A key challenge in the life of most animals is to avoid being eaten. Via effects such as enhanced predator detection (Lima, 1995; Magurran et al., 1985), predator confusion (Landeau and Terborgh, 1986), and risk dilution effects (Foster and Treherne, 1981; Turner and Pitcher, 1986), individuals living and moving in groups can reduce their risk of predation (Ioannou et al., 2012; Krause and Ruxton, 2002; Pitcher, 1993; Ward and Webster, 2016). This helps explain why strong predation pressure is known to drive the formation of larger and more cohesive groups (Beauchamp, 2004; Krause and Ruxton, 2002; Seghers, 1974). However, the costs and benefits of grouping are not shared equally among individuals within groups, and besides differential food intake and costs of locomotion, group members themselves may experience widely varying risks of predation (Handegard et al., 2012; Krause, 1994; Krause and Ruxton, 2002). Where and whom predators attack within groups not only has major implications for the selection of individual phenotypes, and thereby the emergence of collective behaviour and the functioning of animal groups (Farine et al., 2015; Jolles et al., 2020; Ward and Webster, 2016), but also shapes the social behaviour of prey and the properties and structure of prey groups. Hence, a better understanding of the factors that influence predation risk within animal groups is of fundamental importance.

Previous work has identified many different, and sometimes contradictory, factors that predict prey vulnerability in groups. Some of the most long-standing theoretical work suggests that, when predators appear at random and attack the nearest prey, predation risk should be highest on the edge (‘marginal predation’) and front of mobile groups (Bumann et al., 1997; Hamilton, 1971; Morrell and Romey, 2008; Vine, 1971). If such predictions play out in the real world, individuals should try and surround themselves with others to reduce their domain of danger, known as the ‘selfish herd’ effect (Hamilton, 1971). There indeed exists empirical evidence of such behaviour in a range of species, with individuals moving closer together at the moment they perceive increased predation risk (Jakobsen and Johnsen, 1988; King et al., 2012; Krause, 1993; Sosna et al., 2019; Viscido and Wethey, 2002; but see Sankey et al., 2021). While there is also evidence from several studies that predation risk is higher towards the edge (e.g. Krause, 1993; Romenskyy et al., 2020; Romey et al., 2008) and front (e.g. Bumann et al., 1997; Ioannou et al., 2019) of animal groups, the empirical evidence for this is equivocal and seems to be system-dependent. For example, a number of studies report opposite patterns, with predation risk being highest in the group centre (e.g. Brunton, 1997; Hobson, 2011; Parrish, 1989) or towards the back of the group (e.g. Handegard et al., 2012; Krause et al., 2017). Experiments have shown that individuals in such positions actually have poorer access to salient social information, as well as visual information of what happens outside the group (Rosenthal et al., 2015), which could potentially help explain these findings.

Much of the focus in the literature on the spatial effects underlying predation risk has looked at centre-to-edge and front-to-back effects, partly because they are the easiest to measure, and largely concentrated only on a single or a few key potential features (but see e.g. Lambert et al., 2021; Romenskyy et al., 2020). However, it may be more likely that a spectrum of different factors shapes predation risk within groups. In particular, following Hamilton’s selfish herd theory (Hamilton, 1971), factors related to the spacing of individuals with respect to nearby neighbours have been shown to be of importance, with studies showing individuals with fewer neighbours or a larger domain of danger experiencing higher predation risk (De Vos and O’Riain, 2010; Lambert et al., 2021; Quinn and Cresswell, 2006; Romenskyy et al., 2020). But also individuals’ alignment to nearby neighbours (Ioannou et al., 2012), and features related to their visual information can be expected to play a role, two factors known to strongly influence how well individuals respond to others (Davidson et al., 2021; Rosenthal et al., 2015; Sosna et al., 2019; Strandburg-Peshkin et al., 2013).

The majority of previous research has also focused exclusively on prey behaviour. Like in Hamilton’s classic model (Hamilton, 1971), the predator is thereby treated as an abstract source of risk and any predator-related features as well the interactions among the predator and grouping prey typically are not explicitly evaluated. This is problematic as the behaviour and attack strategy of predators – such as to ambush, stealthily approach, or hunt groups of prey – may be some of the most influential factors that affect which prey are ultimately attacked (Hirsch and Morrell, 2011; Lima, 2002; Stankowich, 2003), and in turn are a strong selective force in the evolution of prey and predator traits (Lima, 2002). Furthermore, while considerable empirical work has investigated in detail how predators attack groups of prey, the differential likelihood to be attacked – rather than actual prey survival – is still often used to infer predation risk (Krause, 1994). This is because of difficulties in quantifying successful attacks in the wild (e.g. Handegard et al., 2012) and in the lab (e.g. Ioannou et al., 2019; Milinski, 1977; Parrish et al., 1989; Romenskyy et al., 2020). It therefore remains unclear what factors actually affect attack success and the probability for targeted prey to potentially survive attacks.

To advance our understanding of differential predation risk in animal groups, we need to systematically investigate the different absolute and relative spatial and visual features of both prey and predator while considering the real-time dynamics between live predators attacking groups of prey they can actually capture. This, however, poses a considerable challenge since it requires the simultaneous tracking of all members of a group of prey as well as the predator at a sufficiently high spatial and temporal scale. Here, we present experiments in which we achieve this. Specifically, to gain a detailed mechanistic understanding of when and where predators attack groups of prey and what predicts individual predation risk and survival, we observed Northern pike (Esox lucius), a geographically widespread and ecologically important predator (Craig, 2008), attacking free-swimming schools of 40 golden shiner fish (Notemigonus crysoleucas). We exposed pseudo-randomly composed groups (controlling for pike exposure) of juvenile shiners (8.5 cm, 95% CI: 6.4–11.5 cm) to individual pike (n=13; 22.4 cm ±1.1 cm) in a large open arena (1.05 m x 1.98 m; 6 cm depth), and used custom-developed tracking software to acquire detailed spatial and temporal data at 120 fps for a total of 125 attacks (see Materials and methods). By tracking both the predator and all prey individually over time, we were able to quantify each fish’s spatial position, relative spacing, orientation, and visual field, and analysed their movement kinematics in detail throughout each attack (Figure 1). We then quantified in detail how, when, and where the pike attacked. Subsequently, we used model fitting procedures to infer, both from the prey’s and predator’s perspective, the relative importance of a suite of potential features to predict shiners’ risk to be targeted (see Table 1). Finally, as pike were able to catch and consume their prey, we were able to investigate what factors best predicted the likelihood of attacks to be successful and thus for prey to survive a predator attack or to be eatenTable 1.

Figure 1

Download asset Open asset

Understanding when, where, and how predators attack animal groups, and the types of anti-predator benefits grouping animals may experience, has been of long-standing interest (reviewed in Krause and Ruxton, 2002; Pitcher, 1993; Ward and Webster, 2016). Although it is well appreciated that there is differential predation risk within animal groups, our knowledge has, nonetheless, remained largely centred on marginal predation and selfish herd effects. By employing high-resolution tracking of predators attacking large, dynamically moving groups of prey, here we provide detailed insights into where and when predators attack and reveal a suite of both prey and predator features that best predict the risk of schooling prey to be targeted and survive attacks. Our study shows that consideration of the multi-faceted factors underlying predation risk in combination with predators’ attack strategy and decision-making is important to better understand the broader costs and benefits of grouping, with implications for the evolution of social and collective behaviour.

Following a slow and steady approach of the school, pike tended to attack the groups head-on, in one fast burst of movement. However, rather than attacking individuals in the leading edge of the group, often assumed to be more risky than further back in mobile groups (e.g. Bumann et al., 1997; Ioannou et al., 2019; Krause et al., 2017; but see e.g. Handegard et al., 2012; Lambert et al., 2021), we found that pike often slowly entered the school and then launched their rapid attack towards the group centre. That predators such as pike can get very close to their prey has been described previously (Coble, 1973; Morris et al., 1956; Krause et al., 1998; Nursall, 1973; Webb and Skadsen, 1980) and could potentially be explained by their narrow frontal profile (Webb, 1982) as it makes it very hard for prey to detect movement changes, especially when attacked head-on. This may also explain the suggestion of previous work that pike do not suffer much from the confusion effect (Turesson and Bronmark, 2004). That predators may actually enter groups and strike at central individuals is not often considered (Hirsch and Morrell, 2011), possibly because it contrasts with the long-standing idea that predation risk is higher on the edge of animal groups (Duffield and Ioannou, 2017; Krause, 1994; Krause and Ruxton, 2002; Stankowich, 2003). However, our finding is in line with the predictions of theoretical work that suggest that the extent of marginal predation may depend on attack strategy and declines with the distance from which the predator attacks (Hirsch and Morrell, 2011). Furthermore, increased risk of individuals near the centre of groups may be more widespread than currently thought. Predators not only exhibit stealthy behavioural tactics that enable them to approach and attack central individuals, as we show here, but may also do so by attacking groups from above (Brunton, 1997) or below (Clua and Grosvalet, 2001; Hobson, 2011; but see Romey et al., 2008), and by rushing into the main body of the group (Handegard et al., 2012; Hobson, 2011; Parrish, 1989).

One of the assumptions of Hamilton’s classical model is that the predator appears at random and strikes the nearest prey (Hamilton, 1971). This helps explain why, as shown for other systems, that individuals on the edge and front of groups experience higher predation risk (Bumann et al., 1997; Ioannou et al., 2019; Krause, 1993; Romenskyy et al., 2020; Romey et al., 2008). But although here we also show that pike tended to attack the nearest prey (they were physically capable of attacking), this was importantly not the first prey they encountered. Namely, pike tended to delay their attack until they were very close to and often even surrounded by their prey. Besides potentially having more prey to choose from, pike clearly benefited from this behaviour by being able to generate acceleration and speeding forces considerably higher than that of the prey they targeted (see also Domenici and Blake, 1997). This is supported by the finding that attacks were less likely to be successful when the target was further away and thereby able to achieve a high acceleration response in the moments before the strike. That pike very slowly and carefully approached the grouping prey may thus be to minimize their strike distance, position and angle, until the point where they have to decide to either launch the attack or call off it off based on their perceived likelihood for success (Nilsson and Eklöv, 2008). This type of predator decision-making when to attack, particularly relevant in the context of ambush predators, is rarely considered (Hirsch and Morrell, 2011), but fundamental for a proper understanding of the distribution of predation risk in animal groups across different predator-prey systems.

While many studies have investigated differential predation in animal groups, few have systematically compared and considered the many types of features that can be hypothesized to influence predation risk (but see e.g. Lambert et al., 2021; Romenskyy et al., 2020). Here, using a multi-model inference approach, we found that, in addition to centre-to-edge position, also limited domain of danger and local misalignment were key predictive features of predation risk from the prey’s perspective. In line with predictions from Hamilton’s selfish herd model (Hamilton, 1971), we found that fish with a larger (limited) domain of danger were at higher risk of being targeted. Although some previous studies have shown individual predation risk to be related to their domain of danger (De Vos and O’Riain, 2010; Lambert et al., 2021; Romenskyy et al., 2020), our results highlight the importance of variation in individuals’ domain of danger within groups. This is further reflected by the finding of considerable unexplained variance between fish’s centre-to-edge position and their limited domain of danger. In other words, despite central individuals being more at risk, it still pays for prey to be close to and surrounded by others within the group to reduce their risk of predation. Such spatial heterogeneity within groups and its effects on differential predation is rarely considered (Jolles et al., 2020). Our finding that the local alignment of prey was also a strong predictor of predation risk, with less aligned individuals having a higher chance of being targeted, is in line with experimental work using virtual prey (Ioannou et al., 2012). Less aligned individuals may be predated more simply by those individuals standing out from their surrounding group mates (i.e. the oddity effect; Landeau and Terborgh, 1986) and/or because predators may have evolved to target misaligned individuals because of being less capable of obtaining salient social information (Couzin and Krause, 2003). Together, these findings bring some important nuance to the ‘selfish herd’ phenomenon: rather than moving towards the group centre, individuals should try and position themselves near others, also inside the group, and even when it is of small or medium size, and make sure they do not attract attention based on their orientation.

By rerunning multi-model inference with a predator-focused approach, that is by also including features about the predator and only considering prey within its perceived attack region, we found prey’s distance and angle to the pike were the strongest determinants for being targeted (see also Romenskyy et al., 2020). This shows that, besides moving towards others, as discussed above, it pays for prey to move away and avoid the cone of risk directly in front of the predator. By repeatedly testing the shiners with the pike, we saw that shiners indeed increasingly distanced themselves from the predator and especially avoided the region in front of the pike. This is consistent with previous work that suggests that, in addition to unlearned predispositions, experience with predators is important (Kelley and Magurran, 2003). While much previous work has only focused on predation risk from the prey’s perspective, these findings highlight that, for a proper understanding of predation risk in animal groups it is important to not remove the predator from the equation (Hirsch and Morrell, 2011; Lima, 2002).

Vision is known to be the primary modality for pike’s predatory behaviour (Nilsson and Eklöv, 2008) and for mediating social interactions among species of schooling fish (Kotrschal et al., 1998; Rosenthal et al., 2015; Strandburg-Peshkin et al., 2013). We however found that the extent that a shiner’s vision was occupied by conspecifics, the extent it could see the predator, and the extent the predator could see the shiner were only weak predictors of individual predation risk. This does not mean they are not relevant per se, as they were closely linked to the main predictors, but, being more specific, could potentially explain less of the observed variance overall. Hence, future experiments specifically manipulating the conditions that impact the vision of predator and prey would be valuable in further disentangling the role of visual information in predator-prey interactions. Furthermore, while we unravelled key features that are strongly predictive of which individuals were most likely to be targeted for an attack, further future work, including experimental manipulations and/or directed acyclic graphs (DAGs) (Laubach et al., 2021; McElreath, 2020), is also needed to properly disentangle cause and effect in the predation risk of grouping prey.

Overall, the pike were very successful, with 70% of attacks resulting in the targeted shiner being eaten. This is comparable to previous studies with sit-and-wait predators (Krause et al., 1998; Neill and Cullen, 1974; Turesson and Bronmark, 2004; Webb and Skadsen, 1980). But where previous work was not able to quantify differential predation in terms of mortality risk, such as by using confined or virtual prey (Ioannou et al., 2019; Milinski, 1977), difficulties of field conditions (Handegard et al., 2012), or by the predator simply failing to ever attack successfully (Parrish, 1989; Romenskyy et al., 2020), here we were able to also investigate the factors that may affect the likelihood for an individual to survive an attack. We found that both predator distance and prey acceleration in the moments until the attack were highly predictive of predation success, with targeted prey that were further away, and that showed a faster acceleration response, being more likely to evade capture. These two features have also been found as significant variables affecting survival in a study on predator-prey dynamics with single prey (Walker et al., 2005; see also Lucas et al., 2021). Together, this suggests that targeted prey could sometimes anticipate the strike, highlighting that, despite the very high speeds pike attained, prey response does matter in predator-prey dynamics, and that evasive prey behaviour may be especially successful when prey are further away (see also Ranåker et al., 2012).

Although other ambush predators may employ different attack strategies, and even among pike there may be variation in attack behaviour, the principles we highlight related to the attack sequence and when, where, and how predators attack are relevant for other systems. One thing however that is not clear yet is how the observed effects play out in much larger prey groups. In particular, while group size is not expected to effect much whether ambush predators are likely to attack internal individuals, the specific risk of central individuals could both be hypothesized to decrease with group size, such as if the predator is more likely to attack when surrounded by prey, or to not be affected by it, such as if the predator actively targets central individuals. Whatever the process, the observed findings are likely for prey that move in groups of somewhat intermediate size; for very large groups, such as the huge schools encountered in the pelagic, ambush predators may simply not be able to attack the group centre due to spatial constraints. More generally, the tendency for predators to attack the centre of moving groups may depend on the medium in which the predator-prey interactions occur. As in the air there is potential for (fatal) collisions, and on land it is physically difficult for predators to enter groups and predators’ size advantage tends to be more limited, predators may be less likely to go for the group centre as compared to in aquatic or mixed (e.g. aerial predator hunting aquatic prey) systems. Hence, the important interplay we highlight between predator attack strategy and prey response may have different implications across different predator prey systems and warrants concerted further research effort.

Predation is seen as one of the main factors to shape the collective properties of animal groups (Herbert-Read et al., 2017) and has been shown to drive the formation of larger, more cohesive groups that exhibit collective, coordinated motion (see e.g. Beauchamp, 2004; Ioannou et al., 2012; Seghers, 1974). Our finding that central individuals are more at risk of being predated could actually have the opposite effect, with schooling having a selective disadvantage, which over time could result in weaker collective behaviour and less cohesive schools. However, we do not deem this likely as selection is likely to be group-size dependent, as discussed above. Furthermore, our multi-model inference approach revealed that, despite more central individuals experiencing higher predation risk, being close to others inside the school was still associated with a lower risk of being targeted. As most prey experience many types of predators, including sit-and-wait predators and active predators that hunt for prey, the extent and direction of such selection effects will depend on the broader predation landscape in which prey find themselves. While the finding that pike were more likely to attack the main school may also appear to indicate a selective disadvantage to school, calculating the per-capita-risk for each individual would actually reveal it is still safest to be part of the main school. Nevertheless, as the shiners in our study rarely exhibited fission-fusion dynamics we feel our dataset is not appropriate to make proper inferences about how predation risk is linked to group size.

Laboratory studies on predator-prey dynamics like ours do, of course, have their limitations. Although the size of the arena we used (~2m²) is in line with behavioural studies with large schools of fish (e.g. Sosna et al., 2019; Strandburg-Peshkin et al., 2013) and experiments with live predators attacking schooling prey (Bumann et al., 1997; Magurran and Pitcher, 1987; Neill and Cullen, 1974; Romenskyy et al., 2020; Theodorakis, 1989), compared to conditions in the wild the prey and predator had limited space to move. However, as pike are ambush predators they tend to move relatively little to search for prey and rather rely on prey movement for encounters (Nilsson and Eklöv, 2008). Increasing tank size would have made effective tracking extremely difficult, or impossible, and while a much larger tank is expected to considerably increase latency to attack, we expect it to have relatively little effect on the observed findings. This was primarily done to be able to keep track of individual identities and compute features related to the visual field of the fish. Shiners naturally school in very shallow water conditions as well as near the surface in deeper water in the wild (Hall et al., 1979; Krause et al., 2000b; Stone et al., 2016) and also pike primarily occur in the shallow littoral zone, sometimes only a few of tens of cm deep (Pierce et al., 2013; Skov et al., 2018). Furthermore, pilot experiment showed the pike did exhibit normal swimming and attack behaviour with attack speeds and acceleration comparable to previous work (Domenici and Blake, 1997; Walker et al., 2005). Recent other work on predator-prey dynamics did not find a considerable impact of adding the third dimension to their analyses (Romenskyy et al., 2020). Still, the water depth used is a limiting factor of our study and in the future this type of work should be extended to deeper water while still keeping track of individual identities over time. We expect that adding the third dimension would not change the stealthy attack behaviour of the pike and therefore still put more central individuals most at risk, but possibly attack success would be reduced because of increased predator visibility and prey escape potential in the vertical plane, which remains to be tested.

For our experiments, we used a testing arena without any internal structures such as refuges. This was a strategic decision as providing a more complex environment would have impacted the ability of the shiners to school in large groups and would have led fish to hide under cover. Although studying predator-prey dynamics in more complex environments would be interesting in its own regard, it would not have allowed us to study the questions we are interested in about the predation risk of free-schooling prey. Furthermore, pilot experiments indicated that the pike never used refuges (consistent with previous work, see Turesson and Bronmark, 2004), so they were not further provided during the actual experiment. Experiments were conducted under artificial lights with reduces intensity but at a level high enough to be able to acquire accurate videos of the trials without motion blur. In the wild, pike are however generally found to be most active during dusk and dawn (Raat, 1988; Skov et al., 2018) and to consume most prey at low light intensity (Dobler, 1977). We expect that if a lower light intensity was used, the pike may profit from visual superiority and thereby would have increased predation success, further aided by a likely loosening of the prey schools due to limited light being available (Dobler, 1977). While it is now increasingly possible to obtain detailed data from predation events on grouping prey in the field (see e.g. Handegard et al., 2012; Krause et al., 2017), even with the most sophisticated field-based imaging it would not have been possible to acquire the highly detailed data we obtained here. That is, individual-level characteristics of predator and all grouping prey throughout predator attacks at high spatial and temporal resolution, linked to attack success and survival. Future work is now needed that further considers the different relevant ecological factors, for example deeper water, more heterogeneous environments with vegetation, different light levels, and how they interplay with the observed effects of differential predation risk in schools of fish.

In conclusion, using a quantitative empirical approach in which we acquired highly detailed individual-based characteristics of predators attacking grouping prey, we provide key mechanistic insights into when and where predators attack coordinated mobile groups and what predicts individual predation risk and survival. We demonstrate that ambush predators such as pike can stealthily approach groups and delay their attack until being very close to, and often even until being surrounded by their prey. We also show that, rather than just go for the group centre, it pays for animals to position themselves near others, align with their body orientation, and avoid being positioned close to, and in front of, the predator. Even central individuals, when targeted, have a chance to escape an attack by avoiding the predator’s head and strongly accelerating in response to the attack. Our study provides key insights about differential predation risk in groups of prey and highlights a fundamental role for both predator attack strategy and decision-making and prey behaviour. This may have important repercussions for the costs and benefits of grouping and thereby the distribution of (social) phenotypes in populations, and the collective properties of animal groups (Farine et al., 2015; Herbert-Read et al., 2017; Jolles et al., 2020). It is therefore paramount for future work to consider the multi-faceted features of both predator and prey and the role of the predator in the broader predation landscape to properly understand its role in the evolution of animal grouping.

Associated datasets are available on Mendeley Data (https://doi.org/10.17632/bszk9ztryp.1).

1. 1. Jolle J
   2. Matt S

   (2021) Mendeley Data

   Data for: Both Prey and Predator Features Determine Predation Risk and Survival of Schooling Prey.

   https://doi.org/10.17632/bszk9ztryp.1

1. 1. ASAB/ABS

   (2012) Guidelines for the treatment of animals in behavioural research and teaching

   Animal Behaviour 83:301–309.

   https://doi.org/10.1016/j.anbehav.2011.10.031

   • Google Scholar
2. 1. Ballerini M
   2. Cabibbo N
   3. Candelier R
   4. Cavagna A
   5. Cisbani E
   6. Giardina I
   7. Lecomte V
   8. Orlandi A
   9. Parisi G
   10. Procaccini A
   11. Viale M
   12. Zdravkovic V

   (2008) Interaction ruling animal collective behavior depends on topological rather than metric distance: evidence from a field study

   PNAS 105:1232–1237.

   https://doi.org/10.1073/pnas.0711437105

   • PubMed
   • Google Scholar
3. 1. Beauchamp G

   (2004) Reduced flocking by birds on islands with relaxed predation

   Proceedings. Biological Sciences 271:1039–1042.

   https://doi.org/10.1098/rspb.2004.2703

   • PubMed
   • Google Scholar
4. 1. Brunton DH

   (1997) Impacts of predators: Center nests are less successful than edge nests in a large nesting colony of least terns

   The Condor 99:372–380.

   https://doi.org/10.2307/1369943

   • Google Scholar
5. 1. Bumann D
   2. Rubenstein D
   3. Krause J

   (1997) Mortality risk of spatial positions in animal groups: The danger of being in the front

   Behaviour 134:1063–1076.

   https://doi.org/10.1163/156853997X00403

   • Google Scholar
6. Book

   1. Burnham KP
   2. Anderson D.

   (2002)

   Model selection and multimodel inference: A practical information-theoretic approach

   New York: Springer.

   • Google Scholar
7. 1. Burnham KP
   2. Anderson DR
   3. Huyvaert KP

   (2011) AIC model selection and multimodel inference in behavioral ecology: some background, observations, and comparisons

   Behavioral Ecology and Sociobiology 65:23–35.

   https://doi.org/10.1007/s00265-010-1029-6

   • Google Scholar
8. 1. Clua É
   2. Grosvalet F

   (2001) Mixed-species feeding aggregation of dolphins, large tunas and seabirds in the Azores

   Aquat Living Resour 14:11–18.

   https://doi.org/10.1016/S0990-7440(00)01097-4

   • Google Scholar
9. 1. Coble DW

   (1973) Influence of appearance of prey and satiation of predator on food selection by northern pike (esox lucius)

   Journal of the Fisheries Research Board of Canada 30:317–320.

   https://doi.org/10.1139/f73-057

   • Google Scholar
10. 1. Couzin ID
    2. Krause J

    (2003) Self-organization and collective behavior in vertebrates

    Advances in the Study of Behavior 32:1–75.

    https://doi.org/10.1016/S0065-3454(03)01001-5

    • Google Scholar
11. 1. Craig JF

    (2008) A short review of pike ecology

    Hydrobiologia 601:5–16.

    https://doi.org/10.1007/s10750-007-9262-3

    • Google Scholar
12. 1. Davidson JD
    2. Sosna MMG
    3. Twomey CR
    4. Sridhar VH
    5. Leblanc SP
    6. Couzin ID

    (2021) Collective detection based on visual information in animal groups

    Journal of the Royal Society, Interface 18:20210142.

    https://doi.org/10.1098/rsif.2021.0142

    • PubMed
    • Google Scholar
13. 1. De Vos A
    2. O’Riain MJ

    (2010) Sharks shape the geometry of a selfish seal herd: experimental evidence from seal decoys

    Biology Letters 6:48–50.

    https://doi.org/10.1098/rsbl.2009.0628

    • PubMed
    • Google Scholar
14. 1. Dobler E

    (1977) Correlation between the feeding time of the pike (Esox lucius) and the dispersion of a school of Leucaspius delineatus

    Oecologia 27:93–96.

    https://doi.org/10.1007/BF00345687

    • PubMed
    • Google Scholar
15. 1. Domenici P
    2. Blake RW

    (1997) The kinematics and performance of fish fast-start swimming

    The Journal of Experimental Biology 200:1165–1178.

    https://doi.org/10.1242/jeb.200.8.1165

    • PubMed
    • Google Scholar
16. 1. Duffield C
    2. Ioannou CC

    (2017) Marginal predation: do encounter or confusion effects explain the targeting of prey group edges?

    Behavioral Ecology 28:1283–1292.

    https://doi.org/10.1093/beheco/arx090

    • PubMed
    • Google Scholar
17. 1. Farine DR
    2. Montiglio P-O
    3. Spiegel O

    (2015) From individuals to groups and back: The evolutionary implications of group phenotypic composition

    Trends in Ecology & Evolution 30:609–621.

    https://doi.org/10.1016/j.tree.2015.07.005

    • PubMed
    • Google Scholar
18. 1. Foster WA
    2. Treherne JE

    (1981) Evidence for the dilution effect in the selfish herd from fish predation on a marine insect

    Nature 293:466–467.

    https://doi.org/10.1038/293466a0

    • Google Scholar
19. 1. Grueber CE
    2. Nakagawa S
    3. Laws RJ
    4. Jamieson IG

    (2011) Multimodel inference in ecology and evolution: challenges and solutions

    Journal of Evolutionary Biology 24:699–711.

    https://doi.org/10.1111/j.1420-9101.2010.02210.x

    • PubMed
    • Google Scholar
20. 1. Hall DJ
    2. Werner EE
    3. Gilliam JF
    4. Mittelbach GG
    5. Howard D
    6. Doner CG
    7. Dickerman JA
    8. Stewart AJ

    (1979) Diel foraging behavior and prey selection in the golden shiner (notemigonus crysoleucas)

    Journal of the Fisheries Research Board of Canada 36:1029–1039.

    https://doi.org/10.1139/f79-145

    • Google Scholar
21. 1. Hamilton WD

    (1971) Geometry for the selfish herd

    Journal of Theoretical Biology 31:295–311.

    https://doi.org/10.1016/0022-5193(71)90189-5

    • PubMed
    • Google Scholar
22. 1. Handegard NO
    2. Boswell KM
    3. Ioannou CC
    4. Leblanc SP
    5. Tjøstheim DB
    6. Couzin ID

    (2012) The dynamics of coordinated group hunting and collective information transfer among schooling prey

    Current Biology 22:1213–1217.

    https://doi.org/10.1016/j.cub.2012.04.050

    • PubMed
    • Google Scholar
23. 1. Harrison XA
    2. Donaldson L
    3. Correa-Cano ME
    4. Evans J
    5. Fisher DN
    6. Goodwin CED
    7. Robinson BS
    8. Hodgson DJ
    9. Inger R

    (2018) A brief introduction to mixed effects modelling and multi-model inference in ecology

    PeerJ 6:e4794.

    https://doi.org/10.7717/peerj.4794

    • PubMed
    • Google Scholar
24. 1. Hein AM
    2. Gil MA
    3. Twomey CR
    4. Couzin ID
    5. Levin SA

    (2018) Conserved behavioral circuits govern high-speed decision-making in wild fish shoals

    PNAS 115:12224–12228.

    https://doi.org/10.1073/pnas.1809140115

    • PubMed
    • Google Scholar
25. 1. Herbert-Read JE
    2. Rosén E
    3. Szorkovszky A
    4. Ioannou CC
    5. Rogell B
    6. Perna A
    7. Ramnarine IW
    8. Kotrschal A
    9. Kolm N
    10. Krause J
    11. Sumpter DJT

    (2017) How predation shapes the social interaction rules of shoaling fish

    Proceedings. Biological Sciences 284:20171126.

    https://doi.org/10.1098/rspb.2017.1126

    • PubMed
    • Google Scholar
26. 1. Hirsch BT
    2. Morrell LJ

    (2011) Measuring marginal predation in animal groups

    Behavioral Ecology 22:648–656.

    https://doi.org/10.1093/beheco/arr026

    • Google Scholar
27. 1. Hobson ES

    (2011) Selective feeding by the gafftopsail pompano, trachinotus rhodopus (gill), in mixed schools of herring and anchovies in the gulf of california

    Copeia 1963:595.

    https://doi.org/10.2307/1441506

    • Google Scholar
28. 1. Huntingford FA

    (1984) Some ethical issues raised by studies of predation and aggression

    Animal Behaviour 32:210–215.

    https://doi.org/10.1016/S0003-3472(84)80339-5

    • Google Scholar
29. 1. Ioannou CC
    2. Guttal V
    3. Couzin ID

    (2012) Predatory fish select for coordinated collective motion in virtual prey

    Science 337:1212–1215.

    https://doi.org/10.1126/science.1218919

    • PubMed
    • Google Scholar
30. 1. Ioannou CC
    2. Rocque F
    3. Herbert-Read JE
    4. Duffield C
    5. Firth JA

    (2019) Predators attacking virtual prey reveal the costs and benefits of leadership

    PNAS 116:8925–8930.

    https://doi.org/10.1073/pnas.1816323116

    • PubMed
    • Google Scholar
31. 1. Jakobsen PJ
    2. Johnsen GH

    (1988) Size-specific protection against predation by fish in swarming waterfleas, Bosmina longispina

    Animal Behaviour 36:986–990.

    https://doi.org/10.1016/S0003-3472(88)80057-5

    • Google Scholar
32. 1. Johannes MRS
    2. McQueen DJ
    3. Stewart TJ
    4. Post JR

    (1989) Golden shiner (notemigonus crysoleucas) population abundance correlations with food and predators

    Canadian Journal of Fisheries and Aquatic Sciences 46:810–817.

    https://doi.org/10.1139/f89-101

    • Google Scholar
33. 1. Johnson JB
    2. Omland KS

    (2004) Model selection in ecology and evolution

    Trends in Ecology & Evolution 19:101–108.

    https://doi.org/10.1016/j.tree.2003.10.013

    • PubMed
    • Google Scholar
34. 1. Jolles JW
    2. King AJ
    3. Killen SS

    (2020) The role of individual heterogeneity in collective animal behaviour

    Trends in Ecology & Evolution 35:278–291.

    https://doi.org/10.1016/j.tree.2019.11.001

    • PubMed
    • Google Scholar
35. 1. Kelley JL
    2. Magurran AE

    (2003) Learned predator recognition and antipredator responses in fishes

    Fish and Fisheries 4:216–226.

    https://doi.org/10.1046/j.1467-2979.2003.00126.x

    • Google Scholar
36. 1. King AJ
    2. Wilson AM
    3. Wilshin SD
    4. Lowe J
    5. Haddadi H
    6. Hailes S
    7. Morton AJ

    (2012) Selfish-herd behaviour of sheep under threat

    Current Biology 22:R561–R562.

    https://doi.org/10.1016/j.cub.2012.05.008

    • PubMed
    • Google Scholar
37. 1. Kotrschal K
    2. Van Staaden MJ
    3. Huber R

    (1998) Fish brains: Evolution and environmental relationships

    Reviews in Fish Biology and Fisheries 8:373–408.

    https://doi.org/10.1023/A:1008839605380

    • Google Scholar
38. 1. Krause J

    (1993) The effect of “Schreckstoff” on the shoaling behaviour of the minnow: A test of Hamilton’s selfish herd theory

    Animal Behaviour 45:1019–1024.

    https://doi.org/10.1006/anbe.1993.1119

    • Google Scholar
39. 1. Krause J

    (1994) Differential fitness returns in relation to spatial position in groups

    Biological Reviews of the Cambridge Philosophical Society 69:187–206.

    https://doi.org/10.1111/j.1469-185x.1994.tb01505.x

    • PubMed
    • Google Scholar
40. 1. Krause J
    2. Ruxton GD
    3. Rubenstein D

    (1998) Is there always an influence of shoal size on predator hunting success?

    Journal of Fish Biology 52:494–501.

    https://doi.org/10.1111/j.1095-8649.1998.tb02012.x

    • Google Scholar
41. 1. Krause J
    2. Butlin RK
    3. Peuhkuri N
    4. Pritchard VL

    (2000a) The social organization of fish shoals: a test of the predictive power of laboratory experiments for the field

    Biological Reviews 75:477–501.

    https://doi.org/10.1111/j.1469-185X.2000.tb00052.x

    • Google Scholar
42. 1. Krause J
    2. Hoare DJ
    3. Croft DP
    4. Lawrence J
    5. Ward AJW
    6. Ruxton GD
    7. Godin JG
    8. James R

    (2000b) Fish shoal composition: mechanisms and constraints

    Proceedings. Biological Sciences 267:2011–2017.

    https://doi.org/10.1098/rspb.2000.1243

    • PubMed
    • Google Scholar
43. Book

    1. Krause J
    2. Ruxton GD.

    (2002)

    Living in groups

    Oxford: Oxford University Press.

    • Google Scholar
44. 1. Krause J
    2. Herbert-Read JE
    3. Seebacher F
    4. Domenici P
    5. Wilson ADM
    6. Marras S
    7. Svendsen MBS
    8. Strömbom D
    9. Steffensen JF
    10. Krause S
    11. Viblanc PE
    12. Couillaud P
    13. Bach P
    14. Sabarros PS
    15. Zaslansky P
    16. Kurvers RHJM

    (2017) Injury-mediated decrease in locomotor performance increases predation risk in schooling fish

    Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 372:20160232.

    https://doi.org/10.1098/rstb.2016.0232

    • PubMed
    • Google Scholar
45. 1. Lambert PJ
    2. Herbert-Read JE
    3. Ioannou CC

    (2021) The measure of spatial position within groups that best predicts predation risk depends on group movement

    Proceedings. Biological Sciences 288:20211286.

    https://doi.org/10.1098/rspb.2021.1286

    • PubMed
    • Google Scholar
46. 1. Landeau L
    2. Terborgh J

    (1986) Oddity and the ‘confusion effect’ in predation

    Animal Behaviour 34:1372–1380.

    https://doi.org/10.1016/S0003-3472(86)80208-1

    • Google Scholar
47. 1. Laubach ZM
    2. Murray EJ
    3. Hoke KL
    4. Safran RJ
    5. Perng W

    (2021) A biologist’s guide to model selection and causal inference

    Proceedings. Biological Sciences 288:20202815.

    https://doi.org/10.1098/rspb.2020.2815

    • PubMed
    • Google Scholar
48. 1. Lima SL

    (1995) Back to the basics of anti-predatory vigilance: the group-size effect

    Animal Behaviour 49:11–20.

    https://doi.org/10.1016/0003-3472(95)80149-9

    • Google Scholar
49. 1. Lima SL

    (2002) Putting predators back into behavioral predator–prey interactions

    Trends in Ecology & Evolution 17:70–75.

    https://doi.org/10.1016/S0169-5347(01)02393-X

    • Google Scholar
50. 1. Lucas J
    2. Ros A
    3. Gugele S
    4. Dunst J
    5. Geist J
    6. Brinker A

    (2021) The hunter and the hunted-A 3D analysis of predator-prey interactions between three-spined sticklebacks (Gasterosteus aculeatus) and larvae of different prey fishes

    PLOS ONE 16:e0256427.

    https://doi.org/10.1371/journal.pone.0256427

    • PubMed
    • Google Scholar
51. 1. Magurran AE
    2. Oulton WJ
    3. Pitcher TJ

    (1985) Vigilant behaviour and shoal size in minnows

    Zeitschrift Für Tierpsychologie 67:167–178.

    https://doi.org/10.1111/j.1439-0310.1985.tb01386.x

    • Google Scholar
52. 1. Magurran AE
    2. Pitcher TJ

    (1987) Provenance, shoal size and the sociobiology of predator-evasion behaviour in minnow shoals

    Proceedings of the Royal Society of London. Series B. Biological Sciences 229:439–465.

    https://doi.org/10.1098/rspb.1987.0004

    • Google Scholar
53. Book

    1. McElreath R

    (2020) Statistical Rethinking

    Chapman and Hall/CRC.

    https://doi.org/10.1201/9780429029608

    • Google Scholar
54. 1. Milinski M

    (1977) Experiments on the Selection by Predators against spatial Oddity of their Prey1

    Zeitschrift Für Tierpsychologie 43:311–325.

    https://doi.org/10.1111/j.1439-0310.1977.tb00078.x

    • Google Scholar
55. 1. Morrell LJ
    2. Romey WL

    (2008) Optimal individual positions within animal groups

    Behavioral Ecology 19:909–919.

    https://doi.org/10.1093/beheco/arn050

    • Google Scholar
56. 1. Morris D
    2. Tinbergen N
    3. Hoogland R

    (1956) The spines of sticklebacks (gasterosteus and pygosteus) as means of defence against predators (perca and esox)

    Behaviour 10:205–236.

    https://doi.org/10.1163/156853956X00156

    • Google Scholar
57. 1. Neill SRJ
    2. Cullen JM

    (1974) Experiments on whether schooling by their prey affects the hunting behaviour of cephalopods and fish predators

    Journal of Zoology 172:549–569.

    https://doi.org/10.1111/j.1469-7998.1974.tb04385.x

    • Google Scholar
58. Book

    1. Nilsson PA
    2. Eklöv P.

    (2008)

    Finding Food and Staying AliveBiology and Ecology of Pike

    Taylor & Francis Group.

    • Google Scholar
59. 1. Nursall JR

    (1973) Some behavioral interactions of spottail shiners (Notropis hudsonius), yellow perch (Perca flavescens), and northern pike (Esox lucius)

    J Fish Res Board Canada 30:1161–1178.

    https://doi.org/10.1139/f73-187

    • Google Scholar
60. 1. Parrish JK

    (1989) Re-examining the selfish herd: are central fish safer?

    Animal Behaviour 38:1048–1053.

    https://doi.org/10.1016/S0003-3472(89)80143-5

    • Google Scholar
61. 1. Parrish JK
    2. Strand SW
    3. Lott JL

    (1989) Predation on a school of flat-iron herring, harengula thrissina

    Copeia 1989:1089.

    https://doi.org/10.2307/1446009

    • Google Scholar
62. 1. Pierce RB
    2. Carlson AJ
    3. Carlson BM
    4. Hudson D
    5. Staples DF

    (2013) Depths and thermal habitat used by large versus small northern pike in three minnesota lakes

    Transactions of the American Fisheries Society 142:1629–1639.

    https://doi.org/10.1080/00028487.2013.822422

    • Google Scholar
63. Book

    1. Pitcher TJ

    (1993) Functions of Shoaling Behaviour in Teleosts In

    In: Pitcher TJ, editors. Behaviour of Teleost Fishes. Dordrecht: Chapman and Hall. pp. 363–439.

    https://doi.org/10.1007/978-94-011-1578-0

    • Google Scholar
64. 1. Quinn JL
    2. Cresswell W

    (2006) Testing domains of danger in the selfish herd: sparrowhawks target widely spaced redshanks in flocks

    Proceedings. Biological Sciences 273:2521–2526.

    https://doi.org/10.1098/rspb.2006.3612

    • PubMed
    • Google Scholar
65. Book

    1. Raat AJP

    (1988)

    Synopsis of Biological Data on the Northern Pike: Esox Lucius Linnaeus, 1758Synopsis of Biological Data on the Northern Pike: Esox Lucius Linnaeus, 1758

    FAO Fisheries Synopsis.

    • Google Scholar
66. 1. Ranåker L
    2. Jönsson M
    3. Nilsson PA
    4. Brönmark C

    (2012) Effects of brown and turbid water on piscivore-prey fish interactions along a visibility gradient

    Freshwater Biology 57:1761–1768.

    https://doi.org/10.1111/j.1365-2427.2012.02836.x

    • Google Scholar
67. 1. Romenskyy M
    2. Herbert-Read JE
    3. Ioannou CC
    4. Szorkovszky A
    5. Ward AJW
    6. Sumpter DJT
    7. Dingemanse N

    (2020) Quantifying the structure and dynamics of fish shoals under predation threat in three dimensions

    Behavioral Ecology 31:311–321.

    https://doi.org/10.1093/beheco/arz197

    • Google Scholar
68. 1. Romey WL
    2. Walston AR
    3. Watt PJ

    (2008) Do 3-D predators attack the margins of 2-D selfish herds?

    Behavioral Ecology 19:74–78.

    https://doi.org/10.1093/beheco/arm105

    • Google Scholar
69. 1. Rosenthal SB
    2. Twomey CR
    3. Hartnett AT
    4. Wu HS
    5. Couzin ID

    (2015) Revealing the hidden networks of interaction in mobile animal groups allows prediction of complex behavioral contagion

    PNAS 112:4690–4695.

    https://doi.org/10.1073/pnas.1420068112

    • PubMed
    • Google Scholar
70. 1. Sankey DWE
    2. Storms RF
    3. Musters RJ
    4. Russell TW
    5. Hemelrijk CK
    6. Portugal SJ

    (2021) Absence of “selfish herd” dynamics in bird flocks under threat

    Current Biology 31:3192–3198.

    https://doi.org/10.1016/j.cub.2021.05.009

    • PubMed
    • Google Scholar
71. 1. Seghers BH

    (1974) Schooling behavior in the guppy (poecilia reticulata): An evolutionary response to predation

    Evolution; International Journal of Organic Evolution 28:486–489.

    https://doi.org/10.1111/j.1558-5646.1974.tb00774.x

    • PubMed
    • Google Scholar
72. Book

    1. Skov C
    2. Lucas MC
    3. Jacobsen L

    (2018) Spatial ecology

    In: Skov C, Nilsson PA, editors. Biology and Ecolog of Pike. CRC Press. pp. 1–8.

    https://doi.org/10.1201/9781315119076

    • Google Scholar
73. 1. Sosna MMG
    2. Twomey CR
    3. Bak-Coleman J
    4. Poel W
    5. Daniels BC
    6. Romanczuk P
    7. Couzin ID

    (2019) Individual and collective encoding of risk in animal groups

    PNAS 116:20556–20561.

    https://doi.org/10.1073/pnas.1905585116

    • PubMed
    • Google Scholar
74. 1. Stankowich T

    (2003) Marginal predation methodologies and the importance of predator preferences

    Animal Behaviour 66:589–599.

    https://doi.org/10.1006/anbe.2003.2232

    • Google Scholar
75. 1. Stone NM
    2. Kelly AM
    3. Roy LA

    (2016) A fish of weedy waters: golden shiner biology and culture

    Journal of the World Aquaculture Society 47:152–200.

    https://doi.org/10.1111/jwas.12269

    • Google Scholar
76. 1. Strandburg-Peshkin A
    2. Twomey CR
    3. Bode NWF
    4. Kao AB
    5. Katz Y
    6. Ioannou CC
    7. Rosenthal SB
    8. Torney CJ
    9. Wu HS
    10. Levin SA
    11. Couzin ID

    (2013) Visual sensory networks and effective information transfer in animal groups

    Current Biology 23:R709–R711.

    https://doi.org/10.1016/j.cub.2013.07.059

    • PubMed
    • Google Scholar
77. 1. Theodorakis CW

    (1989) Size segregation and the effects of oddity on predation risk in minnow schools

    Animal Behaviour 38:496–502.

    https://doi.org/10.1016/S0003-3472(89)80042-9

    • Google Scholar
78. 1. Tunstrøm K
    2. Katz Y
    3. Ioannou CC
    4. Huepe C
    5. Lutz MJ
    6. Couzin ID

    (2013) Collective states, multistability and transitional behavior in schooling fish

    PLOS Computational Biology 9:e1002915.

    https://doi.org/10.1371/journal.pcbi.1002915

    • PubMed
    • Google Scholar
79. 1. Turesson H
    2. Bronmark C

    (2004) Foraging behaviour and capture success in perch, pikeperch and pike and the effects of prey density

    Journal of Fish Biology 65:363–375.

    https://doi.org/10.1111/j.0022-1112.2004.00455.x

    • Google Scholar
80. 1. Turner GF
    2. Pitcher TJ

    (1986) Attack abatement: a model for group protection by combined avoidance and dilution

    The American Naturalist 128:228–240.

    https://doi.org/10.1086/284556

    • Google Scholar
81. 1. Vine I

    (1971) Risk of visual detection and pursuitby a predator and the selective advantage of flocking behaviour

    Journal of Theoretical Biology 30:405–422.

    https://doi.org/10.1016/0022-5193(71)90061-0

    • PubMed
    • Google Scholar
82. 1. Viscido SV
    2. Wethey DS

    (2002) Quantitative analysis of fiddler crab flock movement: evidence for ‘selfish herd’ behaviour

    Animal Behaviour 63:735–741.

    https://doi.org/10.1006/anbe.2001.1935

    • Google Scholar
83. 1. Walker JA
    2. Ghalambor CK
    3. Griset OL
    4. McKENNEY D
    5. Reznick DN

    (2005) Do faster starts increase the probability of evading predators?

    Functional Ecology 19:808–815.

    https://doi.org/10.1111/j.1365-2435.2005.01033.x

    • Google Scholar
84. Book

    1. Ward AJW
    2. Webster MM

    (2016) Sociality: The Behaviour of Group-Living Animals

    Cham: Springer International Publishing.

    https://doi.org/10.1007/978-3-319-28585-6

    • Google Scholar
85. 1. Webb PW
    2. Skadsen JM

    (1980) Strike tactics of Esox

    Canadian Journal of Zoology 58:1462–1469.

    https://doi.org/10.1139/z80-201

    • PubMed
    • Google Scholar
86. 1. Webb PW

    (1982) Avoidance responses of fathead minnow to strikes by four teleost predators

    Journal of Comparative Physiology? A 147:371–378.

    https://doi.org/10.1007/BF00609671

    • Google Scholar
87. 1. Webster MM
    2. Rutz C

    (2020) How STRANGE are your study animals?

    Nature 582:337–340.

    https://doi.org/10.1038/d41586-020-01751-5

    • PubMed
    • Google Scholar

Decision letter after peer review:

Thank you for submitting your article "Both Prey and Predator Features Predict the Individual Predation Risk and Survival of Schooling Prey" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Christian Rutz as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this decision letter to help you prepare a revised submission.

Essential revisions:

1. Please discuss the limitations of the experimental design, including limited space for the prey to move and lack of opportunity to seek refuge, as well as the possibility that the stealthy hunting strategy described for the pike is affected by tank conditions.

2. Please discuss the limitations of 2D versus 3D tracking. Although the manuscript includes some arguments for why 2D tracking is sufficient, one of the reviewers noted that the primary supporting citation is for a prey-focused study design for different predator and prey species. Rather than making the assumption that the third dimension does not matter, please evaluate critically how violations of that assumption might affect your inferences/conclusions (or motivate further work) and ensure that this discussion is incorporated in the main text to make it apparent to readers.

3. Please clarify how the present results contribute to our understanding of the evolution of collective behavior in the wild, taking into account the potentially counterintuitive observation that fish in smaller clusters experienced lower predation. And please address Reviewer #2's related question about whether repeated exposure leads to reduced schooling behavior overall.

4. Please clarify all questions related to experimental design, animal numbers, the outcome of prey attacks, and statistical analyses, and add an explanatory figure, as suggested by Reviewer #3.

Reviewer #1 (Recommendations for the authors):

Overall this paper is beautifully written and presented, with appropriate analyses.

It is packed full of information, and as a result, it can be a little dense and hard to follow at times. Some streamlining, signposting and highlighting of key points would be of benefit.

One solution may be clearer hypotheses or at least clearer aims. This would add some further structure to the text. As a reader, going from the Introduction to the first line in the Results section being about describing pike movements, it was a little tricky to immediately follow the logic and flow.

From reading everything, it is still not clear to me if the pike actually caught and killed the shiners? It feels a little like the ethics section dances somewhat around this point. Please clarify.

The authors allude to the captivity aspect in their Discussion but I think this needs to be greater. There is still limited space for the prey to move and react. You mention this, but you don't expand on what you think would happen if the prey had endless options?

Do pike ambush easier when extensive cover is provided?

In the Introduction you state: "This not only has crucial implications for the selection of individual phenotypes, but also for the emergence of collective behaviour, and the dynamics and functioning of animal groups (Farine et al., 2015; Jolles et al.,2020; Ward and Webster, 2016). Hence, better understanding what factors influence individual predation risk within animal groups is of fundamental importance." I think this needs to be clearer. Why is it actually of fundamental importance? What real benefit is there to finding this out, beyond modelling something? Linked to this is a frequent reference to the same work, and/or the same authors. This started to feel a little like an echo chamber. I would suggest some diversification in the literature you refer too.

Based on the speed at which everything is happening, this can change in an instant. What time period was this considered over? Apologies if I missed this. "As most predators only have a specific region that they are biologically capable of attacking, we also only considered shiners found inside the pikes' strike zone, an area of roughly 8 cm wide and 15 cm long directly in front of the pike within which all targeted prey were positioned (Figure 2E)."

Based on Figure 1, how relevant is the centroid?

How much do the lighting conditions (1) influence how pike hunt, and (2) determine how the shiners behave in response to a threat?

"The behavioural type of the shiners and pike was not determined but likely showed a natural range". This seems very vague and somewhat assumptive. I think there's plenty of evidence to suggest this could go either way? Please provide a little more detail here.

"…have been of long-standing interest (Ioannou et al., 2012; Krause and Ruxton, 2002; Ward and Webster, 2016)". I'm not convinced you can say long-standing and then present references which, in the big scheme of things, are all relatively recent.

"However, besides by predators using a stealthy approach as we show here, central individuals may also be more at risk by predators". Very clunky sentence that I had to read a few times to understand what you meant. It might benefit from a minor re-write.

Reviewer #2 (Recommendations for the authors):

1. The second paragraph in the introduction came off as more of a list of conflicting evidence rather than explaining a topic. Perhaps it could be restructured to better motivate the following paragraphs by stating something like "Previous work has identified many different, and sometimes contradictory, factors that predict prey vulnerability in a collective group, likely because predation strategy has been overlooked."

2. What was the top speed of the shiners? This might be relevant to report in line 145.

3. In line 170, it might also be useful to refer to the distances between the pike and the centroid of the school in terms of the percent of total size.

4. Line 142: Consider changing “compromised” to “comprised”.

5. Line 337: Consider changing “However, besides by predators” to “In addition to predators”.

6. The topics of the paragraphs starting on lines 346 and 419 are overlapping. Perhaps these paragraphs can be combined so that the results of the study and the previous work in the field can be discussed together?

7. The sentence starting on line 434 is very long and confusing. Consider rewording.

8. The sentence starting on line 452 is also very long and unfocused. Consider rewording.

9. Consider breaking up the sentence starting on line 468: “…the predator. Even central individuals have a cha”ce".

10. Line 575: What does "we assured perfect tracks" mean?

Reviewer #3 (Recommendations for the authors):

Overall, I enjoyed reading this impressive paper. It was well written and provided a nice example of the modern, highly quantitative analyses of predator-prey attack dynamics in controlled laboratory conditions. I had only a few relatively minor comments.

The authors did highly detailed analyses of metrics that explained which prey individual was targeted by the predator, and whether the targeted individual successfully evaded capture or not. This assumes that it is clear cut that there was a particular individual targeted and that the analysis correctly identified the 'targeted' individual. Given that there were probably some, if not many, attacks where multiple prey were directly in front of the predator and close by, what exact criteria did the authors use to identify which prey was targeted by the predator? In the 70% of attacks where a prey fish was captured, it is reasonably clear that that was the prey that was targeted, but in the other 30%, it seems less clear cut. Were there some, perhaps quite a few, situations where it was not 100% clear which prey was targeted? Conventional wisdom includes the idea that attack success is reduced if predators get 'confused' by having numerous prey that are potential targets, particularly if they escape in different directions. If the authors mis-identified which prey was targeted this could presumably affect the accuracy of their analyses on factors that affect attack success.

Presumably to facilitate having sharp video images and 2-dimensional analyses of encounters, the trials were done in 6 cm of water which is likely much shallower than natural conditions for pike and shiners. The authors suggest that shiners are commonly in flat schools, so quantifying their position and movements in 3 dimensions is not necessary. They also cite a recent paper (Romenskyy et al. 2020) that suggested that the third dimension (variation in the vertical-depth axis) is not critically important in a similar study quantifying the details of attack, but this other study was on different predator and prey fish species, and focused only on which prey are targeted, not on attack success. Overall, the notion that it is valuable to study fish predator-prey interactions in 2 dimensions sounds plausible at a rough level, but I am somewhat skeptical of that conclusion at the very fine, detailed scale of this paper's analyses. Nonetheless, I found this study to be impressive and valuable, even if the detailed dynamics under some, perhaps many natural conditions (deeper water, with some refuge not far away, in different light conditions etc.) might be moderately different.

The experimental design and sample sizes with regard to pattern of re-use of animals across multiple days (lines 544 ff.) might be easier to understand if the authors included a summary figure. How many groups of 40 shiners began each batch? There were 9 pike in batch 1 and 4 in batch 2, does that mean that the experiment began with 360 shiners in batch 1 and 160 in batch 2? Trials continued until there were not enough fish left for another trial with that number of exposures, which I think means until there were fewer than 40 fish. But only 70% of 125 attacks were successful (about 88). If the study started with 520 prey and only 88 were killed, it sounds like the study could go through a very large number of repeated trials before it would not have enough prey to continue. I am clearly not understanding something about the design. Pike had 3 rest days between trials. Was the same rest schedule applied to shiners? Shiners were kept in tanks with fish with the same number of previous exposures, but did this mix animals with different numbers of days since their previous trial? How many shiners participated in the study? There were a total of 125 attacks, with 1.84 attacks/trial which seems to mean 68 trials, spread across 13 pike in 2 batches. But 'a number of pike' never attacked any prey. Does the 13 pike include these or are the 13 pike only the ones that did attack prey?

I will admit that I might not have outlined the paper carefully enough, but the statistical analyses described briefly in lines 193 ff., and in some detail in the Appendix involved reasonably standard model choice methods listing relative model weights, but in the main text, the authors also showed various other statistics: LRT chi-square values, F-values, with p-values, R-squares, and correlation coefficients. Perhaps I missed where they described these, but if indeed they did not do so, they should describe all statistical tests.

Model choice analyses also allow one to assign relative weights not just to each model, but to each factor by summing the weights for each model in which that variable appears. Rather than only show the top 15 models (Appendix 3), it might be useful to calculate the weights for each factor. This would complement the results and conclusions in the main text.

Perhaps I did not study the Appendices carefully enough, but the best listed models (Appendix 3 and 4) did not include number of exposures, pike ID or attack attempt, or more importantly, interactions between these random effects and the predictive features from either the prey or predator view. Appendix 2 shows that exposure number had large effects on pike distance from the group centroid and from the target individual. Did the more detailed features of targeted prey (or attack success) change with repeated exposures? For each exposure number (Appendix 2 - figure) the distances varied quite a lot. I assume that these are for the 13 different pike. Did individual pike differ consistently in their targeting tendencies, or even in their change in targeting tendencies with repeated exposure and learning? If so, this deserves more discussion.

Appendix 2, figure 1 -- what are the darker points in figures B and C?

Line 193: 'were' not 'where'.

Essential revisions:

1. Please discuss the limitations of the experimental design, including limited space for the prey to move and lack of opportunity to seek refuge, as well as the possibility that the stealthy hunting strategy described for the pike is affected by tank conditions.

We have considerably extended our discussion and now included two dedicated paragraphs to discuss the potential limitations of our work and how the stealthy hunting strategy of the pike may be affected by these conditions. This includes a discussion of the effect of tank size (lines 472-480): “Laboratory studies on predator-prey dynamics like ours do, of course, have their limitations. Although the size of the arena we used (~2m²) is in line with behavioural studies with large schools of fish (e.g. Sosna et al., 2019; Strandburg-Peshkin et al., 2013) and experiments with live predators attacking schooling prey (Bumann et al., 1997; Magurran and Pitcher, 1987; Neill and Cullen, 1974; Romenskyy et al., 2020; Theodorakis, 1989), compared to conditions in the wild the prey and predator had limited space to move. However, as pike are ambush predators they tend to move relatively little to search for prey and rather rely on prey movement for encounters (Nilsson and Eklöv, 2008). Increasing tank size would have made effective tracking extremely difficult, or impossible, and while a much larger tank is expected to considerably increase latency to attack, we expect it to have relatively little effect on the observed findings.”

As well as a discussion of the lack of refuges (lines 495-502): “For our experiments we used a testing arena without any internal structures such as refuges. This was a strategic decision as providing a more complex environment would have impacted the ability of the shiners to school in large groups and would have led fish to hide under cover. Although studying predator-prey dynamics in more complex environments would be interesting in its own regard, it would not have allowed us to study the questions we are interested in about the predation risk of free-schooling prey. Furthermore, pilot experiments indicated that the pike never used refuges (consistent with previous work, see Turesson and Brönmark, 2004), so they were not further provided during the actual experiment”.

2. Please discuss the limitations of 2D versus 3D tracking. Although the manuscript includes some arguments for why 2D tracking is sufficient, one of the reviewers noted that the primary supporting citation is for a prey-focused study design for different predator and prey species. Rather than making the assumption that the third dimension does not matter, please evaluate critically how violations of that assumption might affect your inferences/conclusions (or motivate further work) and ensure that this discussion is incorporated in the main text to make it apparent to readers.

We apologize for the misunderstanding as it was not our intention to state the third dimension does not matter. Rather, we decided to focus on two-dimensions as the third dimension would have made it considerably more difficult, both to keep track of the individual fish in the highly cohesive schools before and during the attacks, and to be able to consider features related to the visual field of the fish. We now better clarify the shallow water as a limitation of our study, explain why shallow water was used, and discuss our expectations of what could happen in deeper water (lines 480-494): “In terms of water depth, fish were tested in relatively very shallow water. This was primarily done to be able to keep track of individual identities and compute features related to the visual field of the fish. Shiners naturally school in very shallow water conditions as well as near the surface in deeper water in the wild (Hall et al., 1979; Krause et al., 2000b; Stone et al., 2016) and also pike primarily occur in the shallow littoral zone, sometimes only a few of tens of cm deep (Pierce et al., 2013; Skov et al., 2018). Furthermore, pilot experiment showed the pike did exhibit normal swimming and attack behaviour with attack speeds and acceleration comparable to previous work (Domenici and Blake, 1997; Walker et al., 2005). Recent other work on predator-prey dynamics did not find a considerable impact of adding the third dimension to their analyses (Romenskyy et al., 2020). Still, the water depth used is a limiting factor of our study and in the future this type of work should be extended to deeper water while still keeping track of individual identities over time. We expect that adding the third dimension would not change the stealthy attack behaviour of the pike and therefore still put more central individuals most at risk, but possibly attack success would be reduced because of increased predator visibility and prey escape potential in the vertical plane, which remains to be tested.”

3. Please clarify how the present results contribute to our understanding of the evolution of collective behavior in the wild, taking into account the potentially counterintuitive observation that fish in smaller clusters experienced lower predation. And please address Reviewer #2's related question about whether repeated exposure leads to reduced schooling behavior overall.

We now included a dedicated paragraph in our discussion to better explain how our results may contribute to our understanding of the evolution of animal grouping, including of the potential evolutionary consequences of central individuals experiencing higher predation risk (lines 456-467): “Predation is seen as one of the main factors to shape the collective properties of animal groups (Herbert-Read et al., 2017) and has so far generally been seen as to drive the formation of larger, more cohesive groups that exhibit collective, coordinated motion (see e.g. Beauchamp, 2004; Ioannou et al., 2012; B. H. Seghers, 1974). Our finding that central individuals are more at risk of being predated could actually have the opposite effect, with schooling having a selective disadvantage, which over time could result in weaker collective behaviour and less cohesive schools. However, we do not deem this likely as selection is likely to be group-size dependent, as discussed above. Furthermore, our multi-model inference approach revealed that, despite more central individuals experiencing higher predation risk, being close to others inside the school was still associated with a lower risk of being targeted. As most prey experience many types of predators, including sit-and-wait predators and active predators that hunt for prey, the extent and direction of such selection effects will depend on the broader predation landscape in which prey find themselves.”

We understand that the result that predation risk was higher in the main cluster could be interpreted as suggesting that schooling may have a selective disadvantage. However, we would like to point out that the per-capita-risk would actually be lower in the schools due to the dilution of risk. As in our experiments the shiners were highly cohesive and almost always in one single large school – rather than exhibiting fission-fusion dynamics (merging and splitting of schools) – we feel our dataset is not appropriate for proper interpretations of group-size effects on predation risk. We now also clarify this point in the discussion (lines 467-471): “While the finding that pike were more likely to attack the main school may also appear to indicate a selective disadvantage to school, calculating the per-capita-risk for each individual would actually reveal it is still safest to be part of the main school. Nevertheless, as the shiners in our study rarely exhibited fission-fusion dynamics we feel our dataset is not appropriate to make proper inferences about how predation risk is linked to group size.”

Repeated exposure did not impact the packing fraction of the schools and thus did not lead to reduced schooling behaviour overall, as we also stated in the results of the original manuscript (lines 183-187): “While the shiners did not show a change in their packing fraction (median nearest-neighbour distance) with repeated exposure to the pike (F1,52 = 1.81, p = 0.185), they increasingly avoided the area directly in front of the pike’s head (Appendix 2 – Figure 1A) resulting in the pike attacking from increasingly further away (target distance: F1,52 = 45.52, p < 0.001, see Appendix 2 – Figure 1B,C).”

4. Please clarify all questions related to experimental design, animal numbers, the outcome of prey attacks, and statistical analyses, and add an explanatory figure, as suggested by Reviewer #3.

We have made considerable modifications to the manuscript to accommodate the questions related the experimental design, sample sizes, repeated testing of animals, outcome of prey attacks, and statistical analyses (see detailed changes in our responses below). We thereby put extra attention on the Experimental Procedure section in the methods (lines 590-611): “We ran the experiment with two batches of pike and shiners, with the fish of each batch tested repeatedly in the arena over time. Pike were randomly selected, but only after they had at least three rest days since the previous trial, which was decided based on pilot work that showed pike were less motivated when tested with less time between trials. For each trial, shiners were also randomly selected from the holding tanks to create groups of 40 fish. Shiners were mixed to avoid potential effects of familiarity between individuals and the associated social feedback and learning effects related to group composition. However, as we used a repeated-measures design, we made sure that all shiners in the same group had the same number of exposures to the pike by keeping shiners in separate social holding tanks based on their number of pike exposures. In total we used 20 pike and started with about 1500 shiners. Pike had a mean number of 1.84 ± 0.14 attacks per trial, with 18 trials having more than 1 successful attack. However, not all pike did always attack during the trials, and 7 pike never attacked. As those trials did not provide any data on the attacks they were excluded and the pike not used in further trials. Also the shiners were excluded for subsequent trials as their future behaviour in the assay could be influenced by having had experience with a pike that did not attack. As a result, our sample size decreased with exposure and we stopped at a maximum number of 6 exposures. Like the pike, shiners were also generally tested with three rest days between trials but sometimes groups had one or a few fish that only had two rest days. This was inevitable due to the difficulty of just being able to run a few trials each day, and sometimes pike thus not attacking. Although none of the fish had any previous experience with any experiments, during the repeated exposures both shiners and pike had time to learn about the conditions they were confronted with, thus enabling us to also investigate how this impacted predators’ attack behaviour and prey response.”

While we appreciate the suggestion by Reviewer #3 to include an explanatory figure about the experimental design, we think this would result in a rather complex figure to show the repeated exposures, holding of shiners based on exposure, and fish being excluded due to lack of attacks, so hope our revisions of the text are sufficient.

Reviewer #1 (Recommendations for the authors):

Overall this paper is beautifully written and presented, with appropriate analyses.

It is packed full of information, and as a result, it can be a little dense and hard to follow at times. Some streamlining, signposting and highlighting of key points would be of benefit.

One solution may be clearer hypotheses or at least clearer aims. This would add some further structure to the text. As a reader, going from the Introduction to the first line in the Results section being about describing pike movements, it was a little tricky to immediately follow the logic and flow.

We thank the reviewer for their positive remarks. We have considerably revised the introduction to address the different points of the reviewer. We now better clarify the aims of our study on lines 95-98: “To advance our understanding of differential predation risk in animal groups we need to systematically investigate the different absolute and relative spatial and visual features of both prey and predator while considering the realtime dynamics between live predators attacking groups of prey they can actually capture.”

And further on lines 100-102: “Specifically, to gain a detailed mechanistic understanding of when and where predators attack groups of prey and what predicts individual predation risk and survival."

We now also better discuss previous work, e.g. lines 69-73: “Much of the focus in the literature on the spatial effects underlying predation risk has looked at centre-to-edge and front-to-back effects, partly because they are the easiest to measure, and largely concentrated only on a single or a few key potential features (but see e.g. Lambert et al., 2021; Romenskyy et al., 2020). However, it may be more likely that a spectrum of different factors shapes predation risk within groups.” Line 81: “The majority of previous research has also focused exclusively on prey behaviour.” And lines 92-94: “It therefore remains unclear what factors actually affect attack success and the probability for targeted prey to potentially survive attacks.”

Finally, we now explain in more detail what types of analyses we did and in what order (lines 108-115): “By tracking both the predator and all prey individually over time, we were able to quantify each fish’s spatial position, relative spacing, orientation, and visual field, and analysed their movement kinematics in detail throughout each attack (Figure 1). We then quantified in detail how, when, and where the pike attacked. Subsequently, we used model fitting procedures to infer, both from the prey’s and predator’s perspective, the relative importance of a suite of potential features to predict shiners’ risk to be targeted. Finally, as pike were able to catch and consume their prey, we were able to investigate what factors best predicted the likelihood of attacks to be successful and thus for prey to survive a predator attack (see Table 1).”

From reading everything, it is still not clear to me if the pike actually caught and killed the shiners? It feels a little like the ethics section dances somewhat around this point. Please clarify.

This aspect of the study, the predictors of attack success and pike being able to consume their prey, is actually one of the key novelties of our work, as most previous work only quantified the risk to be attacked or could not quantify the predictors for survival. While this was noted in the original manuscript (e.g. lines 30, 107-108, 288-304, 407-422), we have now made this more explicit and have added further clarifications throughout, including on line 114 “(pike were able to catch and consume their prey …)”, line 294 “(The pike in our study were allowed to catch and consume their prey and were relatively successful…)”, lines 518-519 “(what predicts individual predation risk and survival)”, lines 582583 “(trials were not stopped after the first shiner was captured or eaten)”, and line 717 “(live predators could interact freely with and consume their prey).”

The authors allude to the captivity aspect in their Discussion but I think this needs to be greater. There is still limited space for the prey to move and react. You mention this, but you don't expand on what you think would happen if the prey had endless options?

We would like to note that in the methods of the original manuscript we stated our expectations regarding this point (lines 525-526): “We expect that if a much larger space would have been used, pike would still show the same approach and attack behaviour linked to their stealthy attack strategy.”

Nevertheless, following this, and the other reviewers’ suggestions, we now dedicate two new paragraphs in the discussion to the potential limitations of our experimental design, including about the limited space available. We now state more extensively (lines 472-480): “Laboratory studies on predator-prey dynamics like ours do, of course, have their limitations. Although the size of the arena we used (~2m²) is in line with behavioural studies with large schools of fish (e.g. Sosna et al., 2019; Strandburg-Peshkin et al., 2013) and experiments with live predators attacking schooling prey (Bumann et al., 1997; Magurran and Pitcher, 1987; Neill and Cullen, 1974; Romenskyy et al., 2020; Theodorakis, 1989), compared to conditions in the wild the prey and predator had limited space to move. However, as pike are ambush predators they tend to move relatively little to search for prey and rather rely on prey movement for encounters (Nilsson and Eklöv, 2008). Increasing tank size would have made effective tracking extremely difficult, or impossible, and while a much larger tank is expected to considerably increase latency to attack, we expect it to have relatively little effect on the observed findings.”

Do pike ambush easier when extensive cover is provided?

While we included a relevant statement about this in the methods section of the original manuscript (lines 530-533) – in which we explain that pilot experiments indicated that pike did not use refuges, in line with previous work – we have now expanded this point in the main text (lines 495-502): “For our experiments we used a testing arena without any internal structures such as refuges. This was a strategic decision as providing a more complex environment would have impacted the ability of the shiners to school in large groups and would have led fish to hide under cover. Although studying predator-prey dynamics in more complex environments would be interesting in its own regard, it would not have allowed us to study the questions we are interested in about the predation risk of free-schooling prey. Furthermore, pilot experiments indicated that the pike never used refuges (consistent with previous work, see Turesson and Brönmark, 2004), so they were not further provided during the actual experiment.”

We would like to note that, although pike are sit-and-wait predators that often hide in cover, in the wild it is predominantly younger pike that stay in vegetation, with cannibalism by larger pike being one of the main driving factors (see Skov and Lucas, 2018 Chapter 5). Due to their elongated body and narrow frontal body pike do not necessarily need refuges and are also able to ambush prey outside of cover, as our study shows. We now also better discuss this aspect on lines 338-343 of the discussion: “That predators such as pike can get very close to their prey has been described previously (Coble, 1973; Hoogland et al., 1956; Krause et al., 1998; Nursall, 1973; Webb and Skadsen, 1980) and could potentially be explained by their narrow frontal profile (Webb, 1982) as it makes it very hard for prey to detect movement changes, especially when attacked head-on. This may also explain the suggestion of previous work that pike do not suffer much from the confusion effect (Turesson and Brönmark, 2004).”

In the Introduction you state: "This not only has crucial implications for the selection of individual phenotypes, but also for the emergence of collective behaviour, and the dynamics and functioning of animal groups (Farine et al., 2015; Jolles et al.,2020; Ward and Webster, 2016). Hence, better understanding what factors influence individual predation risk within animal groups is of fundamental importance." I think this needs to be clearer. Why is it actually of fundamental importance? What real benefit is there to finding this out, beyond modelling something? Linked to this is a frequent reference to the same work, and/or the same authors. This started to feel a little like an echo chamber. I would suggest some diversification in the literature you refer too.

We agree that this was unclear. We have considerably revised the first paragraph of the introduction to better clarify the importance of improving our understanding predation risk in animal groups (lines 36-50): “A key challenge in the life of most animals is to avoid being eaten. Via effects such as enhanced predator detection (Lima, 1995; Magurran et al., 1985), predator confusion (Landeau and Terborgh, 1986), and risk dilution effects (Foster and Treherne, 1981; Turner and Pitcher, 1986), individuals living and moving in groups can reduce their risk of predation (Ioannou et al., 2012; Krause and Ruxton, 2002; Pitcher and Parrish, 1993; Ward and Webster, 2016). This helps explain why strong predation pressure is known to drive the formation of larger and more cohesive groups (Beauchamp, 2004; Krause and Ruxton, 2002; B. Seghers, 1974). However, the costs and benefits of grouping are not shared equally among individuals within groups, and besides differential food intake and costs of locomotion, group members themselves may experience widely varying risks of predation (Handegard et al., 2012; Krause, 1994; Krause and Ruxton, 2002). Where and whom predators attack within groups not only has major implications for the selection of individual phenotypes, and thereby the emergence of collective behaviour and the functioning of animal groups (Farine et al., 2015; Jolles et al., 2020; Ward and Webster, 2016), but also shapes the social behaviour of prey and the properties and structure of prey groups. Hence, a better understanding of the factors that influence predation risk within animal groups is of fundamental importance."

We now also included a dedicated paragraph about the potential evolutionary implications of our findings in the discussion (lines 456-471, see above) and included a number of new references about the role of predation in the evolution of animal grouping, including (Herbert-Read et al., 2017; Beauchamp, 2004; Seghers, 1974).

Based on the speed at which everything is happening, this can change in an instant. What time period was this considered over? Apologies if I missed this. "As most predators only have a specific region that they are biologically capable of attacking, we also only considered shiners found inside the pikes' strike zone, an area of roughly 8 cm wide and 15 cm long directly in front of the pike within which all targeted prey were positioned (Figure 2E)."

We considered the shiners that were found in the pike’s strike zone at the frame of attack but followed their position and behaviour from 0.5 s before until 0.1 s after strike initiation. For further information, please see the Quantification of behaviour and Analyses section of the Appendix.

Based on Figure 1, how relevant is the centroid?

The centroid (geometric centre of the group) is a helpful measure as it allows us to calculate the overall and relative motion of the prey group (see lines 155-156 and 198-200), the distance from which the pike attacked (lines 180-181), as well as the hierarchical clustering approach employed to establish the number of groups (lines 143150). We would like to highlight that in the manuscript we provide extensive discussion about the relevance of the group centre-edge positioning and that Figure 1 is a cropped image from a sample trial just moments before the attack that shows the relevance of the centroid in terms of the pike and shiners’ position.

How much do the lighting conditions (1) influence how pike hunt, and (2) determine how the shiners behave in response to a threat?

As pike are visual predators, in general pike display a diel activity pattern, with more activity during the day than at night. However, although quite some variation exists between populations, behavioural types, and age classes, with larger pike being more active throughout the day, pike tend to show their peaks in activity at dusk and dawn (see Skov and Lucas, 2018 Chapter 5). Hence, we expect that at relatively low light intensity pike would show the highest activity and at high light intensity the lowest. For shiners we know that a sharp reduction in light intensity results in visual impairment and thereby reduced ability to respond to their external environment. For our study design we aimed to not use too bright lights that may subdue the attack behaviour of the pike but we could also not use too dim light, both because shiners are a highly visual species and because our tracking software would struggle to differentiate the fish from the tank, especially during their fast movement during attacks. As the pike in our study tended to be active and show normal attack behaviour during the trials, we believe we found a reasonable balance with the light intensity provided. We added a clarification about the light intensity in the methods (lines 560-562): “Diffused light was provided by two LED panels positioned outside the curtains at the far ends of the tank, which provided light of medium intensity.”

And we dedicated some sentences about the potential effect of light intensity in the discussion (lines 502-508): “Experiments were conducted under artificial lights with reduces intensity but at a level high enough to be able to acquire accurate videos of the trials without motion blur. In the wild, pike are however generally found to be most active during dusk and dawn (Raat, 1988; Skov et al., 2018) and to consume most prey at low light intensity (Dobler, 1977). We expect that if a lower light intensity was used, the pike may profit from visual superiority and thereby would have increased predation success, further aided by a likely loosening of the prey schools due to limited light being available (Dobler, 1977).”

"The behavioural type of the shiners and pike was not determined but likely showed a natural range". This seems very vague and somewhat assumptive. I think there's plenty of evidence to suggest this could go either way? Please provide a little more detail here.

We apologize as we did not want to make any assumptions here about the behavioural types of the fish (i.e. consistent behavioural differences among individuals). Rather, we just wanted to clarify that was not part of our study. We have now removed the sentence to not raise any confusion. We do think that fishes behavioural type will likely play an important role in both the attack behaviour of the pike and position and avoidance response of the shiners, but this is something for future work.

"…have been of long-standing interest (Ioannou et al., 2012; Krause and Ruxton, 2002; Ward and Webster, 2016)". I'm not convinced you can say long-standing and then present references which, in the big scheme of things, are all relatively recent.

We apologise for this ambiguity. These works are reviews that cite the previous literature, and we now make this clear, preceding the citations with the term “reviewed in”. We also included Pitcher and Parrish, 1993 as a further reference.

"However, besides by predators using a stealthy approach as we show here, central individuals may also be more at risk by predators". Very clunky sentence that I had to read a few times to understand what you meant. It might benefit from a minor re-write.

We thank the reviewer for highlighting this. We have rewritten the sentence accordingly (lines 349-354): “Furthermore, increased risk of individuals near the centre of groups may be more widespread than currently thought. Predators not only exhibit stealthy behavioural tactics that enable them to approach and attack central individuals, as we show here, but may also do so by attacking groups from above (Brunton, 1997) or below (Clua and Grosvalet, 2001; Hobson, 1963; but see Romey et al., 2008), and by rushing into the main body of the group (Handegard et al., 2012; Hobson, 1963; Parrish et al., 1989).”

Reviewer #2 (Recommendations for the authors):

1. The second paragraph in the introduction came off as more of a list of conflicting evidence rather than explaining a topic. Perhaps it could be restructured to better motivate the following paragraphs by stating something like "Previous work has identified many different, and sometimes contradictory, factors that predict prey vulnerability in a collective group, likely because predation strategy has been overlooked."

We thank the reviewer for this fair suggestion. We now start the second introductory paragraph with the sentence: “Previous work has identified many different, and sometimes contradictory, factors that predict prey vulnerability in groups.”

2. What was the top speed of the shiners? This might be relevant to report in line 145.

The maximum speed of the shiners was observed during the predatory attacks, which is reported on lines 126-127 of the original manuscript. We now report the 90% quantile of the top speed of the school outside of the actual attack as representative measure (line 155-156): 90% quantile: 16.3 cm/s.

3. In line 170, it might also be useful to refer to the distances between the pike and the centroid of the school in terms of the percent of total size.

The schools not necessarily had a circular or oval shape so a straight-forward computation of relative distance based on total school size is not possible. This is also one of the reasons why we further investigate if pike actually entered the groups, see results on lines 190-200.

4. Line 142: Consider changing “compromised” to “comprised”.

Corrected.

5. Line 337: Consider changing “However, besides by predators” to “In addition to predators”.

We have rewritten the sentence to correct this.

6. The topics of the paragraphs starting on lines 346 and 419 are overlapping. Perhaps these paragraphs can be combined so that the results of the study and the previous work in the field can be discussed together?

We thank the reviewer for bringing this up. We agree this is a great suggestion and we have now integrated and considerably rewritten the two paragraphs mentioned.

7. The sentence starting on line 434 is very long and confusing. Consider rewording.

The sentence starting on line 434 is relatively short while the one starting on line 442 is relatively long, so we presume that is the sentence the reviewer is referring to and rephrased it accordingly: “Furthermore, increased risk of individuals near the centre of groups may be more widespread than currently thought. Predators not only exhibit stealthy behavioural tactics that enable them to approach and attack central individuals, as we show here, but may also do so by attacking groups from above (Brunton, 1997) or below (Clua and Grosvalet, 2001; Hobson, 1963; but see Romey et al., 2008), and by rushing into the main body of the group (Handegard et al., 2012; Hobson, 1963; Parrish et al., 1989).”

8. The sentence starting on line 452 is also very long and unfocused. Consider rewording.

We thank the reviewer for pointing this out and have revised the sentence and split it into two, and moved it down in the discussion (lines 508-513): “While it is now increasingly possible to obtain detailed data from predation events on grouping prey in the field (see e.g. Handegard et al., 2012; Krause et al., 2017), even with the most sophisticated field-based imaging it would not have been possible to acquire the highly detailed data we obtained here. That is, individual-level characteristics of predator and all grouping prey throughout predator attacks at high spatial and temporal resolution, linked to attack success and survival.”

9. Consider breaking up the sentence starting on line 468: “…the predator. Even central individuals have a cha”ce".

Revised accordingly.

10. Line 575: What does "we assured perfect tracks" mean?

We apologize for the confusion. We have rephrased the sentence and instead now explain what we meant: “For our analyses we manually checked for constancy of individual shiners’ identities in the large schools at 120 fps from one second before strike initiation through to one second later, and manually corrected identity swaps and centroid and heading positions where needed.”

Reviewer #3 (Recommendations for the authors):

Overall, I enjoyed reading this impressive paper. It was well written and provided a nice example of the modern, highly quantitative analyses of predator-prey attack dynamics in controlled laboratory conditions. I had only a few relatively minor comments.

The authors did highly detailed analyses of metrics that explained which prey individual was targeted by the predator, and whether the targeted individual successfully evaded capture or not. This assumes that it is clear cut that there was a particular individual targeted and that the analysis correctly identified the 'targeted' individual. Given that there were probably some, if not many, attacks where multiple prey were directly in front of the predator and close by, what exact criteria did the authors use to identify which prey was targeted by the predator? In the 70% of attacks where a prey fish was captured, it is reasonably clear that that was the prey that was targeted, but in the other 30%, it seems less clear cut. Were there some, perhaps quite a few, situations where it was not 100% clear which prey was targeted? Conventional wisdom includes the idea that attack success is reduced if predators get 'confused' by having numerous prey that are potential targets, particularly if they escape in different directions. If the authors mis-identified which prey was targeted this could presumably affect the accuracy of their analyses on factors that affect attack success.

We thank the reviewer for this important question. We now include a clear explanation for how we determined the identity of the targeted individual in the methods (lines 622-627): The identity of the targeted individual was acquired by carefully comparing the video and visualisations of the tracking data at the moment of the attack. For the far majority of unsuccessful attacks we also acquired the identity of the targeted shiner with high certainty because the pike attacked the prey from very near and exhibited a clear attack trajectory. In the few cases there was a second potential target we were able to determine the most likely target by carefully going back and forth through the video, frame-by-frame, at the moment of the attack.

Our data suggests pike may not suffer much from the confusion effect, which was also suggested by Turesson & Brönmark (2004). We now further discuss this point in the context of pike being able to get so close to their prey and thereby able to actually focus on a single prey (lines 338-343): That predators such as pike can get very close to their prey has been described previously (Coble, 1973; Hoogland et al., 1956; Krause et al., 1998; Nursall, 1973; Webb and Skadsen, 1980) and could potentially be explained by their narrow frontal profile (Webb, 1982) as it makes it very hard for prey to detect movement changes, especially when attacked head-on. This may also explain the suggestion of previous work that pike do not suffer much from the confusion effect (Turesson and Brönmark, 2004).

Presumably to facilitate having sharp video images and 2-dimensional analyses of encounters, the trials were done in 6 cm of water which is likely much shallower than natural conditions for pike and shiners. The authors suggest that shiners are commonly in flat schools, so quantifying their position and movements in 3 dimensions is not necessary. They also cite a recent paper (Romenskyy et al. 2020) that suggested that the third dimension (variation in the vertical-depth axis) is not critically important in a similar study quantifying the details of attack, but this other study was on different predator and prey fish species, and focused only on which prey are targeted, not on attack success. Overall, the notion that it is valuable to study fish predator-prey interactions in 2 dimensions sounds plausible at a rough level, but I am somewhat skeptical of that conclusion at the very fine, detailed scale of this paper's analyses. Nonetheless, I found this study to be impressive and valuable, even if the detailed dynamics under some, perhaps many natural conditions (deeper water, with some refuge not far away, in different light conditions etc.) might be moderately different.

We appreciate the reviewer’s concerns about the two-dimensionality of our study. We would like to clarify that we do not think or state the third dimension is not necessary. Rather, we decided to focus on two dimensions only in this study to be able to acquire the highly detailed individual-level characteristics and quantify features related to the visual field of the fish. We now clarify this on lines 480-482: In terms of water depth, fish were tested in relatively very shallow water. This was primarily done to be able to keep track of individual identities and compute features related to the visual field of the fish. And lines 489-491: Still, the water depth used is a limiting factor of our study and in the future this type of work should be extended to deeper water while still keeping track of individual identities over time.

We now also better explain that both shiners and pike tend to spent a lot of time in shallow water, and consider how the water could have restricted the pike movements (lines 480-487): Shiners naturally school in very shallow water conditions as well as near the surface in deeper water in the wild (Hall et al., 1979; Krause et al., 2000b; Stone et al., 2016) and also pike primarily occur in the shallow littoral zone, sometimes only a few of tens of cm deep (Pierce et al., 2013; Skov et al., 2018). Furthermore, pilot experiment showed the pike did exhibit normal swimming and attack behaviour with attack speeds and acceleration comparable to previous work (Domenici and Blake, 1997; Walker et al., 2005). Pike had a body height between 3-4cm, which we now explain on lines 562-564 in the methods: The tank was filled with water to a depth of 6 cm, about 1.5x – 2x the body height of the pike. The vertical space was enough for shiners to swim over the (back of the) pike, which we indeed observed occasionally.

We removed the short discussion about the 2D aspect of our study from the methods and now also include a short discussion with our expectations for testing in three dimensions (lines 491-494): We expect that adding the third dimension would not change the stealthy attack behaviour of the pike and therefore still put more central individuals most at risk, but possibly attack success would be reduced because of increased predator visibility and prey escape potential in the vertical plane, which remains to be tested.

The experimental design and sample sizes with regard to pattern of re-use of animals across multiple days (lines 544 ff.) might be easier to understand if the authors included a summary figure. How many groups of 40 shiners began each batch? There were 9 pike in batch 1 and 4 in batch 2, does that mean that the experiment began with 360 shiners in batch 1 and 160 in batch 2? Trials continued until there were not enough fish left for another trial with that number of exposures, which I think means until there were fewer than 40 fish. But only 70% of 125 attacks were successful (about 88). If the study started with 520 prey and only 88 were killed, it sounds like the study could go through a very large number of repeated trials before it would not have enough prey to continue. I am clearly not understanding something about the design. Pike had 3 rest days between trials. Was the same rest schedule applied to shiners? Shiners were kept in tanks with fish with the same number of previous exposures, but did this mix animals with different numbers of days since their previous trial? How many shiners participated in the study? There were a total of 125 attacks, with 1.84 attacks/trial which seems to mean 68 trials, spread across 13 pike in 2 batches. But 'a number of pike' never attacked any prey. Does the 13 pike include these or are the 13 pike only the ones that did attack prey?

We thank the reviewer for asking these detailed questions about our sample size and about the repeated testing of the fish. This made us realize we weren’t clear enough about these points. In total we started with around 1500 shiners and 20 pike. Out of these 7 pike never attacked and a number of them stopped attacking during later trials, resulting in us to exclude that trial and not use those pike and shiners for any further trials. Hence, the number of attacks we acquired decreased quickly with exposure number. Like the pike, shiners were also generally tested with three rest days between trials but sometimes groups had a couple fish that only had two rest days. This was inevitable due to the difficulty of just being able to run a few trials each day, and sometimes pike not attacking.

We have considerably rewritten the Experimental Procedure section to accommodate all of the reviewers questions and help better explain our sample size and retesting of animals across the experiment (lines 589-610): We ran the experiment with two batches of pike and shiners, with the fish of each batch tested repeatedly in the arena over time. Pike were randomly selected, but only after they had at least three rest days since the previous trial, which was decided based on pilot work that showed pike were less motivated when tested with less time between trials. For each trial, shiners were also randomly selected from the holding tanks to create groups of 40 fish. This group size reflects the size of shiner shoals observed in the wild (Hall et al., 1979; Krause et al., 2000a). Shiners were mixed to avoid potential effects of familiarity between individuals and the associated social feedback and learning effects related to group composition. However, as we used a repeated-measures design, we made sure that all shiners in the same group had the same number of exposures to the pike by keeping shiners in separate social holding tanks based on their number of pike exposures. In total we used 20 pike and started with about 1500 shiners. Pike had a mean number of 1.84 ± 0.14 attacks per trial, with 18 trials having more than 1 successful attack. However, not all pike did always attack during the trials, and 7 pike never attacked. As those trials did not provide any data on the attacks they were excluded and the pike not used in further trials. Also the shiners were excluded for subsequent trials as their future behaviour in the assay could be influenced by having had experience with a pike that did not attack. As a result our sample size decreased with exposure and we stopped at a maximum number of 6 exposures. Like the pike,shiners were also generally tested with three rest days between trials but sometimes groups had a couple fish that only had two rest days. This was inevitable due to the difficulty of just being able to run a few trials each day, and sometimes pike thus not attacking. Although none of the fish had any previous experience with any experiments, during the repeated exposures both shiners and pike had time to learn and anticipate about the situation, thus enabling us to also investigate how this impacted predators’ attack behaviour and prey response.

I will admit that I might not have outlined the paper carefully enough, but the statistical analyses described briefly in lines 193 ff., and in some detail in the Appendix involved reasonably standard model choice methods listing relative model weights, but in the main text, the authors also showed various other statistics: LRT chi-square values, F-values, with pvalues, R-squares, and correlation coefficients. Perhaps I missed where they described these, but if indeed they did not do so, they should describe all statistical tests.

We carefully checked all results and description of the analyses again and now state more clearly in the methods how all statistics were acquired in the few cases where this was still missing, on lines 703-706: To investigate how repeated exposure changed the shiners packing fraction and avoidance of the pike’s head and pikes’ distance from the targeted individual, we ran a linear model with exposure, fitted as a cubic function, as a fixed factor and data subsetted to pikes’ first attack attempt during the trials. And lines 709- 711: Thereby estimates, likelihood-ratio tests, and p-valuesreported in the text were acquired from the final models. In some cases we included Pearson correlations and linear models to further investigate the relationship and explained variance (R2 ) between two predictor variables.

Model choice analyses also allow one to assign relative weights not just to each model, but to each factor by summing the weights for each model in which that variable appears. Rather than only show the top 15 models (Appendix 3), it might be useful to calculate the weights for each factor. This would complement the results and conclusions in the main text.

For our analyses into what variables best predicted individual predation risk and survival we did just that: we selected the final predictive features based on both their relative importance score across all models, as the reviewer is suggesting, as well as by considering the features that appeared in the most parsimonious models. Plots of the relative weights of each factor are given in the main text (panel A in figures 4-6).

Perhaps I did not study the Appendices carefully enough, but the best listed models (Appendix 3 and 4) did not include number of exposures, pike ID or attack attempt, or more importantly, interactions between these random effects and the predictive features from either the prey or predator view. Appendix 2 shows that exposure number had large effects on pike distance from the group centroid and from the target individual. Did the more detailed features of targeted prey (or attack success) change with repeated exposures? For each exposure number (Appendix 2 - figure) the distances varied quite a lot. I assume that these are for the 13 different pike. Did individual pike differ consistently in their targeting tendencies, or even in their change in targeting tendencies with repeated exposure and learning? If so, this deserves more discussion.

In our models we always included exposure number, pike ID and attack attempt as random effects, as we stated in the original manuscript on lines 963-965. They are not features that help us better predict which individual may be targeted or survive an attack but factors that may influence the variation among trials. Because of that they are not included in the weighting of the models nor as any interaction with the fixed factors. We did not run further detailed analyses with repeated exposure to see how the importance of different predictive features may change with repeated exposure as this is not the focus of the present study and the amount of results we present is already quite dense, as Reviewer 1 also pointed out. We saw that when exposure was included as a random factor it played only a very minor role in explaining the variance among trials. Although the question if pike differed in their change in targeting with repeated exposure and learning is definitely interesting, this again goes beyond the focus of the presentstudy. A proper investigation of such questions about how predators may learn and adapt their attack tendencies with experience warrants a carefully designed study with that focus in mind.

Appendix 2, figure 1 -- what are the darker points in figures B and C?

Semi-transparent points indicate individual trials. Some points appear darker due to overlap. We now clarify this in the figure legend.

Line 193: 'were' not 'where'.

Author details

1. Jolle Wolter Jolles

   1. Department of Collective Behaviour, Max Planck Institute of Animal Behavior, Konstanz, Germany
   2. Zukunftskolleg, University of Konstanz, Konstanz, Germany
   3. Centre for Ecological Research and Forestry Applications (CREAF), Barcelona, Spain

   Contribution

   Conceptualization, Data curation, Software, Formal analysis, Supervision, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing

   For correspondence

   j.w.jolles@gmail.com

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-9905-2633

2. Matthew MG Sosna

   Department of Ecology and Evolutionary Biology, Princeton University, Princeton, United States

   Contribution

   Conceptualization, Resources, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Project administration, Writing – review and editing

   Competing interests

   No competing interests declared

3. Geoffrey PF Mazué

   School of Life and Environmental Sciences, University of Sydney, Sydney, Australia

   Contribution

   Investigation, Writing – review and editing

   Competing interests

   No competing interests declared

4. Colin R Twomey

   Department of Biology, University of Pennsylvania, Philadelphia, United States

   Contribution

   Software, Writing – review and editing

   Competing interests

   No competing interests declared

5. Joseph Bak-Coleman

   1. eScience Institute, University of Washington, Seattle, United States
   2. Center for an Informed Public, University of Washington, Seattle, United States

   Contribution

   Methodology, Writing – review and editing

   Competing interests

   No competing interests declared

6. Daniel I Rubenstein

   Department of Ecology and Evolutionary Biology, Princeton University, Princeton, United States

   Contribution

   Resources, Supervision, Writing – review and editing

   Competing interests

   No competing interests declared

7. Iain D Couzin

   1. Department of Collective Behaviour, Max Planck Institute of Animal Behavior, Konstanz, Germany
   2. Department of Biology, University of Konstanz, Konstanz, Germany
   3. Centre for the Advanced Study of Collective Behaviour, University of Konstanz, Konstanz, Germany

   Contribution

   Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing – review and editing

   For correspondence

   icouzin@ab.mpg.de

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-8556-4558

• 1,579

  Page views

• 444

  Downloads

• 4

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.