Prior cueing affects the saccadic response to targets in the praying mantis Sphodromantis lineola Open Access

Attention is an essential process by which animals filter out mu­ltiple stimuli to focus on the ones of interest (Johnson and Proctor, 2004; Wolfe and Horowitz, 2004). Spatial attention, the allocation of attention to particular regions in space, has been a particularly well studied aspect of attention and the visual properties that guide it are well established, especially in humans and non-human primates (Chica et al., 2013; Krauzlis et al., 2013; Posner, 1980; Spence and Santangelo, 2009). Yet other animals, including insects, would also benefit from attentional processes – think of a bee choosing one amongst many flowers in a meadow or a mantis choosing between prey. Studies focusing on attention in insects often do so under the framework of selective attention – the ability to choose one amongst several targets (for reviews, see De Bivort and Van Swinderen, 2016; Nityananda, 2016). Insect selective attention is influenced by different properties of targets, including contrast. However, we do not know how prior cueing influences insect selective attention to targets presented with various properties.

A classic paradigm to study visuospatial attention in humans uses cues to bias spatial attention prior to the presentation of a target (Posner, 1980). In these studies, prior cueing influences the detection and perception of subsequent targets, depending on the position of the cues. Only a few studies have applied similar cueing paradigms in insects (Lancer et al., 2019; Sareen et al., 2011; Wiederman et al., 2017), and they have not investigated how the properties of the cue could affect insect attention. These studies also focused on insect spatial attention within a single plane and have not investigated the influence of depth-related cues. In particular, the interaction between insect attention and stereoscopic disparity remains largely unknown. Disparity is the difference in the retinal position of an object on the two eyes. Since this difference is greater for nearer objects and less for objects that are further away, disparity can provide information about the depth of an object independent of other visual characteristics, including angular size and motion. The ability to use disparity to calculate depth is called stereopsis, which is present in several animals (Nityananda and Read, 2017), including one group of insects, the praying mantids.

Stereopsis has been demonstrated behaviourally in mantises by measuring their probability to strike at prey presented with different disparities (Rossel, 1983). Mantises preferentially strike at prey with disparities corresponding to nearer distances, typically 2.5 to 5 cm away (Maldonado et al., 1967; Nityananda et al., 2016a, 2018; Rossel, 1983). Early studies used prisms to manipulate the disparity of prey (Rossel, 1983). More recently, this was done using an insect 3D cinema, which involves affixing colour filters on each eye of the mantis, enabling the presentation of different stimuli to each eye on a computer monitor (Nityananda et al., 2016b) and thus manipulation of the perceived disparity of an object on the screen (for an example, see the schematics in Fig. 1). This method has allowed researchers to investigate the mechanisms underlying stereopsis and its role in detecting and striking at prey. Recent neurophysiological work (Rosner et al., 2019) has also suggested that disparity differences could guide attention in 3D space.

We ran a series of behavioural experiments to investigate the effects of cueing on mantis selective attention to two subsequent stimuli that differed in contrast. Additionally, we also looked at the effect of cue disparity. We used an insect 3D cinema to present mantises with cues that varied in disparity, duration and lateral location, followed by a choice between a focal high-contrast target and another identical stimulus presented at different contrasts, which we called a ‘distractor’. To investigate selective attention, we tested how the properties of the cue and distractor influenced mantis saccades to the target.

The mantises in our experiment were all adult female Sphodromantis lineola (Burmeister 1838), bred in a dedicated insect housing facility or bought as 4th instar nymphs from Bugz UK. They were raised in a dedicated insect facility and kept in individual plastic containers (dimensions 19.8×12.5×15.5 cm) at a constant temperature of 25°C. When individuals did not participate in experiments, they were fed three times a week with an adult cricket. During experiments, mantises were fed only once a week to increase their motivation to react to stimuli (Bertsch et al., 2019; Pickard et al., 2021).

Each mantis was fitted with 3D anaglyph glasses consisting of one red and one purple tear drop-shaped transparent light filter (Fig. 1A, LEE^® colour filters, 135 Deep Golden Amber and 797 Purple, respectively) fixed in front of the eyes of the mantises using a mixture of wax and rosin on the frons (for more details, see Nityananda et al., 2019a). For each mantis, the side of the red and the purple filters was determined randomly. These glasses ensured that each eye could see only one of the two channels in which we presented stimuli. We were then able to manipulate stereoscopic disparity based on the position of the stimuli in each eye (Fig. 1B). An attachment was also fixed with rosin and wax to the pronota of the mantises to allow us to later attach them to the experimental stand. Mantises were allowed to recover for 1 day after this procedure.

On experimental days, mantises were placed hanging upside down with their feet holding a platform and fixed to the experimental stand using the attachment on their pronotum (Fig. 1A). This position is their preferred hunting position and is typical of mantises in other vision experiments (Rossel, 1983, 1996). Our setup maintained the mantises at a fixed position of 10 cm in front of the centre of a computer screen while allowing them to freely move their head and front limbs. Experimental stimuli were displayed on the monitor (DELL U2413f LED monitor, 1920×1200 pixels resolution, 60 Hz refresh rate, 51.8×32.5 cm). A webcam (Kinobo USB B3 HD Webcam) placed under the screen recorded the responses of the mantises and downsampled them to save them at approximately 8.35 frames s⁻¹.

All stimuli were generated using custom-written code with Psychtoolbox (Brainard, 1997; Kleiner et al., 2007; Pelli, 1997) (version 3.0.15) in MATLAB (The MathWorks Inc., version 2012b). Stimuli were presented in 3D by presenting each eye with stimuli in different colour channels in combination with the anaglyph filters (Fig. 1B–D). In order to balance the input of each channel through the filters, the RGB gains were adjusted to be (0 1 0) in the green channel and (0 0 0.13) in the blue channel (for more details, see Nityananda et al., 2019a). Stimuli in all experiments were presented against a background that consisted of 50% of the maximum output of each colour taking into account these adjustments. It was thus a mixture of blue and green that would be perceived as equally ‘grey’ for each eye.

During experiments, mantises were first presented with a centring stimulus consisting of a spiralling black disc (RGB values: 0 0 0) that swirled in from the periphery to the centre of the screen (Fig. 1B, left). After a pause of 10 s, mantises in conditions with a cue were presented with a cue (Fig. 1B, centre) on either the left or the right side of the screen 2.8 cm from the centre line and at height of 24.3 cm from the bottom of the screen (8.05 cm above the line of sight of the mantises). The cue was a disc (subtended angle at the viewing distance of 10 cm: 11.42 deg) consisting of a group of black and white dots (each of diameter 18.5 pixels, subtending an angle of around 2.86 deg). This disc was stationary but in other respects was identical to that used in previous experiments (Nityananda et al., 2019b) to elicit prey capture responses when moving. Based on pilot experiments, we anticipated that a stationary cue would bias spatial attention without eliciting saccades by itself; that is, it would lead to covert and not overt attention. In overt attention, an individual moves their eyes or head to allocate their attention in space whereas covert attention involves the reallocation of one's attention without eye or head movement (see Carrasco, 2011 for further details). The change in covert spatial attention due to the cue was tested by measuring its effect on the overt saccades to subsequent targets on either side.

The positions of the target and distractor relative to the vertical centreline of the screen were shifted compared with the position of the cue to avoid a possible masking effect of the cue. Both stimuli were presented with a binocular disparity simulating a position on a virtual plane situated 2.5 cm from the mantis (Fig. 1B). They appeared with their bottom edges a quarter of the screen height from the bottom of the screen (approximately 8.1 cm) and moved synchronously upward on parallel trajectories with a speed of 324 pixels s⁻¹ (9.72 cm s⁻¹) until they reached the top of the screen. They thus took approximately 2 s after the cue disappeared to reach the vertical position of the cue. This design and movement were chosen because such stimuli (without binocular disparity) have previously been shown to elicit mantis head saccades (Rossel, 1996). In that study, when presented with two of these targets, mantises would have an equal probability of producing saccades towards each of them. However, if the perception of one of the targets was artificially hampered (by partially occluding it), mantises would disproportionately produce saccades towards the other one. As the aim of our own study was to measure how a cue can influence the perception of subsequent stimuli by the mantises through the capture of their attentional process, this paradigm appeared ideal for our experiment. Since mantises were happy to track such simple stimuli, we decided against the use of more photorealist stimuli resembling prey. These were unnecessary to test our hypotheses, could have introduced unpredictable confounding variables and made our study less comparable to older relevant work.

The presentation sides of the cue and target, and the distractor contrast were randomly assigned on a given trial and counterbalanced across trials. One experimental block consisted of two repeats of every combination of the two cue presentation sides, the two target presentation sides and the five distractor contrast values, making a total of 40 trials. A gap of 60 s was given between each trial to prevent habituation to the stimuli.

In all experiments, the motivation of the mantis was first tested by presenting a 3D black disc simulated to be on a virtual plane situated at a distance of 2.5 cm from the animal (Fig. 1A, right). This disc resembled the centring stimulus (but with crossed disparity) and spiralled in from the periphery to the centre of the screen. The experiment started only after the mantis attempted to strike at the target for two consecutive presentations of this stimulus. The same stimulus was also run at the end of the experimental block and the mantises' responses or lack of response to two consecutive presentations of the stimulus was noted. The mantis was deemed unmotivated for the experimental block if it did not track any of the stimuli in the block and also did not respond to the motivational stimulus at the end. In these cases that block was discarded. When possible, the same block was presented again to the mantis either on the same or a different day. In all, 8 out of 286 experimental blocks were discarded.

We ran two experiments investigating the effect of cueing on selective attention and a control experiment designed to check whether the monocular location of the cue was responsible for the effects measured with crossed and uncrossed disparity cues.

In this experiment, the cue was presented for 4 s (N=14 mantises) before the stimuli appeared. The duration of the cue was similar to that previously used to study insect attention (Sareen et al., 2011). To investigate the role of cue disparity in selective attention, we had three disparity conditions (Fig. 1C, Movie 1) presented in separate experimental blocks: (1) a crossed disparity condition, (2) an uncrossed disparity condition and (3) a zero disparity condition. We also had an ‘uncued’ condition where no cue was presented.

In the crossed disparity condition, the cue disparity simulated a cue at a depth of 2.5 cm when viewed directly from 10 cm away (i.e. at the same depth as the subsequent 3D stimuli – the target and distractor). In the uncrossed disparity condition, the cue had the reverse disparity to that in the crossed disparity condition (i.e. the same parallax but with the stimulus positions in each eye swapped). This therefore simulated an ‘impossible’ stimulus without any clear depth and served as a control condition. In the zero disparity condition, the cue was displayed with images of both eyes perfectly overlapping on the screen (i.e. zero screen parallax), at a depth of 10 cm from the mantis (and a direct diagonal distance from the mantis of 10.38 cm).

These conditions were presented in a random order in a blocked design and the entire set of conditions was presented twice to 13 mantises and once to one mantis that died before the end of the experiment. Except for this last mantis, across all experimental blocks, we had eight trials per mantis for every combination of cue condition and distractor contrast value in which the cue was on the same side as the target. We also had eight trials per mantis for every combination of cue condition and distractor contrast value in which the cue was on the opposite side to the target.

In addition to the spatial position and simulated depth of a cue, its duration could also have an impact on the ability to capture attention. To study the effect of cue duration we therefore repeated experiment 1 with a cue duration of 100 ms (Movie 2). We ran this experiment with 16 mantises (6 of which participated to the experiment 1).

In the crossed and uncrossed disparity conditions of the previous experiments, the monocular position of the cue in each eye differed compared with the zero disparity condition. We therefore ran an experiment to test whether any difference in response between conditions was due to disparity or monocular position. Here, we presented each of 13 mantises with a cue of 4 s duration in only the zero disparity condition but displaced 1.05 cm either toward the left or the right of its original position (Fig. 1D, Movie 3). The resultant positions of the cue corresponded to the monocular positions of the cue in the crossed disparity and uncrossed disparity condition. This manipulation offset the cue either toward the centre of the screen (hereafter ‘centred’) or toward the outside of the screen (hereafter ‘excentred’). All other details of the cue and target presentation were as for the zero disparity condition in experiment 1.

Mantis responses on every trial were recorded blind to the stimuli presented. Cue presentations were indicated on the screen by a small black disc at the bottom of the screen that was visible in the recordings but hidden from the mantis by a card barrier.

Video recordings of every trial were visually examined and the presence/absence and the direction of the mantis first head saccade recorded. A first head saccade was defined as a distinct rapid head movement towards one side of the screen made by the mantis in the horizontal plane, allowing the mantis to place either the target or distractor in her fovea (Movie 1). Slow head movements were not counted and trials where the mantis started with its head already tilted toward a side of the screen were excluded from further analysis. After the video analysis, responses were matched with the location of the target and distractor. Trials in which the mantis's first saccade was performed toward the target were recorded as 1 and trials in which the first saccade was toward the distractor or in which mantises did not make saccades were recorded as 0. We also manually recorded the video frame on which the target and distractor appeared, the one on which the mantis started its first saccade for all trials from experiments 1 and 2. We used these to compute the timing of the first saccade.

If the mantis performed a saccade toward the cue before the target and distractor appeared, the trial was excluded from further analysis. In total, 2.4% of trials (229 out of 9400) were excluded from analysis because of saccades to the cue. Statistical analyses were conducted in R 4.0.4. (r-project.org) by fitting generalised linear mixed effect models using the lme4 package (https://CRAN.R-project.org/package=lme4; Bates et al., 2015). All models included the mantis ID as a random effect. We used target saccade probability as the dependent variable, coding trials with a first saccade toward the target as 1 and trials with a first saccade toward the opposite side or no saccade as 0. These models were fitted using a binomial family and a logit link function.

For the analysis of the results, we combined data from experiments 1 and 2 in a model. The independent variables in the model were the contrast of the distractor taken as a continuous variable, the cue duration, and cueing condition. Since we had two possible locations of the cue (target side cued or non-target side cued) for each of the three cueing disparity conditions and an additional category of uncued trials, there were a total of seven different cueing condition levels included in the last variable. Additionally, we also considered the interaction between the three independent variables in our analysis.

To determine the significance of an effect and select the most appropriate model, we compared the goodness of fit of models of increasing levels of complexity. The goodness of fit comparisons were conducted using the likelihood ratio tests from the anova() function in R. We first compared models with only one of the three main effects (distractor contrast, cue duration and cueing condition) to a null model to analyse whether any of these variables influenced the probability of first saccades towards the target. We next examined whether the effect of distractor contrast was different depending on the cueing condition by comparing a model including the main effects of distractor contrast, cueing condition and their interaction (contrast×cueing condition) with a model including only the distractor contrast). Finally, we analysed whether this last interaction was different depending on the duration of the cue. To do so, we analysed the significance of the three-way interaction by comparing the full model (contrast×cueing condition×duration) to a model including the main effects and interactions between only distractor contrast and cueing condition.

Because conditions when the distractor was not visible appeared to be of particular significance, we fitted additional models on the binomial data of the mantises' saccades towards the target when the distractor contrast was equal to 0. We first compared a model with the cueing condition as an independent variable to the null model. We then tested the effect of the cue duration against the same null model. Finally, we tested the effect of the interaction of the cue duration with the cueing condition by comparing a model with this interaction to the model including only the effect of the cueing condition.

We also investigated whether the cue position (displayed on the side of the target or the non-target side) had an effect on the probability of mantis head saccades toward the target, specifically when the distractor was absent. To do this, we focused on the data when the distractor contrast was 0. We excluded all uncued trials (trials without a cue displayed) from these data since cueing position is not meaningful for the uncued condition. We first tested the effect of the cue position and the cue disparity by comparing two models fitted with each of these variables to a null model. We then tested the interaction of the cue position with the cue disparity by comparing a model including this interaction to the model with the cue position only.

Finally, we also used the uncued condition trials from experiments 1 and 2 to analyse the effect of distractor contrast on the overall mantis saccade probability to either stimulus. We used a model fitted with a binomial family and a logit link function on the data from both experiments. The effect of the contrast was assessed by comparing a model with only distractor contrast as an independent variable to the corresponding null model. To confirm that an eventual effect of the distractor contrast detected by our model was not only due to the particular case when the distractor was not displayed, we also performed the same analysis on the data with the zero distractor contrast condition excluded.

To analyse the results from experiment 3, the model selection process was similar to the previous one. First, we compared three models with each of the main effects of distractor contrast, cue position (target side or non-target side) and the offset type (centred or excentred) to a null model. Then, we compared a model with the main effects of distractor contrast and cue position with a model that also included their interaction. Finally, we compared this last model to a full model including the offset side.

To analyse the timing of the first saccade, we used mixed effect models fitted on the number of video frames taken by the mantises to perform their first saccades after the target and distractor appeared on screen. To do so, we used the glmmTMB package (https://cran.r-project.org/package=glmmTMB) to fit models with a negative binomial family and a log link function, and also included the modelling of the heteroscedasticity by including a dispersion formula based on the interaction between the cue disparity and the cue duration. All models included the mantis identity as random effect. With these models we examined the effect of the distractor contrast, the cue duration, cue position (target side or non-target side), cue condition (crossed, uncrossed, zero disparity or uncued) as well as side of the first saccade (towards the target or toward the distractor) by comparing the fit of model with each of these main effect to a null model. We then built a model with all significant main effects and progressively added interaction effects and compared each new model to the one preceding it.

For every analysis, after our model comparison, we examined the significance of the estimates of the best model using Wald tests through the summary() function in R. In addition, we calculated the 95% confidence intervals of these intercepts to confirm these test results (see Tables S1–S5). We used a significance level of α=0.05 throughout our analyses.

The best model of the target saccades in experiments 1 and 2 showed that there was a significant main effect of the distractor contrast (χ²=697.570, d.f.=1, P<0.001) and a significant interaction effect of distractor contrast and cueing condition on saccade probability (χ²=29.287, d.f.=12, P=0.004) but no significant effect of cue duration (χ²=23.205, d.f.=14, P=0.057).

The probability that the first saccade was to the target decreased significantly as distractor contrast increased (Fig. 2A,B, Table S1; estimate±s.e.=−2.894±0.205, Z=−14.096, P<0.001). This effect can likely be explained by the fact that, across conditions, as the distractor became more discernible from the background, mantises were more likely to saccade to the distractor, and therefore the probability that they performed saccades toward the target decreased (for more details and supporting statistics, see Fig. S1).

The effect could further be explained by a second phenomenon: the overall likelihood that the mantises performed any saccade at all, to either stimulus, also decreased as the target and distractor had more similar contrasts (Fig. 3; effect of the distractor contrast: estimate±s.e.=−1.544±0.159, Z=−9.690, P<0.001). The curve presented in Fig. 3 shows that the decrease in saccades is particularly pronounced between the 0 and 0.25 distractor contrasts. This indicates that the effect of the contrast shown in our model is mostly driven by the presence of the distractor. However, the effect of distractor contrast was significant even if we excluded the condition without a distractor. This suggests that it is the distractor contrast and not merely the presence of distractor that reduces the saccades to the target (estimate±s.e.=−0.513±0.230, Z=−2.233, P=0.026). The presence of the distractor thus strongly reduced the overall number of saccades and the increase in distractor contrast reduced the saccades to the target.

When the cue was presented on the side of the target, it did not significantly affect target saccade probability compared with the uncued condition (Fig. 2A, Table S1; crossed versus uncued condition: estimate±s.e.=−0.268±0.157, Z=−1.710, P=0.087; uncrossed versus uncued condition: estimate±s.e.=−0.189±0.156, Z=−1.213, P=0.225; zero disparity versus uncued condition: estimate±s.e.=−0.189±0.156, Z=−1.211, P=0.226). In these cases, there was also no significant interaction effect between distractor contrast and the cue type on target saccade probability compared with the uncued condition (crossed versus uncued condition: estimate±s.e.=−0.006±0.315, Z=−0.018, P=0.985; uncrossed versus uncued condition: estimate±s.e.=0.359±0.300, Z=1.197, P=0.231; zero disparity versus uncued condition: estimate±s.e.=0.334±0.302, Z=1.106, P=0.269).

However, when the cue was presented on the opposite side to the target, it had different effects for different distractor contrasts. When the distractor contrast was zero, only the target was visible. In this case, compared with the uncued condition, the crossed and uncrossed disparity cues reduced the target saccade probability (Figs 2B, 4D,E, Table S1; crossed versus uncued condition: estimate±s.e.=−0.434±0.157, Z=−2.765, P=0.007; uncrossed versus uncued condition: estimate±s.e.=−0.441±0.156, Z=−2.830, P=0.005). This was not true of the zero disparity condition (Fig. 4F; zero disparity versus uncued condition: estimate±s.e.=−0.195±0.156, Z=−1.251, P=0.211) where the cue was presented 10 cm away from the mantis. The effect sizes for crossed and uncrossed disparity cues were correspondingly larger than for zero disparity cues. This suggests that in this situation, cues in two of the three disparity conditions covertly attracted the mantises' attention compared with a condition without a cue, making them less likely to look towards a target subsequently presented on the other side of the screen. Importantly, the cues themselves did not result in saccades, since we had excluded trials where this happened. This result is therefore not explained by an initial change in head direction before the saccade to a target.

The decrease in target saccade probability with increasing distractor contrast was significantly attenuated when these cues were on the non-target side (Table S1; crossed versus uncued condition, estimate±s.e.=0.887±0.295, Z=3.011, P=0.003; uncrossed versus uncued condition: estimate±s.e.=1.112±0.288, Z=3.857, P<0.001). As the contrast of the distractor increased, the crossed and uncrossed disparity cues thus appeared to progressively have a reduced effect, and the target saccade probability became comparable to that in the uncued condition (Fig. 2B). This significant interaction effect between cue and contrast was, however, not seen for a zero disparity cue (zero disparity versus uncued condition: estimate±s.e.=0.389±0.300, Z=1.300, P=0.194).

Despite the different effect of cues with different disparities compared with the uncued condition, we did not find many differences when comparing the effects of these cues with each other. When presented on the non-target side, the main effect of a zero disparity cue was not significantly different to that of both crossed and uncrossed disparity cues (Table S2; crossed disparity versus zero disparity condition: estimate±s.e.=−0.239±0.182, Z=−1.318, P=0.188; uncrossed disparity versus zero disparity condition: estimate±s.e.=−0.247±0.181, Z=−1.366, P=0.172). The interaction with contrast also did not significantly differ for zero disparity and crossed disparity cues (estimate±s.e.=0.498±0.339, Z=1.467, P=0.142). However, the effect of this interaction with contrast did significantly differ between the zero disparity and uncrossed disparity conditions (estimate±s.e.=0.722±0.334, Z=2.165, P=0.030). Taken together, these results show that disparity only seems to have a limited influence on target saccade probability.

To confirm the results for this previous model, we focused our analysis on the trials when the distractor was not visible (distractor contrast=0). In this specific case, the model selection phase confirmed the effect of the cueing condition (Fig. 4; χ²=13.958, d.f.=6, P=0.030). The examination of the estimates of this particular model including the effect of the cueing condition also confirmed that when the cue was on the side opposite to the target with a crossed disparity (Table S3; estimate±s.e.=−0.575±0.200, Z=−2.869, P=0.004) or an uncrossed disparity (estimate±s.e.=−0.556±0.200, Z=−2.775, P=0.006), it decreased the probability of head saccades towards the target compared with the uncued condition. However, when compared with a condition with a zero disparity cue in the same position, the effect of the crossed (Table S4; estimate±s.e.=−0.318±0.232, Z=−1.375, P=0.169) and uncrossed disparity cues (estimate±s.e.=−0.299±0.231, Z=−1.293, P=0.196) did not significantly differ.

Crossed and uncrossed disparity cues appear to have an effect on mantis saccades to the target but only when the cues were on the non-target side and when the distractor was not visible. We therefore decided to analyse the effect of the cue position (on the side of the target or the opposite side) for trials when the distractor was not shown. For these models, we used the data of the trials with a distractor contrast of 0 and excluded the uncued condition because the cue position was not relevant for this condition. The model selection showed that the cue position had a significant effect (χ²=3.851, d.f.=1, P=0.0497). However, in this specific condition, cue position did not interact with cue disparity (χ²=4.450, d.f.=4, P=0.349). Thus, the selected model indicated that the chances of saccades toward the target were lower when the cue was on the opposite side of the screen than when it was on the side of the target (Table S5; estimate±s.e.=−0.269±0.136, Z=−1.975, P=0.048). Overall, cues seemed to attract attention away from targets when on the opposite side but not boost attention when on the same side. In addition, the disparity of the cue only had a limited effect.

The monocular position of stimuli differs for zero disparity cues and the crossed and uncrossed disparity cues (Fig. 1C). We therefore ran an additional experiment to verify whether any differences between cueing conditions observed with the crossed and uncrossed disparity cues was due to the disparity or the monocular position of the cue. We presented a zero disparity cue offset at either of the monocular positions of the cues with disparity used above. Similarly to the results above, target saccade probability significantly decreased with an increase in distractor contrast (Fig. 5; χ²=52.920, d.f.=1, P<0.001; effect of distractor contrast: estimate±s.e.=−1.273±0.179, Z=−7.126, P<0.001). However, the position of the monocular cue did not significantly decrease target saccade probability (χ²=1.795, d.f.=1, P=0.180). There was also no significant effect of the direction of the cue offset (χ²=0.210, d.f.=1, P=0.647). The offset did not have a significant interaction with distractor contrast (χ²=0.859, d.f.=2, P=0.651). Finally, the three-way interaction between distractor contrast, cue position and offset direction was also not significant (χ²=5.819, d.f.=6, P=0.444). Overall, these results confirm that the monocular position of a zero disparity cue does not have an effect on target saccade probability.

Mantises took less time to perform their first saccade when it was directed at the target than when it was towards the distractor (Fig. 6; effect of saccade direction: χ²=49.976, d.f.=1, P<0.001; estimate±s.e.=−0.285±0.050, Z=−5.720, P<0.001). However, a significant interaction with the distractor contrast showed that while the latency to perform a saccade towards the target increased with the distractor contrast (effect of distractor contrast: χ²=133.510, d.f.=1, P<0.001; estimate±s.e.=0.293±0.027, Z=10.780, P<0.001), this increase was not seen for saccades towards the distractor (interaction: χ²=27.143; d.f.=1, P<0.001; estimate±s.e.=−0.345±0.066, Z=−5.21, P=0.395).

In addition, the presence of the cue also affected the latency to perform the first saccade. Mantises took more time to perform their first saccade in trials with a cue of any disparity condition compared with the uncued condition (effect of cue type: χ²=14.460; d.f.=3, P=0.002; crossed disparity cue versus uncued: estimate±s.e.=0.057±0.024, Z=2.330, P=0.020; uncrossed disparity cue versus uncued: estimate±s.e.=0.066±0.024, Z=2.760, P=0.006; zero disparity cue versus uncued: estimate±s.e.=0.072±0.023, Z=3.080, P=0.002). However, there was no difference between the different cue conditions (crossed versus uncrossed disparity cue: estimate±s.e.=0.010±0.023, Z=0.420, P=0.674; crossed versus zero disparity cue: estimate±s.e.=0.016±0.022, Z=0.700, P=0.485; uncrossed versus zero disparity cue: estimate±s.e.=0.006±0.022, Z=0.270, P=0.784). Moreover, there was no interaction with the distractor contrast and the saccade direction (effect of interaction: χ²=12.674, d.f.=9, P=0.178). Finally, the cue duration or saccade side (on the side of the target or of the distractor) did not have an effect on the timing of mantis first saccades (effect of cue duration: χ²=0.032 ; d.f.=1, P=0.858 ; effect of cue side: χ²=0.126 ; d.f.=1, P=0.723).

Our results showed that prior cueing influences selective attention in mantises. The presence of a cue, without any overt behavioural response, nonetheless influenced subsequent selective attention between two stimuli. Mantis selective attention was influenced mostly by the contrast of the distractor and cue position and was not strongly affected by the disparity of the cue.

As both our stimuli grew more identical, mantises struggled to selectively choose either of the stimuli. This might reflect poor selective attention or divided attention to both stimuli. Mantis saccade rates have previously been shown to be reduced when more than one target or distractor was shown simultaneously in a trial (Yamawaki, 2006), which resembles our results. Different neural units may be responsible for triggering saccades towards different areas of the mantis visual field (Yamawaki, 2006). When several units are activated at the same time, they would compete to initiate a saccade and this competition would decrease the chances of saccade. How might a mantis catch prey given this apparent limitation to their ability to choose between stimuli? The presence of two identical equidistant prey would likely be extremely rare in their natural environment. Natural differences in distance or prey traits could resolve the problem for the mantis. The selective attention mechanism of the mantis would thus appear to depend on the natural variation in the environment to enable prey selection. This points towards the importance of considering the role of ecology when studying the cognitive abilities of animals.

Other work (Rossel, 1996) presenting mantises with identical zero disparity stimuli found that their first saccades were not made clearly towards one or the other target but were more evenly distributed, including towards the centre between the stimuli. It is not clear there whether the total proportion of trials with saccades also declined in the trials with two identical stimuli compared with trials with a single target as we observed in our experiment. Flies have also been shown to choose between two stimuli as long as they are far enough apart. With stimuli that are closer together they choose the midpoint of two stimuli (Reichardt and Poggio, 1973, 1975). On the other hand, in dragonflies, which are predatory insects like mantises, an interneuron ensures that one of two paired targets can be selected (Lancer et al., 2019; Wiederman and O'Carroll, 2012).

The presence of multiple stimuli has also been shown to slow down mantis reaction time in some experimental conditions (see part 3.3.2 in Yamawaki, 2006), which is similar to our results (Fig. 6). This effect of the simultaneous presence of a distractor on the hemifield opposite the target resembles the remote distractor effect observed in humans (Walker et al., 1995, 1997) where distractors slow down the execution of eye saccades to targets. However, contrary to our results, this effect in humans does not seem to be affected by the distractor contrast (Born and Kerzel, 2008). Interestingly, the time to perform saccades towards the distractor is consistently high, which suggests that processing a low-contrast stimulus may require more time in the mantises. Although, this result should be considered with caution as the number of saccades toward the distractor drastically decreases at low contrasts, thus reducing the sample size and increasing the variance of our data (as shown in Fig. 6).

Cue position influenced mantis saccades in our experiment. In primates, cue position also changes the attentional performance based on the contrast (Carrasco et al., 2000; Herrmann et al., 2010; Snowden et al., 2001). However, the presence of two variable contrast targets or of a single target in these primate studies means that our results may not be directly comparable. Importantly, our cues had an effect only when presented on the side opposite to the target, when they reduce saccades to the target (Figs 2 and 4). This supported our expectation that target saccades would be lower when the cue was on the distractor side. However, we did not see enhanced attention (which would be indicated in our experiment by an increase in target saccades) when the cue was on the same side as the target, which is often seen in similar experiments with primates (Carrasco et al., 2000; Herrmann et al., 2010). Previous research on insect attention has found cueing effects that enhanced attention, but these experiments typically used cues that were identical to the targets (Lancer et al., 2022; Sareen et al., 2011). In our experiments, the cue and the target were very different, and this might explain why we do not see a boost in attention to the cue in our experiments.

Cues presented on the opposite side of a target did reduce attention to the target. This arguably resembles the effect of the presence of the distractor. However, the cue vanished before the appearance of the target, whereas in trials with a distractor, the target and distractor were presented simultaneously. The cue thus had an effect that persisted after it disappeared for at least a limited duration (approximately 2 s before the target reached the same vertical position). If the cue is impairing selective attention like the distractor does, this effect appears to last longer than the immediate presence of a stimulus. This was similar to previous work showing that a cue can influence target selection by fruit flies both immediately after cueing or after a delay of 2 s (Sareen et al., 2011).

Most studies of spatial attention in humans and non-human primates have investigated attention using cues and stimuli in a single plane. Fewer studies have investigated how attention is allocated in three-dimensional space. Studies that have researched this have found that attention can be guided by cues to depth, especially stereoscopic cues (Andersen, 1990; Andersen and Kramer, 1993; Atchley et al., 1997; He and Nakayama, 1995; Plewan and Rinkenauer, 2020, 2021; Zou et al., 2022). In our experiments, although the presence of a cue slowed down the reaction time of mantises, cue disparity did not affect latency to perform a saccade. This suggests that the decision to saccade is influenced by the presence of a prior cue but may not be affected by the cue properties. However, our results about saccade latency should be considered with caution given the poor temporal resolution of our videos. It is also possible that disparity or other cue properties had a stronger effect on saccade dynamics and the fine-grained details of mantis head movements that we were unable to measure. Therefore, future studies analysing mantis behaviours in response to various cueing paradigms with a finer temporal resolution would be important to investigate these effects.

Cue disparity also had little effect on the probability of saccades in our experiments. Crossed disparity cues simulated a cue nearer to the mantis whereas zero disparity cues simulated a farther cue. The fact that ‘near’ cues had a larger effect compared with ‘farther’ cues suggest that mantis attention could be influenced more by nearer objects than farther ones. This would be similar to some results found in humans (Chen et al., 2012; Finlayson and Grove, 2015; Plewan and Rinkenauer, 2017). It would also seem to fit in with results showing that mantises prefer to strike at stimuli placed 2.5 cm from them and that their probability of striking is dramatically reduced for stimuli farther than 5.63 cm (Nityananda et al., 2016a). However, cues with uncrossed disparity had a similar effect to crossed disparity cues. This similarity in response to cues of crossed and uncrossed disparities is different from previous results about prey capture behaviour. There, mantises were seen to be more likely to strike at crossed disparity targets and not to uncrossed or zero disparity targets (Nityananda et al., 2016b, 2018). The difference in results suggest two things. First, mantises could attend to targets based on the parallax between monocular positions rather than target disparity. Second, the decision to capture prey is governed by a separate, possibly later module which does differentiate between crossed and uncrossed disparities. Previous work (Nityananda et al., 2019a,b) has also argued that prey detection and prey capture are governed by separate mechanisms and our results add further support for these mechanisms. These authors also showed that both crossed and uncrossed cue disparities can prime subsequent strikes to targets and crossed cue disparities had a stronger effect (Nityananda et al., 2019b).

Our results can be understood in light of the neural mechanisms underlying stereopsis in mantises. Neurons tuned to specific crossed disparity signals have been recently found in the optic lobes of praying mantises (Rosner et al., 2019, 2020). These are binocular neurons that connect to the optic lobes of both eyes and then feed into more central regions of the brain. More relevant to our experiments are neurons that relay information from the central complex to the optic lobe. These neurons have been suggested to serve an attentional function and could be involved in the responses we see to cues. Notably, these centrifugal neurons included neurons that responded to both stimuli simulating nearby targets (TMEcen-neurons) and stimuli with diverging lines of sight (TAcen-neurons) as would occur for our cues with crossed and uncrossed disparity, respectively. The neurons that responded to diverging lines of sight responded more strongly to bright stimuli but given that our cue had both white and black dots, they still could have been stimulated in our study. Our results thus add support to the idea that these neurons could be involved in attentional processing. They further suggest that some of these neurons should potentially govern attention based on the parallax between monocular positions, or the presence of any stimulus, rather than only crossed disparities. Other neurons or a subset of these neurons would presumably then respond only to the crossed disparities and send commands to the motor circuits to enable prey capture.

The involvement of central brain structures in mantis attentional processes is also consistent with the few previous studies on insect attention. Most of these studies have shown that some of the neurons involved are present early on in the nervous system, close to the peripheral sensory neurons (Kim et al., 2015; Lancer et al., 2019; Paulk et al., 2014; Pollack, 1988; Römer and Krusch, 2000; Tang and Juusola, 2010; van Swinderen, 2012; Wiederman and O'Carroll, 2012; Wiederman et al., 2017). However, several different brain areas, including more central areas such as the mushroom bodies (Xi et al., 2008; Zhang et al., 2007) and the central complex (Seelig and Jayaraman, 2015) have also been implicated in insect selective attention.

Our results demonstrate that prior cueing can affect mantis selective attention and that these effects are compatible with the known visual neurophysiology of the praying mantis; they thus further our understanding of the sophistication of attention-like abilities in insects.

We thank Adam Simmons for excellent care and rearing of the mantises.