The personality types of key catalytic individuals shape colonies' collective behaviour and success.

Behavioural ecologists often note that one or a few group members appear to shape the collective behaviour of social groups differentially. Our understanding of these keystone individuals is largely taken from meticulous field observations and semi-scientific anecdotes. In this study we experimentally test whether the behavioural tendencies of putative keystone individuals shift the collective behaviour of colonies using the social spider Stegodyphus dumicola. Prior studies on Stegodyphus demonstrated that the single best predictor of colonies' collective behaviour is the behaviour of colonies' boldest individual. Here, we probe the causal relationship between the traits of extremely bold individuals and colonies' collective behaviour by experimentally creating colonies of identical size and personality composition in the laboratory and then adding a single individual of varying boldness (the putative keystone individual). Experimentally adding just one extremely bold individual increased the foraging aggressiveness of entire colonies and altered the total mass gained by fellow group members, relative to the addition of a less bold individual. Additionally, our data suggest that bold individuals are capable of such influence because they catalyse variation in the behavioural tendencies of fellow group members.

Fastidious observers of animal societies often note that one or a few key individuals appear to exert an inordinately large influence over groups. Dominant males in lekking species (Ballard & Robel, 1974), alpha individuals in primate societies (Alberts, Sapolsky, & Altmann, 1992), queens in social insect colonies (Aron, Passera, & Keller, 1994), and superspreaders in disease transmission networks (Meyers, Pourbohloul, Newman, Skowronski, 2005) all share the common feature that they are thought to affect their groups more strongly than standard, more generic individuals. Here, we term these animals ‘keystone individuals’ or merely ‘keystones’, which, analogous to Paine's ‘keystone species’ concept (Paine, 1966, Paine, 1995), are defined as individuals that show an inordinately large influence over their social groups relative to their abundance (Sih et al., 2012, Sih and Watters, 2005). The ecological effects of these individuals vary wildly among study systems, and their mere presence can become major drivers of social groups' success (Modlmeier, Keiser, Watters, Sih, & Pruitt, 2014).

Although behavioural ecologists have often observed the effects of keystone individuals in situ (reviewed in Modlmeier et al., 2014), rigorous and experimental studies of these effects are rare (Flack et al., 2006, Flack et al., 2005). Instead, the evidence behind the role of keystones is more regularly treated as a series of semi-scientific anecdotes or idiosyncratic field observations (e.g. Pyle et al., 1999, Sapolsky and Share, 2004). This is, in part, because manipulating the presence/abundance of putative keystone individuals is intractable for the majority of study systems. For example, predicting which individuals will act as keystones can prove difficult (e.g. during outbreak of severe acute respiratory syndrome, SARS; Shen et al., 2004) and, in some cases, the addition/removal of keystone individuals is impossible (e.g. McComb et al., 2011). Consequently, we maintain a poor understanding of (1) how keystone individuals are maintained within populations, (2) the behavioural and physiological mechanisms by which they exert their influence over their fellow group members and (3) how variation in the traits of the keystone individuals themselves shift the collective behaviour or performance of their associated societies. Exploring these avenues is important for our understanding of animal societies because the data available, although quite limited, suggest that keystones have the potential to become powerful arbiters of collective behaviour and group success (e.g. McComb et al., 2011, Sih and Watters, 2005).

Social spiders, like those of the genus Stegodyphus, are a superb model with which to explore the effects of keystone individuals on collective behaviour. Social Stegodyphus live in cooperative foraging societies characterized by highly female-biased sex ratios, serial within-colony inbreeding and a high incidence of colony turnover (Aviles, 1997, Lubin and Bilde, 2007). Like all social spiders, the majority of colony maintenance tasks are performed by females. Shared colony maintenance tasks include cooperative prey capture, shared web maintenance and alloparental care.

In a previous field study on intercolony variation in collective behaviour, we demonstrated that social Stegodyphus colonies show stable, characteristic differences in their collective-foraging behaviour: some colonies attack prey rapidly with numerous attackers, whereas other colonies attack prey more slowly with fewer attackers (Keiser et al., 2014, Pruitt et al., 2013). Most notably, the single best predictor of intercolony variation is the behavioural tendency (or ‘personality’) of colonies' boldest individuals. Boldness, in this context, is defined as the latency for an individual to resume activity after an aversive threatening stimulus. Strikingly, this variable alone explains more than 60% of the naturally occurring variation in colonies' collective-foraging behaviour (Pruitt et al., 2013). This result is intriguing because it implies that the behavioural tendencies of these extremely bold individuals may somehow drive colonies' collective behaviour. Unfortunately, these correlative data were unable to verify a causal relationship between the presence of these extremely bold individuals and colonies' collective behaviour.

In this study, we investigate the presence of extremely bold individuals on colonies' collective behaviour by manipulating colony composition in the laboratory. Specifically, we asked the following questions. (1) Does the presence of a single extremely bold individual predictably change the collective-foraging behaviour of a colony composed of nonbold individuals? (2) Are extremely bold individuals more likely to engage in foraging behaviour than their nonbold colonymates? (3) How does the presence of an extremely bold individual change the behavioural tendencies of other group members? (4) Is colony performance (i.e. change in mass and colony members' survivorship) associated with the phenotypes of their boldest individuals? Together, our studies were designed to probe more deeply into the interplay of animal personality, keystone individuals and collective behaviour than has been achievable to date.

Methods

Study System

Stegodyphus dumicola (Araneae, Eresidae) is a patchily abundant spider throughout southwestern Africa living in colonies containing 1–2000 individuals (Bilde et al., 2007). Colonies' webs consist of two functionally distinct and connected structures: the capture web and the nest. The nest is composed of a dense three-dimensional matting of silk, dried leafs and prey carcasses, which together serve as the spiders' retreat during the day. In addition, webs contain one or more two-dimensional capture webs composed of sticky cribellate silk that extend out from the nest and serve to intercept prey (Henschel, 1998). When prey make contact with the capture web, one or more spiders emerge from the nest, locate the struggling prey and subdue them. Once subdued, a prey item is either dragged back to the nest and shared with other group members or consumed directly on the capture web.

The spiders used in these experiments were collected as mixture of adults and subadults in the Northern Cape, South Africa along the southern edge of the Kalahari Desert (28°26′S, 21°21′E, 894 m elevation). Colonies were collected by first disturbing the capture web, which resulted in the spiders retreating into their nest. The nest was then plucked off of its substrate in its entirety and placed within a chiffon pillow case. Colonies were transported back to nearby hostels/hotels where they were hand-sorted. The number of colony members and inquilines (i.e. heterospecific arthropods living as social parasites in the nest) were counted and colony members were placed together in a clear 490 ml plastic container for transport to Pittsburgh, PA, U.S.A. In Pittsburgh, we again hard-sorted through colonies, determined the boldness of each individual spider and maintained spiders individually in 2-ounce (6 ml) plastic cups. Animals were kept in isolation for 14 days prior to the experiments outlined herein. Spiders were fed a maintenance diet of one 2-week-old cricket, once per week. Water was provided by misting the webs with a water bottle every 3 weeks.

Procedural Overview

To observe whether the behavioural tendency of a single very bold individual is sufficient to change the collective behaviour of a colony, we created colonies of nine S. dumicola with identical low boldness scores (‘personality’ assay described below). To these colonies we then added one bolder individual from the same source colony (i.e. a putative keystone), determined randomly using statistical software, that varied in its degree of boldness depending on the colony. Some colonies received only a modestly bold individual, while others received a considerably bolder individual, and still other colonies received an extraordinarily bold individual. Care was taken not to mix spiders from multiple source colonies, and thus, relatedness among experimental colony constituents resembled that of naturally occurring colonies. Colonies of Stegodyphus are very highly inbred (r  ≈ 0.70) as a consequence of low dispersal among colonies and serial within-colony inbreeding (Smith, Van Rijn, Henschel, Bilde, & Lubin, 2009). The mass of each spider was measured prior to experimental combinations. A total of 36 experimental colonies were created in this way. We then split these colonies among two experimental treatments. In half of these colonies, we tested their collective behavioural response to an unrewarding vibratory stimulus daily for 5 days. In the other half of the colonies, we tested collective-foraging behaviour with an identical vibratory stimulus but, upon attacking the artificial stimulus, we rewarded spiders with a presubdued prey item. We then compared the relationship between the behavioural tendencies of colonies' boldest individuals and colonies' collective-foraging behaviour: the latency for the first spider to bite the vibratory stimulus and the total number of spiders that were out on the capture web during the time of attack. At the end of our assessment of colonies' collective-foraging behaviour, colonies were taken apart and the boldness of each individual was reassessed. To assess the effect of putative keystone individuals on colony performance, we compared the collective mass gain of the nine generic colony members over the duration of our collective-foraging assays (total mass after − total mass before). For this performance metric we restricted our analysis to colonies that received a rewarding prey stimulus.

Personality Assays

Boldness assays were designed to resemble the sensory cue of a rapidly approaching aerial predator (e.g. avian, chiropteran, wasp, etc.), which web-building spiders detect via a sudden, rapidly moving and directional jet of air (Jones et al., 2011, Riechert and Hedrick, 1993). The individual-based assay described below elicits a highly repeatable behavioural response, and is an informative indicator of individuals' behaviour in a social context and predicts survivorship under increased predation risk (Grinsted et al., 2013, Pruitt et al., 2013, Riechert and Hedrick, 1990). Moreover, this assay has been utilized in an impressive diversity of test systems (including >20 species of spider), and thus, allows for seamless comparisons across species using a consistent methodology (Niemela, DiRienzo, et al., 2012, Niemela, Vainikka, et al., 2012, Pruitt and Riechert, 2012).

Boldness assays were initiated by removing females from their home containers and placing them within a plastic enclosure (13.5 × 13 × 3.5 cm). Females were permitted 60 s to acclimate before applying two rapid jets of air to the dorsal anterior cephalothorax from approximately 5 cm away using an infant ear-cleaning bulb. This stimulus elicits a ‘huddle’ response from S. dumicola, where the spider will tuck its legs under the body and remain motionless. Anecdotally, this seems to be an innate response expressed by every spider we have tested. We then measured spiders’ latency to resume locomotion (i.e. moving at least one body length) following the huddle response over the next 10 min. Individuals with low latencies to resume movement were deemed to be more ‘bold’ than individuals with longer latencies. Individuals’ ‘boldness’ score were subsequently calculated by subtracting their latency to initiate movement from the total trial duration (i.e. 600 s). This calculation makes the presentation of our analyses and results more intuitive, because it ensures that larger numbers correspond to greater boldness.

Establishing Experimental Groups

We created 36 colonies of 10 spiders (nine nonbold, one bold) 3 days after an ad libitum feeding of 2-week-old crickets. Nine of the spiders in each colony showed the same boldness (i.e. latency: 598 ± 2 s) to resume movement following a predator cue; the remaining spider in each colony resumed movement in 2–568 s. Prior to placing the spiders in each colony, we removed each individual from isolation, weighed it and measured its cephalothorax width using imaging software on a Leica M80 light microscope. In our unrewarding stimulus procedure, keystone individuals could be differentiated from nonkeystones by a single blue acrylic paint dot marked atop their abdomen, whereas other colony members remained unmarked. In our rewarding stimulus procedure, all colony constituents were individually marked using a unique pair of coloured dots atop their cephalothorax, which does not appear to influence individuals' participation in foraging activities (Pruitt et al., 2013). Colonies were established in 490 ml clear plastic containers containing a lattice of poultry wiring to facilitate web construction. Colonies were then given 5 days to build a web before their collective-foraging behaviour was tested. Assignment of colonies to the rewarding or unrewarding stimulus treatment was performed haphazardly.

Collective-foraging Assays

Five days after establishing colonies, we initiated our assessments of their collective-foraging behaviour. Colonies' collective behaviour was assessed daily for 5 consecutive days. The order in which colonies were assayed was alternated across days to avoid effects of trial order. Trials were run between 0630 and 0930 hours each morning. Procedures follow those used by previous investigations on social Stegodyphus (Grinsted et al., 2013, Pruitt et al., 2013, Settepani et al., 2013).

Trials were initiated by removing the lids to the colony and placing a 1 × 1 cm piece of white computer paper in the capture web of the colony, approximately 4 cm away from the nearest entrance to the nest. We then provided 120 s of acclimation time before applying a standardized vibratory stimulus. We used a hand-held, battery-operated vibratory device (GoVibe) to generate the vibratory stimulus. A 2 mm wide aluminium alloy rod was wrapped around the base of the device and extended 6 cm beyond the tip. This rod was used to make contact with the 1 × 1 cm paper and resulted in the paper flittering quickly and erratically in the capture web (Grinsted et al., 2013, Keiser and Pruitt, 2014, Pruitt et al., 2013). For both the rewarding and nonrewarding stimulus, we then recorded the latency for the first individual to attack the paper (‘latency to attack’) and the number of attackers that responded to the stimulus by the time of attack (‘number of attackers’). For the unrewarding stimulus, the trials ended at this point.

For the rewarding stimulus trials, we placed a presubdued 3-week-old cricket adjacent to the paper square. In 68% (62/90) of trials, the cricket was seized by the initial attacker, in 21% (19/90) of trials, the cricket was seized by another attacker, and in the remaining trials, the cricket was ignored or the placement of the cricket initiated a flight response from the colony. At the end of the collective-feeding event assays, we isolated all colony constituents, weighed them and reassessed their boldness using the protocol described above.

Statistical Methods

To assess whether the boldness of colonies' boldest individuals influenced collective-foraging behaviour, we generated a generalized linear model. We included the reward treatment (rewarding/unrewarding stimulus), the day of the assay (1–5) and the boldness of the boldest individual as fixed effects. In addition to these effects, we included two interaction terms in our model: day*boldness of the boldest individual and day*boldness of the boldest individual*reward treatment. One model was created to predict colonies' latency to attack a prey, and a separate model was created to predict the total number of attackers that responded to the prey stimulus.

To assess whether colonies' boldest individuals engaged in foraging behaviour more than generic individuals, we used chi-square tests. We performed a single test independently for each day. For each test, we compared the total number of colonies in which the putative keystone individual was the first attacker against a null expectation of one out of every 10 colonies, since there were 10 spiders in each colony. To control for type I error, we used a Bonferroni-adjusted alpha of 0.01 for these analyses. Then, to assess how the participation of colonies' boldest individuals changed over time, we created a generalized linear model with the day of the assay (1–5) and the reward treatment as fixed effects. In addition to these effects, we included the interaction term day*reward treatment as a predictor in our model. Our binary response variable for this model was whether colonies' boldest individuals did or did not participate in prey capture (coded as 1 = attacked, 0 = did not attack).

To assess whether the presence of one extremely bold individual influenced within-colony behavioural variation at the end of our study, we generated a generalized linear model with the following predictor variables: boldness of the boldest individual, the reward treatment and their interaction term. For our response variable, we used the standard deviation in the boldness scores of the nine generic colony members taken at the end of our study. Note that colonies did not differ significantly in their within-colony behavioural variation at the start of our experiment (P  > 0.90).

To assess whether the boldness of colonies' boldest individuals influenced colony mass gain, we used a simple linear regression with the boldness of the boldest individual as the predictor variable and the collective change in mass of the nine generic individuals as our response variable. For this analysis, we restricted our focus to the rewarding stimulus treatment only. To assess whether the boldness of colonies' boldest individuals influenced survivorship of fellow colony members, we performed an ordinal logistic regression with the number of surviving generic group members (5–9) as our response variable, and the boldness of the boldest individual, the reward treatment and their interaction term as predictor variables.

Results

We detected highly significant associations between colonies' collective behaviour and the boldness of their boldest individuals (Table 1 ). In colonies in which the boldest individual was particularly bold, the boldest individual attacked prey four times as rapidly as the boldest individuals in colonies in which the boldest individuals were comparatively shy (Fig. 1 ). We also detected a highly significant interaction term between day and the boldness of the boldest individual, where the relationship between the behaviour of colonies' boldest individuals and their latency to attack only emerged on days 4 and 5 (Table 1, Fig. 1). In contrast, no significant association was detected for days 1–3 (all P  > 0.10). Colonies in which the boldest individual was particularly bold also tended to attack prey with more attackers (Table 1), and the number of attackers that responded to prey increased overall across the 5-day duration of the experiment. Although the interaction between day and the boldness of the boldest individual was not significant (P  = 0.0675), there was a tendency for colonies containing bolder individuals to have more individuals participate in the attack.

Table 1

Summary of the effects tests for a generalized linear model predicting colonies' latency to attack prey and total number of attackers that responded to the prey

Source	df	Estimate	SE	F ratio	P
Latency of attack
Reward treatment	1, 33	27.08	10.57	6.55	0.0152
Day	1, 141	−4.12	7.64	0.29	0.5907
Boldness of boldest individual	1, 33	−0.24	0.07	11.71	0.0017
Boldness of boldest individual*day	1, 141	−0.27	0.05	28.06	<0.0001
Boldness of boldest individual*day*reward treatment	1, 141	−0.05	0.05	0.87	0.3517
Number of attackers
Reward treatment	1, 33	−0.25	0.11	5.43	0.026
Day	1, 141	0.45	0.05	69.48	<0.0001
Boldness of boldest individual	1, 33	0.01	0.01	17.85	0.0002
Boldness of boldest individual*day	1, 141	0.09	0.05277	3.39	0.0675
Boldness of boldest individual*day*reward treatment	1, 141	<0.001	<0.001	0.03	0.8698

Figure 1

Relationship between the boldness score of colonies' boldest individuals and their latency to attack prey for either an unrewarding (left) or a rewarding (right) prey stimulus.

Our combined model predicting the participation of colonies' boldest individuals' in prey capture was highly significant (, R 2  = 0.24, P  < 0.0001). This result was driven by the large effect of day on the participation of colonies' boldest individuals (Table 2 ), where colonies' boldest individuals engaged in prey capture during 60–75% of prey capture events during days 1 and 2, but tapered off to only 10% of capture events on days 4 and 5 (Fig. 2 ). Because we failed to detect an effect of reward treatment on task participation of colonies’ boldest individuals (P  = 0.98; Fig. 2), we pooled the data from both treatments for our chi-square tests comparing the involvement of colonies’ boldest individuals against that of generic colony members. This procedure increased our statistical power. For day 1 (, P  < 0.0001) and day 2 (, P  < 0.0001), colonies' boldest individuals were far more likely to be the first attacker than were colonies' generic individuals. However, for day 3 (, P  = 0.011), day 4 (, P  = 0.97) and day 5 (, P  = 0.97), this bias disappeared. Note, however, that even for days 3–5, colonies' boldest individuals never engaged in less prey capture behaviour than would be expected by chance.

Table 2

Summary of the effects tests for a logistic regression model predicting whether colonies’ boldest group members were or were not the first individuals to attack prey

Source	Estimate	SE	df	L-R χ2	P
Reward treatment	−0.01	0.21	1	0.01	0.9897
Day	−1.12	0.35	4	50.53	<0.0001
Reward treatment*day	0.29	0.38	4	6.18	0.1863

Figure 2

Proportion of trials where colonies' boldest individuals were the first to attack prey for each day of the 5-day experiment.

Our combined model predicting within-group behavioural variation at the end of our 5-day trial was highly significant (F 3,32  = 10.66, R 2  = 0.45, P  < 0.0001; Fig. 3 ). The significance of this model was driven by the effects of reward treatment and the boldness of the boldest individual, where colonies in the rewarding treatment showed greater within-group behavioural variation at the end our experiments than did colonies in the unrewarding treatment (Table 3 , Fig. 3). Moreover, colonies in which the boldest colony members were particularly bold showed greater within-group behavioural variation in both treatments.

Figure 3

Relationship between the boldness of colonies' boldest individuals and within-colony variation in boldness at the end of the 5-day experiment for colonies that received an unrewarding (white dots, dashed line) or a rewarding (black dots, solid line) prey stimulus. The slopes of these two lines were not significantly different.

Table 3

Summary of the effects tests for a generalized linear model predicting within-colony behavioural variability in boldness at the end of the 5-day experiment

Source	Estimates	SE	df	F ratio	P
Reward treatment	39.77	10.26	1	15.0269	0.0005
Boldness of boldest individual	0.23	0.07	1	11.367	0.002
Reward treatment*boldness of boldest individual	0.06	0.07	1	0.9128	0.3465

Both aspects of colony performance investigated were associated with the boldness of colonies' boldest individuals. In our rewarding treatment, generic group members gained twice as much mass when they occupied colonies with an extremely bold individual (F 1,16  = 10.66, R 2  = 0.32, P  = 0.008; Fig. 4 ). Likewise, our combined model predicting the number of surviving generic colony members was significant (, R 2  = 0.08, P  = 0.028). Here again, colonies that contained one very bold individual boasted 40% higher survivorship than rival colonies (Table 4 ).

Figure 4

Relationship between the boldness score of colonies' boldest individuals and the total mass gained by all other group members over the 5-day experiment. Data are from the rewarding stimulus treatment only.

Table 4

Summary of the effects tests for an ordinal logistic regression model predicting the number of surviving individuals at the end of the 5-day experiment

Source	df	Estimate	SE	χ2	P
Reward treatment	1	0.04	0.34	0.01	0.91
Boldness of boldest individual	1	0.01	0.002	8.65	0.01
Reward treatment*boldness of boldest individual	1	−0.001	0.002	0.56	0.45

Discussion

Our results demonstrate that, regardless of whether the prey stimulus was rewarding or not, colonies that contained just one extremely bold individual were faster to attack prey and more individuals participated in foraging bouts. However, the relationship between the personality of these putative keystone individuals and collective behaviour only emerged at the end of our collective-foraging assays, and once the keystones (virtually) ceased participation. Interestingly, the influence of colonies' boldest individuals went beyond collective behaviour per se to affect the behavioural tendencies of fellow group members and colony-wide success. In particular, spiders within colonies that contained just one extremely bold individual showed greater behavioural variation at the end of our foraging assays. Unfortunately, the extent to which these heightened individual differences are repeatable across time remains untested. Finally, in our rewarding stimulus treatment, colonies in which the boldest individual was extremely bold gained more than three times as much mass and more colony members survived the duration of the experiment. Collectively, these data provide compelling evidence that the behavioural tendency of just one individual is capable of shifting an impressive assortment of colony attributes: the behaviour of fellow group members, colonies' collective-foraging behaviour and colony-wide performance.

Colony Success

Numerous studies on social insects, social spiders, and more than a handful of vertebrate models have demonstrated that both group composition and collective behaviour can be major determinants of individual fitness and group success (Dyer et al., 2009, Eldakar et al., 2009, Eldakar et al., 2010, Modlmeier and Foitzik, 2011, Modlmeier et al., 2012, Pradhan and van Schaik, 2008, Pruitt et al., 2012, Pruitt and Riechert, 2011a). Comparatively rarely, however, have such studies been designed to consider the possibility that just one group member can change the success of an entire social group (Flack et al., 2005). For S. dumicola, our results not only suggest that singularly bold individuals can augment colony success (i.e. colony member survivorship, collective mass gained), they also reveal that the influence of these individuals is dramatic. In this study, colonies that contained just one extremely bold individual gained three to four times more mass, on average, compared to colonies in which only one modestly bold spider was added. Spiders within these colonies also enjoyed 40% greater survivorship. In New World social spiders of genus Anelosimus, the efficiency with which a group feeds is the key determinant of collective mass gain. Here, groups that contain a mixture of aggressive and docile individuals are 40% more efficient at prey extraction than are monotypic groups, which leave considerable amounts of food uneaten (Pruitt et al., 2012, Pruitt and Riechert, 2009).

Our results are noteworthy because they may represent the most extreme form of social and/or frequency-dependent selection possible: a situation where the performance of an entire group is intimately linked to the behaviour of just one singularly influential individual. Under these circumstances, the behaviour of just one individual could dictate the success of an entire society, and hence, this individual could become a powerful arbiter of social selection. If this is also the case in situ, then this raises several questions. (1) How are individual differences in boldness determined? (2) Why do the average and extreme boldness scores of constituents differ between colonies, when colonies with bolder individuals seem to receive multiple performance benefits? (3) Do seemingly ‘generic’ group members adopt strategies to manipulate the boldness of the colony’s most extreme individuals, or to avoid detrimental colony compositions (Cote et al., 2011, Harcourt, Sweetman, et al., 2009)?

Collective Behaviour

In S. dumicola, the collective behaviour of an entire colony can change dramatically based on the behavioural tendency of only a single group member. In both our rewarding and unrewarding stimulus treatments, we detected an association between the behavioural tendencies of colonies' boldest individuals and two metrics of colony foraging behaviour: latency to attack and the total number of attackers.

Our original hypothesis for why bold individuals influence colonies' collective-foraging behaviour was that bold individuals merely act as foraging specialists. Our reasoning here was that whichever individuals perform a task would, by default, determine colonies' collective behaviour for that task. Consistent with this hypothesis, we found that bolder individuals were more likely to engage in prey capture both in the laboratory (Fig. 2) and in the field (Grinsted et al., 2013, Settepani et al., 2013). Similar instances of personality-dependent task participation have been observed in other systems as well, frequently involving boldness (Harcourt, Ang, et al., 2009, Kurvers et al., 2010, Le Vin et al., 2011). However, upon finer examination, our results here demonstrate that personality-dependent task participation seems insufficient to explain the influence that extremely bold individuals have on their colony. First, colonies containing just one very bold individual attacked prey with significantly more participants. In our view, this suggests that bold individuals are somehow capable of instigating the involvement of fellow group members, and thus, their influence extends beyond just their personal involvement in prey capture. Superficially, this relationship is akin to the excitatory effect of vibratory signals in honeybees, where ‘activator’ individuals deploy a vibratory signal that instigates and mobilizes the behaviour of fellow group members (Cao et al., 2007, Cao et al., 2009, Hyland et al., 2007). Second, the relationship between collective behaviour and the boldness of colonies' boldest individuals emerged only near the end of our foraging assays, and after the boldest individual had ceased participation. This delay in detectable influence also differentiates our results from normal leader–follower dynamics, where the influence of leaders on the following behaviour of group members is almost immediately recognizable and more ephemeral (Harcourt, Ang, et al., 2009, Webster and Ward, 2011).

Taken together, our results raise the question of how (mechanistically) singularly bold individuals change the collective behaviour of their colonies. In particular, we wonder whether the degree to which simple self-organizational rules versus central control underlie the observed patterns. At present, the prevailing theory from the social insect literature is that collective behaviour is largely organized by the actions of simple, functionally equivalent colony members that adhere to one or a few simple rules (Bonabeau et al., 1997, Mallon et al., 2001, Sumpter, 2006, Sumpter and Pratt, 2003). And this view is certainly sufficient to explain a diversity of impressive collective behaviours. Yet, based on our results, one wonders whether all group members, even those that are morphologically and physiologically similar, follow the same rules or to the same degree. And, we query, does allowing for consistent individual differences in the rules being followed qualitatively change the outcomes of our models of self-organization? Similar questions are now being put forth within the social insect literature (Jandt and Dornhaus, 2014, Jandt, Bengston, et al., 2014, Jeanson and Weidenmüller, 2013). Alternatively, our system may represent another example where the relative organizational and regulative input of keystone individuals is the central organizing unit for collective behaviours (e.g. similar to queens or keystone workers driving worker foraging in paper wasps; Jandt, Tibbetts, et al., 2014, Reeve and Gamboa, 1987).

Behavioural Variation and Division of Labour

One mechanism by which bold individuals may influence the behaviour and performance of their colony is by shifting the behavioural tendencies of fellow group members. As demonstrated in a suite of laboratory and field studies, personality types often play an important organizational role in spider and insect societies, shaping individuals' propensity to perform various tasks, their aptitudes/efficiencies for those tasks and colony success. Thus, at least for arthropod societies, any factors that shape the behavioural tendencies of colony constituents could also have implications for colonies' organization and success. Parallel arguments have been made for the distribution of worker size classes in social insects (Billick and Carter, 2007, Jandt and Dornhaus, 2014, Porter and Tschinkel, 1985). In the present study, colonies that contained one very bold individual tended to show greater interindividual variation in boldness at the end of our foraging assays, regardless of whether the prey stimulus was rewarding or unrewarding. This is in contrast to groups that contained only a moderately bold individual, which showed a more compressed distribution of personality types at the end of our colony foraging assays. And importantly, consistent individual differences in behaviour have been linked to task differentiation in three species of social spider, including S. dumicola (Pruitt and Riechert, 2011a, Pruitt and Riechert, 2011b, Grinsted et al., 2013). Although these results are not absolute, they could provide some clues to how extremely bold individuals exact their influence over colonies' collective behaviour. Specifically, we propose that extremely bold individuals play a sort of catalytic role in the development and persistence of varied personality types within their groups. This, in turn, could influence how task participation is organized among colony constituents. After all, virtually by definition, within-colony behavioural variation is the necessary precondition for division of labour and higher social organization (Beshers & Fewell, 2001). Admittedly, at present we cannot confirm whether the observed shifts in intracolony behavioural variation actually reflect consistent individual differences in behaviour (i.e. if the shifts are long-lasting); however, numerous studies from our laboratory have demonstrated that individuals’ boldness measures are highly repeatable under a diversity of social circumstances (Keiser et al., 2014, Pruitt et al., 2013).

One potential explanation for our results is that the observed shifts in within-colony behavioural variation and collective behaviour are a consequence of differential access to food. If bold individuals are behaviourally dominant, as they often are, then their participation in foraging may have initially restricted shyer subordinate individuals’ access to food early in the trials. However, our results are contrary to this view, since the relationship between the boldness of colonies’ boldest individuals and collective aggressiveness in later trials was identical in colonies that received a rewarding or an unrewarding prey stimulus (Fig. 1, Table 1, Table 2, Table 3, Table 4). Moreover, individual differences in boldness are not associated with body condition or recent feeding history in S. dumicola (Keiser et al., 2014, Keiser et al., in press). Consequently, however intuitive, differential food restriction of shy individuals and satiation of bold individuals does not appear to be a viable explanation for the results herein. Further studies that inspect both task allocation and division of labour in the presence versus absence of catalytic keystone individuals will help to clarify alternative views.

Conclusions

The idea that singularly influential individuals are capable of shifting group dynamics has been posed for a diverse set of social systems, ranging from social insects to cetaceans. By finely manipulating the phenotypic composition of laboratory colonies of the spider S. dumicola, we provide here some of the first experimental evidence that the presence of just one very bold individual is sufficient to change colonies’ collective behaviour and performance. These findings are notable because experimental data sets on the effects of keystone individuals are rare. Moreover, among the limited number of data sets available, it is not clear how variation in the behaviour of the keystones themselves influences either the magnitude or the nature of the effects that they exert over their groups. In contrast, in this study, we were able to associate variation in the behavioural tendencies of putative keystones with the effects that they have on colony attributes, at least in the laboratory. In S. dumicola, the presence of very bold individuals was associated with an increase in behavioural variation in their fellow group members and an increase in colonies’ collective aggressiveness during staged foraging bouts. Finally, when groups were permitted to feed collectively on prey items, groups that contained extremely bold individuals gained more mass than other colonies and enjoyed higher individual survivorship. Although the effects of very bold individuals on colony dynamics in the field are yet unknown, our results suggest that individuals showing rare, extreme phenotypes could be powerful arbiters of social selection and group success in both this and other systems.