Changes driven by the evolution of higher group polarization in the guppy are consistent across different predation pressures and associated with neuroanatomical changes

One of the most spectacular displays of social behavior is the synchronized movements that many animal groups perform to travel, forage and escape from predators. However, elucidating the neural mechanisms underlying the evolution of collective behaviors, as well as their fitness effects, remains challenging. Here, we study anti-predator behavior in guppies experimentally selected for divergence in polarization, an important behavioral aspect of coordinated movement. We find that groups from artificially selected lines remain more polarized than control groups in the presence of a threat. Neuroanatomical measurements of polarization-selected individuals indicated changes in brain regions previously suggested to be important regulators of perception, fear and attention, and motor response. Additional visual acuity and temporal resolution tests performed in polarization-selected and control individuals indicate that observed differences in anti-predator and schooling behavior should not be attributable to changes in visual perception, but rather are more likely the result of the more efficient relay of sensory input in the brain of polarization-selected fish. Our findings highlight that brain morphology may play a fundamental role in the evolution of coordinated movement and anti-predator behavior.


71
Animals regularly gather -for safety, for exploiting resources, or for mating. Group-living 72 often leads to spectacular forms of collective behavior, and individuals in many taxa coordinate 73 their movements in order to increase efficiency in foraging and travelling, or to confuse 74 predators 1 . Collective motion has evolved multiple times in fishes and is widely understood as 75 a behavioral adaptation aimed at reducing the risk of predation 2 . This behavioral adaptation is 76 underpinned by the efficient acquisition of information from external cues through visual and 77 lateral line sensory systems 3-6 . To date, we have a detailed understanding that coordinated 78 movements in animal groups likely emerge from decision rules that individuals use to interact 79 in groups e.g. 7,8 . Similarly, correlation-based analyses have revealed how predation levels are 80 associated with variation in collective motion in wild populations (see for instance 9 ). Yet, the 81 causes of collective motion are still unclear, particularly how evolutionary changes in collective 82 motion contribute to anti-predator specific situations, or what type of sensory information and 83 information processing schooling fishes use to identify and avoid predators as groups. 84 85 The brain, as the central organ controlling locomotion, sensory systems and decision-making, 86 Table 1, Fig. 1A). Previously observed differences between polarization-selected and control 161 groups in these traits were still present in the presence of these stimuli in the experimental arena 162 with approximate differences of 7% (Supplementary Table 2 To further characterize potential differences in how information spreads in polarization-192 selected and control fish groups we assessed the effect of group speed in the alignment of fish 193 when exposed to a predation threat. For this, we generated heatmaps showing the correlation 194 of group speed and group polarization across treatments for our different selection lines (Fig.  195 2). Heatmaps indicated that group polarization of polarization-selected fish in OFTs was 196 constantly close to maximal values of one and differences with control groups were partially 197 independent of the speed of the group (despite larger values of median speed observed in 198 relation to those in control groups). Overall group polarization reduction in the presence of a 199 novel object and a predator model in relation to OFTs were likely associated with a strong 200 reduction in the speed of the group, including periods of no movement of the group. Periods of 201 no movement in response to a threat were observed more frequently in control groups 202 suggesting that they might be an important driver of group polarization differences observed 203 between polarization-selected and control groups. In addition to differences driven by lack of 204 movement, correlations of group polarization and speed indicated that differences between 205 polarization-selected and control groups observed during collective motion resembling motion 206 in OFT's (mean group speed range 2 -5 mm/frame) were maintained in the presence of stimuli 207 in the experimental arena (Fig. 2). 208 209

Predator inspection behavior 210
Group polarization spatial patterns and its relationship to swimming speed suggested 211 differences between polarization-selected and control fish in their attention to stimuli presented 212 in the experimental arena. As such, we quantified inspection behavior of individuals in our 213 groups of eight fish in the presence of a predator model. We scored recorded videos for the 214 start and end point for each predator inspection performed by one randomly selected fish in 215 each video. Analyses of predator inspection data showed that females from control lines 216 presented a higher tendency to inspect the predation threat presented in the experimental arena 217 than polarization-selected females. Specifically, we observed that polarization-selected 218 females spent 21% less time inspecting the predator model and the mean duration of predator 219 inspections were 18% shorter in polarization-selected females (GLMMtime_inspecting: Incidence 220 Analysis of positional data and median distance to the stimulus presented in our assays 228 suggested that most inspection behaviors to the predator model were performed during the 229 initial three minutes of the assays (Supplementary Figure 1). Further, the majority of 230 inspections were performed at a range <200 mm and in the tail area of the predator model 231 presented in the experimental arena ( Fig. 1A; Fig. 3A). Consequently, we filtered our data to 232 evaluate collective motion patterns of fish groups in the time and locations where predator 233 inspections were performed during our assays (see methods). The overall differences in group 234 polarization found between polarization-selected and control groups were maintained in areas 235 within 200 mm of the predator model ( behavior is mainly performed from areas with reduced risk of attack from a predator. In line 238 with such expectation, we found that polarization of all groups was greatly reduced in the area 239 of the predator model tail (Fig. 3B). However, we found no differences in group polarization 240 between selected and control females in the head area of the predator model, but a maintenance 241 of 8% differences in group polarization between polarization-selected and control females in 242 close proximity to the tail of the predator model (LMMpolarization We used microcomputed tomography (micro-CT) to reconstruct the brain anatomy of 13 249 polarization-selected females, and 15 control females and determine overall brain volume and 250 the volumes of 11 major brain regions that could be safely identified in these brain scans 251 covering the whole brain volume: olfactory bulbs, ventral telencephalon, dorsal telencephalon, 252 thalamus, hypothalamus, nucleus glomerulus, torus semicircularis, optic tectum cup, central 253 optic tectum, cerebellum, and medulla oblongata (see methods). We next used this data set to 254 evaluate potential associations between collective behavior and neuroanatomy. Polarization-255 selected and control fish showed no differences in relative brain size (whole brain volume in 256 relation to their body size; LMMrelativebrain: t = -0.41, df = 23.29, p = 0.682; Fig. 4a). However, 257 analyses of relative brain region size (volume of each region in relation to the volume of the 258 rest of the brain) indicated that the thalamus and optic tectum cups are approximately 7% and 259 4% larger in polarization-selected than control females respectively (LMMthalamus: Odds Ratio 260 (OR) polarization = 0.929 (0.866-0.998); t = 2.187, df = 25, p = 0.038; LMMo.tectum: Odds Ratio 261 (OR) polarization = 0.959 (0.924-0.994); t = 2.409, df = 23.09, p = 0.024; Fig. 4a), and the medulla 262 oblongata is an 8% larger in control females (LMMmedulla: Odds Ratio (OR) polarization =1.08 263 (1.01-1.14); t = -2.65, df = 23.91, p = 0.013; Fig. 4a). All other eight brain regions measured 264 presented no differences between polarization-selected and control females in relative region 265 volumes ( Fig. 4a- In parallel, we analyzed brain region volume differences using a more conservative approach 268 and found similar and consistent differences between selection lines. Specifically, we used a 269 multivariate Bayesian model that included the relative size of the 11 brain regions as dependent 270 variables. Posterior samples drawn from the multivariate model indicated that confidence 271 intervals for the difference in relative volume in the medulla oblongata, the optic tectum cups 272 and the thalamus between polarization-selected and control females did not overlap with zero 273 (Supplementary Table 7a).  274   275 We then used the multivariate Bayesian model to evaluate the correlation in relative brain 276 region volume between multiple regions measured. We focused on evaluating correlations with 277 other brain regions for the three regions significantly differentiated between lines following 278 artificial selection (Supplementary Table 7b). We found no correlation between optic tectum 279 cup relative volume and volume of any other region measured. However, we found a significant 280 inverse correlation between thalamus and medulla relative volume (rescorrMedulla-Thalamus [95% 281 CIs]: -0.40 [-0.65 / -0.12]). This finding suggests that the opposite differences observed in the 282 volume of these two brain regions between control and polarization-selected female guppies 283 may be linked to changes in brain development processes associated with artificial selection 284 for higher coordinated motion. 285 286 Information acquisition through the visual system 287 Efficiently acquiring information through the sensory system, mainly through visual cues, is a 288 basic principle of collective motion in shoaling fishes 3 . Given observed differences in the size 289 of the optic tectum cups between polarization-selected and control fish, we investigated 290 potential differences in visual perception between lines. For this, we compared eye size and 291 two key characteristics of the visual system to track movement of conspecifics, visual acuity 292 and temporal resolution. 293 294

Eye size 295
We quantified eye size in females from polarization-selected (n = 57, standard length mean 296 We further assessed potential differences in visual perception between selection and control 311 lines by quantifying visual acuity in the same individuals for which eye size was measured. 312 Visual acuity allows an individual to resolve spatial detail and can be critical for an organism's 313 fitness 29 . We measured visual acuity in our fish by quantifying their innate optomotor response 314 in contrasting rotating gratings. This a widely used method to study visual acuity in multiple 315 fish species, including guppies 30-32 , and we have previously used this approach to evaluate the 316 visual system of guppies in similar contexts 28,33 . Following the methods in ( 28 ), we exposed our 317 fish to a series of six stimuli with rotating and static gratings of different widths at the lower 318 end of the known guppy visual acuity, where thinner widths are more difficult to perceive. If 319 polarization-selected and control groups differed in their ability to resolve spatial ability, 320 differences in their optomotor response towards these rotating stimuli at the lower end of the 321 guppy visual acuity should be expected. However, we found no difference in their average 322 optomotor response combining data from all stimuli (LMMacuity: Estimateselection = 0.00 [-0.08, 323  Although the ability to resolve spatial detail, acuity, is arguably an important visual parameter 329 for guppies to recognize conspecific positions in shoals, it provides no information on an 330 individual's ability to track movement 34 . Similar to many social fish species, guppies swim 331 with a saltatory movement style that features discrete changes in speed and direction 9 . 332 Consequently, we implemented an additional experiment that evaluated potential differences 333 between polarization-selected and control fish in their temporal assessment of speed and 334 direction changes. Using the same experimental apparatus used to evaluate visual acuity, we 335 video recorded female guppies from our selection and control lines and exposed them to a 336 single-width rotating stimulus that moved in multiple directions and at different speeds (see 337 methods). We next used automated tracking to obtain orientation and speed of the fish for each 338 frame and to quantify their direction and speed in relation to the stimuli presented at each time 339 point. If polarization-selected and control groups differed in their ability to track movement, Our work demonstrates that selection for schooling behavior in female guppies has important 367 implications for how this species behaves in the presence of potential threat. Analyses of 368 motion patterns in these fish show that polarization-selected groups maintain higher 369 polarization and speed when exposed to a potential threat. In addition, our analyses indicate 370 that individuals from polarization-selected groups spend less time inspecting the threat than 371 individuals from control lines. We further studied visual capacities in these fish and find no 372 differences between polarization-selected and control individuals in visual acuity, temporal 373 resolution or eye size, suggesting that effective processing of visual information in the brain is key for the observed differences in their ability to synchronize their swimming with close 375 neighbors at multiple contexts. 376 377 In parallel, our results suggest that artificial selection for higher polarization has produced 378 significant changes in the brain anatomy of female guppies. Neuroanatomical measurements 379 indicate that polarization-selected fish exhibit a larger thalamus and a large optic tectum cup, 380 but a smaller medulla oblongata, compared to control fish. These rapid changes in brain region 381 sizes in response to selection for polarization behavior are consistent with previous artificial 382 selection directly on neuroanatomy, which resulted in rapid shifts in both relative brain size 383 and relative telencephalon size, in just a few generations in guppies 35,36 . 384 385 Below, we discuss the implications of these discoveries for our understanding on how the 386 association between brain morphology and anti-predator behavior might drive the evolution of 387 collective behavior. We found two brain regions that were larger in relation to the rest of the brain in polarization-444 selected fish, the optic tectum cup and the thalamus. The optic tectum is the terminus of a vast 445 majority of optic nerve fibers and axons of retinal cells 50 and as such is the primary vertebrate 446 visual center. Despite wide variation in optic tectum size across teleost species, this region is 447 known to control eyes and head movement in fish, as well as forming instant representations 448 of the immediate surroundings to transform visual cues into behavioral responses 50,51 . This 449 function is primarily achieved in superficial layers of the tectum 52 , which corresponds to the 450 optic tectum cup region used in our neuroanatomical parcellation of major brain regions in the 451 guppy 53 . 452

453
The evolved differences in optic tectum cup size between polarization-selected and control 454 female guppies we found are concordant with phenotypic plasticity findings in nine-spined 455 stickleback where it was found that individuals reared in groups developed larger optic tectum 456 that those reared individually 18 . Differences in the ability to acquire sensory input have 457 previously been associated with differences in schooling propensity 54 . In our experiment, rapid 458 evolution of higher polarization did not lead to changes in visual performance or eye size. 459 Together, these findings suggest that differences found between polarization-selected and 460 control lines in this section of the optic tectum should have an effect in their ability to control 461 body orientation during complex social maneuvers such as predator evasion and swimming 462 with low inter-individual distances, but not in sensory information acquisition. This is 463 consistent with current considerations of the tectum as more than a visuomotor relay, with 464 implementation of intrinsic processing of visual information to extract critical features and send 465 processed outputs to the posterior thalamus and then to further decision-making regions of the 466 brain 51 . 467

468
The primary receiving region of processed inputs from the tectum, the thalamus, was also found 469 to be enlarged in polarization-selected females. To date, the level of homology between the 470 mammalian and teleost thalamic regions is under strong debate 55,56 . In mammals, the thalamus 471 plays critical roles in the modification, filtering and distribution of sensory and 472 mechanosensory information into decision-making regions of the brain 56,57 . Recent findings in 473 zebrafish suggest that these tasks are performed mainly by the preglomerular complex, whose 474 developmental origin might be outside the diencephalic area that contains the thalamus in 475 mammals 58 . The neuroanatomical atlas we created to analyze relative region size does not 476 allow us to assess whether this particular region is part of the thalamus area which has been 477 enlarged in polarization-selected females. However, our results of strong gene expression 478 changes in the telencephalon of polarization-selected and control females in response to 479 decision-making across social contexts 59 , are in agreement with a higher efficiency in the relay 480 of processed information towards the telencephalon, potentially facilitated by an enlarged In contrast to the thalamus and the optic tectum cup, we found that the medulla oblongata was 495 smaller in polarization-selected lines. The medulla oblongata is an important relay center of 496 nervous signals between the spinal cord and ascendant brain regions with well-described 497 functions in teleosts. First, it has an important role in basic motor function through processing 498 of mechanosensory stimuli from hydrodynamic information 61 . Consistently found across fish 499 species, the lateral line nerves terminate in the medial and caudal octavo lateralis nucleus in 500 the medulla 62 . As the lateral line is crucial for schooling through cues that allow fishes to assess 501 neighbor changes in speed and direction, the reduced relative size of the medulla in 502 polarization-selected lines could be associated with potential differences in the ability to 503 integrate and process information received through these nerves. Yet, our previous studies 504 evaluating motor control capabilities found no difference between polarization-selected and 505 control female guppies 21,63 . Further studies evaluating information processing of 506 mechanosensory input in these selection lines is nonetheless paramount, with special focus on 507 low light and high turbidity conditions. 508 509 Finally, studies in fishes indicate that the medulla has an important role for processing 510 somatosensory signals, with special emphasis in auditory and gustatory signals 64,65 . 511 Specifically, the vagal lobe, part of the medulla, is described to have a prominent role in 512 processing taste. Consistent with our results, the vagal lobe of larval reef fishes is larger in 513 solitary as compared to more social species 66 . While not tested in this study, it may be that the 514 reduction in medulla observed in polarization-selected lines might be associated with important 515 changes in the auditory system and the ability to perceive different tastes. In line with this 516 reduction in the size of the medulla oblongata, our results show that three other brain regions 517 have significant hypoallometric relationships with the medulla (see Supplementary Table 4b): 518 the cerebellum (motor control center), the thalamus and the hypothalamus (hormonal 519 regulation center). Gene expression of angiopoietin-1, a locus implicated in brain tissue 520 development, showed contrasting expression levels between the medulla and the thalamic and 521 hypothalamic regions 67 . Based on this, we hypothesize that selection for more coordinated 522 motion leads to a trade-off between general sensory capabilities that are not important in 523 coordinated movements and specific sensory capabilities required to coordinate movement 524 with neighbors. In the future, it will be interesting to investigate the association between 525 schooling propensity, brain anatomy and potential trade-offs between sensory and 526 mechanosensory capacities. 527 528 Conclusion 529 530 Our empirical approach with behavioral assays on artificial selection lines with divergence in 531 polarization show that collective motion differences are consistent in the presence of a predator 532 threat and that predator inspection behavior varies between the selection lines and the control 533 lines. Moreover, we reveal differences in neuroanatomy that could provide a mechanistic 534 explanation to the observed behavioral differences. Based on our discoveries, we propose that 535 changes in behavior are intimately intertwined with matching changes in brain morphology 536 during the evolution of collective behavior. 537 538 Methods 539 540

Artificial selection for higher group polarization 541
We evaluated the association between brain anatomy and collective motion in female guppies 543 following artificial selection for higher polarization. Extensive detail on the selection procedure 544 can be found in ( 21 ). In short, groups of female guppies were tested in repeated open field tests 545 and sorted in relation to the mean polarization of the group, the degree to which the individuals 546 of a group move with higher alignment 21,24,68 . For three generations, females from groups with 547 higher polarization were bred with males from those cohorts to generate three up-selected 548 polarization lines. In parallel, random females were exposed to the same experimental 549 conditions and bred with unselected males to generate three control lines. Third generation 550 polarization-selected females presented on average a 15% higher polarization, 26% higher 551 median speed and 10% higher group cohesiveness (i.e. 10% shorter nearest neighbor distances) 552 in comparison to control females 21 . Further tests in these lines showed that selection for higher 553 directional coordination was not only driven for selection for faster moving fish. This is 554 because polarization-selected lines were still 5.7% more polarized after statistically controlling 555 for speed differences, and the distance to conspecifics was an important factor driving 556 differences in speed between polarization-selected and control lines 21 . The selection procedure 557 targeted polarization on female groups and we found a weaker response to selection in males, 558 and therefore subsequent neuroanatomical, behavioral and physiological studies focused on 559 females. All fish were removed from their parental tanks after birth, separated by sex at the 560 first onset of sexual maturation, and afterwards transferred to single-sex groups of eight 561 individuals in seven liter tanks. We kept all fish used for anti-predator response experiments, 562 visual capacity tests and brain morphology measurements in these groups throughout their life 563 span. Fish were not reused for experiments across these three categories. All tanks contained 2 564 cm of gravel with continuously aerated water, a biological filter, and plants for environmental 565 enrichment. We allowed for visual contact between the tanks. The laboratory was maintained 566 at 26°C with a 12-h light:12-h dark schedule. Fish were fed a diet of flake food and freshly 567 hatched brine shrimp daily. data from these open field assays was previously used to analyze differences in social 578 interactions 21 . After ten minutes, we sequentially introduced a novel object and a predator 579 model for 6-minute periods in the centre of the experimental arena. In half the assays, we 580 introduced the novel object first and the predator model second, with the order reversed in the 581 other half of the assays. We used a blue coffee mug as a novel object and a fishing lure (18 x 582 3 cm) custom-painted to resemble the pike cichlid Crenicichla frenata, a natural predator on 583 the guppy, as the predator model. These objects have been previously used to successfully 584 reproduce natural behaviors of the guppy in response to a novel object and a predation threat 47 . 585 To facilitate automated data collection, the position and orientation of the predator model was We euthanized animals with an overdose of benzocaine and fixated the whole fish in 2% 685 glutaraldehyde and 4% paraformaldehyde in phosphate buffered saline (PBS) for five days. 686 Following two PBS washes, the brains were dissected out and stained for 48 h in 1% 687 osmiumtetraoxide. We embedded the stained brains in 3% agar and scanned them using 688 microcomputed tomography (microCT, Skyscan 1172, Bruker microCT, Kontich, 689 Belgium). The scanner operated at a voltage of 80 kV, a current of 125 μm, with a 0.5 mm 690 aluminum filter. Images were acquired using an isotropic pixel size of 2.4 μm. We 691 reconstructed cross-sections from scanned images following a in NRecon (Bruker microCT) 692 following a protocol successfully implemented in a previous study evaluating neuroanatomical 693 differences between guppies up-and down-selected for relative brain size 53 . This protocol 694 allowed us to obtain measurements of whole brain size volume and brain region volume in 11 695 major brain regions in the guppy brain: olfactory bulbs, ventral telencephalon, dorsal 696 telencephalon, thalamus, hypothalamus, nucleus glomerulus, torus semicircularis, optic tectum 697 cup, central optic tectum, cerebellum, and medulla oblongata (Fig. 4B). We chose all regions 698 that we could safely identify in the brain scans based on an adult swordtail brain atlas 80 and our 699 own knowledge in fish neuroanatomy. Extended details on guppy brain region reconstruction 700 from digital images can be found in ( 53 ). Two brains from polarization-selected lines were 701 damaged during the protocol, which reduced the sample size to 28 samples. 702

703
We tested for overall differences in relative brain size between polarization-selected and 704 control lines using a linear mixed model (LMM) with brain volume as dependent variable, 705 body size (standard length) as covariate, selection regime as fixed effect, and replicate as 706 random effect. For each brain region, we used two different approaches to determine whether 707 polarization-selected and control lines differ in relative brain region size. First, we used 11 708 independent LMMs with each region's volume as dependent variable, whole brain volume 709 (excluding the volume of the region of interest) as covariate, selection regime as fixed effect, 710 and replicate as random effect. LMMs were implemented in in R (v4.1.3) and RStudio (v2022.07.0) 71,73 using lme4 and lmerTest packages 74,75 . Second, to take into consideration that 712 brain region volumes may be interdependent, we used a more conservative approach and 713 analyzed the data using a Bayesian multilevel model that included 11 brain regions as 714 dependent variables in a fully multivariate context. The full model included an analogous 715 structure to those used in the independent LMMs for each brain region. Parameter values were 716 estimated using the brms interface 81,82 to the probabilistic programming language Stan 83 . We 717 used default prior distributions with student-t distribution (3, 0, 2.5) for all parameters. The 718 model estimated residual correlations among all brain region volumes with a Lewandowski- interests. Data and materials availability: All data and code needed to evaluate the 845 conclusions in the paper will be deposited in an online repository.   Table S1. Statistical tests for overall comparisons in tests evaluating polarization-selected and control female guppies in their shoaling patterns when exposed to an open field test (OFT), a novel object (cup) and a predator model.  Table S3. Statistical tests for comparisons in tests evaluating polarization-selected and control female guppies in their anti-predator behavior when exposed to a predator model  Table S4a. Statistical tests for overall comparisons of group polarization in polarization-selected and control female guppies when the average position of the group was shorter than 200 mm to the stimulus presented in the arena in tests that exposed these fish to a predator model and a novel object (cup).  Table S4b. Statistical tests for independent contrasts of group polarization in polarization-selected and control female guppies when swimming at a distance closer than 200 mm to the stimulus presented in the arena in tests that exposed these fish to a predator model and a novel object (cup).  Table S6. Results from independent Linear Mixed Models evaluating differences in relative brain and relative brain region size between polarization-selected and control female guppies.  Table S7a Results from a Bayesian multilevel model evaluating differences in relative brain region size between polarization-selected and control female guppies. Stars indicate estimates that do not include zero in the confidence interval range based on the posterior samples drawn from the model.  Table S7b. Residual correlations of thalamus, optic tectum cups and medulla oblongata relative volume to other brain regions estimated from a Bayesian multilevel model evaluating differences in relative brain region size between polarization-selected and control female guppies. Stars indicate estimates that do not include zero in the confidence interval range based on the posterior samples drawn from the model.   Table S11. Statistical results using a robust linear mixed model approach for comparisons in the proportion of time following the correct direction of the stimulus in visual temporal assays between polarization-selected and control female guppies.