Predator cue studies reveal strong trait-mediated effects in communities despite variation in experimental designs

Nonconsumptive or trait-mediated effects of predators on their prey often outweigh density-mediated interactions where predators consume prey. For instance, predator presence can alter prey behaviour, physiology, morphology and/or development. Despite a burgeoning literature, our ability to identify general patterns in prey behavioural responses may be in ﬂ uenced by the inconsistent methodologies of predator cue experiments used to assess trait-mediated effects. We therefore conducted a meta-analysis to highlight variables (e.g. water type, predator husbandry, exposure time) that may in ﬂ uence invertebrate prey ’ s behavioural responses to ﬁ sh predator cues. This revealed that changes in prey activity and refuge use were remarkably consistent overall, despite wide differences in experimental methodologies. Our meta-analysis shows that invertebrates altered their behaviour to predator cues of both ﬁ sh that were fed the focal invertebrate and those that were fed other prey types, which suggests that invertebrates were not responding to speci ﬁ c diet information in the ﬁ sh cues. Invertebrates also altered their behaviour regardless of predator cue addition regimes and ﬁ sh satiation levels. Cue intensity and exposure time did not have signi ﬁ cant effects on invertebrate behaviour. We also highlight that potentially confounding factors, such as parasitism, were rarely recorded in suf ﬁ cient detail to assess the magnitude of their effects. By examining the likelihood of detecting trait-mediated effects under large variations in experimental design, our study demonstrates that trait-mediated effects are likely to have pervasive and powerful in ﬂ uences in nature. 2013 The Authors. Study Animal

Nonconsumptive or trait-mediated effects of predators on their prey often outweigh density-mediated interactions where predators consume prey. For instance, predator presence can alter prey behaviour, physiology, morphology and/or development. Despite a burgeoning literature, our ability to identify general patterns in prey behavioural responses may be influenced by the inconsistent methodologies of predator cue experiments used to assess trait-mediated effects. We therefore conducted a meta-analysis to highlight variables (e.g. water type, predator husbandry, exposure time) that may influence invertebrate prey's behavioural responses to fish predator cues. This revealed that changes in prey activity and refuge use were remarkably consistent overall, despite wide differences in experimental methodologies. Our meta-analysis shows that invertebrates altered their behaviour to predator cues of both fish that were fed the focal invertebrate and those that were fed other prey types, which suggests that invertebrates were not responding to specific diet information in the fish cues. Invertebrates also altered their behaviour regardless of predator cue addition regimes and fish satiation levels. Cue intensity and exposure time did not have significant effects on invertebrate behaviour. We also highlight that potentially confounding factors, such as parasitism, were rarely recorded in sufficient detail to assess the magnitude of their effects. By examining the likelihood of detecting trait-mediated effects under large variations in experimental design, our study demonstrates that trait-mediated effects are likely to have pervasive and powerful influences in nature. Ó 2013 The Authors. Published on behalf of The Association for the Study of Animal Behaviour by Elsevier Ltd. All rights reserved.
The impact of nonconsumptive or trait-mediated effects of predators on their prey can be strong, often outweighing the effect of density-mediated interactions where predators directly consume prey (Preisser et al. 2005). Trait-mediated effects have an impact on prey populations because predators influence prey behaviour, development, morphology and/or physiology (Peacor & Werner 2001;Werner & Peacor 2003;Frommen et al. 2011). Additionally, trait-mediated indirect effects may radiate throughout the community as predators affect competitors of the prey and resources (Schmitz et al. 2004;Mowles et al. 2011;Gosnell & Gaines 2012). For example, increased refuge use by small-mouthed salamanders, Ambystoma barbouri, in response to predation risk was shown to have positive effects on their isopod prey (Huang & Sih 1991). There is much current interest in the role such trait-mediated indirect effects play in community ecology; they may be important drivers of population dynamics (Alexander et al. 2013) and community structure (Ohgushi et al. 2012), influential components of hoste parasite interactions (Hatcher & Dunn 2011) and drivers of biological invasions (White et al. 2006;Dunn et al. 2012).
A standard method for quantifying trait-mediated effects is measuring behavioural responses to predator cues (e.g. Richmond & Lasenby 2006;Dalesman et al. 2007;Dunn et al. 2008). Despite a burgeoning literature in this research field (>180 predator cue studies in aquatic environments, ISI Web of Science), considerable variation in prey responses to predator cues exists. For instance, some studies report increased prey activity in response to predator cues (e.g. Scrimgeour & Culp 1994;Miyasaka & Nakano 2001), whereas others report decreased prey activity (e.g. Åbjörnsson et al. 2000;Dezfuli et al. 2003). Although these differences may be partially explained by predator-specific responses of prey (e.g. refuge use by aquatic snails increases in response to a pelagic fish predator, but decreases to avoid a benthic crayfish predator, Turner et al. 1999), variation in experimental design may further confound the outcome of predator cue studies. These confounding factors include cue intensity, degradation rate, addition regime and presence of predator diet cues or alarm substances from consumed conspecifics or heterospecifics released during predation events, the water type, prey functional feeding group and familiarity with the predator and satiation level of the predator, among others.
Predator cue intensity varies widely among studies, and therefore may affect the ability of prey to detect predators and estimate their relative proximity (Dickey & McCarthy 2007;Ferrari et al. 2007). Similarly, cue degradation time frames are likely to be influenced by differences in sunlight and microbial activity affecting cue breakdown rates (Ferrari et al. 2007), coupled with varying cue exposure times (e.g. 4 weeks, Åbjörnsson et al. 2000;5 min, Dunn et al. 2008). Despite this, few studies assess predator cue efficacy (e.g. Hazlett 1999;Ferrari et al. 2007;Wisenden et al. 2009), with most studies relying instead on the prompt use of a cue after its production. Although long-term studies may avoid cue degradation effects by housing predators with focal prey, additional problems of habituation to predator cues may confound results (e.g. Gammarus pulex amphipods no longer reduced leaf consumption following 4 weeks of continuous exposure to sculpin, Cottus gobius, Åbjörnsson et al. 2000). Furthermore, some water types (e.g. indoor experiments using dechlorinated tap water) may alter natural degradation processes to extend cue efficacies beyond their natural 'shelf lives' (Ferrari et al. 2007), offering an explanation as to why prey exposed to old/frozen cues display antipredator responses (e.g. Wudkevich et al. 1997;Pettersson et al. 2000).
Predator cue studies rarely consider how prey functional feeding group (e.g. carnivore, omnivore, filter-feeder; MacNeil et al. 1997) may influence whether prey respond to cues as a predation threat or a potential food resource. Additionally, the information that the cue conveys about the predator, and thus the potential risk of predation to the prey, varies with predator satiation level (e.g. Åbjörnsson et al. 1997), as well as the presence/absence of diet or alarm cues from consumed conspecifics or heterospecifics (e.g. Huryn & Chivers 1999). Indeed, studies may provide predators with either the focal invertebrates (e.g. Åbjörnsson et al. 2000;Bernot & Turner 2001) or heterospecific invertebrates as a food source (e.g. Gyssels & Stoks 2005;Wohlfahrt et al. 2006), or hold predators without food entirely (e.g. Mathis & Hoback 1997;Miyasaka & Nakano 2001). Furthermore, predator identity may be important for prey to mount appropriate behavioural responses to known predators (Henry et al. 2010), whereas prey may be unable to recognize predation risks posed by novel predators (Cox & Lima 2006). However, prey exposed to unfamiliar predators may benefit from diet information provided in the cue to convey predation risk or, alternatively, display fixed antipredator responses that can be activated with novel predators (Sih et al. 2010).
To determine whether the experimental design of predator cue studies influences whether trait-mediated effects will be detected, we undertook a quantitative literature review using a 'flexible' (sensu Nakagawa et al. 2007) meta-analytical approach. Specifically, we examined the influence of 10 experimental design factors, including water type, fish satiation, cue intensity and exposure time, on invertebrate prey activity and refuge use observed in fish predator cue experiments. We also assessed publication bias, which is a common source of criticism in meta-analyses since studies with significant results are more likely to be published (the 'file drawer' problem, Rosenthal 1979).

Data Collection
Studies investigating the behavioural responses of aquatic invertebrates to predator cues were obtained from literature databases and internet searches (pre June 2012), and were primarily selected according to the following criteria: (1) published in English; (2) predator cues derived from fish; (3) macroinvertebrate prey; (4) experimental study of a freshwater system rather than field-based observations (meta-analysis search terms: (fish*) AND (aquatic OR freshwater) AND (cue OR kairomones OR odour) AND (invertebrate* OR macroinvertebrate* OR insect*)). We focused on chemical odour cues since turbidity and/or a prey's visual ability in aquatic environments often impairs visual recognition of predators (Chivers & Smith 1998;Wisenden 2000). We included only those studies that measured the effect of predator cue on invertebrate activity or refuge use, because we did not consider other behavioural measures, such as latency of pairing, to be immediate responses to predation threats. Furthermore, our final data set includes only those studies that reported the control and treatment sample sizes, and the effect size, or another measure from which the effect size could be calculated (e.g. test statistic, mean and standard deviation or error). We also contacted corresponding authors of publications where data required to calculate effect sizes could not be extracted from published text or figures.

Calculation of Effect Sizes
We calculated the effect size Cohen's d (also known as Hedge's g, maximum likelihood estimator) for each measure of activity or refuge use (e.g. change in drift rate or position on substrate), then converted each effect size into the standardized mean difference effect size g. As effect sizes (the standardized mean difference between control and treatment group) were seldom reported in published papers, we calculated the effect size for each study by (1) transforming the reported statistic (e.g. t, F), or (2) the reported mean and SE or SD of the control and treatment groups using methods outlined by Rosenthal (1994). As F statistics were often reported from more than one treatment (e.g. control versus cue from multiple predator types; df > 1), effect sizes were also calculated from control and treatment means extracted from figures using DataThief (Tummers 2006).

Moderator Variables
We selected 10 moderator variables (fixed effects) from the original studies that potentially influence aquatic invertebrate responses to fish cue (see Appendix Table A1). Another unaccounted variable, parasite infection status, may be relevant but was seldom reported unless the influence of parasitism was the focus of the investigation, with such studies removed from further analysis.

Statistical Procedures
All statistical analyses were computed in R (version 2.13.1, R Development Core Team 2011). Linear mixed-effect models were used to conduct mixed-effects meta-analyses (Pinheiro et al. 2013). Outliers were removed (by visual inspection of funnel plots) before we fitted models for g using the restricted maximum likelihood estimation. Our preliminary analysis demonstrated that effects of fish cue were unlikely to be revealed from the pooled invertebrate data set because pooling effect sizes from invertebrates that respond differently to the threat of predation would generate 95% confidence intervals that bounded zero (see Appendix Tables A2,  A3). Therefore, we used absolute effect sizes to examine the effect of experimental design, as the magnitude of the change in behaviour rather than the direction of change (i.e. increasing or decreasing activity or refuge use appropriate to the invertebrate) was of interest.
To estimate between-study variability, we used Study ID as a random factor in our analysis. Although prey species and predator species might be considered random factors (see Nakagawa et al. 2007;Nakagawa & Hauber 2011), there were insufficient observations to fit these predictors without overparameterizing the model. The I 2 statistic (Higgins et al. 2001;Nakagawa & Santos 2012) was used to calculate the heterogeneity (degree of consistency among studies). Delta Akaike information criterion (DAIC; mixed modele random only model) fitted with maximum likelihood estimation was used to examine whether any of the a priori fixed effects improved model fits (see Appendix Table A1). Each fixed effect was included in a separate meta-analytical model, because few studies provided information on all predictor variables, with a minimum of eight studies for each predictor considered necessary for analysis (Nakagawa et al. 2007). Continuous variables (cue intensity, exposure time) were centred on the mean and scaled by two times the standard deviation (Gelman 2008). We report the effect size estimates for each model representing intercepts for categorical factors, and slopes for continuous variables. To determine whether estimates were different from zero (i.e. no effect) we used 95% confidence intervals and tested statistical significance using P values from z approximations of t values because degrees of freedom are difficult to specify from mixed-effect models. Contrast analyses were constructed for each model to assess whether the factors in each predictor variable differed, with significant contrasts indicated in the results only (see Appendix Tables A4, A5).
Publication bias was assessed by constructing funnel plots to examine graphically the relationship between effect size (original g) and sample size for activity and refuge use, with absence of publication bias indicated by decreasing effect sizes with increasing sample size (Sterne et al. 2005). We also calculated the Spearman rank correlation to examine statistically the relationship between effect size and sample size. If a significant relationship was detected, we then used the Rosenberg (2005) fail-safe number calculator (metafor package, Viechtbauer 2010) to estimate the number of additional studies averaging null results that would be required to reduce the significance level of the average effect size to the commonly accepted level of statistical significance of a ¼ 0.05. We assumed that, if the fail-safe number was larger than 5n þ 10 where n is the number of studies, the results were robust regardless of publication bias.

Meta-analysis
Twenty-eight original studies met the criteria for inclusion in the meta-analysis. These involved a total of 28 invertebrate and 29 fish species, from which 66 effect size estimates of activity and 39 refuge use responses were obtained (see Appendix Tables A2, A3). The majority of studies involved Ephemeroptera (N ¼ 7), Gastropoda (N ¼ 7), Amphipoda (N ¼ 5) and Odonata (N ¼ 5).

Activity
Overall, we found that fish cues altered invertebrate prey activity (t test: z ¼ 6.05, P < 0.0001), with the I 2 statistic indicating that Study ID accounts for most of the heterogeneity in the data (Table 1, Fig. 1a). Of the three invertebrate types for which there were sufficient studies, Amphipoda and Ephemeroptera altered activity in the presence of a cue (t test: z ¼ 4.11, P < 0.0001; z ¼ 4.53, P < 0.0001), while Odonata did not (z ¼ 0.93, P ¼ 0.352). All invertebrate functional feeding groups altered activity in the presence of a cue (Table 1, Fig. 1a).
Invertebrates altered their activity in response to cues from familiar fish species (Table 1; insufficient data to test for a response to novel fish), regardless of whether the fish were fed conspecific invertebrates or other food sources, whether or not the fish was starved, or whether the fish cue was added once or continuously, with no difference in the magnitude of the effects within each predictor. Invertebrates were more likely to alter their activity when the fish cue was provided from a fish not physically present in the experimental tank (contrast [effect size Fish in tank Yes À effect size Fish in tank No ]: t test: z ¼ 2.12, P ¼ 0.034). Fish cues provided in tap water resulted in highly variable, nonsignificant effect sizes, whereas invertebrates exposed to a fish cue in dechlorinated, ground or stream water showed altered activity. Neither cue intensity nor exposure time showed a relationship with activity effect sizes.

Refuge Use
Fish cues altered invertebrate refuge use overall, with the I 2 statistic also indicating that the random factor Study ID accounts for much of the heterogeneity between studies (Table 2, Fig. 1b). Gastropoda and Ephemeroptera (insufficient data for Amphipoda) both altered refuge use in the presence of a cue; however, the cue had a greater influence on Gastropoda (contrast [effect size Ephemeroptera À effect size Gastropoda ]: t test: z ¼ 2.02, P ¼ 0.004). Invertebrates in the functional feeding group 'grazer' also altered their refuge use in the presence of a fish cue (t test: z ¼ 5.02, P < 0.0001; insufficient studies for other groups).
Invertebrates altered their refuge use regardless of familiarity to the fish species, whether or not the fish was in the experimental tank, fish satiation levels or cue addition regime, with no difference in the magnitude of the effects within each predictor. Cues from fish that were fed invertebrate conspecifics and cues provided in stream water significantly altered refuge use effect sizes (insufficient data for fish that were fed other invertebrates and other water types). Cue intensity and exposure time did not have a significant effect on invertebrate refuge use.

Publication Bias
The Spearman rank correlation coefficient for activity suggested a relationship between effect size and sample size across studies (r S ¼ 0.349, N ¼ 66, P ¼ 0.004). However, visual inspection of the funnel plot (Fig. 2a) showed that this publication bias was not severe. This conclusion was also supported by the Rosenberg fail-safe number, which indicated an additional 1214 studies averaging null results would be required to reduce the significance of the average effect size below a ¼ 0.05. For refuge use, the funnel plot (Fig. 2b) and Spearman rank correlation coefficient (r S ¼ À0.250, N ¼ 39, P ¼ 0.124) indicated the absence of publication bias.

DISCUSSION
Predator cue studies are a frequently utilized approach when assessing the potential trait-mediated effects of predators on prey (e.g. Trussel et al. 2003;Dalesman et al. 2007;Griffen et al. 2012). Our meta-analyses indicate that, despite the very considerable differences in methodologies employed in predator cue experiments, effect sizes were remarkably consistent (with the exception of tap water), indicating that predator cue experiments are relatively robust to differences in experimental design. Variation in tap water quality offers an explanation of the inconsistency of tap water effect sizes, since tap water may be chlorinated in some locations, whereas it may be sourced directly from ground water elsewhere. The consistent signal of predator cue effects on prey behaviour, despite variations in experimental design, lends further weight to current proposals that trait-mediated indirect effects are pervasive and powerful influences in nature (Dunn et al. 2012;Ohgushi et al. 2012).
When the original effect sizes of invertebrates in predator cue studies are examined, it may appear that few invertebrate taxa or functional feeding groups show consistent behavioural responses to predator cues (see Appendix Tables A2, A3). However, these differences are likely to reflect both the prey-and/or predatorspecific responses (e.g. fast-moving prey increase activity to escape predators; prey increase refuge use to avoid pelagic predators). Prey exhibiting inappropriate or unnecessary predator avoidance behaviour may face penalties in terms of reduced foraging and reproductive outputs (Dunn et al. 2008), in addition to increased predation risk from other predators (Chivers & Smith 1995;Åbjörnsson et al. 2004). Therefore, prey benefit from the ability to detect and respond appropriately to cues that indicate potential predation risk (e.g. Wisenden et al. 1997;Mirza & Chivers 2003;Richmond & Lasenby 2006).
However, the appropriateness of a particular behavioural response of an invertebrate to a 'predator' cue may not be fully evaluated since few studies consider the functional feeding group of the invertebrate species itself. This is of particular importance in studies that focus on the behaviour of invertebrates known to consume tissues of live and/or dead fish (e.g. Gammarus amphipods, reviewed in MacNeil et al. 1997; notonectid waterbugs, Papá cek 2001; odonates, Mobley et al. 2013). With such omnivorous 'prey' species, conclusions must be cautiously drawn from cue studies, since observed behaviour may not be strictly that of an invertebrate prey avoiding a fish predator, and may in fact be a feeding response.
Invertebrates showed behavioural responses to the cues of both familiar and unfamiliar fish species, indicating a general ability to perceive and respond to the potential risk of predation posed by novel predators, which may become increasingly important as freshwater communities face mounting pressure from the introduction of exotic species (Strayer 2010). Previous studies have suggested that invertebrates may display innate (general) predator responses to novel predation threats, or use diet cues to learn and respond rapidly to novel predator cues (Wisenden & Millard 2001;Sih et al. 2010). Our meta-analysis shows that invertebrates altered their behaviour to predator cues of both fish that were fed the focal invertebrate and those that were fed other prey types, which suggests that invertebrates were not responding to specific diet information in the fish cues. Additionally, satiation levels of the fish did not have a strong influence on whether invertebrates altered their behaviour.
Both the presence and absence of the predatory fish in the experimental tank resulted in invertebrates altering their refuge use, whereas invertebrates altered their activity only when fish were not in the tank. This suggests that invertebrates may adjust their predator avoidance strategies based on additional information obtained from their physical environment. If the exact location of the fish is unknown (i.e. is outside the experimental tank or behind an opaque barrier), and only a chemical cue of its presence is available, then the best strategy for an invertebrate to avoid predation may be to alter its behaviour.
Previous studies have suggested that changes in cue intensity provide prey with a method of assessing predation risk based on the density of the predators, as well as their temporal and spatial proximity (Ferrari et al. 2006). In contrast, our results suggest that invertebrates respond in a similar fashion regardless of the intensity of the cue. This behavioural trait is likely to be advantageous in avoiding being consumed since the appropriate behavioural response required to avoid a single predator is likely to be relevant if there are multiple predators (of the same species) present. If prey respond differently to predator number or proximity, then our results suggest that cue intensity alone may not be sufficient for prey to distinguish between these threats. Indeed, prey may respond to predation threats by utilizing multiple cues in an additive manner as proposed in the 'sensory complement' hypothesis (Lima & Steury 2005). However, we cannot discount the possibility that the intensity of cues used in these studies was sufficiently high to mask otherwise subtle effects of predator number or proximity (i.e. studies should use more realistic (low) concentrations of predator cue). In this study, our ability to evaluate fully the influence of a number of experimental design factors was limited owing to a lack of studies, which in some cases was further confounded by available studies failing to report effect sizes or statistics and/or figures from which effect sizes could be estimated. For example, fewer than eight refuge use studies used water types other than 'stream' and thus the influence of other water types could not be evaluated. In contrast, sufficient invertebrate activity studies were available for four different water types, which indicated that experiments should avoid tap water since highly variable effect sizes were likely to be generated. The ability for meta-analyses to assess the overall effect of predator cues on prey behaviour relies directly on the access to effect size statistics, and thus their inclusion should be encouraged in future studies. In other instances, factors such as cue degradation are not routinely assessed when designing predator cue studies, and thus little inference could be made on their effect. Likewise, we found parasite infection status was rarely reported, despite trophically transmitted parasites frequently altering the behaviour of their intermediate hosts to enhance their transmission to the predatory definitive host (e.g. Thomas et al. 2005). For example, G. pulex amphipods infected with the fish acanthocephalan Pomphorhynchus laevis prefer water containing the odour of perch, Perca fluviatilis (a known definitive host, Baldauf et al. 2007); while Medoc & Beisel (2008) demonstrated increased escape performance of Polymorphus minutus infected with Gammarus roeseli amphipods in response to a nonhost predator. Indeed, there is growing evidence that many parasites, including many that are not trophically transmitted, influence host behaviour and thereby induce trait-mediated indirect effects on species with which the host interacts (reviewed in Hatcher & Dunn 2011). This is particularly relevant for predatoreprey studies because parasites can alter both host vulnerability to predation and, for predatory host species, their predation rate. Thus, future predator cue studies would benefit from ensuring prey are not parasitized when the influence of parasitism is not of interest.
In conclusion, our study highlights that when variations resulting from choice of cue and response variables, and adaptive underpinning of response in relation to prey functional or taxonomic group, are properly accounted for, fish predatoreinvertebrate prey studies are remarkably robust to differences in experimental design. Thus, the standardization of predator cue experimental designs may not be required in order to assess the strong influences of predator cue on prey behaviour. Furthermore, this study provides evidence to suggest that trait-mediated effects are powerful drivers of ecological and evolutionary processes that define prey populations, and the resources with which they interact.

Acknowledgments
We thank Shinichi Nakagawa and Mathieu Lundy for statistical advice, and Robert Elwood and two anonymous referees for constructive comments. This manuscript was funded by NERC grant NE/G015201/1.     Invertebrate (