Simulated evolution of selfish herd behavior

https://doi.org/10.1016/j.jtbi.2004.11.035Get rights and content

Abstract

Single species aggregations are a commonly observed phenomenon. One potential explanation for these aggregations is provided by the selfish herd hypothesis, which states that aggregations result from individual efforts to reduce personnel predation risk at the expense of group-mates. Not all movement rules based on the selfish herd hypothesis are consistent with observed animal behavior. Previous work has shown that herd-like aggregations are not generated by movement rules limited to local interactions between nearest neighbors. Instead, rules generating realistic herds appear to require delocalized interactions. To date, it has been an open question whether or not the necessary delocalization can emerge from local interactions under natural selection. To address this question, we study an individual-based model with a single quantitative genetic trait that controls the influence of neighbors as a function of distance. The results indicate that predation-based selection can increase the influence of distant neighbors relative to near neighbors. Our results lend support for the idea that selfish herd behavior can arise from localized movement rules under natural selection.

Introduction

Gregarious behavior occurs in animals across a wide variety of taxa, including invertebrates (Ritz, 1994), fish (Shaw, 1970), birds (Lack, 1968), wildebeests (Gueron and Levin, 1993), lions (Bertram, 1975), and primates (Nakagawa, 1990), among others. The ubiquity of gregarious behavior across taxa implies that it has evolved independently at least several times (Williams, 1966), and therefore one might justifiably expect aggregations to confer strong advantages upon their members. Indeed, individual benefits may include improved reproductive success (Lack, 1968, Burger and Gochfeld, 1991), increased foraging success (Cody, 1971, Krebs et al., 1972), or an improved chance to survive predation (Hamilton, 1971, Vine, 1971, Watt et al., 1997). The last of these benefits, the lowering of predation risk, has received much attention over the years (Morton et al., 1994).

Social groups can decrease individual predation risk by (1) confusing the predator so that predator success declines (Hall et al., 1986, Smith and Warburton, 1992), (2) providing improved vigilance through many sets of eyes (Lima, 1995, Roberts, 1995), (3) providing a means of cooperative defense to “fight off” the predator (Wilson, 1975), (4) diluting risk so that the per-individual chance of being killed declines (Foster and Treherne, 1981), or (5) providing the individual with “cover” in the form of its group-mates (Hamilton, 1971). Which defense mechanism is employed depends on the size and mobility of the prey relative to that of the predator (Pulliam, 1973), and in some cases the mechanisms cannot operate simultaneously. For example, a very dense group might well dilute risk, while at the same time decreasing vigilance by blocking the line of sight for all but those on the periphery (Parrish et al., 2002).

An individual's ability to derive “cover” from its group-mates was famously proposed by Hamilton (1971). According to Hamilton's (1971) “selfish herd” hypothesis, animal groups form because individuals try to interpose their group-mates between themselves and potential danger. Hamilton (1971) assumed that the predator could appear anywhere on the field occupied by the prey, and would attack the nearest target. He defined a “Domain of Danger” for each prey individual as the area closer to that individual than to any other member of the herd. Because the predator simply attacks the closest target, if it first appears inside an individual's Domain of Danger, that individual will be attacked. Smaller Domains of Danger are therefore better than larger ones, and there should be a strong advantage to moving in such a way that an individual's Domain of Danger decreases relative to those of the individual's neighbors (Hamilton, 1971, Vine, 1971, Viscido et al., 2002).

The selfish herd hypothesis provides an explanation for why an individual would want to be in the midst of many group-mates when threatened with danger, but it does not explain how an individual would go about doing so. The mechanism Hamilton (1971) originally proposed—movement toward the nearest neighbor—does not achieve the desired positioning for the individual, nor does it result in a true aggregation forming (Morton et al., 1994, Viscido et al., 2002). Thus, a key question remains unanswered: What “movement rule” must the individuals follow to obtain a densely packed aggregation where most Domains of Danger are small? Viscido et al. (2002) proposed a rule called the “local crowded horizon” rule, where individuals would scan the horizon, mentally “tallying up” the positions of group-mates and giving more weight to nearby individuals than to those farther away. The local crowded horizon rule achieved the desired result: a densely packed aggregation formed, where most individuals had small Domains of Danger.

However, a movement rule such as the local crowded horizon rule, where individuals average the influence of many neighbors, is rather complex. We assume that such a rule would not suddenly spring into existence, any more than a complex organ such as the vertebrate eye would. Rather, we envision a rather simplistic starting condition, where animals pay very strong attention to their nearest neighbor, and position themselves relative to only that individual. From such a simple rule, is it possible that the more complicated situation—with individuals evaluating the position of dozens of neighbors—could arise by natural selection? To date, only one study has attempted to tackle this problem. Morton et al. (1994) conducted a simulated evolutionary game between individuals moving randomly and individuals moving toward their nearest neighbor, and showed a slight benefit to those using the latter rule. The difference was judged sufficient to provide a selective advantage (Morton et al., 1994), and so we have a beginning: how nearest-neighbor movement could arise from a random movement habit is established. However, as we noted above, nearest-neighbor movement does not really produce a cohesive group, such as a fish school (Shaw, 1970)—many more neighbors must be considered before large aggregations are observed (Viscido et al., 2002, Warburton and Lazarus, 1991).

In addition to the movement rule question, a further difficulty with the selfish herd hypothesis is that, by definition, not everyone in the herd benefits from the behavior. For each individual who manages to reduce his domain of danger, there is another individual, somewhere else in the group, whose predation risk increases as a result (Viscido, 2003). During a single predator attack, therefore, there will always be some proportion of the flock that benefits, and some proportion that is harmed, by the behavior. Since a predator only makes a single kill at a time, many individuals in a large domain of danger will survive a single attack, and the evolutionary question then becomes whether, over time, individuals benefit from the behavior more often than they are harmed. In other words, does the behavior allow one to reach a safe position more often than it leaves one at greater risk? If so then we would expect selfish herd behavior to be favored by natural selection; if not, then we would not expect it to arise. Because all previous models (Hamilton, 1971, Morton et al., 1994, Viscido et al., 2001, Viscido et al., 2002, James et al., 2004) have only looked at changes to individual risk during a single attack, the answer to this question remains unclear. We attempt to solve this problem by tracking populations over thousands of generations, through many attacks, and following the evolution of a “gene” for selfish herd behavior.

The goal of this study is to determine whether a population composed of individuals paying attention only to one or a few neighbors when deciding where to move could, through the process of natural selection, evolve into a population that pays attention to the entire group. We assume, for the sake of simplicity, that a single quantitative locus controls how strongly an individual is influenced by its neighbors in a parthenogenic population, and that there is a small amount of genetic variance with respect to this trait in the population. We further assume that the predator can appear anywhere and, upon doing so, will kill the nearest target. Given these conditions, we use individual-based simulations to test the hypothesis that the selective pressure provided by the predator will drive the “neighbor influence” gene such that individuals will consider increasing numbers of neighbors in their decisions.

Section snippets

Methods

We begin with a summary of our model. The model is individual-based, consisting of an asexual population and a series of iterations corresponding to predation events. At the start of each iteration, a group of individuals is randomly distributed on a surface. Each individual then moves across the surface for a fixed number of steps. At each step, individuals choose their movement direction using a weighted average of the bearings to all other group members. The weights are functions of the

Analysis of the influence genotype

To understand how changes in the “influence genotype” parameter β alter the macroscopic behaviors of individuals, consider a reference individual located at the origin. We define the absolute influence η of a neighboring individual at distance ρ from our reference individual in correspondence with Eq. (2.2) asη=11+(r/ρ)β,where ρ is the half-saturation level, r is distance, and β is the parameter experiencing selection. Suppose we have two neighbors C1 and C2 with respective distances r1 and r2

Discussion

Hamilton's selfish herd hypothesis is an a priori model for evolution (Maynard Smith, 1972). Like most a priori models, the selfish herd hypothesis was not formulated independent of observational data, can only explain a limited range of behavioral adaptation, and may be incorrect despite its elegance. There are many two-dimensional movement rules consistent with the selfish herd hypothesis, but only some of these lead to behaviors matching field observations. An important but unaddressed

Acknowledgements

The authors would like to thank two anonymous reviewers, the participants of the Mathematical Biology seminar series for helpful discussion, and T. Daniel for suggesting Eq. (2.3). M. Miller and M. Kot provided helpful critical comments. This work was supported in part by NSF VIGRE Grant DMS-9810726.

References (33)

  • S.V. Viscido et al.

    The dilemma of the selfish herdthe search for a realistic movement rule

    J. Theor. Biol.

    (July 2002)
  • K. Warburton et al.

    Tendency-distance models of social cohesion in animal groups

    J. Theor. Biol.

    (1991)
  • P.J. Watt et al.

    Toad tadpole aggregation behaviourevidence for a predator avoidance function

    Anim. Behav.

    (1997)
  • M. Abramowitz et al.

    Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables

    (1972)
  • C.R. Bertram

    Social factors influencing reproduction in wild lions

    J. Zool.

    (1975)
  • J. Burger et al.

    The Common Ternits Breeding Biology and Social Behavior

    (1991)
  • Cited by (50)

    • Cooperative predation on mutualistic prey communities

      2020, Journal of Theoretical Biology
      Citation Excerpt :

      Similar positive interaction in the form of mutualism may also be present in the prey population. Although prey mutualism is generally known to reduce the risk of predation via various mechanisms like group vigilance or dilution effect (Barnard and Thompson, 1985; Reluga and Viscido, 2005; Courchamp et al., 2008), it can also enable an additional per-capita growth independent of the predatory effect. Such evidences of possible indirect consumer-resource mutualism among terrestrial herbivores has been reported in an earlier study where removal of one of the species may result in low reproduction and growth in the associated species (Boucher et al., 1982).

    • Living in groups: Spatial-moment dynamics with neighbour-biased movements

      2020, Ecological Modelling
      Citation Excerpt :

      Stemming from this original idea, James et al. (2004) considered a model with greater biological realism, by incorporating a limited domain of danger, representing either a limited detection range or attack range of predators, that could be applied to animal groups of finite size. Further work by Reluga and Viscido (2005) pointed out that rules for generating realistic selfish herds need interactions beyond an individual's nearest neighbours, and showed how predation-based selection could increase the influence of distant neighbours. Other models explored animal aggregation behaviour by introducing sensory zones of individuals, for example zones of repulsion or attraction that drive animals towards or away from neighbouring individuals, giving rise to higher order structure in the population (Couzin et al., 2002; Wood and Ackland, 2007; Bode, 2011; Herbert-Read et al., 2011).

    • Collective movement in alarmed animals groups: A simple model with positional forces and a limited attention field

      2019, Physica A: Statistical Mechanics and its Applications
      Citation Excerpt :

      The second is that many different species should be able to act in accordance with it, from Crustacea [2] to Mammals [4]. Finally, a behavioural ecology perspective, suggests that the rule must be simple enough for allowing natural selection to favour and fix it [14]. Usually considered interactions, as for example molecular type forces, with a weighted sum of a large number of neighbours, clearly do not satisfy these assumptions.

    View all citing articles on Scopus
    View full text