Minnows trust conspecifics more than themselves ...

Minnows trust conspecifics more than themselves when faced with conflicting information about predation risk Adam L. Cranea† and Maud C.O. Ferrarib a

Department of Biology, University of Saskatchewan, 112 Science Place, Saskatoon, SK, S7N 5E2, Canada Email: [email protected] Phone: 306-370-0816 b

Department of Veterinary Biomedical Science, University of Saskatchewan, 52 Campus Drive, Saskatoon, SK, S7N 5B4, Canada Email: [email protected] Phone: 306-966-4317 †

corresponding author

Abstract Prey often face uncertainty when learning about predation risk because stimuli indicating risk can vary in reliability. However, the way this uncertainty is expressed at the individual level is often poorly understood. Here, we compared how prey fish (Pimephales promelas) responded to information conflicting with their previous experience when this information came from contrasting sources. First, fish had the opportunity to learn that a novel odour was safe from repeated exposure to the odour in the absence of negative consequences, or they received pseudo-exposures. Then, using one of two learning paradigms, we conditioned fish to recognize that the “safe” odour was actually the odour of a predator. Fish were exposed to the odour paired with either (1) cues released from injured conspecifics (alarm cues), allowing for the fish to learn alone, or (2) cues from a knowledgeable (frightened) conspecific responding to the risky odour, allowing fish to acquire the information via social learning. Fish were tested individually following conditioning, and movement, foraging, shelter use, and freezing were quantified.

Learned antipredator responses were similar between the two mechanisms for individuals with no prior exposure to the odour. However, fish that knew the odour as safe did not acquire a fearful response to the odour following conditioning with alarm cues, whereas interaction with frightened conspecifics appeared to cause fish to ignore their prior learning of safety, suggesting that learning from a live conspecific was more persuasive than individual assessment via alarm cues. This study adds to the body of literature contrasting the reliability of information sources and their consequences on cognition, communication and group dynamics.

Keywords: social learning, uncertainty, latent inhibition, alarm-cue learning, decision making, Pimephales promelas

Introduction Group-living animals, and those in close aggregations, have the potential to acquire knowledge or skills from observing others (Hoppitt & Laland 2013). While acquiring information on one’s own may sometimes be more relevant and more reliable, it can also be time consuming, dangerous, and quickly outdated (Rieucau & Giraldeau 2011). Social learning, however, can be a fast learning mechanism that helps animals adjust to complex and changing environments with limited exposure (Rendell et al. 2010). By eavesdropping on publiclyavailable information, naïve animals can quickly learn to find and capture food, how to choose mates, and how to successfully avoid predators (Crane & Ferrari 2013; Galef & Giraldeau 2001; White 2004). Predation is a major evolutionary force that shapes many characteristics of prey including their behavioural defences (Lima & Dill 1990). However, the risk of predation can be unpredictable for prey because threats can fluctuate in time and space, and thus, prey often face uncertainty about predation risk (Ferrari et al. 2010a; Sih 1992). Although prey can often avoid predation through avoidance of risky habitats, or risky times, they often cannot be confined to those safe niches because these are usually associated with no or low food gain (Ferrari, Sih & Chivers 2009; Lima & Bednekoff 1999). Therefore, it is critical for prey to correctly distinguish between situations of risk and safety and adjust their behaviours accordingly to maximize gain while minimizing their risk of being attacked (Lima & Dill 1990). Prey encounter a variety of stimuli that potentially indicate predation threat, and while some species innately recognize cues from some predators, many others must learn to recognize these cues as risky (Berejikian, Tezak & LaRae 2003; Brown & Chivers 2005; Ferrari, Wisenden & Chivers 2010b). Learning from direct experience with predators should reduce uncertainty

about risk but also poses great costs, including death (Arai et al. 2007; Griffin & Boyce 2009; Griffin & Haythorpe 2011). In contrast, learning indirectly, and without being exposed to attack, should decrease the chance of mortality, but it potentially brings the cost of learning something inaccurate or irrelevant (Crane & Ferrari 2013; Danchin et al. 2004). Learning from frightened conspecifics (i.e., models or demonstrators) is usually demonstrated with a standard methodological approach where a predator-naïve individual (the observer) is exposed to novel predator stimuli that are paired with a predator-experienced individual (the model) (Mathis, Chivers & Smith 1996). During this conditioning phase, the observer has the opportunity to learn from the model displaying a fright response, and subsequently, the observer is tested with exposure to the predator stimulus (or live predators) in the absence of the model to determine whether learning occurred (Crane & Ferrari 2013; Mathis et al. 1996). This form of learning commonly referred to as social learning. Once social learning has occurred, the observer may now serve as an experienced model for inexperienced individuals, thereby initiating a chain of transmission where the information can be passed multiple times (Crane & Ferrari 2013). Previous work suggests the intensity of the learned response weakens as the chain becomes longer (Cook et al. 1985). However, the preservation of the fright response has been shown in a 3× chain of zebrafish (Brachydanio rerio) (Suboski et al. 1990) and a 7× chain of blackbirds (Turdus merula) (Curio 1988). Another learning mechanism that is available in a wide variety of aquatic organisms is alarm-cue learning (Ferrari et al. 2010b). This occurs when chemical cues (usually from conspecifics) are released by injury during a predation event. Prey then have the opportunity to detect these alarm cues, associate them with other stimuli (e.g., the sight, smell or sound of a predator), and then learn these as indicators of risk. In recent years, this form of learning has

become viewed as a form of social learning because the alarm cues are social in that they are released by companion individuals (Griffin 2004; Lindeyer & Reader 2010; Zentall & Galef Jr 2013). However, contrary to true social cues, the information transmission (timing of release, quantity, quality) is not controlled by the sender, and the information content is not modifiable based on the interpretation of the sender, and hence, cannot be sent dishonestly. For these reasons, we tend to consider alarm cues to be non-social cues, in much the same way the -odour of a predator is not considered a social cue. In the context of this study, we will refer to social learning when the learning results from interacting with live conspecifics, which has the ability to modulate the cues detected by the observer. While there are a variety of cues available for prey to use in predator-recognition learning, few comparisons have been made among different learning modalities. Some work has compared learning from visual cues from models to learning via auditory cues from models (alarm calls), finding learned responses from these types of cues were similar (Curio, Ernst & Vieth 1978; Vieth, Curio & Ernst 1980). However, we might expect prey to learn better (and/or have less uncertainty) with some types of cues compared to others (Crane & Ferrari 2013). For instance, visual cues from models can be highly accurate in space and time and usually provide information about the target of the response, but they also require the observer to recognize that a model has changed its normal behaviour, which may be unlikely in complex habitats or at night. Chemical cues, however, are available all the time, and can travel long distances if moved by air or water, but may be less reliable in space and time, as they persist long after predation occurs, and are moved by currents that can create erroneous pairings (Ferrari et al. 2010b). In contrast to learning about risk, prey can also learn safety (i.e., recognizing stimuli as non-threatening) via repeated encounter with novel stimuli in the absence of negative

consequences (Ferrari & Chivers 2011). When this prior experience subsequently prevents a learned association between the stimuli and risk, latent inhibition has occurred (Acquistapace, Hazlett & Gherardi 2003; Lubow 1973). For example, when damselfish (Pomacentrus moluccensis) were repeatedly exposed (6×) to a novel odour, they failed to learn it as risky during a subsequent pairing with alarm cues. However, repeated pairing with alarm cues (≥ 3×) reversed (i.e., released) the learned safety (Mitchell et al. 2011). Latent inhibition of alarm-cue learning has been shown in other aquatic species (e.g., Acquistapace et al. 2003; Ferrari & Chivers 2006b, 2011) and for auditory learning via alarm calls in birds (Vieth et al. 1980). When blackbirds (Turdus merula) were previously exposed to mounted owls (predators) and then exposed to a mounted owl paired with conspecific alarm calls, they subsequently ignored the mock predator (Vieth et al. 1980). However, when the mounted owl was paired with the sight of a conspecific mobbing the owl, observers became wary of the predator indicating the visual cues from models were more persuasive than auditory cues (Vieth et al. 1980). Similarly, rhesus monkeys (Macaca mulata), that had previously shown no fear toward a toy snake (predator) when alone, immediately learned that it was dangerous when interacting with conspecifics that were frightened of the snake (Mineka & Cook 1986). Interaction with live conspecifics also appears particularly persuasive in other contexts such as communication (song) learning and the length of the sensitive learning period in birds (Zonotrichia leucophrys) (Baptista & Petrinovich 1984, 1986; Nelson 1998; Petrinovich & Baptista 1987). However, socially-learned food preferences in rats (Rattus norvegicus) waned after individually experiencing negative consequences and having more time with alternatives (Galef & Whiskin 2001). These studies draw attention to the dilemmas animals face from encountering different types of information.

In this study, we sought to compare the value of information learned via a live conspecific to that learned via alarm cues, using fathead minnows (Pimephales promelas). Both forms of learning can occur after only a single conditioning and can also increase survival in a predation context (Manassa & McCormick 2013; Mirza & Chivers 2000). This comparison involves some inherent differences between these two forms of learning. For instance, social learning requires the presence of another individual whereas alarm-cue learning does not. Moreover, interacting with frightened models should provide more information (Seyfarth et al. 2010) about risk, compared to alarm-cue learning. The presence of frightened models can provide visual, chemical and mechanical cues for the observers (Johnston & Johnson 2000; Vavrek et al. 2008), although chemical disturbance cues do not mediate predator-recognition learning (Ferrari et al. 2008). In contrast, only chemical cues are provided during alarm-cue learning and are not modifiable by the model, whereas social cues are, whereas alarm cues. We first gave minnows the opportunity to learn a novel odour (pike, Esox Lucius, odour) as safe, by being repeatedly exposed to the odour without negative reinforcement. We then used a conflicting context to force minnows to make a choice between the value of their previously learned information (the cue is safe) and new information (the cue is risky) acquired via the alarm-cue or social learning mechanisms. After this conflicting phase, the minnows were exposed alone to the odour, and their antipredator behaviour was measured. Our control group allowed us to compare the intensity of the learned response via the two mechanisms. We expected minnows that learned from models to respond either (1) as much as those learning from alarm cues because the models learned from the same alarm-cue concentration or (2) slightly less than those learning from alarm cues because the intensity of the response usually weakens

through a chain of transmission (Crane & Ferrari 2013). We also expected a single pairing of alarm cues would not be enough to override the prior learning of safety. Methods Fish collection and maintenance Fathead minnows are a small fish species, living in shoals in ponds, lakes and rivers throughout North-America, and are prey to many young fish piscivores, birds, snakes and aquatic invertebrates (Warren, Burr & Tomelleri 2014). In September 2013, we trapped adult minnows (unsexed; total length: 25–50 mm) at Feedlot Pond on the University of Saskatchewan campus using Gee’s inverted minnow traps set over night. Minnows at this pond are prey for invertebrate and avian predators, but extensive trapping over the past two decades has revealed no fish predators at this site (i.e., these minnows are naïve to fish predation), and this population lacks innate recognition of fish predators, such as pike or trout. Several studies with minnows from this naïve population have demonstrated recognition of novel pike odour via alarm-cue learning (Chivers & Smith 1994; Ferrari et al. 2005; Kelley & Magurran 2003) or social learning (Chivers & Smith 1995; Mathis et al. 1996), both with only a single pairing of the conditioned and unconditioned stimuli. Minnows were housed in 76-L flow-through tanks, containing a gravel substrate, an airstone and maintained under a 15:9 L:D cycle. They were fed flake food and received ~30% water change every day. Minnow alarm cues We used standard procedures for making alarm cues (Crane et al. 2011; Ferrari & Chivers 2006a) and sacrificed five minnows (35–43 mm total length) with a blow to the head (in accordance with the Canadian Council on Animal Care). We then removed a total of 11.7 cm2 of skin from the minnows. The skin was placed into a beaker with 20 mL of system water and

homogenized (Polytron PT-2500E). The resulting solution was then diluted according to an established protocol for diluting minnow alarm cues to reach a concentration of ~1 cm2 per 40 L (Ferrari, Capitania-Kwok & Chivers 2006a; Ferrari & Chivers 2006a; Ferrari et al. 2005). The water containing alarm cues (AC) was then frozen (-20 C) in 100 mL aliquots. Predator odour We collected three pike (size: 15-25 cm total length) from Pike Lake, SK, using a seine. While pike do not occur at Feedlot Pond, they are one of the primary native predators of minnows in the region. The pike were housed in 76-L tanks that were wrapped in black polyethylene sheeting to ensure visual isolation from surrounding tanks. Prior to stimulus collection, each pike was fed two swordtail (Xiphophorus hellerii) (total length: 30-40 mm) to help expulsion of any minnow diet cues (Ferrari, Messier & Chivers 2006b; Mathis & Smith 1993). After 4 d, pike were placed individually into 38-L tanks with clean water, filled with a volume of water proportional to the size of the pike (50 mL per g of fish). After 24 h without water filtration, pike were removed and the water was frozen in plastic bags (600 mL aliquots). This process was repeated three times to ensure a sufficient amount of predator odour (PO) was available for the experiment. The pike were returned in their holding tanks and fed immediately after the end of the odour collection. Ethical note In our laboratory, we fed pike with live prey because of the arduousness of training wildcaught pike to consume other foods. This study was approved by our University Committee on Animal Care and Supply (protocol: 20130079). We collected minnows under a Saskatchewan Ministry of Environment Special Collection Permit. Experimental protocol

The experiment consisted of three phases: (1) a prior-exposure phase, where solitary minnows were given either safe experience with previously-unknown PO (i.e., the latent inhibition group) or a sham experience (control group), then (2) a conditioning phase, where the minnows (observers) were exposed to PO paired with a pike-experienced model (i.e., social conditioning) or pike odour paired with AC (i.e., AC conditioning), and (3) a subsequent testing phase where the observers’ antipredator responses were recorded when exposed to water (control) or PO in the absence of models. Prior-exposure: Minnows were placed individually into 37-L experimental tanks containing gravel substrate and a shelter (10 × 10 cm ceramic tile with three 2-cm plastic legs) and were allowed to acclimate for 24 h before the first exposure. Each tank was equipped with an injection hose (150 cm piece of air tubing) where stimuli could be added slowly with a syringe and then flushed with tank water that had been withdrawn just prior to the injection (Ferrari et al. 2006a; Ferrari et al. 2006b). Over the next 3 d, minnows were exposed to either 20 mL of PO or blank water (W) twice a day (Table 1). In previous studies with fishes, 5–6 prior-exposures to PO was enough to ensure safety learning (Ferrari & Chivers 2006b; Mitchell et al. 2011). In this study, exposures occurred in the morning (08:00-10:00) and afternoon (13:00-16:00), and a 100% water change was conducted 1 h following each injection. Conditioning: After the sixth prior-exposure (and the subsequent water change), the shelters were removed from all the tanks to facilitate shoaling rather than agonistic behaviours among the fish in the social-learning group. Then, in half of the tanks one model was added (at ~1600 h), while a similar disturbance (dipping of a net in the tank) was given to the other tanks. Models were always slightly larger than observers (size difference < 10 mm total length), so that we could distinguish them. The models had been conditioned ~ 5 h earlier with a single exposure

to PO (20 mL) paired with AC (5 mL). Observers had the opportunity to fully interact with these models over the next 18-24 h. We then injected 20 mL of PO, giving observers the opportunity to learn from the models that pike odour was risky (i.e., social cues: “SC”, Table 1). At the same time for the alarm cue group, 20 mL of PO paired with 5 mL of AC was injected into each tank. All tanks received a water change and had shelters returned 1-4 h following conditioning. Models were removed with a dipnet, and we controlled for this disturbance by waving a dipnet next to the AC-conditioned individuals for ~10 s. Table 1 Summary of prior-exposure and conditioning treatments Treatment

Day 1

Day 2

Day 3

Day 4

Prior water × alarm cue conditioning

W,W

W,W

W,W

AC+PO

Prior water × social conditioning

W,W

W,W

W,W

SC+PO

Prior odour × alarm cue conditioning

PO,PO

PO,PO

PO,PO

AC+PO

Prior odour × social conditioning

PO,PO

PO,PO

PO,PO

SC+PO

W water; PO pike odour; AC alarm cue; SC social cue Testing: Testing took place the day following risk conditioning. Minnows were fed 1 h before testing, because hunger weakens responses to predation risk (Brown & Smith 1996; Whitham & Mathis 2000). One human observer, with a tally denominator and multiple digital timer, measured the behaviour of each individual during an 8-min pre-stimulus period. The number of lines crossed on a grid (~ 6.3 × 6.3 cm) was quantified, along with the time spent under shelter, and the time spent foraging (striking at the gravel). Following this pre-stimulus period, either 20 mL of PO or W was added via the injection hose, and the response variables were recorded again during an 8-min post-stimulus period. We also quantified whether freezing (yes/no) occurred (body pressed against a tank surface and the centre of the body not moving).

Observations were conducted blind and the order of the testing was randomized across treatments. Each day, following the end of the trials, minnows were moved to a new housing tank, and experimental tanks were cleaned for another round of trials. No fish were tested more than once. Sample sized were 20–32 per group. Statistical analysis Data from pre- and post-stimulus behaviours were computed into change in behaviour (post-pre) and used as response variables in our analysis. Data for lines crossed, however, varied greatly among fish, and we used a proportional change [(post-pre)/pre] to better standardize the response among individuals. We analyzed the three response variables (lines crossed, time under shelter, and time spent foraging) using a 3-way MANOVA. The fully-factorial 2×2×2 design included the prior-exposure treatment (safety or no safety), the conditioning treatment (alarm cue or conspecific learning), and the testing treatment (PO or W) as fixed factors. Two-way MANOVAs and t-tests with Bonferroni corrections were used as post-hoc analyses. Because MANOVA assumptions were not fully met, we used Pillai’s Trace because of its robustness to non-normality and covariance heterogeneity (Olson 1976) which occurred for some treatment groups for some response variables. For freezing behaviour (binary variable), we used a 4-way contingency table (G test) with the prior-exposure treatment, the conditioning treatment, and the testing treatment as factors, and the binary outcome (yes or no) as the response variable. This was followed by targeted post-hoc G tests with Bonferroni corrections. We used α=0.05 except where noted otherwise. Results Baseline activity during the pre-stimulus period did not differ among the treatments (all main-effect and interaction P-values > 0.3). When analysing behavioural change, a significant 3-

way interaction indicated that the effect had by the prior-exposure treatment on responses to testing stimuli was dependent on the type of conditioning (F3,190 = 3.95, P = 0.01; Fig. 1). When examining each type of conditioning separately, we found that alarm-cue learning was inhibited by the prior-exposure to PO (prior-exposure × testing cue: F3,94 = 11.54, α = 0.025, P < 0.001), whereas social learning was not (prior-exposure × testing cue: F3,94 = 0.36, α = 0.025, P = 0.79, Fig. 1). The pattern for freezing responses was different, however (Fig. 2). While there was an overall difference among the treatments (G4 = 25.26, P < 0.001), post-hoc tests revealed that freezing responses to PO were inhibited by prior exposure to PO in both the alarm-cue and social-cue conditioning treatments (prior PO and AC conditioning: G2 = 2.25, α = 0.013, P = 0.32; prior PO and SC conditioning: G2 = 1.36, α = 0.013, P = 0.51; prior water and AC conditioning: G2 = 12.59, α = 0.013, P = 0.002; prior water and SC conditioning: G2 = 10.46, α = 0.013, P = 0.005; Fig. 2). The overall learned responses from alarm and social cues for fish with no prior exposure to PO were similar, with both responding to the predator odour with the same intensity, as expected (prior water only, conditioning x testing cue: F3,96 = 0.27, α = 0.025, P = 0.85). While data for the proportional change in lines crossed were larger for alarm-cue conditioned minnows (AC vs. SC, responses to PO following prior water: t64 = 3.03, α = 0.013, P = 0.004; Fig. 1A), there were no statistical differences for the other responses (AC vs. SC, responses to PO following prior water:shelter time: t64 = 0.0, α = 0.013, P = 0.99, Fig.1B; AC vs. SC, responses to PO following prior water: foraging time: t64 = 0.56, α = 0.013, P = 0.58, Fig 1C; AC vs. SC, responses to PO following prior water: freezing: G2 = 2.25, α = 0.013, P = 0.32, Fig. 2).

Figure 1 Mean (± SE) proportional change in lines crossed (A), change in time spent under shelter (B), and change in time spent foraging (C) for minnows with prior exposure (6×) to

predator odour or blank water, conditioned with alarm cues or social cues, and then tested with

Prop. where freezing occurred

predator odour (dark bars) or blank water (white bars).

0.5 0.4 0.3 0.2 0.1 0.0 prior water

prior odour

alarm cue conditioned

prior water

prior odour

social conditioned

Figure 2 Proportion of trials where minnows performed freezing behaviour following prior exposure (6×) to predator odour or blank water, conditioning with alarm cues or social cues, and then testing with predator odour (dark bars) or blank water (white bars). Discussion As predicted, minnows in this study demonstrated learned safety. Learning from a single alarm-cue conditioning did not occur when preceded by repeated exposure to the novel odour in the absence of negative consequences (i.e., latent inhibition). This affirms that information about risk does not universally override learned safety. However, after interacting with models, minnows appeared to ignore their prior assessment of safety. This pattern was consistent for 3 of 4 response variables (Fig. 1), indicating that cues from live conspecific models are more persuasive than alarm cues in conflicting situations. Thus, the behaviour of live conspecifics appears to play a key role in managing uncertainty about predation risk for fathead minnows.

The fundamental differences between social and alarm-cue learning (e.g., presence of a conspecific, amount of information) prevent us from teasing apart the mechanisms behind our results. Nevertheless, we can conclude that learning about risk from live conspecifics was more persuasive than alarm cues. The alarm-cue learning in our study was considered a form of individual learning in the sense that it occurred in the absence of social companions, but individual learning via other mechanisms likely leads to different outcomes in the context of uncertainty. For instance, learning about predation risk by directly escaping a predator attack should provide prey with multiple cues and presumably more certainty about risk. We are not aware of any study on learning from direct experience with predators, but we expect this form of individual learning would be persuasive enough to override any prior expectations. Any weakening of responses through the chain of transmission (1× in this study) appeared negligible in our experiment. When minnows had no prior exposure to predator odour, their learned responses were similar (in terms of response intensity) following social and alarmcue conditioning. While alarm cue learning resulted in a significantly larger proportional change in lines crossed (~25%), the learned responses from alarm-cue and social conditioning were congruent overall (Fig. 1–2). We are not aware of any previous work that has compared social and alarm-cue learning, but several studies have explored the use of chemical versus visual stimuli in recognizing predation threat (Chivers et al. 2001; Hickman, Stone & Mathis 2004; Mathis & Vincent 2000). While the learned responses from alarm cues and social cues were similar in intensity for the minnows in our study, priority use of stimuli can vary across populations (Mathis, Chivers & Smith 1993). We caution that similar intensities of learned responses may not indicate similar levels of certainty. In an experiment with tadpoles (Lithobates sylvaticus), individuals that were conditioned one time to novel risk exhibited similar learned

responses (1-d post-conditioning) compared to individuals conditioned multiple times, but tadpoles conditioned multiple times retained their learned responses several days longer, indicating they had been more certain about the threat initially (Ferrari et al. 2012). In this case, it was reinforcement of learning that decreased uncertainty, but other factors like predator predictability and the reliability of cues from models are also thought to contribute to uncertainty (Crane & Ferrari 2013; Ferrari et al. 2010a). The size and reliability of models. Several factors likely affect the reliability of models (Crane & Ferrari 2013); for instance in some species, socially-learned responses are stronger when models are familiar or closely-related to the observers (Huebner et al. 1979; Kavaliers, Colwell & Choleris 2005). In this study, models were always larger than observers, indicating that they were older and had more experience with predators. Would learned responses from smaller models be weaker than those from larger models? Because minnows are unlikely to “outgrow” the gape of many predators, small models should be just as reliable as large individuals (i.e., they are vulnerable to the same predators). However, in other species, growth and life-stage changes presumably play an important role, but perhaps only in contexts where the cost of being wrong is relatively small (e.g., learning locations of food). In contrast, the high cost of being wrong about predation might dictate that social reliability is less important, and learned responses about risk will be conservative in a broad sense. Does conflicting information cause uncertainty for minnows? Mineka and Cook’s (1986) learning experiments with rhesus monkeys demonstrated that prior learning of safety was overridden by social conditioning, but only when the prior information (mock predators) was assessed alone. When the prior learning of safety was transmitted socially (mock predators + non-frightened models), the subsequent attempt to socially condition monkeys to risk was

unsuccessful. While this study is not directly comparable to ours due to the nature of alarm-cue learning, both suggest conflicting information from different modalities (e.g., individual-social) causes uncertainty. This hypothesis is supported by the inconsistent behaviour of minnows in our experiment. When prior assessment of safety was followed by cues from live conspecifics indicating risk, minnows exhibited reduced movement, increased shelter usage, and reduced foraging, but they did not increase freezing behaviour, the most costly of these responses (Magurran & Pitcher 1987; Smith 1992; von Frisch 1941). When a minnow freezes, it typically limits its field of view, and thus its vigilance, as an extreme last-ditch effort to escape predation. The inconsistent nature of these responses (reducing activity while maintaining vigilance by not freezing) may indicate that minnows recognized the possibility of danger but were uncertain about it. How conflicting information from different learning modalities relates to uncertainty, and whether social information overrides individual assessments in other contexts, deserves further attention. Acknowledgements This work was funded by the Natural Sciences and Engineering Research Council of Canada to MCOF. We thank the University of Saskatchewan, Department of Biology and the Department of Veterinary Biomedical Science for support. Katherine Raes made significant contributions to laboratory and animal maintenance. Literature Cited Acquistapace, P., Hazlett, B. A. & Gherardi, F. 2003. Unsuccessful predation and learning of predator cues by crayfish. Journal of Crustacean Biology, 23, 364-370. Arai, T., Tominaga, O., Seikai, T. & Masuda, R. 2007. Observational learning improves predator avoidance in hatchery-reared Japanese flounder Paralichthys olivaceus juveniles. Journal of Sea Research, 58, 59-64. Baptista, L. F. & Petrinovich, L. 1984. Social interaction, sensitive phases and the song template hypothesis in the white-crowned sparrow. Animal Behaviour, 32, 172-181.

Baptista, L. F. & Petrinovich, L. 1986. Song development in the white-crowned sparrow: social factors and sex differences. Animal Behaviour, 34, 1359-1371. Berejikian, B. A., Tezak, E. P. & LaRae, A. L. 2003. Innate and enhanced predator recognition in hatchery-reared chinook salmon. Environmental Biology of Fishes, 67, 241-251. Brown, G. E. & Chivers, D. P. 2005. Learning as an adaptive response to predation. In: Ecology of Predator-Prey Interactions (Ed. by P. Barbosa & I. Castellanos), pp. 34-54. Oxford: Oxford University Press. Brown, G. E. & Smith, R. J. F. 1996. Foraging trade-offs in fathead minnows (Pimephales promelas, Osteichthyes, Cyprinidae): Acquired predator recognition in the absence of an alarm response. Ethology, 102, 776-785. Chivers, D. P., Mirza, R. S., Bryer, P. J. & Kiesecker, J. M. 2001. Threat-sensitive predator avoidance by slimy sculpins: understanding the importance of visual versus chemical information. Canadian Journal of Zoology, 79, 867-873. Chivers, D. P. & Smith, R. J. F. 1994. The role of experience and chemical alarm signaling in predator recognition by fathead minnows, Pimephales promelas. Journal of Fish Biology, 44, 273-285. Chivers, D. P. & Smith, R. J. F. 1995. Chemical recognition of risky habitats is culturally transmitted among fathead minnows, Pimephales promelas (Osteichthyes, Cyprinidae). Ethology, 99, 286-296. Cook, M., Mineka, S., Wolkenstein, B. & Laitsch, K. 1985. Observational conditioning of snake fear in unrelated rhesus-monkeys. Journal of Abnormal Psychology, 94, 591-610. Crane, A. L. & Ferrari, M. C. O. 2013. Social learning of predation risk: a review and prospectus. In: Social learning theory: Phylogenetic considerations across animal, plant, and microbial taxa (Ed. by K. Clark), pp. 53-82. New York: Nova Science Publisher. Crane, A. L., Fritts, A. K., Mathis, A., Lisek, J. C. & Barnhart, M. C. 2011. Do gill parasites influence the foraging and antipredator behaviour of rainbow darters, Etheostoma caeruleum? Animal Behaviour, 82, 817-823. Curio, E. 1988. Cultural transmission of enemy recognition by birds. Curio, E., Ernst, U. & Vieth, W. 1978. The adaptive significance of avian mobbing. Zeitschrift für Tierpsychologie, 48, 184-202. Danchin, É., Giraldeau, L.-A., Valone, T. J. & Wagner, R. H. 2004. Public information: from nosy neighbors to cultural evolution. Science, 305, 487-491. Ferrari, M., Vrtělová, J., Brown, G. & Chivers, D. 2012. Understanding the role of uncertainty on learning and retention of predator information. Animal Cognition, 15, 807-813. Ferrari, M. C. O., Brown, G. E., Bortolotti, G. R. & Chivers, D. P. 2010a. Linking predator risk and uncertainty to adaptive forgetting: a theoretical framework and empirical test using tadpoles. Proceedings of the Royal Society B-Biological Sciences, 277, 2205-2210. Ferrari, M. C. O., Capitania-Kwok, T. & Chivers, D. P. 2006a. The role of learning in the acquisition of threat-sensitive responses to predator odours. Behavioral Ecology and Sociobiology, 60, 522-527. Ferrari, M. C. O. & Chivers, D. P. 2006a. Learning threat-sensitive predator avoidance: how do fathead minnows incorporate conflicting information? Animal Behaviour, 71, 19-26. Ferrari, M. C. O. & Chivers, D. P. 2006b. The role of latent inhibition in acquired predator recognition by fathead minnows. Canadian Journal of Zoology, 84, 505-509. Ferrari, M. C. O. & Chivers, D. P. 2011. Learning about non-predators and safe places: the forgotten elements of risk assessment. Animal Cognition, 14, 309-316.

Ferrari, M. C. O., Messier, F. & Chivers, D. P. 2006b. The nose knows: minnows determine predator proximity and density through detection of predator odours. Animal Behaviour, 72, 927932. Ferrari, M. C. O., Sih, A. & Chivers, D. P. 2009. The paradox of risk allocation: a review and prospectus. Animal Behaviour, 78, 579-585. Ferrari, M. C. O., Trowell, J. J., Brown, G. E. & Chivers, D. P. 2005. The role of learning in the development of threat-sensitive predator avoidance by fathead minnows. Animal Behaviour, 70, 777-784. Ferrari, M. C. O., Vavrek, M. A., Elvidge, C. K., Fridman, B., Chivers, D. P. & Brown, G. E. 2008. Sensory complementation and the acquisition of predator recognition by salmonid fishes. Behavioral Ecology and Sociobiology, 63, 113-121. Ferrari, M. C. O., Wisenden, B. D. & Chivers, D. P. 2010b. Chemical ecology of predatorprey interactions in aquatic ecosystems: a review and prospectus. Canadian Journal of Zoology, 88, 698 - 724. Galef, B. G. & Giraldeau, L. A. 2001. Social influences on foraging in vertebrates: causal mechanisms and adaptive functions. Animal Behaviour, 61, 3-15. Galef, B. G. & Whiskin, E. E. 2001. Interaction of social and individual learning in food preferences of Norway rats. Animal Behaviour, 62, 41-46. Griffin, A. S. 2004. Social learning about predators: A review and prospectus. Learning & Behavior, 32, 131-140. Griffin, A. S. & Boyce, H. M. 2009. Indian mynahs, Acridotheres tristis, learn about dangerous places by observing the fate of others. Animal Behaviour, 78, 79-84. Griffin, A. S. & Haythorpe, K. 2011. Learning from watching alarmed demonstrators: does the cause of alarm matter? Animal Behaviour, 81, 1163-1169. Hickman, C. R., Stone, M. D. & Mathis, A. 2004. Priority use of chemical over visual cues for detection of predators by graybelly salamanders, Eurycea multiplicata griseogaster. Herpetologica, 60, 203-210. Hoppitt, W. & Laland, K. N. 2013. Social Learning: An Introduction to Mechanisms, Methods, and Models: Princeton University Press. Huebner, D. K., Lentz, J. L., Wooley, M. J. & King, J. E. 1979. Responses to snakes by surrogate- and mother-reared squirrel monkeys. Bulletin of the Psychonomic Society, 14, 33-36. Johnston, C. E. & Johnson, D. L. 2000. Sound production in Pimephales notatus (Rafinesque)(Cyprinidae). Copeia, 2000. Kavaliers, M., Colwell, D. D. & Choleris, E. 2005. Kinship, familiarity and social status modulate social learning about ''micropredators'' (biting flies) in deer mice. Behavioral Ecology and Sociobiology, 58, 60-71. Kelley, J. L. & Magurran, A. E. 2003. Learned predator recognition and antipredator responses in fishes. Fish and Fisheries, 4, 216-226. Lima, S. L. & Bednekoff, P. A. 1999. Temporal variation in danger drives antipredator behavior: The predation risk allocation hypothesis. American Naturalist, 153, 649-659. Lima, S. L. & Dill, L. M. 1990. Behavioral decisions made under the risk of predation - a review and prospectus. Canadian Journal of Zoology, 68, 619-640. Lindeyer, C. M. & Reader, S. M. 2010. Social learning of escape routes in zebrafish and the stability of behavioural traditions. Animal Behaviour, 79, 827-834. Lubow, R. E. 1973. Latent inhibition. Psychological Bulletin, 79, 398.

Magurran, A. & Pitcher, T. 1987. Provenance, shoal size and the sociobiology of predatorevasion behaviour in minnow shoals. Proceedings of the Royal Society B-Biological Sciences, 229, 439-465. Manassa, R. & McCormick, M. 2013. Social learning improves survivorship at a life-history transition. Oecologia, 171, 845-852. Mathis, A., Chivers, D. P. & Smith, R. J. F. 1993. Population differences in responses of fathead minnows (Pimephales promelas) to visual and chemical stimuli from predators. Ethology, 93, 31-40. Mathis, A., Chivers, D. P. & Smith, R. J. F. 1996. Cultural transmission of predator recognition in fishes: Intraspecific and interspecific learning. Animal Behaviour, 51, 185-201. Mathis, A. & Smith, R. J. F. 1993. Fathead minnows, Pimephales-promelas, learn to recognize northern pike, Esox-lucius, as predators on the basis of chemical stimuli from minnows in the pike's diet. Animal Behaviour, 46, 645-656. Mathis, A. & Vincent, F. 2000. Differential use of visual and chemical cues in predator recognition and threat-sensitive predator-avoidance responses by larval newts (Notophthalmus viridescens). Canadian Journal of Zoology, 78, 1646-1652. Mineka, S. & Cook, M. 1986. Immunization against the observational conditioning of snake fear in rhesus monkeys. Journal of Abnormal Psychology, 95, 307-318. Mirza, R. S. & Chivers, D. P. 2000. Predator-recognition training enhances survival of brook trout: evidence from laboratory and field-enclosure studies. Canadian Journal of Zoology, 78, 2198-2208. Mitchell, M. D., McCormick, M. I., Ferrari, M. C. O. & Chivers, D. P. 2011. Friend or foe? The role of latent inhibition in predator and non-predator labelling by coral reef fishes. Animal Cognition, 14, 707-714. Nelson, D. A. 1998. External validity and experimental design: the sensitive phase for song learning. Animal Behaviour, 56, 487-491. Nocera, J. J. & Ratcliffe, L. M. 2010. Migrant and resident birds adjust antipredator behavior in response to social information accuracy. Behavioral Ecology, 21, 121-128. Olson, C. L. 1976. On choosing a test statistic in multivariate analysis of variance. Psychological Bulletin, 83, 579. Perreault, C., Moya, C. & Boyd, R. 2012. A Bayesian approach to the evolution of social learning. Evolution and Human Behavior, 33, 449-459. Petrinovich, L. & Baptista, L. F. 1987. Song development in the white-crowned sparrow: modification of learned song. Animal Behaviour, 35, 961-974. Rendell, L., Boyd, R., Cownden, D., Enquist, M., Eriksson, K., Feldman, M. W., Fogarty, L., Ghirlanda, S., Lillicrap, T. & Laland, K. N. 2010. Why copy others? Insights from the social learning strategies tournament. Science, 328, 208-213. Rieucau, G. & Giraldeau, L.-A. 2011. Exploring the costs and benefits of social information use: an appraisal of current experimental evidence. Philosophical Transactions of the Royal Society B: Biological Sciences, 366, 949-957. Seyfarth, R. M., Cheney, D. L., Bergman, T., Fischer, J., Zuberbühler, K. & Hammerschmidt, K. 2010. The central importance of information in studies of animal communication. Animal Behaviour, 80, 3-8. Shier, D. M. & Owings, D. H. 2007. Effects of social learning on predator training and postrelease survival in juvenile black-tailed prairie dogs, Cynomys ludovicianus. Animal Behaviour, 73, 567-577.

Sih, A. 1992. Prey uncertainty and the balancing of antipredator and feeding needs. American Naturalist, 139, 1052-1069. Smith, R. J. F. 1992. Alarm signals in fishes. Reviews in Fish Biology and Fisheries, 2, 33-63. Suboski, M. D., Bain, S., Carty, A. E., McQuoid, L. M., Seelen, M. I. & Seifert, M. 1990. Alarm reaction in acquisition and social transmission of simulated-predator recognition by zebra danio fish (Brachydanio rerio). Journal of Comparative Psychology, 104, 101-112. Vavrek, M. A., Elvidge, C. K., DeCaire, R., Belland, B., Jackson, C. D. & Brown, G. E. 2008. Disturbance cues in freshwater prey fishes: do juvenile convict cichlids and rainbow trout respond to ammonium as an 'early warning' signal? Chemoecology, 18, 255-261. Vieth, W., Curio, E. & Ernst, U. 1980. The adaptive significance of avian mobbing. III. Cultural transmission of enemy recognition in blackbirds: Cross-species tutoring and properties of learning. Animal Behaviour, 28, 1217-1229. von Frisch, K. 1941. Die bedeutung des geruchsinnes im leben der fische. Naturwissenschaften, 29, 321-333. Warren, M. L., Burr, B. M. & Tomelleri, J. R. 2014. Freshwater Fishes of North America: Volume 1: Petromyzontidae to Catostomidae: Johns Hopkins University Press. White, D. J. 2004. Influences of social learning on mate-choice decisions. Animal Learning & Behavior, 32, 105-113. Whitham, J. & Mathis, A. 2000. Effects of hunger and predation risk on foraging behavior of graybelly salamanders, Eurycea multiplicata. Journal of Chemical Ecology, 26, 1659-1665. Zentall, T. R. & Galef Jr, B. G. 2013. Social learning: psychological and biological perspectives: Psychology Press.