Visual noise from caustic flicker does not affect the hunting success of cuttlefish

Many animals rely on their visual systems to detect, locate or discriminate information in their environment. Environmental ‘ visual noise ’ , however, may interfere with an animal's ability to detect visual information, affecting decision-making processes. A ubiquitous form of visual noise in aquatic environments is caustic ﬂ icker: moving light patterns caused by the refraction of light through surface waves. While caustics impair the ability of ﬁ shes to detect prey, the impacts of caustics on the ability of non-vertebrates to target prey remains untested. In the present study, we asked whether the hunting success of the common cuttle ﬁ sh, Sepia of ﬁ cinalis , is affected by the presence of caustic ﬂ icker. To do this, we tested whether both the spatial (de ﬁ nition) and temporal (speed) components of caustic ﬂ icker affected the ability of cuttle ﬁ sh to detect and catch a common prey, the brown shrimp, Crangon crangon . Neither the spatial nor temporal components of caustic ﬂ icker affected the detection latency or the capture time of prey. Moreover, cuttle ﬁ sh did not adapt their hunting behaviour, including their approach speed, movement bouts, attack distance or angle, as a function of caustic ﬂ icker. Our results show that visual noise from caustic ﬂ icker does not affect the ability of cuttle ﬁ sh to hunt their prey or their hunting behaviour. We provide multiple explanations, including the role of

For a wide range of behavioural tasks, such as avoiding predators or finding resources (e.g. food, shelter or mates), animals rely on their sensory systems to detect, localize or discriminate information in their environment. However, other environmental stimuli that interfere with the ability of sensory systems to detect information (so-called 'noise'; Corcoran & Moss, 2017) can reduce an animal's likelihood of perceiving information in three ways. First, information may be masked by noisy stimuli, increasing the difficulty of detecting or identifying the information. Second, noise can distract an animal which may limit or delay its detection or response to information. Third, animals may misidentify noise as biologically relevant information, which may elicit an inappropriate and potentially maladaptive behavioural response (Dominoni et al., 2020). Through these mechanisms, noise may affect intraspecific communication, navigation, hunting success or predator avoidance behaviour (Kunc & Schmidt, 2019;Shannon et al., 2016;Williams et al., 2015), ultimately affecting an animal's survival and reproductive success.
While noise is usually considered in the acoustic domain (Merchant et al., 2016;Shannon et al., 2016), noise also occurs in other sensory channels, including chemical (Nehring et al., 2013) and electrical domains (Benda et al., 2013). Moreover, noise is particularly prevalent in the visual domain, where it can take static or dynamic forms (Matchette et al., 2018;Peters, 2013). For example, turbidity or mist can absorb and scatter light, thereby reducing light availability and contrast details (Lythgoe, 1979;Utne-Palm, 2002), whereas light passing through naturally swaying foliage can create dynamically moving patches of light and shadows (Matchette et al., 2018(Matchette et al., , 2019. These sources of visual noise can either come as a hindrance or can be exploited by animals. For example, turbidity can reduce the foraging success of pinfish, Lagodon rhomboides, and rosyside dace, Clinostomus funduloides (Lunt & Smee, 2015;Zamor & Grossman, 2007), whereas three-spined sticklebacks, Gasterosteus aculeatus, associate with areas of increased turbidity to reduce their likelihood of being detected by predators (Engstr€ om-€ Ost et al., 2009;Sohel & Lindstr€ om, 2015). Dynamic visual noise can further mask the movements of animals, thereby reducing their likelihood of detection (Matchette et al., 2019), but may also be misidentified by animals interpreting changes in luminance as being generated by other animals (Dominoni et al., 2020). Some animals, on the other hand, do not appear to be affected by visual noise in their environment. For example, several cuttlefish species (Sepia spp.) use their polarization vision (see below) to improve their object detection in turbid waters (Cartron, Josef, et al., 2013;Cartron, Shashar, et al., 2013). Whether, and to what degree, an animal's ability to detect information is impaired in visually noisy environments, therefore, allows us to understand physiological and behavioural adaptations that have evolved to mitigate the impacts visual noise can have on perception.
One ubiquitous form of visual noise in shallow marine environments is caustic flicker, a moving mesh-like pattern of lowintensity polygonal patches of light enclosed by high-intensity light bands on the substrate (Fig. 1a). This dynamic form of visual noise results from the fluctuation of the focal points of light beams when refracted by surface waves varying in their amplitude, frequency and directional orientation (Lock & Andrews, 1992;McFarland & Loew, 1983). Caustic flicker (or caustics) affects the ability of three-spined sticklebacks and Picasso triggerfish, Rhinecanthus aculeatus, to detect prey (Attwell et al., 2021;Matchette et al., 2020). Moreover, sticklebacks avoid areas of their environment with faster moving caustic patterns where prey detection is more difficult (Attwell et al., 2021). Whereas most fish species have camera-type eyes similar to those of terrestrial vertebrates, the visual systems of marine invertebrates range, among others, from convergently evolved camera-type eyes in cephalopods (Packard, 1972) to compound eyes in crustaceans (Meyer-Rochow, 2001). While caustic flicker affects the ability of fishes to detect prey in their environment, whether caustics similarly interfere with the visual systems of marine invertebrates remains unknown.
One marine invertebrate species that frequently occurs in shallow marine habitats which are prone to caustic flicker is the common cuttlefish, Sepia officinalis. Cuttlefish are visual predators (Messenger, 1968), mainly feeding on teleosts, crustaceans and polychaetes (Alves et al., 2006;Castro & Guerra, 1990). Despite their well-developed camera-type eyes (Packard, 1972), cuttlefish visual systems are vastly different to those of vertebrates (Nixon et al., 2003). In contrast to many shallow water fish species (Marshall et al., 2019), cuttlefish are physiologically and behaviourally colourblind (Marshall & Messenger, 1996;M€ athger et al., 2006; but see Stubbs & Stubbs, 2016). However, due to a parallel alignment of the microvilli in the photoreceptors of their retina (Shashar, 2014), cuttlefish are able to discriminate the polarization of light (Shashar, 2014;Temple et al., 2012)  light for a range of visually guided behaviours, such as navigation (Cartron et al., 2012) or object detection (Cartron, Josef, et al., 2013;Shashar et al., 2000). As caustic flicker shows almost no modulation in polarization, it has been suggested that polarization-sensitive animals possess a visual channel that is likely to be unaffected by caustic noise, thereby mitigating the impacts that caustics may have on visual systems based only on intensity (Venables et al., 2022). In this study, therefore, we asked whether the hunting behaviour of cuttlefish is affected by caustic flicker, predicting that the visual adaptations of cephalopods may mitigate against the impact of this type of visual noise on prey detection. By recording the hunting behaviour of S. officinalis towards a natural prey, the brown shrimp, Crangon crangon, in different caustic conditions, we first assessed whether prey detection latency and capture time are prolonged by caustic noise. Subsequently, we tested whether cuttlefish changed their hunting behaviour when exposed to caustic flicker noise. In particular, we established whether cuttlefish altered their prey approach behaviour in noisy conditions, for example by approaching the prey more rapidly to avoid losing visual contact, or more slowly and with more resting periods to visually reassess the location of the prey. Last, we analysed the attack distance and angle between S. officinalis and its prey C. crangon to establish whether caustic flicker induces cuttlefish to attack their prey from closer distances or certain angles.

Animals
Common cuttlefish (N ¼ 35; average mantle length at beginning of experimental period: 52 ± 9 mm (mean ± SD); at the end of the experimental period: 60 ± 11 mm (mean ± SD)) were reared at the laboratory of the Marine Biological Association (Plymouth, U.K.; 50 21 0 51.2 00 N, 4 08 0 21.1 00 W) from eggs collected from the British south coast in September 2020. Animals were kept in 10 opaque holding tanks (internal dimensions 1560 Â 640 mm, 310 mm deep, water level at 250 mm), each divided into four individual compartments (390 Â 640 mm). All holding tanks were connected to a flow-through system supplied by natural sea water and equipped with aerators and mesh lids. Each compartment contained artificial plants and rocks for enrichment. Cuttlefish were fed three times daily with live river shrimp, Palaemon varians, as well as once weekly with live shore crab, Carcinus maenas. Brown shrimp, which were used as prey in the hunting experiments and are a natural prey of common cuttlefish, were periodically collected from Wonwell Beach (Devon, U.K.; 50 18 0 48.9 00 N, 3 56 0 23.7 00 W) and kept in two opaque holding tanks (internal dimensions 410 Â 340 mm, 250 mm deep, water level at 150 mm) supplied with natural sea water. Brown shrimp were fed with frozen mysid shrimp twice per week. All holding tanks were exposed to ambient light and followed the natural light regime over the period when experiments were conducted (MayeJuly). The water temperature in all holding tanks ranged between 14.0 and 17.5 C throughout the experimental period.

Caustic Flicker Playbacks
To test the effect of caustic flicker on the hunting success and behaviour of cuttlefish in controlled conditions, computergenerated animations of caustic patterns were rendered using Caustics Generator Pro software (Dual Heights, www.dualheights. se/caustics/). We manipulated two properties of the caustic patterns: the speed at which the caustics flickered and the definition of the patterns. Both the speed and the definition of caustic flicker vary under natural conditions (Lock & Andrews, 1992;McFarland & Loew, 1983). Here, speed is defined as the rate at which an animation looped, with faster looping caustics having higher flicker speeds (see Video S1). Definition refers to the diffuseness of the caustic patterns, with caustics ranging from well-defined to more diffuse patterns (Fig. 1a, b, c). We chose to manipulate both the flicker speed and definition because higher flicker frequency results in false motion cues that can mask the movements of animals (Matchette et al., 2018), and different definitions create backgrounds with more, or less, pronounced edges, which could interfere with the edge detection of an animal's visual system (Matchette et al., 2019). Therefore, we defined animations with higher flicker speeds and sharper definition as having increased visual noise. To create 12 playbacks with different levels of caustic noise (three different definition levels each with four different speeds) we used the Caustic Generator Pro software to render three caustic animations each consisting of 200 unique frames (aspect ratio 1920 Â 1920 pixels) and differing only in their definition levels (high, medium or low; resulting from Caustics Generator Pro's depth value 5, 3 or 1 m; Fig. 1a, b, c; see Appendix Table A1 for detailed software settings). We then manipulated the speed at which these animations looped by repeating or skipping frames using a bespoke code in MATLAB (The Mathworks Inc; Natick, MA, U.S.A.). Per definition level, four different animations (played-back at 30 fps) were rendered: fast flicker speed (animation looped every 1.66 s), medium flicker speed (animation looped every 6.66 s), slow flicker speed (animation looped every 26.66 s) and static (animation did not loop). For visualization of these animations see Video S1. We ensured that the average light intensity of these animations was consistent (Appendix Fig. A1). In addition to these caustic animations, we also created a control playback of a plain greyscale image with no caustics (uint8 value ¼ 87; Fig. 1d), sharing the same average light intensity as the caustic animations (Appendix Fig. A1).

Experimental Set-up
The hunting success and behaviour of common cuttlefish was assessed in an arena (internal dimensions 1800 Â 800 mm, 835 mm deep, water level at 600 mm), connected to the same flowthrough natural sea water system as the holding tanks. We partitioned two areas on each short side of the tank using opaque Perspex sheets to accommodate the tank inflow and outflow ( Fig. 1e and f), resulting in an experimental arena of 1550 Â 800 mm with a water depth of 600 mm. Two projectors (both CP-WX3541WN, Hitachi Ltd, Tokyo, Japan) were mounted 2040 mm above one long side of the tank (Fig. 1f) and connected via an HDMI-splitter to a computer (Optiplex 5070, Dell Technologies, Round Rock, TX, U.S.A.). As the caustic animations generated with Caustic Generator Pro could be tiled, the projectors were mounted at a fixed distance of 765 mm apart and positioned so that the caustic animations from each projector were aligned at the bottom of the arena. Two cameras (UI-3240CP, IDS Imaging Development Systems GmbH, Obersulm, Germany) were placed in between both projectors and connected to a laptop (G3, Dell Technologies) to film each side of the arena simultaneously (at 20 fps; 1280 Â 1024 aspect ratio) including roughly a 50 cm wide overlap (Fig. 1f). We attached laminated photographs of a homogeneous, uniform sand sample (Appendix Fig. A2) to the base as well as the bottom 100 mm of all four sides of the arena, mimicking a natural substrate bottom . In both corners of one of the short sides of the arena, we suspended two 650 mm long, vertical, opaque, black tubes with an internal diameter of 77 mm, so that the end of the tube sat 100 mm above the arena floor ( Fig. 1e and f). These tubes were used to introduce the brown shrimp into the arena. Directly under the tubes, no caustics were visible owing to the tubes blocking the projected playbacks. A visual barrier made of black cloth was placed above the openings of the tubes to prevent the cuttlefish detecting the experimenter when the shrimp were introduced into the arena. On the opposite short side of the arena to the tubes, we placed a 100 Â 100 mm wide piece of artificial turf on the arena floor to encourage cuttlefish to rest in this location. This aimed to maintain the furthest possible distance between the cuttlefish and the shrimp when the shrimp entered the arena through the tube. To keep the light intensity conditions consistent and minimize external disturbances, a black cubicle, consisting of a black gazebo and black sheets of felt, surrounded the experimental set-up.

Experimental Protocol
For each trial, a single cuttlefish that had not been fed for at least 16 h was collected from its housing tank and released into the arena close to the patch of artificial turf. One of the 13 playbacks was then projected into the arena. Each playback first consisted of a greyscale image being played for 10 min as an acclimation period. This was followed by a 30 s fading transition into an 11 min caustic pattern or the greyscale control. One minute after the caustic pattern (or control) had appeared at full intensity, a shrimp was introduced into the arena ( Fig. 1g and h) through one of the two tubes, giving the cuttlefish 10 min to catch the shrimp in caustic conditions. Shrimp were usually dropped into the tube furthest away from the location of the cuttlefish. In situations in which the distance between the cuttlefish and either tube was approximately equal, we chose the tube that had been used the least in previously conducted trials to maintain an approximately equal use of tubes throughout the experimental period (227 left and 228 right usages). After 11 min of the caustic patterns or greyscale control, another 5 min of a greyscale image were projected into the arena to allow the cuttlefish, if it had not done so before, to catch the shrimp in non-noisy conditions. This resulted in a total playback length of 26 min 30 s ( Fig. 1g and h). We recorded all behavioural activity inside the arena from the beginning of each playback until either 30 s after the cuttlefish caught the shrimp, or the end of the playback if the cuttlefish did not catch the shrimp. All cuttlefish (N ¼ 35) were tested in a repeating set order to maintain a consistent minimal time interval of at least 3 days per animal; however, we ensured that each cuttlefish was tested at different times of the day throughout the experimental period to avoid any daytime-based effects. Each cuttlefish was tested in each playback (N ¼ 13) once (totalling 455 trials), and received a different, unique order of the 13 playbacks, whereby the order cuttlefish received different playbacks was balanced across trials.

Video Editing
As each trial was filmed with two cameras to record the activity in the whole arena, we stitched both video recordings together into one video showing the whole experimental arena for subsequent analysis of the recorded hunting behaviour. To do so, we filmed a standardized checkerboard in three positions along the overlapping area of the camera's field of views once a day. These recordings were used to provide daily calibrated reference points of the positions of both cameras relative to each other. By temporally aligning both trial videos and using the corresponding reference points and an adapted MATLAB code based on Brown and Lowe (2007), we then stitched both trial recordings together on a frame-by-frame basis.

Analysis of cuttlefish hunting success
For each trial, we manually determined the frame numbers when (1) the shrimp entered the arena, (2) the cuttlefish detected the shrimp and (3) the cuttlefish caught the shrimp. We also determined whether the shrimp was detected and/or caught while underneath one of the tubes or in the open arena. The frame of prey detection was defined by the first indication of the cuttlefish shifting its orientation towards the prey (either full body, head or eye movement, often accompanied by a change in its behavioural state, e.g. change in coloration or fin beating; Messenger, 1968). The frame of prey capture was defined by the moment the cuttlefish tentacles first touched the shrimp's body. All videos were scored by the same observer, but blind scoring was not possible as the prevailing caustic noise levels were visible in each video recording. Video recording the trials, however, allowed the timings of these events to be accurately identified and tracking cuttlefish movements in the videos (see below) quantitatively measured their responses to prey. To establish whether the prey detection and hunting success of the cuttlefish were impaired by caustic flicker, we calculated the prey detection latency as the time between the shrimp entering and being detected for all trials in which the shrimp was detected within the 10 min long caustic/greyscale control exposure (N ¼ 409 trials out of 455), the capture time as the time between the detection and capture of the shrimp for all trials in which the shrimp was caught within the 10 min long caustic/ greyscale control exposure (N ¼ 406 trials out of 455), and whether the shrimp was caught or not within the 10 min long caustic/ greyscale control exposure for all trials in which the shrimp was eventually caught (i.e. the shrimp was caught after the exposure had ended, thereby accounting for motivational effects; N ¼ 424 trials out of 455). See Appendix Table A2 for full details of how trials were categorized.
We manually identified the coordinates of the cuttlefish's and shrimp's head (defined as the central point between the eyes) and tail (cuttlefish: posterior end of the mantle; shrimp: transition between pleuron and telson) for the frames in which the shrimp was detected using a bespoke MATLAB script. This was done for all trials in which the shrimp was detected within the 10 min caustic/ greyscale control display (N ¼ 409 out of 455 trials). We calculated the distance between the cuttlefish and the shrimp at the moment of detection as the shortest distance between the cuttlefish's head and the shrimp's body defined by the shrimp's head -tail vector. For trials where either the cuttlefish or the shrimp was underneath one of the tubes at the moment of detection (N ¼ 268 trials out of 409), the centre of the tube was used as a proxy for the animal's position.

Analysis of cuttlefish approach behaviour
For trials in which the shrimp was both detected and caught in the open arena (i.e. not under one of the tubes; N ¼ 176 of 455 trials; Appendix Table A2) and therefore fully exposed to caustic flicker throughout the hunting sequence, we uploaded these videos to the tracking software Loopy (Loopbio, Vienna, Austria; www. loopbio.com/loopy/). We used a supervised machine learning approach to track the position of the cuttlefish across all frames of each trial. To do this, we manually annotated examples of the cuttlefish 'head' (set as the central point between the eyes) in approximately 1900 frames across 91 videos (N ¼ 7 per caustic playback). We then trained a model to detect the head of the cuttlefish and used this model to predict the x-y coordinates of these objects on all frames of the videos. Because the cameras were not directly above the centre of the arena, we used Loopy's inbuilt 'geometric distortion correction' function to transform these x-y coordinates to real-world dimensions (mm) based on the known dimensions of the experimental arena and imported those into MATLAB. There, we smoothed the cuttlefish trajectories using MATLAB's smooth function (Savitzky-Golay method; span ¼ 15; degree ¼ 3). Using these trajectories, we quantified the approach behaviour of the cuttlefish in the period between prey detection and capture. We first calculated the instantaneous speed of the cuttlefish as the distance travelled by the cuttlefish's head between two consecutive frames. Log-transforming these speeds across all trials revealed a bimodal distribution (Appendix Fig. A3), showing that the cuttlefish were either stationary (speed 20 mm/s) or moving (speed > 20 mm/s) during their approach to the shrimp. Using these definitions, we calculated (1) the median approach speed of the cuttlefish towards the prey when moving (median of all speeds > 20 mm/s), (2) the proportion of time spent stationary and (3) the number of movement bouts (moving phases of the cuttlefish separated by at least 1 s of stationary behaviour).

Analysis of cuttlefish attack behaviour
For all trials in which both the cuttlefish and the shrimp were fully visible and not obscured by the tubes at the moment of prey capture (N ¼ 233 of 455 trials; Appendix Table A2), we manually identified the coordinates of both the cuttlefish's and shrimp's head and tail at the moment of prey capture using a bespoke MATLAB code. Using these coordinates, we calculated the attack distance and angle of the cuttlefish towards the shrimp. While the attack distance was calculated identically as the detection distance above, the attack angle was defined as the dot product between the vector of the cuttlefish's and shrimp's heading, as defined by their head and tail positions. These values were converted to degrees, where values closer to zero degrees indicated the cuttlefish attacked the shrimp from its posterior end ('tail'), whereas values closer to 180 refer to an attack towards the shrimp's anterior end ('head').

Statistical Analysis
All statistics were performed in R v. 4.2.2 (www.R-project.org) and included mixed-effects models using the packages lme4 (Bates et al., 2015) and glmmTMB (Brooks et al., 2017). We checked the assumptions for all linear mixed models (LMMs) and generalized linear mixed models (GLMMs) using the performance package (Lüdecke et al., 2021) and the DHARMa package (Hartig, 2022), respectively. Significant effects of each factor in a model were determined by using the 'drop1' call from the lme4 package (Bates et al., 2015) along with c 2 statistics. The effect sizes and corresponding 95% confidence intervals for each model component, represented by Cohen's D values (with higher D values denoting larger statistical effect size), were calculated using the EMAtools package (Kleiman, 2017) and effectsize package (Ben-Shachar et al., 2020), respectively. All box plots were created using ggplot2 (Wickham, 2016), whereas polar histograms were created in MATLAB.
To investigate whether caustic flicker affected the hunting success of the cuttlefish, we tested whether (1) the detection latency, (2) the capture time and (3) whether the shrimp was caught or not (0 or 1) could be modelled as a function of the spatial (definition) and temporal (speed) component of the caustic playbacks. For both (1) and (2), LMMs were used and both data sets were logtransformed to meet normality assumptions, whereas for (3) a GLMM with a binomial family error structure was used. All three models included the spatial and temporal components of the caustic playbacks as ordinal, four-scaled fixed effects, both ranging from 0 to 3 (see Appendix Fig. A4 for classification). The shrimp were often detected (N ¼ 268 of 409 trials) and caught (N ¼ 149 of 406 trials) underneath one of the tubes. When underneath the tubes, the shrimps were shaded from the caustics, whereas outside the tube, caustics were projected over the shrimp's body. We thus tested whether caustic flicker impacted the detection or capture of the shrimp when in these two conditions. To do this, we included an interaction term between the location of the shrimp at the moment of detection/capture (detected/caught underneath tube: yes/no) and the spatial and temporal components of caustic flicker in models 1 and 2. We report these interactions as well as the results from the final models after nonsignificant interaction terms were removed. To control for an effect of the distance between the cuttlefish and the prey at the moment of detection on both (1) the detection latency and (2) the capture time, we added this distance as a continuous covariate to both models. As cuttlefish individuals were used across multiple trials, we included individual ID as a random effect in all three models. We also tested whether (4) the probability of detecting a shrimp underneath the tube increased over time (testing whether the cuttlefish learned where shrimps would appear in the arena). To do so, we used a GLMM with a binomial family error structure and included the trial number as a fixed effect and the cuttlefish ID as a random effect.
We then investigated the effect of caustic flicker on the cuttlefish approach behaviour towards its prey. To do this, we tested whether the spatial and temporal components of the caustic playbacks affected (5) the median approach speed of the cuttlefish in nonstationary phases, (6) the proportion of time spent stationary while approaching the shrimp and (7) the number of movement bouts when approaching the prey. For (5), we used an LMM with speed square-root transformed to meet normality assumptions. For (6) and (7), we used a GLMM with a beta regression or Poisson error structure, respectively.
Finally, we tested whether caustic flicker affected cuttlefish attack behaviour. To do this, two LMMs were used to test whether (8) the attack distance and (9) the attack angle of the cuttlefish changed as a function of the spatial and temporal components of caustic flicker. For the above-mentioned models 5e9, we included the spatial and temporal components of the caustic playbacks as ordinal, four-scaled fixed effects, both ranging from 0 to 3 (see Appendix Table A3 for classification). Moreover, individual cuttlefish ID was included as a random effect in each model as we reused the same individuals across multiple trials.

Ethical Note
This study adhered to the ASAB/ASB guidelines for use of animals in behavioural research. All procedures were approved by the University of Cambridge Animal Welfare and Ethical Review Body (UBS reference number: NR2020/43). Cuttlefish were kept and handled following current care and welfare recommendations (Cooke et al., 2019;Fiorito et al., 2015;Tonkins et al., 2015). On a daily basis, cuttlefish were examined for signs of stress (based on locomotor, postural and chromatic displays; Andrews et al., 2013;Fiorito et al., 2015), illness (e.g. injuries, changes in skin and eye appearance, unusual activity levels or ventilation rate; Andrews et al., 2013;Fiorito et al., 2015) or inadequate nutrition (protruding eyes, poor body condition, floating; Fiorito et al., 2015). Animals did not show any of these signs during the experiment. Although caustic flicker is a naturally occurring phenomenom which cuttlefish experience throughout their life, we limited the number of trials and length of trials as presented above to reduce handling time and time spent outside their housing tanks.

RESULTS
We first investigated whether caustics affected the time it took for the cuttlefish to detect the shrimp. Shrimps that were underneath the tubes were detected earlier than shrimps in the open arena ( Fig. 2a and b; LMM: c 2 1 ¼ 50.80, P < 0.001). This is expected given the shrimps entered through the tubes into the arena and took time to leave the tubes. Therefore, to test whether caustics affected the detection latency while controlling for the shrimp's location, we looked at the interaction between the caustic patterns and the shrimp's location on the detection latency. There was no interaction between the spatial (LMM: c 2 3 ¼ 0.33, P ¼ 0.95) or temporal (LMM: c 2 3 ¼ 4.62, P ¼ 0.20) components of caustic flicker and the location of the shrimp (underneath a tube versus in the open arena) on the detection latency. Therefore, caustic flicker did not affect the time it took cuttlefish to detect the shrimp either when the shrimp was shaded or when it was exposed to caustics in the arena. After removing these interaction terms from the model, the time taken for the cuttlefish to detect the prey was still unaffected by either the spatial (Fig. 2a; LMM: c 2 3 ¼ 4.54, P ¼ 0.21) or temporal ( Fig. 2b; LMM: c 2 3 ¼ 1.96, P ¼ 0.58) components of the caustic flicker playbacks. The distance between the cuttlefish and shrimp did not affect the detection latency of the cuttlefish to the prey (LMM: c 2 1 ¼ 1.13, P ¼ 0.29). The time it took cuttlefish to capture the shrimp was significantly longer when the cuttlefish was further away from the shrimp at the moment of detection (LMM: c 2 1 ¼ 70.48, P < 0.001) and when the prey was located underneath one of the tubes (as opposed to not underneath the tubes; LMM: c 2 1 ¼ 33.42, P < 0.001). Again, this is partly expected given the cuttlefish had to swim further upon prey detection to capture the shrimp, therefore prolonging the capture time. There was no interaction, however, between the spatial (LMM: c 2 3 ¼ 4.12, P ¼ 0.25) or temporal (LMM: c 2 3 ¼ 0.80, P ¼ 0.85) components of caustic flicker and the location of the shrimp (underneath a tube versus in the open arena) on the capture time, demonstrating that caustic flicker did not affect the time it took cuttlefish to catch the shrimp when either the shrimp was exposed or when it was shaded from the caustics. After removing these interaction terms from the model, we found that the prey capture time of cuttlefish was still unaffected by either the spatial ( Fig. 2c;  Because neither the prey detection latency nor the hunting success of cuttlefish was affected by the presence of caustic flicker noise, we asked whether the cuttlefish changed their hunting behaviour to mitigate the impacts of caustic flicker on their hunting success. First, we tested whether the approach behaviour of cuttlefish towards their prey changed in different levels of caustic flicker. We found no evidence that cuttlefish altered their prey approach behaviour in the presence of caustic flicker. Neither the median approach speed ( Fig. 3a and b; LMM: spatial component: temporal component: c 2 3 ¼ 1.51, P ¼ 0.21) between cuttlefish and shrimp varied as a function of either the spatial or temporal aspect of caustic flicker. Therefore, cuttlefish did not modify their attack behaviour as a function of visual noise.
To assess the confidence in our nonsignificant effects, we calculated the 95% confidence intervals for the effect sizes of all models given above (Colegrave & Ruxton, 2003). Average  T0  T1  T2  T3   T0  T1  T2  T3   T0  T1  T2   nonsignificant effect sizes were small and the confidence intervals for nonsignificant effects never spanned into large effect sizes (Appendix Fig. A5). Therefore, our results suggest that caustics are unlikely to have a strong effect on any aspects of cuttlefish hunting success or behaviour.

DISCUSSION
The hunting success of the common cuttlefish was not impaired in the presence of caustic flicker, a ubiquitous form of dynamic visual noise in shallow aquatic environments. Neither an increase in the definition (spatial component) nor an increase in the speed (temporal component) of caustic flicker affected the prey detection latency or prey capture time. Moreover, there were no changes to the hunting behaviour of the cuttlefish in the different levels of caustics. In particular, the approach speed, the time spent stationary and the number of movement bouts towards the prey, as well as the attack distance and angle, did not change in different levels of caustics. Overall, our results show that cuttlefish hunting success or behaviour was not impacted by visual noise from caustics.
Our findings that caustic noise does not impair the ability of cuttlefish to detect or catch prey contrasts with other observations  on the impacts of caustics on prey detection in fishes. Caustic flicker reduces the likelihood that prey are detected in three-spined sticklebacks and Picasso triggerfish (Attwell et al., 2021;Matchette et al., 2020). Similarly, other types of dynamic visual noise, such as dappled light, have been reported to reduce or delay object detection in both humans and fowl chicks, Gallus gallus domesticus (Matchette et al., 2018(Matchette et al., , 2019. These findings in fishes, fowl chicks and humans suggest that either dynamic visual noise affects the ability to detect prey or that noise reduces the motivation of a predator to attack prey (Attwell et al., 2021). In our study, however, cuttlefish hunting success and motivation were not impacted by visual noise from caustic flicker. We outline several potential reasons for these differences below.
In previous studies investigating the impacts of caustics on visual perception, predators were tasked with detecting virtual prey (Attwell et al., 2021;Matchette et al., 2020). In the present study, however, we investigated the hunting behaviour of cuttlefish towards live prey. With live prey, additional sensory cues may have been available to the cuttlefish that would not have been available if we had used virtual prey. Cuttlefish are able to detect chemical cues of their prey (Boal & Golden, 1999) and possess statocysts that detect low-frequency particle motion (Samson et al., 2014). While such cues may have provided additional sensory information about the presence of prey to the cuttlefish, there is little evidence such cues are used by cuttlefish for prey detection, with hunting behaviour being predominantly visually guided (Brauckhoff et al., 2020;Messenger, 1968). In our study, therefore, it is unlikely that in the presence of visual noise cuttlefish relied more heavily on other sensory cues for detecting or hunting shrimp.
Another explanation for why caustic flicker did not affect cuttlefish hunting success could be that the task of detecting the live prey in caustics was not perceptually challenging enough for the cuttlefish. Indeed, in previous studies that identified a reduction in prey detection with increased caustic noise (Attwell et al., 2021;Matchette et al., 2020), predators were tasked with detecting twodimensional (2D) virtual prey in caustics. In our study, however, the three-dimensional (3D) form of the live prey may have facilitated its detection, particularly because the prey were presented against uniform images of a sandy substrate lacking 3D structure. While C. crangon can be well camouflaged on sandy substrates (Siegenthaler et al., 2018), the shrimp in our study were presumably more conspicuous than in more natural and complex scenes. This may have allowed the cuttlefish to detect the shrimp even in the treatments with the most visual noise. Moreover, research from computer vision has suggested that caustic flicker may even facilitate the detection of 3D objects with stereovision (Swirski et al., 2009(Swirski et al., , 2010. While there was no improvement in prey detection in the presence of caustics in our experiment, whether caustics facilitate or hinder the detection of 3D prey in more complex visual scenes remains to be tested. Future research, therefore, should address whether caustics affect the detection of 2D and 3D objects on backgrounds with and without 3D structure.
We also tested whether cuttlefish hunting success was unaffected by caustic flicker because cuttlefish could have adapted their hunting behaviour to compensate for the impacts that visual noise was having on their perception. Again, however, we found no evidence for alterations to prey approach or capture behaviour as a function of visual noise. Indeed, cuttlefish did not approach their prey more quickly to avoid losing visual contact with it, nor approached it more slowly or with more resting phases between their movement bouts to potentially reassess the prey's location. Similarly, cuttlefish did not alter their attack distance or angle in more noisy conditions, indicating that the visual assessment of the prey's location to calculate the angle and distance to shoot their tentacles was not impeded by caustic flicker noise.
Cuttlefish hunting success may not have been affected by caustics because of adaptations to their visual system which can mitigate against the impacts of a fluctuating light environment. Owing to a parallel alignment of the microvilli in the photoreceptors of their retina (Shashar, 2014), cuttlefish are able to discriminate the polarization of light (Shashar, 2014;Temple et al., 2012). Polarization sensitivity allows animals to detect the orientation of the electrical vector properties of light, adding a second contrast level to the recipient's visual system (Cronin et al., 2014;Horv ath, 2014;Marshall & Cronin, 2011). As the polarization of light does not change with water depth, it is hypothesized that for cuttlefish and other marine animals, polarization vision might be an effective way of detecting visual information in aquatic habitats (Marshall & Cronin, 2011). Indeed, cuttlefish are known to use their polarization vision for navigation, object detection and potentially intraspecific communication (Shashar, 2014). Recently, Venables et al. (2022) reported that caustic flicker shows very little modulation in polarization, and so does not interfere with the underwater polarization scene. While the intensity-based visual channels of animals (such as fishes) are impaired by the intensity fluctuations of caustic flicker noise, polarization-sensitive animals possess a second, potentially even independently processed visual channel (Smithers et al., 2019), that is unlikely to be affected by caustic noise. Indeed, caustic flicker does impair the ability of common cuttlefish and the crab C. maenas (which also possesses polarization vision) to detect looming stimuli that only have an intensity component, but detection of these stimuli is not impacted by caustics when these stimuli are polarized (Venables et al., 2022). These results may explain why neither the prey detection latency nor the prey capture time of the cuttlefish tested in the present study were negatively affected by caustic flicker. Many visual backgrounds in aquatic environments are either low in polarization (e.g. coral reefs, sandy bottoms, etc.) or have a constant, predominantly horizontal, linearly polarized background in the water column (Cronin & Shashar, 2001;How & Marshall, 2014). In our study, therefore, polarized light reflected by the shrimp would reveal a contrast against the weakly polarized background (substrate bottom). Owing to their highly acute polarization vision (Temple et al., 2012), even subtle polarization contrasts between the shrimp and the background are likely to be detectable by the cuttlefish. This sensitivity may allow them to detect their prey without caustic flicker noise impacting this detection.
Whereas some fish species are less likely to detect prey when exposed to caustic flicker (Attwell et al., 2021;Matchette et al., 2020) and actively avoid visually noisy areas (Attwell et al., 2021), the results of both Venables et al.'s (2022) study and the present study demonstrate that caustics do not appear to affect the ability of cuttlefish to detect information in their environment. Along with their camouflage capabilities (Hanlon, 2007), associating with visually noisy environments may benefit cuttlefish even further, as their prey or predators may be less likely to detect them. This may provide an advantage when searching for prey or when avoiding predators. Therefore, future experiments should address a potential preference for habitats with different levels of caustic flicker to establish whether, and to what degree, cuttlefish exploit visually noisy habitats.
Our findings demonstrate that the hunting success of cuttlefish is not impaired by the presence of a type of dynamic visual noise, caustic flicker. Compared to some species of fish that are less responsive to detecting prey in caustic environments, neither the prey detection latency nor the capture time of cuttlefish was significantly prolonged in visually noisy conditions. While no behavioural adaptions in both approaching and attacking the prey could be observed in response to the increased noise levels, we suggest that cuttlefish polarization vision helps mitigate the impact of the intensity-based visual noise imposed by caustic flicker.  Log (Instantaneous speed between two consecutive frames (mm/s)) 2 4 6 8 Figure A3. Histogram of the log-transformed instantaneous speeds between two consecutive frames across all trials. The bimodal distribution shows that the cuttlefish were either stationary (left peak) or moving (right peak).  Figure A4. Classification of the 13 playbacks used in this study along two axes representing the spatial (definition) and temporal (speed) components.   Approach behaviour To calculate median approach speed, time spent stationary and number of movement bouts of cuttlefish for the approach sequence (frame shrimp detected e frame shrimp caught) for prey fully exposed to the experimental stimuli Trials in which either the shrimp or the cuttlefish was beneath the tube at any point during the approach sequence and trials in which the cuttlefish left the bottom and approached the shrimp while high in the water column (as opposed to the bottom of the arena) as the trajectory analysis was based on 2D tracking data on a horizontal plane)

176
Attack behaviour To calculate the distance and angle between the cuttlefish and shrimp at the moment of prey capture for shrimp that were fully exposed to the experimental stimuli Trials in which either the cuttlefish or the shrimp was underneath the tube at the moment of prey capture and trials in which the prey capture occurred in the water column (as opposed to the arena bottom, see above)

233
'Experimental stimuli' refers to the 10 min caustic/greyscale control time period starting with the introduction of the shrimp into the arena but excludes the subsequent 5 min of greyscale (see Fig. 1g and h).