Impacts of predation risk on learning and memory of free-living mice

Abstract

In predator–prey interactions, responses to predation risk typically involve behavioural, morphological or physiological changes. Laboratory-based studies have also shown changes in prey cognition (i.e. learning and memory), with individuals often showing impairment. However, an ecological perspective predicts that wild animals should conserve their cognitive ability, given that many risk responses require robust cognition. Here, we simulated predation risk and used a field-adapted version of the Morris Water Maze (MWM) to investigate how chronic predation risk affects cognition in wild white-footed mice (Peromyscus leucopus). We found that 24 days' exposure to predation risk did not impair learning. However, those exposed to risk had a 25% reduction of their short-term memory. Twelve days post-risk exposure, we found no performance differences between risk-exposed and control mice. Additionally, risk-exposed mice displayed greater exploration with a higher probability of completing the MWM in their initial trial. Given that prey integrate multiple pieces of information to shape their behaviour, the lack of learning impairment and altered exploration strategies may help mice respond to predation risk. However, the tendency of memory impairment suggests there are consequences for cognition when experiencing increased predation risk.

1. Introduction

Predators can influence prey in two non-exclusive ways: through consumptive effects—where the predator influences the density of prey by direct consumption—and through non-consumptive, predation risk effects—where the mere presence of the predator can affect prey through fear [1,2]. Prey responses to the risk of predation can be morphological [3], physiological [4] and behavioural [5], which may include changes in habitat use [6,7], vigilance [8,9], foraging [10,11] and movement patterns [12,13]. For example, Bond and colleagues showed that predator presence induced habitat shifts in southern stingrays, which tend to spend more time in safer flat habitats when sharks are present [14]. Kelleher and colleagues demonstrated a reduction in foraging behaviour in wild white-footed mice when exposed to predation risk, but the magnitude of response varied depending upon an interaction among habitat type, refuge availability and predator hunting mode [15]. A context-dependent risk response in wild prey has been shown numerous times [16–18] and suggests that prey are able to integrate multiple pieces of information from their environment to inform their response [7,19,20]. While much is known about how predation risk may alter the behaviour and physiology of wild animals, far less is known about how predation risk may alter their cognitive ability (i.e. learning and memory), which is a critical aspect if they are to integrate multiple pieces of information.

The ability to learn and memorize is a key aspect for survival in nature, allowing animals to perform fundamental tasks such as locating food and water [21–23], mating [24,25] and avoiding predators [26,27]. Learning is broadly defined as the acquisition of new information, while memory is the retention of the newly learned information [28,29]. Learning can pertain to spatial configuration (i.e. spatial learning), predictive links between cues and reward or punishment (i.e. associative learning) or about a sequence of actions (i.e. motor learning) [29]. Memory can be distinguished depending on the duration of the information stored. Short-term memory is the retention of information in an accessible temporary state (e.g. minutes to hours), while long-term memory involves the storage of information for an extended period of time (e.g. days or months) [30]. The adaptive benefits of specific types of learning and memory will depend on the life history and system in which the species exists. For example, an appropriate response to predation risk usually comes from associative learning and memorization of potential predators [31]. Failure to recognize or appropriately respond to a potential predator can be fatal, while unnecessary or exacerbated anti-predatory responses may result in expensive energy costs and reduced fitness [5,31,32]. Thus, it has been suggested that wild animals should have evolved the ability to maintain their cognitive performance given its impact on fitness, even when encountering an ecological or environmental stressor in the wild (such as predation risk [33,34]). For example, Heathcote and colleagues investigated the benefits of spatial knowledge in pheasants and found that those with better spatial memory were less likely to be killed by a predator [35]. However, laboratory-based studies have shown that predation risk exposure frequently reduces cognitive performance. For example, Diamond and colleagues showed that rats exposed to predation risk had reduced learning and memory, although this impairment was lost in easier tasks [36]. Park and colleagues demonstrated that adult rats exposed to predators showed impairment not only in learning ability but also in memory consolidation and retrieval [37]. Thus, while laboratory-based evidence shows a risk-induced reduction in cognitive performance, an evolutionary perspective suggests that individuals should maintain their cognitive performance when exposed to predators as this should increase survival. Yet, we have little evidence testing the impact of predation risk on cognitive performance in wild mammals.

Testing the impact of predation risk on learning and memory in wild animals can be challenging. The lack of control of potential cofounding variables is usually the major downside in investigating animals’ cognition in the wild [38,39]. While there are many robust laboratory techniques (e.g. [36,40,41]), far fewer have been employed for use with wild animals, particularly those with relevance to their natural world. However, this scenario has been changing over the past decade [29,38]. For example, Henk-von der Malsburg and colleagues used a battery of behavioural and cognitive tests to investigate the cognitive abilities of two wild mouse lemur species with distinct ecological adaptations (generalist and specialist) [42]. Rochais and colleagues studied cognitive performance in free-living African striped mice, also using a number of cognitive tests that included attention, problem-solving and learning tasks [43]. Despite being challenging, there is an increasing focus on taking laboratory-based experimental studies into the field to test cognition in a natural habitat (e.g. [39,44]). Among the tests and techniques applied to studies in wild animals, GPS collars and tags (e.g. [45,46]) and maze tests (e.g. [42,47]) stand out for their adaptability to the natural environment and the correlation between cognitive performance and fitness components in the wild [29,48,49].

The Morris Water Maze (MWM) is one of the most prevalent tests used to investigate learning and memory in rodents [50–52]. In its most basic form, the maze consists of a small pool filled with opaque water and a platform that is hidden right below the water surface, which can be used to escape from the water but is not visible while swimming [50,52]. Using beacon cues (an indicator positioned near to the platform, which enables the subject to reach the goal by aiming at the beacon [53]) or landmark cues (multiple cues equally distant from the goal that serve as spatial points of reference [54]), subjects have multiple subsequent attempts within a set amount of time to find the platform and escape from the water [52,55]. Learning is assessed as a decrease in time or distance to find the platform between attempts, while memory is displayed by maintaining time or distance to find the platform after a prolonged delay between attempts (which can range from hours to days). Here, we used an adapted version of the MWM specifically validated for use in the field to test the influence of predation risk on learning and memory in a free-living mammal [49].

We tested the hypothesis that chronic predation risk causes an impairment in learning and memory in free-living white-footed mice (Peromyscus leucopus). We simulated predation risk for 24 days using a standard playback experiment [56–58]. Free-living mice were live-trapped and their cognitive behaviour was measured using the field-adapted version of the MWM [49]. We predicted that mice exposed to chronic predation risk, as compared to control mice, would have (i) reduced learning and (ii) reduced memory. Furthermore, we also predicted that (iii) mice exposed to chronic predation risk would show a greater impairment in long-term memory when compared with control mice. This study may provide important insights into the effects of predation risk on cognitive functions, i.e. learning and memory, of free-living mammals.

2. Methods

(a). Study area

The study was conducted in the Southeastern Massachusetts Bioreserve (41°’4'21”9"N 71°’1'11”2"W), one of the largest unfragmented forests in eastern Massachusetts. The area covers approximately 2225 hectares and is mainly composed of white pine (Pinus strobus), mixed oaks (Quercus spp.), pitch pine (Pinus rigida), red maple (Acer rubrum) and black birch (Betula lenta). Understory vegetation consists of low bush blueberry (Vaccinium angustifolium), bracken fern (Pteridium aquilinum), black huckleberry (Gaylussacia baccata) and greenbrier (Smilax spp.). Grid selection (n = 4) was made based on the similarity between areas, where factors such as vegetation diversity and complexity (e.g. fallen trees and tree density) were accounted for. After initial selection of the grids, the appointment of treatment for each grid was random.

(b). Trapping

We set up four 150 m × 150 m grids, with a minimum distance of 500 m between grids to eliminate noise transfer and rodent movement [59,60]. Within each grid, we deployed 50 traps, with 15 m between traps. In addition, grids were a minimum of 200 m from any major cutline through the forest. We live-trapped white-footed mice (P. leucopus) from 27 June to 9 September 2020. We employed standard capture techniques [61] using Longworth small mammal traps that were pre-baited with millet for at least two days before trapping. At dusk, we set the traps with millet, carrot and cotton and checked each trap at dawn (8−10 h later). Trapping events occurred three times: one before the start of the playback experiment, a second time immediately after the end of the playback experiment and a third approximately two weeks after the end of the playback experiment. At each capture, mice were moved to a common test area approximately 100 m from the grid (different test areas were set up for each grid). At the testing site, mice were weighed, had their sex and reproductive status determined and each new mouse was tagged with a unique numerical identifier (National Band and Tag Co., Newport, USA; only at first capture). For trapping events two and three, in addition to the procedure previously described, recaptured individuals were identified using the unique numerical identifier. Mice then had their learning and memory tested using the field-adapted version of the MWM [49].

All methods in this study were approved by our University’s IACUC (IACUC 22-02) and followed guidelines of the American Mammalogy Society [62].

(c). Predation risk manipulation

We simulated chronic predation risk on two grids via auditory playbacks (using Kicker KB6000 Speaker System, 6.5” woofer and 2 × 5” tweeter, specifically designed for this experiment) of local predators. Following established protocols, calls were broadcasted in a 4 days-on, 4 days-off schedule in order to avoid habituation to the sound [63]. During the day, calls were broadcast for 10% of the time and consisted of the diurnal predators red-tailed hawk (Buteo jamaicensis) and barred owl (Strix varia). During the night, calls were broadcast for 40% of the time and consisted of barred owl, great horned owl (Bubo virginianus) and eastern screech owl (Megascops asio). On each grid, calls were broadcast at approximately 80 dB (at the source of sound) using two speakers positioned at opposite corners; speakers were moved to adjacent corners between the days-on schedule [64]. The predation risk treatment was designed to mimic noise exposure that prey would experience from a natural predator. Control grids (n = 2) consisted of grids with no playback during the entire experiment. Two additional control grids were sampled at the onset of the experiment, but owing to time and resource constraints, we could not sample these during further trapping occasions. Similar or identical playback set-ups have been used in previous studies and have shown that mice not only detected the sound as a risk but also had altered their behaviour [15,65] and physiology [66] in response to it. To ensure that sound itself was not the source of disturbance, we previously tested for differences in faecal glucocorticoid metabolites (FGM; a proxy for stress levels in wild animals) [67] in mice exposed to songbird/amphibian sounds (i.e. positive control) and no sound (i.e. negative control). We found no difference in FGM levels between mice exposed to no sound and songbird/amphibian playbacks (electronic supplementary material, table S1; figure S1), indicating that sound itself was not a source of disturbance. The main reason why we opted to use no sound controls rather than sound controls is because studies show that the use of non-predator calls can actually act as safety cues, instead of as an uninformative cue (i.e. no sounds) [68]. Lastly, although we did not measure the effects of predator playbacks on predators themselves, we did notice (L McKay and CC Ganci, personal observation, 2020) that the use of predator playbacks seemed to attract the presence of real predators to those grids, further increasing predation risk on ‘predation risk’ grids.

We performed the MWM on three different experiment days on all grids (see electronic supplementary material, figure S2 ). The first experiment day happened before any audio playbacks started (i.e. day 0, pre-risk treatment day). The second experiment day occurred on the last day of the audio playbacks (i.e. day 24, during-risk treatment day). The third experiment day occurred approximately two weeks after the end of the audio playbacks (i.e. day 36, post-risk treatment day).

(d). Cognitive testing

We used a field-adapted version of the MWM to test cognitive ability in the field with free-living mice [49]. The maze consisted of a small pool (1.5 m diameter; ≈ 4.7 m circumference; walls 0.5 m high) filled with approximately 15 cm of water made opaque with powdered milk. A small (10 cm diameter) platform was hidden just below the water level, so that the mice could climb on it to avoid the water but could not see it while swimming. A 15 cm high wire with a small flag on the top was attached to the platform to serve as a beacon cue. A live recording GoPro camera (GoPro Hero7) [69] was positioned at the top of the pool so that the observer could follow the experiment through without any interference.

For each test, mice were subjected to 6 consecutive trials, with 15 s separating the first 5 trials and 2 h between the 5th and 6th trials. Each test lasted a maximum of 60 s, where mice had to learn to associate the flag with the location of the platform. Once the mouse located the platform, we let it rest for 15 s to solidify the idea of the goal. In cases where mice did not find the platform within 60 s, we gently guided them to the platform by hand and also let them rest for 15 s. Between each trial, both the platform and the mouse entry point were moved into preset locations that had been randomly chosen before the start of the MWM test, but these locations were kept consistent among mice; i.e. every mouse experienced the same order of both mouse entry point and platform position. After the 5th trial, we let the mice rest in their trap with fresh cotton, apple and millet for 2 h, until the final (6th) trial. Learning was estimated by comparing the distance travelled between 1st and 5th trial, while short-term memory was estimated by comparing the distance travelled between 5th and 6th trial. Long-term memory was assessed by comparing maze performance between mice that had been captured multiple times (~20 days between each capture event).

All trials were recorded using a video camera (GoPro Hero7) suspended above the pool (720 p @ 24fps). The MWM test is a well-established protocol to test learning and memory and has been validated for use with wild mice. We made all attempts to reduce unnecessary impacts, selecting only healthy non-pregnant adults, removing individuals immediately if problems arose, providing ad libitum food during rest periods and drying animals prior to release. Tests were only conducted on warm days (>22°C). At the end of the experiment, we released the mice at the capture site.

(e). Video analysis

For the video analysis, we used a tracking software developed by Hunninck L. in Python (v. 3.7.4) [70], using the PyCharm user interface (v. 2020.3) [71] and used a computer vision approach to track mouse movement in recorded videos [49]. The main Python library used for this software was Open CV (v. 4.4.0.44; [72]). Other libraries used included pandas (v. 1.1.3) [73] and numpy (v. 1.19.2) [74]. The software is based on a tracking algorithm known as Discriminatory Correlation Filter with Channel and Spatial Reliability, or CRST for short (see [75]). The complete code is available in electronic supplementary information; the packaged program for Mac OS is available on GitHub [76]. This tracking software requires a manual specification of the pool wall location (by clicking on three different locations on the wall), the position of the flag and the initial mouse location before tracking can start. After the tracking program was started, the x–y location of the mouse was determined for each frame of the video (frame rate = 24 fps).

The distance covered was calculated as the total distance swum from where the mouse was initially placed into the water until it stood on the platform. To reduce the error in the distance estimates, distances were calculated every 0.5 s; that is, the distance between the mouse location at frame n and the mouse location at frame n + 12 (24 fps). The sum of these distances resulted in a total distance travelled.

(f). Statistical analysis

We constructed two generalized linear mixed effects models (glmmTMB v.1.1.5) [77] to study the effects of predation risk on learning and memory in white-footed mice. The first model measured the performance of mice through distance travelled until the trial was completed [49]. The second model assessed the variation in a mouse’s probability of successfully completing a trial (i.e. reaching the platform within 60 s).

In model 1, distance travelled (in metres; n = 1057) was log-transformed to comply with the assumption of normally distributed model residuals. Trial (i.e. trials 1–6), treatment (i.e. control and predation risk), experiment day (i.e. pre-treatment, during-treatment and post-treatment) and their three-way interaction terms were included in the model. To investigate the potential for long-term memory in mice, we included the number of times an individual mouse was captured (i.e. capture event; range: 1−3) as a fixed effect. Lastly, we added sex of the mouse (i.e. female or male) and trial outcome (i.e. success or fail) as fixed effects to control for differences between sex and trial outcome, respectively. In model 2, success rate (binomial variable) was analysed in relation to trial, treatment and experiment day, while including their three-way interactions. Similar to model 1, capture event and sex of the mouse were also included as fixed explanatory variables.

To assess baseline performance, we tested whether test performance on day 0 (i.e. distance travelled within a trial and success rate of trials) differed between grids assigned to either control or treatment before the treatment was active.To test whether there was a difference between mice assigned to predator and control grids in both learning ability and short-term memory, we also performed an ANOVA test on similar models as described above, but only using data from day 0.

In both models, a measure of relative initial individual performance was included as a fixed effect to account for differences in individual performance; i.e. some individuals might have an overall better or worse performance than most individuals. We accounted for this variation by calculating a relative measure of initial performance. Overall mean performance in trial 1 was calculated for those individuals that were first captured without having experienced the treatment (all control mice and mice in predator grids on day 0). Then we calculated the difference between this overall mean performance and each individual’s performance in trial 1 at first capture to obtain relative individual performance. Relative initial individual performance where first capture was during the treatment (not day 0) was set at 0, or average performance. In this way, the initial performance of 26 individuals was set to 0; 42% of these individuals were in the control group, while 58% were in the predator treatment. This minor discrepancy should not affect our analysis as we further include individual ID as a random effect.

In both models, date was included as a random effect to control for potential variation in weather and environment, while individual mouse ID nested within grid ID were included as random effects to account for further individual differences between mice, and variation owing to study site, respectively. Model residuals and assumptions were checked with DHARMa (v.0.4.5) [78]. We used emmeans (v.1.7.2) [79] to extract and back-transform model estimates and confidence intervals and to conduct posthoc tests. R-squared measures were extracted with MuMin (v.1.42.17) [80]. All tests with a p-value below 0.05 were considered statistically significant. We also report all p-values below 0.1 because these differences may be biologically significant [81]. We completed all analyses in R (v. 4.4.0) [82].

3. Results

Across the study, we captured and tagged 317 unique mice, with multiple captured across trapping events, resulting in 570 total mice trapped. For cognitive testing, lactating or pregnant females (n = 99 captures) and juveniles (n = 151 captures), as well as those deemed in poor condition (n = 60) were not tested, resulting in 157 unique mice (n = 260 captures) eligible for testing. Additionally, several mice were unable to complete a trial, some videos were corrupted or stopped early and the tracking algorithm was unable to follow mice movements for some videos. This resulted in unequal sample sizes between trials, where e.g. more individuals were included in the analysis in trials 5 and 6 compared with 1 (electronic supplementary material, table S2). The final sample size—including all trials—consisted of 1057 videos, from 198 captures of 128 unique mice (electronic supplementary material, table S2). Note that while there may be unequal sample sizes for trials, only mice that completed trials 1 and 5 were measured for learning, while mice that completed trials 1, 5 and 6 could be used for memory testing. Furthermore, trials 2–4 were not included (but the data are presented in figure 1). We found no difference in the likelihood of recapture between the control and predator treatment grids (χ² = 1.85, df = 2, p = 0.397).

Figure 1.

Performance of mice in the field-adapted version of the Morris Water Maze. (A) Average distance travelled (in metres) to find the platform of control (blue) and predation risk (red) treatment mice on day 0 (pre-experiment). (B) Average distance travelled (in metres) to find the platform of control (blue) and predation risk (red) treatment mice on day 24 (during-experiment). (C) Average distance travelled (in metres) to find the platform of control (blue) and predation risk (red) treatment mice on day 36 (post-experiment). Bars represent the 95% confidence interval. Bold bars show the main trials (trials 1, 5 and 6) used to estimate learning and memory. First, second or third individual capture events are represented by circles, triangles and crosses, respectively.

(a). Learning and short-term memory—day 0

Across treatments (on day 0, i.e. before the experiment), mice travelled 51% less in trial 5 compared with trial 1 (t = 3.98, df = 390, p < 0.001; figure 1A), indicating learning. In addition, mice travelled 22% more in trial 6 compared to trial 5 (t = −1,83, df = 390, p = 0.163) and 37% less when compared with trial 1 (t = 2.65, df = 390, p < 0.023; day 0, pre-experiment; figure 1A), indicating partial retention of their learned ability after 2 h (i.e. short-term memory). As expected, mice increased ther success rate in finding the platform from 87% success in trial 1 to 99% success in trial 6 (t = −5.02, p < 0.001; day 0, pre-experiment; figure 2A).

Figure 2.

Probability of mice finding the platform in the field-adapted version of the Morris Water Maze. (A) Probability of success in finding the platform on day 0 (i.e. pre-experiment) for control (blue) and predation risk (red) treatment mice. (B) Probability of success in finding the platform on day 24 (i.e. during-experiment) for control (blue) and predation risk (red) treatment mice. (C) Probability of success in finding the platform on day 36 (i.e. post-experiment) for control (blue) and predation risk (red) treatment mice. Bars represent the 95% confidence intervals. Bold bars show the main trials (trials 1, 5 and 6) used to estimate learning and memory.

There was no difference in distance travelled between mice assigned to the control or predator treatment in trial 1 (t = 0.95, df = 149, p = 0.345), trial 5 (t = 0.68, df = 149, p = 0.499) or trial 6 (t = −0.62, df = 149, p = 0.536). Similarly, there was no difference in success rate between mice assigned to the control or predator treatment in trial 1 (z = 0.13, p = 0.901), trial 5 (z = −1.19, p = 0.235) or trial 6 (z = −0.30, p = 0.765).

We also found that the treatment that mice underwent (i.e., grid) from did not affect either their learning—calculated as difference between performances in trial 5 and trial 1 (F_1,56 = 0.88, p = 0.351, on day 0), or their short-term memory—calculated as difference between performances in trial 5 and trial 6 (F_1,62 = 0.27, p = 0.603, on day 0). Similarly, there was no difference in success rates between mice assigned to the control or predator treatment (z = −0.654, p = 0.513, on day 0).

(b). Predation risk effects—day 24

Predation risk did not significantly affect mice’s ability to learn or inhibit their ability to memorize (on day 24, table 1; figure 1B). At the end of the five trials (i.e. learning), both control and predation risk treatment mice showed a similar decrease in distance (56% and 45%, respectively) to reach the platform when compared with trial 1 (considering data of day 24 only, when the treatment was active; figure 1B). At the end of trial 6 (short-term memory), however, mice in the control treatment tended to show a better short-term memory, retaining 72% of their learned ability—calculated as the difference in performance (i.e. distance travelled) between trial 1 and trial 6 divided by the difference in performance between trial 1 and trial 5 (trial 1 vs trial 6, control, day 24; t = 1.94, df = 390, p = 0.128), compared with mice in the predation risk treatment, which retained only 22% of their learned ability (trial 1 vs trial 6, predator, day 24; t = 0.55, df = 390, p = 0.846). In other words, mice in the control treatment travelled much shorter distances to find the platform in trial 6 when compared with mice in the predator treatments (on day 24, during-risk treatment).

Table 1.

ANOVA (type III) results from the model explaining the variation in distance travelled and the variation in the success rate of trials. Estimates in bold indicate a significant relation on an alpha level <0.05.

variable	distance travelled			success rate of trial
	χ2	df	p	χ2	df	p
trial	53.83	2	<0.001	25.58	2	<0.001
treatment	00.25	1	0.614	02.35	1	0.126
day of capture	17.48	2	<0.001	01.00	2	0.606
cap. event	09.55	2	0.008	11.89	2	0.003
sex	04.13	1	0.042	00.04	1	0.843
initial performance	14.36	1	<0.001	15.11	1	<0.001
trial : treatment	04.89	2	0.087	00.15	2	0.930
trial : day	01.22	4	0.875	05.21	4	0.266
treatment : day	01.21	2	0.546	00.28	2	0.871
trial : treatment : day	00.56	4	0.967	07.19	4	0.126

Predation risk did affect the probability of success in finding the platform (table 1; figure 2B). Mice in the predation risk treatment had an initial success rate (i.e. trial 1) 40% higher than mice under control treatment (on day 24, during-risk experiment; figure 2B); while this might not be an indicator of learning or memory, it may indicate differences in exploratory behaviour.

(c). Long-term memory—days 24 and 36

Recaptured mice performed much better in trial 1 than naive mice, regardless of treatment (figures 1C and 3A). On average, mice that were recaptured once (t = 3.03, df = 390, p = 0.007) or twice (t = 1.936, df = 390, p = 0.130) travelled a significantly shorter distance than naive mice, tested on their first capture (table 1). There was no difference in performance between mice that were recaptured once or twice (t = −0.264, df = 390, p = 0.962).

Figure 3.

Mice performance results in the field-adapted version of Morris Water Maze according to capture event, sex and initial performance. Upper panels (A, B, C) show the average distance travelled (in metres) to find the platform, and lower panels (D, E, F) show the probability of success in finding the hidden platform within 60 s. Left panels (A, D) show the effect for mice captured once, twice or three times during the entire experiment. Centre panels (B, D) show the effect for females and males. Right panels (C, F) show the effect of initial performance. Bars and grey areas show 95% CI around the model estimate.

Similarly, success rates of mice recaptured once (i.e. capture event 2; z = −2.922, p = 0.010) or twice (i.e. capture event 3; z = 2.78, p = 0.015; figure 3D) were 12 and 14% higher, respectively, compared with naive mice (i.e. capture event 1) in trial 1. The success rate increased from an average of 81% (SE = 0.04%) at first capture, to 98% (SE = 0.02%) in the third capture event.

Additionally, day of experiment had a significant effect on distance travelled, but not on success rate (table 1). The overall performance—across all trials and both treatments—of mice was better on day 36 as compared with day 24 (t = 4.06, df = 390, p < 0.001) but performance on day 24 was not significantly better than on day 0 (t = −1.21, df = 390, p = 0.450). There were no significant interactions between day and trial or day and treatment (table 1).

(d). Sex

Overall, female mice (n = 53) travelled 16% less than males (n = 75; table 1; figure 3B). However, there was no difference in success rate between males and females (table 1; figure 3E).

(e). Initial performance

Individual relative initial performance had a clear negative effect on distance travelled across tests, with individuals that had a good initial performance (initial performance score = + 1) travelling on average 20% less than individuals with a poor initial performance (initial performance score = −1; figure 3C). Initial performance also had a clear positive effect on success rate (table 1; figure 3F). These tests were done across all individuals.

4. Discussion

We tested whether chronic predation risk would affect the cognitive ability of free-living white-footed mice. We found that, regardless of treatment, mice had the ability to learn and they retained memory of their learned performance, over both short-term (hours) and long-term (weeks) periods (days 24 and 36). However, those in the predation risk treatment lost 3.3 times more of their recently learned ability compared with control mice, indicating risk-induced reduction in short-term memory (on day 24, during-risk experiment). We also found that mice exposed to predation risk had a greater probability of success in their initial trial of the MWM (on day 24, during-risk experiment) as compared with control mice, indicating possible differences in exploratory and search strategies. Below we discuss the potential that learning and memory are critical for prey exposed to predation risk and, as such, they have evolved mechanisms to reduce risk-induced impairments and maintain their cognitive abilities.

Contrary to our predictions and to many laboratory studies (e.g. [33,34,65]), we found that free-living mice exposed to predation risk did not have impaired learning (days 24 and 36). Unlike laboratory experiments, where animals are generally restricted in space and time, free-living animals can respond to predation risk in a multitude of ways [83] that may reduce exposure to and perception of predation risk [84]. For example, studies show that ungulates under predation risk tend to change their movement and space use, reducing visitation and foraging in places with high predation risk [12,85]. Kelleher and colleagues showed similar results for white-footed mice, with individuals reducing foraging time and increasing refuge use when exposed to predation risk [15]. Thus, mice in our experiment may not have perceived as high a level of predation risk as those in laboratory studies, given their ability to respond behaviourally, reducing potential risk-induced outcomes. Furthermore, wild animals may conserve the ability to learn, requiring a higher threshold of exposure to a stressor in order to develop stressor-induced consequences in learning [43,86]. The lack of impairment in learning is consistent with the ecological and evolutionary prediction that wild animals respond to chronic stressors in an adaptive manner that promotes increased fitness [33].

Our findings also suggest a partial reduction in short-term memory (on day 24, during-risk experiment). These results are consistent with several other studies that show declines in short-term memory in response to exposure to stressors for laboratory [37,87–89] and wild [90,91] animals. A reduction in memory could be a negative pathology but it could also be an adaptive response. As a pathological outcome, chronic exposure to a stressor (i.e. predation risk) could lead to a short-term memory impairment owing to elevated glucocorticoid hormone levels, which may alter biochemical processes and cause accelerated cell loss in the hippocampus—one of the main areas responsible for memory formation and retrieval [92–97]. Alternatively, a reduction in memory may be adaptive, i.e. forgetting can be potentially beneficial depending on the biological importance of the event in relation to what is occurring in your immediate environment [98,99]. Similar to time- or energy-budget trade-offs, mice in the predation-risk treatment may be allocating more of their memory capacity to information relevant to predation risk and less to irrelevant information like escaping the MWM, i.e. it is more important to save information related to predation risk than to the MWM.

Unlike short-term memory differences, we found no difference in long-term memory between predation risk-exposed and control mice. These results also differ from most findings of laboratory-based studies [94,100–102]. For example, Elhage and colleagues demonstrated that mice exposed to unavoidable predatory stimuli had long-term memory impairments that persisted for extended periods (up to 28 days) after the predator stressor exposure [103]. Long-term memory may have been unaffected in our study given its critical importance for survival in the wild [27,104]. From simple tasks such as locating food and water to more complex tasks like spatial navigation, long-term memory likely plays a fundamental role in wild animals in supporting behaviours essential for survival [105,106]. For example, Heathcote and colleagues tested how spatial navigation abilities affect the survival rates in wild pheasants and found that individuals with greater spatial reference developed larger home ranges that led to a decrease in mortality rate [35]. Alternatively, it is also possible that our results were biased, as we were unable to test long-term memory in all individuals in our population. Given that long-term memory measurements required recapture after a 2-week period, it is possible that those with poor long-term memory may have died while those with good long-term memory lived, and we were able to re-test only those latter individuals. This seems to be at least part of the explanation, since mice performed significantly better on day 36 compared to day 24 (regardless of the treatment). Further examination of the links between predation risk, long-term memory and survival is clearly required.

We also found that mice exposed to predation risk had a greater probability of success in the first trial of the MWM test (on day 24, during risk experiment). This may be owing to a change in exploratory and search strategies in mice exposed to predation risk. Previous studies have shown that mice chronically exposed to stressors shifted from using landmark cues (i.e. multiple cues that serve as spatial reference points towards the goal) to beacon cues (i.e. those directly associated with the goal) in spatial learning tasks (e.g. [107–109]). Our modified MWM test uses a beacon cue (i.e. flag) and mice have limited access to clear landmark cues, which may provide the initial advantage to risk-exposed mice. This altered search strategy was first proposed by Easterbrook, who suggested that a reduction in the range of cues used by an organism is associated with an improvement in central performance when under stress [110]. In parallel, studies also suggest that cognitive flexibility is reduced in individuals under stress [111,112]. In other words, the amount of information that is processed and registered by the brain is reduced when under stress. Therefore, shifts in search strategies from landmark cues to beacon cues are believed to be an adaptative response because it is easier to process and enable faster responses—which are two crucial factors for surviving under a predation risk situation [109].

Lastly, we found that sex and initial performance (day 0, pre-experiment) also had an effect on the overall performance in the MWM test. Male mice travelled longer distances than female mice, but both were equally successful in finding the platform (and there was no risk effect). This is unlikely to be related to learning or memory but to the naturally occurring exploratory and search patterns of males vs females. Sex differences in exploratory behaviour have been found across a wide variety of species, with males most often having a greater exploratory behaviour relative to females [113–116]. White-footed mice are a polygynous species (one male mates with more than one female) and a greater exploratory behaviour in males would be advantageous in the natural world, where it likely evolved as a component of this mating system [116]. Additionally, individuals with good initial performance travelled shorter distances and had greater probabilities of success when compared with individuals with bad initial performance. It is somewhat expected that different individuals have different performances, with some individuals being naturally better or worse than others [117]. Several factors, such as body condition [118], social environment [119] and personality traits [118,120], are considerable causes of individual differences in cognitive tests. Even though the explicit mechanisms underlying cognitive individual differences are not well understood, acknowledging this variation provides some basis for future approaches to better understand cognition.

One caveat to discuss is the possibility that our predation risk manipulation was not adequate to simulate the threat of being eaten and consequently it was not appropriately perceived by mice, resulting in the lack of impairment that we found. While, in this study, we did not measure additional metrics of response, we have used this same design in other studies and found significant responses by mice. Kelleher et al. [15] exposed free-living white-footed mice to barred owl playbacks (one of the predator-playback sounds in the current study) for 2 consecutive days and found a 31% reduction in food intake compared with controls [17]. Giordano et al. [65] used a 4-day-on playback manipulation with the same avian predators as the current study and found that risk-exposed mice significantly altered their foraging behaviour and significantly reduced their food intake [68]. Thus, we believe that our manipulation was sufficiently robust to be perceived as a threat by mice. However, we additionally suggest that future work should include at least one metric (e.g. quantifying a behavioural or physiological response) that intentionally tests the efficacy of the risk simulation, while studying its effects on cognition.

5. Conclusions

Overall, our findings suggest that wild white-footed mice may use different adaptative responses to cope with predation risk. Since learning and memory are crucial to survival, a significant impairment of those cognitive abilities could, ultimately, cost an individual’s life. The shifts in exploration and search strategies may be an adaptative response to help mice to respond to the risk of predation. Yet, the tendency towards impairment of short-term memory also points to the possibility of a negative consequence of chronic predation risk exposure in wild mammals. If that is the case, the big question may not be ‘whether predation risk causes negative effects on cognitive abilities’ but ‘what is the threshold where predation risk shifts from adaptive to maladaptive outcomes’. Clearly, the effects of predation risk on free-living prey’s cognition require further exploration.

Ethics

Animals involved in this study were cared for in accordance with the IACUC guidelines of the University of Massachusetts Dartmouth, which reviewed and approved methods used (IACUC #19-2).

Data accessibility

The data collected are available as open data via the Dryad data repository [121].

Supplementary material is available online [122].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

C.C.G.: writing—original draft, writing—review and editing; L.McK.: data curation; L.H.: formal analysis, software, visualization; M.J.S.: conceptualization, project administration, resources, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

No funding has been received for this article.