Revisiting the social brain hypothesis: contest duration depends on loser’s brain size

Background Brain size is expected to evolve by a balance between cognitive benefits and energetic costs. Several influential hypotheses have suggested that large brains may be especially beneficial in social contexts. Group living and competition may pose unique cognitive challenges to individuals and favor the evolution of increased cognitive ability. Evidence comes from comparative studies on the link between social complexity and brain morphology, but the strength of empirical support has recently been challenged. In addition, the behavioral mechanisms that would link cognitive ability to sociality are rarely studied. Here we take an alternative approach and investigate experimentally how brain size can relate to the social competence of individuals within species, a problem that so far has remained unresolved. We use the unique guppy brain size selection line model system to evaluate whether large brains are advantageous by allowing individuals to better assess their performance in a social contest situation. Based on theoretical literature, we predict that contest duration should depend on the brain size of the loser, as it is the capitulation of the losing individual that ends the fight. Results First, we show that studying the movement of competitors during contests allows for precise estimation of the dominance timeline in guppies, even when overt aggression is typically one-sided and delayed. Second, we staged contests between pairs of male that had been artificially selected for large and small relative brain size, with demonstrated differences in cognitive ability. We show that dominance was established much earlier in contests with large-brained losers, whereas the brain size of the winner had no effect. Following our prediction, large-brained individuals gave up more quickly when they were going to lose. Conclusions These results suggest that large-brained individuals assess their performance in contests better and that social competence indeed can depend on brain size. Conflict resolution may therefore be an important behavioral mechanism behind macro-evolutionary patterns between sociality and brain size. Since conflict is ubiquitous among group-living animals, the possible effects of the social environment on the evolution of cognition may be more broadly applicable than previously thought.

Brain size is expected to evolve by a balance between cognitive benefits and energetic costs. Several 23 influential hypotheses have suggested that large brains may be especially beneficial in social 24 contexts. Group living and competition may pose unique cognitive challenges to individuals and favor 25 the evolution of increased cognitive ability. Evidence comes from comparative studies on the link 26 between social complexity and brain morphology, but the strength of empirical support has recently 27 been challenged. In addition, the behavioral mechanisms that would link cognitive ability to sociality 28 are rarely studied. Here we take an alternative approach and investigate experimentally how brain 29 size can relate to the social competence of individuals within species, a problem that so far has 30 remained unresolved. We use the unique guppy brain size selection line model system to evaluate 31 whether large brains are advantageous by allowing individuals to better assess their performance in 32 a social contest situation. Based on theoretical literature, we predict that contest duration should 33 depend on the brain size of the loser, as it is the capitulation of the losing individual that ends the 34 fight. 35

36
First, we show that studying the movement of competitors during contests allows for precise 37 estimation of the dominance timeline in guppies, even when overt aggression is typically one-sided 38 and delayed. Second, we staged contests between pairs of male that had been artificially selected for 39 large and small relative brain size, with demonstrated differences in cognitive ability. We show that 40 dominance was established much earlier in contests with large-brained losers, whereas the brain size 41 size. Since conflict is ubiquitous among group-living animals, the possible effects of the social 48 environment on the evolution of cognition may be more broadly applicable than previously thought. 49 50 Keywords: brain evolution, social competence, contest behavior, guppy, social brain hypothesis 51 52 Background 53 Brain size, relative to body size, shows a fascinating amount of variation among vertebrates 54 [1]. In the past decades, there has been great interest in how this variation could be related to the 55 social environment in which species live. This idea has intuitive appeal, as dealing with conspecifics 56 who themselves are cognitive actors may be challenging [2]. Several related hypotheses have been 57 constructed around this central premise where each places a focus on different aspects of social 58 living, such as cunning minds of social deception in the Machiavellian intelligence hypothesis [3], pair 59 bonding in the social brain hypothesis [4] or social learning in the cultural brain hypothesis [5]. While 60 recent publications [6-10] debate the strength of comparative evidence linking sociality and brain 61 size in a macro-evolutionary context, a rigorous way forward is to conduct experimental studies at 62 the micro-evolutionary level that directly test putative mechanisms and social benefits of brain size. 63 In sum, comparative studies provide insight via large-scale patterns and have generated strong 64 hypotheses, yet lack the ability to demonstrate causality and thus experiments are needed to 65 investigate putative mechanisms. 66 Conflict is a central aspect of social behavior in most animals and competitors often incur 67 large costs in the contest over dominance. These costs, including physical costs such as attack 68 van der Bijl et al. -Contest duration depends on loser's brain size 4 damage as well as time costs and vulnerability to predation, can be reduced when contests are 69 settled quickly, which should be facilitated by fast assessment of the expected outcome of the 70 contest [11]. Cognitive ability is thus likely important in animal contests [12]. Here we test this 71 recently posited hypothesis [13] that enlarged brains may provide an advantage by allowing for 72 faster assessment during contest. More specifically, based on theoretical models of animal contests 73 [14,15] we hypothesized that the duration of a contest should decrease with the brain size of the 74 loser when brain size is associated with cognitive ability [13]. This follows from the notion that the 75 conflict is settled with the decision of the loser to abandon the fight, and it is therefore the loser's 76 assessment that affects contest duration [16]. The assessment ability of the winner should therefore 77 be inconsequential. While assessment mechanisms during contests can be simple [17], such as giving 78 up when physical exertion reaches a threshold or using direct strength comparisons during physical 79 contact, conflicts in many species can be resolved without overt aggression and physical attacks. In 80 contests where resolution is also not aided by display behavior and ritualization, such as the roaring 81 and parallel walks of red deer stags [18], cognitive ability is especially likely to be important in 82 assessment. Here, we used the guppy (Poecilia reticulata) as a model to study the role of brain size in 83 behavior during conflict. We first established, combining classic and novel approaches, how 84 dominance is formed in this system, and then we compared the assessment speed between guppies 85 with small and large brains, which differ in cognitive ability [19,20]. 86

87
Guppy dominance is established before overt aggression 88 In the first experiment we hosted 16 contests between pairs of male guppies, establishing 89 residency for both fish by acclimating them overnight in two separated sides of a tank with an 90 opaque divider. In the morning, the divider was removed, and the males were free to interact for two 91 hours while being recorded from an overhead camera. Long recordings were necessary, as guppies 92 are known to show little to no aggression in the first hour of contact [21]. We kept track of the 93 van der Bijl et al. -Contest duration depends on loser's brain size 5 individual identities using computer vision [22] and scored all overt aggressive interactions (attacks) 94 in these contests, defined as chases, lunges and bites. We found that the attacks within pairs were 95 typically extremely one-sided, with on average 90% of the attacks performed by one of the two 96 individuals. More importantly, attacks were one-sided at the onset of aggression as there was no 97 discernable change in the identity of the aggressor over the duration of the trial, either measured as 98 time or attack number (binomial GLMM with random intercept and slope per trial; time: z = -0.017, p 99 = 0.987; attack number: z = -0.020, p = 0.984). This contrasts with classic theoretical concepts of 100 competitive situations where opponents exchange blows while collecting information [14]. Instead, 101 aggression was one-sided and invariable with time, and so assessment must have been occurring 102 before the onset of attacks. 103 A dominance relation may be apparent from other behavioral cues than overt aggression. 104 The relative movement between individuals is often heavily informed by social status, where 105 subordinates will avoid dominant individuals but not vice versa. This asymmetry in displacement has 106 a long history in being used to assess dominance in ethological studies of e.g. children   Our experimental design mimicked a scenario of random encounters, and therefore consisted of 157 small-brained versus small-brained, large-brained versus large-brained and small-brained versus 158 large-brained, with 18, 18 and 36 contests respectively. This full factorial design allowed us to 159 disentangle brain size effects of winners and losers. Within the contests of mixed brain size, small-160 and large-brained competitors were equally likely to win (19 vs 17, binomial test: 53%, p = 0.87). the goal of this study is to investigate brain size effects, we elected to remain agnostic to what 185 variables should be important and used a broad model selection approach. Additionally, we included 186 total displacement (as an estimate of total social activity) as a covariate to control for possible effects 187 of aggressive activity. We fitted linear mixed models for different combinations of predictors, 188 excluding models that contained the winner, loser and difference trait value of the same trait (since 189 these models are rank deficient), and models with more than six terms (to reduce overfitting). We 190 controlled for replicate selection line by including random intercepts for the interaction between the brain size of the winner and the brain size of the loser, nested within replicate (lme4 syntax: (1 | 192 replicate / (brain_size_winner : brain_size_loser)). 193 We found considerable model uncertainty, with 43 models that were within 4 AICc of the 194 best model (table S1). Brain size of the loser was the most important variable and present in all 195 models, but the inclusion of covariates varied. To take this covariate uncertainty into account we 196 averaged coefficients over the 43 models, using full model averaging [33]. In accordance with our 197 hypothesis, contests with large-brained losers lasted on average 36 minutes (or 34%) shorter than 198 contests with small-brained losers (SEadjusted = 13.52, z = 2.66, p = 0.008, figure 3, table S2). The effect 199 of loser brain size was robust against making different choices in cut-off values (figure S1). The brain Our results show that contest duration in the guppy depends on the brain size of the loser 221 but not on the brain size of the winner. This demonstrates experimentally that brain size can affect 222 the performance in a social task. Since dominance and contests are common among social animals, 223 cognitive variation in assessment ability has the potential to be important across many taxa. Indeed, 224 a recent study [34] showed that the amount of agonism in primate groups is positively associated 225 with brain and neocortex size, independent of group size. unclear to what extent overt aggression occurs in the wild, which has led some to suggest that male-231 male competition is perhaps not very important in this species [35]. That dominance is usually 232 established before overt aggression, however, suggests that the absence of clearly aggressive 233 behavior should not be interpreted as a lack of social roles since subordinate males may simply 234 escape after dominance has become established. This means that male-male competition may play a 235 more prominent role in the guppy than previously thought, with potentially important consequences 236 for the study of this model system in aspects of mate choice and sexual selection. 237 Since contest duration could not be established by the classical method, i.e. simple 238 quantification of discrete behaviors such as attacks, we developed a novel method of tracking the 239 development of social relationships. By formalizing the well-established concept of displacement in a 240 strongly quantitative and continuous form using modern computer vision tools, we could analyze the 241 establishment of dominance in detail. Since in many animals overt aggression is not necessary for 242 hierarchies to form, similar approaches could prove valuable to the broader study of contests and 243 dominance. 244 Interestingly, in addition to the assessment during conflict discussed here, we have 245 previously found guppies from the selection lines to perform better in assessment of potential mates 246 [26,30], as well as of the threat of predation [28, 29]. These differences were not due to differences 247 in color perception [30] or visual acuity [36], making it unlikely that behavioral differences stem from 248 alterations to low-level perceptual systems. Instead, volumetric increases in both the optic tectum 249 and telencephalon [37] could have led to better integration of visual information and/or decision 250 making in these contexts. 251

252
There is an ongoing debate about macro-evolutionary correlations between measures of 253 social complexity and brain size [8]. Till now, studies have focused on wide categorization of sociality 254 such as social system, group size or social repertoire size, with speculations about mechanisms. Here 255 van der Bijl et al. -Contest duration depends on loser's brain size 14 we presented results from an experimental assessment of the link between brain size and 256 assessment during contest, confirming our hypothesis the brain size can aid in conflict resolution. We 257 provide a novel and complementary approach to the study of social cognitive evolution, where the 258 experimental study of brain size effects on social cognition within species can generate bottom-up 259 predictions to explain inter-specific variation in brain size. We hope to see further investigations into 260 the interplay between cognition, brain size and these types of social interactions in the future. This is 261 necessary to provide exhaustive testing of the intuitively attractive idea of a link between sociality 262 and the brain [2, 4]. 263

264
Experimental set-up 265 The first experiment used sexually mature male wildtype guppies (Poecilia reticulata). All 266 individuals came from long-term lab populations whose founders were imported from Trinidad in 267 1998 and since then kept in large populations (>500 individuals) to avoid genetic bottlenecks. 268 The second experiment used fifth generation males from an artificial selection experiment on 269 relative brain size [19]. The initial selection experiment was performed by using indirect selection; in 270 short, random pairs were formed from the same population we used in experiment 1, and allowed to 271 breed. After reproduction, the parents were sacrificed and their brains were weighed. The offspring 272 from parents that had the 20% largest and 20% smallest combined relative brain size (residuals from 273 a regression of brain weight on standard length) were then used to start the "up" and "down" 274 selected lines. This process was repeated three times to form the three independent replicates. For a 275 further 2 generations, the top and bottom 20% was used for further selection. The fourth and fifth 276 generations were bred randomly to maintain the selected populations. For more details see [19]. The contests 285 The behavioral experimental procedure was identical in both tasks, with the exception that 286 we recorded contests for two hours in experiment 1, and three hours in experiment 2. Contestants 287 were chosen randomly from non-adjacent tanks, so that there had been no previous visual contact. 288 In the late afternoon, competitors were placed in opposite ends of an acrylic experimental tank, 289 measuring 15x15x40 cm with 2 liters of water (3.3 cm water depth). The shallow water facilitated 290 accurate tracking and identification. In the center of two long walls, U-shaped brackets were glued to 291 the tank between which a wall could be tightly fit, creating two 15x20 cm compartments with no 292 contact between the fish. The walls of the tank were covered with white plastic adhesive material 293 and placed on a white plate to maximize contrast and aid visibility in overhead recordings. Both 294 compartments were aerated during the night. The following morning, the aeration and wall were 295 removed, and the fish were free to interact. After the end of the trial, the animals were separated 296 again by placing back the wall. 297 All competitors were taken randomly from their home tanks. We maximized the 298 independence of the individuals by sampling the 144 competitors from 89 difference source tanks. 299 We randomized the brain size and replicates between days, test tanks and starting side, by running 6 300 batches of 12 combinations (3 replicates and four brain size combinations) and completely 301 randomizing within each batch. All contests were between males from the same replicate selection 302 line. We hosted 4 contests each day. 303 The experiment was done blindly to the brain size of competitors, as the tanks of males were 304 marked with running numbers as soon as they were caught from their source tank. The experimenter 305 van der Bijl et al. -Contest duration depends on loser's brain size 16 remained blind during the moving of fish, the start and end of the trial, the measuring of 306 competitors, the analysis of photographs, and the tracking of the videos. All contest durations were 307 assigned algorithmically. 308 Competitor traits 309 Directly after the trials, the competitors were photographed, their size was measured using 310 digital calipers to the nearest 0.01 mm and weighed to the nearest 0.0001 gram. Both standard 311 length (from the tip of the snout to the end of the caudal peduncle) and total length (from the tip of 312 the snout to the most caudal part of the tail fin) was measured, and tail size was calculated as their 313 difference. Both sizes and weights were measured in duplo and averaged. Coloration was measured 314 by taking a photograph of the fishes left flank using a Canon EOS 1200, and then manually outlined 315 the color patches in paint.NET and taking their area in pixels. We then calculated % body area by 316 comparing it to the total area of the left flank in pixels. We separately quantified black, orange, 317 yellow and iridescent coloration, but collated the latter three into a single "bright coloration" 318 variable to reduce the number of variables in the analysis and thus avoid over-parameterization. We 319 did not obtain pictures for two males, which reduced our sample size to 70 trials. 320 Identification and tracking of animals 321 We started an overhead recording just before removing the opaque barrier and ended the 322 recording after placing the barrier back. We then used idTracker [22] to keep consistent identities 323 during the trial, and noted afterwards whether the fish had returned to their starting compartment 324 or had been switched, in order to correctly assign the size, weight and coloration measures 325 afterwards. In order to also obtain information about orientation and shape measurements in the 326 tracks, we also tracked the video's with Ctrax [38]. We matched the identities of idTracker to the 327 detailed tracks of Ctrax by minimizing the squared errors. This allowed to also use the double 328 tracking to automatically correct errors, by marking frames where idTracker did not corroborate the Ctrax results as missing. Leveraging the strengths of both algorithms, we obtained tracks with a 330 median completeness of 100% and 97% in experiments 1 and 2 respectively. 331 Overt aggression 332 We quantified overt aggression for experiment 1 by manually reviewing the two hours in 333 each of the 16 videos and recording the time of each attack, as well as the identity of the aggressor. 334 We scored bites and lunges, two behaviors that both are based on aggressive fast approaches but 335 while contact was made between the individuals during a 'bite', no contact was observed during a 336 'lunge'. Additionally, we identified chases where one individual closely and rapidly followed the other 337 over at least several centimeters. 338

339
We calculated the displacement metric for both individuals separately. Speed v was taken as 340 the number of pixels the centroid had moved since the last frame. Distance d was taken as the 341 distance from the front of the body (the ellipsoid shape estimated by Ctrax) to the nearest point on 342 the body of the competitor. Distance from fish a to b is therefore not necessarily the same as the 343 distance from fish b to a. Finally, we determined the social angle φ as the difference in angle 344 between the orientation of the focal fish (i.e. not the heading, but the physical orientation as 345 estimated by Ctrax) and the angle towards the centroid of the competitor. Its cosine is therefore 1 346 when oriented towards the competitor, and -1 when facing away. We then combined the three 347 variables in a displacement score as displacement = (log(v + 1) * cos(φ)) / d. Speed was log 348 transformed and we added 1 to preserve its sign. The displacement metric is strongly positive when 349 the focal fish is quickly swimming towards the opponent at close distance, and strongly negative 350 when it's doing the same away from the opponent. 351

352
To determine the contest duration for a trial, we calculated the displacement statistic for 353 each individual across three hours. For each point in time we then evaluated the past displacement