Survival selection imposed by predation on a physiological trait underlying escape speed


Correspondence author. E-mail:


1. In contrast to other phenotypic traits, selection on physiological traits remains largely undocumented. We have evaluated survival selection imposed by predation by dragonflies on the activity of arginine kinase (Ak), a key enzyme delivering energy for escape performance in invertebrates.

2. To accomplish this, we conducted a semi-natural field enclosure experiment in which we manipulated predation by large dragonfly predators, and quantified escape swimming speed and Ak in the prey, the damselfly Enallagma vesperum. To avoid confounding selection on Ak with selection on other swimming speed-related variables, we also scored all morphological and behavioural traits thought to underlie swimming speed in these damselflies.

3. Dragonfly predators imposed considerable mortality and selected for faster swimming speed and higher activity levels of Ak. Furthermore, higher Ak levels contributed to higher swimming speeds, confirming the mechanistic role of Ak for escape performance. Although morphological (size of the caudal lamellae which generate trust) and behavioural (number of beats made by the abdomen during swimming and the start angle of the C-start) variables contribute to increasing swimming speed, we detected no selection on these variables. This may be due to functional redundancy.

4. Taken together, our results indicated selection on Ak and suggested that selection on physiological traits may be as strong as selection on morphology and life history traits.


The study of selection has a central role in evolutionary ecology as selection is the process driving adaptive evolution potentially leading to population divergence and ultimately speciation (Schluter 2000, 2009). There is a long tradition of quantifying the strength of natural selection in wild populations, with a strong bias toward life history and morphology traits (reviews in Endler 1986; Hoekstra et al. 2001; Siepielski, DiBattista & Carlson 2009). Although adaptive population and species differences have also been documented in physiological traits (e.g. McPeek 1999; Odell, Chappell & Dickson 2003; Pauwels et al. 2007), selection on physiological traits remains largely unknown (reviews in Kingsolver et al. 2001; Siepielski, DiBattista & Carlson 2009; but see e.g. Watt et al. 2003).

Predation is a particularly important selective agent which may drastically affect fitness and is known to have shaped many traits in prey organisms (e.g. Mikolajewski et al. 2006; Svensson & Friberg 2007; Carlson, Rich & Quinn 2009). Several studies have demonstrated survival selection imposed by predators on escape performance (overview in Irschick et al. 2007; Janzen, Tucker & Paukstis 2007; Strobbe et al. 2009). Many phenotypic traits may underlie this whole-organism performance measure and therefore experience such survival selection. As is typically the case for studies of selection, selection on morphological traits but not on physiological traits has been documented. Yet, physiological traits like enzyme kinetics involved in energy delivery, may also contribute to escape performance (Tremblay, Guderley & Frechette 2006).

We explored selection on a physiological trait underlying escape performance in an Enallagma damselfly. These damselflies have been intensively studied with regard to predator-prey interactions and the evolution of antipredator traits (McPeek 1995, 1997, 1999; McPeek, Schrot & Brown 1996; Stoks, McPeek & Mitchell 2003; Stoks & McPeek 2006). When confronted with large dragonfly predators they typically escape by swimming away (McPeek 1990b, 2000; McPeek, Schrot & Brown 1996). Enallagma larvae swim by moving their abdomens from side-to-side and using their three caudal lamellae to generate thrust (Brackenbury 2002). Both morphological and behavioural traits influencing swimming speed have been identified. Larvae with larger, more circular lamellae and that wave their abdomens at a faster rate swim faster (McPeek, Schrot & Brown 1996; Strobbe et al. 2009). Consistent with this, large dragonfly predators impose survival selection for larger and more circular lamellae (McPeek 1997; Strobbe et al. 2009).

A good candidate physiological trait underlying escape performance via swimming and hence under survival selection is the rate at which the enzyme arginine kinase (Ak) phosphorylates ADP into ATP and thus resupplies ATP pools during the first few seconds of strenuous activity (Morrison 1973). For example, higher Ak activities have been linked to higher swim activities in scallops (Tremblay, Guderley & Frechette 2006). In Enallagma damselflies, differences among species in average Ak activity is consistent with adaptation to living with dragonfly predators by evolving faster swimming speeds (McPeek 1999).

We tested for survival selection for increased swim escape performance and increased Ak activity in the damselfly Enallagma vesperum in a field enclosure experiment with large dragonfly predators. To avoid confounding selection on Ak with selection on other swimming speed-related traits we also scored all major morphological and behavioural traits that have been shown to underlie swimming speed in these damselflies (see Strobbe et al. 2009).

Materials and methods

Setup selection experiment

As in previous selection studies of Enallagma damselflies (McPeek 1997; Strobbe et al. 2009), we identified selection based on differences in trait values between caged predator and free-ranging predator treatments at the end of a field enclosure experiment. We did not include a treatment with no predators as previous studies showed that predator cues did not induce a higher escape performance in Enallagma larvae (McPeek 1997; M. A. McPeek, F. Strobbe and R. Stoks, unpublished data). Conventional methods of measuring selection by comparing the distribution of phenotypes before and after the action of a selective agent are not possible in this system for a number of reasons. First, methods for marking are not available to allow us to identify individuals. Without such a technology, we cannot follow free-ranging individuals in cages over time periods (i.e. weeks to months) necessary for selection agents to act. In addition, Ak enzyme activity cannot be quantified without killing the animals. Thus, we could only score the phenotype of interest after the selective agent has acted on the population.

We performed the selection experiment in enclosures in Sylvester Pond (Norwich, VT, USA), a small pond where large dragonfly larvae like Anax junius are the top predators (Stoks & McPeek 2003) (Fig. 1). Enallagma vesperum larvae were collected in McDaniels Marsh (Enfield, NH, USA). We worked with the natural available distribution of larval instars at the time of the experiment. Larval densities used were within the natural density ranges as described in McPeek (1990b).

Figure 1.

 Study pond with field enclosures; the inset shows a larva of the study species Enallagma vesperum.

We used semi-permeable enclosures that allowed small prey items of the damselflies to enter, but prevented damselfly larvae from escaping. These enclosures were 1·2 m high × 30 cm diameter cylindrical cages. Enclosures were made of 2 cm mesh size chicken wire covered with mosquito netting (0·6 × 1·2 mm mesh size). Cages were sealed at the bottom ends with plastic dishes containing c. 1 cm of pebbles. The tops of the enclosures extended 30 cm out of the water and were left uncovered. All enclosures were linearly arranged in the pond at a depth of 90 cm. The macrophyte Chara vulgaris from the same pond was added to each cage in natural density. When free predators are present, Enallagma mortality rates and growth rates in the type of enclosures used has been shown to be indistinguishable from those in natural populations (McPeek 1990b, 1998), suggesting that phenotypic selection measured in these enclosures should accurately reflect selection in natural populations.

For the caged-dragonfly treatment, one antepenultimate instar Anax junius larva was placed inside a small cage in the enclosure. The dragonfly cage was 11 × 11 × 6 cm, and was constructed by placing a small, coarse-mesh (openings 1·7 × 1·0 cm) plastic container inside a bag constructed of mosquito netting. This type of container allowed damselflies visual and olfactory cues for detecting that a large dragonfly larva was present, but prevented the dragonfly from eating the damselflies. Identical cages without a dragonfly were placed in all other enclosures. For the free-ranging dragonfly treatment, one antepenultimate instar Anax junius larva was placed inside the enclosure; dragonflies in this treatment were free to feed on the Enallagma larvae.

Twenty enclosures were installed in the pond on 29 August 2005. Six replicates were run for the caged-dragonfly treatments and eleven replicates for the free-ranging-dragonfly treatment. We performed more replicates for the free-ranging predator treatments, because fewer individuals were expected to survive in this treatment. Enclosures were installed 8 days prior to the addition of odonates to allow colonization by prey for damselflies and dragonflies through the netting. On 5 September, 30 E. vesperum larvae were randomly added to each enclosure, together with a random background sample of 30 larvae of other Enallagma species and one Anax larva. By doing so we avoided any potential differences in the initial trait distributions across enclosures. This is also supported by the fact that at the end of the experiments in the caged-predator treatment, none of the traits differed among enclosures (all > 0·37). Because damselfly larvae use their caudal lamellae to generate thrust (McPeek, Schrot & Brown 1996), only larvae with three intact lamellae were used. We also placed three control enclosures into which we introduced only Chara to check for possible immigration of Enallagma larvae. No immigrants were detected in any of the control enclosures at the end of the experiment.

After 35 days, the contents of the enclosures were returned to the laboratory and all surviving larvae were immediately removed from the samples and kept individually in 100 mL cups in a room at 20 °C and the natural photoperiod, until all swimming trials were conducted (within 36 h).

Response variables

Mortality rate was calculated separately in each enclosure as mortality rate = − [ln (recovered number) − ln (initial number)]/(duration of the experiment). This equation assumes a constant mortality rate throughout the experiment.

For each larva recovered alive (n = 120), we measured wet mass to the nearest 0·01 mg on a Mettler Toledo electrobalance. For subsets of larvae (see below), we also measured swimming speed, and a set of swimming speed-related variables. These included, besides the mass-specific activity of arginine kinase, aspects of behaviour during swimming, and the size and shape of the caudal lamellae. We briefly describe the followed methodologies below; for more detailed descriptions of these methods, we refer to McPeek (1995), McPeek, Schrot & Brown (1996) and Strobbe et al. (2009).

We measured swimming speed by videotaping individual larvae swimming in a 34 × 26 × 6 cm container in the laboratory. A single larva was placed in the container, allowed to settle, and then gently prodded to swim by lightly tapping behind the larva with blunt forceps. The fastest swimming speed from up to four separate swims was determined by digitizing the videotaped swim bouts. Because caudal lamellae are so important in determining swimming speeds (McPeek, Schrot & Brown 1996), we quantified swimming speed only for larvae with three intact lamellae that could be coaxed to swim (n = 59). The percentage of larvae that could be coaxed to swim did not differ between predator treatments (caged Anax: 83%, Free Anax: 70%, Fisher Exact test, P = 0·16) so this variable was unlikely to bias our results. Caudal lamellae are routinely damaged and lost under natural conditions; up to 60% of larvae in any given Enallagma population may have at least one caudal lamella which has been lost or regenerated (McPeek 1990a). Moreover, lamellae could be damaged and lost while retrieving larvae from the enclosures.

Based on the videotaped sequences we also quantified two behavioural variables on these larvae. We calculated the rate of abdomen beats (hereafter swimming beats) as the number of cycles through which the tip of the abdomen moved from side to side per duration of swim, which has units of beats per second (McPeek, Schrot & Brown 1996). One complete cycle is defined as the tip of the abdomen swinging to one side, then to the other, then returning to the initial position. The rate at which the larva swings its abdomen is a measure of the effort of the larva during the swimming event. We also measured the smallest start angle in the bent abdomen (C-start, Brackenbury 2002) before starting their fastest swim bout.

We measured the enzymatic activity of arginine kinase (Ak) spectrophotometrically on all recovered larvae (n = 118, two samples accidently thawed before processing). Arginine kinase activity of each sample (whole body homogenate) was assayed as the rate of production of NADPH as measured by the change in absorbance at 340 nm during the linear phase of the reaction. All enzyme assays were run on a Bio-Rad Benchmark Plus microplate reader (Bio-Rad, Hercules, CA, USA) at 25 °C. Mass specific arginine kinase activity was expressed in units of μm NADPH produced × g body mass−1 × min−1.

We quantified two morphological variables on the median lamella for every larva with an unregenerated median lamella (n = 83). From previous work on Enallagma damselflies, we know that the morphometrics of the median and lateral lamellae can be captured in two principal component axes which summarize the size and shape of the lamellae, and that these axes are highly correlated with the area and circularity of the median lamella (McPeek 1995, 1997; Stoks & McPeek 2006). We digitized each lamella while viewing its lateral surface using Image-Pro Plus 5.0 (Media Cybernetics Inc., Bethesda, MD, USA) and recorded the lamella area and perimeter. We calculated an index of lamellar circularity using the formula circularity = (perimeter)²/area; this index has a minimum value of 4π for a perfect circle and becomes larger as the shape of the object becomes less circular.

Statistical analyses

We could not run an overall path analysis including phenotypes (e.g. arginine kinase), performance (swimming speed) and fitness (survival). This is because we did not have at the individual level information on phenotype, performance and fitness, we only have this for the link between phenotype and performance; fitness is measured at the enclosure level. Instead we ran separate models between these variables. We separately tested for an effect of predator treatment on mortality rate and swimming speed with mixed model an(c)ovas using proc mixed in sas 9.1; SAS Institute, Inc., (2009). For swimming speed, enclosure nested within treatment was added as a random variable to take advantage of the full dataset while overcoming the problem with pseudoreplication within enclosures (Brown & Prescott 1999; Millar & Anderson 2004). Mass was included as a covariate when analysing the effects on swimming speed. Initially, we also included interactions with mass in the model, but as none were significant, these interactions were removed from the final model.

To test whether various traits influenced swimming speed, we ran a mixed model ancova with predator treatment as categorical predictor and all swimming speed-related variables and mass as continuous predictor variables (covariates) and swimming speed as dependent variable. Initially, we also included interactions between predator treatment with the swimming speed-related variables; but as none were significant, these interactions were removed from the final model. Next, we tested for effects of predator treatment on the swimming speed-related variables in a mixed model mancova with mass as a covariate and enclosure as a random factor. Additionally, we ran separate univariate ancovas per swimming speed-related variable. Again, interactions with mass were not significant, and were removed from the final model. Because of the prior expectation for the direction of phenotypic selection, i.e. predators should select for higher swimming speeds, larger and more circular caudal lamellae, more beats, smaller start angle and higher Ak values (McPeek 1995, 1997; McPeek, Schrot & Brown 1996; Strobbe et al. 2009), statistical tests on these variables are one-tailed.

Given that we could not score the phenotype of the animals before selection, we could not run conventional logistic regressions to evaluate the presence of directional and stabilizing selection (Janzen & Stern 1998; Stinchcombe et al. 2008). Instead, we tested for the presence of directional selection on traits like arginine kinase activity by evaluating whether the regression lines of the trait against body mass differed in intercept between the caged Anax and free-ranging Anax treatments. If dragonfly predation imposed directional selection for higher Ak activity, the regression of the free-ranging Anax treatment will be displaced above the regression line of the caged Anax treatment and the main effect of treatment will be significant in the ancova. The value of the associated directional selection coefficient was estimated by dividing the difference in intercepts between the free-ranging predator and caged dragonfly treatments by the standard deviation in residuals around the caged predator regression; this metric measures the displacement of the free-ranging predator regression from the caged predator regression in phenotypic standard deviation units (see McPeek 1997). Furthermore, we tested for the presence of stabilizing selection by evaluating whether the variance of the residuals around both regression lines differed using F-tests. Under stabilizing selection, we expect the variance in residuals around the free-ranging Anax regression line to be smaller than the variance in residuals around the caged Anax regression line.


The presence of a free-ranging dragonfly predator considerably increased mortality of E. vesperum larvae (anova, F1,15 = 15·05, P = 0·0015, Fig. 2). Swimming speed was higher in larvae recovered from the enclosures with a free-ranging predator compared to larvae from enclosures with a caged predator (ancova, F1,56 = 3·66, P = 0·031, Fig. 2). The associated selection coefficient on swimming speed was +0·28 standard deviation units. Swimming speed did not differ among enclosures of the same predator treatment (Likelihood ratio test of the random effect enclosure nested in predator treatment: inline image = 0·00, = 1·00). We did not detect stabilizing selection on swimming speed (F-test, F37,20 = 1·10, P = 0·42). The effect of predator treatment on swimming speed disappeared when correcting for all swimming speed-related variables (ancova, F1,25 = ·17, P = 0·29). Besides Ak activity (partial correlation coefficient +0·54) also lamellae size (+0·75) and swimming beats (+0·62) had a significant positive contribution, while start angle (−0·84) had a significant negative contribution to swimming speed (Table 1).

Figure 2.

 Daily mortality rates and swimming speed of Enallagma vesperum larvae in the caged and free Anax predator treatments. Given are least square means (± 1 SE); swimming speed corrected for mass.

Table 1.   Results of multiple regression of swimming-related variab-les on swimming speed of Enallagma vesperum
 t8P-valueSlope (SE)Partial R
Mass (mg)−1·020·17−1·88 (1·85)−0·33
Lamella size (mm²)3·190·006415·75 (4·94)0·75
Lamella shape0·010·500·012 (1·61)0·0027
Angle−4·360·0012−0·44 (0·10)−0·84
Beats (beats s−1)2·260·0277·53 (3·34)0·62
Arginine kinase activity (μm NADPH produced × g body mass−1 × min−1)1·820·0500·51 (0·28)0·54

A mancova with mass as a covariate showed a predator treatment effect (F5,26 = 28·15, P < 0·0001) on the set of swimming speed-related variables. Separate an(c)ovas showed that predator treatment had no significant effect on mass or any of the mass-corrected swimming speed-related variables (all P > 0·25), except for Ak activity (ancova, F1,115 = 3·13, P = 0·040, Fig. 3). Ak activity was higher in larvae recovered from the enclosures with free-ranging predators compared to those from the enclosures with caged predators. The associated selection coefficient was +0·38 standard deviation units. We detected no stabilizing selection for any of the swimming speed-related variables (all P > 0·13).

Figure 3.

 Swimming speed-related variables of Enallagma vesperum larvae recovered in the caged and free Anax predator treatments. Except for mass, least square means (± 1 SE) are given corrected for mass. Arginine kinase activity is expressed as μm NADPH produced × g body mass−1 × min−1.


As in previous semi-natural field enclosure studies (McPeek 1990a, 1997, 1998; Stoks, De Block & McPeek 2005; Strobbe et al. 2009), large dragonfly larvae of the genus Anax, imposed significant mortality on Enallagma larvae. As expected, this was associated with survival selection for increased escape swimming speed and for a higher Ak activity. It is important to note that our experimental approach, i.e. comparing phenotypic distributions between caged predator and free-ranging predator treatments at the end of the exposure period, is unlikely to have biased our results. The observed patterns were all expected based on a priori knowledge of the role of swimming to survive dragonfly attacks (e.g. McPeek 1990b, 1997; McPeek, Schrot & Brown 1996), and are consistent with the biochemical role of Ak (Ellington 2001) and a macro-evolutionary reconstruction showing that increases in swimming speeds were associated with increases in Ak activity (McPeek 1999).

The alternative explanation that the observed differences in these variables are confounded with predator-induced plasticity in these traits, potentially linked with training where damselflies in the free-ranging predator enclosures that managed to avoid these attacks learned to improve their escapes via faster swimming, seems unlikely. In previous experiments designed to evaluate the degree of phenotypic plasticity in these traits, no plasticity has ever been detected (McPeek 1997; M. A. McPeek, F. Strobbe and R. Stoks, unpublished results; for Lestes damselfly larvae see Stoks et al. 1999). In addition after having videotaped thousands of larvae of these and other damselfly species and where we simulated a predator attack by coaxing larvae to swim multiple times (across a wide range of time intervals, from minutes to days), we never found a pattern consistent with training (i.e. increased swimming speed after repeated simulated attacks) (M. A. McPeek, F. Strobbe and R. Stoks, unpublished data).

Dragonfly predation selected for higher swimming speeds. This is consistent with previous studies on other Enallagma species both using field enclosures (Strobbe et al. 2009) and during staged encounters in short-term laboratory experiments (McPeek 1990b; McPeek, Schrot & Brown 1996). More generally, the few other studies, all on vertebrates, also show positive predator-imposed selection for escape speed under semi-natural conditions (overviews in Irschick et al. 2007, 2008; Janzen, Tucker & Paukstis 2007).

In line with its biochemical role in quickly recharging the ATP pool (Ellington 2001), we documented survival selection for a higher Ak activity. Typically, Ak is the first ATP-generating enzyme utilized during escape responses in invertebrates. Specifically, it converts phosphoarginine to arginine with production of ATP (Ellington 2001). Inter-individual variation in Ak activity in insects probably does not reflect allelic variation at the Ak locus as this locus has been shown to be almost monomorphic in the few species where it was studied (Collier 1990). Instead inter-individual variation in the activity of this enzyme more likely reflects differences in the quantity of the enzyme produced per unit of muscle mass, which in turn probably reflect variation in regulation (discussed more fully in McPeek 1999). The role of Ak in generating fuel for quick muscle contractions is reflected by its positive contribution to swimming speed (Tremblay, Guderley & Frechette 2006; this study). Additional proof for survival selection on Ak operating through swimming speed itself comes from the observation that when adding swimming speed as a covariate to the analysis, selection on Ak was no longer significant (ancova, F1,53 = 3·13, P = 0·45). We consider the selection on swimming speed therefore to be direct selection and the selection on Ak activity to be indirect consequence of selection on swimming speed. The reported selection coefficient (+0·38 standard deviation units) on Ak is relatively strong but falls within the range observed in other studies, mostly on morphological and life history traits (median = 0·15–0·16 standard deviation units, Kingsolver et al. 2001; Siepielski, DiBattista & Carlson 2009). In contrast, in a similar study on two other Enallagma species we could not detect survival selection on Ak activity (Strobbe et al. 2009). In the latter study, however, Ak also was not related to swimming speed. One reason for differences between studies may be that the latter study included Enallagma species that evolved higher swimming speeds after a habitat shift and in association evolved more than double the levels of Ak activity (McPeek 1999; Strobbe et al. 2009). In these species, no ongoing directional selection on Ak may be detectable as Ak levels may have been optimal for most larvae thereby showing less variation for selection to act upon.

While other morphological (lamellae size) and behavioural (start angle of the C-start and swimming beat frequency) also contributed considerably to swimming speed as indicated by their partial correlations, we could not detect selection on them. Potentially functional redundancy (also known as many-to-one mapping, Alfaro, Bolnick & Wainwright 2004; Wainwright et al. 2005) may have played a role. Under this scenario, animals may achieve a swimming speed high enough to survive attacks by large dragonfly larvae through different trait combinations. A larva with suboptimal values for one swimming speed-related variable (e.g. small lamellae) may end up with a high enough swimming speed when having optimal values for one or more other variables (e.g. a small start angle) (see also Strobbe et al. 2009). A comparable situation has been described in Anolis lizards where similar levels of sprint performance can be achieved through different morphological traits (limb segment lengths and muscle mass) (Vanhooydonck et al. 2006). Note that two of the traits where we could not detect selection were behavioural traits. So far, very few studies have considered selection on behavioural traits (Kingsolver et al. 2001; Siepielski, DiBattista & Carlson 2009).

Our study adds to the few others that demonstrate survival selection by predation on a performance trait and the underlying phenotypic trait (Jayne & Bennett 1990; Downes & Shine 1999; Miles 2004; Janzen, Tucker & Paukstis 2007; Strobbe et al. 2009). More general, selection on physiological traits remains largely unknown (Kingsolver et al. 2001; Siepielski, DiBattista & Carlson 2009). A notable exception is the study by Artacho & Nespolo (2009), showing negative selection (−0·11 standard deviation units) on resting metabolic rate in the garden snail Helix aspersa. Together with our study, this suggests that selection on physiological traits may be as high as selection on morphology and life history traits. Obviously, more studies are needed to confirm this. We hypothesize that enzymes involved in energy replenishment and thus may contribute to escape performance traits like swimming (for an example in vertebrates see Watkins 2000), may frequently be under strong survival selection by predation.


F.S. was supported as a PhD fellow of the Institute for the Promotion of Innovation through Sciences and Technology in Flanders (IWT) and M.D.B. is postdoctoral fellow of the Research Foundation Flanders (FWO). This study was funded by research grants from FWO and the KULeuven Research Fund to L.D.M. and R.S. and from NSF to M.A.M.