Abstract
Over the last 50 years, point-light displays have been successfully used to explore how animals respond to dynamic visual stimuli—specifically, differentiation of the biological from the non-biological. These stimuli are designed to preserve movement patterns while minimizing static detail, with single dots representing each of the main joints of a moving animal. Imposed by their internal skeleton, vertebrate movements follow a specific semi-rigid dynamic pattern, termed “biological-motion”, which can be used to distinguish animate from inanimate objects. Although biological motion detection has not been studied in invertebrates, rigid exoskeletons force many species to also follow semi-rigid movement principles. Due to their highly developed visual system and complex visual behaviors, we investigated the capability of jumping spiders to discriminate biological from non-biological motion using point-light display stimuli. By constraining spiders so that they could rotate but not move directionally, we simultaneously presented two point-light display stimuli with specific dynamic traits and registered their preference by observing which pattern they turned towards. Jumping spiders clearly demonstrated the ability to discriminate between stimuli. However, spiders showed no preference when both stimuli presented patterns with semi-rigid movements, results that are directly comparable to responses in vertebrate systems. This represents the first demonstration of biological motion recognition in an invertebrate, posing crucial questions about the evolutionary history of this ability and complex visual processing in non-vertebrate systems.
For the vast majority of animals, determining if an object in the environment is another animal is crucial for survival. However, recognizing every single object individually quickly becomes computationally impractical, requiring more resources for minimal gains. In these cases, we expect evolution to favor simple but robust systems with general rules exploiting common characteristics able to cover most situations (Chittka and Niven, 2009). In most vertebrates there are characteristic spatio-temporal relationships between different body parts while in motion—relationships imposed by the overall body-plans of these animals. Because the vertebrate body consists of linked rigid segments, movements between joints are accordingly semi-rigid—while some joints may seem independent from one another (e.g. wrist to knee) with their relative distances varying across time, other joint-pairs are not (e.g. wrist to elbow) with relative distances that are fixed. Thus when observed visually, the movements of these animals result in an identifiable statistical pattern.
Indeed, even when presented with just eleven dots moving in correspondence with the position of the main joints of the human body, observers can correctly identify the presence of an agent (Johansson, 1976, 1973; Neri et al., 1998; Troje, 2013; Troje and Westhoff, 2006). Such dot-only stimuli (called point-light displays) are completely devoid of structural information, but retain motion-based information. The detection of these patterns in humans is not confined to stimuli depicting conspecifics, but enables the detection of many other animals (Mather and West, 1993). Even “scrambled” displays, where dots maintain a semi-rigid relationship, are perceived as unknown, yet alive, entities (Troje and Westhoff, 2005). Moreover, the discrimination of these patterns seems to be innate (Simion et al., 2008). Beyond humans, studies in many other species have revealed similar results (Brown et al., 2010; Nakayasu and Watanabe, 2014; Regolin et al., 2000; Vallortigara et al., 2005), suggesting that this visuo-cognitive strategy is widespread across vertebrates and is thus evolutionary ancient. As such, biological motion detection is widely considered to be a key mechanism in enabling the detection of living animals within a visual scene (Troje and Westhoff, 2006, 2005). That this system relies on motion (Neri et al., 1998; Poom and Olsson, 2002) is important, as it has been proposed that motion-based visual cues can be extracted more quickly than many static cues, are less vulnerable to disruption, and are potentially governed by smaller networks of neurons (Borst and Egelhaaf, 1989).
In some invertebrates, body plans and movements follow the same rules of semi-rigidity described above for vertebrates, including most arthropods where the exoskeleton is composed of connected, rigid components. Further, like many vertebrate species, the ability to visually differentiate moving animals from other visual stimuli is likely to provide a strong selective advantage. Because of these similarities in both form and function, we set out to explore whether biological motion cues might also be used by invertebrates, suggesting an even more widespread ability across animals.
We used jumping spiders (Salticidae) to address this question, as these animals are among the most visually adept of all arthropods, with vision playing a central role in a wide range of behaviors (Harland et al., 2012). The visual systems of these animals is also unique, with a total of eight eyes and with different pairs thought to be specialized for different visual tasks (Harland et al., 2012). The large anterior-medial eyes (AME, or primary eyes) are forward facing and characterized by a narrow visual field (<5°) and high visual acuity (Harland et al., 2012), and are believed to play a central role in figure recognition (Land, 1969a, 1969b). The anterior-lateral eyes (ALE) have lower visual acuity but a much wider visual field (≈±50°) and are believed to be specialized for motion detection (Land, 1972a; Zurek and Nelson, 2012a). The remaining two pairs of eyes, posterior-medial and posterior-lateral (PME and PLE), although smaller, are characterized by a wide visual field, and together with the ALE grant the spider a near 360° field of view. The function of the PME and PLE seem to be similar to the ALE, with all three pairs classified as secondary eyes. Behavioral observations support the functional division of labor hypothesis: when a moving target is detected by the secondary eyes, the spider will pivot to face it directly and track it with the AMEs (Jakob et al., 2018; Land, 1972a; Zurek and Nelson, 2012b). These rotational pivots are so rapid, precise, and robust that they are referred to as saccades (Land, 1972b), analogous to the eye movements of other animals, including humans. Neuroanatomically, the two sets of eyes (primary and secondary) possess completely separate early visual-processing pathways, and as such the information coming from these two sources is integrated in higher-order brain centers only after separated processing, if at all (Steinhoff et al., 2020). However, salticids do not turn towards all moving targets in the visual field, suggesting a selective-attention-like process and/or that secondary eyes can inform target differentiation, such as the detection/discrimination of biological motion cues, rather than simply acting solely as motion detectors (Spano et al., 2012).
We tested Menemerus semilimbatus spiders in a forced choice paradigm using point-light display stimuli. For a detailed description of the procedure, see Methods section. Individuals were suspended above a polystyrene sphere so that their legs could contact it, while the sphere was supported and allowed to spin freely by a constant stream of compressed air from below. In spiders, the legs attach to the cephalothorax (the head)—thus in this setup, the spider remained in a fixed position and maintained a fixed head orientation, but was able to move its legs freely, transferring its intended locomotor actions to the sphere. Stimuli were then presented on a computer monitor placed 14cm in front of the spider (Figure 1A). Two stimuli were presented simultaneously, entering the monitor space from opposite sides and moved towards the center, where they disappeared. Using a video-based sphere-tracking system (FicTrac) (Moore et al., 2014), we then measured how spiders moved the sphere in response to the stimuli presentation. As stated above, jumping spiders produce rotations, or “saccades” with their body upon detection of a stimulus with the secondary eyes, turning rapidly about a single location (like a top, or a military tank). Thus, the measured output was the rotational locomotor behavior—which of the two targets the spider chose to rotate towards.
We designed 5 different stimuli, paired in a total of 4 different conditions (Figure 1B; see Methods section for details; videos are available as supplements) Stimuli were designed as spider-versions of standard point-light display stimuli. The “biological motion” stimulus consisted of a point-light display following the motion of a walking spiders from a side-view perspective. In the “scrambled motion” stimulus the starting point of each dot was randomized, but it then followed the same path as in the biological motion stimulus—a pattern perceived as biological in other systems (Troje and Westhoff, 2005). The “random motion” stimulus consisted of the same number of dots constrained to the same overall area as the biological motion stimulus, moving with the same velocities, but in randomized directions. A spider silhouette, based on the same video used to create the biological motion stimulus and thus following the same movements, and an ellipse translating at the same speed of the other stimuli but without the finer-scale joints movement were also used. In conditions 1 and 2 (Figure 2C), the random motion stimulus was paired against the biological and the scrambled stimuli respectively, testing responses to biological versus non-biological motion. In condition 3, the biological stimulus was paired against the scrambled stimulus, thus presenting two semi-rigid stimuli. In condition 4, the spider silhouette was paired against the ellipse, a control condition to observe responses towards structurally detailed stimuli compared to moving stimuli of similar size but with minimal detail.
Results
Only the main results are reported here. For the full analysis see SI. Firstly, we checked whether the observed rotations of the sphere did truly describe the reaction of the spider to the presence of any stimulus on screen. We found a strong correlation between absolute rotational speed and the stimuli angular location (both occupied the same angle, being them symmetric to the center) (GLMM analysis of deviance, Chi-square=3330.8, p-value<0.0001) across all experimental conditions (difference between conditions: Chi-square=2.52, p-value=0.47) (Figure 2A and 2B). We also observed a significant interaction between stimulus position and condition (Chi-square=46.77, p-value<0.0001): specifically we found this correlation to be stronger in the silhouette vs. ellipse condition than in the other conditions (post-hoc analysis, Tukey correction. Silhouette-Ellipse vs Scrambled-Random: est.=0.004, SE=0.001, t=4.73, p<0.0001; vs Biological-Scrambled: est.=0.005, SE=0.001, t=6.31, p<0.0001; vs Biological-Random: est.=0.004, SE=0.001, t=5.12, p<0.0001). This was likely due to the fact that the silhouette and the ellipse stimuli are composed of more black pixels, resulting in higher contrast between the stimuli and the white background than the light point displays. This would have allowed those stimuli to be detected earlier, strengthening the correlation between rotational speed and stimulus position. All together these observations suggest that the observed Z rotation of the sphere is indicative of saccades produced by the spiders towards the stimuli on screen.
Having established that, we proceeded by setting as positive the rotation produced towards the more biological stimulus, and as negative the ones congruent with the other. We then used this variable to compute the spider’s preference (Figure 2C). During stimulus presentation, the spiders preferred the ellipse over the silhouette (est.=−23.73, SE=3.41, t=−6.96, p<0.0001), the random over the scrambled (est.=−15.1, SE=3.41, t=−4.435, p=0.0001), the random over the biological (est.=−13.37, SE=3.38, t=−3.96, p<0.0006), and showed no preference in the biological vs scrambled condition (est.=− 1.04, SE=3.3, t=−0.32, p=1). When considering the 30 seconds between-stimulus period instead we observed no difference for all conditions (post-hoc analysis, Tukey correction. biological-random: est.=0.52, SE=1.59, t=0.32, p=1; biological-scrambled: est.=1.68, SE=1.53, t=1.1, p=0.922; scrambled-random: est.=2.46, SE=1.68, t=1.47, p=0.706; silhouette-ellipse: est.=3.4, SE=1.72, t=1.97, p=0.328).
Discussion
These results clearly demonstrate the ability of jumping spiders to discriminate between biological motion cues. The results also show strong internal consistency, with preferences across all stimulus pairs for the less realistic stimuli, with the exception of the paired semi-rigid stimuli where no preference was observed. We initially found this “reverse choice” surprising, as we expected spiders to turn towards the stimulus with the highest probability of being a living organism, as has been previously found in other systems (Simion et al., 2008; Vallortigara et al., 2005). However, orientation towards the less-biological stimulus appears consistent with the functional organization of the visual system of the jumping spiders. As described above, these animals produce saccades upon detection of a target with the lateral eyes, allowing further inspection with the AME. As per our initial hypothesis, the secondary eyes may be immediately able to decode motion-based information, enabling them to determine which of the stimuli requires more detailed investigation. In a forced choice paradigm it may be advantageous to focus the AME on the stimulus which cannot be decoded with the lateral eyes alone, particularly since the other target will still remain in the visual field of the secondary eyes following rotation. Moreover, it is crucial to consider that many of the species in which a preference for biologically moving stimuli has been observed are highly social. Indeed, it has been amply argued that an innate “agent detection” system may exist because of the need of these animals to quickly aggregate with social companions (Rosa-Salva et al., 2021). In such species it would be advantageous to show an early attentional preference to movements which are more likely to be generated from an agent. However, jumping spiders are not social, thus we should not necessarily have the same expectation for stimulus preference. It is crucial to point out that motion cues in point-light displays can be used for much more than just “agent detection”. It has been shown that humans can not only detect the presence or absence of a living being in the stimulus, but also much finer details, like for example the sex of the actor (Kozlowski and Cutting, 1977). It has been suggested that also jumping spiders could detect and use more subtle motion cues, to discriminate between different species to approach or avoid, like conspecifics or ants (Cross and Jackson, 2018). The lack of a preference between the scrambled and the biological displays in our experiment seems to point against the presence of such finer discrimination, but more direct inquiries are needed to provide a definitive answer.
In our setup spiders could only observe the target with both AME and ALE eyes when the stimuli reached the center of the screen (see Figure 1A). Specifically, this opportunity for primary-secondary overlap corresponded with the third period of stimulus movement, yet the majority of during-stimulus turns, and thus choices, occurred before this phased (see Figure 2B). That spiders demonstrate this preference even when targets can only be viewed by the secondary eyes is striking. It has been suggested that these eyes are capable of feats beyond mere motion detection (Spano et al., 2012), but our results directly demonstrate that these eyes can solve a discrimination task without aid from the primary eyes. As point-light displays are designed to contain minimal visual detail, the discrimination operated by the secondary eyes must be based on motion. How this might be implemented in the nervous system, however, remains unknown. Regardless, it is noteworthy that the jumping spider brain is far more compact, with orders of magnitude fewer neurons, than the brains of the vertebrate species where biological motion detection has been reported.
Our findings provide evidence for the discrimination of point-light display biological motion stimuli (Johansson, 1973) in an invertebrate. If this is evidence of convergent evolution or of deep homology (e.g. presence in the shared common ancestor), still remains unclear. However, it is widely believed that the complexity of high-acuity vision in the vertebrates and invertebrates (e.g. vertebrates, insects, spiders, and mollusks) have evolved independently (Gehring, 2014), suggesting that this is more likely a result of convergent evolution. Indeed, the detection of biological motion represents a solution or strategy to the problem of “agent detection” (Troje and Westhoff, 2006, 2005), which could have been adopted by spiders and vertebrates alike. We suspect that given this, future work in other lineages (e.g. insects and mollusks) may reveal similar results.
Indeed, the use of motion perception by insects to extract other forms of information has been amply studied: for example, flies, bees and wasps use optic-flow information to calculate ground-speed and thus navigate the environment (Egelhaaf and Borst, 1993), while locusts use of motion parallax to estimate distance (Kral and Poteser, 1997). These systems remain different from the detection of biological motion, as thy rely on changes in the visual scene caused by one’s own movement, while responses to dynamic targets still function when the observer is stationary.
The presence of a biological motion-based detection system in jumping spiders deepens questions regarding the evolutionary origins of this visual-processing strategy, and opens the possibility that such mechanisms might be widespread across the animal kingdom and not necessarily related to sociality.
Materials and Methods
Subjects
We used 60 Menemerus semilimbatus in the experiment, of which 31 were females, 10 males and 19 juveniles. The spiders were collected in the wild, in the garden of Esapolis’ living insect museum, Padua, Italy, between June and August 2020. Only animals with a body length bigger than 7 mm were collected to guarantee the proper functioning of the methodology. Once caught, the spiders were maintained in clear plastic boxes measuring 80×65×155 mm and immediately fed a small Tenebrio molitor to ensure a shared level of satiation before the test. The day after capture, a magnet was fixed to the head of each subject to allow us to constrain the animal on top of the treadmill for the duration of the experiment (see next paragraph). Subjects were first constrained between a sponge and a latex film. The latter presented a hole in correspondence with the location of the spider’s head. Here a 1×1×1 mm neodymium magnet was applied using a UV activated resin. Each spider underwent its first test between 1 and 3 days after receiving the magnet. At the end of the full experiment, each spider was again constrained under the latex film to remove the magnet. Spiders were then freed in the same place they were captured. Magnets did not appear to negatively affect the animals during the short period in which they were housed in the lab and spiders that had the magnet removed appeared to move and behave normally.
Experimental apparatus
In order to ensure that spiders would always see the stimuli from the same position, orientation, distance, and most importantly both stimuli at the same time, we built a sphere-based treadmill, similar to that described in Moore et al. (Moore et al., 2014). Similar procedures have been used in jumping spiders, and these animals appear to react to digital stimuli as if they were freely moving in the environment (Humphrey et al., 2018; Melrose et al., 2018; Peckmezian and Taylor, 2015; Zurek and Nelson, 2012a, 2012b).
To construct the treadmill, a polystyrene sphere (diameter = 38 mm) was placed on top of a shallow concave holder. A second piece of plastic with a 24 mm hole in the center was placed on top of the sphere, leaving 1 mm of clearance where this plate met the sphere. In this setup, the sphere was free to rotate, but its overall position was fixed in place. The plastic holder was designed so that the side of the sphere was still mostly visible, with the plastic pieces covering only around 30% of its surface, and the top portion of the sphere was left exposed so that the spider could contact/hold it. This plastic holder also contained 16, 1 mm diameter air inlets directed towards the sphere. When a compressed air source set to a constant pressure of 50mbar was connected to the plastic holder, it produced an air cushion that the sphere floated on, providing almost friction-less rotational movement. The pressure was low enough not to cause the sphere to rotate by itself. The rotational inertia of the ball was also such that the spider could rotate it freely.
To initiate each trial, we used a stick with a magnet glued at one of its ends to pick up a spider. After adjusting the position/orientation of this stick, we positioned the spider on top of the sphere so that the center of the spider’s cephalothorax was at the top of the sphere, its body aligned towards the screen, and at a height above the sphere that enabled a normal walking gait. In this setup, the subject could not move nor rotate, as its cephalothorax was fixed in place, but it would still behave as if it was able to move, acting directly on the sphere. By measuring sphere rotations, we could infer the spider’s intended action.
To enable tracking of the sphere, we used acrylic paint to draw uneven shapes across its surface. From the side view, we recorded video of each experiment at 120 fps. By running the FicTrac software by Moore et al. (Moore et al., 2014) on these videos, we were able to extract the exact orientation of the sphere in each frame, expressed by its X (left-right), Y (fore-aft), and Z (rotational) rotational components, and in turn, the rotational speed.
After every 4 trials, the system was disassembled to clean the sphere and the plastic pieces from any silk residue, which might interfere with sphere rotation.
Stimuli of the experiment
All stimuli were based on a video recording of the side view of a Salticus scenicus walking from the right to the left of the frame. The video was captured with a Nikon D7200 DLSR camera, focused on a 5 cm wide runway recorded at 60 fps and 1080p. From the video we extracted 72 frames, which contained a total of 8 full steps of the subject, with step defined as a movement of all 8 legs starting and terminating with each limb occupying the same position relative to each other. In each frame, we manually registered the position of the eyes, the pedicel, the spinneret, and for every leg the end of the tarsus, the joint between the metatarsus and the tibia, the joint between the patella and the femur. If any of the points were not visible in a specific frame (which did often happen especially for the right side legs, being covered by the body) the point position was registered as not available.
All stimuli were presented to the spider on a 1080p screen, with a pixel size of 0.248 mm, positioned 14 cm away from the center of the treadmill sphere. In each condition (except for the single-stimulus control trials) two stimuli were presented simultaneously, one entering from the left side and the other entering from the right, both moving towards the center, disappearing behind an ideal white screen in the exact center of the monitor. The presentation of the stimuli was controlled by a script written in python3.8 (Van Rossum and Drake, 2009), using the package psychopy (Peirce et al., 2019). The same script also controlled the camera recording of the treadmill, with the package opencv (Bradski, 2000). This way, stimulus presentation and camera recording were matched, in order to know the position of the sphere for every frame of the display. All the stimuli were presented on the screen at 30fps.
Biological motion
For this point light display stimuli, a black dot was presented at each anatomical location registered in the original video, re-scaled in order for the stimulus to occupy a size of 170×60 px on the screen (the total width and height changed slightly following the extension and retraction of the spider legs). The starting position of the stimulus was set outside the screen, so that the picture would not be shown suddenly on the screen but would instead appear to move in from outside. After every 72 frames, (all the frames extracted from the original video) the stimulus would remain static for 45 frames, then loop over the original 72 frames, but this time translated in the X direction to center the first frame over the current position. This 45 frames pause was repeated another time and was followed by a third loop iteration, that terminated with the stimulus disappearing progressively behind a white box at the center of the screen. A full stimulus presentation, from start to end position, was a total of 270 frames, or 9 seconds at the display rate of 30 frames per second. The total time the stimulus was visible to the spider, thus excluding the time it took to appear from outside of the screen, was 8 seconds.
Scrambled motion
In the scrambled motion stimulus, the path of every dot was maintained, but their relative positions to one another in the first frame was randomized. In other species tested, scrambled motion stimuli are still perceived as biological, giving the impression to the observer of seeing an unknown living creature (Troje and Westhoff, 2005). To generate this stimulus, we calculated for every point in each frame its distance to the position it occupied in the previous one, but while maintaining the center of the stimulus in a fixed position. Then, the position of every dot was randomized for the first frame. The position of the dots in the following frames, would be determined by re-applying the distance and direction extracted from the original stimulus to the random position of the dot. We then re-applied the translation, moving all the dots by placing the center of the picture in correspondence with the center of the original stimulus. This way, the scrambled motion stimulus retained the same speed as the original, by globally translating the same amount, and each dot would follow the same path, but the relative position of the dots in relation to one-another was eliminated. We created 4 different versions of this stimulus, with 4 different randomizations. For each trial one of these 4 was chosen, so that each spider would never see the same randomization in different trials (for example, two different conditions present the scramble motion – biological vs scrambled and scrambled vs random – any one spider will see two different versions of the scrambled stimuli, one for each condition) and different spiders would see different versions for the same condition.
Random motion
In the random motion stimulus, the position of every dot in every frame was randomized, and according to the literature this should not be perceived as biological (12). To build this stimulus, we began with the initial, full stimulus used in the biological motion stimuli. We then followed the same procedure as for the scrambled motion but defined the position of the dots in frames 2 to 72 using the same distance from the previous frame as the biological motion stimulus, but in a random direction. Thus the magnitude, or amount of motion is maintained, but each dot takes a random path. To prevent dots from shifting too far from the center, we recalculated a new random direction if the resulting position of the dots exceeded the perimeter of an ideal rectangle of 170×60 px, exactly as the one enclosing all the dots in the biological motion stimulus. We created 4 different versions of this stimulus, with 4 different randomizations, following the same presentation as per scrambled motion.
Silhouette
This stimulus was intended as a control, to test whether spiders would show a preference in this setup at all. Previous experiments have shown that jumping spiders respond differently to different silhouettes (Shamble et al., 2017), but we wanted to evaluate responses to stimulus presentation on our spherical treadmill. For this stimulus, we connected the dots from the biological motion stimulus belonging to the same leg, resulting in a thick black line representing the full leg in its movements. We also added black shapes to represent the spider cephalothorax and opisthosoma, based on the original walking spider video. One edge of each body segment shape was then centered on the picture and each shape tilted according to the eyes and spinneret dots. Thus, this stimulus moved in the same way as the biological motion stimuli, but provided a full silhouette of the target in terms of static appearance.
Ellipse
This shape matched exactly the silhouette in terms of the number of black pixels. For every frame, the center of the ellipse was placed to the coordinates of the center of the biological motion, and in turn all the other stimuli.
Detailed procedure
Each spider underwent a total of 4 trials, 2 of which were administered on a given day, and the other two were administered after a 3-day break. Every trial started with 5 minutes of habituation, in which no stimulus was shown, just a white screen. This phase was included in the experiment to give the animals enough time to decrease their level of arousal from the manipulation before the experiment started and revert to normal behavior. After habituation, the first presentation started, where two stimuli would appear, one from the left and one from the right, move towards the center of the screen, and then disappear. As described, the stimuli stop twice, for around 1.5 seconds, before reaching the center. After the stimulus disappearance, a 30 second pause, with no presentation, followed. This pattern of presentation-pause was repeated another 9 times, for a total of 10 stimuli presentation. Thus, a full trial lasted 11 minutes and 30 seconds. In a given trial, the same stimulus pair was shown for all the 10 stimuli presentation and the starting position of each stimulus (left or right) would follow a semi-random pattern (left, right, left, right, left, left, right, right, left, right; or vice-versa for the other stimulus). This way, each trial was part of a single condition. Each spider was subjected to a different condition (namely, a specific stimuli pair) in every trial, and was therefore presented with all experimental conditions. The order of the trials was randomized for each spider.
After the end of the experiment, we added a 5th condition: ellipse vs nothing. Only one subject was used to test it, as it was only intended to retrieve a baseline measure of preference, to inform the interpretation of the data of the other conditions.
Data analysis
The sphere-tracking software FicTrac (Moore et al., 2014) provided, among other measures, the radians per frame that the sphere rotated around its X, Y, and Z axis, for each video frame. A rotation around each of these axes represented a different kind of motion of the spider:
X-axis: Parallel to the screen, rotation of the sphere around this axis would occur when the subject would move forward or backwards. From the perspective of the side-view camera that was used, a counterclockwise rotation would happen when the spider moved forward, and vice-versa.
Y-axis: Perpendicular to the screen, such that rotation around this axis would happen if the spider moved directly sideways (left or right).
Z-axis: Vertically, through the spider, such that rotation would be registered when the spider rotated around its center. A turn to the left of the spider would correspond to a clockwise rotation and vice-versa.
Other types of movements (e.g. the spider moving diagonally) would result in a combination of the three rotations. Before further analysis the values of each axis were smoothed with a Butterworth filter (order = 3, critical frequency = 0.5) from the python package scipy (Virtanen et al., 2020). What we were interested in were the turns towards the left and towards the right, which would correspond with Z axis rotation peaks in the absence of X and Y components. Due to noise in the system and to individual differences of the spiders and videos, movements typically contained rotational velocities across all three axes. To reduce the impact of these cross-axes effects, for every frame we subtracted from the absolute value of Z the absolute value of X and Y, with 0 as the minimum allowable resulting value. Then, the original sign of Z component (+ or −) was reapplied. This way, Z peaks with low concurrent X and Y components remained high, while highly coupled rotations would disappear. Thus, saccadic movements (large Z rotations, low X, Y components) could be clearly separated from non-saccadic movements such as walking forwards (large X, low Y and Z) or curved walking (large X, large Y, medium Z).
To analyze the impact of stimuli on saccadic turns, we focused on the stimuli presentation part of the trials, including the 30 seconds pauses between stimuli-pair presentations, and changed the sign of each rotation according to the position of the expected preferred stimulus: instead of positive numbers meaning left and negative meaning right, positive numbers represented rotations towards the more biologically realistic stimulus and vice-versa. For example, in condition biological motion vs. scrambled motion, biological motion was coded as positive. Note that because of this conversion, the scrambled motion stimulus was negative in one condition (biological vs. scrambled) but positive in another (scrambled vs. random). The same coding was maintained for the pause section following each stimulus, with rotation towards the side in which the correct stimulus was previously present being positive. We then isolated peaks higher than 0.001rad/sec for both the positive and the negative space and used these peak heights and times in our analysis.
With this coding, a preference for the more realistic stimulus (i.e. more turns towards it) would result in a positive peak average, preference for the less realistic stimulus will result in a negative average, and an average of 0 would signify no preference. If we instead decided a cutoff angular value we would have risked to delete meaningful data. This system instead Indeed, we expected to observe no preference for the between-stimulus sections, regardless of condition, given the a priori expectation that rotational movements in the absence of stimuli should be directed randomly.
Statistical analysis
All data and graphs were prepared using python3.8 (Van Rossum and Drake, 2009) with the libraries pandas (Jeff Reback et al., 2020; McKinney, 2010), numpy (Oliphant, 2006; van der Walt et al., 2011), scipy (Virtanen et al., 2020), and matplotlib (Hunter, 2007). Analyses were carried out with R4.0.2 (R Core Team, 2020), using the packages glmmTMB (Brooks et al., 2017), car (Fox and Weisberg, 2011), emmeans (Lenth, 2018), DHARMa (Hartig, 2018), readODS (Schutten et al., 2020). For the full script see supplemental text. The full database is available as a supplement. As suggested by Forstmeier and Schielzeth (Forstmeier and Schielzeth, 2011), we included in the models only factors that we had an a priori reason for including. First, we calculated a model with the absolute (without the sign coding the direction) peak height as a dependent variable, and the angular position of the stimulus on the screen as a predictor. We employed a generalized linear mixed-effect model, with subject as a random effect. This analysis was performed to check whether extracted peak height represent rotational speed of the spider. Indeed, we know from previous literature (Zurek and Nelson, 2012b, 2012a) that when moving on similar spherical treadmills the rotation caused by the detection of a target (saccadic turns) performed by spiders match very closely the actual angular position of the target. Moreover, these saccadic rotations tend to take similar amounts of time, resulting in faster rotation when the target angle is higher. Subsequently, we calculated a second model, using peak height (with positive values when its direction was congruent with the more realistic target and vice-versa) and the dependent variable, condition, and stimulus presence (visible or not visible) as predictors. Here, we expected the average peak height to be indistinguishable from zero when the stimulus is not visible (as saccadic rotations should be less frequent, and even if they did appear, they should be towards a random direction), and to be higher than 0 for conditions biological vs. random, scrambled vs. random and silhouette vs. ellipse, signifying a preference for the biologically structured stimulus.
Methods validation results
To check whether our scoring based on the average Z rotation was indeed an effective proxy for preference, in condition 5 we presented spiders with a choice between a moving ellipse and nothing on the other side. As expected, we observed a clear preference for the ellipse position when the stimulus was on screen (post-hoc analysis, Tukey correction est.=107.0, SE=12, t=8.89, p<0.0001), and no preference in the 30 second between-stimulus time (est.=−24.5, SE=13.9, t=1.76, p=0.155), supporting the utilized score as representative of spiders turning preference.
Competing Interests
The authors declare no competing interests.
Acknowledgments
We thank Esapolis’ Living Insect Museum of the Padua Province, Padua, Italy, and its director Enzo Moretto, for providing the extra facilities unexpectedly needed due to the COVID-19 pandemic. We also thank Maria Loconsole and Lucia Regolin for suggestions regarding the manuscript.