To Olfactory and Visual Cues from Multiple Predators

Oecologia (2010) 162:893–902 DOI 10.1007/s00442-010-1564-x

BEHAVIORAL ECOLOGY - ORIGINAL PAPER

DiVerential habitat use and antipredator response of juvenile roach (Rutilus rutilus) to olfactory and visual cues from multiple predators

Charles W. Martin · F. Joel Fodrie · Kenneth L. Heck Jr. · Johanna Mattila

Received: 16 June 2009 / Accepted: 6 January 2010 / Published online: 3 February 2010 © Springer-Verlag 2010

Abstract The indirect, behavioral eVects of predation and open-water habitat, but entered structured habitat when predator–predator interactions can signiWcantly alter the presented with a vision cue of this predator. Opposite trophic ecology of many communities. In numerous responses were elicited from roach for both olfactory and instances, the strength of these eVects may be determined visual cues of perch. Interestingly, roach defaulted to selection by the ability of prey to identify predation risk through of structured habitat when presented with vision + olfaction predator-speciWc cues and respond accordingly to avoid cues of either predator. Moreover, when presented individ- capture. We exposed juvenile roach (Rutilus rutilus), a ual cues of both predators together, roach responded by common forage Wsh in many brackish and freshwater envi- choosing open-water habitat. Upon being presented with ronments, to vision and/or olfactory cues from two preda- vision + olfaction cues of both predators, however, roach tors with diVerent hunting methods: northern pike (Esox strongly favored structured habitat. DiVerences in habitat lucius, an ambush predator) and European perch (Perca selection of roach were likely in response to the alternative Xuviatilis, a roving predator). Our results demonstrated that foraging strategies of the two predators, and suggest that responses of roach to perceived risk (as evidenced by their prey species may not always use structured habitats as pro- selection of structured or open-water habitats) were highly tection. This appears particularly true when a threat is per- dependent on cue type and predator identity. For instance, ceived, but cannot immediately be located. These results roach responded to olfactory cues of pike by entering provide insight to the complex and variable nature by which prey respond to various cues and predators, and oVer a mechanistic guide for how behaviorally mediated and Communicated by Anssi Laurila. predator–predator interactions act as structuring processes in aquatic systems. C. W. Martin (&) · K. L. Heck Jr. Dauphin Island Sea Lab, 101 Bienville Boulevard, Dauphin Island, AL 36528-0369, USA Keywords Antipredator behavior · Predator–prey e-mail: [email protected] interactions · Olfaction · Vision · Multiple predator eVects · Non-consumptive eVects C. W. Martin · K. L. Heck Jr. University of South Alabama, 307 University Boulevard N, Mobile, AL 36688-0002, USA Introduction

F. J. Fodrie The non-consumptive eVects of predation (also referred to Department of Marine Sciences, Institute of Marine Sciences, as trait or behaviorally mediated indirect interactions) are University of North Carolina at Chapel Hill, increasingly recognized to be an important, yet understud- 3431 Arendell Street, Morehead City, NC 28557, USA ied, component of predator–prey interactions and food-web dynamics (Dill et al. 2003; Grabowski 2004). Recent meta- J. Mattila Husö Biological Station, Åbo Akademi University, analyses have indicated that the mere threat of predation Tykistökatu 6, 20520 Turku, Finland can have impacts that are as strong, if not stronger, than 123 894 Oecologia (2010) 162:893–902 direct consumption (Dill et al. 2003; Werner and Peacor predator eVects on shared prey, especially relative to the 2003; Pressier et al. 2005). For instance, indirect interac- impacts of predator species operating alone (Sih et al. tions that alter the foraging behavior of mid-trophic-level 1998). In some cases, multiple predators can operate syner- species can initiate trophic cascades (Trussell et al. 2002; gistically, resulting in enhanced risk for their shared prey Schmitz et al. 2004). A key element of these non-lethal (Fodrie et al. 2008). Nearly all reports of predator facilita- predator–prey exchanges is the ability of prey to recognize tion that result in enhanced risk for a shared prey are based and evade risk once a threat is perceived through visual, on two elements: one predator alters the habitat selection of olfactory, mechanoreception, electroreception, or social prey, and this makes prey highly vulnerable to attacks from cues. Thus, separating the behavioral responses of prey to another predator (Sih et al. 1998). In other scenarios, preda- various cue types should lead to a better understanding of tors interfere with one another resulting in reduced risk for the circumstances in which trait-mediated interactions will prey (Crowder et al. 1997). Often, this occurs if prey are dominate trophic exchanges and regulate energy Xow capable of rapidly switching between predator-speciWc through food webs. defense behaviors/mechanisms (Sih et al. 1998). Regard- Although the roles of all antipredator sensory cues have less of the outcome, the ability of prey to receive, process, been evaluated in aquatic systems [e.g., mechanoreception and act upon multiple cues is put to a stern test in instances via lateral line sensitivity (Montgomery and MacDonald when there is more that one threatening predator species. 1987) and social cues via schooling (Hager and Helfman Although tests of multiple predator eVects have thus far 1991; Brown et al. 2009)], many experiments have focused focused on prey mortality as the response variable, multiple on the role of vision (Mathis et al. 1993; Heithaus et al. predator eVects should be equally important determinants 2002, 2006; Wirsing et al. 2007) and olfactory cues in of non-consumptive eVects. To our knowledge, the speciWc mediating predator–prey exchanges (Smith 1992; Kats and cues used by prey in multiple predator scenarios and Dill 1998; Chivers and Smith 1998; Brown 2003; Amo subsequent eVects on habitat selection have not been et al. 2004; Kusch et al. 2004; Ylönen et al. 2007). Within evaluated. aquatic environments, we still know little about the relative To investigate cue-mediated predator–prey interactions, roles of each in governing predator–prey interactions we designed an experiment to address the following ques- (Wahle 1992; Mathis and Vincent 2000; Hickman et al. tions: (1) what are the relative roles of olfactory and visual 2004; Kim et al. 2009), even though it could be expected cues from two common predators on the habitat selection of that prey rely heavily on olfaction within relatively turbid a widespread prey species, and (2) do interspeciWc combi- environments. In the future, eutrophication (Utne-Palm nations of predator cues induce responses in prey habitat 2002; Engström-Öst and Mattila 2008) and the loss of sedi- selection that could be predicted by averaging the responses ment-stabilizing foundation species such as submerged of prey to each predator individually? aquatic vegetation (Waycott et al. 2009) should lead to increasingly turbid conditions. This may force prey to rely even more heavily on olfaction for guiding defensive Materials and methods responses to predator detection (Lehtiniemi et al. 2005). Therefore, data are needed to evaluate the relative eVects of Study organisms vision and olfaction, which are recognized as the most important sensory cues used by prey within shallow-water We explored cue-driven predator–prey interactions using environments (Mikheev et al. 2006), on the predator-avoid- commonly encountered species in brackish and freshwater ance responses of prey. Moreover, quantifying the magni- habitats throughout Europe. Juvenile roach (Rutilus ruti- tude and direction of prey responses elicited by these cues, lus), a numerically dominant forage Wsh that inhabits both both individually and in combination, has the potential to open-water and structured habitats [e.g., emergent reeds reveal new insights for predicting how food webs will (Phragmites australis) as well as numerous species of sub- respond to anthropogenic stressors aVecting water quality merged aquatic vegetation], were used as prey. While roach (sensu Lindquist and Bachmann 1982; Shivik 1998). are known to school, especially under threat of predation Since omnivory is common in natural systems (Pimm (Christensen and Persson 1993; Eklöv and Persson 1995), and Lawton 1978), most species at lower and intermediate we used individual roach in this experiment to tease apart positions in food webs (i.e., forage Wsh in estuarine envi- their responses to various cue inputs as a starting point to ronments) not only receive multiple cues from single preda- understanding their behavior. We selected northern pike tors, but also simultaneous threat stimuli from multiple (Esox lucius) and European perch (Perca Xuviatilis), two of predator species. Therefore, prey must respond to an amal- the most commonly encountered piscivores in this region gam of risk cues. Where predator species diVer in foraging (BonsdorV and Blomqvist 1993; Ådjers et al. 2006) as pre- behaviors, there is growing interest in quantifying multiple dators. These two species have diVerent foraging strategies: 123 Oecologia (2010) 162:893–902 895 northern pike hide in vegetation and ambush prey (Savino drained, rinsed thoroughly, and reWlled with Wltered water and Stein 1989), while perch forage by continually roving to eliminate any lingering olfactory cues from previous among habitats (Eklöv and Persson 1995). Despite these trials. Salinity and temperature were checked periodically divergent foraging strategies, pike and perch are both capa- throughout the experiment and found to range between ble predators of roach and could be deWned by qualitatively 4.7–5.0 and 15.5–18.5°C. Fluorescent lighting, placed similar interaction coeYcients. Furthermore, these preda- approximately 20 cm above each aquarium, was used for tors are known to elicit strong antipredator responses in illumination. other species of prey (Christensen and Persson 1993; The behavioral responses of roach [95% conWdence Mathis and Smith 1993; Mathis et al. 1993). interval (CI): 31.6 § 5 mm; n = 162] to pike (95% CI: 157.8 § 13 mm; n = 10) and/or perch (95% CI: Experimental design and procedures 144.2 § 14 mm; n = 6) cues were investigated by examin- ing habitat selection (open water vs. structured) in a two- Our experiment was conducted during 14 August–7 Sep- way experimental manipulation within experimental tember 2008 at Husö Biological Station in the Åland archi- aquaria. First, we allowed individual roach to select pelago. We collected roach, pike, and perch by seining or between habitats during exposure to three diVerent cue dip netting in Husö Bay adjacent to Husö Biological Sta- types: olfaction, vision, and vision + olfaction. Second, we tion, in the Åland archipelago (60°04’N, 20°48’E). Follow- manipulated the predator Weld that generated each of our ing collection, Wshes were held in 75-l holding tanks three cue types: no predator (control), one pike, one perch, (654 cm long £ 29 cm wide £ 9.5 cm tall) until their use in one pike + one perch, two pike, two perch. This resulted in experimental trials (each species in separate tanks). Hold- a 6 £ 3 orthogonal design with 16 unique treatments (we ing tanks were devoid of structure and completely sur- were unable to include the one-pike and one-perch treat- rounded with plastic dividers to minimize Wsh stress. We ments during trials made with just olfactory cues because of exchanged water daily in holding tanks and provided aera- predator availability). We conducted six replicates of each tion with a single airstone in each tank. Predators (pike and unique treatment combination, resulting in 96 trials in perch) were fed two roach (31.6 § 5 mm) each day until which we observed the behavioral response of roach. Runs needed for experiments. Roach were always collected the of cue type were blocked by time for logistic reasons (num- day before their use in experimental trials and therefore not ber of available pike and perch to generate olfactory or fed while in the holding tank. Individual prey were used visual cues), while the six predator Welds were completely once and then released in Husö Bay. randomized among aquaria and experimental cycles. Experimental trials were conducted within six 17-l Including six levels of predator Weld allowed us to evaluate experimental aquaria (33 cm long £ 24.5 cm wide £ 21 cm if multiple predator eVects (i.e., do pike and perch foraging tall). In each aquarium, we placed artiWcial seagrass units together aVect the behavior of prey diVerently than the (ASUs) over half of the bottom (16.5 cm long £ 24.5 cm additive eVects of single predators?) inXuenced the habitat wide £ 21 cm tall) to provide structured habitat. The other selection of roach using both substitutive (comparing half of the aquarium remained unstructured, open-water results from the one pike + one perch treatment with results habitat. ASUs were assembled by tying green ribbon to of the two-pike and two-perch treatments) and additive Vexar mesh to mimic the morphology of many common (comparing results from the one-pike and one-perch treat- aquatic plants [e.g., eelgrass (Zostera marina)] at a density ments against the one pike + one perch treatment) designs of approximately 1,000 shoots m¡2, a value within the (GriVen 2006). Both methods have advantages. The substi- range of typical eelgrass beds in the Baltic Sea (Baden and tution design allows for the teasing apart of total density Boström 2001). During each experimental run, Wltered and species richness eVects (e.g., Schmitz and Sokol-Hess- water (180 m) from Husö Bay was used to Wll each aquar- ner 2002; Vance-Chalcraft et al. 2004). The additive design ium. Although predator olfactory cues could be present in acknowledges the reality that species richness and overall the incoming water, we assumed these to be at trace levels density usually scale together, and does not alter the and equally introduced among all experimental aquaria and strength of intraspeciWc interactions among treatments holding tanks. Since chemical cues were involved in these (e.g., Finke and Denno 2002; Lang 2003; Warfe and experiments, we did not circulate water in aquaria during Barmuta 2004). these trials, but did use an airstone to ensure the aquaria We designated six additional holding tanks (also without were well aerated. The airstone was placed in the ASU sec- structure and with opaque partitions) for generating and tion of each aquarium to prevent the introduction of struc- recovering olfaction cues from each level of predator Weld. ture to the open half of the aquaria. A pilot study veriWed Water from nearby Husö Bay was Wltered through a 180- that prey were neither attracted nor deterred by the presence m sieve and used to Wll tanks (24.5 cm long £ 16.5 cm of the airstone. Between trials, experimental aquaria were wide £ 21 cm tall; Wlled to »3–1), and then we added one 123 896 Oecologia (2010) 162:893–902 of the six combinations of predators to each tank. To Statistical analyses recover predator-derived olfactory cues, we kept predators inside these tanks for 24 h to standardize scent accumula- We used a series of one-way ANOVAs to determine if tion. Between trials, we rinsed tanks thoroughly. This diVerent cue types and predator treatments aVected the procedure was completed each time an experimental cycle roach’s selection of ASU habitat (not having olfactory was run. treatments from the one-pike or one-perch treatments pre- We also placed six additional aquaria immediately cluded the use of a two-way analysis). First, three separate behind our six experimental aquaria (devoid of any struc- ANOVAs were conducted to determine if predator Weld ture, and with an airstone). In treatments requiring vision or generated diVerences in roach habitat selection. For these vision + olfaction cues from predators, we randomly placed analyses, we considered the data from each cue treatment one of the six combinations of predators in these aquaria (i.e., olfaction, vision, or vision + olfaction) individually. (moving pike and perch out of the holding tanks as needed). Similarly, we conducted six additional ANOVAs to assess Thus, roach were able to see these predators but not receive how cue type aVected roach habitat selection. This required olfactory cues (unless separately added, see below). parsing the data among the individual predator treatments For each trial, we Wrst allowed a single roach to accli- [i.e., no predator (control), one pike, one perch, one mate to the experimental aquarium for 1 h. In treatments pike + one perch, two pike, two perch]. In cases with sig- that included olfactory cues, we sampled a 100-ml aliquot niWcant ANOVA results ( = 0.05), Tukey’s post hoc tests from each of the six tanks in which predators had been were performed to make pairwise comparisons between soaking for 24 h. These aliquots were immediately poured treatments. into the center of the appropriate experimental aquaria. In The presence of multiple predator eVects on the habitat trials that employed visual cues, we placed predators in the selection of roach for both the additive and substitutive tanks adjacent to the experimental aquaria at the same time designs was evaluated by comparing “expected” habitat roach were added. To allow acclimation for both predator selection of roach (based on single-predator species and prey, we placed an opaque divider between the two treatments) to observed habitat selection of roach in the aquaria and then removed it after 1 h. In vision + olfaction mixed-predator treatment. If predator cues aVected roach cue trials, we combined the steps from the olfaction and independently in the one pike + one perch treatment, vision treatments. expected habitat selection of prey in the mixed-predator To observe the behavioral response of roach to cues and aquaria for the additive experimental design can be pre- minimize handling artifacts, we recorded Wsh movements in dicted by averaging ASU selection (%) by roach in the the aquaria using a Sony digital video camera (model DCR- separate one-pike and one-perch treatments. A similar H36). Video recordings were initiated following the accli- approach was true for the substitution design: expected mation period (immediately after the addition of predator habitat selection of roach in the 1 pike + 1 perch treatment cues) and the experiment then ran for 1 h. Pairs of roach/ was calculated by averaging results from the individual predator aquaria were placed immediately beside each two-pike and two-perch treatments. We averaged the other so we could capture all six in the video recording, and results of separate predator treatments to calculate expected therefore opaque dividers were placed between the six pairs habitat selection, rather than using the additive or multipli- to avoid transmission of visual cues among aquaria and cative models suggested in Sih et al. (1998) or GriVen achieve independent replicates. Using a white background (2006), because we measured a non-consumptive variable behind the aquaria pairs, the silhouettes of roach were and our measured response could be bi-directional. For highly visible in all videos, making it easy to determine if example, roach could shift away from one habitat for one each individual was in the structured or open-water half of predator, and toward that habitat for the other, relative to each of the six aquaria. After the experiment, all aquaria our controls, with the expectation that in combined predator and holding tanks were reset as described above. During treatments these two eVects should cancel one another. the Wrst six trials, roach occupancy was recorded every Thus, the methods proscribed in consumptive multiple minute and compared at intervals of 1, 2, or 3 min. No sta- predator eVect experiments were not logically or quantita- tistical diVerence was detected among sampling regimes tively appropriate and had to be amended. For both the (two-way ANOVA with trial number and time interval as additive and substitution designs, we used two-sample factors; F1,16 = 39.350, P = 0.125; Tukey’s post hoc: 1 ver- unpaired t-tests to compare observed outcomes in multiple sus 2 min, P = 0.985; 1 versus 3 min, P = 0.932; 2 versus predator treatments (one pike + one perch) to calculated, 3min, P = 0.979), thus 3-min intervals were used to quan- expected outcomes as prescribed above. Using this statisti- tify roach occupancy. The frequency of ASU selection by cal approach, a signiWcant diVerence ( · 0.05) between roach was used as the response variable in all subsequent observed and expected values would indicate that the com- analyses. bined eVects of pike and perch on the habitat selection 123 Oecologia (2010) 162:893–902 897

Table 1 Results of one-way ANOVAs testing for the eVects of either Parametric assumptions were satisWed in all instances cue type or predator Weld on selection of structured habitat by roach and therefore data transformations were not required. Source df SS FP Because each statistical test applied to separate and easily distinguishable hypotheses, we also made no corrections to Predator Weld (olfaction 3 11,319.8 4.10 0.020 experiment-wise during this study (Moran 2003). only treatments) Error 20 18,404.2 W Predator eld (vision 5 2,939.2 7.50 <0.001 Results only treatments) Error 35 27,448.6 W Juvenile roach responded strongly to the cues of pike and Predator eld (vision + 5 10,007.1 6.49 <0.001 W olfaction treatments) perch, but the speci c responses (i.e., habitat selection) of W Error 36 11,107.1 roach varied based on the predator eld and cue type. W Cue type (control) 2 21.3 0.01 0.989 Within all levels of predator eld except the control and V Error 16 15,970.8 one-pike treatment, the e ect of cue type on roach habitat selection was highly signiWcant (Table 1). Likewise, preda- Cue type (one pike) 1 7 0.03 0.872 tor Weld had a signiWcant eVect on the response of roach Error 12 3,186 during trials run with the three cue combinations (Table 1). Cue type (one perch) 1 10,314 12.92 <0.001 For instance, roach were observed in structured habitat only Error 12 9,579 »30% of the time when provided olfactory cues from two Cue type (two pike) 2 12,686.1 13.95 <0.001 pike (we did not have a one-pike treatment with olfactory Error 17 7,732.5 cues). In contrast, in both the one-pike and two-pike treat- Cue type (two perch) 2 13,570.9 12.76 <0.001 ments, vision and vision + olfaction cues drove roach into Error 17 9,042.9 the structured habitat more than 80% of the time (Fig. 1). Cue type (one pike + 2 11,430.6 8.49 0.003 For the treatments that included only perch as predators, one perch) olfaction (two perch) and vision + olfaction (one and two Error 17 11,449.4 perch) cues resulted in roach selecting structured habitat SigniWcant values (P · 0.05) are shown in bold nearly 80% of the time. However, roach only utilized structured habitat 25–30% of the time when presented with just (behavioral response) of roach were not independent. For a visual cue of one or two perch (Fig. 1). If roach were pre- the additive design, separate analyses were conducted sented with only olfactory or visual cues from one for the trials run with vision and vision + olfaction cues. For pike + one perch together, the prey selected structured hab- the substitution design, separate analyses were run for all itat 20–25% of the time. Conversely, when roach were three cue types. introduced to the vision + olfaction cues of one pike + one

2 pike 2 perch 1 pike+1 perch 1 pike 1 perch control 100 A,1 A,1 A,2 A,2 B,1 A,2 A,1 A,2 80

AB AB 60 B B,1 A,1 40 A,1 B,2 B,1

Time in structured habitat (%) habitat Time in structured No No Data Data 0 Olfaction Vision Vision + olfaction Cpue ty e

Fig. 1 Frequency of roach (Rutilus rutilus) selection of artiWcial sea- DiVerent letters indicate diVerences at P · 0.05 among predator treat- grass habitat (ASU; mean + 1 SE) when exposed to diVerent cue types ments within each cue treatment and numbers indicate diVerences at (olfaction, vision, and vision + olfaction) and predator Welds (no pred- P · 0.05 among each cue treatment within each predator Weld (ANO- ator, one pike, one perch, one pike + one perch, two pike, two perch). VA results provided in Table 1) 123 898 Oecologia (2010) 162:893–902 perch they selected the ASU half of the aquaria 70% of the 100 time (Fig. 1). Notably, no diVerences in prey response were Expected (additive) Expected (substitutive) found between single predator species treatments regard- Observed less if cues from one or two individuals were used (i.e., ) 80 % ( vision or vision + olfaction cues from either one pike and t a t i two pike or one perch and two perch), suggesting that cue b a h saturation was achieved with the addition of a single preda- 60 d e tory Wsh (Fig. 1). r u t c

Statistical support for the occurrence of multiple preda- u r t

V s 40 tor e ects was highly dependent on the cue type presented n i to roach (Table 2). For olfactory cues, only the substitutive e m design was used to evaluate if pike and perch together gen- i T erated non-linear behavioral responses from the roach. 20 Based on the separate two-pike and two-perch treatments, we estimated that roach should select structured habitat »55% of the time in the one pike + one perch treatment. 0 Olfaction Vision Vision + Interestingly, observed roach selection of ASU habitat olfaction was signiWcantly less than what we expected (»25%; Cue Type P =0.036; Fig.2). Fig. 2 “Expected” and observed results for roach habitat selection in When roach were presented with visual cues of one one pike + one perch treatments. Both additive (black bars) and substi- pike + one perch, both the additive and substitutive designs tutive (gray bars) approaches were employed to calculate expected and demonstrated that the roach again used structured habitat assess multiple predator eVects on the frequency of ASU habitat selection by roach (mean + 1 SE) less than would be predicted based on the separate pike- prey and perch-prey interactions. We expected roach to select structured habitat »60% and »55% of the time in the one pike + one perch treatment based on the additive and designs, we expected that roach would select structured substitutive designs, respectively. However, we observed habitat »90% and »85% of the time in the mixed-predator that roach selected structured habitat in the mixed-predator treatment, respectively. Ultimately, we observed that roach treatment only »25% of the time, which was unambigu- selected ASU habitat »75% of the time, which was not a ously diVerent from our expectations (additive, P =0.017; signiWcantly diVerent result (additive, P = 0.118; substitu- substitutive, P = 0.024). tive, P = 0.362). We did note that in all Wve analyses, When roach were presented with vision + olfaction cues, observed selection of structured habitat in mixed-predator both the additive and substitutive experimental approaches treatments was less than expected by averaging the separate revealed that the behavioral response of roach to multiple pike-roach and perch-roach results. predators could be predicted by averaging individual predator–prey interactions. Using the additive and substitutive Discussion

Juvenile roach responded strongly to the cues of the pike Table 2 Results of t-tests for the diVerences in “expected” and ob- and perch predators, but the speciWc responses (habitat served habitat selection of roach using both the additive and substitu- selection) of roach were varied and partially counter-intuitive designs tive (Fig. 3). While fear is known to inXuence the behavior Source df t P of prey (Brown 2003; Wirsing et al. 2007), our pike-perch- roach data indicate that fear-driven responses are context- Additive speciWc. SpeciWcally, our results suggest that roach are not Vision 12 2.77 0.017 only able to distinguish between species-speciWc predator Vision + olfaction 12 1.68 0.118 cues, but also demonstrate unique responses based on the Substitutive type of cue that is detected. For instance, roach responded Olfaction 10 2.43 0.036 to the olfactory cues of pike by entering open-water habitat, Vision 12 2.57 0.024 but entered structured habitat when presented with a visual Vision + olfaction 12 0.95 0.362 cue of this predator. When presented with visual or olfac- SigniWcant values (P · 0.05) are shown in bold and are indicative of tory cues of perch, the opposite responses were elicited non-linear multiple predator eVects from roach. Interestingly, roach defaulted to selection of 123 Oecologia (2010) 162:893–902 899

be able to determine the direction from which a camou- Xaged, stationary pike will attack. As a result, they seek open habitat to increase their opportunity to locate pike visually (sensu Horinouchi et al. 2009; Schultz et al. 2009). Conversely, once a stationary pike is located visually (alone or in combination with olfaction), roach can use structured habitat for the mechanical defense it provides against directed pike attacks (Fig. 3). Alternatively, perch move continuously while foraging in structured and unstructured habitats (Eklöv and Persson 1995). As a result, olfactory cues may be received by roach as an unambiguous signal that perch have entered the immediate vicinity, and therefore present a pressing, short-term threat. In this circum- stance, our data suggest that roach prefer structured habitat as refuge. Within structured habitats, visual encounters with perch are informative but potentially less reliable than Fig. 3 Conceptual diagram illustrating the strength of habitat prefer- with ambush predators since roving predators will likely ence for open or structured habitat by roach with olfactory and/or vi- move in and out of the visual Weld routinely. Therefore, sual cues from pike and/or perch. The length of the arrow scales with once a perch has been visually detected, roach may be con- the magnitude of observed preference for either habitat type during trials with each combination of cue type and predator Weld (calculated in ditioned to seek open habitats in an attempt to maintain comparison to the average of all control treatments in which no pred- more constant visual contact with this roving predator ator cues were included) (unless the immediate threat associated with perch scent cues is also perceived). In fact, we frequently encountered roach in open habitats adjacent to structured habitats while structured habitat when presented with vision + olfaction capturing organisms for our trials. The emergent theme of cues of either predator. Moreover, when roach were pre- our results is that structured habitat serves as optimal refuge sented with olfactory cues of both predators, the roach habitat only if roach are able to detect the speciWc location behaved as though they were ignoring the perch, and (direction) from which a predation threat will come from responded just as they did during the pike-only treatments (Horinouchi et al. 2009; Schultz et al. 2009). (choosing open-water habitat). When visual cues of both Although aquatic vegetated habitats can decrease the predators were presented to the prey, roach responded as foraging eYciency of predators (Nelson and BonsdorV they did in the perch-only treatments (again selecting 1990; Mattila 1992), our results suggest that prey may not open-water habitat). Upon being presented with always perceive structure as optimal refuge space: a pattern vision + olfaction cues of both predators, however, roach previously documented in terrestrial studies. For example, strongly favored structured habitat; consistent with their Valeix et al. (2009) found that lions utilized tall grass as responses to detecting pike visually or perch through olfac- cover during the day while stalking/ambushing prey. In tion. Our results indicating increased use of structured hab- instances that prey (multiple ungulate species) became itat by roach when presented with vision + olfaction cues aware of threatening lions, they spent more time during the agree with previous studies that have examined the forag- day in open habitat away from grasslands. In this case, tall ing eYciency of roach predators (i.e., both visual and olfac- grass did serve as the preferred refuge for these prey spe- tory stimuli are available to roach; Nelson and BonsdorV cies at night, when lions were able to stalk/ambush prey 1990; Mattila 1992), but also highlight the more detailed, equally well in open habitat (FischhoV et al. 2007). cue- and predator-speciWc responses of this common forage Recently, several marine studies have identiWed speciWc Wsh. predator–prey scenarios that can result in prey beneWting The speciWc antipredator responses of roach likely reX- from less-structured habitats. Notably, these studies and our ect two interacting factors: the foraging strategies of preda- own experiment all suggest that if structure increases tory pike and perch and the diVerence between only uncertainty for prey in locating predation threats, prey spe- perceiving a predation threat versus perceiving and locating cies do not necessarily associate complex habitats with suit- that threat. Northern pike forage primarily by occupying able refuge. Studies on gobies under pressure from cryptic structured habitats and ambushing prey (Savino and Stein ambush predators (sculpin) have shown that prey can expe- 1989). We hypothesize that roach may escape structured rience elevated mortality while occupying dense seagrass habitat when a pike is detected through only olfactory cues, patches (Horinouchi et al. 2009). Similarly, James and because, although roach are aware of threat, they may not Heck (1994) showed that increasing density of artiWcial 123 900 Oecologia (2010) 162:893–902 vegetation had no impact on the foraging eYciency of responded to multiple predators by shifting toward open another well-camouXaged ambush predator, the seahorse. water. However, we also noted anecdotally that individual In tropical environments, the Weld of vision for damselWsh roach switched between unstructured and structured habi- can be limited by complex coral reefs. Subsequently, the tats more often when simultaneously presented with cues behavior of damselWsh is altered such that mating and feed- from both pike and perch. Such behavior has been previ- ing opportunities are lost, negatively aVecting individual ously documented with roach under predation threat (Chris- Wtness (Rilov et al. 2007). tensen and Persson 1993). We propose that this increase in While behaviors can be observed, determining the deci- emergence rates would increase their risk of being captured sion rules that produce those behaviors is extremely diY- (Sih 1997). As a result, we predict that multiple predator cult. With this in mind, we are left to speculate on the role scenarios will result in enhanced mortality for roach despite of uncertainty in mitigating the defensive responses of its Wne-tuned defensive behaviors. roach to evaluate behavior-related multiple predator eVects. While GriVen (2006) has demonstrated that additive and In both olfaction- and vision-only treatments, we observed substitutive experimental approaches can lead to quantita- a non-linear response by roach to the simultaneous detec- tively and qualitatively diVerent results, we found that these tion of one pike + one perch. In each instance, we suspect approaches supported one another in investigating the anti- this occurred because roach eVectively ignored the predator predator responses of roach (Table 2). Although these for which the introduced cue should reveal the predator’s approaches diVer in the way predator density and identity position (perch in olfaction trials, pike in vision trials). are manipulated, roach responded similarly whether we Rather, the roach behaved as if only the predator that could included one or two pike, or one or two perch in single- be detected but not consistently located determined the predator treatments (Fig. 1). Essentially, the presence of a response that should minimize overall risk (Table 2; single individual pike or perch was suYcient to induce Fig. 3). Conversely, when vision + olfaction cue combina- roach to demonstrate their full antipredator response. This tions of both predators were presented to roach, thereby is not surprising given the relative size of our experimental providing the prey with information to detect and locate mesocosms and that of predators/prey in these systems, but both species, roach responded as expected by balancing its it may also suggest that roach do not respond as though response to account for two equally capable predators there is intraspeciWc facilitation or interference among pike (Table 2; Fig. 3). Thus, it is not only the fear of predators or perch. As a result, there was little diVerence in expected that aVects the behavior of prey (Pressier et al. 2005), but habitat use for the one pike + one perch treatment regard- also the fear of ambiguity in knowing where attacks will less of whether we employed the additive or substitutive come from. design. More broadly, our data contribute to a growing literature Ultimately, this study should be a gateway to further that indicates community dynamics cannot be predicted by experiments of roach-pike-perch interactions. We con- simply summing or averaging all pairwise predator–prey ducted our experiments on single roach, ignoring the social interactions. Although previous studies have mostly cues that might be relied on by this schooling species to focused on changes in prey abundance (e.g., Fodrie et al. avoid predators (Magurran and Seghers 1994). Schooling, 2008), quantifying non-consumptive eVects in scenarios for example, should increase the awareness of each individ- with multiple predators should also elucidate important tro- ual roach to the presence and location of predators by the phic dynamics. First, several studies have demonstrated increased probability of detecting predator olfactory and/or distinct and complex behavioral responses to visual and visual cues when multiple prey species are present. While olfactory cues that are known to aid prey in escaping single collecting roach for our experiments, we routinely observed predators (e.g., Brown and Cowan 2000; Vincent et al. individuals swimming together in open water near emer- 2005; Brown et al. 2006). Second, emergent multiple pred- gent or submerged vegetation. Thus, additional tests are ator eVects typically follow from the behavioral response(s) needed to determine whether schooling and solitary roach of a shared prey after detecting one or more predators (Sih respond diVerently to threats perceived though olfactory et al. 1998). Ultimately, quantifying these non-consumptive and visual cues. Also, a future experiment should examine eVects are valuable because: (1) a shift in prey activity can if multiple predators (pike and perch) interact to generate have impacts that propagate through the food web enhanced or reduced mortality risk for juvenile roach in (Grabowski 2004), and (2) understanding the defensive landscape mosaics of open and structured habitats. responses (habitat selection) of prey against predators Lastly, we recognize that the habitat selection of roach provides a framework for predicting whether multiple may be the result of instinctual responses or learned anti- predators should have independent, risk-reducing, or risk- predator behaviors (Ferrari et al. 2005; Brown and Chivers enhancing eVects. For instance, our results strongly suggest 2006; Leduc et al. 2007) prompted by earlier attacks. Thus, that, for all of the sensory cues we investigated, roach conditioning experiments could investigate if the responses 123 Oecologia (2010) 162:893–902 901 we observed are plastic, and therefore capable of evolving Brown GE, Harvey MC, Leduc AOHC, Ferrari MCO, Chivers DP to meet the demands of temporally and spatially varying (2009) Social context, competitive interactions and the dynamic W nature of antipredator responses of juvenile rainbow trout. J Fish predator elds and changing environmental conditions Biol 75:552–562 (e.g., increasing turbidity or loss of vegetated habitats). For Chivers DP, Smith RJF (1998) Chemical alarm signaling in aquatic instance, as turbidity increases, roach are likely to demon- predator–prey systems: a review and prospectus. Eucoscience strate a bias for olfactory cues to sense predators. In 5:338–352 Christensen B, Persson L (1993) Species-speciWc antipredatory behav- instances with nearby multiple predators, this bias should iours: eVects on prey choice in diVerent habitats. Behav Ecol So- lead to greater pike-avoidance behaviors by roach ciobiol 32:1–9 (Table 2), regardless of the distribution of perch. As a Crowder L, Squires D, Rice J (1997) Non-additive eVects of terrestrial result, roach-perch encounters may increase (ultimately, and aquatic predators on juvenile estuarine Wsh. Ecology V 78:1796–1804 predation rates will also be a ected by other co-occurring Dill LM, Heithaus MR, Walters CJ (2003) Behaviorally mediated indi- changes in submerged aquatic vegetation cover and capture rect interactions in marine communities and their conservation eYciency in turbid water). Clearly, there is considerable implications. Ecology 84:1151–1157 opportunity to further explore the complexity of predator– Eklöv P, Persson L (1995) Species-speciWc antipredator capacities and prey refuges: interactions between piscivorous perch (Perca Xu- prey interactions and the non-intuitive outcomes that can viatilis) and juvenile perch and roach (Rutilus rutilus). Behav result from defensive behaviors of prey beyond simple hab- Ecol Sociobiol 37:169–178 itat preference (although this likely represents one of the Engström-Öst J, Mattila J (2008) Foraging, growth and habitat choice W foremost antipredator defenses). Although applied to a in turbid water: an experimental study with sh larvae in the Bal- tic Sea. Mar Ecol Prog Ser 359:275–281 freshwater and estuarine ecosystem here, our results may be Ferrari MCO, Trowell JJ, Brown GE, Chivers DP (2005) The role of broadly applicable to many other environments. learning in the development of threat-sensitive predator avoidance by fathead minnows. Anim Behav 70:777–784 Acknowledgments Support for this project was provided through Finke DL, Denno RF (2002) Intraguild predation diminished in com- general funds from the University of South Alabama’s Department of plex-structured vegetation: implications for prey suppression. Marine Science, as well as Husö Biological Station, Åbo Akademi Ecology 83:643–652 University. We thank S. Scyphers, M. Scheinin, M. Ajemian, and M. FischhoV IR, Sundaresan SR, Cordingley J, Rubenstein DI (2007) Kenworthy for assistance in collecting Wsh and running trials, as well Habitat use and movement of plains zebra (Equus burchelli) in re- as the staV and students at Husö Biological Station for their logistical sponse to predation danger from lions. Behav Ecol 18(4):725–729 support throughout our visit. We also thank J. Valentine, M. Ajemian, Fodrie FJ, Kenworthy MK, Powers SP (2008) Unintended facilitation B. Toscano, and two anonymous reviewers for their constructive criti- between marine consumers generates enhanced mortality for their cism and improvements to this manuscript. All experiments were in shared prey. Ecology 89(12):3268–3274 compliance with the laws of Finland. Grabowski JH (2004) Habitat complexity disrupts predator-prey interactions yet preserves the trophic cascade in oyster-reef communities. Ecology 85:995–1004 GriVen BD (2006) Detecting emergent eVects of multiple predator References species. Oecologia 148:702–709 Hager MC, Helfman GS (1991) Safety in numbers: shoal size choice Ådjers K, Appelberg M, Eschbaum R, Lappalainen A, Minde A, by minnows under predatory threat. Behav Ecol Sociobiol Repecka R, Thoresson G (2006) Trends in coastal Wsh stocks of 29:271–276 the Baltic Sea. Bor Env Res 11:13–25 Heithaus MR, Dill LM, Marshall GJ, Buhleier B (2002) Habitat use Amo L, Lopez P, Martin J (2004) Chemosensory recognition and behav- and foraging behavior of tiger sharks (Baleocerdo cuvier) in a ioural responses of wall lizards, Podarcis muralis, to scents of seagrass ecosystem. Mar Biol 140:237–248 snakes that pose diVerent risks of predation. Copeia 2004:691–696 Heithaus MR, Hamilton IM, Wirsing AJ, Dill LM (2006) Validation of Baden S, Boström C (2001) The leaf canopy of seagrass beds: faunal a randomization procedure to assess animal habitat preferences: community structure and function in a salinity gradient along the microhabitat use of tiger sharks in a seagrass ecosystem. J Anim Swedish coast. Ecol Stud 151:214–236 Ecol 75:666–676 BonsdorV E, Blomqvist EM (1993) Biotic couplings on shallow water Hickman CR, Stone MD, Mathis A (2004) Priority use of chemical soft bottoms: examples from the northern Baltic Sea. Oceanogr over visual cues for detection of predators by neotenic graybelly Mar Biol Annu Rev 31:153–176 salamanders, Eurycea multiplicata griseogaster. Herpetologica Brown G (2003) Learning about danger: chemical alarm cues and local 60:203–210 risk assessment in prey Wshes. Fish Fish 4:227–234 Horinouchi MN, Jo Y, Fujita M, Sano M, Suzuki Y (2009) Seagrass Brown GE, Chivers DP (2006) Learning about danger: chemical alarm habitat complexity does not always decrease foraging eYciencies cues and local risk assessment in prey Wshes. In: Brown C, Laland of piscivorous Wshes. Mar Ecol Prog Ser 377:43–49 KN, Krause J (eds) Fish cognition and behaviour. Blackwell, James PL, Heck KL (1994) The eVects of habitat complexity and light London, pp 49–69 intensity on ambush predation within a simulated seagrass habi- Brown GE, Cowan J (2000) Foraging trade-oVs and predator inspec- tat. J Exp Mar Biol Ecol 176:187–200 tion in an Ostariophysan Wsh: switching from chemical to visual Kats LB, Dill LM (1998) The scent of death: chemosensory assessment cues. Behaviour 137:181–196 of predation risk by prey animals. Ecoscience 5:361–394 Brown GE, Rive AC, Ferrari MCO, Chivers DP (2006) The dynamic Kim J, Brown GE, Dolinsek IJ, Brodeur NN, Leduc AOHC, Grant nature of anti-predator behaviour: prey Wsh integrate threat-sensi- JWA (2009) Additive and interactive eVects of chemical and tive anti-predator responses within background levels of preda- visual information in eliciting antipredator behaviour in juvenile tion risk. Behav Ecol Sociobiol 61:9–16 Atlantic salmon, Salmo salar L. J Fish Biol 74:1280–1290 123 902 Oecologia (2010) 162:893–902

Kusch RC, Mirza RS, Chivers DP (2004) Making sense of predator Schmitz OJ, Sokol-Hessner L (2002) Linearity in the aggregate eVects scents: investigating the sophistication of predator assessment of multiple predators in a food web. Ecol Lett 5:168–172 abilities of fathead minnows. Behav Ecol Sociobiol 55:551 Schmitz OJ, Krivan V, Ovadia O (2004) Trophic cascades: the primacy Lang A (2003) Intraguild interference and biocontrol eVects of gener- of trait-mediated indirect interactions. Ecol Lett 7:153–163 alist predators in a winter wheat Weld. Oecologia 134:144–153 Schultz ST, Kruschel C, Bakran-Petricioli T (2009) InXuence of sea- Leduc AOHC, Roh E, Breau C, Brown GE (2007) Learned recognition grass meadows on predator-prey habitat segregation in an Adriat- of a novel odour by wild juvenile Atlantic salmon (Salmo salar) ic lagoon. Mar Ecol Prog Ser 374:85–99 under fully natural conditions. Anim Behav 73:471–477 Shivik JA (1998) Brown tree snake response to visual and olfactory Lehtiniemi M, Engström-Öst J, Viitasalo M (2005) Turbidity decreas- cues. J Wildl Manage 62(1):105–111 es anti-predator behaviour in pike larvae (Esox lucius). Environ Sih A (1997) To hide or not to hide? Refuge use in a Xuctuating envi- Biol Fish 73:1–8 ronment. Trends Ecol Evol 10:375–376 Lindquist SB, Bachmann MD (1982) The role of visual and olfactory Sih A, Englund G, Wooster D (1998) Emergent impacts of multiple cues in the prey catching behavior of the tiger salamander, Ambys- predators on prey. Trends Ecol Evol 13:350–355 toma tigrinum. Copeia 1982(1):81–90 Smith RJF (1992) Alarm signals in Wshes. Rev Fish Biol Fish 2:33–63 Magurran AE, Seghers BH (1994) Predator inspection behaviour Trussell GC, Ewanchuk PJ, Bertness MD (2002) Field evidence of covaries with schooling tendency amongst wild guppy, Poecilia trait-mediated indirect interactions in a rock intertidal food web. reticulata, populations in Trinidad. Behaviour 128:121–134 Ecol Lett 5:241–245 Mathis A, Smith RJF (1993) Fathead minnows (Pimephales promelas) Utne-Palm AC (2002) Visual feeding of Wsh in a turbide environment: learn to recognize pike (Esox lucius) as predators on the basis of physical and behavioural aspects. Mar Freshwater Behav Physiol chemical stimuli from minnows in the pike’s diet. Anim Behav 35:111–128 46:645–656 Valeix M, Loveridge AJ, Chamaillé-Jammes S, Davidson Z, Murinda- Mathis A, Vincent F (2000) DiVerential use of visual and chemical gomo F, Fritz H, MacDonald DW (2009) Behavioral adjustments cues in predator recognition and threat-sensitive antipredator of African herbivores to predation risk by lions: spatiotemporal behaviour by larval central newts, Notophthalmus viridescens. variations inXuence habitat use. Ecology 90:23–30 Can J Zool 78:1646–1652 Vance-Chalcraft HD, Soluk DA, Ozburn N (2004) Is prey predation Mathis A, Chivers DC, Smith RJF (1993) Population diVerences in re- risk inXuenced more by increasing predator density or predator sponses of fathead minnows (Pimephales promelas) to chemical species richness in stream enclosures? Oecologia 139:117–122 and visual stimuli from predators. Ethology 93:31–40 Vincent SE, Shine R, Brown GP (2005) Does foraging mode inXuence Mattila J (1992) The eVect of habitat complexity on predation eY- sensory modalites for prey detection? A comparison between ciency of perch (Perca Xuviatilis L.) and ruVe (Gymnocephalus males and female Wlesnakes (Acrochordus arafurae Acrochordi- cernuus L.). J Exp Mar Biol Ecol 157:55–67 dae). Anim Behav 70:715–721 Mikheev VN, Wanzenböck J, Pasternak AF (2006) EVects of predator- Wahle RA (1992) Body-size dependent anti-predator mechanisms of induced visual and olfactory cues on 0+ perch (Perca Xuviatilis the American lobster. Oikos 65:52–60 L.) foraging behavior. Ecol Freshwater Fish 15:111–117 Warfe DM, Barmuta LA (2004) Habitat structural complexity medi- Montgomery JC, MacDonald JA (1987) Sensory tuning of lateral line ates the foraging success of multiple predator species. Oecologia receptors in antarctic Wsh to movement of planktonic prey. Sci- 141:171–178 ence 235:195–196 Waycott M, Duarte CM, Carruthers TJB, Orth RJ, Dennison WC, Moran MD (2003) Arguments for rejecting the sequential Bonferroni Olyarnik S, Calladine A, Fourqurean JW, Heck KL Jr, Hughes in ecological studies. Oikos 100:403–405 AR, Kendrick GA, Kenworthy WJ, Short FT, Williams SL (2009) Nelson WG, BonsdorV E (1990) Fish predation and habitat complexity: Accelerating loss of seagrasses across the globe threatens coastal are complexity thresholds real? J Exp Mar Biol Ecol 141:183–194 ecosystems. Proc Natl Acad Sci 106(30):12377–12381 Pimm SL, Lawton JH (1978) On feeding on more than one trophic lev- Werner EE, Peacor SD (2003) A review of trait-mediated indirect el. Nature 275:542–544 interactions. Ecology 84:1083–1100 Pressier EL, Bolnick DI, Benard MF (2005) Scared to death? The Wirsing AJ, Heithaus MR, Dill LM (2007) Living on the edge: dug- eVects of intimidation and consumption in predator–prey interac- ongs prefer to forage in microhabitats that allow escape from rath- tions. Ecology 86:501–509 er than avoidance of predators. Anim Behav 74:93–101 Rilov G, Figueira WF, Lyman SJ, Crowder L (2007) Complex habitats Ylönen H, Kortet R, Myntti J, Vainikka A (2007) Predator odor recog- may not always beneWt prey: linking visual Weld, reef Wsh behav- nition and antipredatory response in Wsh: does the prey know the ior and distribution. Mar Ecol Prog Ser 329:225–238 predator diel rhythm? Acta Oecol 31:1–7 Savino JF, Stein RA (1989) Behavioural interactions between Wsh predators and their prey: eVects of plant density. Anim Behav 37:311–321

123