Visual signals in the wing display of a tephritid fly deter jumping spider attacks

For visual communication to be effective, a signal must attract and hold the attention of the targeted receiver, as well as be interpreted accurately. A number of constraints modulate the efficacy of the signal, such as the receiver's visual capabilities and sensory biases, the medium of transmission and the distance between individuals (Rosenthal, 2007). The incredible variety of movements and colours in peacock spiders (Otto and Hill, 2017) or in the birds of paradise (Ligon et al., 2018) is a testament to the effect of selection in the generation of such displays. The key to triggering the receiver's attention is a combination of display and colour (Rosenthal, 2007), but this has rarely been studied together.

Visual signals have led to the evolution of different behaviours in various animals in the context of sexual selection and prey attraction, among others. In some cases, prey species find it useful to attract and hold the attention of their potential predators either to advertise their toxicity (aposematism) or to distract predators (Robledo-Ospina and Rao, 2022; Ruxton et al., 2018). This is counterintuitive because prey signalling to the predators is inherently risky. However, prey signalling to predators may reduce the likelihood of a successful attack, either before the attack is launched or during the attack (Ruxton et al., 2018). Prey may also transmit information regarding their body condition (Caro, 1995) or seek to deceive the predator by adopting exaggerated postures that modify their appearance (Brodie, 1977). Signals that are used to deter attacks have been broadly characterised as pursuit deterrence signals (Hasson, 1991). Despite a longstanding interest in the function of these signals, there have been few empirical tests, and fewer still using ecologically relevant predators (Ruxton et al., 2018).

For pursuit deterrent signals to be effective, the movement, form or colouration of the body part must be conspicuous to the predator's visual system. Conspicuousness may be enhanced if the prey has structural colours produced by interference reflection, where visibility is influenced by the angle of light and movement (Kelley et al., 2019; Parker, 1998; Stuart-Fox et al., 2020), but structural colours may also be used as a form of camouflage (Kjernsmo et al., 2020). Wing interference colours (WIC) and wing specularity (i.e. gloss or shine) generated by the transparent part of insect wings may have an anti-predatory function, similar to that seen in the metallic colours of greenbottle flies (Pike, 2015), or constitute a signal for conspecific communication (Eichorn et al., 2017; Hawkes et al., 2019; Schultz and Fincke, 2009; Shevtsova et al., 2011).

We explored the structure and function of predator–prey signalling in the jumping spider–tephritid fly system, where the fly performs a wing waving display that deters an attack from the spider. Tephritid fly wings are generally banded, with pigmented sections interspersed with hyaline segments. The fly's defence against conspecifics and heterospecifics is mainly with wing waving displays termed as supination, where it makes semi-circular loops while waving its wings in a synchronous or asynchronous manner as it approaches the target (see Fig. 1A). The supination display is triggered by the movement of the fly's opponent (either another fly or a predator) and is similar irrespective of the identity of the opponent (Aguilar-Argüello et al., 2015). It has been shown in several studies (Greene et al., 1987; Hasson, 1995; Mather and Roitberg, 1987) that wing displays from tephritid flies deter attacks (up to 90% of spider attacks in the fly Anastrepha ludens), but attention has usually been focused on the pigmented fraction of the fly wings (Rao and Díaz-Fleischer, 2012), and there is no information about the effect of fly motion with respect to jumping spider vision.

Jumping spider vision has been reviewed in detail (Harland et al., 2012; Hill, 2022; Morehouse, 2020). Jumping spiders have an unusual distributed visual system with four pairs of eyes, two pairs of which are forward facing (Fig. 2A). Of these forward-facing eyes, the anterior lateral eyes have larger retinae that do not move and serve as excellent motion detectors. The principal eyes have small, boomerang-shaped retinae situated at the proximal end of moveable eye tubes inside the spider's cephalothorax. The eye tubes can move to direct the gaze of the boomerang-shaped retinae to different areas of the visual field. The anterior lateral and principal eyes work in close collaboration, as the anterior lateral eyes are necessary to direct the gaze of the principal eyes (Jakob et al., 2018). When detecting a stimulus of interest, a spider pivots its body to bring the stimulus into the field of view of the forward-facing eyes. The principal eyes can then move to track a moving stimulus or investigate a stimulus while the body remains motionless (Land, 1969), a potentially useful trait for a visually hunting predator.

We hypothesised that the pursuit deterrent effect of a displaying fly can be attributed to changes in the spider's visual attention. To address this, we: (1) analysed the change in body orientation of untethered spiders in an arena as they encountered displaying and non-displaying flies; (2) used a custom-built eyetracker to analyse the gaze direction of a tethered spider's principal eyes as they observed videos of displaying, moving and non-moving flies; and (3) used multispectral digital photography, visual modelling of the acuity and colour perception of the spider, and empirical behavioural assays to evaluate the efficacy of chromatic and achromatic cues in deterring attacks under different light conditions.

We used adult female Phidippus audax (Hentz 1845) (Araneae: Salticidae) spiders and the Mexican fruit fly (Anastrepha ludens Loew 1873; Diptera: Tephritidae). Both species are known to co-occur in citrus orchards in Mexico (Gonzalez-Lopez et al., 2015). The fruit flies frequently congregate on the leaves of citrus plants and are known to employ a lekking mating system where males defend non-resource territories (Aluja et al., 2006). Female flies defend oviposition sites.

Spiders were collected from abandoned farms around Xalapa, Mexico, and kept under environmentally enriched conditions (Carducci and Jakob, 2000) in the arthropod laboratory at INBIOTECA, Universidad Veracruzana. Flies were obtained as pupae from the MoscaFrut facility in Metapa de Dominguez, Chiapas, Mexico, and were reared in cages. They were fed with a yeast hydrolosate/sugar mixture and given water ad libitum upon emergence. Predator–prey experiments were carried out in the laboratory under natural light conditions (i.e. next to a window with sunlight) by placing a spider and a fly in a glass Petri dish (10 cm diameter) arena, with an opaque partition separating the two. They acclimated for 1 min and the partition was removed for the experiment to begin. The interaction was filmed from above at 25 frames s⁻¹ at a resolution of 1920×1080 pixels with a SONY HDR-PJ790V video camera.

From the videos, using a custom MATLAB program (courtesy of Jan Hemmi and Robert Parker; MathWorks, Natick, MA, USA), we manually tracked two points on the spider's cephalothorax and one point on the fly's head. We digitised the anterior (between the spider's principal eyes) and posterior position of the spider cephalothorax and generated x,y coordinates to determine the approximate gaze direction of the spider's forward-facing principal eyes. Gaze direction is a useful measure of selective attention in invertebrates (Winsor et al., 2021). We note here that cephalothorax movement is a rough proxy of retinal movements, as jumping spider retinae have been shown to have an angular travel of up to 50 deg (Land, 1969; Morehouse, 2020) and it was not possible to control for movement of the eye tubes of the spiders. Nevertheless, a field of view of ±10 deg from the midpoint of the sightline, i.e. a line joining the centre of the principal eyes and the back of the cephalothorax has been used in other studies (Rößler et al., 2021).

We manually separated videos into two categories: (1) where the fly performed a display (n=9; Movie 1) and (2) where there was no display (n=10). For an illustration of a typical display, see Fig. 1A. We analysed these behaviours only when the fly and spider were subjectively observed to be oriented towards each other. To determine the gaze direction of spiders relative to the fly (Fig. 1B,C inset), we calculated the angle formed by the body axis of the spider (i.e. the line connecting the anterior and posterior position of the cephalothorax of the spider) and the line formed between the anterior position of the cephalothorax of the spider and the head of the fly (also known as the line of sight). This angle has also been referred to as the ‘error angle’ (Collett and Land, 1975). When the spider's gaze was directed to the head of the fly, the angle recorded would be 0 deg. We determined these spider error angles for displaying and non-displaying flies. Because the bout duration varied between individuals (different flies displayed for different periods), we used the first 38 frames of all spiders to determine the error angles. Thirty-eight frames were used because it was the smallest number of frames in common between the videos. The first frame was determined when both spider and fly oriented towards each other. From this, we determined the mean heading direction (ø), length of the mean vector (r) and circular standard deviation in Oriana v 4.0 (Kovach Computing Services, UK). We compared the distribution of the error angle of the spider with respect to displaying and non-displaying flies using a Watson U² test (Landler et al., 2021).

For this experiment, P. audax spiders were collected as penultimates or adults from farming fields in western Massachusetts, USA, and housed in enriched environment cages at the University of Massachusetts Amherst. Spiders were fed crickets (Acheta domesticus) and had access to water ad libitum. Adult female spiders were randomly chosen from the lab population.

To observe the movement of the retinae of the two principal eyes (see Fig. 2A), we used a custom-built spider retina tracking apparatus (hereafter ‘eyetracker’) at the University of Massachusetts Amherst. The eyetracker and other experimental design details are described in detail elsewhere (Canavesi et al., 2011; Jakob et al., 2018). We provide a brief summary of the setup here.

The eyetracker consists of two main optical elements (Canavesi et al., 2011). Spiders were positioned so that they could look through the eyetracker and view stimuli presented in visible light. The video stimuli were projected onto a Roscoclux frosted diffusion filter (Gel/116, transmission 9%) by an Aaxa P4x Pico Projector. These stimuli did not contain UV information, but previous work confirmed that P. audax respond to video lacking UV (Bednarski et al., 2012; Bruce et al., 2021; Jakob et al., 2018). The retinae were illuminated by an IR light, invisible to the spiders, shone into the spider's principal eyes (Thorlabs IR 850 nm Mounted High-Power LED). An EO-1312 mol l⁻¹ CMOS monochrome USB camera captured the reflections in IR of the retinae and recorded their motion.

To position spiders in the eyetracker, we tethered them to a plastic micro-brush by a warmed wax mixture (beeswax:resin :: 1:1 mixture) applied to their cephalothorax. The end of the microbrush was clipped to a manual five-way positioner (Thor Labs). To ensure that spiders were positioned so they could see every area of the stimulus screen, we ran a calibration protocol (written in Processing v2.2) in which we checked that spider retinae tracked an image of a square moved to each corner of the screen.

Spiders watched three types of stimulus videos: a still fly (n=14), a moving fly (i.e. non-displaying fly that was walking; n=15) and a displaying fly (n=14). We used three different exemplars of the same behaviour of each stimulus treatment. The flies were filmed from the front (i.e. with the fly facing the camera; see Movie 1) for the still and displaying treatments and from the side for the movement treatment. All flies were filmed with a uniform background. Each spider was presented all three fly video types in a randomised order and individual spiders were tested only once on the treatments. Spiders were exposed to a white screen for 1 min followed by 3 min of a stimulus video, with approximately 5 min of blank screen between treatments in order to allow the spider to rest.

To ensure that the stimulus video and retinal positions were aligned in time, we displayed both simultaneously on the same computer monitor and recorded them together in real time using the screen capture software Debut (v3.07). In post-processing, we then overlaid the stimulus video and retinal video using Final Cut Pro (Apple Inc., Cupertino, CA, USA), using the calibration video to ensure that the alignment was correct. These composite videos were exported to Compressor (v.4) and then saved as image sequences (.jpg files) at 10 frames s⁻¹ (see Movie 2 for a sample video). Using the image sequences, we tracked the x,y position of the centres of the left and right retinae and the head of the fly (the stimulus) using the MTrackJ plugin for ImageJ (Meijering et al., 2012) (Fig. 2B,C). We chose the head of the fly as the target as a previous study showed that jumping spiders target the head of the prey (Bartos and Minias, 2016). Only the x coordinates (i.e. horizontal displacement) were used for subsequent analysis because there is little or no vertical displacement of the fly in the videos as well as under natural conditions (see Fig. S1 for analysis of vertical displacement).

The movements of the two principal-eye retinae of P. audax are highly synchronized and spiders direct the high-acuity ‘elbow’ of the boomerang-shaped retinae toward areas of interest (Jakob et al., 2018), so we used the midpoint of the distance between the retinal centres as a proxy for retinal movement. The midpoint was calculated as the centre of the Euclidian distance between the centres of the two retinae.

We used time-series analysis to analyse the data by determining the distance (i.e. separation in pixels) between stimulus and retina coordinates at each time period (i.e. video frame, 10 frames s⁻¹). A distance value of 0 implies that the spider retinae were perfectly tracking the fly's head. Distance data were pooled according to treatments. Because the order of stimulus videos was randomised, there is very little probability that any two spiders would view the fly in the same position at exactly the same time. Missing data (owing to noise in video of retinae) in time and retinal position were dealt with by a third-order Hermite interpolation. We then compared the frequency distributions of distances of all spiders between the three treatments (i.e. display versus moving, display versus still and moving versus still) with a Kolmogorov–Smirnov test (this tests the hypothesis whether the two distributions are drawn from the same population) using Mathematica (Wolfram Research, Inc., Champaign, IL, USA). One outlier in the still treatment was removed owing to a lack of sufficient trackable data.

To quantify and visualise the motion of components of the fly display and movement (including wings and body movements), we used an optical flow algorithm using the ImageDisplacements function in Mathematica to generate a dense motion field. This process compares horizontal and vertical displacement of individual pixels across consecutive frames of a short representative video and computes the magnitude and direction of pixel displacement (Raudies, 2013). Subsequently, these were represented as vectors known as the optical flow vector field. We then used the mean vector over the whole image sequence to present the results in a composite image for a displaying fly and a moving fly.

In this experiment, we simulated the appearance of the fly wing in different angles from the perspective of a jumping spider using psychophysical visual modelling techniques. We took photos of detached fly wings at different opening angles relative to the longitudinal body axis of the fly to measure colour variation during the display behaviour. We staged the opening angles with a wing attached to an entomological pin such that the plane of the wing was perpendicular to the ground in order to simulate different wing angles during a display. All other components of the wing position (i.e. yaw, pitch) were kept constant. We photographed the following wing opening angles against a standard grey background: 90, 100, 110, 120, 130, 140 and 150 deg. We measured the relative area of the pigmented part of the wing, the wing specularity and the WIC at different opening angles in ImageJ.

We took two photos at each angle in the UV and visible spectra (∼300 to 400 nm and ∼400 to 700 nm, respectively) of both wing and head of A. ludens from a front view with an Olympus Pen E-PM2 camera (converted to full spectrum, Lifepixel.com) with a UV-transmitting EL-Nikkor 80 mm f/5.6 lens attached. All the photos were taken in laboratory conditions with an Iwasaki EYE Color arc lamp (70 W 1.0 A; Venture Lighting Europe Ltd, Hertfordshire, UK) as light source. The lamp was modified by manually removing the integrated UV filter, which allows the emission of light in the UV spectrum. We did not use a diffusion filter for the lamp.

The ultraviolet photo was created by using a Baader UV pass and infrared (IR)/visible blocking filter, transmitting from ∼300 to 400 nm, and the visible spectrum photo, using a UV/IR blocking filter, transmitting between ∼400 and 700 nm. Those sets were combined for each angle using the Multispectral Image Calibration and Analysis (MICA) plugin (Troscianko and Stevens, 2015) to create a multispectral image that is a stack of images corresponding to different parts of the spectrum: UV, short wavelength (SW), medium wavelength (MW) and long wavelength (LW).

The multispectral images, with reflectance values in each channel, were transformed to the predicted photoreceptor responses (quantum catch values) for P. audax vision using the MICA plugin. To do this, we modelled the UV and MW photoreceptors of the spider visual system using the standard D65 illuminant, and by taking into account the spectral sensitivities of the camera and the spider. The spectral sensitivity of the camera was calculated previously (sensitivities peaks were UV 369 nm, SW 477 nm, MW 556 nm, and LW 596 nm; see Robledo-Ospina et al., 2017), whereas for P. audax sensitivity, we used previously reported values (UV: 338 nm, MW: 544 nm) for the genus Phidippus (Peaslee and Wilson, 1989; de Voe, 1975). The regression model fitted well with the photoreceptor-mapping model from camera sensitivities (R²=0.999 for both spider photoreceptors).

We created a pseudo-colour image (for details, see Troscianko and Stevens, 2015), which is a subset of the spider vision channels without transformation, with the MW channel shown as yellow and the UV channel as blue (Fig. 5B).

Thus, positive values indicate that the wing is perceived as brighter than the head.

We manipulated the light environment in order to generate different wing appearances from the perspective of the spider. We used an UV-transmitting Teflon sheet as a light diffuser to enhance WICs.

Four light conditions were used for the behavioural assays, taken in a darkroom with artificial full spectrum light (Iwasaki Eye Color Arc, Venture Lighting Europe Ltd, Watford, Hertfordshire, UK). The treatments were chosen to enhance different components of the appearance of the wing and are as follows: (1) control (to enhance perception of pigmented part of the wings): white background, direct light (i.e. no diffuser), light bulb without UV filter; (2) specular reflection (to enhance perception of glossy or shiny part of the wings): black background, direct light, light bulb without UV filter; (3) wing interference colours (WIC) (to enhance perception of wing interference colours): black background, diffuser between the light source and the wings, light bulb without UV filter; and (4) WIC without ultraviolet (to test whether UV plays a significant role in perception of wing interference colours): black background, diffuser, light bulb with UV filter.

With these treatments, if spiders were deterred by the presence of specular reflection or WIC, we would expect the probability of attack would be lower than in the control treatment. Furthermore, if UV played a role in deterrence, we would expect a lower attack rate for spiders encountering flies with enhanced WIC under light with a UV component in comparison to light without a UV component.

In each of these four light conditions, we recorded videos of encounters (n=15 for each treatment) between P. audax and A. ludens inside a Petri dish (15 cm diameter). Although the Petri dish transmitted UV, we have no information about other light features such as polarisation. Before the experiment began, both the spider and the fly were placed inside the Petri dish with an opaque cardboard partition (15 cm in length) in the middle for acclimatisation for 1 min. We then removed the partition and recorded the interaction for 3 min or until the spider and the fly made contact in the first interaction. We only used flies that performed the supination behaviour because previous experiments have shown that non-displaying flies are likely to be attacked at higher rates than displaying flies (Rao and Díaz-Fleischer, 2012). We quantified the number of attacks.

In arena trials, spiders (i.e. a field of view of ±10 deg from the midpoint of the sightline) were oriented directly towards the head of both non-displaying and displaying flies. However, the distribution of the error angles between the two groups was significantly different (display group: n=342 observations, 9 spiders, mean vector µ=351.29 deg, length of mean vector r=0.898, circular s.d.=26.51 deg; non-display group: n=380 observations, 10 spiders, mean vector µ=7.2 deg, length of mean vector r=0.848, circular s.d.=32.84 deg; Watson's U² test, U²=1.59, P<0.001; Fig. 1B,C).

In the eyetracker, the retinae of the principal eyes of the spiders were directed more closely at the heads in videos of moving flies compared with spiders that viewed videos of non-moving flies (Fig. 3A). Spiders facing displaying flies tracked the fly heads with the least precision (Fig. 3A). All distributions were significantly different from one another (Kolmogorov–Smirnov test: display versus moving, D=0.098, P<0.01; display versus still, D=0.085, P<0.05; moving versus still, D=0.072, P<0.05; Fig. 3B).

The different motion components (i.e. a vector flow field) of a displaying fly were substantially different from those of a moving fly (Fig. 4; Movie 3). In general, a displaying fly produced motion in different directions and magnitudes (Fig. 4A) whereas a moving fly produced motion in one direction and at lower magnitudes (Fig. 4B). With this analysis, we noted the relative stillness of the fly's head with respect to the wings and the ovipositor.

We created a composite image to show the different visual components of the fly wing (Fig. 5A). According to our simulation of the fly wing appearance (Fig. 5B), spiders were likely to detect the wing specularity and bands when at a close distance and when the wing was more open (150 deg; Fig. 5C). Only the green and UV components of the total colour information produced by the transparent portion of the wing were likely to be detected by the spider, with information from the red channel unlikely to be recorded as red owing to the lack of a red receptor in this species. At greater distances (∼8 cm), the overall appearance of the fly may be detected but details were not conspicuous. The percentage of wing area covered by WIC, UV and specularity in general followed a hump-shaped pattern (Fig. 6A). The data were best explained by a polynomial fit to WIC (R²=0.82, F=9.61, P<0.05), specularity (R²=0.71, F=5.08, P<0.05), UV (R²=0.91, F=19.54, P<0.0001) and pigment (R²=0.92, F=26.53, P<0.005). The significant fit suggests that information from these components is nonlinear in nature.

The achromatic contrast (difference in luminance between wing and head) varied significantly according to the wing angle (polynomial fit: R²=0.97, F=95.54, P<0.0005). The contrast followed a hump-shaped pattern, with a peak at 110–120 deg (Fig. 6B). A similar pattern was seen with the chromatic contrasts (polynomial fit: R²=0.85, F=33.02, P<0.005; Fig. 6B). These results suggest that visual information as perceived by the spider fluctuates nonlinearly during the wing display.

There was a significant difference in the percentage of flies that were attacked in different light conditions (generalized linear regression; logit link function, binomial distribution, χ²=8.09, d.f.=3, P<0.044). With respect to the control, flies in the WIC with UV and the WIC without UV treatments were attacked significantly less (post hoc contrasts; z=2.11, P<0.05 and z=2.33, P<0.05, respectively), while there was no significant difference between the specularity treatment and the control (z=−1.2, P=0.23; see Supplementary Materials and Methods for the complete model).

In this paper, we bring together three different lines of evidence that suggest that the tephritid fly's wing display generated a multitude of motion and colour cues which contributed to the hesitation by jumping spiders to attack. Using data from orientation in untethered spiders, retinal tracking in tethered spiders and visual modelling of wing appearance, we have presented a broad picture of a deterrent visual display. The deterrent effect of a tephritid fly's display against a jumping spider predator has been previously attributed to the mimetic markings on the fly's wings (Greene et al., 1987). According to this hypothesis, spiders misidentify flies as other salticid spiders. However, salticids can be deterred even when the wing markings are artificially blackened out, suggesting that the wing motion has an important deterrent effect (Rao and Díaz-Fleischer, 2012). Our results provide a mechanism that can explain this deterrence. The hesitation to attack may be attributed to the failure by the spider to accurately track the movements of the fly. In the arena experiments, there is a fluctuation in the spider's orientation when the fly is displaying, while the retina analysis showed that there was a significant lack of precision in tracking the motion of the fly. These subtle shifts could cause the spider to slow or break off its attack, thereby allowing a displaying fly to survive an encounter with a predator (Rao and Díaz-Fleischer, 2012). In addition, the fly's head does not move as much as the ovipositor and wings when the fly is displaying, perhaps to reduce the chances of the head being detected and to attract attention to other, less vital, components. This behaviour is consistent with results from experiments on other jumping spider species that use the head component of potential prey to target attacks (Bartos and Minias, 2016). From the fly's point of view, reducing the motion of its head could allow the fly to gather more information about the nature of the threat.

To quantify the potential motion cues available from the fly movements, we calculated the optical flow of the movement from representative videos (Movie 3). The resulting vector field from the optical flow shows stark differences in magnitude and direction between a displaying fly and a moving fly. A displaying fly generates motion components in various directions and magnitudes, which may make it more difficult for the spider to quickly identify the fly as prey. A moving fly, in contrast, generates a uniform and single direction flow, which potentially makes it easier for the spider to assess the best attack angle.

The hyaline parts of wings have been recently shown to produce bright structural colouration, and it has been suggested that predators may be dissuaded from attacking insects sporting such colours (Pike, 2015). In the jumping spider–tephritid system, there are special features that we need to consider. Firstly, jumping spiders are very sensitive to movement (Drees, 1952); their principal eyes have colour vision, but their anterior lateral eyes are ‘motion detection’ eyes (Jakob et al., 2018). Furthermore, orientation behaviour in salticids has been shown to be mediated by the anterior lateral eyes and is influenced by the speed and size of the stimuli (Zurek et al., 2010). The supination display of the fly produces a series of motion information: the wings themselves, the changing contrasts in multiple chromatic and achromatic perspectives as the wings are extended, and the movement of the head and legs as the display is performed. Note that because we only used one wing for the visual modelling, the number of colour and movement cues produced by the fly display is likely to be doubled under natural conditions.

The behavioural assays showed that spiders are less likely to attack flies in the WIC treatment, though they are likely to use only achromatic information in targeting prey. We detected no effect of UV on the likelihood of attacks by spiders. However, we note here that a limitation of this study is that the control treatments were with a white background whereas the other treatments had a black background, which could affect the results given that there may be poorer detectability against black backgrounds. As these spiders are dichromatic, with spectral sensitivity in the UV and green wavelengths, we suggest that only achromatic contrast is being used to target prey, i.e. the spiders possibly see the colour patterns as a ‘grayscale’ gradient (de Voe, 1975). A similar effect may be operating in the red jumping spider Saitis barbipes, which does not possess a red receptor (Glenszczyk et al., 2022).

Our results may be extended to other forms of displays. Male jumping spiders use elaborate courtship displays with an abundance of motion components, and mating success is correlated with the complexity of courtship display in at least some groups (Girard et al., 2015). Given that courtship can be dangerous for males, it may be that some complex visual displays are difficult for females to quickly assess, and thus buy time for a courting male to approach. These ideas need to be tested further.

We thank Pablo Núñez Berea for help in collection of spiders in Mexico. Mary Emma Searles and Ashley Carey helped in the collection of retina data at University of Massachusetts Amherst. We thank Nathan Morehouse and David Outomuro for helpful discussions. We thank Jolyon Troscianko and two anonymous reviewers for comments that greatly improved an earlier draft of this paper. The eyetracker was designed by Cristina Canavesi and Jannick Rolland (University of Rochester) and built by Stingray Optics (Keene, North Hampshire).