Body posture, gaze, and predator detection: how stalking behaviour may influence colour vision evolution

The success of a predatory attack is related to how much a predator manages to approach a prey without being detected. Some carnivore mammals use environmental objects as visual obstacles during stalking behaviour, allowing them to get closer to their prey while only showing parts of their coat or faces during movement or visual monitoring. Here, we investigate the influence of carnivores’ body postures and gaze on their detection by potential prey. To do so, we photographed taxidermized carnivore models (cougar, ocelot and lesser grison) in natural scenes and presented them to human dichromats (i.e., colourblind) and trichromats (i.e., normal colour vision). Our findings highlight the importance of predators’ complete body outline and gaze as search images during predator detection tasks. We also demonstrate how the coat and facial colour pattern of predators may camouflage their body outline and gaze, hampering predator detection. Furthermore, we observed that carnivore coat colour patterns may serve as an additional cue for trichromats, particularly in hidden carnivore detection tasks that proved to be more difficult for dichromats. We discuss our results within the context of a predator-prey arms race scenario, considering the evolutionary processes that may have generated the evidence presented in this study.


8
The ability to accurately infer environmental predation risk is fundamental for the survival 3 9 of animals, given the drastic adaptive consequences of not responding adequately to a predatory 4 0 attack [1][2][3]. Predators commonly use cryptic behaviours while trying to capture a prey (i.e., 4 1 stalking behaviour), moving stealthily and using visual obstacles (e.g., vegetation) to hide their 2 We followed the camera settings of previous studies that also used digital photographs to 2 2 3 study animal colouration [61,62,76]. In summary, we set camera to AV mode (automatic shutter 2 2 4 speed) with an ISO of 100 or 200, depending on available light. We used EF-S 18-55 mm lens 2 2 5 (Canon, Inc) and set them at 55 mm focal length (88 mm equivalent). We set aperture to f/36 to 2 2 6 obtain the maximum field depth. The decrease in sharpness caused by small aperture was not 2 2 7 considered critical. We captured photographs in sunny cloudless days, between 9:00 and 15:30.

8
We positioned the camera with our backs to the sun, so the camera always faced the sunlit side 2 2 9 of carnivore models. We created four body posture categories to simulate different types of movement of a 2 3 1 potential predator (Fig. 1). In the first category, we positioned the carnivore model in the 2 3 2 background environment with one side of its body facing the camera, so that the animal's full right side of the image, and never faced the camera (i.e., both eyes were never visible to the 2 3 5 observer). In the second category, we positioned the carnivore model facing the camera so that 2 3 6 its face and eyes were fully visible to the observer. In this posture, only the model's head, 2 3 7 shoulders, and front legs were visible in the photo; we used scenery objects, such as bushes, logs 2 3 8 and lianas, to hide the posterior portion of the body when necessary. In the third category, we 2 3 9 positioned the carnivore model with its back to the camera, so that only its dorsum, hind legs, 2 4 0 and tail were visible in the photo. We also used scenery objects to obstruct the anterior portion of 2 4 1 the body when necessary. Finally, in the fourth category, we positioned the carnivore model with 2 4 2 one of its sides facing the camera, but we visually obstructed parts of its body with objects from the lateral posture, presenting complete body outline, represents the task of easier detection; (2) 2 4 7 the anterior posture, presenting face and eyes but not body complete outline, represents the 2 4 8 second easiest task; (3) the posterior posture, presenting dorsum but neither face nor body 2 4 9 outline, represents a considerably more difficult task; and (4) the hidden posture, showing only 2 5 0 parts of pelage, represents the most difficult detection task. We used the Pantone ColorChecker Passport Camera Calibration Software (X-Rite Inc.) to 2 6 0 create custom DNG profiles based on colour chart photos, and Adobe Camera RAW to apply 2 6 1 these profiles to RAW pictures for colour adjustments. RAW colour adjusted photos were 2 6 2 converted to TIFF and edited in Adobe Photoshop CS5. We controlled for real size difference 2 6 3 between the carnivore models by creating two stimulus dimension categories, in which the TIFF 2 6 4 images were edited so that all models had one of the following dimensions: (1) small targets, in 2 6 5 which model's length (from snout to hindquarter) was set to two centimetres of the monitor 2 6 6 screen; and (2) large targets, in which the model's length was set to five centimetres of the 2 6 7 monitor screen. In photographs where specimens were not positioned on their lateral side (i.e., 2 6 8 anterior and posterior posture categories), the photograph has been edited so that carnivore 2 6 9 model presented the same shoulder height or the same hip height as in its lateral posture photo. These stimulus dimension categories simulate different detection distances: while small 2 7 1 specimens appear to be farther away from the observer, larger specimens appear to be closer. We conditions, small targets stimulate a retinal area smaller than the human fovea, while larger  heads at a distance of 40 cm from the screen with the aid of a chin and forehead support. We Leopardus sp.) and this model was excluded from experimental session to avoid the creation of a 2 9 0 search image by subjects. Experimental sessions started with the presentation of a grey screen for three seconds, after 2 9 2 which a + symbol appeared in the centre of the monitor. Subjects were instructed to touch the + 2 9 3 to move to the next stage of the experiment, fixing their eyes on the centre of the grey screen while performing this action. After touching +, a set of four photographs was displayed on the 2 9 5 monitor, with one photo in each quadrant. Three of the four photos represented a negative 2 9 6 discriminative stimulus (i.e., natural scenes without the carnivore model) and the remaining 2 9 7 photo represented the positive discriminative stimulus (i.e., natural scene with the carnivore 2 9 8 model). The software displayed the photos for one minute or until subjects responded by 2 9 9 touching the screen. We instructed subjects to identify the positive stimulus and touch its 3 0 0 corresponding photo as quickly as they could. After the subjects' response or after one minute, 3 0 1 the photo set disappeared and was replaced by the grey screen. After three seconds, the + symbol  We selected a total of 72 photographs to be presented to experimental subjects as positive randomly between each photo set. We believe that this experimental procedure minimized or 3 1 4 nullified any observer bias during behavioural sampling, since experiments were conducted and 3 1 5 data were recorded by a computer software. The statistical analyses followed a previous study that employed a similar method [62]. We analysed data with linear mixed models (LMM, response latency) and generalized linear mixed the models to account for the repeated measure design. Here, our primary focus was to examine 3 2 2 whether the response of subjects (response latency and detection accuracy) varied with carnivore 3 2 3 body posture depending on the visual phenotype (dichromat or trichromat). Therefore, nearly all 3 2 4 models include body posture and visual phenotype as fixed effects plus the interaction between 3 2 5 these two variables.

2 6
We were also interested in evaluating whether body posture affects differentially the 3 2 7 responses of dichromats and trichromats according to background scenario (grassland, savanna, 3 2 8 or forest), carnivore species (lesser grison, ocelot, or cougar), and stimulus dimension (small or variables as covariates when appropriate. For example, we added carnivore species and stimulus 3 3 5 dimension as fixed effects in the models built for each level of background scenario.

6
Response latency was log-transformed (natural log) before the analyses to meet the 3 3 7 normality assumption. For the detection accuracy analysis, we analysed whether and how the  Using the same dataset, de Moraes et al. [62] found that a ceiling effect driven by the long trial 3 4 2 duration (one minute) masked the differences between phenotypes in detection accuracy.  We also assessed whether facial colour patterns affect gaze detection of different 3 4 6 carnivores. For this purpose, we built two models (one for each response latency and detection 3 4 7 accuracy) using only data from the anterior posture of taxidermized models. We included visual 3 4 8 phenotype, carnivore species, and their interaction term as fixed effects in these two models. We considered background scenario and stimulus dimension as covariates in the models. post-hoc tests on full models or models without non-significant interaction terms using emmeans  were hidden (which led to slower responses) (Fig. 3, Table S2). Irrespective of body posture (Figs 2-7, Tables S1-S6), trichromats outperformed 3 7 8 dichromats when detecting cougars, but both visual phenotypes performed similarly when 3 7 9 detecting lesser grisons. Ocelots were equally easy to detect by both visual phenotypes but only 3 8 0 when positioned laterally; under the remaining postures, trichromats outperformed dichromats 3 8 1 (Fig. 3, Table S2). In several conditions, dichromats, but not trichromats, performed worse when 3 8 2 detecting predators in the posterior posture than in the anterior posture (Figs. 2, 3, and 7, Tables   3  8  3 S1-S2, S6). Both trichromats and dichromats performed better in detecting the gaze of cougars (latency 3 9 2 and accuracy) than the gaze of ocelots and lesser grisons (Fig. 8, Tables S7-S8). Trichromats (latency), but these phenotypes did not differ in the ability to detect the gaze of lesser grisons 3 9 5 (Fig. 8, Tables S7-S8). Our results show a broad trichromatic advantage for the detection of carnivoran mammals.   performances only in some of the easier tasks (e.g., detection of larger targets with visible body 4 2 8 outline or gaze; Fig. 7.A) and during lesser grison detection, the only carnivore model with 4 2 9 cryptic coat (i.e., the hardest detection tasks) (Figs. 3.C and 6.C). The first study to observe this trichromatic advantage on predator detection was conducted   predicted values during the first three seconds of each trial (mean and 95% CI). Differences in detection 4 5 7 accuracy between visual phenotypes are highlighted by asterisks. Differences in detection accuracy 4 5 8 between carnivore models are given by different letters. naïve prey that does not know a new predator, we can expect that the latter is not interpreted as a  The display of predator's gaze can generate a new evolutionary pressure on prey, in which 4 7 6 natural selection should favour those that have the neural capacity to infer a higher predation risk a high predation risk cue. Therefore, two distinct search images can be equally relevant in 4 9 0 predator detection tasks: the predator's body contour and its gaze. These results also point to the 4 9 1 importance of gaze recognition for primates (at least for humans), indicating that predation risk  A probable next step in this sensory arms race is the evolution of gaze camouflage. As prey 4 9 5 are able to infer high predation risk from staring eyes, natural selection may have favoured 4 9 6 predators that add noise to the visual perception of their faces [14,28]. This noise likely hampers 4 9 7 gaze identification, allowing predators to get even closer to prey during stalking behaviour and 4 9 8 achieve higher capture success [8]. Our results seem to corroborate this reasoning, considering 4 9 9 that the predator that presents facial uniform colouration (i.e., cougar) is more easily detected 5 0 0 than models that present complex facial colourations (i.e., ocelots and lesser grison) (Fig. 8). Mustelidae, the two families with animal models used in the present study. According to Caro et 5 0 7 al. [28], the anterior colouration of mustelids is associated with the ability to spray noxious anal An important step in the arms race described here is represented by the evolution of 5 1 7 trichromacy. Although this visual phenotype probably evolved in feeding contexts, such as the 5 1 8 detection of reddish or orange fruits, leaves and other food items [40,42,43,46,[50][51][52]86], it is 5 1 9 possible that an exaptation process [87] has occurred once trichromacy helped some primates to parvorder Catarrhini (i.e., the uniformly trichromats primates). Our results, in general, demonstrate the superiority of trichromacy in this context, but some specific situations make other mammals with a dichromatic visual phenotype [33]. Therefore, the coat colour of some 5 3 5 carnivores represents a search image for trichromats that is not available to dichromats, since this 5 3 6 visual phenotype does not present the necessary sensory channels to perceive this environmental 5 3 7 cue [61]. This trichromatic advantage is also observed when we evaluate the effect of gaze 5 3 8 camouflage (Fig. 8). Although complex colour patterns hampered trichromats' performance 5 3 9 during predators' gaze detection, in comparison to uniform coloured faces, trichromats 5 4 0 performed better than dichromats when targets presented complex orange/yellowish facial 5 4 1 colours.

4 2
The colour pattern presented by lesser grison seems to be immune to trichromacy, since the able to nullify the effect of trichromacy on gaze detection (Fig. 8). Therefore, the cryptic colours presented by this carnivore appear to be the final (or contemporary) step in the sensory arms race 5 4 7 described so far, as it allows these animals to overcome the evolutionary novelty presented by investigate whether primate specialist predators (or those that prey upon any other trichromatic  In conclusion, we demonstrated the efficiency of primate trichromacy in predator detection that indicate their presence on the environment (i.e., body contour and gaze), trichromatic prey 5 5 8 can easily identify them by using coat colour as a high-risk cue. An important factor not 5 5 9 evaluated by our experiments is the influence of movement [88,89] in carnivore detection. The predator's colour pattern difficult its identification by prey during movement would be very with complex patterns of spots and stripes, or disruptive colouration, etc.) on the detection of 5 6 6 moving predators.

6 7
Going further, our results point to a comprehensive advantage of trichromacy in different situations of greater detection difficulty. A productive step for future studies is to investigate 5 7 0 whether this trichromatic advantage in predator detection also applies to animals from other non-5 7 1 mammalian groups that also prey on primates, such as birds of prey [90] and snakes [91]. It is 5 7 2 especially relevant to consider these groups since orange hues are uncommon among avian or 5 7 3 snake mammalian predators, which could reduce the importance of trichromacy in these 5 7 4 contexts. Therefore, visual modelling studies or behavioural experiments that use birds of prey  We once again demonstrate that the research of primate colour vision evolution can take multiple 5 7 9 paths besides the investigation of relative phenotype advantages in foraging, so we emphasize intraspecific communication (e.g., [59,60,93]) and predation risk. This study was financed in part by the Coordenacao de Aperfeicoamento de Pessoal de