Independence and synergy of spatial attention in the two visual systems of jumping spiders Free

Animals are continuously bombarded with sensory input, to the point that it would be impossible with their limited cognitive resources to compute every stimulus, especially as only a handful of them may eventually be relevant, requiring a behavioural response. To possibly face this evolutionary pressure, a set of mechanisms known as ‘attention’ developed, enabling animals, among other advantages, to selectively direct their cognitive resources toward relevant stimuli or parts of their environment, while filtering out irrelevant information (Evans et al., 2011). Similar attentional mechanisms appear to be widespread among different species and even clades, including mammals (Bowman et al., 1993; Marote and Xavier, 2011; Rizzolatti, 1983; Wagner et al., 2014), birds (Johnen et al., 2001; Lazareva et al., 2012; Shimp and Friedrich, 1993; Sridharan et al., 2014), amphibians (Greenfield and Rand, 2000; Tárano, 2015), fishes (Gabay et al., 2013; Parker et al., 2012) and even invertebrates (Earl, 2023; Eckstein et al., 2013; Humphrey et al., 2018; Morawetz and Spaethe, 2012; Wiederman and O'Carroll, 2013; Winsor et al., 2021). In this work, we investigate a particular instance of attention, i.e. visual spatial attention in a jumping spider (Menemerus semilimbatus).

Individuals are generally in control of which stimulus to focus on; for instance, when we want to concentrate on a specific stimulus in the environment, we shift our gaze towards it. This movement puts the eyes' fovea on the target, maximising the collectable information in that specific area of the visual field. This mechanism is termed in the literature ‘overt attention’, referring to the fact that the focus of attention of the individual matches their gaze direction (Yantis and Jonides, 1984). It has, however, been observed that animals can pay attention to specific portions of the visual periphery, without having to shift their gaze. This happens, for example, in the presence of an unexpected stimulus with a relatively high intensity, such as a brief flash of light. This allows individuals to quickly react to subsequent unexpected environmental changes, without having to abandon the current target of overt attention (Posner, 2016). This process is termed ‘covert attention’, referring to the fact that the attentional shift happens without any visible change in behaviour. This type of attention is less precise in processing fine details (as it does not correspond to the fovea position), yet it enhances monitoring and subsequent responses (for instance, subsequent eye movement) to the attended location. Overt and covert attention are two distinct mechanisms, ascribed to two separate cognitive processes (Hunt and Kingstone, 2003a,b), yet they work synergistically, allowing the individual a more efficient distribution of cognitive resources.

During the last decades, jumping spiders have been established as an interesting model for the study of vision, having been found capable of producing complex behaviours in response to visual stimulation (Cross et al., 2020; De Agrò, 2020, 2017, 2021; Dolev and Nelson, 2014, 2016; Harland and Jackson, 2000; Mannino et al., 2023; Rößler et al., 2022,b). Jumping spiders offer the chance to draw a direct link with the psychological construct of separated and dedicated attentional mechanisms for stimuli detection and processing, thanks to the unique modular organisation of their visual system (Foelix, 2011; Harland and Jackson, 2000; Land, 1985; Menda et al., 2014; Morehouse, 2020; Zurek and Nelson, 2012a; Zurek et al., 2010). Jumping spiders' visual system is split across four different eye pairs (Chong et al., 2024; Winsor et al., 2024). The two large anteromedial eyes (AMEs, ‘principal eyes’) are characterised by a narrow visual field (∼5 deg for each eye), but are characterised by the highest spatial acuity among all pairs (Blest et al., 1981; Foelix, 2011; Jackson and Cross, 2011; Land, 1969a,b; Morehouse, 2020; Zurek et al., 2015). The long eye tubes of these principal eyes can be moved by the spider thanks to a set of muscles (Land, 1969b). These characteristics together suggest that the AMEs are specialised in the detection of fine details and figure recognition (Harland et al., 2012). The other three pairs of eyes, anterolateral, posteromedial and posterolateral eyes (ALEs, PMEs and PLEs, or ‘secondary eyes’), are unmovable, and are characterised by a poorer spatial acuity but a much wider visual field, covering a combined ∼350 deg around the spider. These eyes are specialised in motion detection and recognition (Bruce et al., 2021; De Agrò et al., 2021; Duelli, 1978; Fenk and Schmid, 2010; Jackson and Cross, 2011; Jakob et al., 2018; Spano et al., 2012; Zurek and Nelson, 2012a; Zurek et al., 2010). Among secondary eyes, the ALEs are the biggest two and are forward facing, covering the area between ±60 deg. PLEs instead point backwards, each covering from the end of the ALE field to around ±170 deg. PMEs are the smallest and considered to be vestigial for many, but not all, salticids (Land, 1985). The different eyes project into anatomically separate brain regions (Steinhoff et al., 2020), and are seemingly independent from each other, apart from late high-level integration (Strausfeld and Barth, 1993; Strausfeld et al., 1993).

The division of labour between the eyes is also reflected in the spider's behaviour. Whenever a moving object passes across the fields of view of the secondary eyes, the spider can detect it. In this case, the animal may perform a full body pivot, rotating the exact angle necessary to face the target frontally (Zurek et al., 2010). These rotations are very robust: even stimuli without any resemblance to natural objects consistently stimulate pivots (Zurek and Nelson, 2012b; Zurek et al., 2010). In contrast, when multiple objects are simultaneously in the field of view of the secondary eyes the pivot can be selective, prioritising the most relevant object to face according to different characteristics (De Agrò et al., 2021; Spano et al., 2012). Once the target is directly in front of the spider, the AMEs begin a ‘scanning’ process, by which they move their retinas across the whole target, possibly to process its shape and characteristics (Land, 1969a). If the target moves, the spider can smoothly follow (i.e. track) it using the AMEs, aided by the ALEs, which, having a wider visual field, can keep track of the changing location of the target (Jakob et al., 2018; Land, 1969a).

We propose that jumping spider orienting and scanning behaviour is an instance of overt attention, as it is possible to infer where the animal is allocating attentional resources by the direction of its gaze. In this situation, the spider actively orients its gaze to a desired location, so that the stimulus falls in the area of the full field of view with maximal spatial acuity (i.e. the fovea in vertebrate system, the centre of AMEs in spiders). Conversely, as it is not possible to observe a behavioural manifestation of covert attention, it remains unclear whether covert attentional mechanisms are present to aid target detection by the secondary eyes. In other words, it is yet to be determined whether detection and recognition by the secondary eyes is equally spread across the totality of their visual field, or whether the spiders can instead selectively allocate resources to a specific portion of this wider visual field.

In the present study, we aimed to investigate spatial attention mechanisms in the jumping spider modular visual system. To this end, we adapted a behavioural paradigm commonly used to study spatial attention in humans and other vertebrates (Bowman et al., 1993; Bushnell et al., 1981; Golla et al., 2004; Marote and Xavier, 2011; Wagner et al., 2014): the spatial cuing task, also known as the Posner task, named after its inventor (Posner et al., 1980). In the classical paradigm, two stimuli are employed: a cue and a target. Subjects are required to only respond to a specific target stimulus, while ignoring another stimulus irrelevant to the task, namely the cue, that is briefly presented before the target. Interestingly, the cue could appear either in the same location where the target will appear (valid trials) or in a different one (invalid trials). The task-irrelevant cue has been shown to cause an automatic and involuntary shift of the covert attentional focus toward the location where it appeared: subjects become faster and more accurate in locating the target in valid trials, at the cost of lower speed and accuracy in invalid trials.

In a first experiment, we investigated whether it is possible to prime covert attentional monitoring in the secondary eyes using a sudden visual stimulation. We presented jumping spiders with a spatial cue (i.e. a brief flash of light) to the right or left side on a computer monitor, followed by a target stimulus (i.e. a black dot moving vertically) in the same or opposite position. If the presentation of the cue can enhance the attentional system via the secondary eyes, we hypothesised a reduction in reaction time and/or an increase in body pivot frequency toward the target in valid (i.e. target appearing in the same location as the cue) versus invalid (i.e. target appearing in the opposite location to the cue) trials. In other words, we predicted that a flashing cue may have an ALE-specific priming effect on the detection of the target, when presented in the same spatial location. This would indicate that the secondary eyes could benefit from a covert attentional mechanism for which animals could be pre-alerted to allocate cognitive resources on a specific portion of the space rather than be constrained to equally spread their attention along the entire visual field.

In a second experiment, we investigated whether the principal eyes' overt attention towards a certain portion of space can influence the secondary eyes' detection of stimuli, by enhancing their covert attention in the area near the visual fields of the AMEs. To do so, we presented the spiders with two types of moving cues to stimulate AMEs tracking either toward the left or the right hemispace. After the cues reached either side of the monitor, they disappeared, and this was followed by the appearance of a target either in the vicinity of the location from which the cue disappeared (valid trials) or in the contralateral field (invalid trials). If the two eye systems work synergistically, we hypothesised that directing the overt attention of the AMEs to a certain spatial location would also result in enhanced covert attention in the ALEs. In a Posner task paradigm, this would consist of a reduction in reaction time and/or an increase in detection rate toward the target in valid trials versus invalid ones. In other words, we predicted that a moving cue presented to the AMEs may synergistically prime the ALEs to detect a novel stimulus in a similar spatial location.

A total of 136 Menemerus semilimbatus Hahn 1829 were used in the experiments, with 65 spiders used in experiment 1 and the remaining 71 used in experiment 2. All animals were collected from the wild, from the park outside the Esapolis' Living Insects Museum, Padua, Italy, between March and September 2021. Once the spiders had been caught, we applied to the cephalothorax a cylindrical, 1×1 mm neodymium magnet, using UV-activated resin. This magnet was needed to fix the animal to the apparatus (see the next section). The spiders were then housed individually in plastic boxes and tested for the first time the next day. At the end of the experiment for each animal, the magnet was removed and the subject was released back into the wild.

Both experiments were performed using a spherical treadmill, as fully described in De Agrò et al. (2021). The spider was attached to the end effector of a 6-axis micro-manipulator. The magnet glued to the spider's cephalothorax was connected to an opposite-polarity magnet on the end effector. Just below, a polystyrene sphere was present. This ball was maintained suspended by constant streams of compressed air, to make its motion frictionless. By adjusting the micromanipulator, the spider was moved towards the sphere, until it contacted it with its legs. In this configuration, the spider's orientation remained fully fixated, while it could still impress its intended movements on the sphere. By recording and extracting the rotation of the sphere, using the software FicTrac (Moore et al., 2014), we could infer the animal’s motion direction and speed. The full apparatus was put at a distance of 250 mm from a computer screen, directly in front of the spider's visual perspective. The monitor had a 1080p resolution, a refresh rate of 60 Hz and a pixel pitch of 0.248 mm. The stimuli were presented on the monitor at 30 frames s⁻¹, at 30 deg left or right from the centre of the visual field. The recording and the stimuli presentation were programmed in Python 3.8 (Van Rossum and Drake, 2009), using the packages PsychoPy (Peirce et al., 2019), OpenCV (Bradski, 2000) and NumPy (Oliphant, 2006; van der Walt et al., 2011). As stated in the Introduction, jumping spiders perform full-body pivots towards a target perceived in the secondary eyes' visual field, to observe it with the principal eyes. By observing the z-axis rotations of the sphere, we were able to extract the spiders’ intended rotations (i.e. body pivots) towards each stimulus.

In our first experiment, we inquired about the effect of a peripheral cue presented in the visual field of the secondary eyes on the speed and probability of detection of the target by the same eyes. To start, the background of the monitor was set to grey, apart from two ‘windows’, at ±30 deg from the centre, respectively. The centre of these windows was just outside the field of view of the AMEs, even including gaze shifting maximal span, but still inside the visual fields of the ALEs (De Agrò et al., 2024; Land, 1969a). These windows had a white background and were 20 deg wide and 20 deg tall (Fig. 1).

After 3 min of habituation, in which the spiders observed this static background while harnessed to the treadmill apparatus, the first stimulus appeared. One of the two windows would ‘flash’, alternating between black and white for 0.15 s at a speed of 30 Hz. This flash would act as a spatial cue, aiming to increase the spider's covert attentional monitoring toward that portion of the space. After the cue disappeared, a random time interval between 0.2 and 5 s would elapse. Randomisation was performed logarithmically: larger intervals were sparser than shorter ones. This was done on the assumption that the interval effect is logarithmic itself, like many psychophysical effects (Portugal and Svaiter, 2011). After this interval, the target stimulus was presented: this consisted of a 4 deg wide black circle, moving up and down in alternating frames for a span of 2 deg, for a total of 0.66 s (Fig. 1A). This target could appear at the centre of one of the two lateral windows, either the one previously pre-alerted by the cue (i.e. valid trials) or the opposite one (i.e. invalid trials). After the target disappeared, a 25 s pause was presented, followed by the next cue–interval–target set. In a full session, the spiders were presented with a total of 12 trials.

We also designed two control conditions. In the first condition, the spiders were presented directly with the target stimulus, without this being preceded by the flashing light cue. This was used to collect a baseline measure of response rate and speed. In the second condition, the spiders were presented with the cue only, without it being followed by a target. This was used to check whether the cue alone was sufficient to elicit body pivots. It is possible in fact that in the main condition, the spiders would perform a pivot towards the cue but slow enough to be mistakenly recorded as a fast reaction to the subsequently presented target. By testing the reaction to cue alone, we could control for this possibility. The trials of these two conditions also started with the 3 min of habituation and presented a total of 12 trials. Each spider was subjected to all three conditions, in three different sessions in random order, on the same day.

In our second experiment, we investigated whether directing the AMEs’ attention toward a spatial location could favour stimulus detection in the visual field of the secondary eyes in that same portion of space. To this aim, we presented a central spatial cue in the visual field of the AMEs. This was designed to trigger a visual shift in the AMEs, toward either the left or right. We designed two different versions of such cues for two separate experimental conditions. Both conditions maintained the same overall structure as the first experiment (background colour, window position, initial habituation section).

In the first condition, the cue, named ‘moving dot’, consisted of a 4 deg wide circle, appearing in the centre of the screen and moving toward one of the two windows, for a span of 15 deg, at a speed of 15 deg s⁻¹, before disappearing again (Fig. 1B).

In the second condition, the employed cue was named ‘five dots’. It consisted of five white dots of 4 deg diameter at the centre of the visual field. The dots were equally spaced, 7.5 deg from each other, ranging from −15 deg to +15 deg. These five dots remained visible on the screen for the whole duration of the experiment. The cue consisted of a change in luminance of the left or right circle, from white to black. After 0.15 s, the dot switched back to white and the adjacent dot turned black. This proceeded until the black dot reached the opposite side, leaving only white dots. Notably, the visual field of the AMEs covers only 10 deg (5 deg from each side to the centre); each dot appeared at 7.5 deg distance from the previous one, forcing the spider to use the ALEs for detection and then re-anchor the AMEs' fields onto it (Fig. 1B). This created a substantial difference with respect to the ‘moving dot’ condition, where the spider could follow the target smoothly with the AMEs, as it moved for less than 1 deg per frame, always remaining in view. Moreover, different from the first condition, the first dot shifting from white to black was not centred but already oriented in space (i.e. either to the left or to the right). As such, the probability of the AME starting with the dot in the visual field decreased even more. Overall, this condition was intended to maintain the same indication towards one of the two sides as the previous one, but without stimulating the AMEs, forcing the spiders to rely on the ALEs' visual fields.

In both conditions, following a random time interval (see experiment 1), a target appeared either in the window on the same side that the cue disappeared from (i.e. valid trial) or the other one (i.e. invalid trial). We expected the cue to be tracked by the spider's AMEs, causing a shift of the animal's overt attention to a specific direction. Importantly, it was not possible for the spider to see the two lateral windows with the AMEs, as they were positioned outside their visual fields. As such, any stimulus appearing in the two lateral windows had to be detected by the secondary eyes.

The rotations of the spherical treadmill, collected with the software FicTrac (Moore et al., 2014), were further analysed through an algorithm developed in Python 3.8 (Van Rossum and Drake, 2009) (see Dataset 2).

We analysed z-axis rotations in the 5 s after the appearance of each stimulus (i.e. the cue or the target). Cumulative changes in absolute orientation of at least 20 deg (the angle was chosen to be consistent with the stimulus position at 30 deg, but still accounting for errors) in a direction consistent with the position of the stimulus (e.g. if the stimulus was to be located on the left, we would expect clockwise rotations of the sphere) were considered as body pivots toward that stimulus.

We also scored the latency of movement by calculating the time that elapsed between the appearance of the stimulus and the beginning of the rotation. For experiment 1, in the ‘cue only’ condition, the latency was calculated from the presentation of the cue, whereas in the two other conditions, it was calculated from the presentation of the target. In these latter cases, we also checked for rotations occurring from the presentation of the cue up to 5 s after the target appearance. Regarding experiment 2, we calculated latency only following the presentation of the target.

All analyses were performed in R4.1.2 (http://www.R-project.org/), using the packages readODS (https://CRAN.R-project.org/package=readODS), glmmTMB (Bolker et al., 2009; Brooks et al., 2017), car (Fox and Weisberg, 2019), DHARMa (https://CRAN.R-project.org/package=DHARMa), emmeans (https://CRAN.R-project.org/package=emmeans), ggplot2 (Wickham, 2009) and reticulate (https://CRAN.R-project.org/package=reticulate). Raw data are available in Dataset 1.

For the first experiment, we initially observed the differences between the three conditions, in terms of both response probability and response delay. For the pivot probability, we employed a generalised linear mixed model (GLMM) with a binomial error structure. We also included session and trial number: we expected to observe a decrease in pivots with the passing of time, as it has been observed that spiders' response rate decreases with successive presentation (Beydizada et al., 2024; Humphrey et al., 2018, 2019; Melrose et al., 2018). We included the identity of the subjects as a random intercept, and trial number as a random slope. After evaluating the goodness of fit for the model, we observed the effect of the predictors using an analysis of deviance, and subsequently performed a Bonferroni-corrected post hoc analysis on factors that had a significant effect. When analysing the response delay, we followed the same procedure, but used a Gaussian error structure. Notably, the analysis was performed on the log of the delay, rather than on the raw value. This decision followed the same logic we used when deciding the time intervals between cues and targets: in the realm of psychophysics, the magnitude of change increases logarithmically, rather than linearly (Portugal and Svaiter, 2011; see Dataset 2 for the distributions).

After observing the difference between conditions, we concentrated on the effects in the main experimental condition, where both the cue and the target were presented. We again analysed both the probability of response and the response delay, using the GLMM binomial and Gaussian error structures, respectively. As predictors, we included trial validity (cue in the same or opposite spatial location to the target) and time interval (between the appearance of the cue and the appearance of the target), in addition to trial and stimulus number as before. We included the identity of the subjects as a random intercept, and stimulus number as a random slope. The analysis then followed the same procedure described above.

The analysis for the second experiment followed the exact same structure, observing as dependent variables both the probability of pivots and the response delay. For both models, we included as predictor the experimental condition (‘moving dot’ or ‘five dots'), the validity of the trial, the time interval and the stimulus number (the trial number was not included because it was shown to have no effect for this experiment, see Dataset 2).

As expected, we found an overall effect of both the trial number (GLMM analysis of deviance, χ²=13.939, d.f.=1, P=0.0002) and the session number (χ²=32.947, d.f.=2, P<0.0001) on the probability of producing a pivot toward any stimulus. Specifically, the response probability remained the same in sessions 1 and 2 (corrected post hoc Bonferroni, estimate=0.0139, s.e.=0.124, t=0.121, P=1), while for session 3 it was lower than for the other two sessions (1 versus 3, estimate=0.707, s.e.=0.138, t=5.131, P<0.0001; 2 versus 3, estimate=0.0.693, s.e.=0.137, t=5.07, P<0.0001). Across trials in each session, response probability significantly decreased (trend=−0.056, s.e.=0.0153, t=−3.633, P=0.0003).

Regarding the differences between the three conditions, the response probability across all trials was the same for the ‘target only’ condition and the ‘both cue and target’ condition (post hoc Bonferroni corrected, odds ratio=1.0656, s.e.=0.274, t=0.247, P=1). In contrast, both conditions presented a higher response probability relative to the ‘cue only’ condition (both cue and target versus cue only, odds ratio=31.585, s.e.=10.7, t=10.191, P<0.0001; cue only versus target only, odds ratio=0.0337, s.e.=0.0116, t=−9.823, P<0.0001). Similarly, the conditions of ‘both cue and target’ and ‘target only’ were characterised by a faster response delay over the ‘cue only’ condition (both cue and target versus cue only, ratio=0.151, s.e.=0.0164, t=−17.347, P<0.0001; cue only versus target only, ratio=5.075, s.e.=0.5933, t=13.897, P<0.0001). The condition ‘both cue and target’ also had a faster response rate than the ‘target only’ condition, although the difference was much smaller (ratio=0.765, s.e.=0.0709, t=−2.885, P=0.0113). These results suggest that the cue alone does not trigger a pivot, for which the target is required instead.

Regarding the response delay in the main condition (Fig. 2A), we found no effect of the time between the cue and the target appearance (GLMM analysis of deviance, χ²=0.3866, d.f.=1, P=0.534). Instead, there was an effect of trial validity (χ²=7.9906, d.f.=1, P=0.0047). Specifically, spiders showed faster responses (0.505 s versus 0.584 s) during invalid trials over valid ones (post hoc Bonferroni corrected, ratio=0.866, s.e.=0.0432, t=−2.894, P=0.0041).

The response probability (Fig. 2B) was strongly influenced by both the validity of the trial (GLMM analysis of deviance, χ²=17.8217, d.f.=1, P<0.0001) and the interaction between trial validity and the time between the cue and target appearance (χ²=5.2378, d.f.=1, P=0.0221). Specifically, the response probability was higher for invalid trials than for valid ones (post hoc Bonferroni corrected, odds ratio=2.58, s.e.=0.628, t=3.889, P=0.0001). Furthermore, while the response probability remained constant over time for invalid trials (trend=−0.108, s.e.=0.157, t=−0.69, P=0.4905), it increased significantly for valid ones (trend=0.471, s.e.=0.165, t=2.853, P=0.0045), starting from a much lower response for short intervals and reaching an equal probability of invalid trials for longer ones.

For the second experiment, we found no differences between the two sessions in terms of response probability (GLMM analysis of deviance, χ²=2.61, d.f.=1, P=0.106). This is consistent with the result of experiment 1, as the difference observed there was only evident in session 3. We also found an overall effect of trial number independently of condition (χ²=76.497, d.f.=1, P<0.0001), as previously observed.

Regarding the response delay in both conditions (Fig. 3A), we found an overall effect of the validity of the trial (GLMM analysis of deviance, χ²=4.756, d.f.=1, P=0.0292). There was no effect of the time interval between the cue and target (χ²=0.8371, d.f.=1, P=0.3602), nor did we observe any differences between the two conditions (χ²=0.651, d.f.=1, P=0.4197). Specifically, spiders appeared to be marginally faster for invalid trials than for valid ones (post hoc, ratio=0.87, s.e.=0.0625 t=−1.941, P=0.053).

Regarding the probability of response (Fig. 3B), we found a difference between the two conditions (GLMM analysis of deviance, χ²=15.618, d.f.=1, P<0.0001), but no effect of trial validity (χ²=0.6137, d.f.=1, P=0.433) nor of the time interval between cue and target (χ²=3.5913, d.f.=1, P=0.058). Specifically, spiders had a much higher probability of responding to targets in the ‘moving dot’ condition over the ‘five dots' one (post hoc, odds ratio=0.254, s.e.=0.0784, t=−4.482, P<0.0001).

In this paper, we investigated the presence of a covert visual attentional mechanism in jumping spiders, and whether this relies on a synergistic activation of both principal and secondary eye systems.

In the first experiment, we investigated whether a spatial cue presented to the secondary eyes could affect the detection of a target stimulus also presented to the same eyes.

It is crucial to point out that in our experiment, the stimuli that we selected as cue and target are both unnatural, and as such probably bear the same ecological value (if any) for the spiders. The target mimics more closely a natural type of stimulus, being composed of a moving object translating across the screen, exploiting the innate tendency of spiders to pivot towards it. However, the flashing stimulus also simulates a type of motion in its most basic form, being composed of pixels changing luminance across time, and we cannot fully exclude that the spiders could have detected it as a moving target. For this reason, there is no infallible a priori reason why the animals should have considered the moving dot a ‘target’ and the flashing light a ‘cue’. This is of course not an issue where the Posner's paradigm is administered to humans, as in this case the task can be verbalised and the experimenter can explain which stimulus the participant should look for. Here, we ensured the correct interpretation of the two stimuli with the two control conditions: the ‘cue only’ and ‘target only’. The very low response rate in the ‘cue only’ condition demonstrates that the flashing stimulus alone is insufficient to trigger a pivot response, differently from what happens with the moving dot alone. This indicates that the spiders do not treat the flashing light as a meaningful ‘target’, which they do for the moving dot. However, it is possible that the flashing stimulus was completely irrelevant for the animals. Our results seem to also exclude this possibility, as even though the response rate in the ‘target only’ and ‘both cue and target’ conditions was identical, the response delay changed significantly, demonstrating an effect of the cue's presence.

In accordance with the literature on vertebrates (Posner, 2016; Posner et al., 1980), we hypothesised that in the ‘both cue and target’ condition, a spatial cue would support performance when appearing in the same position as the target (valid trials) and hinder it when appearing in the opposite hemispace (invalid trials), in terms of both probability of detection and response times. We observed, contrary to our expectations, that subjects performed better in invalid trials, where the probability of detecting the target remained stable across time intervals and spiders were overall faster in responding than for valid trials. In these latter trials, the probability of response was affected by the time interval, with low scores for short intervals (around 0.2 s) and an increasing improvement until the performance in the invalid trials was matched in longer delays (around 5 s).

The initially unexpected drop in performance for valid trials could be explained with the split role of spider AMEs and ALEs, and the possibly split nature of attention between these two sets of eyes. As stated in the Introduction, following a first detection by the ALEs, there should be an overt shifting (i.e. the spider performs a pivot to orient toward the stimulus) of the AMEs. These eyes would then anchor to the stimulus detected by the ALEs and start a detailed scan. We hypothesise that, to maximise the efficiency of the allocation of attentional resources, after the initial detection of the stimulus, the ALEs will ‘dis-anchor’ from the previously detected stimulus and redirect their (covert) attention to a different portion of the space. Note that this may be achieved without dis-anchoring the (overt) attention from the target currently being scanned by the AMEs, and without producing any pivot, but rather (covertly) focusing on a new section of the space with the ALEs. At the same time, the ALEs would also be inhibited from re-anchoring attention on the previously attended hemispace, avoiding an overlap with the area currently attended by the AMEs, and maximising the efficiency in detecting novel stimuli.

In the case of our experiment, this resulted in a worsening of detection in valid trials (in terms of both probability of detection and reaction times), while maintaining an optimal performance in invalid ones, as ALEs would still be responsive to portions of the space other than that occupied by the cue. Note that when the spiders did respond to the targets, they were significantly faster with respect to when no cue was presented at all, and consequentially no covert attentional shift of ALEs occurred. This shows that shutting off a portion of the visual field is indeed advantageous for performance. This ‘attentional blindness’ of the ALEs with respect to the previously inspected spatial location probably has a limited duration, as the performance is restored after longer delays between the cue and the target, in the order of 5 s.

In the ‘moving dot’ condition of the second experiment, we addressed whether a translating visual cue presented to the AMEs could improve detection of a spatial target in the visual field of the secondary eyes. The cue appeared in the centre of the screen, where it could be detected by the AMEs. We know from previous literature that smooth tracking of a target with AMEs is directed by the ALEs (Zurek and Nelson, 2012a). As such, while the cue moved across the screen, we can assume that the attention of both AMEs and ALEs shifted with it. Similar to experiment 1, spiders responded faster to invalid trials than to valid ones. This is again contrary to our initial hypothesis, yet it well fits our interpretation of the results in terms of the ALEs being hindered in maintaining visual attention in a previously inspected spatial location.

In terms of reaction times, spiders were overall faster in invalid versus valid trials, hinting at an enhanced attention in the previously unattended location (or, conversely, a reduced attention in the already inspected one). However, while this effect was found to be significant in the ANOVA, it was lost in the post hoc analysis. This is probably a result of the low response rate in experiment 2, for which, despite the number of subjects being equal, we had fewer observations to include in the analysis. Differently from experiment 1, when considering the probability of response, we did not find a difference between valid and invalid trials. This could be explained as being a result of methodological differences between the two experiments. In fact, in experiment 1, the target appeared at exactly the same location as the cue, hence being affected by the hypothesised attentional blindness of the ALEs; however, in experiment 2, valid trials presented the target stimulus in the same direction as the cue (left or right) but not in the exact same spatial position. As such, the target stimulus could be still considered as novel rather than completely overlapping with the cue, and consequently deemed relevant and attended to, even if with slower reaction times as a consequence of the attentional shift.

Interestingly, we found a difference between the ‘moving dot’ and the ‘five dots’ conditions, the latter scoring a significantly lower response rate. We interpreted this as being due to a different involvement of the principal and secondary eye systems. In the ‘moving dot’ condition, as stated before, the AMEs could smoothly track the cue toward the left or right hemispace. Therefore, we can assume that the attentional focus of the ALEs moved together with that of the AMEs in the cued direction as follows: (i) detection of the cue by the ALEs; (ii) subsequent involvement of the AMEs for sustained attention; (iii) attentional tracking of the moving cue by both the AMEs and the ALEs; (iv) detection of the target by the ALEs. However, in the ‘five dots’ condition, the cue could not be smoothly followed, as it did not proceed along a continuous movement, but rather haltingly appeared and disappeared, progressively shifting toward a certain spatial location. Compared with the ‘moving dot’ condition, this would require repeated effort to detect the dot reappearing in the novel location and re-anchoring the attention of the AMEs and ALEs to a newly appeared object a total of six times for each trial (five times for the five dots acting as a cue, plus one additional time for the target). This could constitute a highly demanding task for the spiders, resulting in the observed drop in response rate. The previously hypothesised re-anchoring refractoriness to previously attended hemispaces might have also played a role in further enhancing this effect. Similarly, we observed a reduction in the response rate between trials within the same session in both our experiments. This is also in line with previous literature showing that jumping spiders' detection rate drops following repeated stimulus presentation (De Agrò et al., 2021; Humphrey et al., 2018, 2019; Melrose et al., 2018).

Overall, our study hints at the presence of two independent systems for visual attention, for which the jumping spider's principal and secondary eyes work independently, but synergistically, in attentional monitoring and stimulus detection. When the ALEs detect a stimulus in space, the spider produces a pivot and the AMEs start a scanning process of it. This frees attentional resources of the ALEs with respect to that portion of space, allowing for monitoring of different locations not overlapping with the current attentional focus of the AMEs. In our study, it was not possible for spiders to coordinate the attentional focus of the two systems, as the animals were constrained on a fixed position on the sphere, leading to the discussed poor performance in valid trials, as the AMEs could not overtly direct attention to the cue location while the ALEs were covertly scanning different portions of the space. However, at the same time, we could observe an advantage in invalid trials (experiment 1) for which spiders were faster in responding to the target when it was followed by the cue compared with when the cue was not presented. This is consistent with our hypothesis: the spider narrows its attentional focus to spatial locations that were not previously inspected. Indeed, even though the split attention mostly hindered the spiders’ performance in our experiment, it would probably be highly advantageous to them in a natural context.

Bruce et al. (2021) demonstrated that a distractor presented to the ALEs can cause a gaze shift of the AMEs, even when these were already scanning a previous stimulus. The probability of causing such a gaze shift is linked to the survival relevance of the new distractor, having to be higher than that of the scanned target. This means that even during AME scanning, ALEs are still computing other stimuli (otherwise the distractor could be detected but its valence could not be determined), and are capable of driving a selective attention shift, even when AMEs are in use. In our experiment, we demonstrated that this extends to spatial attention, pertaining not only on the stimulus nature but also on its location. It remains to be seen how the two dimensions of stimulus characteristics and location interact with one another. In our study, we employed an extremely impoverished version of natural stimuli, aiming at describing spiders' response to a minimal amount of information. Future work might better clarify how the attentional response can be further modulated by multifaceted characteristics of the stimuli, such as their ecological relevance or complexity.

Given our results and the wider literature, we suggest that jumping spiders have employed an ingenious solution to their limited brain resources, by splitting attentive resources between the two eye systems. However, they maintain synergistic coordination, to avoid wasting resources in attentional overlaps. What remains unknown is to what extent these two systems remain separate and how much communication is possible between the two.