Prey state shapes the effects of temporal variation in predation risk

Abstract

The ecological impacts of predation risk are influenced by how prey allocate foraging effort across periods of safety and danger. Foraging decisions depend on current danger, but also on the larger temporal, spatial or energetic context in which prey manage their risks of predation and starvation. Using a rocky intertidal food chain, we examined the responses of starved and fed prey (Nucella lapillus dogwhelks) to different temporal patterns of risk from predatory crabs (Carcinus maenas). Prey foraging activity declined during periods of danger, but as dangerous periods became longer, prey state altered the magnitude of risk effects on prey foraging and growth, with likely consequences for community structure (trait-mediated indirect effects on basal resources, Mytilus edulis mussels), prey fitness and trophic energy transfer. Because risk is inherently variable over time and space, our results suggest that non-consumptive predator effects may be most pronounced in productive systems where prey can build energy reserves during periods of safety and then burn these reserves as ‘trophic heat’ during extended periods of danger. Understanding the interaction between behavioural (energy gain) and physiological (energy use) responses to risk may illuminate the context dependency of trait-mediated trophic cascades and help explain variation in food chain length.

1. Introduction

All organisms must acquire energy to survive, grow and reproduce. For species in the middle of food chains, obtaining food can be especially risky because individuals are often more vulnerable to predators while foraging [1–3]. Solutions to this growth/predation risk trade-off often involve changes in prey behaviour, physiology, morphology or life history [1,4–6] that can have rapid, widespread and diverse ecological consequences [7–9]. By causing prey to shift to safer habitats, reduce foraging rates and/or experience physiological stress, predation risk can limit prey fitness and population size, drive trophic cascades via trait-mediated indirect interactions and modify the flow of energy and nutrients within food webs [7–14]. Thus, the ecological impacts of predators scaring prey (‘non-consumptive predator effects’) can be more substantial than those elicited by predators consuming prey [9,15,16].

Physiological and environmental conditions that shift how prey balance the costs (susceptibility to predation) and benefits (energy intake and growth) of foraging probably influence how prey respond to predation risk [1,8]. Theoretical and empirical work suggests that prey may be more willing to accept increased predation risk while foraging if the risk of starvation is sufficiently high [17–21], thereby weakening the strength of non-consumptive predator effects on prey and emergent indirect effects on other species or ecological processes. Alternatively, because risk is inherently variable over space and time, prey may ‘wait out’ more dangerous periods, shifting all foraging activity to periods or places of relative safety [22–25]. However, both the quantity (duration or frequency) and quality (food availability) of intervening periods of safety can influence the capacity of prey to wait out periods of danger [26,27]. The ‘predation risk allocation hypothesis' [27] argues that the degree of antipredator behaviour exhibited by prey at any given moment depends on the temporal pattern of risk exposure surrounding that moment [27]. For example, risk allocation counterintuitively predicts that prey foraging activity during both safe and risky periods will increase as the duration or frequency of risky periods increases. This prediction arises because prey must acquire some minimum amount of energy in order to survive [27]. Therefore, prey must fully compensate for reductions in foraging activity during risky periods with increased foraging during periods of safety. As safe periods become less frequent, and risky periods become longer, prey must also increase foraging activity during risky periods in order to meet energetic demands.

Despite its influence on the study of predation risk, empirical support for risk allocation has been mixed, perhaps because many tests of the model have failed to satisfy its assumption that prey live on the energetic ‘edge’ of survival [28,29]. Indeed, state-dependent versions of the model predict risk allocation behaviour to be less intense for prey with greater energy reserves [26,27]. However, the predicted ecological impacts of risk allocation behaviour still rely on this underlying assumption: if prey forage at a minimum rate in order to survive, then the proportion of time under high predation risk should not affect their mean foraging rate when averaged across all periods of risk and safety [27,30].

Clearly, estimates of the strength and relative importance of non-consumptive predator effects will depend on where and when we measure prey responses to predation risk. Here, we use a rocky intertidal food chain to test the behavioural predictions of risk allocation across multiple prey states and examine the ecological impacts of temporally variable predation risk. On rocky shores in the Gulf of Maine, waterborne risk cues from the predatory green crab (Carcinus maenas) cause its gastropod prey (Nucella lapillus) to increase refuge use and reduce foraging activity [31], thereby generating strong trait-mediated indirect effects on basal resources such as mussels (Mytilus edulis) and barnacles (Semibalanus balanoides) [15,16,32]. Past experiments in this system indicate that even infrequent exposure to green crab risk cues can have large effects on Nucella foraging and growth [32], but it remains unknown whether Nucella allocate foraging effort across safe and risky periods, and whether these behaviours are shaped by energetic status.

Importantly, recent work indicates that for Nucella and other prey species, conversion of acquired energy into new biomass may be less efficient under predation risk because the physiological stress imposed by risk can be energetically costly [10–12,33–35]. Such energetic costs can reduce prey fitness to alter prey populations [10,35] and limit the flow of energy to higher trophic levels [12,13]. Here, we use foraging and growth data to estimate these energetic costs, which we describe using a new metric called ‘trophic heat’ [13]. Trophic heat serves as an index of how prey respond physiologically to predation risk and describes the rate at which energy is lost from the system by an intermediate consumer (prey) due to inefficiencies in secondary production and/or the use of bodily energy reserves. We found that the production of trophic heat depends strongly on the temporal pattern of exposure to predation risk and the energetic state of prey. Our results show that temporal and energetic contexts interact to influence the behavioural and physiological responses of prey to predation risk, and may shape the community- and ecosystem-level consequences of these responses.

2. Material and methods

Our experiment examined the effects of prey state on the response of prey (the Atlantic dogwhelk, N. lapillus, hereafter ‘snail’) to temporally variable predation risk from the green crab (C. maenas). Two levels of baseline predation risk (low and high) were fully crossed with two levels of temporal variability (constant and variable) and two levels of prey energetic state (starved and fed). The resulting eight treatment combinations were each randomly assigned to 64 independent, flow-through mesocosms (n = 8) at Northeastern University's Marine Science Center in Nahant, MA, USA. Replicate mesocosms consisted of modified plastic utility boxes (27 × 15 × 5 cm, length × width × height) individually housed in 6 l containers to prevent water exchange among units. Each utility box had two sections separated by a perforated barrier: an upstream crab chamber (11 × 15 × 5 cm) used to manipulate predation risk and a downstream snail chamber (16 × 15 × 5 cm). The snail chamber held a granite tile to mimic natural substrate, and was stocked with four tagged and measured experimental snails (all starved or all fed) plus an abundant supply of small mussels as food (M. edulis, shell length range 8.5–14.0 mm, n = 80 mussels per mesocosm). Continuously flowing seawater entered the crab chamber through a vinyl hose, flowed through the perforated barrier and exited the snail chamber through a mesh roof.

We manipulated predation risk by adding and removing green crabs to/from the crab chambers of appropriate mesocosms on a 4-day cycle for 16 days beginning 19 August 2010. During the first 3 days of each cycle (‘baseline’ days), mesocosms either held a single green crab (high baseline risk) or no crab (low baseline risk). For the final 24 h of each cycle (‘switch’ day), baseline risk levels either remained constant (temporal variability = constant) or were switched to the opposite risk state (temporal variability = variable) by adding or removing crabs. The resulting predation risk treatment combinations controlled the proportion of time prey spent under high risk (P_risk = 0, 0.25, 0.75 or 1). Green crabs that were in mesocosms for all 4 days in the cycle (P_risk = 1) were replaced at the end of each cycle to control for any effects of crab addition/removal. Each crab was fed two snails upon addition to a mesocosm, and we removed any uneaten snails or shell fragments when crabs were removed.

All snails were collected from a semi-exposed shore near New Harbor, ME, USA. To manipulate prey state, snails were held in flow-through aquaria with either no food (starved) or small mussels supplied ad libitum (fed). After 20 days under these conditions, we selected 128 starved and 128 fed snails of similar shell lengths (mean ± s.d., 12.3 ± 1.1 and 12.5 ± 1.2 mm, respectively) and glued an identifying tag on each snail's shell. A starvation period of 20 days is reasonable because N. lapillus snails have been observed to shelter in crevices for at least 39 consecutive days without foraging [36], often forego foraging due to environmental stressors other than predation risk [37] and can survive several months with a negative scope for growth [38]. Based upon previous work [39], we estimate that starved individuals lost 5–10% of their body mass, whereas fed snails probably gained an additional 60–70% of their initial body mass during the 20-day conditioning period.

To estimate tissue growth, tagged snails were measured at the beginning and end of the experiment using Palmer's [40] buoyant weight technique to obtain wet tissue mass (see the electronic supplementary material, appendix A). We calculated each snail's net tissue production (T_p) as the difference between the final (T_f) and initial (T_i) energetic value of its tissue mass. T_f and T_i were found by converting wet tissue mass to dry tissue (electronic supplementary material, appendix A) then multiplying by an energetic conversion factor of 22.7 J mg⁻¹ [41]. The initial wet tissue mass of starved and fed snails differed by 15 mg or 96 J (mean ± s.d., 75 ± 18 and 90 ± 24 mg, respectively).

We observed and recorded the foraging activity (foraging or not foraging) of individual snails 24 h after placement in mesocosms and every approximately 24 h thereafter. On days when crabs were added/removed, behavioural observations were made just prior to addition/removal. A snail was considered foraging if its body was in contact with a mussel and it remained stationary for a minimum 20 s observation period. Although consuming a mussel can be a complex process [42], assessing foraging behaviour to a higher resolution within our mesocosms would have required handling the snail or mussel with undesired effects on their behaviour. The number of mussels consumed and their energetic value were determined at the end of the experiment by counting and measuring the shell lengths of consumed mussels. We estimated the energetic value of consumed mussels in each mesocosm by converting shell lengths to dry flesh mass and energetic equivalents using an empirically derived equation and conversion factor of 19.5 J mg⁻¹ [43,44]. To estimate the per capita amount of energy acquired (A) by a given snail within a mesocosm, we multiplied the total energy from consumed mussels by the proportion of total foraging occurrences made by that snail within its mesocosm.

We define trophic heat as the proportion of energy lost from the system by an intermediate consumer (due to growth inefficiencies and/or the use of bodily energy reserves) that may have otherwise become available to its predators. We calculated trophic heat for each snail as 1 − T_p/(T_i + A), where T_i reflects the snail's previous foraging gains or energy reserves. Minimum trophic heat and maximum flow of energy up the food chain are produced with perfect prey growth efficiency (T_p = A). As trophic heat increases, the net rate at which energy flows up the food chain decreases. Trophic heat values less than 1 indicate positive energy flow (T_p > 0), while values greater than 1 indicate negative energy flow (T_p < 0).

Statistical analyses were performed in R (v. 2.15.2) [45], and all data are available via the Dryad data repository [46]. We analysed daily snail foraging activity (proportion of snails foraging) with a binomial generalized linear mixed model (GLMM). The model included baseline risk, temporal variability, prey state, cycle and cycle day as fully crossed fixed effects, and mesocosm, mesocosm × cycle and mesocosm × cycle day as random effects. Because there were days when no snails were observed foraging in one or more treatment groups, we added 0.5 to all observations (number foraging and number not foraging) to avoid degenerate confidence intervals in the GLMM.

We also performed a more formal time-series analysis on the behaviour of snails exposed to variable risk. Using spectral analysis, we estimated the magnitude (maximum spectral density) and frequency (frequency of maximum spectral density) of changes in snail foraging activity over time in each mesocosm. We analysed spectral data (ln-transformed to satisfy parametric assumptions) with factorial analyses of variance (ANOVAs) that included prey state and the proportion of time spent under high risk (P_risk = 0.25 or 0.75) as fixed effects.

Using the mean foraging activity of snails during safe and risky periods, we tested the predictions of the risk allocation hypothesis that as the proportion of time under high risk (P_risk) increases, foraging activity will increase during (a) risky periods and (b) safe periods. We tested these predictions separately for starved and fed snails using quasi-binomial generalized linear models (GLMs) that included P_risk as a fixed effect, with P_risk = 0.25, 0.75 or 1 for (a) and P_risk = 0, 0.25 or 0.75 for (b).

We analysed the mean per capita energy acquired within each mesocosm using a generalized least-squares (GLS) regression model that included baseline risk, temporal variability and prey state as fully crossed fixed effects. Tissue production and trophic heat of individual snails were analysed with separate linear mixed-effects (LME) models that included the same fixed effects as the GLS plus ‘mesocosm’ as a random effect to avoid pseudoreplication because there were multiple snails per mesocosm. Weighted variance structures (REML-estimated) were included in the GLS model of per capita foraging rates and the LME model of tissue production to account for heteroscedasticity among the baseline risk treatments and the baseline risk × temporal variability treatment combinations, respectively [47]. While taking down the experiment, three of the four snails in one mesocosm were accidentally crushed, and therefore were excluded from tissue production and trophic heat analyses.

3. Results

The daily foraging activity of snails (figure 1) varied with the temporal pattern of predation risk (baseline risk × temporal variability × cycle day: Wald χ² = 31.02, d.f. = 3, p < 0.0001; see also the electronic supplementary material, table B1 in appendix B). Foraging activity was generally low when snails were exposed to crab risk cues (figure 1, shaded areas) and high during periods of safety (figure 1, white areas). Under constant safety (figure 1a) or constant risk (figure 1b), snails maintained relatively high or low foraging activity, respectively, throughout the risk cycle (linear contrast of baseline versus switch days under constant safety: p = 0.22; under constant risk: p = 0.61). Snails under variable risk altered their foraging activity when baseline conditions were switched to the opposite risk state: foraging activity declined during brief pulses of risk (p < 0.0001; figure 1c) and increased during brief pulses of safety (p < 0.0001; figure 1d). Despite significant variation in foraging activity among risk cycles (cycle: Wald χ² = 99.92, d.f. = 3, p < 0.0001), prey responses to different temporal patterns of risk did not vary significantly among cycles (baseline risk × temporal variability × cycle × cycle day: Wald χ² = 6.72, d.f. = 9, p = 0.67; electronic supplementary material, table B1).

Figure 1.

Mean (±s.e.) foraging activity (proportion of snails foraging) of starved (open circles) and fed (filled circles) snails during each day of the experiment while in the presence (shaded areas) or absence (white areas) of a predatory green crab. The proportion of time snails were exposed to green crab risk cues (P_risk) varied so that snails were under (a) constant low risk, (b) constant high risk, (c) variable low risk or (d) variable high risk (P_risk = 0, 1, 0.25 or 0.75, respectively).

Differences in the foraging activity of starved and fed snails varied over time (prey state × cycle × cycle day: Wald χ² = 48.98, d.f. = 9, p < 0.0001) and with baseline risk (baseline risk × prey state: Wald χ² = 7.86, d.f. = 1, p = 0.005). Averaged across all risk levels, the foraging activity of starved snails was greater than that of fed snails on day 1 (linear contrast: p < 0.0001), but this overall effect of prey state disappeared by day 3 (p = 0.26; days 4–16: all p > 0.3; figure 1). Averaged across all time periods, however, starved snails foraged 2.6 times more than fed snails while under high baseline risk (linear contrast: p < 0.0001), but prey state had no effect on the average foraging activity of snails under low baseline risk (p = 0.26).

Spectral analysis revealed that, under variable risk, prey state affected the magnitude of changes in snail foraging activity over time (spectral density), but this effect depended on the duration of risky periods (prey state × P_risk: F_1,28 = 5.43, p = 0.027; see also electronic supplementary material, table B2a in appendix B). Starved snails exhibited large fluctuations in foraging activity (high spectral density) regardless of P_risk (Tukey HSD: p > 0.9). Fed snails exhibited similarly large fluctuations when exposed to short periods of risk (P_risk = 0.25), but foraging activity was more stable when exposed to longer periods of risk (P_risk = 0.75; Tukey HSD: p < 0.01; electronic supplementary material, table B2a), indicating that fed snails responded less strongly to brief pulses of safety than to brief pulses of risk. The dominant frequency of changes in snail foraging activity aligned with the frequency of changes in risk (0.25 or once every 4 days) and did not vary among treatment groups (p > 0.6 for all effects; see the electronic supplementary material, table B2b).

As predicted by the risk allocation hypothesis, foraging activity during risky periods increased as the proportion of time under high risk (P_risk) increased, but only when snails were starved (F_2,21 = 9.42, p = 0.001; figure 2a). Fed snails maintained low foraging rates during risky periods regardless of P_risk (F_2,21 = 0.19, p = 0.83; figure 2a). However, foraging activity during safe periods did not increase with increasing P_risk as predicted by risk allocation. During safe periods, P_risk had no effect on the foraging activity of starved snails (F_2,21 = 2.31, p = 0.13), but the foraging activity of fed snails declined as P_risk increased (F_2,21 = 6.81, p = 0.005; figure 2b).

Figure 2.

The average foraging activity (proportion of snails foraging) of starved (open circles) and fed (filled circles) snails during (a) risky (shaded areas in figure 1) and (b) safe periods (white areas in figure 1) that varied in duration. The proportion of time spent under high predation risk (P_risk) increases along the horizontal axis. Values and error bars are predicted means and 95% CIs from the GLMs (symbols offset for visual clarity).

Reductions in foraging activity during the experiment caused snails to acquire less energy (figure 3a) and produce less body tissue (figure 3b) with increasing exposure to predation risk (baseline risk × temporal variability: both p < 0.0001; table 1a,b). When baseline risk was low, exposure to brief periods of high risk (increase in P_risk from 0 to 0.25) caused snails to acquire 40% less energy (linear contrast: p < 0.0001) and produce 68% less body tissue (p < 0.0001). The negative effects of additional exposure to risk depended on prey state (baseline risk × prey state: both p < 0.01; table 1a,b and figure 3a,b). For example, compared with constant safety, constant high risk caused starved snails to acquire 68% less energy and produce 97% less body tissue, but had even stronger effects on fed snails, causing 90% and 125% reductions in foraging and growth, respectively. On average, brief periods of safety (decrease in P_risk from 1 to 0.75) allowed snails to acquire an additional 99 J (approx. 0.5 mussels) snail⁻¹ (p = 0.012; figure 3a; see also electronic supplementary material, figure B1 in appendix B), but did not yield a significant increase in tissue production (p = 0.14; figure 3b).

Figure 3.

Mean (a) per capita energy acquired by snails, (b) snail tissue production and (c) trophic heat in replicates where starved (open circles) or fed (filled circles) snails were exposed to different temporal patterns of predation risk (symbols offset for visual clarity). The proportion of time spent under high predation risk (P_risk) increases along the horizontal axis. Error bars (sometimes smaller than symbols) represent 1 s.e. (calculated from pooled replicate means, n = 8). The dotted line in (b) corresponds to zero tissue production. In (c), values below and above the dotted line (trophic heat = 1) correspond to net positive and net negative energy flow, respectively.

Table 1.

Summary of ANOVA results for (a) per capita energy acquired (J snail⁻¹), (b) tissue production (J snail⁻¹) and (c) trophic heat. Baseline risk (base), temporal variability (temp) and prey state (prey) were fixed effects.

effect	(a) energy acquired		(b) tissue produced		(c) trophic heat
	F1,56	p-value	F1,56	p-value	F1,56	p-value
base	191.50	<0.0001	325.69	<0.0001	260.57	<0.0001
temp	12.87	0.0007	53.20	<0.0001	16.16	0.0002
prey	8.26	0.0057	12.30	0.0009	40.24	<0.0001
base × temp	40.17	<0.0001	83.49	<0.0001	44.54	<0.0001
base × prey	8.49	0.0051	10.68	0.0019	20.79	<0.0001
temp × prey	0.48	0.4901	2.22	0.1419	0.86	0.3586
base × temp × prey	0.05	0.8262	3.94	0.0520	3.51	0.0662

The production of trophic heat increased with increasing exposure to predation risk (baseline risk × temporal variability: p < 0.0001; table 1c and figure 3c). In the absence of risk (P_risk = 0), 81% of the energy available to snails for growth and maintenance was lost as trophic heat, while 19% was invested in new tissue production. Brief pulses of risk (P_risk = 0.25) increased trophic heat to 93% (linear contrast: p < 0.0001). With greater exposure to predation risk, the production of trophic heat depended on prey state (baseline risk × prey state: p < 0.0001; table 1c and figure 3c). Under high baseline risk (P_risk = 0.75 or 1), 97–99% of the energy available to starved snails was lost as trophic heat, effectively halting the flow of energy up the food chain. Fed snails produced even more trophic heat (108–111%), burning 100% of the energy acquired from mussels plus an additional 8–11% of energy stored prior to the experiment as tissue mass.

4. Discussion

Snails responded quickly to fluctuating levels of green crab predation risk by increasing their foraging activity during low-risk periods and decreasing foraging activity during high-risk periods. Prey state interacted with the temporal pattern of risk to shape the magnitude of changes in foraging activity between low- and high-risk periods: all snails responded strongly to brief pulses of high risk in an otherwise safe environment (figure 1c), but behaviour in a relatively risky environment depended on prey state (figure 1d). As predicted by the risk allocation hypothesis [27], foraging activity during risky periods increased as those periods became longer, but only when snails were starved (figure 2a). By contrast, fed snails maintained low foraging rates during long periods of risk and foraged less during brief pulses of safety (figure 2), dampening the temporal pattern in foraging activity and reducing their impact on mussel resources.

Our results provide good support for traditional models of risk-sensitive foraging [8,20,21,24], but only mixed support for risk allocation. Risk allocation predicts that prey will forage more during high-risk periods as the proportion of time under high risk (P_risk) increases because of diminishing opportunities to forage under safety [27]. The behaviour of starved snails was more consistent with this prediction probably because they better satisfied the model's assumption that prey are on the energetic edge of survival. Learning the temporal pattern of risk also may also influence whether empirical tests support risk allocation [29]. However, recent modelling work suggests that state-dependent behaviour can lead to typical risk allocation patterns (B. Luttbeg 2014, personal communication). Prey may recall information about temporal patterns of risk by ‘looking it up in their gut’ [48] because risk-induced foraging reductions in the past can affect current hunger levels. Hence, past exposure to high P_risk should result in depleted energy reserves and therefore greater foraging activity during future periods of safety or danger [28]. In our experiment, it is likely that starved snails were foraging at a maximum rate during safe periods. Thus, the only way they could meet critical energetic demands when safe periods were short was to forage more during extended periods of danger (figure 2a,b). Creel et al. [22] found similar patterns while observing elk vigilance in response to wolf predation risk at different sites in the Greater Yellowstone Ecosystem. When wolves were nearby, elk vigilance was lower at sites visited more frequently by wolves. Importantly, these observations were made during winter months, when elk have limited access to food and face a greater risk of starvation [22]. As the risk of starvation increases, prey are more likely to behave in the ‘paradoxical’ manner predicted by risk allocation [27,29].

The functional relationship between foraging and fitness may ultimately determine the larger impacts of risk on ecological communities because it can shape the integrated effects of temporally variable risk on prey foraging rates [48,49]. The original risk allocation model [27] assumes that fitness and foraging are related by a step function: prey forage at some minimum average rate, R, in order to survive and thus increase fitness from 0 to 1 [27]. In this case, P_risk should not affect R [27,30]. Alternatively, if fitness increases linearly with foraging, then R should increase as P_risk decreases so that prey can maximize future reproductive output [48]. This latter prediction is better supported by our results, which show that snails consume more mussels (R increases) as P_risk decreases (figure 3a; electronic supplementary material, figure B1), and by growing evidence that increased exposure to risk reduces the foraging impact of prey on basal resources, leading to strong trait-mediated indirect interactions [7,9,16,32,50–52]. For example, chemical cues signalling predation risk in freshwater ponds increase the abundance of periphyton by causing herbivorous snails to reduce grazing activity [52]. However, the magnitude of this response and the relative strength of resulting trait-mediated indirect interactions depend on the temporal pattern of exposure to risk cues [53,54].

We found that increased exposure to risk led to reduced foraging and growth, which is tightly coupled with fitness in N. lapillus [43]. Compared with individuals under constant safety, exposure to predation risk for just 25% of the time led to a 40% reduction in foraging and a 68% reduction in growth. Additional exposure to risk caused further reductions in foraging and growth, but the magnitude of these effects depended on prey state (figure 3a,b). Under constant risk, for example, starved snails reduced foraging by 68% and tissue growth dropped to approximately 0. High risk had even stronger effects on fed snails, which reduced foraging and growth by 90% and 125%, respectively, consuming less than 1 mussel snail⁻¹ and losing more than 10% of their initial tissue mass (figure 3a,b; electronic supplementary material, figure B1). Because prey are more likely to be ‘fed’ when resources are abundant, our results suggest that resource levels may interact with temporal patterns of risk to ultimately shape the role that trait-mediated cascades play in community dynamics [18,52,55].

Trophic heat provides an estimate of how prey respond physiologically to predation risk and describes the rate at which energy is lost from the system by prey due to inefficiencies in secondary production and the burning of energy reserves. Trophic heat increased with risk exposure (figure 3c), indicating that the energetic consequences of predation risk exceeded those imposed by reduced prey foraging gains alone. In relatively safe environments, snails exposed to brief pulses of risk were able to maintain positive energy flow (trophic heat less than 1) by using energy from consumed mussels to produce new tissue mass, albeit at a lower efficiency (higher trophic heat) than snails under constant safety. When exposed to longer periods of risk, fed snails produced more trophic heat than starved snails. On average, a fed snail under constant risk acquired an additional 94 J from consumed mussels but lost 73 J of tissue mass, whereas a starved snail acquired 285 J from mussels and produced 8 J of tissue. Starved snails, which were likely to be in survival mode, foraged just enough to maintain current reserves/body mass (trophic heat ≈ 1). By contrast, negative growth and high trophic heat indicate that fed snails burned all of the energy acquired during the experiment plus a substantial proportion of the energy reserves they had established prior to the experiment. While energy reserves had no effect on the production of trophic heat under relatively safe conditions, reserves were clearly an important source of fuel for respiration, stress responses or other physiological costs in riskier environments [10–12,33–35]. These results suggest that when a predator enters the system, not only does it affect the movement of energy from basal resources to prey by reducing conversion efficiency [12], but it also affects the fate of energy that had been stored by prey prior to its arrival.

How prey use different sources of available energy (reserves versus resources) under different risk regimes has important implications for risk allocation and trophic dynamics. Increased production of trophic heat under risk suggests that risk-exposed prey will require more energy to achieve the same fitness as prey under safer conditions. These additional energetic demands could exacerbate risk allocation behaviour: in addition to reduced foraging gains during risky periods, foraging rates during safe periods must also compensate for the energetic costs of risk-induced stress [11,13]. Differences in the costs of short- versus long-term stress responses [56] may ultimately cause variable risk to be more energetically costly than constant risk [32].

The duration and quality (e.g. food availability) of risky and safe periods, rather than the relative proportion of time spent under each condition, can influence how prey acquire and use energy [26] and affect trophic heat. Abundant resources during long periods of safety (e.g. our pre-experiment feeding period) allow prey to build up energy reserves, but this energy is lost as trophic heat during subsequent long periods of risk (trophic heat greater than 1). Although prey should eventually forage in order to avoid starvation, the energy lost as trophic heat cannot be restored. Under such conditions, prey are effectively an energy sink. By increasing trophic heat, energy reserves may exacerbate the negative effects of predation risk on energy flow up the food chain [12] and potentially explain why some food chains are short even when basal resources are abundant [57,58]. As argued by Pimm & Lawton [57], food chains may be short because longer chains are more sensitive to environmental perturbations (the dynamic stability hypothesis). Exhaustion of energy reserves and reduced growth efficiency may increase the vulnerability of prey populations to environmental perturbations (e.g. future predation risk and resource limitation). Thus, trophic heat may destabilize the link between basal resources and higher trophic levels, and shorten food chains.

Because most species are in the middle of food chains [59] and have to balance eating versus being eaten, understanding the factors that govern individual foraging choices and energy use will be key to predicting the ecological consequences of predation risk. We found that temporal and energetic contexts interact to shape the behavioural and physiological responses of prey to predation risk. Our results also suggest that the indirect effects of predation risk on community structure (via trait-mediated indirect interactions) and ecosystem function (trophic heat) will be weaker in systems where prey live on the edge of starvation. By contrast, predation risk may lead to stronger trophic cascades and grossly reduce the flow of energy up the food chain in more productive systems where prey are able to establish energy reserves. Hence, understanding how physiological and behavioural responses to predation risk interact may elucidate the context-dependency of trophic cascades and food chain length.

Data accessibility

All data are available from the Dryad (http://datadryad.org) data repository (doi:10.5061/dryad.86n98).

Funding statement

Research was supported by the National Science Foundation through a Doctoral Dissertation Improvement grant (IOS-1110675) to C.M.M., grant nos. OCE-0648525 and 0727628 to G.C.T. and grant OCE-0963010 (Academic Research Infrastructure Recovery and Reinvestment Program) to the Marine Science Center (MSC). This is contribution 317 from the MSC.