Box 1 : State dependent foraging games between gerbil prey and owl predators
The interaction of predator and prey is a state-dependent foraging game where the prey must manage risk using time allocation and vigilance (Brown 1999), and the predators must manage fear: as prey become more afraid, they become less catchable. The predator’s tools include time allocation and derring-do; a more daring predator is more willing to risk injury in order to capture its prey (Brown et al . 2016). Here we focus on Allenby’s gerbil (Gerbillus andersoni allenbyi ), a nocturnal rodent of sand dunes in the Middle East, and its barn owl (Tyto alba ) predator. Within an outdoor vivarium (17 x 34 x 4.5 m), it is possible to manipulate the energetic states, and subsequently quantify the foraging behavior, of both gerbils and owls (Kotleret al . 2004).
In theory, a forager should exploit depletable resource patches until the benefits of its harvest rate no longer exceed the sum of energetic, predation, and missed opportunity costs of foraging (Brown 1988). The food density at which this occurs is called the giving-up density (GUD) and is a behavioral indicator of foraging costs for that context. Energetic costs of foraging and risk factors should all lead to higher GUDs, and do so in gerbils (Kotler et al . 1991; Kotler et al . 1993). The predation cost is highly state-dependent as it equals predation risk multiplied by the survivor’s fitness divided by the marginal value of the food. Hungry animals and those in a low state or with poor prospects should be less fearful and have lower GUDs.
In vivarium experiments, gerbils that received supplemental food, relative to those that did not, used food patches less intensively, had higher GUDs, and avoided risky open microhabitat (Kotler 1997; Kotleret al . 2004). These effects carried over into the subsequent night when no gerbils received supplemental food. Gerbils that had received supplemental food previously responded more strongly to owls than those that did not (Kotler 1997). These results show how a higher energetic state acts to magnify foraging costs and alter behaviors, ultimately leading to diminished risk taking during phase two.
Tracking gerbil foraging over the course of lunar cycles revealed the dynamic nature of risk management and feedbacks with state (Kotleret al . 2010). Starting at new moon, as the moon waxes, gerbils increased vigilance to counter the greater ease of predator encounter, and reduced their time allocation to limit their exposure to predators; they sacrificed state to buy safety. By full moon, the gerbils upped vigilance even more, but increased time foraging; they defended state to guard against starvation. As the moon waned, gerbils decreased vigilance and increased foraging time to rebuild state. By new moon, vigilance was at a minimum, and foraging time began to decline; state had been rebuilt in time for another cycle (Kotler et al . 2010).
Prey foraging behavior also depends on the interaction between the state of the prey and that of predator. Using vivarium experiments, Berger-Tal & Kotler (2010) showed that hungry barn owls (Tyto alba ) were 4-7 times more active than their satiated counterparts. Gerbils responded to this increase in predator activity by visiting fewer patches and leaving them at higher GUDs, but only when in high energetic state (Berger-Tal et al . 2010).
Predators, too, consider their state as well as that of their prey. Hungry owls, for example, showed derring-do by performing dangerous attack maneuvers (plunging into areas with stiff, spikey experimental shrubs) more than twice as often as well-fed conspecifics (Embaret al . 2014a). Moreover, owls choose between well-fed and hungry gerbils (Embar et al . 2014b). In spring when gerbils were reproductive, owls favored well-fed gerbils; in the summer when they were months away from breeding, owls favored hungry gerbils. That may seem odd, but well-fed gerbils are more active in spring when energy supports offspring, and hungry gerbils are more active than well-fed gerbils in summer when survivorship to the next reproductive season is paramount. Owls, when given the choice between gerbils with fleas and gerbils without, chose the more active flea-free gerbils (Raveh 2018). In all cases, then, owls sought more active prey.
In summary, foraging games between gerbils and their predators are contingent on environmental factors such as microhabitat and moon phase and biotic factors such as the energetic states of predators and prey. Prey manage risk, predators manage fear, and these actions feed back between the players and the environment throughout each night (Kotleret al . 2002), across moon phases (Kotler et al . 2002, 2010), and over the seasons (Kotler et al . 2004).
Box 2 : The timing of predation risk as an emergent driver of contingency in NCEs
How prey invest in defense at any given time during phase two (prey response to perceived risk) may depend on the temporal pattern of intrinsic predation risk. Namely, according to the risk allocation hypothesis, defensive investment should be greatest in response to transient pulses of high risk against a background of relative safety (given that periods during which safe feeding can occur should soon return), and reduced when pulses of safety occur against a background of elevated danger (Lima & Bednekoff 1999). By implication, prey in systems where predation danger is highly punctuated may be able to compensate for heavy anti-predator investment when predators are most active (and/or lethal) by feeding during periods of predator inactivity. For example, vicuñas (Vicugna vicugna ) exploit puma (Puma concolor ) downtimes (during the day) to utilize their feeding grounds but avoid these densely-vegetated areas when low light levels and ample stalking cover combine to enhance puma lethality (Smith et al . 2019). Under these circumstances, demographic risk effects experienced by prey populations and the potential for prey to transmit indirect NCEs during phase three may be limited (Kohl et al . 2018).
To date, empirical support for the risk allocation hypothesis has been mixed (Ferrari et al . 2009), perhaps in part because prey condition in some assessments has been high enough to allow for continuous anti-predator investment even when risk is chronic (Matassa & Trussell 2014), or because some focal prey species were not given sufficient time to learn the risk regime (Moll et al . 2017). Our review offers an additional, non-mutually exclusive explanation. Namely, the temporal pattern of intrinsic risk experienced by a prey individual is an emergent outcome of the interaction between the properties (e.g., activity) of the predator(s) by which it is threatened and setting in which an encounter might take place. Moreover, as outlined earlier, the response of any prey individual/species to perceived intrinsic danger cues during phase two hinges on its own properties (e.g., escape tactics). Thus, proper quantification of the temporal pattern of risk and how prey should respond to perceived stimuli in any situation requires explicit consideration of each of these drivers of context dependence, as well as their interplay. It is possible that, lacking the capacity to be this comprehensive, some prior tests of the risk allocation hypothesis may have misrepresented the temporal pattern of risk. We view studies exploring this possibility as a fruitful line of inquiry. In the meantime, a recent investigation by Dröge et al . (2018) offers a path forward, at least in terms of accounting for predator properties. Namely, their ability to explain vigilance responses by African ungulates was greatest when immediate risk stimuli (predator proximity) were considered in relation to patterns of long-term risk associated specifically with the approaching predator species rather than the predator guild overall.
Box 3 : Methodology for the meta-analysis exploring contingent relationships between prey energetic state and risk taking
As the basis for our meta-analysis, we searched the literature using the bibliographic sources BIOSIS Previews, Google Scholar, and Web of Science using search terms for each engine were ‘predator’ OR ‘predation’ AND ‘prey’ AND ‘risk taking’ OR ‘behavior’ AND ‘state’ OR ‘condition’ OR ‘hunger’ OR ‘nutrition’ OR ‘food’. We then screened all resulting studies to ensure that they (i) manipulated prey state directly (e.g., via starvation or nutritional supplementation of individuals) or indirectly (e.g., by contrasting behavior of individuals in environments offering different amounts of food); (ii) exposed prey to direct (e.g., live or model predators) or indirect (different amounts of cover) predation risk cues; (iii) presented comparisons of some form of risk-taking behavior by prey individuals with divergent energetic states that were not confounded by differences with respect to development, experience/learning, parasite load, personality/temperament, reproductive state, sex, or size; and (iv) included means, measures of variance, and sample sizes required for calculating effect sizes. Separate experiments addressing different species within the same study were considered to be independent.
After filtering, our review of the literature yielded a total of 145 studies spanning a broad range of taxa and presenting 156 unambiguous, species-specific contrasts of risk-taking behavior manifested by prey individuals with disparate energetic states (see Appendix S1 in Supporting Information). The risk-taking behaviors addressed by these studies clustered into four categories, which we analyzed separately: activity (n = 30 studies making 34 species-specific contrasts); exposure (n = 68 studies making 71 contrasts); foraging investment (at the expense of apprehension; n = 37 studies making 41 contrasts); and isolation (versus grouping/shoaling; n = 10 studies presenting 10 contrasts). For the studies quantifying prey activity (movement), response metrics included time spent active or latency to resume activity after exposure to a predation risk cue for individuals, as well as the proportion of individuals in a group engaged in activity. Risk-taking responses addressed by studies of exposure were either time spent or the proportion of individuals not hiding, outside of a refuge, or in a dangerous versus safe area/patch. Studies addressing foraging investment quantified several feeding behaviors expected to influence vulnerability to predation: namely, feeding motivation (attack distance or latency to feed), feeding time, giving-up densities (GUDS; with greater amounts of food consumed corresponding to heightened foraging activity), and the proportion of individuals feeding. Studies of isolation quantified risk taking in terms of group cohesion (e.g., nearest neighbor distance) or size. For each of these four dimensions of risk taking, we first scaled all measures such that higher values corresponded with increased risk taking. Then, we assessed the across-study effects of prey energetic state on risk taking using two effect size metrics: Hedges’ d (Hedges & Olkin 1985) and log response ratios (lnRR, Hedges et al . 1999). For Hedges’ d , we interpreted effect sizes of 0.2 to be “small”, 0.5 to be “medium”, 0.8 to be “large”, and any exceeding 1 to be “very large” (Gurevitch & Hedges 1993). Estimates of both effect sizes were considered to be statistically significant if their 95% confidence intervals did not overlap zero. For each of the four dimensions of risk taking, we calculated mean effect sizes for each effect size metric using random effects models in OpenMEE (Wallace et al . 2017). We also tested for publication bias with Spearman’s rank-order correlations; no test was significant (all p ≥ 0.076), so we did not use funnel plots to remove outliers from any of our four sub-analyses (Begg & Mazumdar 1994). Finally, our meta-analyses of activity, exposure, and foraging investment included a few papers presented behavioral contrasts of more than one species. Given that these contrasts may not have been independent (Gurevitch & Hedges 1999), we conducted truncated meta-analyses for the three aforementioned risk-taking dimensions that included only one randomly-selected contrast from each study addressing multiple species. Owing to close agreement between our full and truncated analyses, we only present results from the full analyses below.