Discussion

The principal result of this study is that the effect of predation risk on the abundance of a given prey taxon varied strongly across experiments conducted under similar conditions, objectives, measurables and implementation; the effect of caged fish on the abundance of the same taxa ranged among experiments from negative or no effect, positive or no effect, and negative, positive or no effect. A multivariate analysis similarly indicated an influence of caged fish on the zooplankton community composition that varied among experiments. Our approach is conservative in nature: variability in unmeasured factors that influence risk effects is expected to be much larger in nature than what we found in our experiment. Our study indicates that if a predator is found to have a positive or negative risk effect on prey abundance in a particular system, or at a given time, this alone may yield limited information on the existence or magnitude of a risk effect on prey abundance in another location or time.
Predation-risk effects result from a predator inducing changes in prey phenotype (e.g., behavior, morphology), which can lead to changes in prey fitness components (e.g., fecundity, survival), which in turn influences prey population growth rate and abundance (Sheriff et al. 2020, Wirsing et al. 2020), and contingency could arise at any one of these scales. Empirical studies have shown that the magnitude, and even existence, of a trait response is contingent on the balance of costs and benefits associated with the trait change, each of which is strongly dependent on biotic and abiotic factors, such as resource level, the presence of other predators, con- and heterospecific density, and abiotic factors such as temperature and landscape features (Peacor et al. 2013, Thaker et al. 2011, Tollrian et al. 2015, Donelan et. al. 2017, Wirsing et al. 2020). For example, prey are predicted to show a lower response at lower resource levels, which can be affected by many factors including competitors.
The effect of a trait response on fitness components (i.e., NCEs) has also been shown to be strongly contingent on abiotic and biotic factors. For example, the cost of a habitat shift (the trait response) will depend on consequent experienced changes in temperature, resources, and mortality risk from other causes (e.g. parasites or other predators). Further, it has been shown that the level of trait responses of intraspecific or interspecific competitors to predators can cause the NCE on a focal prey’s fitness to be magnified, dampened and even reversed (Relyea 2000, Peacor and Werner 2004).
Previous studies that intentionally manipulate aspects of the abiotic or biotic environment thus clearly show contingency in risk effects on both trait and fitness components, and altering either response could alter predation-risk effects on prey abundance. Indeed, theoretical studies show very clearly that altering responses “upstream” of prey abundance can alter abundance responses (Abrams and Vos 2003; Bolker et al. 2003; Peacor and Cressler 2012). However, this is the first empirical study to show contingency in risk effects on prey abundance.
Moreover, there was high within-treatment replicate variability compared to typical NCE experiments. For example, different replicate mesocosms within an experiment often varied in characteristics across experiments such as periphyton growth and zooplankton taxa density (density coefficient of variation equal to 1.2 ± 0.62 [mean ± stdev] within treatments for all taxa across the four experiments). The high levels of variability observed within treatments, despite the relatively simplicity of the environment, has two important implications. First, high variation within treatments would make significant risk effects less likely to be observed, as any effect of the caged fish on prey density had to be large to be statistically detectable over the high within-replicate variation. Second, the fact that we observed contingency despite the high similarity among experiments relative to differences among natural systems, and despite the high within treatment variation relative to typical experiments of risk effects, suggests that this study provides a conservative test of the contingency of risk effects on prey abundance in natural settings.
Nearly all of the important species in our experimental communities have been reported to be components of bluegill diet in lakes and ponds. In particular, Daphnia , Ceriodaphnia , Bosmina ,Chydorus , copepods, ostracods, and Simocephalus can at times be a dominant or substantial component of bluegill diets (e.g., Hall et al. 1970, Keast 1985, Rettig and Mittelbach 2002, Dewey 1997, Bremigan and Stein 1994). Consequently, these species have an evolutionary history of risk to bluegill predation, and indeed, many have been shown to respond phenotypically to presence of fish, including in some cases bluegill, and to other predators. For example, a large number of studies have documented phenotypic response in Daphniato a wide array of predators including bluegill (see reviews in Lass and Spaak (2003) and Diel et al. (2020)). Diel et al. (2020) also report evidence of phenotypic responses to predators in Bosmina(including to fish), Diaphanosoma , Simocephalus (including to fish), Ceriodaphnia (including to fish), copepods, andChydorus (to injured conspecifics, Pecor et al. 2016). In experiments of the same venue used here, we have further observed fish kairomones to affect the spatial position of Alona, Bosmina, Chydorus, Diaphanosoma, Scapholeberis and ostracods (unpublished data). Thus, many of the important components of the zooplankton communities in the experiments have the potential to respond to the presence of bluegill kairomones and initiate effects we see of bluegill presence on community structure.
Perhaps surprisingly, there was no risk effect of caged fish on the abundance of some vulnerable zooplankton prey in some or multiple experiments. For example, using bluegill of the same size and origin, we found that bluegill preference (as measured by Chesson α selectivity index) was highest for D. pulex and Ceriodaphnia (Rafalski et al. 2023) which is consistent with other studies that report preferred prey of small bluegill (Bremigan and Stein 1994). Yet, there was no NCE on D. pulex or Ceriodaphnia abundance in any of the four experiments, (though there is a strong negative trend in Exp. 2 on Ceriodaphnia ). The lack of an effect cannot be explained by the absence of a cue of predation risk in the experimental venue (e.g., kairomones can be affected by properties of the water (Peacor 2006, Turner and Chislock 2010)) because other species were affected by caged fish in Exp.1, 2 and 4, while in Exp. 3 multiple species responded behaviorally to caged fish (Rafalski et al. unpublished). A power analysis confirmed that the lack of any significant effects in Exp. 3 was not due to a lower sample size or higher variance in that experiment, but rather to a true lack of effect of fish on the densities of any species (Supporting Information). Our results therefore highlight that predation risk may have little or no effect on the abundance of highly vulnerable prey (Hoverman and Relyea 2012), and the lack of any significant responses on any species in one of four experiments further highlights the contingency of the risk effects found in this study.
It is beyond the scope of this study to identify the specific mechanisms responsible for the observed contingency in NCEs in our study, but we can speculate. Incidental differences in factors such as temperature and initial relative abundances could lead to variation in community assembly processes across experiments (Drake 1991, Chase 2003) that could lead to the differences in species densities observed among experiments, and those differences in species densities could influence predation risk effects. For example, in Exp. 2 Ceriodaphniadensity was higher in the no-fish treatment than it was in the other experiments, and it was the most abundant zooplankton taxon (Table 1, Fig. 2). We have found that Ceriodaphnia responds behaviorally to caged bluegill by moving horizontally away from the tank center (unpublished data) that would likely be associated with reduced foraging rates or freeing of resources in the center of the tanks. The indirect benefit to other zooplankton taxa due to competitive release would increase as a function of Ceriodaphnia density, and thus would be most likely observed in Exp. 2 where indeed positive effects were observed on a number of taxa. There was a strong non-significant trend for predation risk to reduce Ceriodaphnia density in Exp. 2, which could have further contributed to the observed positive effects on other species densities. Additionally, the length of the experiments (relative to generation times; even the species with the longest generation time experienced at least three generations; Peacor et al 2012), allowed ample time for incidental differences among experiments to generate ecological feedbacks, for example, generating indirect effects on resources. Indirect positive effects on resources could also vary among experiments due to differences in species densities among experiments, or a predator-induced reduction in foraging on particular phytoplankton species that facilitated growth of other phytoplankton species favored by different zooplankton species.
The presence of Hydra in Exp. 1 could add an additional mechanism by which caged fish influence zooplankton and hence cause contingency. If caged fish induce a habitat shift in zooplankton to be closer or further from surfaces where Hydra reside, this would make them more or less vulnerable to Hydra predation, respectively. Ensuing effects on density of one zooplankton species could then have indirect effects on the abundance of other zooplankton species. Note that effects of Hydra cannot be responsible for all of the contingencies observed, as different effects of caged fish remain if Exp. 1 is omitted (top box, Table 1).
It is also possible that some contingency is manifest due to species composition differences in broader taxonomic groups in our study (e.g., cyclopods and calanoids) across experiments (and such species reacted differently to the bluegills). This possibility, however, does not apply to more circumscribed categories, for example there is only one species of Chydorus in the region (Chydorus sphaericus) , and caged fish had a positive, neutral, and negative effect on Chydorusdensity.
It is important to establish the degree to which risk effects on prey abundance are contingent for at least two reasons. First, contingency will make it challenging to translate findings of strong risk effects in simple experimental settings to natural communities, or translate strong trait-responses of prey seen in the field to a consistent prediction of ensuing effects on prey fitness and abundance. Second, much of the literature on risk effects has addressed whether they are strong relative to consumptive effects (CEs). For example, empirical (Werner and Peacor 2006) and meta-analysis studies (Preisser et al. 2005) have compared the influence of risk effects and CEs, with an implicit assumption that they contribute different relative fractions of the net effect of a predator. But if risk effects on prey abundance are strongly contingent, then it is more difficult to make inferences from single reports of the magnitude of risk effects. If contingencies of risk effects on prey abundance are important, this highlights the need to understand the mechanisms underlying risk effects, and how environmental factors will influence the consequences of risk effects. Resolving such questions has clear implications for ecological theory and attempts to include the effects of risk effects into conservation and management.
We have demonstrated that predation risk effects on abundance of prey species can be highly contingent on conditions, varying even in sign, using methodology that likely underestimates the influence of contingency in natural systems. Many investigators state that risk effects are known to be important, implicitly implying that risk effects influence the abundance of species in natural systems (Sheriff et al. 2020). But few studies, in very few ecological systems, have actually tested for such effects on abundance in natural systems (Peacor et al. 2022), and until such tests are performed, our results on the contingency of risk effects suggest that results from experimental work may not translate to natural systems. Rather, our results highlight the necessity of examining the underlying mechanisms before predicting the consequences of predation risk to natural systems or applying these ideas to problems in conservation and management. Indeed, theoretical studies have argued that risk effects have been exaggerated by omitting crucial factors such as stage structure that can decrease the influence of risk effects (Luttbeg et al. 2003; Persson and De Roos 2003), or by methodological problems used in experimental studies (Abrams 2010). Identifying the mechanisms underlying risk effects will allow us to understand the contingencies that can arise in the cascading effects of predation risk on prey traits, ensuing effects on prey fitness components and indirect effects on resources, competitors and other predators, which combine to influence prey abundance, and make informed predictions of the influence of risk.