Moving beyond linear food chains: trait-mediated indirect interactions in a rocky intertidal food web

Abstract

In simple, linear food chains, top predators can have positive indirect effects on basal resources by causing changes in the traits (e.g. behaviour, feeding rates) of intermediate consumers. Although less is known about trait-mediated indirect interactions (TMIIs) in more complex food webs, it has been suggested that such complexity dampens trophic cascades. We examined TMIIs between a predatory crab (Carcinus maenas) and two ecologically important basal resources, fucoid algae (Ascophyllum nodosum) and barnacles (Semibalanus balanoides), which are consumed by herbivorous (Littorina littorea) and carnivorous (Nucella lapillus) snails, respectively. Because crab predation risk suppresses snail feeding rates, we hypothesized that crabs would also shape direct and indirect interactions among the multiple consumers and resources. We found that the magnitude of TMIIs between the crab and each resource depended on the suite of intermediate consumers present in the food web. Carnivorous snails (Nucella) transmitted TMIIs between crabs and barnacles. However, crab–algae TMIIs were transmitted by both herbivorous (Littorina) and carnivorous (Nucella) snails, and these TMIIs were additive. By causing Nucella to consume fewer barnacles, crab predation risk allowed fucoids that had settled on or between barnacles to remain in the community. Hence, positive interactions between barnacles and algae caused crab–algae TMIIs to be strongest when both consumers were present. Studies of TMIIs in more realistic, reticulate food webs will be necessary for a more complete understanding of how predation risk shapes community dynamics.

1. Introduction

In simple, linear food chains, top predators can have positive indirect effects on basal resources by causing changes in the density and/or traits of intermediate consumers [1–4]. For example, predators may indirectly affect plants by consuming and limiting the density of herbivore populations (density-mediated indirect interactions or DMIIs) or by causing a change in herbivore traits, such as reduced per capita grazing rates (trait-mediated indirect interactions or TMIIs). Such TMIIs drive trophic cascades in a variety of terrestrial, marine and freshwater systems [4–6]. Theory and a large body of empirical work has shown that TMIIs can be of equal or greater strength than DMIIs [7–9].

Classic, density-mediated trophic cascades are predicted to be weaker in complex food webs that include more trophic diversity (i.e. omnivory) or a greater proportion of weak species interactions [10–13]. Indeed, this classic view suggests that the prevalence of strong trophic cascades in aquatic systems is a result of the relatively simple food chains that predominate these systems [11,14]. Despite growing appreciation of the role of TMIIs in driving trophic cascades [1,4], few studies have examined how trait-mediated cascades operate in more complex food webs having multiple intermediate and basal trophic compartments (but see [15]). Studies of multiple predator effects suggest that increased complexity may impact trophic cascade strength when one predator increases or decreases the susceptibility of prey to other predators [16–21]. Similarly, intermediate and basal species may interact with one another in positive (e.g. facilitation) or negative (e.g. intraguild predation, competition) ways, and the number and sign of these interactions will determine whether increased complexity strengthens or weakens TMIIs between top predators and basal resources [15].

Here, we examine the effects of a single top predator (the green crab, Carcinus maenas, ‘Carcinus’) on a rocky intertidal community consisting of two basal resources: fucoid algae (germlings of Ascophyllum nodosum) and sessile invertebrates (barnacles, Semibalanus balanoides), both of which are important foundation species on rocky shores [22–24]. Carcinus exerts positive indirect effects on barnacles and algae by causing reductions in the feeding activity of its gastropod prey: the carnivore Nucella lapillus (Nucella) and the herbivore Littorina littorea (Littorina), respectively [25–27]. TMIIs between Carcinus and barnacles via Nucella and between Carcinus and fucoids via Littorina have each been studied independently in linear food chains [25,26], but this approach abstracts the complexity that may emerge under more realistic and reticulate food web scenarios (figure 1). Indeed, interactions between barnacles and fucoids (facilitation or competition [28–30]) and between Nucella and Littorina (predation [31,32]) can potentially alter the direct and indirect effects of Carcinus on each species in the food web. For example, barnacles can have positive effects on fucoid algae by providing additional substrate for the attachment of recruits, and recruits that settle within the structurally complex interstices between the tests (shells), of individual barnacles have a spatial refuge from herbivory [29,33–35]. By consuming barnacles, Nucella may have negative density-mediated indirect effects on fucoids because (i) the tests of consumed barnacles and any fucoids attached to them are readily dislodged from the substratum and (ii) removal of barnacles may reduce structural complexity, thereby enhancing Littorina's access to interstitial fucoids. Whether increased access to fucoids translates to a positive effect on Littorina grazing or growth rates should also depend on any direct effects of Nucella on Littorina because Nucella consume Littorina and other mobile gastropods when preferred prey (barnacles or mussels) are unavailable [31,32,36]. In either case, risk-induced reductions in Nucella foraging rates should enhance fucoid density. Thus, the magnitude of TMIIs between Carcinus and fucoid algae may depend not only on the presence of barnacles, but also on the composition of intermediate consumer community.

Figure 1.

Hypothetical interaction webs depicting negative (light red) and positive (dark blue) direct (solid lines) and indirect (dashed lines) effects between species in our experiment. The webs consist of two basal resources (fucoid algae (Ascophyllum nodosum) and barnacles (Semibalanus balanoides)) and one or two intermediate consumers (the herbivore Littorina littorea and/or the carnivore Nucella lapillus) in the absence (top row) or presence (bottom row) of predation risk from a top predator (the green crab Carcinus maenas). Arrow thickness corresponds to the relative per capita interaction strength between species. In the absence of Carcinus (top row), each consumer exerts strong negative effects on their respective resources. Because settlement ceased prior to our experiment, barnacle density could have positive effects on fucoids via two pathways: (1) by providing substrate for fucoids that had settled on barnacle tests prior to the experiment (solid blue arrow) and (2) by interfering with Littorina grazing on fucoids that had settled between barnacle tests (dashed blue arrow). By consuming barnacles, Nucella may have negative, density-mediated indirect effects on fucoids via pathway 1 in the absence of Littorina or by pathways 1 and/or 2 in the presence of Littorina. Nucella may also consume Littorina when preferred resources (barnacles or mussels) are unavailable, but, we observed no predation on Littorina in our experiment. The presence of Carcinus risk cues (bottom row) reduces the per capita foraging and growth rates of Littorina and Nucella, which lead to positive, trait-mediated indirect interactions (TMIIs) on each of their respective resources. By reducing the negative direct effects of Littorina and indirect effects of Nucella on fucoids, we hypothesized that Carcinus would exert stronger net TMIIs on fucoids when both consumers are present in the food web. For visual clarity, numbers identifying the two barnacle–fucoid facilitation pathways are omitted from the +Carcinus diagrams. (Online version in colour.)

We conducted a laboratory mesocosm experiment that independently manipulated the presence/absence of each consumer to examine how these two linear food chains interact with one another to structure the basal resource community. Our goal was to determine whether the independent TMIIs between green crabs and each basal resource were modified by the increased complexity arising from the combination of both food chains in the food web. We expected TMIIs between Carcinus and fucoids via reduced Littorina grazing rates to be stronger when Nucella were also present whereas TMIIs between Carcinus and barnacles would not depend on the presence of Littorina.

2. Material and methods

We tested the effects of predation risk on a rocky intertidal food web by exposing one or two species of intermediate consumers, the carnivorous snail Nucella lapillus and/or the herbivorous snail Littorina littorea, to waterborne risk cues from a top predator, the green crab C. maenas (species hereafter referred to by their generic names only). The three treatments (Carcinus risk cues, Littorina, Nucella), each having two levels (present, absent), were randomly applied to 40 independent, outdoor mesocosms (n = 5 replicate mesocosms per treatment combination) located at the University of Maine's Darling Marine Center in Walpole, ME, USA. The experiment ran for 60 days, from late June to late August, 2001.

Each mesocosm (35 × 15 × 15 cm, l × w × h) received its own supply of running seawater and was stocked with a pair of granite settlement tiles (15 × 15 × 1 cm each, l × w × h) that served as a natural substratum for the two snails and an abundant supply of their respective resources: acorn barnacles (S. balanoides) and fucoid algae (germlings of A. nodosum). Mesocosms assigned to receive Nucella and/or Littorina were stocked with five experimental Nucella (shell length range 7–11 mm) and/or five Littorina (shell length range 6–9 mm), respectively. Carcinus risk cues were applied by placing a perforated ‘crab container’ (plastic tub, 15 × 5 cm, d × h) inside the mesocosm and stocking it with a single male green crab. Each crab was fed 10 Nucella and/or 10 Littorina food snails to match the assigned consumer treatment. In the absence of Carcinus risk cues, crab containers received only the corresponding food snails, which remained alive. All food snails were replaced or replenished weekly for the duration of the experiment. Crabs would consume some of these snails right away and most of the remaining snails by the end of the week. Our previous work demonstrates that even brief pulses of risk can produce strong effects on prey and their resources [37,38]. While risk cues come from both the crab and the consumed prey items, the effects of crab cues are much stronger and account for the majority of the risk effect [39].

Prior to their placement in mesocosms, granite settlement tiles (N = 190) were anchored on a rocky outcropping in the intertidal zone of the Damariscotta River during the annual barnacle and fucoid recruitment seasons. This site, like many in the estuary, has a dense Ascophyllum canopy with barnacles on the substratum in the canopy understory. Barnacle recruitment began in mid-March and lasted several weeks, whereas fucoid recruitment began in early May and continued through June. Because both snail species begin to forage actively in late spring, all tiles were enclosed in galvanized hardware cloth and left in the field until mid-June to ensure adequate recruitment and survival of both barnacles and fucoid algae until the start of our experiment. Once retrieved from the field, we selected 40 pairs of tiles (from the initial supply of 190) with uniformly high fucoid recruitment and similar barnacle densities (mean ± s.d.: 8.66 ± 1.65 barnacles cm⁻², N = 80) to standardize the initial community composition across all mesocosms. We estimated initial barnacle density by counting the number of barnacles within five 3 × 3 cm quadrats randomly placed on each tile. Initial counts of fucoids were not possible because of their extremely high densities, often in excess of 300 individuals cm⁻². Rather, all tiles were scanned under a dissecting microscope, and fucoid density was qualitatively characterized. Each pair of tiles was randomly assigned to a mesocosm, and analysis of initial barnacle densities revealed no significant differences among our eight experimental treatment combinations (ANOVA: F_7,32 = 0.23 p = 0.97; all pairwise comparisons: p ≥ 0.31).

At the end of the experiment, we counted the number of live barnacles remaining in five 3 × 3 cm quadrats and the number of fucoids remaining in five 1 × 1 cm quadrats on each tile (total sampling area per mesocosm = 90 cm² and 10 cm² for barnacles and fucoids, respectively). The placement of quadrats was randomized using non-overlapping grid coordinates and a random number generator. We used per-tile estimates of final barnacle density and final fucoid density to estimate the strength of TMIIs, or the proportional increase in resource density due to predation risk. For each of the four combinations of consumers (no consumers, Littorina only, Nucella only, or Littorina + Nucella), we calculated TMII strength between Carcinus and each basal resource using the formula TMII = (Y_tr(+C)/Ȳ_(−C)) – 1, where Y_tr(+C) is the resource density on tile t in replicate r of the +Carcinus treatment, and Ȳ_(−C) is the mean density of that resource (Ȳ) across all five replicates of the corresponding −Carcinus control [9]. Hence, we obtained two estimates of TMII strength per tile: Carcinus–barnacle TMII and Carcinus–fucoid TMII. For example, a TMII strength of 0.7 indicates that predation risk caused a 70% increase in resource density compared with the no risk control.

To determine whether treatment-specific differences in snail feeding rates on barnacles and fucoids translated into differences in snail growth, we measured the final shell lengths of all experimental snails (N = 200) at the end of the experiment. Prior to the start of the experiment, we randomly sampled five individuals from each treatment combination and measured their initial shell length. Because we used a narrow size range of snails in the experiment and there were no significant differences among treatment combinations for either snail species (mean ± s.e. shell length of Littorina = 7.48 ± 0.16 mm, F_3,16 = 0.14, p = 0.93; of Nucella = 9.27 ± 0.19 mm, F_3,16 = 0.10, p = 0.96), we consider final shell length measurements as a proxy for growth during the experiment. We also measured final shell thickness of Littorina to examine whether the presence of either Carcinus or Nucella induced the formation of relatively thicker shells [25]. Shell thickness for each snail was calculated as an average of two thickness measurements taken at two locations along the lip margin. We did not measure Nucella shell thickness because heavy sculpture along the lip margin made it difficult to obtain reliable and consistent values.

All statistical analyses were performed using JMP Pro software v. 11.1.1 (SAS Institute Inc., Cary, NC, USA) [40]. Final shell lengths of individual Nucella and Littorina were analysed with a two-way nested ANOVA that included Carcinus risk cues and the presence/absence of heterospecific consumers as fixed effects. Individual mesocosms (nested within the Carcinus and consumer treatments) were included in the model as random intercepts via restricted maximum likelihood (REML) to avoid pseudoreplication because there were multiple snails within each replicate mesocosm. Nucella from one replicate of the –Carcinus/+Littorina treatment combination were mistakenly not measured at the end of the experiment and were therefore excluded from the analysis. No snails died during the experiment. Final shell thickness of Littorina was analysed with a two-way nested ANCOVA that included the same fixed and random effects as above plus final shell length as the covariate. The full ANCOVA model with all covariate interactions confirmed that the assumption of homogeneity of slopes was satisfied (all p > 0.17), so the slope terms (covariate interactions) were pooled with residual error in the final model [41,42]. Denominator degrees of freedom in the nested models were calculated using the Kenward–Roger first-order approximation [40].

Final barnacle density and final fucoid density (individuals cm⁻²) on each experimental tile were analysed with separate three-way nested ANOVAs that considered the Carcinus, Littorina and Nucella treatments as fixed effects. Because there were two tiles in each mescosm, mesocosm (nested within the consumer and predator treatments) was included in the model as a random effect using REML.

TMII magnitudes were analysed with a mixed model ANOVA that included the two-consumer treatments (presence/absence of Littorina and Nucella) and the resource type (barnacles or fucoids) as fully-crossed fixed effects. Because there were two tiles per mesocosm and two estimates of TMII per tile (one for each resource), both mesocosm (nested within the consumer treatments) and tile (nested within mesocosm) were included in the model as random effects using REML [42]. Data are available at the Dryad Data Repository [43].

3. Results

(a). Effects of predation risk on intermediate consumers

Waterborne risk cues from Carcinus caused reductions in snail growth rates such that Nucella (F_1,15 = 73.78, p < 0.0001, table 1a) and Littorina (F_1,16 = 10.98, p < 0.0001, table 1b) under predation risk were ultimately 17% and 7% smaller, respectively, than those in the absence of risk cues (figure 2a,b). Heterospecific consumers had no independent or interactive effects on the growth of either Nucella (both p > 0.4) or Littorina (both p > 0.8, table 1a,b).

Table 1.

Results from nested two-way (a,b) ANOVA and (c) ANCOVA testing effects of the presence/absence of Carcinus risk cues and heterospecific consumers on the final shell morphology of experimental (a) Nucella and (b,c) Littorina. Because there were five snails per mesocosm, mesocosm was included in the models as a random effect to avoid pseudoreplication. In (c), final shell length was the covariate, and non-significant slope terms (all p ≥ 0.17) were pooled with the residual error in the final model.

response variable	source	d.f.n	d.f.d	F	p-value
(a) Nucella shell length (mm)	Carcinus risk cues	1	15	73.7773	<0.0001
	Littorina	1	15	0.1533	0.7009
	C × L	1	15	0.5190	0.4824
(b) Littorina shell length (mm)	Carcinus risk cues	1	16	10.9756	0.0044
	Nucella	1	16	0.0574	0.8136
	C × N	1	16	0.0582	0.8124
(c) Littorina shell thickness (mm)	Carcinus risk cues	1	17.5	17.1515	0.0006
	Nucella	1	15.8	3.8409	0.0679
	C × N	1	15.8	4.4506	0.0512
	final shell length	1	90.7	62.2651	<0.0001

Figure 2.

(a) Final shell length of Nucella, (b) final shell length of Littorina and (c) adjusted shell thickness of Littorina in the presence (filled symbols) or absence (open symbols) of Carcinus risk cues and with or without heterospecific consumers in the community (horizontal axis). In (a,b), values are pooled replicate means ± s.e., while values in (c) are adjusted LS means ± s.e. from ANCOVA (n = 5 except in (a), where n = 4 for the +Littorina/+Nucella/−Carcinus treatment combination). Within each panel, different lower-case letters denote significant differences among group means according to Tukey HSD post hoc tests (p < 0.05).

The presence of Carcinus risk cues caused Littorina to produce relatively thicker shells (F_1,17.5 = 17.15, p = 0.0006, table 1c). In the absence of Carcinus, Nucella also caused an increase in Littorina shell thickness (C × N: F_1,15.8 = 4.45, p = 0.051, table 1c and figure 2c). Compared with Littorina in the absence of both predators, exposure to Carcinus only or Nucella only caused Littorina to produce, on average, 15% thicker shells. Exposure to both Carcinus and Nucella did not cause an additional increase in shell thickness. No Littorina were consumed by Nucella during the experiment.

(b). Effects of consumers on basal resources in the absence and presence of predation risk

The presence of Carcinus risk cues reduced the impacts of Nucella and Littorina on their respective barnacle (C × N: F_1,32 = 8.96, p = 0.0053, table 2a) and fucoid resources (C × L: F_1,32 = 6.93, p = 0.0129, table 2b). In the absence of Carcinus, Nucella alone caused a 62% decrease in barnacle density (Tukey HSD: p < 0.0001, figure 3a(−Carcinus)) but had no effect on barnacle density in the presence of Carcinus (Tukey HSD: p = 0.64, figure 3a(+Carcinus)). Similarly, Littorina alone caused a 59% decline in fucoid density in the absence of Carcinus (Tukey HSD: p < 0.0001, figure 3b(−Carcinus)) but had no effect on fucoid density in the presence of Carcinus (Tukey HSD: p = 0.24, figure 3b(+Carcinus)).

Table 2.

Summary of results from three-way ANOVAs testing effects of the presence/absence of Carcinus risk cues, Littorina, and Nucella on the final density of (a) barnacles and (b) fucoids on two experimental community tiles within each mesocosm. Mesocosms were included in the model as a random effect.

	(a) barnacles		(b) fucoids
source	F1,32	p-value	F1,32	p-value
Carcinus risk cues	15.5532	0.0004	17.7991	0.0002
Littorina	0.0417	0.8395	28.4770	<0.0001
Nucella	21.8205	<0.0001	14.7024	0.0006
C × L	0.8327	0.3683	6.9303	0.0129
C × N	8.9575	0.0053	5.3986	0.0267
L × N	0.0499	0.8247	0.4139	0.5246
C × L × N	0.0163	0.8992	8.3810	0.0068

Figure 3.

Final (a) barnacle and (b) fucoid density (individuals cm⁻²) in communities with different combinations of consumers (±Littorina, open and closed circles, respectively, and ±Nucella, horizontal axis) in the absence (−) or presence (+) of risk cues from Carcinus. Values are means ± s.e. (n = 5).

Although Littorina had no effect on final barnacle density (all p > 0.8, table 2a and figure 3a), Nucella had negative effects on fucoid density, especially when Littorina were also present or when Carcinus was absent (C × L × N: F_1,32 = 8.38, p = 0.0068, table 2b and figure 3b). In the absence of Carcinus (figure 3b(i)), Nucella alone caused a 48% decline in fucoid density (Tukey HSD: p = 0.0007), and the negative effects of Nucella and Littorina combined additively (interaction contrast: F_1,32 = 2.53, p = 0.12) to cause an 81% decrease in fucoid density compared with no-snail controls. In the presence of Carcinus (figure 3b(+Carcinus)), Nucella and Littorina had non-additive effects on fucoid density (interaction contrast: F_1,32 = 6.26, p = 0.018): in the absence of heterospecifics, neither Littorina nor Nucella affected fucoid density (linear contrasts: F_1,32 = 0.17, p = 0.68 and F_1,32 = 1.03, p = 0.32, respectively), but together, the two snails caused a 31% decline in fucoid density relative to no-snail controls (linear contrast: F_1,32 = 4.44, p = 0.043).

(c). Strength of trait-mediated indirect interactions between Carcinus and basal resources

Nucella transmitted strong positive TMIIs between Carcinus and basal resources (F_1,16 = 35.28, p < 0.0001, table 3), regardless of resource type (R × N: F_1,36 = 1.62, p = 0.21, table 3 and figure 4a,b). Littorina transmitted TMIIs between Carcinus and fucoids (figure 4b), but not between Carcinus and barnacles (TMII not significantly different from 0; figure 4a; R × L: F_1,36 = 21.02, p < 0.0001, table 3). The contributions of Littorina and Nucella to the strength of TMIIs were additive (L × N: F_1,16 = 0.54, p = 0.47) regardless of resource type (R × L × N: F_1,36 = 0.32, p = 0.58, table 3). In the absence of both consumers, TMIIs were not significantly different from 0 (figure 4).

Table 3.

Summary of results from a mixed model ANOVA on the magnitude of TMIIs between Carcinus and the two different resources (barnacles and fucoids). The two-consumer treatments (presence/absence of Littorina and Nucella) and the resource type (barnacles or fucoids) were included in the model as fixed effects. Mesocosm (nested within the consumer treatments) and tile (nested within replicate) were included in the model as random effects.

source	d.f.n	d.f.d	F	p-value
Littorina	1	16	4.6982	0.0456
Nucella	1	16	35.2829	<0.0001
L × N	1	16	0.5372	0.4742
resource	1	36	0.4693	0.4977
R × L	1	36	21.0203	<0.0001
R × N	1	36	1.6235	0.2108
R × L × N	1	36	0.3169	0.5770

Figure 4.

Magnitude of TMIIs between (a) Carcinus and barnacles, and (b) Carcinus and fucoids mediated by different combinations of consumers (±Littorina, open and closed circles, respectively, and ±Nucella, horizontal axis). Values are LS means and 95% CIs from the mixed model ANOVA. The dashed line indicates TMII = 0.

4. Discussion

A longstanding and influential view in ecology is that the strength of trophic cascades should diminish in more speciose or reticulate food webs [10,11]. The logic of this view has been partly shaped by the prevalence of strong trophic cascades in relatively simple food chains or those having keystone predators (e.g. [44]) and the expectation that increased food web complexity (versus simple linear food chains) dampens cascade strength by enhancing the likelihood of trophic heterogeneity or differentiation [11]. Such heterogenity can emerge via a number of processes, but variation in the edibility of species [45] produced by inducible defences [46–48] and size-structured interactions [49,50] appear to be particularly important. However, our understanding of these issues is far from complete, as evidenced by the growing body of work [20,21,51] that highlights the challenges associated with defining the relationship between trophic cascade strength and food web diversity and/or complexity.

Our results revealed that consumer community composition and food web complexity can alter the strength of TMIIs between a top predator and basal resources. In our experimental single-consumer food webs, the presence of Carcinus risk cues caused both Nucella and Littorina to consume fewer resources, leading to positive indirect effects on both barnacles (TMII = 1.73, figure 4a) and fucoids (TMII = 1.03, figure 4b), respectively. These risk-induced foraging reductions were large enough to eliminate the impact of either consumer on its resource when Carcinus was present (figure 3). However, the carnivore, Nucella, was also able to transmit TMIIs to fucoid algae independently of the herbivore, Littorina (TMII = 0.75, figure 4b). Carcinus–fucoid TMIIs transmitted by Nucella were similar in magnitude to those transmitted by Littorina because Nucella alone caused reductions in fucoid density that were comparable with the impact of Littorina grazing in the absence of risk (figure 3b(−Carcinus)), but Nucella alone had no effect on fucoids in the presence of risk (figure 3b(+Carcinus), open symbols). These negative effects of Nucella on fucoids arose due to Nucella's consumption and removal of barnacles from the experimental communities (figure 3a).

On rocky shores, barnacles facilitate the development of fucoid communities by buffering desiccation stress, creating additional substrate for settlement and providing spatial refuges that interfere with herbivory, all of which may increase the density of fucoid algae [29,30,33–35]. In the field, and in our experimental mesocosms, the tests of consumed barnacles readily slough off the substrate, taking along any fucoids that originally settled upon them. In the absence of both Carcinus and Littorina, Nucella reduced barnacle density by 62% and caused a 52% reduction in fucoid density, whereas Nucella alone had no measurable impact on either barnacle or fucoid density in the presence of risk (figure 3, open circles). Hence, Nucella foraging likely reduced fucoid density in our experiment by incidentally removing fucoids that were attached to consumed barnacles. This density-mediated indirect interaction (DMII) between Nucella and fucoids was weakened by the negative effects of Carcinus on Nucella foraging rates, allowing the transmission of a positive TMII between Carcinus and fucoids via Nucella (figure 1).

In two-consumer webs (figure 1, top right), Nucella may have facilitated Littorina grazing by removing barnacles and the associated structural complexity that interferes with herbivory, thus enhancing Littorina access to fucoids [29]. In this case, assuming no other interactions between Nucella and Littorina, we would expect Littorina and Nucella to have synergistically negative effects on fucoid density in the absence of risk. However, we found that the negative indirect effects of Nucella and direct negative effects of Littorina combined additively to reduce fucoid density by 81% compared with no-consumer controls (figure 3b(−Carcinus)) thereby increasing the net TMII between Carcinus and fucoids (TMII = 1.91, figure 4b).

That the effects of Littorina and Nucella on fucoids were additive in the absence of Carcinus risk cues suggests that these two consumers impacted fucoid density through independent pathways, i.e. there was no evidence that Nucella facilitated Littorina grazing via reduced interference by barnacles. This interpretation is further supported by the fact that Littorina did not exhibit the increase in growth that would be expected with grazing facilitation by Nucella (figure 2b). However, Nucella is also a known predator of Littorina, and although Nucella did not consume any Littorina in our experiment, Nucella may have had negative non-consumptive effects on Littorina grazing and/or growth efficiency that counteracted any positive bottom-up effects [37,52]. Differences in Littorina shell thickness indicate that Nucella did have non-consumptive effects on Littorina morphology and provide some support for this hypothesis. In the presence of Nucella, Littorina produced shells that were 15% thicker than those produced by Littorina in the absence of predators (figure 2c). Such an increase in shell thickness, which is quantitatively similar to the increase in shell thickness induced by Carcinus risk cues, may provide protection from drilling by Nucella. In previous experiments, induced shell defences are accompanied by reductions in grazing and growth rates [25].

In contrast with the additive effects of Nucella and Littorina in the absence of Carcinus, neither consumer had an independent effect on resources in the presence of Carcinus. However, when both consumers were together in the presence of Carcinus they caused a 35% decrease in fucoid density (figure 3b(+Carcinus)). This result is somewhat surprising for two reasons. First, Nucella had no effect on barnacle density in the presence of Carcinus, and therefore reduced barnacle interference of herbivory cannot explain how Nucella and Littorina synergistically reduced fucoid density. Secondly, Nucella caused Littorina to produce thicker shells (figure 2c), indicating that Littorina experienced and responded to predation risk from Nucella. As argued above, predation risk typically causes reductions in prey foraging rates, yet here we found that Nucella had a positive effect on Littorina grazing rates. One possible explanation is that the associated increase in consumer density in the two-consumer food web (n = 10 snails per mesocosm as opposed to n = 5 snails in the linear food chain) alleviated some of the effects of predation risk from Carcinus through a ‘many eyes’ effect. Having ‘many eyes’ allows prey to be less vigilant or perceive less predation risk as conspecific density increases [53–55]. In previous work, we found that high conspecific density can reduce the impact of predation risk on snail foraging and growth [52], and increased consumer density, including heterospecifics, may have produced similar effects here. This interpretation suggests that the risk imposed by a shared predator may reduce the cascade-dampening effects of omnivory or intraguild predation within diverse assemblages of intermediate consumers.

Alternatively, Carcinus and Nucella may have different non-consumptive effects on Littorina because they impose different levels of predation risk, have different attack rates and hunting modes, and share different degrees of spatial and temporal overlap with Littorina [1]. Compared with Carcinus, Littorina's behavioural responses to Nucella (i.e. reduced grazing rates) are probably less effective than induced morphological defences because of the high degree of habitat overlap between Nucella and Littorina, including shared spatial refuges from Carcinus such as cracks and crevices. By contrast, reduced grazing activity, though costly, may be necessary for Littorina to avoid the more frequent and rapid attack rates of Carcinus that occur outside spatial refuges. Clearly, multiple predator effects (i.e. [16,19]) may be operating on Littorina in this system and warrant further study.

Although our study cannot fully identify the mechanisms by which all the species in this food web interacted to shape TMIIs, we clearly demonstrate that interactions between multiple species at lower trophic levels can modify top-down control via predation risk. Previous work has shown that the effects of predation risk on prey and resulting TMIIs can drive trophic cascades in linear food chains [1,5,25–27,56], but such simplified scenarios are rare in nature. Our results suggest that the trait-mediated cascades observed in three-species, linear food chains may persist and even be exacerbated in more complex, realistic food webs. Although the current state of knowledge prevents robust generalizations, recent evidence suggests that trait-mediated cascade strength may increase with increasing food web complexity or diversity. For example, in terrestrial food webs with varying levels of predator diversity, trophic cascade strength increased via an interaction between non-consumptive predator effects and predator diversity [21]. Moreover, there is growing evidence that diversity effects can operate through non-consumptive pathways [57,58] and that consumptive diversity effects can be shaped in ecologically important ways by the predator avoidance behaviours of prey [51]. The emergence of these patterns may partly reflect the fact that predation risk can simultaneously affect multiple prey species, their impact on basal resources, and their interactions with each other [1,4]. Hence, we predict that positive relationships between top-down control and food web complexity will most often be observed in systems where predator avoidance is a key factor governing species interactions [15]. Regardless of the mechanisms involved, it is clear that future modelling and empirical work must incorporate the reticulate nature of natural food webs if we are to obtain a more robust understanding of how predation risk structures natural communities.

Ethics

This work was conducted in accordance with the guidelines of the Association for the Study of Animal Behavior and the animal care guidelines of Northeastern University's Institutional Animal Care and Use Committee (IACUC).

Data accessibility

Data from this experiment are available from the Dryad Data Repository at: http://dx.doi.org/10.5061/dryad.2vt31 [43].

Authors' contributions

G.C.T. and P.J.E. conceived and designed the study and collected the data, and C.M.M. performed the analyses. G.C.T. and C.M.M. wrote the manuscript. All authors contributed to revisions of the manuscript.

Competing interests

We declare we have no competing interests.

Funding

This work was supported by the USA National Science Foundation (OCE-0727628, OCE-1355873) and Division of Ocean Sciences (OCE-0240265).