohr et al. 2010; Rossberg, Brännström & Dieckmann 2010), but a clear relationship between these abstract trophic traits and observations is yet to be established. Hence, we need to find empirically observable traits that could give a specific meaning to these unknown traits.

Predatory aquatic insects are suitable to identify such observable traits in predator–prey interactions. They vary in morphology, foraging strategy, activity and microhabitat use (reviewed in Peckarsky 1984). As top predators, they can exert strong predation pressure and consequently affect the structure and dynamics of the whole food web in fishless water bodies (Murdoch, Scott & Ebsworth 1984; Arner et al. 1998; Cobbaert, Bayley & Greter 2010). Even if many early studies (reviewed in Peckarsky 1984) reported that these predators do not select prey by size, classic studies on Chaoborus larvae, Notonecta nymphs and stonefly larvae showed the opposite (Pastorok 1981; Allan, Flecker & McClintock 1987a,b; Streams 1994). Previous studies further indicated possible roles of predator foraging mode (ambush or searching) and of prey vulnerability to capture in the diet composition of predatory aquatic insects (Peckarsky 1984; Allan, Flecker & McClintock 1987b; Woodward & Hildrew 2002a). Nevertheless, currently available data do not provide detailed insights into the mechanisms of predation and prey choice by aquatic insects (reviewed in Klecka & Boukal 2012).

Here, we use laboratory experiments on a diverse set of predatory aquatic insects and their prey to resolve the importance of selected traits for the strength of predator–prey interactions, measured as prey mortality and mass-specific predation rate. We ask three main questions: to what extent do predator–prey size allometries in freshwater macroinvertebrates conform to the commonly observed patterns? Do foraging traits of the predators (microhabitat use, foraging mode and feeding mode) and vulnerability traits of the prey (microhabitat use, activity and escape ability) modify the expected size-dependent relationships? Do predators with different foraging traits impose equally strong predation pressure on their prey? We also relate our results to recently developed food web models incorporating multiple traits and suggest how empirical data such as ours could guide further theoretical developments.

Materials and methods

Laboratory Experiments

We performed multiple choice predation experiments with 13 predatory aquatic insects (adult and larval Coleoptera, adult Heteroptera and larval Odonata), which were offered a mixture of seven prey species. Species dominating the communities of small fishless water bodies in the temperate zone of Europe were selected for the experiments. The predators ranged from 2·7 to 526·0 mg and prey from 0·04 to 7·82 mg in body mass. The experimental protocol is described in full detail in (Klecka & Boukal 2012).

Individuals were collected in various small fishless water bodies in South Bohemia, Czech Republic. Experiments were carried out in a climate room with a regular day-night temperature (day: max. 22 °C, night: min. 18 °C) and photoperiod (18L:6D) cycle corresponding to the prevailing environmental conditions at the time of capture (May–June). Animals collected in the field were acclimated for 2–5 days prior to experiments. Predators were kept individually in small containers (0·25–0·7 L) and fed daily ad libitum with prey different from that used in the experiments, mainly caddisfly larvae. Each predator was starved for 24 hours prior to the experiment to standardize hunger levels.

Experiments were performed in plastic boxes filled with 2·5 L of tap water (depth c. 8 cm) aged for one day before the experiment. The vessels had no substrate on the bottom; four narrow stripes of white plastic mesh (mesh size 2 mm) suspended in the water column provided a simple perching structure. We eliminated refuges, which are known to affect predation rates (Peckarsky 1984) and thus ensured that prey mortality reflected its size and escape behaviour, the traits we focused on in the analyses. In each experiment, all prey (six individuals of Rana tadpoles, six Lymnaea, 10 Chironomus, 10 Cloeon, 10 Culex, 10 Asellus and 30 Daphnia; approximating a natural setting with small prey outnumbering large prey) were released first in the vessel; the predator individual was added after several minutes when the prey settled in the arena and resumed their normal behaviour (except Chironomus larvae, which were not allowed to build tubes and hence simply crawled/rested on the bottom). Each individual predator was used only once.

We classified the predators and prey according to their microhabitat use and behaviour, based on qualitative observations of individuals during the predation experiments. Predator foraging characteristics included foraging mode (ambush/searching), feeding mode (chewing/suctorial) and foraging microhabitat (bottom/water column); perching predators were classified as foraging in the water column (Table 1). Prey vulnerability was classified by its activity (sedentary/active), ability to escape the predator with a sudden rapid movement (referred to as slow-escape and fast-escape prey) and its preferred microhabitat (bottom/water column; almost no prey individuals used the mesh and only two microhabitats were thus recognized) (Table 2). Differences in both behaviour and microhabitat use were strong between species, and we thus used discrete categories for the traits as in Peckarsky (1984). In particular, all searching predators found their prey by active search, while all ambush predators spend most of their time nearly immobile and attacked prey after its detection. They occasionally employed a mixed stalking-and-ambush tactics, but their overall activity was always much lower than that of active predators. Active prey spent a significant proportion of time moving rather fast in the experimental vessel, while sedentary prey mostly assumed a fixed position and moved chiefly when challenged by a predator (Cloeon, Chironomus) or moved very slowly (Lymnaea).

Table 1. Body mass and other traits of predators used in the experiments
Species N Body mass (mg) Foraging microhabitat Foraging mode Feeding mode
Mean SD
Coleoptera
 Acilius canaliculatus adult 8 61·1 1·13 0 1 0
 Acilius canaliculatus L2 7 2·7 1·18 1 0 1
 Acilius canaliculatus L3 8 13·9 1·36 1 0 1
 Dytiscus marginalis adult 9 526·0 1·10 0 1 0
 Dytiscus marginalis L3 5 161·6 1·36 1 0 1
 Hydaticus seminiger adult 8 64·5 1·11 0 1 0
Hemiptera
 Ilyocoris cimicoides adult 8 33·8 1·25 0 1 1
 Notonecta glauca adult 8 38·7 1·19 1 0 1
Odonata
 Anax imperator F-0 9 261·9 1·23 1 0 0
 Coenagrion puella F-0 9 4·70 1·27 1 0 0
 Libellula depressa F-0 6 55·4 1·41 0 0 0
 Libellula depressa F-2 7 20·3 1·26 0 0 0
 Sympetrum sanguineum F-0 8 20·3 1·25 1 0 0
  • Geometric mean and geometric standard deviation are reported. Foraging microhabitat: 0 = bottom, 1 = water column; foraging mode: 0 = ambush, 1 = searching; feeding mode: 0 = chewing, 1 = suctorial. L2 = second instar larvae, L3 = third instar larvae, F-0 = last instar larvae, F-2, the second before the last instar larvae, N = number of replicates.
Table 2. Body mass and other traits of prey species used in the experiments. Lymnaea snails were weighed without shell. Geometric mean and geometric standard deviation are reported
Species Body mass (mg) Microhabitat Activity Escape Taxon (order)
Mean SD
Asellus aquaticus adult 1·70 1·34 0 1 1 Isopoda
Chironomus sp. larva 0·33 1·35 0 0 1 Diptera
Cloeon dipterum larva 0·98 1·21 0 0 0 Ephemeroptera
Culex sp. larva 0·54 1·31 1 1 0 Diptera
Daphnia sp. adult 0·04 1·94 1 1 1 Cladocera
Lymnaea stagnalis juvenile 7·82 1·38 1 0 1 Pulmonata
Rana sp. tadpoles 3·04 1·19 0 1 0 Anura
  • Microhabitat: 0 = bottom, 1 = water column; activity: 0 = sedentary, 1 = active; escape from predator: 0 = fast, 1 = slow.

All surviving prey were counted after 24 hours to determine predation mortality; see Table S1 in Klecka & Boukal (2012) for detailed data. Number of prey eaten by a predator was estimated by counting missing prey at the end of the experiment. We assumed that prey dying from natural causes would not be eaten by the predator: mean number of prey specimens which died during predator-free control trials (= 4) was subtracted from the numbers of missing prey individuals in each predation trial when calculating the number of eaten prey. Natural mortality was lower than one individual for all prey but Daphnia and Chironomus (Klecka & Boukal 2012) and potential bias stemming from its effect did not affect our main results.

All predators and 20 randomly chosen individuals of each prey species were subsequently stored in 80% ethanol and weighed to the nearest 0·001 mg after 72 hours of drying at 50 °C to quantify dry body mass. Data on predator–prey body mass ratios (hereafter referred to as PPMR) in the analyses are based on the individual dry mass of each predator individual and average body mass of each prey species; the average prey size for each predator is weighted by the total number of prey eaten in all replicate experiments.

Data Analyses

All analyses were carried out using R 2.15.1 (R Core Team 2012). Significance level for all tests was set to = 0·05. Body mass and PPMR data were ln-transformed prior to the analyses because the values were log-normally distributed. For each predator, we report average prey size urn:x-wiley:00218790:media:jane12078:jane12078-math-0001, measured as the (geometric) mean of the body mass of killed prey weighted by the number of killed individuals and its variance Var(ln(w)) as a measure of the width of the size selectivity curve of the predator. The broad range of available prey body masses w (Table 2) justifies our assumption that urn:x-wiley:00218790:media:jane12078:jane12078-math-0002 reflects the preferred prey size of each predator. We tested the dependence of urn:x-wiley:00218790:media:jane12078:jane12078-math-0003, Var(ln(w)) and PPMR on predator mass and foraging traits using a standardized major axis regression (SMA) implemented in the smatr package for R (Warton et al. 2012). Results for each predator species and for each developmental stage (where 2–3 were studied, i.e. in Dytiscus, Acilius and Libellula) were treated as separate data in these calculations. Total mass-specific predation rate Ptot, measured as dry mass of prey killed during the experiment per unit dry mass of the predator, was calculated for each predator species to test the allometry of predator food intake. Effect of ln-transformed predator mass and foraging traits on ln-transformed Ptot values was compared using SMA regression. Since both predator and prey weight could have been confounded by measurement errors (different gut fullness and possible effect of the postexperimental treatment in predators; indirect weight measurement in the prey), SMA regression is the preferred method to estimate the allometries (Warton et al. 2006).

To assess how predator and prey traits affect the strength of trophic links in our experimentally assembled food web, we characterized both predators and prey by four raw trophic traits in the sense of Rossberg, Brännström & Dieckmann (2010). Traits ti of predator i included ln-transformed body mass ln (wi) and three qualitative traits: microhabitat use Hi, feeding mode Gi and foraging mode Fi (Table 1), while traits tj of prey j included ln-transformed body mass ln (wj) and three qualitative traits: microhabitat use Hj, activity Aj and escape ability Ej (Table 2). Rossberg, Brännström & Dieckmann (2010) assumed that values of all raw phenotypic traits can be ascribed to both predators and prey. Here, only the first two raw phenotypic traits (prey size and habitat use) are common for the predators and prey, and the latter two, characterizing foraging of the predator and vulnerability of the prey, are different. This modification has no bearing on the subsequent analyses. All qualitative traits had two levels that were set at 0 and 1 in the model fitting procedure (Tables 1 and 2).

Strength of each trophic link was measured by mean prey mortality mij, defined as the probability of an individual prey being killed by the predator during the experiment. Following the general method outlined in Rossberg, Brännström & Dieckmann (2010), we tested how ln-transformed prey mortality depends on the predator and prey traits using a quadratic polynomial model. We first considered a model that included only predator and body size,
urn:x-wiley:00218790:media:jane12078:jane12078-math-0004(eqn 1)
and compared it with the full model
urn:x-wiley:00218790:media:jane12078:jane12078-math-0005(eqn 2)
that included up to 15 unknown parameters: positive scalar a0, vector b = (b1, …,b6) covering the main statistical effects of predator and prey traits ti and tj and elements urn:x-wiley:00218790:media:jane12078:jane12078-math-0006 of the predator–prey interaction matrix urn:x-wiley:00218790:media:jane12078:jane12078-math-0007 (Appendix S1). Remaining elements of the interaction matrix were set to zero because we wanted to test how microhabitat use and the interactions of the additional two foraging and vulnerability traits modify a baseline mortality given by the predator–prey size relationship. Main effects of predator and prey microhabitat use were not included in model eqn 2 because some combinations of the three qualitative predator and prey traits were not available in our data set and inclusion of all traits resulted in overfitting (not shown). Finally, only the interaction effect of microhabitat use was included in model eqn 2 to test whether predators from the same and the other microhabitat caused different prey mortality.
We also investigated variants of models eqn 1 and eqn 2 in which the influence of predator and prey body sizes was fully characterized by PPMR (Rohr et al. 2010) by setting b2 = − b1c3 = − c1 and c2 = 0:
urn:x-wiley:00218790:media:jane12078:jane12078-math-0008(eqn 3)
and
urn:x-wiley:00218790:media:jane12078:jane12078-math-0009(eqn 4)
Parameters in models eqn 1-eqn 4 were estimated directly using nonlinear regression (nls function in R 2.15.1) of mortality values, that is mij instead of urn:x-wiley:00218790:media:jane12078:jane12078-math-0010, because of zero mortality for some predator–prey pairs (Rossberg, Brannstrom & Dieckmann 2010). For each model, residual errors ɛij relating model predictions mij given by exponentiating the right-hand side in the model to the observed mean mortality values urn:x-wiley:00218790:media:jane12078:jane12078-math-0011 were minimized under the assumption of normally distributed and independent errors,that is the sum urn:x-wiley:00218790:media:jane12078:jane12078-math-0012 was minimized. Akaike Information Criterion with small sample size correction (AICc) was used to select among the best-fitting models differing in model complexity (Burnham & Anderson 2002). Results are reported for models eqn 1-eqn 4 with the lowest AICc value within each model group, for the full models eqn 2 and eqn 4 and for models for which the difference of the AICc value from the lowest value was at most 2 and hence their evidence ratio did not deviate too strongly from unity (Burnham & Anderson 2002).

Results

The Role of Predator and Prey Traits for Selective Predation

As expected, body size played a prominent role in the selectivity of individual predators. Values of PPMR observed across all predators and predation events, which correspond to data commonly presented in studies of prey selectivity (e.g. Chrzanowski & Šimek 1990; Gaymer, Dutil & Himmelman 2004) and aggregated across all predators were approximately log-normally distributed (black bars in Fig. 1). Predators were on average c. 125 times heavier than their prey (median PPMR = 124·7). Predators did not kill their prey randomly with respect to size. To illustrate this, the distribution of PPMRs in observed predation events was compared with the distribution of all potential PPMRs calculated for all available combinations of individual predators and individual prey (white bars in Fig. 1). The two distributions differed significantly (two-sample Kolmogorov–Smirnov test, = 0·14, < 0·0001).

Details are in the caption following the image
Distribution of predator–prey body mass ratios for observed (black bars) and potential (white bars) individual feeding events. Potential feeding events mean that all predators would consume all prey.

These PPRM values translate into a relatively tight predator–prey mass allometry (SMA, R2 = 0·65, = 0·001; Fig. 2a) modified further by the feeding mode of the predator. Body mass and feeding mode of the predator together explained 79% of variance in body mass of killed prey, while no other predator traits (foraging microhabitat and foraging mode) had a significant effect on the allometry. The estimated slope of the allometry (mean value 0·46, 95% CI = 0·34–0·63) did not differ significantly between suctorial and chewing predators (Likelihood ratio statistic = 0·345, d.f. = 1, = 0·56). On the other hand, the allometry was shifted between both groups; suctorial predators consumed on average 1·8 times larger prey than equally sized chewing predators (Wald statistic = 6·22, d.f. = 1, = 0·01; Fig. 2a). Correspondingly, PPMR increased with predator body mass (SMA, R2 = 0·87, = 4·10−6; Fig. 2b) with an estimated slope of 0·62 (95% CI = 0·48–0·79). The width of the size selectivity curve of a predator measured as variance in ln-transformed prey body mass (Fig. 2c) did not depend on predator body mass (SMA, R2 = 0·10, = 0·28), feeding mode (SMA, Wald statistic = 0·002, d.f. = 1, = 0·96) or any other trait.

Details are in the caption following the image
Dependence of mean prey body mass wj (a), predator–prey mass ratio (b) and prey mass variance (c) on predator body mass wi and feeding mode. Empty symbols and solid line are chewing predators, and filled symbols and dashed line are suctorial predators. Empty triangles = larval Odonata, empty squares = adult Dytiscidae, filled squares = larval Dytiscidae, filled circles = adult Heteroptera. Regression equations are as follows: (a), chewing predators: ln(wj) = −3·01 + 0·46 ln(wi), suctorial predators: ln(prey mass) = −2·44 + 0·46 ln(predator mass); (b), chewing predators: ln(predator–prey body mass ratios (PPMR)) = 2·67 + 0·62 ln(predator mass), suctorial predators: ln(PPMR) = 2·14 + 0·62 ln(predator mass); (c), both predator groups: Var(ln(prey mass)) = 0·49–0·06 ln(predator mass). PPMR = predator–prey mass ratio.

Similar to preferred prey size, mass-specific predation rates Ptot were affected only by predator body mass and feeding mode, which together explained 85% of variance in Ptot. Smaller predators had significantly higher Ptot (SMA, R2 = 0·67, = 0·0006; Fig. 3). Mass of consumed prey ranged from 1·8% (Coenagrion) to 40% (Dytiscus L3) of total available prey mass. The most voracious predators, killing more prey than their own body mass per day, were L2 and L3 larvae of the diving beetle Acilius canaliculatus (2·2 mg.mg−1day−1 and 1·2 mg.mg−1day−1, respectively), while the largest predators used in the experiment, adult diving beetles Dytiscus marginalis, consumed only a fraction of their body mass per day (0·024 mg day−1). Allometries linking Ptot to body mass in suctorial and chewing predators differed in the intercept (SMA, Wald statistic = 11·56, d.f. = 1, = 0·0007) but not in the slope (SMA, Likelihood ratio statistic = 0·008, d.f. = 1, = 0·93). Suctorial predators had on average 2·7 times higher Ptot than chewing predators and the common slope of the allometries was −0·63 (95% CI = −.91–−0·44). Sample sizes (number of species) were too small to compare the allometries between different higher taxa.

Details are in the caption following the image
Data and regression lines showing the dependence of total mass-specific killing rate on predator body mass wi in chewing (empty symbols and solid line) and suctorial (filled symbols and dashed line) predators. Empty triangles = larval Odonata, empty squares = adult Dytiscidae, filled squares = larval Dytiscidae, filled circles = adult Heteroptera. Regression equations are ln(mass-specific killing rate) = 0·51–0·63 ln(wi) for chewing predators and ln(mass-specific killing rate) = 1·51–0·63 ln(wi) for suctorial predators.

The Fit of Trait-Based Food Web Models (1)– (4)

The observed dependence of PPMR on predator body mass impacted the performance of models eqn 1-eqn 4 describing prey mortality. Among models eqn 1 and eqn 4 that considered only predator and prey body mass, model eqn 1 allowing for a variable PPMR fitted the data much better than model eqn 3 using a fixed PPMR (Table 3). Model eqn 1 predicted that mortality should be highest for medium-sized predators feeding on medium-sized to large prey (Fig. 4a) and steeply decrease as predator or prey size increases or decreases.

Table 3. Best fits of models eqn 1-eqn 4 to our experimental data
Model Var ∆AICc wA wB Parameter estimates
a0 b1 b2 b3 b4 b5 b6 c1 c2 c3 c4 c5 c6 c7 c8
(2) 0·59 0 0·57 0·03 (0·02) 0·81 (0·32) 1·45 (0·36) 0·33 (0·13) −0·97 (0·56) 0·48 (0·16) −0·06 (0·04) 0·23 (0·07) 0·31 (0·06) 1·47 (0·54) 0·74 (0·35)
(2)* 0·59 1·51 0·63 0·26 0·03 (0·02) 0·84 (0·33) 1·48 (0·36) 0·32 (0·13) −0·89 (0·56) 0·18 (0·17) 0·52 (0·18) −0·07 (0·04) 0·23 (0·07) 0·34 (0·07) 1·29 (0·56) 0·74 (0·35)
(2) 0·57 3·03 0·30 0·13 0·08 (0·03) 0·28 (0·06) 1·25 (0·31) 0·37 (0·14) −0·97 (0·62) 0·15 (0·17) 0·52 (0·18) 0·18 (0·06) 0·32 (0·07) 1·4 (0·62) 0·82 (0·36)
(2) 0·61 5·87 0·07 0·03 0·03 (0·02) 0·89 (0·32) 1·62 (0·37) 0·11 (0·30) −0·85 (0·54) 0·16 (0·24) 0·35 (0·25) −0·07 (0·04) 0·25 (0·07) 0·36 (0·07) −0·24 (0·19) −0·08 (0·29) 1·41 (0·55) 0·46 (0·29) −0·67 (0·35)
(1)* 0·37 26·2 0·00 0·00 0·03 (0·03) 1·04 (0·47) 1·51 (0·50) −0·09 (0·05) 0·22 (0·10) 0·28 (0·07)
(4) 0·22 45·8 0·00 0·00 0·15 (0·04) 0·49 (0·22) −1·34 (0·71) −0·23 (0·23) 0·04 (0·01) =−c1 1·25 (0·76)
(3) 0·08 53·7 0·00 0·00 0·15 (0·04) 0·06 (0·06) =−b1 0·01 (0·01) =−c1
(4) 0·26 56·5 0·00 0·00 0·12 (0·05) 0·06 (0·07) =−b1 0·39 (0·44) −1·16 (0·92) −0·41 (0·37) 0·26 (0·38) 0·03 (0·01) =−c1 0·03 (0·22) 0·14 (0·43) 1·55 (0·91) 0·05 (0·45) −0·69 (0·57)
  • Parameters for traits increasing or decreasing prey mortality have positive or negative values, respectively. Standard deviations are given in parentheses. Parameters significantly different from 0 (< 0·05; t-test) are shown in bold. Asterisk = models shown in Fig. 3. Var = proportion of explained variance in data, ∆AICc = difference in the corrected Akaike information criterion from the best-fitting model, wA = Akaike weights of the models excluding the best-fitting model eqn 2 in the first row, wB = same as wA but including the best-fitting model.
Details are in the caption following the image
Best fits of models eqn 1 and eqn 2. Panel (a): prey mortality predicted by the purely mass-based model eqn 1; panels (b–e): best fit of model eqn 2 including other traits of predators and prey. Panel (b) shows predicted mortality values corresponding to predator–prey combinations with reference traits (i.e. with trait values set to 0) and panels (c–e) show the multiplicative effect of different combinations of predator and prey traits on prey mortality compared with the reference combination (mean values ± SE are plotted).

Including phenotypic traits other than body mass, that is using model eqn 2 instead of eqn eqn 1 and model eqn 4 instead of eqn eqn 3 improved the fit considerably (Table 3). The best-fitting model eqn 2 including all linear terms that were also involved in the interaction terms, indicated by asterisk in Table 3, contained all terms involving predator and prey body masses (with parameters b1, b2 and c1c3), main effect of predator feeding mode and main effects and interactions of predator foraging mode with prey escape ability and activity. The predicted dependence of prey mortality on predator and prey body mass was similar as in model eqn 1 (Fig. 4). The model further predicted that, all else being equal, suctorial predators cause higher prey mortality than chewing predators (Fig. 4c) and searching predators cause lower mortality to fast-escape than to slow-escape prey, although the escape ability does not affect the mortality of prey when exposed to ambush predators (Fig. 4d). On the other hand, ambush predators kill more active prey and cause much higher mortality than searching predators according to the model (Fig. 4e).

Removal of the nonsignificant linear term b5 Ej describing the effect of prey escape ability further improved the AICc score (first row in Table 3), while the removal of the quadratic term involving predator body mass urn:x-wiley:00218790:media:jane12078:jane12078-math-0013 lead to a slightly worse fit (third row in Table 3). Values of the shared parameters were similar in these three models, and their qualitative effect was the same as in the full model (fourth row in Table 3), indicating a robust modelling approach. Microhabitat use did not contribute to the best fit in any of the three reduced models. Finally, model eqn 4 remained clearly inferior to model eqn 2, included partly different terms, and the predicted effect of the quadratic body mass term was opposite (Table 3).

Discussion

The Role of Predator and Prey Traits for Predation

Body size is among key traits for understanding predator–prey relationships (e.g. Brose et al. 2005, 2006; Woodward & Warren 2007; Bersier & Kehrli 2008; Gilljam et al. 2011; Naisbit et al. 2011; Riede et al. 2011). The mean predator–prey mass ratio observed in our experiments (PPMR = 125) falls within other studies (Brose et al. 2006) and individual-level data from the Broadstone Stream (Woodward & Warren 2007) and several other aquatic food webs (Gilljam et al. 2011). We did not measure body mass of individual prey but our species-level PPMR data provide a close approximation of the underlying individual-level values, because similarly sized individuals within each species were selected for the experiments.

Large predators ate much smaller prey relative to their size compared with smaller predators. This is in line with a general pattern emerging from a large data base of predator–prey interactions based on field-collected gut content data (Brose et al. 2006), although some types of food webs may exhibit the opposite trend depending on predators and prey taxa and possibly also on habitat type (Bersier & Kehrli 2008; Naisbit et al. 2011; Riede et al. 2011). Given our much smaller sample size and narrower predator and prey size ranges, the observed steeper increase in PPMR with predator size should be taken cautiously. Nevertheless, our standardized experimental approach allowed us to identify traits that, at least in aquatic invertebrates, may modify the general PPMR allometry. Bersier & Kehrli (2008) and Naisbit et al. (2011) demonstrated that the allometry differs between taxonomical groups; we suggest that this could be caused by differences in other key traits unrelated to body size, similar to our case.

In addition to body mass, feeding mode of the predator (suctorial or chewing) affected preferred prey size and, as discussed in detail below, total predation rates in our experiments. We conclude that feeding mode modifies predator–prey size allometry and can potentially alter the impact of predators on the prey assemblage. Other predator traits may have similar consequences. For example, raptorial zooplankton predators consume larger prey than suspension feeders when corrected for body size (Wirtz 2012) and insect predators differing in microhabitat use have at least partly different diets (epibenthic and interstitial species in streams: Woodward & Hildrew 2002b; benthic and planktonic species in pools: Klecka & Boukal 2012). Comparative experiments and more detailed, trait-based analyses of field-collected diet data may hence boost our understanding of predator–prey interactions and help improve the realism of many current descriptions of food webs.

Such additional and easily obtainable data can also inform modellers about the minimum set of trophic traits (Rossberg, Brännström & Dieckmann 2010) that could be included in the currently prevailing size- or species-based food web models and their predictions of the distribution of feeding links. These models fall in several broad categories inspired by optimal foraging theory (Beckerman, Warren & Petchey 2006; Petchey et al. 2008) and the standard niche model (Williams & Martinez 2000; Williams, Anandanadesan & Purves 2010; Williams & Purves 2011). Both categories assume that predators feed on a limited range of prey sizes, which corresponds well to empirical data, although they do not explicitly consider predator–prey mass allometry, except for the allometric diet breadth model with ratio handling time function (Petchey et al. 2008), which assumes that handling time depends on PPMR. Even if the influence of foraging behaviour of the predator and anti-predator behaviour of the prey on predator–prey relationships is widely acknowledged (e.g. Miner et al. 2005), modifications of these food web models incorporating multiple traits are yet to be published.

We characterized the strength of the predator–prey links by prey mortality in our experimental data and used two recent, more theoretically driven models (Rohr et al. 2010; Rossberg et al. 2010) to predict link strength by a combination of raw phenotypic traits. Our data were fitted well by models based on the general approach developed by Rossberg, Brännström & Dieckmann (2010). The best-fitting model based on body masses explained 37% of variance in prey mortality. Inclusion of two additional predator foraging traits and two prey vulnerability traits explained further 22% of variance in the data (Table 3). This demonstrates that traits other body mass are important determinants of prey mortality and that model of Rossberg, Brännström & Dieckmann (2010), for which we provide the first empirical test, is a promising development in a much needed effort to include multiple traits into food web models.

In particular, we showed that the assumption of constant PPMR across different predators and prey, relaxed in the model of Rossberg, Brännström & Dieckmann (2010) but used in Rohr et al. (2010), is likely to yield unsatisfactory predictions. Constant PPMR is not supported by empirical data (Brose et al. 2006; Bersier & Kehrli 2008; Riede et al. 2011; Naisbit et al. 2011; this paper). Assumption of constant PPMR not only decreased overall model performance in our comparison but also changed the predictions of the impact of additional traits on the strength of trophic links. Incomplete description of the predator–prey size allometry could thus distort our understanding of the effects of other traits.

Our study also shows that the interaction of predator foraging mode and escape ability of the prey strongly affects its mortality. This interaction is intriguing but logical: ambush predators can kill most prey of suitable size, but searching predators capture mainly slow-escape prey because fast-escape prey can be alerted and avoid attack. Only few empirical studies have addressed this phenomenon. Allan, Flecker & McClintock (1987b) and Tikkanen et al. (1997) showed that stonefly larvae, which search for prey and feed mostly on sedentary blackfly larvae (Simuliidae), only rarely capture actively moving mayfly nymphs. Unlike our experiment, these studies could not separate prey activity and escape ability: they used one active prey capable of fast-escape and one sedentary prey that could not escape. Our results suggest that overall activity plays much lesser, and in fact opposite, role in prey mortality – active prey was predated more but only when exposed to ambush predators.

Interacting behavioural traits of prey and predators may also have implications for the susceptibility of different animal personalities in species with intraspecific trait variation in the form of behavioural syndromes (e.g. shy and bold phenotypes) to predation (Pruitt, Stachowicz & Sih 2012). More research is needed to test the generality of these results and to understand population-level implications of such variations in behaviour of both predators and prey.

Absolute Predation Rates: Life-History Implications

Body size stood out as an important trait determining absolute and relative predation rates in our experiments. Larger predators killed larger amounts of prey but the mass-specific predation rate decreased with predator mass, which is in agreement with previous studies (e.g. Peters 1983). Foraging activity reflects total metabolic requirements of the individual, that is the sum of metabolic maintenance costs and energy required for growth and reproduction. The observed slope of predation rate allometry in our experiment (−0·63) is steeper than the theoretically proposed (Brown et al. 2004; Brown, Allen & Gillooly 2007) and empirically measured (Riveros & Enquist 2011) slope of the resting metabolic rate (−0·25), even if this value is vigorously debated (Chown et al. 2007; Makarieva et al. 2008). Our results point in the same direction as a recent meta-analysis, which showed that handling times (a surrogate for predation rates at infinite prey densities) scale with predator body mass as a power law with an exponent of −0·56 (Rall et al. 2012). This corresponds to a scaling exponent of −0·44 for mass-specific predation rates. Both results imply the presence of an energetic constraint on maximum body size in a given taxon: individuals exceeding a given threshold mass would be unable to meet their metabolic demands.

Steeper slope of predation rate allometry may also reflect the allometry of additional resources required for growth and reproduction in addition to resting metabolism. In particular, larvae and adults of the predator may differ in a number of ways including different metabolic rates or gut clearance rates; our experiments could not address this issue but future work on ontogenetic effects in predator–prey interactions in aquatic insects and other groups could provide important insights. Potentially limiting amounts of prey for larger predators probably had at most minor influence on the result. No predator depleted all prey although the most voracious predator, Dytiscus L3 larvae, consumed on average 40% of total prey mass and the largest predator, adult Dytiscus, consumed 17% of total prey mass. The fact that different predators left varying proportions of killed prey uneaten might have also contributed to the steeper predation rate allometry: our estimate of prey consumption was based on the amount of killed prey rather than ingested energy (we could not measure the latter due to logistic and methodological limitations).

More interestingly, chewing predators killed about three times larger mass of prey than suctorial predators after correcting for body size. Conceivable explanations include pace-of-life syndromes (Ricklefs & Wikelski 2002; Réale et al. 2010) and wingedness (Chown et al. 2007) leading to higher metabolic requirements, wasteful killing in suctorial predators and the inability to consume whole prey or incomplete prey consumption. Among the predators used in the experiments, diving beetles including Acilius have the shortest larval development of about 4 weeks under the experimental conditions (Inoda et al. 2009; D. S. Boukal & J. Klecka, unpublished data). Faster growth requires higher food intake, which could explain why the suctorial diving beetle larvae consumed more prey than odonate larvae per unit of body mass. Similar reasoning for adult heteropterans (suctorial) and diving beetles (chewing predators) would suggest the heteropterans have much higher mass-specific reproductive output, but the fecundity of both groups is comparable (Wesenberg-Lund 1943). The alternative that locomotion and hunting require more energy in adult heteropterans is also unlikely. Wingedness cannot explain the differences as all predators used in the experiment are winged and capable of flight as adults. Wasteful killing without consumption was observed, for example in damselfly larvae (Johnson, Akre & Crowley 1975) but did not occur in our experiments.

Incomplete or differential prey consumption provides the most likely explanation for the difference between predation rates of chewing and suctorial predators. The former consume whole prey (although the fluid contents may be partly lost), while the latter consume only the fluids and soft tissues liquefied by digestive enzymes injected into the prey and the mass of prey killed always exceeds consumed prey mass. Moreover, discarding incompletely consumed prey was described in water bugs under high food supply and may be related to optimal foraging behaviour (Cook & Cockrell 1978; Giller 1980; Bailey 1986), but the phenomenon has not been studied in the other predator groups. In any case, our results suggest that suctorial predators may be more capable than chewing predators of regulating prey populations.

Conclusions

The prospect that one could describe food web processes based solely on individual or species body size (Andersen & Beyer 2006; Woodward & Warren 2007; Ings et al. 2009) is attractive. Recent work, however, shows that food web structure may be influenced more by phylogenetic constraints than size structure alone (Naisbit et al. 2012), and modellers have already considered the role of additional abstract trophic traits. We add realism to the ongoing debate by demonstrating that predictive power of purely size-based models can be greatly enhanced by inclusion of other phenotypic traits. Feeding and foraging modes of predators and behaviour of prey were the most important modifiers of the underlying body mass allometries characterizing predation and trophic links in our experiments. These foraging and vulnerability traits are qualitative and easily observable. Their inclusion in food web descriptions could provide a relatively easy way towards more powerful explanations of predator–prey interactions and ultimately yield more reliable descriptions of real food webs. Replicating our experiments with different taxa with a broader range of foraging and feeding strategies could confirm the validity and generality of the patterns reported here and provide better guidance for a much needed synthesis of multi-trait food web models with empirical data.

Acknowledgements

We thank A. Beckerman, A. Thierry, K. Yarlett, D. Mikolajewski and L. Berec for comments on earlier drafts of the manuscript, O. Petchey and P. Warren for discussions on the relevance of the results and A. Rossberg for comments on abstract trophic traits and access to unpublished work. Comments by two anonymous reviewers improved the presentation of our results. The study was supported by the EU Marie Curie European Reintegration Grant ‘AquaMod’ to D.S.B. (PERG04-GA-2008-239543) and the Grant Agency of the University of South Bohemia (grant no. GAJU 145/2010/P). Use of tadpoles in the experiment was permitted by the regional authority (permit no. KUJCK 12524/2010 OZZL/2/Do) and the Ministry of Education of the Czech Republic (permit no. 7947/2010 30).

      Journal list menu