Natural warning signals unexpectedly shape human metamemory ratings but not image recognition success

Natural warning signals unexpectedly shape human metamemory ratings but not image recognition success | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Natural warning signals unexpectedly shape human metamemory ratings but not image recognition success Federico De Filippi, Olivier Penacchio, Akira R. O’Connor, Julie M. Harris This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8175452/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 25 Feb, 2026 Read the published version in Scientific Reports → Version 1 posted 4 You are reading this latest preprint version Abstract Although the bold warning patterns of prey (known as aposematic) have been shown to facilitate predator learning through repeated encounters, it remains unclear to what extent these patterns support memory retention. Here, we tested whether aposematic patterns appear more memorable to human observers, and whether they have an intrinsic advantage in being recognised from memory—even after a single exposure. Observers viewed images of aposematic and non-aposematic butterflies and moths, judged how likely they were to remember each one (metamemory rating), and later completed a test distinguishing previously seen species from novel ones (recognition memory). While aposematic species elicited higher metamemory ratings upon first sight, we found no evidence that they were more likely to be recognised when seen again. Despite this apparent metacognitive failure, for aposematic species the observers tend to remember and forget the same images as one another. This suggests that these images exhibit ‘memorability’, an intrinsic property of an image that allows one to predict how well images can be remembered. These findings raise the possibility that an effective visual warning may hinge less on recognition of a previously seen signal, and more on perceptual processes at play when it is first encountered. Biological sciences/Ecology Earth and environmental sciences/Ecology Biological sciences/Neuroscience Biological sciences/Psychology Social science/Psychology Biological sciences/Zoology aposematism warning signal metamemory memorability receiver psychology vision Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction When we glance at an image, it can leave a lasting trace in memory, making it easy to recognise when encountered again [ 1 , 2 ] . This apparently effortless encoding of visual information is not only a hallmark of human memory but may also have an evolutionary significance in nature. One adaptation thought to exploit these processes is aposematism—the use of conspicuous colours and patterns as honest ‘warning signals’ of prey toxicity or unpalatability [ 3 , 4 ] . How this visual defence strategy influences predators’ foraging decisions is still not fully understood [ 5 ] . The effectiveness of a warning signal is commonly attributed to the ease with which it is remembered by predators [ 6 , 7 ] . Conspicuous patterns combining red, orange, and yellow with black, typically in striped or spotted arrangements, are successful aposematic signals, reliably avoided by predators in both natural and experimental contexts [ 8 – 13 ] . Most studies have focussed on how the conspicuousness of a signal promotes avoidance when predators are repeatedly exposed to it. To the best of our knowledge, there have been no attempts to directly compare how aposematic and non-aposematic species are recognized from memory on a purely visual level. Here, we drew on well-known properties of human visual memory for images [ 14 – 16 ] to test whether images of Lepidoptera (butterfly and moth species) that display warning signals are easier to remember than those that do not. A growing trend in sensory ecology leverages human vision science to understand how animal patterning influences the perceptual mechanisms of the observer (reviewed in [ 17 ] ). There is evidence that the effectiveness of warning signals is rooted in early vision mechanisms, such as edge detectors [ 18 ] and colour opponency processes [ 19 ] , which are broadly shared by humans and birds [ 20 , 21 ] , arguably the primary predators of Lepidoptera [ 13 ] . Researchers have used human observers to test what disrupts or enhances the detectability of prey targets, revealing key principles of camouflage and conspicuousness [ 22 – 29 ] . This comparative approach allows traits that have proved adaptive in nature to be characterised in detail through measuring human visual behaviour. The ability to be remembered is considered crucial for effective animal signalling [ 6 , 7 ] and can be easily measured in humans. However, humans have not yet been used to assess whether warning signals provide an advantage in recognising prey from memory. In this study, we will investigate three aspects of human memory relevant to understanding aposematism, defined below. (1) Metamemory : When we view a stimulus, we can form intuitions about how likely we are to remember it in the future, i.e., we ask people to make a metamemory rating [ 30 , 31 ] . Metamemory ratings for photographs can be accurate when based on specific cues (e.g., whether images ‘tell a story’ [ 32 ] ), but can also be highly inaccurate when observers lack these cues [ 14 ] . Possibly related to this measure, warning-coloured stimuli for predator avoidance experiments are often chosen according to how conspicuous or distinctive they appear to human observers. (2) Recognition memory : Humans can store detailed representations of thousands of images in memory, even after seeing them only once [ 1 , 2 ] . When an image is viewed for the second time, determining whether it has been seen before (i.e., measuring recognition ) can be thought of as being effortless ( [ 1 , 2 ] , though see also [ 33 ] ) This ability makes humans a useful model for testing whether aposematic signals exploit memory mechanisms. (3) Memorability : Recognition is not always successful, the chances that people will correctly recognise an image can vary across images—some images are reliably recognised by most people, but others only by a few. Remarkably, whether one person recognises an image can be predicted from whether others do [ 14 ] , indicating that recognition is not entirely idiosyncratic, but partly image-specific [ 34 ] . This is how memorability is defined: an intrinsic property of an image, whose likelihood of recognition can be predicted [ 15 , 16 , 35 ] . Neuroimaging [ 15 ] and electrophysiology [ 36 ] suggest that memorability has a perceptual basis, related to how the visual system responds to images, and it can also be predicted computationally from image content alone [ 14 , 37 , 38 ] . Memorability thus provides a framework to test whether aposematic species possess visual features that intrinsically influence predator decisions, revealing a previously unexplored dimension of their adaptive function. Previous work from our group constructed a database of hyperspectral images of aposematic and non-aposematic Lepidoptera and used computational neuroscience methods to characterise how coloration patterns may stimulate the brain of predator birds [ 39 ] . The modelling demonstrates that the patterns of aposematic species evoke stronger and less sparse neural responses than those from non-aposematic ones, providing a framework to quantify warning signals based on visual inputs alone. Here, we asked whether these same warning signals might also influence human memory. We selected species from the database predicted to evoke the strongest and weakest responses in the avian visual pathway and measured human memory in response to these images. We tested whether aposematic species would (1) be rated by participants as easier to remember (metamemory rating), (2) be better remembered in a recognition task, and (3) promote consistent memory performance across individuals (memorability). Our aim was to assess whether visual features considered adaptive in ecological contexts truly enhance these aspects of memory. Results Human observers viewed images of aposematic (AP) and non-aposematic (non-AP) Lepidoptera species and provided subjective estimates of how likely it was they would remember each species if encountered again (i.e., metamemory rating). Later, observers completed a recognition memory test, where they viewed a mix of previously seen images (i.e., ‘targets’) and novel ones (i.e., ‘lures’) and had to decide whether each image was ‘old’ or ‘new’ (i.e., recognition judgement). For details on the experimental task and selection of visual stimuli, see Materials and Methods . We compared the effects of AP and non-AP species on observers’ metamemory ratings and recognition. Subsequently, we analysed the consistency of recognition across observers, to estimate the amount of intrinsic memorability delivered by AP and non-AP species. Metamemory rating: Do aposematic species appear more memorable? When each target species was viewed for the first time, observers provided a metamemory rating, a score describing the perceived likelihood of remembering the image when encountered in the future (see Materials and Methods—Metamemory rating block ). Figure 1 A shows examples of species that were rated as most memorable, typical, and forgettable. Figure 1 B shows the average metamemory rating for each image, grouped by whether items were AP or non-AP. To test whether AP species were perceived as more likely to be remembered than non-AP ones, we fitted a linear mixed-effects model predicting metamemory ratings from Lepidoptera class (AP vs. non-AP), including both random intercepts and slopes for participants [[rating ~ class + (class | participant)]]. There was a significant effect of class on metamemory (β = 0.16, SE = 0.02, t = 8.75, p < .001), indicating on average, AP species (mean = 0.61, SE = 0.02) were rated 0.16 units higher than non-AP species (mean = 0.45, SE = 0.02). Random effect estimates showed individual differences in ratings: the standard deviation of the participant intercepts was 0.16, and the standard deviation of the Lepidoptera class slopes was 0.12. This indicates that most observers rated AP species between approximately 0.04 and 0.28 units higher than non-AP species. The proportion of variance in metamemory ratings explained by the fixed effect (marginal R²) was 10.8%. The proportion of variance explained by fixed plus random effects (conditional R²), was 47.5%. These results confirm that, despite individual differences, on average, AP species were perceived as more likely to be remembered than non-AP species upon first sight. Recognition: Are aposematic species easier to remember? We next measured whether images were recognised when viewed again. In an old/new recognition test, a previously seen target is either successfully recognised as familiar (i.e., ‘hit’) or not recognised (i.e., ‘miss’); similarly, a lure can either be successfully recognised as novel (i.e., ‘correct rejection’) or mistakenly judged as familiar (i.e., ‘false alarm’). To evaluate whether AP target species were more likely to be recognised than non-AP ones, we focused on hits. Figure 2 A shows examples of species sorted by the hit rate. Figure 2 B shows the average hit rate for each image, grouped by whether items were AP or non-AP. To test whether AP species were more likely to be recognised from memory than non-AP ones, we fitted a logistic mixed-effects model predicting memory outcome (hit vs. miss) from Lepidoptera class (AP vs. non-AP), including both random intercepts and slopes for participants [[recognition ~ class + (class | participant)]]. There was no significant effect of Lepidoptera class on recognition outcome (β = 0.06, SE = 0.13, z = 0.45, p = .651), meaning that AP species (mean = 0.65, SE = 0.03) were not more likely to be recognised than non-AP species (mean = 0.63, SE = 0.02) (Fig. 2 B). The proportion of variance in recognition outcomes explained by the fixed effect alone was close to zero (0.02%), while the variance explained by the full model including random effects was 10.84%. These results indicate that, contrary to our expectations, and to prevailing suggestions in the literature regarding the effect of warning signals on predators in natural systems, AP species did not confer an overall advantage in human memory using a recognition task, eliciting, on average, the same hit rate as non-AP species (Fig. 2 B). Comparing recognition and metamemory ratings: Is memory for Lepidoptera intuitive? Can we predict whether species would be remembered or forgotten in the future simply based on how ‘memorable’ they appear to us at first exposure? The answer is hinted at by the evident discrepancy between the observed metamemory and memory results (see Figs. 1 and 2 ). Here we provide a more detailed analysis. We calculated the difference between metamemory ratings and recognition hit rates for each image. We then plotted the species with the smallest (‘expected’: strongest matches between the two measures) and largest (‘unexpected’: weakest matches) differences, shown in Fig. 3 A. Among the ‘expected’ cases, AP species with visually distinctive markings and colours were judged memorable and later recognised (‘memorable–expected’), whereas non-AP species that seemingly lack warning colouration were judged most forgettable and later forgotten (‘forgettable–expected’). By contrast, the most ‘unexpected’ species showed less obvious visual distinctions between recognised and forgotten images. To test observers’ metamemory accuracy, we calculated the correlation between the metamemory scores of each image and recognition hit rates (Fig. 3 B). Overall, metamemory and recognition across both classes were uncorrelated (Spearman’s r = 0.01, p = .930). There was a non-significant correlation for AP species (r = 0.21, p = .280), and there was a small but significant negative correlation for non-AP species (r = -0.39, p = .037). See Fig. 3 B. Memorability: Is memory for Lepidoptera predictable across observers? Having found clear differences in metamemory between AP and non-AP species, but not in memory (as measured by recognition), we next asked whether warning signals might influence recognition in ways not captured by overall performance measures. To assess this point, it is necessary to quantify individual differences using a ‘consistency analysis’ [ 14 ] , a standard method in memorability research for evaluating if memory is predictable across observers. The consistency analysis involves repeatedly splitting the full subject pool into two random halves (which we label group j and group k), calculating hit rates per image separately within each group, and assessing consistency in hit rates across groups using two complementary metrics. The first is split-half reliability, a metric which indicates how stable hit rates are across different observers, calculated as the Spearman-Brown corrected correlation between hit rates from each group j and k. We calculated an overall split-half reliability of r = 0.80 (95% CI [0.80, 0.81]) across all Lepidoptera species. Next, to assess differences in split-half reliability between Lepidoptera classes, we calculated it separately for AP and non-AP species. The average split-half reliability for AP species was ρ = 0.90 (95% CI [0.90, 0.90]) and for non-AP species it was ρ = 0.56 (95% CI [0.56, 0.55]). This is shown in Fig. 4 A. Note that split-half reliability quantifies consistency in rank-ordering of hit rates between groups but is independent of overall recognition accuracy. Correlations can be high whether participants tend to remember or forget the same images [ 40 ] . Conversely, low split-half reliability suggests that participants disagreed about which images were remembered or forgotten. The difference in split-half reliability between AP and non-AP species was statistically significant (Δρ = 0.35, permutation test, p < .001, see Materials & Methods—Statistical testing of consistency measures ). The second metric, shown in Fig. 4 B, uses cumulative hit rates across the different subject splits [ 14 ] to estimate how well performance from group j predicts that of group k, providing a more detailed breakdown of consistency as a function of recognition accuracy. If particular images are especially likely to be remembered (or forgotten) by one group, will a different group show the same pattern? To answer this question, image-wise hit rates were sorted in descending order (i.e., from most memorable to forgettable) according to group j and compared against those from group k, or against a baseline shuffled group, created by randomly permuting the association between images and their hit rates in group k, while keeping the image ordering for group j fixed. The curves in Fig. 4 B were constructed by calculating the cumulative hit rate (mean hit rate of top-n images) for each group, while moving across this fixed image ordering (i.e., from the top-2 to the top-28 most remembered images), for each of the 1000 random subject splits shown in Fig. 4 A. High consistency yields a close match between the curves for group j and group k, and a separation from the shuffled group curve, indicating that different individuals remembered and forgot the same images in a systematic, non-random way. For example, Fig. 4 B shows that the top three most-remembered AP images were recognised by approximately 95% of observers in one group and by 85% in another, whereas the top three non-AP images were recognised approximately by 80% and 65%, respectively. To assess the predictability of hit rates across observers, we calculated the difference in cumulative hit rates between groups at each image rank (see Materials & Methods—Statistical testing of consistency measures ). For AP species, we found significant differences between group k and the shuffled group at all image ranks except the last rank (Bonferroni-corrected p < .001 for all ranks, p = 1.000 for rank 28). For non-AP species, group k significantly differed from the shuffled group at all ranks except the first and last rank ( p = 1.000) and marginally at rank 2 ( p = .057; all other ranks p < .001). Between classes, AP and non-AP species differed significantly in how much the cumulative hit rate of group k exceeded its shuffled baseline ( p < .001 at all ranks except rank 26, p = 1.000). This is shown in Fig. 4 B, where the group k and j curves for AP species remained closely aligned across most ranks, whereas non-AP species showed weaker alignment, especially for the most memorable images. Overall, the consistency analysis confirmed that hit rates were significantly predictable for both classes, but AP species strongly increased the inter-observer consistency of hit rates, across the entire spectrum of memory accuracy, i.e., both for remembered and forgotten images. Discussion This study tested whether the warning signals of aposematic (AP) Lepidoptera species enhance visual memory in human observers. Our measurement of metamemory rating showed that AP patterns appear to participants as though they will be more memorable than non-AP patterns. Surprisingly, this was not backed up by actual memory performance: recognition was not better for AP than for non-AP patterns. Yet, our analysis showed that people do tend to remember the same images as one another, and their recognition performance is more similar to those of others when the species carry AP patterns. In the following, we discuss how these results fit within the current literature and offer a novel contribution to understanding the cognitive processes underlying visual warning signals. For this study we chose to compare various aspects of memory performance for 2 sets of patterns, those known to be from AP species of Lepidoptera, and those from non-AP. In previous research, we used computational modelling of both the colour and luminance channels involved in visual processing, to show that the AP species used in this study evoke stronger activity in visual brain areas than the non-AP species [ 39 ] . We wanted to test whether warning signal (AP) patterns are also processed differently by memory systems than patterns likely to offer camouflage (non-AP). We first measured metamemory by asking participants to rate how well they thought they would remember each pattern. The AP patterns were the ones that our participants chose as most likely to be remembered. This result did not surprise us. It is reminiscent of the long-standing idea, dating back to Darwin and Poulton [ 3 ] and still reiterated today [ 41 ] , that warning signals appear particularly ‘striking’ to the human eye. However, as discussed in more detail below, our metamemory impression proved inaccurate: AP patterns did not lead to better performance in the recognition task. We do not know why visually distinctive patterns drive us to think they will be easily remembered. Our results do concur with some related human memory literature. For example, Isola et al. [ 14 ] studied memory for photographs of complex scenes. Participants were unable to predict which photograph would be remembered later in a repeat-detection task, showing no significant correlation between the metamemory measure and a repeat-detection memory measure. Thus, akin to our study, participants intuition for what will be well remembered was incorrect. Finally, to bring us back to the idea that warning signals are ‘striking’ [ 3 ] , Isola et al also found that measures of interestingness and image aesthetics correlated with metamemory. We speculate that whatever strikes us as potentially ‘memorable’ about some patterns or photographs is at least partly driven by systems closely linked to vision, rather than solely by our memory systems. Ours is the first study (to our knowledge) to directly measure in humans whether natural AP patterns are better remembered than non-AP patterns. Despite our two classes of image delivering different visual activity in a modelled visual system [ 39 ] , and our metamemory ratings also differing for the two pattern classes, in a recognition task, we found no evidence that AP species were recognised more accurately than non-AP species. Thus, the metamemory differences we observed did not clearly extend to subsequent recognition. Such a pattern of data is the signature of a form of memory known as memorability, a memory system thought to be linked closely to high level visual areas [ 15 , 35 ] . We will discuss this system in more detail below. Note, however, that there is another emerging literature on metamemory for images, exploring how cues about the meaning of images can deliver highly accurate metamemory. For example, Undorf and Broder [ 32 ] , showed accurate metamemory measures linked to specific attributes of scenes (e.g., peacefulness, inclusion of people, semantic distinctiveness). In our study, images were presented in an intentionally abstracted context: museum specimens on black backgrounds, so that cues to meaning were minimised. Hence, we consider this contrasting literature to be beyond the scope of our investigation. Our findings do suggest that warning signals are doing something more subtle than triggering the memory systems involved in recognition. Some earlier authors [ 5 ] have emphasised a similar complexity, in understanding predator learning and decision making when faced with aposematic prey. In a comprehensive review, these authors highlight that avoidance learning (where a predator learns over time that a prey item is toxic or is otherwise not valuable to eat) is governed by a complex balance between gathering information about the prey (e.g., via conspicuous warning signals) and the cost of eating the prey (e.g., a build-up of toxicity level). If gathering information about a prey which has a strong AP pattern is easier, then learning will likely be faster, but this does not require the patterns to be specifically easier to remember . If the ‘striking’ perception of humans when faced with AP patterns transfers to other predators, perception-led cognition, linked to the memorability of specific images, could drive learning, alongside or independent of traditional memory systems. Overall, our results suggest that a strong AP pattern is best characterised by how it appears at first exposure, rather than how easily it is recognised from memory later, shedding some light on what makes warning signals effective. In recent years, the concept of memorability, in particular applied to images, has been developed by the work of Bainbridge and others (see recent reviews [ 16 , 35 ] ). The key evidence used to define the notion of “intrinsic” memorability is that, when faced with many images, some are easy to remember, and some difficult, yet crucially, the same images tend to be remembered and forgotten by different individuals. We analysed our data with this concept in mind. When comparing split-half reliability between AP and non-AP patterns, we found that it was significant for both pattern classes, but higher for AP patterns: people are more likely to remember or forget the same patterns from this class, compared to non-AP. Furthermore, following [ 14 ] , we showed that there was high inter-observer consistency of both remembering and forgetting patterns, across the full range of performance (from very well remembered to very poorly remembered). This is reflected in our results by a near-perfect alignment in recognition performance across individuals, both for remembered and forgotten AP species (see Fig. 4 B). This alignment was less pronounced for non-AP species. Consistency in memory has been reported across a growing range of human-centred images designed for mass audiences, such as complex photographs (r = 0.75 [ 14 ] ), data visualisations (including infographics and graphs, r = 0.83 [ 42 ] ), paintings (r = 0.53 [ 43 ] ), and dance moves (r = 0.51 [ 44 ] ). However, comparing measures of consistency should be done with caution. Our image sample size was necessarily small compared to most other studies (see Methods), which could potentially result in unreliability of correlations. With that caveat in mind, we did find stronger correlations here (r = 0.80 for all species combined) than in most other studies. A key point for us is that consistency was considerably higher when patterns contained warning signals (r = 0.90 for AP, r = 0.56 for non-AP). Unlike the other studies, our image classes contained ecologically adaptive traits, thus we provide a first example of memory consistency being modulated by such traits, as well as being amongst the highest measured in any study (to our knowledge). Such a pattern suggests that animal patterns might engage the perceptual processes responsible for memorability, known to operate independently of traditional memory systems [ 15 ] . In an attempt to follow up this idea further, we applied a deep-learning model to our images, ResMem [ 38 ] . ResMem has been shown to successfully predict memory for complex scenes [ 43 , 45 ] , but is less successful for single objects on artificial backgrounds (e.g., robot faces [ 46 ] ). For our Lepidoptera dataset on black backgrounds, the model produced trends sometimes opposite to human memory data (the Supplementary Figure S6 shows these results). This suggests that what drives memory for our animal images is not captured by the ResMem deep learning model based on complex scenes. Furthermore, note that our images do not fully fit the standard memorability framework, at least in terms of the classes of images that have been studied so far. Recall that our two image classes were chosen not only on the basis that the species depicted were known to be aposematic, or not, but also that each member of the class provided a ‘good’ example of that class, as determined by our neural modelling of the population response of early visual processing ( [ 39 ] , see Methods). Previous work has shown that basic pixel statistics (hue, saturation, intensity [ 14 ] ) are not predictive of memorability, and that memorable vs. forgettable images matched on colour and spatial frequency do not evoke different responses in early visual areas [ 15 ] . These findings have led to the notion that low-level vision may not significantly influence memorability. However, our work differs from these approaches because it is based on low-level vision modelling performed at a neural population level [ 39 ] , rather than on simple measurements of image statistics. The differences we found while using this approach suggest that future memorability studies may benefit from quantifying low-level image statistics in a more biologically inspired fashion. To sum up, our data offer the first evidence that evolved patterns could be processed by memorability mechanisms, although our image set and data differ from current memorability studies. We finish this Discussion with some thoughts and speculations about how our work fits into the issue of what brain regions and pathways are responsible for the memory effects we have found using AP and non-AP patterns. The standard ‘receiver psychology’ view that warning signals are effective for animal predators partly because they are well remembered [ 6 ] was put forward without the knowledge we have discussed above, of the subtle ways in which visual and memory brain pathways are distinct. We highlight some of these subtleties below. Bainbridge [ 35 ] and her group recently found evidence for specific memorability processing areas of the brain. In an influential human neuroimaging paper [ 15 ] , they identified a distinct neural processing stream for memorability. They used fMRI to compare processing for images (faces and natural scenes) known to be memorable or forgettable. Regions more sensitive to highly memorable stimuli included a variety of high-level visual areas, including the fusiform face area (FFA), the lateral occipital complex (LOC) and the parahippocampal place area (PPA). Yet, there was no activity differences between the image classes in early visual cortex. Sensitivity was also found in a range of memory-related areas, but the effects of memorability were distinct from those found for subsequent memory. Consistent with these findings, our behavioural data reveal a clear dissociation between visual ratings and subsequent memory, but also high memory consistency across observers. This suggests that our stimuli may be differentially processed within a distinct memorability network, offering insight into how and where warning signals might be represented (at least in the human brain). In conclusion, this study suggests that the visual processing of certain animal patterns may influence the initial viewing of an image in ways that make it appear more ‘memorable’ to the human eye. Given the signature behavioural pattern of intrinsic memorability that our images elicited, they are likely to be processed by vision-specific memorability pathways. Our work has begun to unpack the idea that AP patterns could be ‘easily remembered’ to reveal which brain mechanisms are involved in responding to animal warning signals. Methods Materials andmethods Participants 57 participants recruited via the Prolific (www.prolific.com [ 47 ] ) participant pool took part in the study. 7 participants who failed one or more vigilance checks (see Materials & Methods—Vigilance check ) were excluded from the data pool, leaving us with N = 50 (25 male, 25 female, mean age = 28.98, SD = 8.30). The Prolific reimbursement rate for taking part in the study was set to £9 per hour (median completion time = 27 minutes). Participant selection criteria were normal or corrected-to-normal vision and no colour blindness. Informed consent was obtained from participants before the study. Participants were informed that they would be asked to look at pictures of butterflies and moths and provide subjective judgements, but information about which species were AP was not shared. Prior to being made available on Prolific, the study was approved by the local research ethics panel (UTREC, School of Psychology & Neuroscience, University of St Andrews). Selection of stimuli When selecting Lepidoptera species to serve as stimuli in the memory experiment, our specific aim was to compile two sets of images (AP and non-AP species) predicted to be represented differently within the visual system of avian predators [ 39 ] . All images implemented in the study were selected from the publicly available St Andrews Hyperspectral Lepidoptera database ( https://arts.st-andrews.ac.uk/lepidoptera/documentation.html ). For details on image acquisition, see Penacchio et al. [ 39 ] . For details on how these images were displayed within the standard RGB colour space and presented in PsychoPy [ 48 ] , see Supplementary Information . The full database contains images of 125 butterfly and moth species from 12 Lepidoptera families: 96 aposematic (AP) and 29 non-aposematic (non-AP), sampled from British museum collections. For each species, the database provides statistics of in silico neural activity (see Supplementary Figure S1 ) that characterise how it is represented in a model avian visual system. These statistics, which have been shown to effectively discriminate AP from non-AP species [ 39 ] , were used here to select 29 AP species predicted to produce the strongest visual responses, and 29 non-AP species predicted to produce the weakest ones. These 58 species served as targets for memorisation in the experiment (shown in Supplementary Figure S2) , while the remaining species were used as lures ( Supplementary Figure S3 ). Given the small number of non-AP species in the original database, we used only AP species as lures and always presented non-AP species as targets. Note that an a-priori selection for targets and lures is unlike standard memory paradigms, where targets and lures are typically randomised across participants. Instead, our approach ensured that the two classes of targets were always maximally differentiated in their predicted neural representations, enabling a more direct test of whether these ecologically meaningful differences in visual encoding translate into differences in human memory. Procedure Figure 5. Examples of one trial from the metamemory and recognition blocks. ( A ) In the metamemory block, participants viewed all target images and rated each one based on how likely they thought they would remember it later. ( B ) In the recognition block, participants were shown a mix of previously seen images (‘targets’), novel ones (‘lures’), as well as task-irrelevant vigilance-check images, and had to decide whether each stimulus was ‘old’ (seen before) or ‘new’ (not seen before). Metamemory rating block The experiment started with a ‘metamemory rating’ study block, followed by a recognition test block (Fig. 5). In the metamemory block (Fig. 5A), participants viewed all aposematic and non-aposematic targets and provided metamemory ratings. Participants were instructed to look at each image carefully and provide their subjective impression of how memorable it looked. The metamemory rating was explained to participants in the instructions as follows: ‘ A memorable image is one that you feel you would be likely to pick out as having seen before, if you saw it again soon, so you would give it a high score’. At the start of each study trial, a central fixation marker appeared for 1 second, followed by one image that remained on screen for 2 seconds. Subsequently, the image disappeared, and the question ‘ How memorable is this image? ’ appeared on screen, along with a 10-point clickable rating scale. Participants were instructed to select a score that best described their subjective impression of how memorable the stimulus looked, from 1 (‘very forgettable’) to 10 (‘very memorable’). Each target was studied and rated only once, and the order of presented targets was randomised for each subject. Note that previous work that has compared memory and metamemory for photographs [ 14 ] used binary memorable/forgettable ratings for scenes from many different classes (e.g., landscapes, urban, persons). Given that our targets were all of one image-type (i.e., different butterfly/moth species), we chose to implement a broader rating scale to measure finer-grained, within-class differences. At the end of the study block, a break was offered before the test block. Recognition block In the ‘recognition’ test block (Fig. 5B), participants were instructed to look at images and state whether they had previously seen each one. Each test trial started with a 1-second central fixation marker, followed by one test image that remained on screen until the end of the trial (Fig. 4 ). Unbeknownst to participants, the test image could be either a target (i.e., previously studied image), a lure (i.e., never seen before image) or a vigilance check (i.e., a task-irrelevant picture of a flower). After 1 second, the question ‘ Have you seen this image before? ’ appeared on screen. Participants pressed the ‘Y’/‘N’ keys to indicate their decision. No feedback was provided. After response, the statement ‘Rate your confidence’ appeared on screen, along with a 3-point (low, medium, high) clickable rating scale. Given our image set constraints, the rating could not be analysed using signal detection methods, hence it is not reported here. The old/new response and confidence rating were self-paced. In contrast to the metamemory block, where a constant presentation time was chosen to standardize encoding time, the test image was kept on screen indefinitely to allow participants to interrogate their memory with no time constraints. At the end of the recognition block, participants were fully debriefed. Vigilance checks The study and test blocks started with practice blocks that contained 5 arbitrary pictures from a database of flowers [ 49 ] , unrelated to the experiment. One flower picture that was shown both in the study and test practice blocks was also scheduled to randomly repeat 5 times during the real test block as a vigilance check, among Lepidoptera pictures. If participants classified the flower as ‘new’ one or more times, it was taken as evidence that they did not sufficiently engage with the task. Therefore, their data was excluded from the analyses (see Participants ). This method is similar to the ‘vigilance repeats’ [ 14 ] used in continuous recognition memory tasks to screen out inattentive participants. Statistical analysis For statistical analysis, the R environment was used [R Core Team, 2021]. After excluding participants who failed any vigilance checks, a total of 2,850 analysable trials were available for each of the two blocks (metamemory rating and recognition). One AP image was excluded from the analyses because of a technical error in programming the experiment. For each target image (i) we calculated recognition performance aggregated across observers. Each observation was coded as 1 for a hit or 0 for a miss, resulting in N⁽ⁱ⁾ total observations for image i, with H⁽ⁱ⁾ hits. Hit rates were then calculated as the proportion of hits per image (HR⁽ⁱ⁾ = H⁽ⁱ⁾ / N⁽ⁱ⁾), providing a measure of recognition success across the whole subject pool [ 14 ] . Note that early memorability work [ 14 ] used the hit rate per image as a measure of recognition performance (as we do here). More recent work, however, recommends subtracting an image’s false alarm rate from the hit rate to obtain a more accurate performance estimate [ 16 ] . In our image set, lure images were only available for AP species, meaning that false alarm rates could not be calculated for both image classes. For this reason, we used hit rates in all analyses (but see Supplementary Figure S3 for data on lure images). To analyse metamemory ratings and recognition responses, mixed models were fitted to the raw trial-by-trial data using the functions lmer and glmer (respectively) from the lme4 package [ 50 ] . Mixed models allow the inclusion of individual intercepts and slopes for each subject. They are useful to account for the contribution of random variability in data due to individual differences in baseline behaviour (intercepts) and how subjects respond to experimental manipulations (slopes). This is beneficial for online studies such as ours, as viewing conditions and user focus levels cannot be strictly controlled. For the selection of mixed models we used the Akaike information criterion (AIC). ANOVA comparisons between models were conducted and those with the lowest AIC were reported in Results. Full model comparisons are included in Supplementary Tables S5 and S6 . To verify model assumptions, the check_model() function from the performance R package [ 51 ] was used. To estimate marginal and conditional R-squared values from mixed model fits, we used the r.squaredGLMM() function from the MuMIn R package [ 52 ] . Statistical significance was determined using a two-tailed significance threshold of α = .05. In all correlation-based analyses (metamemory, memory, and consistency analysis), Spearman’s rank-order correlation (non-parametric) was used to allow for direct comparisons to results of previous image memorability research, where non-parametric correlations are most frequently reported. Statistical testing of consistency measures For the consistency analysis, as recommended by current guidelines on memorability estimation, correlations between random splits of subjects were corrected using the Spearman-Brown correction for split-half reliability [ 53 , 54 ] . This correction adjusts the correlation to estimate the reliability of the full dataset, compensating for the reduced data size in each split (in our case, N/2 = 25). For the split-half reliability analysis (Fig. 4 A), we assessed statistical significance using a permutation test. We calculated the difference (observed Δρ) between the observed split-half reliabilities for AP and non-AP images. To test whether differences could arise by chance, we randomly permuted the AP vs. non-AP class labels 1000 times, recalculating the reliability difference between AP and non-AP (shuffled Δρ). A p-value was calculated by measuring the rate at which the shuffled Δρ exceeded the observed Δρ. For the cumulative hit rate analysis (Fig. 4 B), we compared hit rates across groups both within classes (for AP and non-AP separately) and between classes (comparing AP and non-AP). Within classes, we calculated two differences: (i) the difference between cumulative hit rates of group j and group k (independent subject splits), and (ii) the difference between group k and its shuffled baseline (chance). Between classes, we compared the extent to which the cumulative hit rate of group k exceeded the shuffled baseline for AP versus non-AP species. Because there was one additional non-AP image, the last rank (n = 29) of the non-AP condition was excluded from between-class comparisons. To assess statistical significance, we used bootstrapped confidence intervals and permutation tests. For each image rank number, the difference between cumulative hit rates across groups was computed, and 95% confidence intervals were obtained from 1,000 bootstrap resamples of the differences. This observed mean difference was then compared to a null distribution. To calculate p-values, the sign of each difference was randomly flipped (i.e., sign-flip permutation test), generating a null distribution of differences. The p-value was defined as the proportion of permuted values with equal or greater magnitude than the observed difference. Because multiple tests were conducted (one for each rank), the Bonferroni correction was used to adjust p-values, controlling for the familywise error rate. Declarations Acknowledgements We thank Jasna Martinovic, Justin Ales, Michael Oram, and Wilma Bainbridge for helpful discussions. Author contributions F.D.F. conceived the study, collected and curated the data, conducted the analyses, and prepared all figures and the original manuscript draft. O.P. provided software and, together with A.R.O., contributed to the study’s conceptualisation, methodology, validation, and supervision. J.M.H. contributed to conceptualisation, methodology, funding acquisition, and supervision, and contributed to both the original and revised manuscript text.All authors read and approved the final manuscript. Data availability The data are available online from the Open Science Framework: https://osf.io/syf2w/ Additional information Rights retention In order to meet institutional and research funder open access requirements, any accepted manuscript arising shall be open access under a Creative Commons Attribution (CC BY) reuse licence with zero embargo. Competing Interests The authors declare no competing interests. Funding This work was funded by the UK Research and Innovation Biotechnology and Biological Sciences Research Council (UKRI-BBSRC) through the East of Scotland Bioscience Doctoral Training Partnership (EASTBIO DTP) as part of F.D.F’s PhD studentship. References Brady, T. F., Konkle, T., Alvarez, G. A. & Oliva, A. Visual long-term memory has a massive storage capacity for object details. Proc. Natl. Acad. Sci. 105, 14325–14329 (2008). Standing, L. Learning 10000 pictures. Q. J. Exp. Psychol. 25 , 207–222 (1973). Poulton, E. B. The Colours of Animals, Their Meaning and Use, Especially Considered in the Case of Insects (D. Appleton and Company, 1890). Stevens, M. & Ruxton, G. D. Linking the evolution and form of warning coloration in nature. Proc. R. Soc. B Biol. Sci. 279, 417–426 (2011). Skelhorn, J., Halpin, C. G. & Rowe, C. Learning about aposematic prey. Behav. Ecol. 27 , 955–964 (2016). Guilford, T. & Dawkins, M. S. Receiver psychology and the evolution of animal signals. Anim. Behav. 42 , 1–14 (1991). Speed, M. P. Warning signals, receiver psychology and predator memory. Anim. Behav. 60 , 269–278 (2000). Shettleworth, S. J. The role of novelty in learned avoidance of unpalatable ‘Prey’ by domestic chicks (Gallus Gallus). Anim. Behav. 20 , 29–35 (1972). Roper, T. J. & Redston, S. Conspicuousness of distasteful prey affects the strength and durability of one-trial avoidance learning. Anim. Behav. 35 , 739–747 (1987). Gamberale-Stille, G. & Tullberg, B. S. Experienced chicks show biased avoidance of stronger signals: an experiment with natural colour variation in live aposematic prey. Evol. Ecol. 13 , 579–589 (1999). Aronsson, M. & Gamberale-Stille, G. Domestic chicks primarily attend to colour, not pattern, when learning an aposematic coloration. Anim. Behav. 75 , 417–423 (2008). Aronsson, M. & Gamberale-Stille, G. Evidence of signaling benefits to contrasting internal color boundaries in warning coloration. Behav. Ecol. 24 , 349–354 (2013). Medina, I. et al. Global selection on insect antipredator coloration. Science 389 , 1336–1341 (2025). Isola, P., Xiao, J., Parikh, D., Torralba, A. & Oliva, A. What Makes a Photograph Memorable? IEEE Trans. Pattern Anal. Mach. Intell. 36 , 1469–1482 (2014). Bainbridge, W. A., Dilks, D. D., Oliva, A. & Memorability A stimulus-driven perceptual neural signature distinctive from memory. NeuroImage 149 , 141–152 (2017). Rust, N. C. & Mehrpour, V. Understanding Image Memorability. Trends Cogn. Sci. 24 , 557–568 (2020). Harris, J. M., Penacchio, O. & Osorio, D. C. The Role of Vision Science in Understanding Animal Camouflage. Annu. Rev. Vis. Sci. 11 , 331–357 (2025). Cuthill, I. C. & Camouflage. J. Zool. 308 , 75–92 (2019). Osorio, D. & Vorobyev, M. A review of the evolution of animal colour vision and visual communication signals. Vis. Res. 48 , 2042–2051 (2008). Pinto, L. & Baron, J. Spatiotemporal frequency and speed tuning in the owl visual wulst. Eur. J. Neurosci. 30 , 1251–1268 (2009). Vorobyev, M. & Osorio, D. Receptor noise as a determinant of colour thresholds. Proc. R. Soc. Lond. B Biol. Sci. 265, 351–358 (1998). Anderson, A. J. & McOwan, P. W. Humans deceived by predatory stealth strategy camouflaging motion. Proc. R. Soc. Lond. B Biol. Sci. 270, (2003). Scott-Samuel, N. E., Baddeley, R., Palmer, C. E. & Cuthill, I. C. Dazzle Camouflage Affects Speed Perception. PLoS ONE . 6 , e20233 (2011). Penacchio, O., Harris, J. M. & Lovell, P. G. Establishing the behavioural limits for countershaded camouflage. Sci. Rep. 7 , 13672 (2017). Penacchio, O., Lovell, P. G. & Harris, J. M. Is countershading camouflage robust to lighting change due to weather? R Soc. Open. Sci. 5 , 170801 (2018). Webster, R. J., Hassall, C., Herdman, C. M., Godin, J. G. J. & Sherratt, T. N. Disruptive camouflage impairs object recognition. Biol. Lett. 9 , 20130501 (2013). Stevens, M. et al. Revealed by conspicuousness: distractive markings reduce camouflage. Behav. Ecol. 24 , 213–222 (2013). Troscianko, J., Lown, A. E., Hughes, A. E. & Stevens, M. Defeating Crypsis: Detection and Learning of Camouflage Strategies. PLoS ONE . 8 , e73733 (2013). Troscianko, J., Skelhorn, J. & Stevens, M. Quantifying camouflage: how to predict detectability from appearance. BMC Evol. Biol. 17 , 7 (2017). Flavell, J. H. Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. Am. Psychol. 34 , 906–911 (1979). Dunlosky, J. & Metcalfe, J. Metacognition. ix, 334 (Sage Publications, Inc, 2009). Undorf, M. & Bröder, A. Metamemory for pictures of naturalistic scenes: Assessment of accuracy and cue utilization. Mem. Cognit . 49 , 1405–1422 (2021). Yonelinas, A. P. The Nature of Recollection and Familiarity: A Review of 30 Years of Research. J. Mem. Lang. 46 , 441–517 (2002). Bylinskii, Z., Isola, P., Bainbridge, C., Torralba, A. & Oliva, A. Intrinsic and extrinsic effects on image memorability. Vis. Res. 116 , 165–178 (2015). Bainbridge, W. A. Chapter One - Memorability: How what we see influences what we remember. in Psychology of Learning and Motivation (eds (eds Federmeier, K. D. & Beck, D. M.) vol. 70 1–27 (Academic, (2019). Jaegle, A. et al. Population response magnitude variation in inferotemporal cortex predicts image memorability. eLife 8 , e47596 (2019). Khosla, A., Raju, A. S., Torralba, A. & Oliva, A. Understanding and Predicting Image Memorability at a Large Scale. in IEEE International Conference on Computer Vision (ICCV) 2390–2398 (IEEE, Santiago, Chile, 2015)., 2015). (2015). 10.1109/ICCV.2015.275 Needell, C. D. & Bainbridge, W. A. Embracing New Techniques in Deep Learning for Estimating Image Memorability. Comput. Brain Behav. 5 , 168–184 (2022). Penacchio, O. et al. A computational neuroscience framework for quantifying warning signals. Methods Ecol. Evol. 15 , 103–116 (2024). Bainbridge, W. A. The memorability of people: Intrinsic memorability across transformations of a person’s face. J. Exp. Psychol. Learn. Mem. Cogn. 43 , 706–716 (2017). Stevens, M. & Ruxton, G. D. Linking the evolution and form of warning coloration in nature. Proc. R. Soc. B Biol. Sci. 279, 417–426 (2012). Borkin, M. A. et al. What Makes a Visualization Memorable? IEEE Trans. Vis. Comput. Graph . 19 , 2306–2315 (2013). Davis, T. M. & Bainbridge, W. A. Memory for artwork is predictable. Proc. Natl. Acad. Sci. U. S. A. 120, e2302389120 (2023). Ongchoco, J. D. K., Chun, M. M. & Bainbridge, W. A. What moves us? The intrinsic memorability of dance. J. Exp. Psychol. Learn. Mem. Cogn. 49 , 889–899 (2023). Wakeland-Hart, C. D., Cao, S. A., deBettencourt, M. T., Bainbridge, W. A. & Rosenberg, M. D. Predicting visual memory across images and within individuals. Cognition 227 , 105201 (2022). Geiger, A. R. & Balas, B. Robot face memorability is affected by uncanny appearance. Comput. Hum. Behav. Rep. 4 , 100153 (2021). Palan, S. & Schitter, C. Prolific.ac—A subject pool for online experiments. J. Behav. Exp. Finance . 17 , 22–27 (2018). Peirce, J. et al. PsychoPy2: Experiments in behavior made easy. Behav. Res. Methods . 51 , 195–203 (2019). Nilsback, M. E. & Zisserman, A. Automated Flower Classification over a Large Number of Classes. in Sixth Indian Conference on Computer Vision, Graphics & Image Processing 722–729 (2008). 722–729 (2008). (2008). 10.1109/ICVGIP.2008.47 Bates, D., Mächler, M., Bolker, B. & Walker, S. Fitting Linear Mixed-Effects Models Using lme4. J. Stat. Softw. 67 , 1–48 (2015). Lüdecke, D., Ben-Shachar, M., Patil, I., Waggoner, P. & Makowski, D. performance: An R Package for Assessment, Comparison and Testing of Statistical Models. J. Open. Source Softw. 6 , 3139 (2021). Bartoń, K. & MuMIn Multi-Model Inference. 1.48.11 (2010). Brown, W. Some experimental results in the correlation of mental abilities. Br. J. Psychol. 3 , 296–322 (1910). Spearman, C. Correlation calculated from faulty data. Br. J. Psychol. 3 , 271–295 (1910). Additional Declarations No competing interests reported. Supplementary Files SupplementaryInformationDeFilippietal.pdf Cite Share Download PDF Status: Published Journal Publication published 25 Feb, 2026 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 25 Nov, 2025 Editor assigned by journal 23 Nov, 2025 Submission checks completed at journal 23 Nov, 2025 First submitted to journal 21 Nov, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8175452","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":550569353,"identity":"0f502aeb-7042-43f8-b4d3-84fb20fed58b","order_by":0,"name":"Federico De Filippi","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA4klEQVRIiWNgGAWjYBAC9obEBwc+QNgGcFHGBjxaeA4kGx6cAVFNvBbjwzykaWFPZjhs8+cPA//s5m0fPrbVMfC3H2CTnIFPC89jhsO5bQYMEneOFc+c2XaYQeJMApvkBjxa7CXyDxzObTCob7iRY8zMu+0AA8MNBjbJB/hskQA6zOKPAYM8SMvfbXVABjFaGNgMGAxAWhi3MQMZDPgdBvLLwd42YwbDG2nFjL3/DvMYnklstsTrffZk5g8//sgxyN1I3szw40ydnNzxwwdv9uDRgmkGgVgZBaNgFIyCUUAMAACI4UxJy6YsDQAAAABJRU5ErkJggg==","orcid":"","institution":"University of St Andrews","correspondingAuthor":true,"prefix":"","firstName":"Federico","middleName":"","lastName":"De Filippi","suffix":""},{"id":550569354,"identity":"b327bf40-8f6f-465a-b136-b3ba4ea86423","order_by":1,"name":"Olivier Penacchio","email":"","orcid":"","institution":"Computer Vision Center","correspondingAuthor":false,"prefix":"","firstName":"Olivier","middleName":"","lastName":"Penacchio","suffix":""},{"id":550569355,"identity":"55940fd7-b148-4899-a9c1-49f859e64dbd","order_by":2,"name":"Akira R. O’Connor","email":"","orcid":"","institution":"University of St Andrews","correspondingAuthor":false,"prefix":"","firstName":"Akira","middleName":"R.","lastName":"O’Connor","suffix":""},{"id":550569356,"identity":"78e6ba21-ff54-42dc-a50d-058eb76de4c2","order_by":3,"name":"Julie M. Harris","email":"","orcid":"","institution":"University of St Andrews","correspondingAuthor":false,"prefix":"","firstName":"Julie","middleName":"M.","lastName":"Harris","suffix":""}],"badges":[],"createdAt":"2025-11-21 16:38:20","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8175452/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8175452/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-026-41178-y","type":"published","date":"2026-02-25T15:58:13+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":97122658,"identity":"951c2b88-14e0-4187-a122-a0e18816d084","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1265844,"visible":true,"origin":"","legend":"","description":"","filename":"Naturalwarningsignalsunexpectedlyshapehumanmetamemoryratingsbutnotimagerecognitionsuccess.docx","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/8cf08ec7532c5a344ae190a2.docx"},{"id":97122650,"identity":"ba4a4945-bc97-45fd-a38a-321f4d43b2c8","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"json","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6105,"visible":true,"origin":"","legend":"","description":"","filename":"1caa03bd222e47c88507bb3cbf51b817.json","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/d14d38cc5fa6f1f6f4c0f271.json"},{"id":97122642,"identity":"d5cc6674-e183-402d-9fc2-0282ebd9f1e6","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"pdf","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1196797,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformationDeFilippietal.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/bdafd2c7d0606355e083ed92.pdf"},{"id":97122649,"identity":"6f4ba4a1-f1c8-4d38-96d9-93a8b95ca51a","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":125530,"visible":true,"origin":"","legend":"","description":"","filename":"1caa03bd222e47c88507bb3cbf51b8171enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/7b1bd89013a66b048c286ad3.xml"},{"id":97122646,"identity":"925fb91c-9bbc-4d07-aae3-b6fb33512b65","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"pdf","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":478152,"visible":true,"origin":"","legend":"","description":"","filename":"FIGURE1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/2d3e112b0206ecd35cd3678a.pdf"},{"id":97122659,"identity":"f7be2d51-4b45-41e7-95d0-c8a4ec4a6cb4","added_by":"auto","created_at":"2025-12-01 08:01:54","extension":"pdf","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":681624,"visible":true,"origin":"","legend":"","description":"","filename":"FIGURE2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/7911d9e9b0188edc30a023db.pdf"},{"id":97141703,"identity":"e74c462d-6369-4a54-a6b4-f16b4e387e09","added_by":"auto","created_at":"2025-12-01 10:06:54","extension":"pdf","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":252213,"visible":true,"origin":"","legend":"","description":"","filename":"FIGURE3.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/275e54bbdf9a0239a6f2c3a4.pdf"},{"id":97122643,"identity":"ff74b7fd-13f8-4230-b0f7-ffdd04e17dcc","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"pdf","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":414520,"visible":true,"origin":"","legend":"","description":"","filename":"FIGURE4.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/0fc643a3449deddc8d62fa65.pdf"},{"id":97122661,"identity":"abec12a5-ded9-48b8-8a85-9a08d7eefbd9","added_by":"auto","created_at":"2025-12-01 08:01:54","extension":"pdf","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1180380,"visible":true,"origin":"","legend":"","description":"","filename":"FIGURE5.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/a8bb898c768a966782158b4e.pdf"},{"id":97122660,"identity":"f1aa2c79-3e69-45e8-a612-7531ad8e3be6","added_by":"auto","created_at":"2025-12-01 08:01:54","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":52214,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/5230b2e1c60abefb2fc08df1.png"},{"id":97141673,"identity":"acc29ff9-a700-4f33-8d27-b79b96f9345f","added_by":"auto","created_at":"2025-12-01 10:06:53","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":65567,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/9f03de6bd9bd003a2c21a104.png"},{"id":97122645,"identity":"21768eef-2d99-4b6d-8e9e-adaad9d0ed77","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57565,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/7d4ce0f6f8a7fad58b3f362b.png"},{"id":97141691,"identity":"6b44b707-ef79-489f-a1d6-6e5ea3842e49","added_by":"auto","created_at":"2025-12-01 10:06:53","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":53876,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/ae5852c401fce4b00a1e4860.png"},{"id":97122655,"identity":"4148b720-b364-48c3-9235-08a92632fb98","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33707,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/874fc6daccdcfbbf63f52ad5.png"},{"id":97122653,"identity":"1d65d889-6ded-473e-b6b3-9c423503a91b","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"xml","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126014,"visible":true,"origin":"","legend":"","description":"","filename":"1caa03bd222e47c88507bb3cbf51b8171structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/075cad8a60e5d1a73440b48f.xml"},{"id":97122654,"identity":"af72a188-57a2-45bd-a82a-6db1ebc98cf9","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"html","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":139061,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/82151231fbee22e9c2c43792.html"},{"id":97142019,"identity":"930ccd60-2e82-4501-a92c-d30022e4c30a","added_by":"auto","created_at":"2025-12-01 10:07:17","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":256866,"visible":true,"origin":"","legend":"\u003cp\u003e(\u003cstrong\u003eA\u003c/strong\u003e): Examples of species from the St Andrews Hyperspectral Lepidoptera Database (\u003ca href=\"https://arts.st-andrews.ac.uk/lepidoptera/index.html\"\u003ehttps://arts.st-andrews.ac.uk/lepidoptera/index.html\u003c/a\u003e), sorted by the average metamemory rating given by human observers (0 = ‘very forgettable’, 1 = ‘very memorable’). The red dots mark aposematic (AP) species. The percentage is the average metamemory rating for that row of images. The ‘memorable’ and ‘forgettable’ rows show the images with the highest and lowest mean metamemory rating. The ‘typical’ row shows the images with the smallest differences from the median of scores. The mean metamemory rating for each species was calculated by averaging ratings across the whole subject pool. (\u003cstrong\u003eB\u003c/strong\u003e): Metamemory rating for each image, grouped by whether species were aposematic (AP, red dots) or not (non-AP, grey dots). Each point shows the mean rating of one image across the subject pool. Points are randomly jittered along the x-axis. Error bars display the mean rating ± 1 standard error of the mean.\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/efbf70f8be3fa45121fc0380.png"},{"id":97142304,"identity":"b5d0013b-fc8e-470f-b083-641ccea795d5","added_by":"auto","created_at":"2025-12-01 10:07:29","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":323514,"visible":true,"origin":"","legend":"\u003cp\u003e(\u003cstrong\u003eA\u003c/strong\u003e) Examples of species from the St Andrews Hyperspectral Lepidoptera Database, sorted by the proportion of correct recognitions (i.e., hit rate) they elicited across the subject pool. Red dots mark aposematic (AP) species. The percentage for each row shows the mean recognition hit rate for that row of images. The ‘memorable’ and ‘forgettable’ rows show the images with the highest and lowest mean hit rates. The ‘typical’ row shows the images with the smallest differences from the median hit rate. (\u003cstrong\u003eB\u003c/strong\u003e): Recognition hit rate for each image, grouped by whether species were aposematic (AP, red dots) or not (non-AP, grey dots). Each point shows the mean rating of one image across the subject pool. Points are randomly jittered along the x-axis. Error bars display the mean rating ± 1 standard error.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/f9e74d56a4e05d75ffb75ba4.png"},{"id":97122647,"identity":"4a9419d0-7847-48dc-a43d-6dc8e8197107","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":247796,"visible":true,"origin":"","legend":"\u003cp\u003e(\u003cstrong\u003eA\u003c/strong\u003e) Examples of the strongest matches (‘expected’) and mismatches (‘unexpected’) between observer’s perception of how memorable each species appeared when seen for the first time (metamemory rating), and recognition success (hit rate). The red dot highlights aposematic (AP) species. For example, species from the top-right set were firstly rated as highly memorable and subsequently recognized by 79% of observers. In contrast, species from the bottom-right set were also rated as highly memorable but recognised only by 42% of observers. (\u003cstrong\u003eB\u003c/strong\u003e): scatterplot and best linear fits of the correlation between recognition and metamemory ratings, grouped by Lepidoptera class.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/e849ed4c33a46fbf93401829.png"},{"id":97141739,"identity":"d6e597c1-7e51-4e84-9c67-632ff8c29720","added_by":"auto","created_at":"2025-12-01 10:06:55","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":199024,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of the consistency analysis used to estimate the intrinsic memorability of aposematic (AP) and non-aposematic (non-AP) Lepidoptera. (\u003cstrong\u003eA\u003c/strong\u003e) Split-half reliability (i.e., Spearman-Brown corrected correlation coefficient) calculated from 1000 random splits of subjects, grouped by Lepidoptera class. Each point shows the split-half reliability between two random splits from one iteration of the consistency analysis. Error bars show the mean split-half reliability ± bootstrapped 95% confidence intervals for each class. (\u003cstrong\u003eB\u003c/strong\u003e) Memory consistency across observers as a function of the hit rate of images. Hit rates for each image were sorted in descending order according to one group (x-axis) and plotted against the cumulative average of the hit rates (y-axis) in another group, or in a randomly shuffled distribution of hit rates. When group j and group k curve overlap and differ the most from the shuffled group curve, consistency between observers is high. Each point is the average hit rate across all 1000 iterations of the analysis. Error bars show means ± bootstrapped 95% confidence intervals of the cumulative hit rates across all iterations.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/dcb8a43b787925cd92d4d5ff.png"},{"id":97122663,"identity":"71172e45-6a82-4999-8a5e-bc1df8a1bca9","added_by":"auto","created_at":"2025-12-01 08:01:55","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":147964,"visible":true,"origin":"","legend":"\u003cp\u003eExamples of one trial from the metamemory and recognition blocks. (\u003cstrong\u003eA\u003c/strong\u003e) In the metamemory block, participants viewed all target images and rated each one based on how likely they thought they would remember it later. (\u003cstrong\u003eB\u003c/strong\u003e) In the recognition block, participants were shown a mix of previously seen images (‘targets’), novel ones (‘lures’), as well as task-irrelevant vigilance-check images, and had to decide whether each stimulus was ‘old’ (seen before) or ‘new’ (not seen before).\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/cbee5c063097fd9a77e1c004.png"},{"id":103765668,"identity":"2bf269f9-1a83-4c5c-9734-74373f5428fa","added_by":"auto","created_at":"2026-03-02 16:07:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1993111,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/8da453ba-fddd-4bfd-8a11-e6eab2691360.pdf"},{"id":97122641,"identity":"1f4e9bde-b543-4273-bfc3-20ffae4896cb","added_by":"auto","created_at":"2025-12-01 08:01:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1196797,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformationDeFilippietal.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8175452/v1/5ff1914c070242fd9379ceef.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Natural warning signals unexpectedly shape human metamemory ratings but not image recognition success","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWhen we glance at an image, it can leave a lasting trace in memory, making it easy to recognise when encountered again \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. This apparently effortless encoding of visual information is not only a hallmark of human memory but may also have an evolutionary significance in nature. One adaptation thought to exploit these processes is aposematism\u0026mdash;the use of conspicuous colours and patterns as honest \u0026lsquo;warning signals\u0026rsquo; of prey toxicity or unpalatability \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. How this visual defence strategy influences predators\u0026rsquo; foraging decisions is still not fully understood \u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eThe effectiveness of a warning signal is commonly attributed to the ease with which it is remembered by predators \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. Conspicuous patterns combining red, orange, and yellow with black, typically in striped or spotted arrangements, are successful aposematic signals, reliably avoided by predators in both natural and experimental contexts \u003csup\u003e[\u003cspan additionalcitationids=\"CR9 CR10 CR11 CR12\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. Most studies have focussed on how the conspicuousness of a signal promotes avoidance when predators are repeatedly exposed to it. To the best of our knowledge, there have been no attempts to directly compare how aposematic and non-aposematic species are recognized from memory on a purely visual level.\u003c/p\u003e\u003cp\u003eHere, we drew on well-known properties of human visual memory for images \u003csup\u003e[\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e to test whether images of Lepidoptera (butterfly and moth species) that display warning signals are easier to remember than those that do not. A growing trend in sensory ecology leverages human vision science to understand how animal patterning influences the perceptual mechanisms of the observer (reviewed in \u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e). There is evidence that the effectiveness of warning signals is rooted in early vision mechanisms, such as edge detectors \u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e and colour opponency processes \u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e, which are broadly shared by humans and birds \u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e, arguably the primary predators of Lepidoptera \u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. Researchers have used human observers to test what disrupts or enhances the detectability of prey targets, revealing key principles of camouflage and conspicuousness \u003csup\u003e[\u003cspan additionalcitationids=\"CR23 CR24 CR25 CR26 CR27 CR28\" citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e. This comparative approach allows traits that have proved adaptive in nature to be characterised in detail through measuring human visual behaviour. The ability to be remembered is considered crucial for effective animal signalling \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e and can be easily measured in humans. However, humans have not yet been used to assess whether warning signals provide an advantage in recognising prey from memory.\u003c/p\u003e\u003cp\u003eIn this study, we will investigate three aspects of human memory relevant to understanding aposematism, defined below.\u003c/p\u003e\u003cp\u003e(1) \u003cem\u003eMetamemory\u003c/em\u003e: When we view a stimulus, we can form intuitions about how likely we are to remember it in the future, i.e., we ask people to make a metamemory rating\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/sup\u003e. Metamemory ratings for photographs can be accurate when based on specific cues (e.g., whether images \u0026lsquo;tell a story\u0026rsquo; \u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e), but can also be highly inaccurate when observers lack these cues \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Possibly related to this measure, warning-coloured stimuli for predator avoidance experiments are often chosen according to how conspicuous or distinctive they appear to human observers.\u003c/p\u003e\u003cp\u003e(2) \u003cem\u003eRecognition memory\u003c/em\u003e: Humans can store detailed representations of thousands of images in memory, even after seeing them only once \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. When an image is viewed for the second time, determining whether it has been seen before (i.e., measuring recognition ) can be thought of as being effortless (\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e, though see also \u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e) This ability makes humans a useful model for testing whether aposematic signals exploit memory mechanisms.\u003c/p\u003e\u003cp\u003e(3) \u003cem\u003eMemorability\u003c/em\u003e: Recognition is not always successful, the chances that people will correctly recognise an image can vary across images\u0026mdash;some images are reliably recognised by most people, but others only by a few. Remarkably, whether one person recognises an image can be predicted from whether others do \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e, indicating that recognition is not entirely idiosyncratic, but partly image-specific \u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/sup\u003e. This is how memorability is defined: an intrinsic property of an image, whose likelihood of recognition can be predicted \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e. Neuroimaging \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e and electrophysiology \u003csup\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e suggest that memorability has a perceptual basis, related to how the visual system responds to images, and it can also be predicted computationally from image content alone \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e. Memorability thus provides a framework to test whether aposematic species possess visual features that intrinsically influence predator decisions, revealing a previously unexplored dimension of their adaptive function.\u003c/p\u003e\u003cp\u003ePrevious work from our group constructed a database of hyperspectral images of aposematic and non-aposematic Lepidoptera and used computational neuroscience methods to characterise how coloration patterns may stimulate the brain of predator birds \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. The modelling demonstrates that the patterns of aposematic species evoke stronger and less sparse neural responses than those from non-aposematic ones, providing a framework to quantify warning signals based on visual inputs alone. Here, we asked whether these same warning signals might also influence human memory. We selected species from the database predicted to evoke the strongest and weakest responses in the avian visual pathway and measured human memory in response to these images. We tested whether aposematic species would (1) be rated by participants as easier to remember (metamemory rating), (2) be better remembered in a recognition task, and (3) promote consistent memory performance across individuals (memorability). Our aim was to assess whether visual features considered adaptive in ecological contexts truly enhance these aspects of memory.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eHuman observers viewed images of aposematic (AP) and non-aposematic (non-AP) Lepidoptera species and provided subjective estimates of how likely it was they would remember each species if encountered again (i.e., metamemory rating). Later, observers completed a recognition memory test, where they viewed a mix of previously seen images (i.e., \u0026lsquo;targets\u0026rsquo;) and novel ones (i.e., \u0026lsquo;lures\u0026rsquo;) and had to decide whether each image was \u0026lsquo;old\u0026rsquo; or \u0026lsquo;new\u0026rsquo; (i.e., recognition judgement). For details on the experimental task and selection of visual stimuli, see \u003cem\u003eMaterials and Methods\u003c/em\u003e. We compared the effects of AP and non-AP species on observers\u0026rsquo; metamemory ratings and recognition. Subsequently, we analysed the consistency of recognition across observers, to estimate the amount of intrinsic memorability delivered by AP and non-AP species.\u003c/p\u003e\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eMetamemory rating: Do aposematic species appear more memorable?\u003c/h2\u003e\u003cp\u003eWhen each target species was viewed for the first time, observers provided a metamemory rating, a score describing the perceived likelihood of remembering the image when encountered in the future (see \u003cem\u003eMaterials and Methods\u0026mdash;Metamemory rating block\u003c/em\u003e). Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eA shows examples of species that were rated as most memorable, typical, and forgettable. Figure\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003eB shows the average metamemory rating for each image, grouped by whether items were AP or non-AP.\u003c/p\u003e\u003cp\u003eTo test whether AP species were perceived as more likely to be remembered than non-AP ones, we fitted a linear mixed-effects model predicting metamemory ratings from Lepidoptera class (AP vs. non-AP), including both random intercepts and slopes for participants [[rating\u0026thinsp;~\u0026thinsp;class + (class | participant)]]. There was a significant effect of class on metamemory (β\u0026thinsp;=\u0026thinsp;0.16, SE\u0026thinsp;=\u0026thinsp;0.02, t\u0026thinsp;=\u0026thinsp;8.75, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001), indicating on average, AP species (mean\u0026thinsp;=\u0026thinsp;0.61, SE\u0026thinsp;=\u0026thinsp;0.02) were rated 0.16 units higher than non-AP species (mean\u0026thinsp;=\u0026thinsp;0.45, SE\u0026thinsp;=\u0026thinsp;0.02). Random effect estimates showed individual differences in ratings: the standard deviation of the participant intercepts was 0.16, and the standard deviation of the Lepidoptera class slopes was 0.12. This indicates that most observers rated AP species between approximately 0.04 and 0.28 units higher than non-AP species. The proportion of variance in metamemory ratings explained by the fixed effect (marginal R\u0026sup2;) was 10.8%. The proportion of variance explained by fixed plus random effects (conditional R\u0026sup2;), was 47.5%. These results confirm that, despite individual differences, on average, AP species were perceived as more likely to be remembered than non-AP species upon first sight.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eRecognition: Are aposematic species easier to remember?\u003c/h3\u003e\n\u003cp\u003eWe next measured whether images were recognised when viewed again. In an old/new recognition test, a previously seen target is either successfully recognised as familiar (i.e., \u0026lsquo;hit\u0026rsquo;) or not recognised (i.e., \u0026lsquo;miss\u0026rsquo;); similarly, a lure can either be successfully recognised as novel (i.e., \u0026lsquo;correct rejection\u0026rsquo;) or mistakenly judged as familiar (i.e., \u0026lsquo;false alarm\u0026rsquo;). To evaluate whether AP target species were more likely to be recognised than non-AP ones, we focused on hits. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA shows examples of species sorted by the hit rate. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB shows the average hit rate for each image, grouped by whether items were AP or non-AP.\u003c/p\u003e\u003cp\u003eTo test whether AP species were more likely to be recognised from memory than non-AP ones, we fitted a logistic mixed-effects model predicting memory outcome (hit vs. miss) from Lepidoptera class (AP vs. non-AP), including both random intercepts and slopes for participants [[recognition\u0026thinsp;~\u0026thinsp;class + (class | participant)]]. There was no significant effect of Lepidoptera class on recognition outcome (β\u0026thinsp;=\u0026thinsp;0.06, SE\u0026thinsp;=\u0026thinsp;0.13, z\u0026thinsp;=\u0026thinsp;0.45, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.651), meaning that AP species (mean\u0026thinsp;=\u0026thinsp;0.65, SE\u0026thinsp;=\u0026thinsp;0.03) were not more likely to be recognised than non-AP species (mean\u0026thinsp;=\u0026thinsp;0.63, SE\u0026thinsp;=\u0026thinsp;0.02) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB). The proportion of variance in recognition outcomes explained by the fixed effect alone was close to zero (0.02%), while the variance explained by the full model including random effects was 10.84%. These results indicate that, contrary to our expectations, and to prevailing suggestions in the literature regarding the effect of warning signals on predators in natural systems, AP species did \u003cem\u003enot\u003c/em\u003e confer an overall advantage in human memory using a recognition task, eliciting, on average, the same hit rate as non-AP species (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB).\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eComparing recognition and metamemory ratings: Is memory for Lepidoptera intuitive?\u003c/h3\u003e\n\u003cp\u003eCan we predict whether species would be remembered or forgotten in the future simply based on how \u0026lsquo;memorable\u0026rsquo; they appear to us at first exposure? The answer is hinted at by the evident discrepancy between the observed metamemory and memory results (see Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Here we provide a more detailed analysis. We calculated the difference between metamemory ratings and recognition hit rates for each image. We then plotted the species with the smallest (\u0026lsquo;expected\u0026rsquo;: strongest matches between the two measures) and largest (\u0026lsquo;unexpected\u0026rsquo;: weakest matches) differences, shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA. Among the \u0026lsquo;expected\u0026rsquo; cases, AP species with visually distinctive markings and colours were judged memorable and later recognised (\u0026lsquo;memorable\u0026ndash;expected\u0026rsquo;), whereas non-AP species that seemingly lack warning colouration were judged most forgettable and later forgotten (\u0026lsquo;forgettable\u0026ndash;expected\u0026rsquo;). By contrast, the most \u0026lsquo;unexpected\u0026rsquo; species showed less obvious visual distinctions between recognised and forgotten images. To test observers\u0026rsquo; metamemory accuracy, we calculated the correlation between the metamemory scores of each image and recognition hit rates (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). Overall, metamemory and recognition across both classes were uncorrelated (Spearman\u0026rsquo;s r\u0026thinsp;=\u0026thinsp;0.01, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.930). There was a non-significant correlation for AP species (r\u0026thinsp;=\u0026thinsp;0.21, \u003cem\u003ep\u0026thinsp;=\u003c/em\u003e\u0026thinsp;.280), and there was a small but significant negative correlation for non-AP species (r = -0.39, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.037). See Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\n\u003ch3\u003eMemorability: Is memory for Lepidoptera predictable across observers?\u003c/h3\u003e\n\u003cp\u003eHaving found clear differences in metamemory between AP and non-AP species, but not in memory (as measured by recognition), we next asked whether warning signals might influence recognition in ways not captured by overall performance measures. To assess this point, it is necessary to quantify individual differences using a \u0026lsquo;consistency analysis\u0026rsquo; \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e, a standard method in memorability research for evaluating if memory is predictable across observers. The consistency analysis involves repeatedly splitting the full subject pool into two random halves (which we label group j and group k), calculating hit rates per image separately within each group, and assessing consistency in hit rates across groups using two complementary metrics.\u003c/p\u003e\u003cp\u003eThe first is split-half reliability, a metric which indicates how stable hit rates are across different observers, calculated as the Spearman-Brown corrected correlation between hit rates from each group j and k. We calculated an overall split-half reliability of r\u0026thinsp;=\u0026thinsp;0.80 (95% CI [0.80, 0.81]) across all Lepidoptera species. Next, to assess differences in split-half reliability between Lepidoptera classes, we calculated it separately for AP and non-AP species. The average split-half reliability for AP species was ρ\u0026thinsp;=\u0026thinsp;0.90 (95% CI [0.90, 0.90]) and for non-AP species it was ρ\u0026thinsp;=\u0026thinsp;0.56 (95% CI [0.56, 0.55]). This is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA. Note that split-half reliability quantifies consistency in rank-ordering of hit rates between groups but is independent of overall recognition accuracy. Correlations can be high whether participants tend to remember or forget the same images \u003csup\u003e[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/sup\u003e. Conversely, low split-half reliability suggests that participants disagreed about which images were remembered or forgotten. The difference in split-half reliability between AP and non-AP species was statistically significant (Δρ\u0026thinsp;=\u0026thinsp;0.35, permutation test, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001, see \u003cem\u003eMaterials \u0026amp; Methods\u0026mdash;Statistical testing of consistency measures\u003c/em\u003e).\u003c/p\u003e\u003cp\u003eThe second metric, shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB, uses cumulative hit rates across the different subject splits \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e to estimate how well performance from group j predicts that of group k, providing a more detailed breakdown of consistency as a function of recognition accuracy. If particular images are especially likely to be remembered (or forgotten) by one group, will a different group show the same pattern? To answer this question, image-wise hit rates were sorted in descending order (i.e., from most memorable to forgettable) according to group j and compared against those from group k, or against a baseline shuffled group, created by randomly permuting the association between images and their hit rates in group k, while keeping the image ordering for group j fixed. The curves in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB were constructed by calculating the cumulative hit rate (mean hit rate of top-n images) for each group, while moving across this fixed image ordering (i.e., from the top-2 to the top-28 most remembered images), for each of the 1000 random subject splits shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA. High consistency yields a close match between the curves for group j and group k, and a separation from the shuffled group curve, indicating that different individuals remembered and forgot the same images in a systematic, non-random way. For example, Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB shows that the top three most-remembered AP images were recognised by approximately 95% of observers in one group and by 85% in another, whereas the top three non-AP images were recognised approximately by 80% and 65%, respectively.\u003c/p\u003e\u003cp\u003eTo assess the predictability of hit rates across observers, we calculated the difference in cumulative hit rates between groups at each image rank (see \u003cem\u003eMaterials \u0026amp; Methods\u0026mdash;Statistical testing of consistency measures\u003c/em\u003e). For AP species, we found significant differences between group k and the shuffled group at all image ranks except the last rank (Bonferroni-corrected \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001 for all ranks, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.000 for rank 28). For non-AP species, group k significantly differed from the shuffled group at all ranks except the first and last rank (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.000) and marginally at rank 2 (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.057; all other ranks \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001). Between classes, AP and non-AP species differed significantly in how much the cumulative hit rate of group k exceeded its shuffled baseline (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001 at all ranks except rank 26, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1.000). This is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB, where the group k and j curves for AP species remained closely aligned across most ranks, whereas non-AP species showed weaker alignment, especially for the most memorable images.\u003c/p\u003e\u003cp\u003eOverall, the consistency analysis confirmed that hit rates were significantly predictable for both classes, but AP species strongly increased the inter-observer consistency of hit rates, across the entire spectrum of memory accuracy, i.e., both for remembered and forgotten images.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study tested whether the warning signals of aposematic (AP) Lepidoptera species enhance visual memory in human observers. Our measurement of metamemory rating showed that AP patterns appear to participants as though they will be more memorable than non-AP patterns. Surprisingly, this was not backed up by actual memory performance: recognition was not better for AP than for non-AP patterns. Yet, our analysis showed that people do tend to remember the same images as one another, and their recognition performance is more similar to those of others when the species carry AP patterns. In the following, we discuss how these results fit within the current literature and offer a novel contribution to understanding the cognitive processes underlying visual warning signals.\u003c/p\u003e\u003cp\u003eFor this study we chose to compare various aspects of memory performance for 2 sets of patterns, those known to be from AP species of Lepidoptera, and those from non-AP. In previous research, we used computational modelling of both the colour and luminance channels involved in visual processing, to show that the AP species used in this study evoke stronger activity in visual brain areas than the non-AP species \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. We wanted to test whether warning signal (AP) patterns are also processed differently by memory systems than patterns likely to offer camouflage (non-AP). We first measured metamemory by asking participants to rate how well they thought they would remember each pattern. The AP patterns were the ones that our participants chose as most likely to be remembered. This result did not surprise us. It is reminiscent of the long-standing idea, dating back to Darwin and Poulton \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e and still reiterated today \u003csup\u003e[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/sup\u003e, that warning signals appear particularly \u0026lsquo;striking\u0026rsquo; to the human eye. However, as discussed in more detail below, our metamemory impression proved inaccurate: AP patterns did not lead to better performance in the recognition task. We do not know why visually distinctive patterns drive us to think they will be easily remembered. Our results do concur with some related human memory literature. For example, Isola et al. \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e studied memory for photographs of complex scenes. Participants were unable to predict which photograph would be remembered later in a repeat-detection task, showing no significant correlation between the metamemory measure and a repeat-detection memory measure. Thus, akin to our study, participants intuition for what will be well remembered was incorrect. Finally, to bring us back to the idea that warning signals are \u0026lsquo;striking\u0026rsquo; \u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e, Isola et al also found that measures of interestingness and image aesthetics correlated with metamemory. We speculate that whatever strikes us as potentially \u0026lsquo;memorable\u0026rsquo; about some patterns or photographs is at least partly driven by systems closely linked to vision, rather than solely by our memory systems.\u003c/p\u003e\u003cp\u003eOurs is the first study (to our knowledge) to directly measure in humans whether natural AP patterns are better remembered than non-AP patterns. Despite our two classes of image delivering different visual activity in a modelled visual system \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e, and our metamemory ratings also differing for the two pattern classes, in a recognition task, we found no evidence that AP species were recognised more accurately than non-AP species. Thus, the metamemory differences we observed did not clearly extend to subsequent recognition. Such a pattern of data is the signature of a form of memory known as memorability, a memory system thought to be linked closely to high level visual areas \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e. We will discuss this system in more detail below.\u003c/p\u003e\u003cp\u003eNote, however, that there is another emerging literature on metamemory for images, exploring how cues about the meaning of images can deliver highly accurate metamemory. For example, Undorf and Broder \u003csup\u003e[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e, showed accurate metamemory measures linked to specific attributes of scenes (e.g., peacefulness, inclusion of people, semantic distinctiveness). In our study, images were presented in an intentionally abstracted context: museum specimens on black backgrounds, so that cues to meaning were minimised. Hence, we consider this contrasting literature to be beyond the scope of our investigation.\u003c/p\u003e\u003cp\u003eOur findings do suggest that warning signals are doing something more subtle than triggering the memory systems involved in recognition. Some earlier authors \u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e have emphasised a similar complexity, in understanding predator learning and decision making when faced with aposematic prey. In a comprehensive review, these authors highlight that avoidance learning (where a predator learns over time that a prey item is toxic or is otherwise not valuable to eat) is governed by a complex balance between gathering information about the prey (e.g., via conspicuous warning signals) and the cost of eating the prey (e.g., a build-up of toxicity level). If gathering information about a prey which has a strong AP pattern is easier, then learning will likely be faster, but this does not require the patterns to be specifically easier to \u003cem\u003eremember\u003c/em\u003e. If the \u0026lsquo;striking\u0026rsquo; perception of humans when faced with AP patterns transfers to other predators, perception-led cognition, linked to the memorability of specific images, could drive learning, alongside or independent of traditional memory systems. Overall, our results suggest that a strong AP pattern is best characterised by how it appears at first exposure, rather than how easily it is recognised from memory later, shedding some light on what makes warning signals effective.\u003c/p\u003e\u003cp\u003eIn recent years, the concept of memorability, in particular applied to images, has been developed by the work of Bainbridge and others (see recent reviews \u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e). The key evidence used to define the notion of \u0026ldquo;intrinsic\u0026rdquo; memorability is that, when faced with many images, some are easy to remember, and some difficult, yet crucially, the same images tend to be remembered and forgotten by different individuals. We analysed our data with this concept in mind. When comparing split-half reliability between AP and non-AP patterns, we found that it was significant for both pattern classes, but higher for AP patterns: people are more likely to remember or forget the same patterns from this class, compared to non-AP. Furthermore, following \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e, we showed that there was high inter-observer consistency of both remembering and forgetting patterns, across the full range of performance (from very well remembered to very poorly remembered).\u003c/p\u003e\u003cp\u003eThis is reflected in our results by a near-perfect alignment in recognition performance across individuals, both for remembered and forgotten AP species (see Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB). This alignment was less pronounced for non-AP species.\u003c/p\u003e\u003cp\u003eConsistency in memory has been reported across a growing range of human-centred images designed for mass audiences, such as complex photographs (r\u0026thinsp;=\u0026thinsp;0.75 \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e), data visualisations (including infographics and graphs, r\u0026thinsp;=\u0026thinsp;0.83 \u003csup\u003e[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/sup\u003e), paintings (r\u0026thinsp;=\u0026thinsp;0.53 \u003csup\u003e[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/sup\u003e), and dance moves (r\u0026thinsp;=\u0026thinsp;0.51 \u003csup\u003e[\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]\u003c/sup\u003e). However, comparing measures of consistency should be done with caution. Our image sample size was necessarily small compared to most other studies (see Methods), which could potentially result in unreliability of correlations. With that caveat in mind, we did find stronger correlations here (r\u0026thinsp;=\u0026thinsp;0.80 for all species combined) than in most other studies. A key point for us is that consistency was considerably higher when patterns contained warning signals (r\u0026thinsp;=\u0026thinsp;0.90 for AP, r\u0026thinsp;=\u0026thinsp;0.56 for non-AP). Unlike the other studies, our image classes contained ecologically adaptive traits, thus we provide a first example of memory consistency being modulated by such traits, as well as being amongst the highest measured in any study (to our knowledge). Such a pattern suggests that animal patterns might engage the perceptual processes responsible for memorability, known to operate independently of traditional memory systems \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn an attempt to follow up this idea further, we applied a deep-learning model to our images, ResMem \u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e. ResMem has been shown to successfully predict memory for complex scenes \u003csup\u003e[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e, \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]\u003c/sup\u003e, but is less successful for single objects on artificial backgrounds (e.g., robot faces \u003csup\u003e[\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]\u003c/sup\u003e). For our Lepidoptera dataset on black backgrounds, the model produced trends sometimes opposite to human memory data (the \u003cem\u003eSupplementary Figure S6\u003c/em\u003e shows these results). This suggests that what drives memory for our animal images is not captured by the ResMem deep learning model based on complex scenes.\u003c/p\u003e\u003cp\u003eFurthermore, note that our images do not fully fit the standard memorability framework, at least in terms of the classes of images that have been studied so far. Recall that our two image classes were chosen not only on the basis that the species depicted were known to be aposematic, or not, but also that each member of the class provided a \u0026lsquo;good\u0026rsquo; example of that class, as determined by our neural modelling of the population response of early visual processing (\u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e, see Methods). Previous work has shown that basic pixel statistics (hue, saturation, intensity \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e) are not predictive of memorability, and that memorable vs. forgettable images matched on colour and spatial frequency do not evoke different responses in early visual areas \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. These findings have led to the notion that low-level vision may not significantly influence memorability. However, our work differs from these approaches because it is based on low-level vision modelling performed at a neural population level \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e, rather than on simple measurements of image statistics. The differences we found while using this approach suggest that future memorability studies may benefit from quantifying low-level image statistics in a more biologically inspired fashion. To sum up, our data offer the first evidence that evolved patterns could be processed by memorability mechanisms, although our image set and data differ from current memorability studies.\u003c/p\u003e\u003cp\u003eWe finish this Discussion with some thoughts and speculations about how our work fits into the issue of what brain regions and pathways are responsible for the memory effects we have found using AP and non-AP patterns. The standard \u0026lsquo;receiver psychology\u0026rsquo; view that warning signals are effective for animal predators partly because they are well remembered \u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e was put forward without the knowledge we have discussed above, of the subtle ways in which visual and memory brain pathways are distinct. We highlight some of these subtleties below.\u003c/p\u003e\u003cp\u003eBainbridge \u003csup\u003e[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e and her group recently found evidence for specific memorability processing areas of the brain. In an influential human neuroimaging paper \u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e, they identified a distinct neural processing stream for memorability. They used fMRI to compare processing for images (faces and natural scenes) known to be memorable or forgettable. Regions more sensitive to highly memorable stimuli included a variety of high-level visual areas, including the fusiform face area (FFA), the lateral occipital complex (LOC) and the parahippocampal place area (PPA). Yet, there was no activity differences between the image classes in early visual cortex. Sensitivity was also found in a range of memory-related areas, but the effects of memorability were distinct from those found for subsequent memory. Consistent with these findings, our behavioural data reveal a clear dissociation between visual ratings and subsequent memory, but also high memory consistency across observers. This suggests that our stimuli may be differentially processed within a distinct memorability network, offering insight into how and where warning signals might be represented (at least in the human brain).\u003c/p\u003e\u003cp\u003eIn conclusion, this study suggests that the visual processing of certain animal patterns may influence the initial viewing of an image in ways that make it appear more \u0026lsquo;memorable\u0026rsquo; to the human eye. Given the signature behavioural pattern of intrinsic memorability that our images elicited, they are likely to be processed by vision-specific memorability pathways. Our work has begun to unpack the idea that AP patterns could be \u0026lsquo;easily remembered\u0026rsquo; to reveal which brain mechanisms are involved in responding to animal warning signals.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eMaterials andmethods\u003c/h2\u003e\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\u003ch2\u003eParticipants\u003c/h2\u003e\u003cp\u003e57 participants recruited via the Prolific (www.prolific.com\u003c/a\u003e\u003c/span\u003e\u003cspan address=\"http://www.prolific.com\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003csup\u003e[\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]\u003c/sup\u003e) participant pool took part in the study. 7 participants who failed one or more vigilance checks (see \u003cem\u003eMaterials \u0026amp; Methods\u0026mdash;Vigilance check\u003c/em\u003e) were excluded from the data pool, leaving us with N\u0026thinsp;=\u0026thinsp;50 (25 male, 25 female, mean age\u0026thinsp;=\u0026thinsp;28.98, SD\u0026thinsp;=\u0026thinsp;8.30). The Prolific reimbursement rate for taking part in the study was set to \u0026pound;9 per hour (median completion time\u0026thinsp;=\u0026thinsp;27 minutes). Participant selection criteria were normal or corrected-to-normal vision and no colour blindness. Informed consent was obtained from participants before the study. Participants were informed that they would be asked to look at pictures of butterflies and moths and provide subjective judgements, but information about which species were AP was not shared. Prior to being made available on Prolific, the study was approved by the local research ethics panel (UTREC, School of Psychology \u0026amp; Neuroscience, University of St Andrews).\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e\n\u003ch3\u003eSelection of stimuli\u003c/h3\u003e\n\u003cp\u003eWhen selecting Lepidoptera species to serve as stimuli in the memory experiment, our specific aim was to compile two sets of images (AP and non-AP species) predicted to be represented differently within the visual system of avian predators \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. All images implemented in the study were selected from the publicly available St Andrews Hyperspectral Lepidoptera database (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://arts.st-andrews.ac.uk/lepidoptera/documentation.html\u003c/span\u003e\u003cspan address=\"https://arts.st-andrews.ac.uk/lepidoptera/documentation.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). For details on image acquisition, see Penacchio et al. \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e. For details on how these images were displayed within the standard RGB colour space and presented in PsychoPy \u003csup\u003e[\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]\u003c/sup\u003e, see \u003cem\u003eSupplementary Information\u003c/em\u003e. The full database contains images of 125 butterfly and moth species from 12 Lepidoptera families: 96 aposematic (AP) and 29 non-aposematic (non-AP), sampled from British museum collections. For each species, the database provides statistics of in silico neural activity (see Supplementary Figure \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e) that characterise how it is represented in a model avian visual system. These statistics, which have been shown to effectively discriminate AP from non-AP species \u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e, were used here to select 29 AP species predicted to produce the strongest visual responses, and 29 non-AP species predicted to produce the weakest ones. These 58 species served as targets for memorisation in the experiment (shown in \u003cem\u003eSupplementary Figure S2)\u003c/em\u003e, while the remaining species were used as lures (\u003cem\u003eSupplementary Figure S3\u003c/em\u003e). Given the small number of non-AP species in the original database, we used only AP species as lures and always presented non-AP species as targets. Note that an a-priori selection for targets and lures is unlike standard memory paradigms, where targets and lures are typically randomised across participants. Instead, our approach ensured that the two classes of targets were always maximally differentiated in their predicted neural representations, enabling a more direct test of whether these ecologically meaningful differences in visual encoding translate into differences in human memory.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eProcedure \u003c/h2\u003e\u003cp\u003e\u003cb\u003eFigure 5.\u003c/b\u003e Examples of one trial from the metamemory and recognition blocks. (\u003cb\u003eA\u003c/b\u003e) In the metamemory block, participants viewed all target images and rated each one based on how likely they thought they would remember it later. (\u003cb\u003eB\u003c/b\u003e) In the recognition block, participants were shown a mix of previously seen images (\u0026lsquo;targets\u0026rsquo;), novel ones (\u0026lsquo;lures\u0026rsquo;), as well as task-irrelevant vigilance-check images, and had to decide whether each stimulus was \u0026lsquo;old\u0026rsquo; (seen before) or \u0026lsquo;new\u0026rsquo; (not seen before).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eMetamemory rating block\u003c/h2\u003e\u003cp\u003eThe experiment started with a \u0026lsquo;metamemory rating\u0026rsquo; study block, followed by a recognition test block (Fig.\u0026nbsp;5). In the metamemory block (Fig.\u0026nbsp;5A), participants viewed all aposematic and non-aposematic targets and provided metamemory ratings. Participants were instructed to look at each image carefully and provide their subjective impression of how memorable it looked. The metamemory rating was explained to participants in the instructions as follows: \u0026lsquo;\u003cem\u003eA memorable image is one that you feel you would be likely to pick out as having seen before, if you saw it again soon, so you would give it a high score\u0026rsquo;.\u003c/em\u003e At the start of each study trial, a central fixation marker appeared for 1 second, followed by one image that remained on screen for 2 seconds. Subsequently, the image disappeared, and the question \u0026lsquo;\u003cem\u003eHow memorable is this image?\u003c/em\u003e\u0026rsquo; appeared on screen, along with a 10-point clickable rating scale. Participants were instructed to select a score that best described their subjective impression of how memorable the stimulus looked, from 1 (\u0026lsquo;very forgettable\u0026rsquo;) to 10 (\u0026lsquo;very memorable\u0026rsquo;). Each target was studied and rated only once, and the order of presented targets was randomised for each subject. Note that previous work that has compared memory and metamemory for photographs \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e used binary memorable/forgettable ratings for scenes from many different classes (e.g., landscapes, urban, persons). Given that our targets were all of one image-type (i.e., different butterfly/moth species), we chose to implement a broader rating scale to measure finer-grained, within-class differences. At the end of the study block, a break was offered before the test block.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003eRecognition block\u003c/h2\u003e\u003cp\u003eIn the \u0026lsquo;recognition\u0026rsquo; test block (Fig.\u0026nbsp;5B), participants were instructed to look at images and state whether they had previously seen each one. Each test trial started with a 1-second central fixation marker, followed by one test image that remained on screen until the end of the trial (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Unbeknownst to participants, the test image could be either a target (i.e., previously studied image), a lure (i.e., never seen before image) or a vigilance check (i.e., a task-irrelevant picture of a flower). After 1 second, the question \u0026lsquo;\u003cem\u003eHave you seen this image before?\u003c/em\u003e\u0026rsquo; appeared on screen. Participants pressed the \u0026lsquo;Y\u0026rsquo;/\u0026lsquo;N\u0026rsquo; keys to indicate their decision. No feedback was provided. After response, the statement \u003cem\u003e\u0026lsquo;Rate your confidence\u0026rsquo;\u003c/em\u003e appeared on screen, along with a 3-point (low, medium, high) clickable rating scale. Given our image set constraints, the rating could not be analysed using signal detection methods, hence it is not reported here. The old/new response and confidence rating were self-paced. In contrast to the metamemory block, where a constant presentation time was chosen to standardize encoding time, the test image was kept on screen indefinitely to allow participants to interrogate their memory with no time constraints. At the end of the recognition block, participants were fully debriefed.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eVigilance checks\u003c/h2\u003e\u003cp\u003eThe study and test blocks started with practice blocks that contained 5 arbitrary pictures from a database of flowers \u003csup\u003e[\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]\u003c/sup\u003e, unrelated to the experiment. One flower picture that was shown both in the study and test practice blocks was also scheduled to randomly repeat 5 times during the real test block as a vigilance check, among Lepidoptera pictures. If participants classified the flower as \u0026lsquo;new\u0026rsquo; one or more times, it was taken as evidence that they did not sufficiently engage with the task. Therefore, their data was excluded from the analyses (see \u003cem\u003eParticipants\u003c/em\u003e). This method is similar to the \u0026lsquo;vigilance repeats\u0026rsquo; \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e used in continuous recognition memory tasks to screen out inattentive participants.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eStatistical analysis\u003c/h2\u003e\u003cp\u003eFor statistical analysis, the \u003cem\u003eR\u003c/em\u003e environment was used [R Core Team, 2021]. After excluding participants who failed any vigilance checks, a total of 2,850 analysable trials were available for each of the two blocks (metamemory rating and recognition). One AP image was excluded from the analyses because of a technical error in programming the experiment. For each target image (i) we calculated recognition performance aggregated across observers. Each observation was coded as 1 for a hit or 0 for a miss, resulting in N⁽ⁱ⁾ total observations for image i, with H⁽ⁱ⁾ hits. Hit rates were then calculated as the proportion of hits per image (HR⁽ⁱ⁾ = H⁽ⁱ⁾ / N⁽ⁱ⁾), providing a measure of recognition success across the whole subject pool \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Note that early memorability work \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e used the hit rate per image as a measure of recognition performance (as we do here). More recent work, however, recommends subtracting an image\u0026rsquo;s false alarm rate from the hit rate to obtain a more accurate performance estimate \u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. In our image set, lure images were only available for AP species, meaning that false alarm rates could not be calculated for both image classes. For this reason, we used hit rates in all analyses (but see \u003cem\u003eSupplementary Figure S3\u003c/em\u003e for data on lure images). To analyse metamemory ratings and recognition responses, mixed models were fitted to the raw trial-by-trial data using the functions \u003cem\u003elmer\u003c/em\u003e and \u003cem\u003eglmer\u003c/em\u003e (respectively) from the \u003cem\u003elme4\u003c/em\u003e package \u003csup\u003e[\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]\u003c/sup\u003e. Mixed models allow the inclusion of individual intercepts and slopes for each subject. They are useful to account for the contribution of random variability in data due to individual differences in baseline behaviour (intercepts) and how subjects respond to experimental manipulations (slopes). This is beneficial for online studies such as ours, as viewing conditions and user focus levels cannot be strictly controlled. For the selection of mixed models we used the Akaike information criterion (AIC). ANOVA comparisons between models were conducted and those with the lowest AIC were reported in Results. Full model comparisons are included in \u003cem\u003eSupplementary Tables S5 and S6\u003c/em\u003e. To verify model assumptions, the \u003cem\u003echeck_model()\u003c/em\u003e function from the \u003cem\u003eperformance\u003c/em\u003e R package \u003csup\u003e[\u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]\u003c/sup\u003e was used. To estimate marginal and conditional R-squared values from mixed model fits, we used the \u003cem\u003er.squaredGLMM()\u003c/em\u003e function from the \u003cem\u003eMuMIn\u003c/em\u003e R package \u003csup\u003e[\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e]\u003c/sup\u003e. Statistical significance was determined using a two-tailed significance threshold of α\u0026thinsp;=\u0026thinsp;.05. In all correlation-based analyses (metamemory, memory, and consistency analysis), Spearman\u0026rsquo;s rank-order correlation (non-parametric) was used to allow for direct comparisons to results of previous image memorability research, where non-parametric correlations are most frequently reported.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eStatistical testing of consistency measures\u003c/h2\u003e\u003cp\u003eFor the consistency analysis, as recommended by current guidelines on memorability estimation, correlations between random splits of subjects were corrected using the Spearman-Brown correction for split-half reliability \u003csup\u003e[\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e, \u003cspan citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e]\u003c/sup\u003e. This correction adjusts the correlation to estimate the reliability of the full dataset, compensating for the reduced data size in each split (in our case, N/2\u0026thinsp;=\u0026thinsp;25). For the split-half reliability analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA), we assessed statistical significance using a permutation test. We calculated the difference (observed Δρ) between the observed split-half reliabilities for AP and non-AP images. To test whether differences could arise by chance, we randomly permuted the AP vs. non-AP class labels 1000 times, recalculating the reliability difference between AP and non-AP (shuffled Δρ). A p-value was calculated by measuring the rate at which the shuffled Δρ exceeded the observed Δρ.\u003c/p\u003e\u003cp\u003eFor the cumulative hit rate analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eB), we compared hit rates across groups both within classes (for AP and non-AP separately) and between classes (comparing AP and non-AP). Within classes, we calculated two differences: (i) the difference between cumulative hit rates of group j and group k (independent subject splits), and (ii) the difference between group k and its shuffled baseline (chance). Between classes, we compared the extent to which the cumulative hit rate of group k exceeded the shuffled baseline for AP versus non-AP species. Because there was one additional non-AP image, the last rank (n\u0026thinsp;=\u0026thinsp;29) of the non-AP condition was excluded from between-class comparisons. To assess statistical significance, we used bootstrapped confidence intervals and permutation tests. For each image rank number, the difference between cumulative hit rates across groups was computed, and 95% confidence intervals were obtained from 1,000 bootstrap resamples of the differences. This observed mean difference was then compared to a null distribution. To calculate p-values, the sign of each difference was randomly flipped (i.e., sign-flip permutation test), generating a null distribution of differences. The p-value was defined as the proportion of permuted values with equal or greater magnitude than the observed difference. Because multiple tests were conducted (one for each rank), the Bonferroni correction was used to adjust p-values, controlling for the familywise error rate.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Jasna Martinovic, Justin Ales, Michael Oram, and Wilma Bainbridge for helpful discussions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eF.D.F. conceived the study, collected and curated the data, conducted the analyses, and prepared all figures and the original manuscript draft. O.P. provided software and, together with A.R.O., contributed to the study\u0026rsquo;s conceptualisation, methodology, validation, and supervision. J.M.H. contributed to conceptualisation, methodology, funding acquisition, and supervision, and contributed to both the original and revised manuscript text.All authors read and approved the final manuscript.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data are available online from the Open Science Framework: https://osf.io/syf2w/\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdditional information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRights retention\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn order to meet institutional and research funder open access requirements, any accepted manuscript arising shall be open access under a Creative Commons Attribution (CC BY) reuse licence with zero embargo.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting Interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded by the UK Research and Innovation Biotechnology and Biological Sciences Research Council (UKRI-BBSRC) through the East of Scotland Bioscience Doctoral Training Partnership (EASTBIO DTP) as part of F.D.F\u0026rsquo;s PhD studentship.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBrady, T. F., Konkle, T., Alvarez, G. A. \u0026amp; Oliva, A. Visual long-term memory has a massive storage capacity for object details. \u003cem\u003eProc. Natl. Acad. Sci.\u003c/em\u003e 105, 14325\u0026ndash;14329 (2008).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStanding, L. Learning 10000 pictures. \u003cem\u003eQ. J. Exp. Psychol.\u003c/em\u003e \u003cb\u003e25\u003c/b\u003e, 207\u0026ndash;222 (1973).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePoulton, E. B. \u003cem\u003eThe Colours of Animals, Their Meaning and Use, Especially Considered in the Case of Insects\u003c/em\u003e (D. Appleton and Company, 1890).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStevens, M. \u0026amp; Ruxton, G. D. Linking the evolution and form of warning coloration in nature. \u003cem\u003eProc. R. Soc. B Biol. Sci.\u003c/em\u003e 279, 417\u0026ndash;426 (2011).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSkelhorn, J., Halpin, C. G. \u0026amp; Rowe, C. Learning about aposematic prey. \u003cem\u003eBehav. Ecol.\u003c/em\u003e \u003cb\u003e27\u003c/b\u003e, 955\u0026ndash;964 (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuilford, T. \u0026amp; Dawkins, M. S. Receiver psychology and the evolution of animal signals. \u003cem\u003eAnim. Behav.\u003c/em\u003e \u003cb\u003e42\u003c/b\u003e, 1\u0026ndash;14 (1991).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSpeed, M. P. Warning signals, receiver psychology and predator memory. \u003cem\u003eAnim. Behav.\u003c/em\u003e \u003cb\u003e60\u003c/b\u003e, 269\u0026ndash;278 (2000).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShettleworth, S. J. The role of novelty in learned avoidance of unpalatable \u0026lsquo;Prey\u0026rsquo; by domestic chicks (Gallus Gallus). \u003cem\u003eAnim. Behav.\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e, 29\u0026ndash;35 (1972).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRoper, T. J. \u0026amp; Redston, S. Conspicuousness of distasteful prey affects the strength and durability of one-trial avoidance learning. \u003cem\u003eAnim. Behav.\u003c/em\u003e \u003cb\u003e35\u003c/b\u003e, 739\u0026ndash;747 (1987).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGamberale-Stille, G. \u0026amp; Tullberg, B. S. Experienced chicks show biased avoidance of stronger signals: an experiment with natural colour variation in live aposematic prey. \u003cem\u003eEvol. Ecol.\u003c/em\u003e \u003cb\u003e13\u003c/b\u003e, 579\u0026ndash;589 (1999).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAronsson, M. \u0026amp; Gamberale-Stille, G. Domestic chicks primarily attend to colour, not pattern, when learning an aposematic coloration. \u003cem\u003eAnim. Behav.\u003c/em\u003e \u003cb\u003e75\u003c/b\u003e, 417\u0026ndash;423 (2008).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAronsson, M. \u0026amp; Gamberale-Stille, G. Evidence of signaling benefits to contrasting internal color boundaries in warning coloration. \u003cem\u003eBehav. Ecol.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 349\u0026ndash;354 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMedina, I. et al. Global selection on insect antipredator coloration. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e389\u003c/b\u003e, 1336\u0026ndash;1341 (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIsola, P., Xiao, J., Parikh, D., Torralba, A. \u0026amp; Oliva, A. What Makes a Photograph Memorable? \u003cem\u003eIEEE Trans. Pattern Anal. Mach. Intell.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 1469\u0026ndash;1482 (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBainbridge, W. A., Dilks, D. D., Oliva, A. \u0026amp; Memorability A stimulus-driven perceptual neural signature distinctive from memory. \u003cem\u003eNeuroImage\u003c/em\u003e \u003cb\u003e149\u003c/b\u003e, 141\u0026ndash;152 (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRust, N. C. \u0026amp; Mehrpour, V. Understanding Image Memorability. \u003cem\u003eTrends Cogn. Sci.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 557\u0026ndash;568 (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHarris, J. M., Penacchio, O. \u0026amp; Osorio, D. C. The Role of Vision Science in Understanding Animal Camouflage. \u003cem\u003eAnnu. Rev. Vis. Sci.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e, 331\u0026ndash;357 (2025).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCuthill, I. C. \u0026amp; Camouflage. \u003cem\u003eJ. Zool.\u003c/em\u003e \u003cb\u003e308\u003c/b\u003e, 75\u0026ndash;92 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOsorio, D. \u0026amp; Vorobyev, M. A review of the evolution of animal colour vision and visual communication signals. \u003cem\u003eVis. Res.\u003c/em\u003e \u003cb\u003e48\u003c/b\u003e, 2042\u0026ndash;2051 (2008).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePinto, L. \u0026amp; Baron, J. Spatiotemporal frequency and speed tuning in the owl visual wulst. \u003cem\u003eEur. J. Neurosci.\u003c/em\u003e \u003cb\u003e30\u003c/b\u003e, 1251\u0026ndash;1268 (2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eVorobyev, M. \u0026amp; Osorio, D. Receptor noise as a determinant of colour thresholds. \u003cem\u003eProc. R. Soc. Lond. B Biol. Sci.\u003c/em\u003e 265, 351\u0026ndash;358 (1998).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAnderson, A. J. \u0026amp; McOwan, P. W. Humans deceived by predatory stealth strategy camouflaging motion. \u003cem\u003eProc. R. Soc. Lond. B Biol. Sci.\u003c/em\u003e 270, (2003).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eScott-Samuel, N. E., Baddeley, R., Palmer, C. E. \u0026amp; Cuthill, I. C. Dazzle Camouflage Affects Speed Perception. \u003cem\u003ePLoS ONE\u003c/em\u003e. \u003cb\u003e6\u003c/b\u003e, e20233 (2011).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePenacchio, O., Harris, J. M. \u0026amp; Lovell, P. G. Establishing the behavioural limits for countershaded camouflage. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e, 13672 (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePenacchio, O., Lovell, P. G. \u0026amp; Harris, J. M. Is countershading camouflage robust to lighting change due to weather? \u003cem\u003eR Soc. Open. Sci.\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e, 170801 (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWebster, R. J., Hassall, C., Herdman, C. M., Godin, J. G. J. \u0026amp; Sherratt, T. N. Disruptive camouflage impairs object recognition. \u003cem\u003eBiol. Lett.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e, 20130501 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStevens, M. et al. Revealed by conspicuousness: distractive markings reduce camouflage. \u003cem\u003eBehav. Ecol.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 213\u0026ndash;222 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTroscianko, J., Lown, A. E., Hughes, A. E. \u0026amp; Stevens, M. Defeating Crypsis: Detection and Learning of Camouflage Strategies. \u003cem\u003ePLoS ONE\u003c/em\u003e. \u003cb\u003e8\u003c/b\u003e, e73733 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTroscianko, J., Skelhorn, J. \u0026amp; Stevens, M. Quantifying camouflage: how to predict detectability from appearance. \u003cem\u003eBMC Evol. Biol.\u003c/em\u003e \u003cb\u003e17\u003c/b\u003e, 7 (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFlavell, J. H. Metacognition and cognitive monitoring: A new area of cognitive\u0026ndash;developmental inquiry. \u003cem\u003eAm. Psychol.\u003c/em\u003e \u003cb\u003e34\u003c/b\u003e, 906\u0026ndash;911 (1979).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDunlosky, J. \u0026amp; Metcalfe, J. \u003cem\u003eMetacognition. ix, 334\u003c/em\u003e (Sage Publications, Inc, 2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eUndorf, M. \u0026amp; Br\u0026ouml;der, A. Metamemory for pictures of naturalistic scenes: Assessment of accuracy and cue utilization. \u003cem\u003eMem. Cognit\u003c/em\u003e. \u003cb\u003e49\u003c/b\u003e, 1405\u0026ndash;1422 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYonelinas, A. P. The Nature of Recollection and Familiarity: A Review of 30 Years of Research. \u003cem\u003eJ. Mem. Lang.\u003c/em\u003e \u003cb\u003e46\u003c/b\u003e, 441\u0026ndash;517 (2002).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBylinskii, Z., Isola, P., Bainbridge, C., Torralba, A. \u0026amp; Oliva, A. Intrinsic and extrinsic effects on image memorability. \u003cem\u003eVis. Res.\u003c/em\u003e \u003cb\u003e116\u003c/b\u003e, 165\u0026ndash;178 (2015).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBainbridge, W. A. Chapter One - Memorability: How what we see influences what we remember. in Psychology of Learning and Motivation (eds (eds Federmeier, K. D. \u0026amp; Beck, D. M.) vol. 70 1\u0026ndash;27 (Academic, (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJaegle, A. et al. Population response magnitude variation in inferotemporal cortex predicts image memorability. \u003cem\u003eeLife\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, e47596 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKhosla, A., Raju, A. S., Torralba, A. \u0026amp; Oliva, A. Understanding and Predicting Image Memorability at a Large Scale. in \u003cem\u003eIEEE International Conference on Computer Vision (ICCV)\u003c/em\u003e 2390\u0026ndash;2398 (IEEE, Santiago, Chile, 2015)., 2015). (2015). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICCV.2015.275\u003c/span\u003e\u003cspan address=\"10.1109/ICCV.2015.275\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNeedell, C. D. \u0026amp; Bainbridge, W. A. Embracing New Techniques in Deep Learning for Estimating Image Memorability. \u003cem\u003eComput. Brain Behav.\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e, 168\u0026ndash;184 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePenacchio, O. et al. A computational neuroscience framework for quantifying warning signals. \u003cem\u003eMethods Ecol. Evol.\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 103\u0026ndash;116 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBainbridge, W. A. The memorability of people: Intrinsic memorability across transformations of a person\u0026rsquo;s face. \u003cem\u003eJ. Exp. Psychol. Learn. Mem. Cogn.\u003c/em\u003e \u003cb\u003e43\u003c/b\u003e, 706\u0026ndash;716 (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eStevens, M. \u0026amp; Ruxton, G. D. Linking the evolution and form of warning coloration in nature. \u003cem\u003eProc. R. Soc. B Biol. Sci.\u003c/em\u003e 279, 417\u0026ndash;426 (2012).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBorkin, M. A. et al. What Makes a Visualization Memorable? \u003cem\u003eIEEE Trans. Vis. Comput. Graph\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 2306\u0026ndash;2315 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDavis, T. M. \u0026amp; Bainbridge, W. A. Memory for artwork is predictable. \u003cem\u003eProc. Natl. Acad. Sci. U. S. A.\u003c/em\u003e 120, e2302389120 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOngchoco, J. D. K., Chun, M. M. \u0026amp; Bainbridge, W. A. What moves us? The intrinsic memorability of dance. \u003cem\u003eJ. Exp. Psychol. Learn. Mem. Cogn.\u003c/em\u003e \u003cb\u003e49\u003c/b\u003e, 889\u0026ndash;899 (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWakeland-Hart, C. D., Cao, S. A., deBettencourt, M. T., Bainbridge, W. A. \u0026amp; Rosenberg, M. D. Predicting visual memory across images and within individuals. \u003cem\u003eCognition\u003c/em\u003e \u003cb\u003e227\u003c/b\u003e, 105201 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGeiger, A. R. \u0026amp; Balas, B. Robot face memorability is affected by uncanny appearance. \u003cem\u003eComput. Hum. Behav. Rep.\u003c/em\u003e \u003cb\u003e4\u003c/b\u003e, 100153 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePalan, S. \u0026amp; Schitter, C. Prolific.ac\u0026mdash;A subject pool for online experiments. \u003cem\u003eJ. Behav. Exp. Finance\u003c/em\u003e. \u003cb\u003e17\u003c/b\u003e, 22\u0026ndash;27 (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePeirce, J. et al. PsychoPy2: Experiments in behavior made easy. \u003cem\u003eBehav. Res. Methods\u003c/em\u003e. \u003cb\u003e51\u003c/b\u003e, 195\u0026ndash;203 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNilsback, M. E. \u0026amp; Zisserman, A. Automated Flower Classification over a Large Number of Classes. in \u003cem\u003eSixth Indian Conference on Computer Vision, Graphics \u0026amp; Image Processing\u003c/em\u003e 722\u0026ndash;729 (2008). 722\u0026ndash;729 (2008). (2008). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/ICVGIP.2008.47\u003c/span\u003e\u003cspan address=\"10.1109/ICVGIP.2008.47\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBates, D., M\u0026auml;chler, M., Bolker, B. \u0026amp; Walker, S. Fitting Linear Mixed-Effects Models Using lme4. \u003cem\u003eJ. Stat. Softw.\u003c/em\u003e \u003cb\u003e67\u003c/b\u003e, 1\u0026ndash;48 (2015).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eL\u0026uuml;decke, D., Ben-Shachar, M., Patil, I., Waggoner, P. \u0026amp; Makowski, D. performance: An R Package for Assessment, Comparison and Testing of Statistical Models. \u003cem\u003eJ. Open. Source Softw.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, 3139 (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBartoń, K. \u0026amp; MuMIn Multi-Model Inference. 1.48.11 (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrown, W. Some experimental results in the correlation of mental abilities. \u003cem\u003eBr. J. Psychol.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 296\u0026ndash;322 (1910).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSpearman, C. Correlation calculated from faulty data. \u003cem\u003eBr. J. Psychol.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 271\u0026ndash;295 (1910).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"aposematism, warning signal, metamemory, memorability, receiver psychology, vision","lastPublishedDoi":"10.21203/rs.3.rs-8175452/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8175452/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAlthough the bold warning patterns of prey (known as aposematic) have been shown to facilitate predator learning through repeated encounters, it remains unclear to what extent these patterns support memory retention. Here, we tested whether aposematic patterns appear more memorable to human observers, and whether they have an intrinsic advantage in being recognised from memory\u0026mdash;even after a single exposure. Observers viewed images of aposematic and non-aposematic butterflies and moths, judged how likely they were to remember each one (metamemory rating), and later completed a test distinguishing previously seen species from novel ones (recognition memory). While aposematic species elicited higher metamemory ratings upon first sight, we found no evidence that they were more likely to be recognised when seen again. Despite this apparent metacognitive failure, for aposematic species the observers tend to remember and forget the same images as one another. This suggests that these images exhibit \u0026lsquo;memorability\u0026rsquo;, an intrinsic property of an image that allows one to predict how well images can be remembered. These findings raise the possibility that an effective visual warning may hinge less on recognition of a previously seen signal, and more on perceptual processes at play when it is first encountered.\u003c/p\u003e","manuscriptTitle":"Natural warning signals unexpectedly shape human metamemory ratings but not image recognition success","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-01 08:01:47","doi":"10.21203/rs.3.rs-8175452/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-25T12:19:21+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-11-24T01:25:58+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-11-24T01:25:04+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-11-21T16:33:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c28ed7fd-1449-4bb8-915c-4b22ab3d6c1b","owner":[],"postedDate":"December 1st, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":58580926,"name":"Biological sciences/Ecology"},{"id":58580927,"name":"Earth and environmental sciences/Ecology"},{"id":58580928,"name":"Biological sciences/Neuroscience"},{"id":58580929,"name":"Biological sciences/Psychology"},{"id":58580930,"name":"Social science/Psychology"},{"id":58580931,"name":"Biological sciences/Zoology"}],"tags":[],"updatedAt":"2026-03-02T16:03:50+00:00","versionOfRecord":{"articleIdentity":"rs-8175452","link":"https://doi.org/10.1038/s41598-026-41178-y","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2026-02-25 15:58:13","publishedOnDateReadable":"February 25th, 2026"},"versionCreatedAt":"2025-12-01 08:01:47","video":"","vorDoi":"10.1038/s41598-026-41178-y","vorDoiUrl":"https://doi.org/10.1038/s41598-026-41178-y","workflowStages":[]},"version":"v1","identity":"rs-8175452","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8175452","identity":"rs-8175452","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}