The effect of group size on sleep in a neotropical bat, Artibeus jamaicensis

Abstract

Sleep is associated with many costs, but is also important to survival, with a lack of sleep impairing cognitive function and increasing mortality. Sleeping in groups could alleviate sleep‐associated costs, or could introduce new costs if social sleeping disrupts sleep. Working with the Jamaican fruit bat (Artibeus jamaicensis), we aimed to: (1) describe sleep architecture, (2) assess how sleeping in groups affects sleep, and (3) quantify total sleep time and identify rapid eye movement (REM) sleep using behavioral indicators that complement physiological evidence of sleep. Twenty‐five adult bats were captured in Panama and recorded sleeping in an artificial roost enclosure. Three bats were fitted with an electromyograph and accelerometer and video recorded sleeping alone in controlled laboratory settings. The remaining 22 bats were assigned to differing social configurations (alone, dyad, triad, and tetrad) and video recorded sleeping in an outdoor flight cage. We found that sleep was highly variable among individuals (ranging from 2 h 53 min to 9 h 39 min over a 12‐h period). Although we did not detect statistically significant effects and our sample size was limited, preliminary trends suggest that male bats may sleep longer than females, and individuals sleeping in groups may sleep longer than individuals sleeping alone. We also found a high correspondence between total sleep time quantified visually and quantified using actigraphy (with a 2‐min immobility threshold) and identified physiological correlates of behaviorally‐defined REM. These results serve as a starting point for future work on the ecology and evolution of sleep in bats and other wild mammals.

Research Highlights

We combine behavioral and physiological data to show that Jamaican fruit bats (Artibeus jamaicensis) sleeping in groups may sleep longer than individuals sleeping alone, and that behavior can be used to measure some sleep metrics in wild bats.

1. INTRODUCTION

Sleeping behavior can be described as a reversible state of sustained quiescence, characterized by a relaxed posture and an increased arousal threshold (Zepelin et al., 2005). When sleeping, animals are more vulnerable to predation (Lima et al., 2005) and are unable to engage in many activities that are essential for survival and reproduction (e.g., foraging and mating). Despite the costs of sleeping, sleep is highly conserved across animals (Ungurean et al., 2020). Further, although the function of sleep remains largely unknown, research has demonstrated that restricting sleep has significant costs, including decreased alertness (Dinges et al., 1997), impaired neurocognitive functioning (Lowe et al., 2017), and increased mortality (Shaw et al., 2002). Together these factors suggest a trade‐off in which individuals must contend with opposing selective pressures, balancing the costs and benefits of sleeping behavior.

Many animals sleep in groups, which could alleviate some costs, but also introduce new ones. Sociality and environmental context could have large implications for sleeping animals, with social sleep potentially reducing the risk of predation (Lendrem, 1984), conferring benefits such as behavioral thermoregulation (Gilbert et al., 2010), and serving a sexual function (Anderson, 1998). Conversely, social sleeping could result in increased resource competition (Génin, 2010) or parasite transmission among individuals (Cote and Poulin, 1995) or could allow for greater rates of physical disturbance (Loftus et al., 2022). Previous work has noted a number of potential trade‐offs to social animals living in groups (Davies et al., 2012). One aspect of sociality that remains largely unexplored, however, is the effect that sleeping in a group has on sleep itself (Gravett et al., 2017). If sleep is disrupted for individuals in groups, such that the total time spent sleeping or time in specific sleep stages is reduced, then social sleeping could be an additional cost of group‐living that counter social benefits. Conversely, if sleep duration increases for individuals in groups, then social sleeping could be an additional, underexplored benefit of living in groups.

Most mammals cycle through two stages of sleep, rapid eye movement (REM) and non‐REM (NREM). Although the specific functions of REM and NREM remain largely unknown, both stages have been separately thought to participate in memory consolidation (Yamazaki et al., 2020). Additionally, NREM has been thought to assist with ‘housekeeping’ processes like metabolic clearing, and REM has been associated with reinforcing cognitive processes (Yamazaki et al., 2020). In many mammal species, REM sleep can be identified by behavioral indicators other than the namesake eye movements, including twitching of extremities and postural atonia (Zepelin et al., 2005). Some of these REM characteristics, such as twitching or irregular breathing, can be observed visually, whereas the detection of other REM characteristics requires more sophisticated equipment. For instance, a nearly ubiquitous feature of REM sleep in mammals is a loss of muscle tone, which can be accurately measured using electromyography (EMG) (Blumberg et al., 2020).

1.1. Sleep in bats

Similar to other mammals, bats cycle through REM and NREM sleep stages. Although components of bat sleep architecture were first characterized over 50 years ago (Brebbia & Pyne, 1969, 1972), to our knowledge only one study in recent years has attempted to further characterize sleep states in bats (Zhao et al., 2010). In a comparative study investigating differences in electroencephalogram (EEG) and EMG data between two southeast Asian frugivorous bats (Cynopterus sphinx and Eonycteris spelaea), Zhao et al. (2010) found that C. sphinx exhibited a greater number of REM bouts than E. spelaea, however these REM bouts were shorter in duration. The authors suggested that a higher number of social interactions in C. sphinx could be one explanation for the observed sleep differences between the two species (Zhao et al., 2010). Because the role of sociality in bat sleep architecture has yet to be explicitly examined, this suggestion remains speculative and calls for further research.

Bats sleep in a range of social configurations, from large interspecific colonies to small single‐species groups (Kunz and Fenton, 2003) and are therefore a compelling group for investigating associations between sociality and sleep. Bats have previously been described as “extreme sleepers” (Siegel, 2005), although recent work has tempered this and revealed a high level of individual variation in sleep (Harding et al., 2022). Further examining if different factors (e.g., the social environment) are correlated with sleep components (e.g., sleep duration) could provide important insight into the underlying functions of sleep. There are, however, few studies on bat sleep. The few studies that explicitly assess sleep in bats have largely focused on total sleep duration (e.g., Siegel, 2005; Zepelin and Rechtschaffen, 1974) and sleep duration in response to ambient temperature changes (e.g., Brebbia & Payne, 1972; Downs et al., 2015). Other studies have assessed various roosting behaviors or metabolic processes, without explicitly investigating sleep. For example, studies have explored the wild roosting behavior of whole colonies (e.g., Cayunda et al., 2004; Connell et al., 2006), harems (e.g., Muñoz‐Romo, 2006; Muñoz‐Romo et al., 2008) and female maternity colonies (e.g., Kerth et al., 2003; Markus & Blackshaw, 2002). Few comparisons exist investigating intraspecific sex differences in roosting behavior (but see García‐Rawlins et al., 2021; Muñoz‐Romo, 2006) or in sleep architecture. Further, although studies have investigated hibernation and torpor (e.g., Harmata, 1987; Salinas et al., 2014), torpor and sleep are often studied as separate phenomena, which is in large part a result of the former being defined using metabolic criteria and the latter using behavioral and electrophysiological criteria (Harding et al., 2022). Clear evidence of overlap between these phenomena, such as the presence of NREM sleep rhythms during torpor bouts in multiple species (Krystal et al., 2013; Walker et al., 1977), merits further investigation of their interaction in bats.

In this study, we quantify total sleep time in the Jamaican fruit bat (Artibeus jamaicensis), a highly social medium‐sized bat and one of the most abundant phyllostomids in the Neotropics (Ortega & Castro‐Arellano, 2001). The sleeping behavior of A. jamaicensis is highly suitable for investigating the effects of social sleeping because this species roosts in different social configurations (Ortega & Castro‐Arellano, 2001). A single dominant male will roost together with several females (typically between 4 and 18 individuals; Ortega & Arita, 1999). Additionally, satellite males often sleep singly or in small groups of two to three (Morrison, 1979). We leverage this natural variation in behavior to investigate the extent to which sleep differs for A. jamaicensis sleeping singly, in pairs, and in small groups. We also quantify a REM‐like state based on unique behavioral patterns that we observed (periods of irregular twitching) to try and understand to what extent these behavioral patterns might reflect REM sleep in this species. In doing so, we aim to achieve three objectives: (1) describe the sleep architecture of A. jamaicensis; (2) assess the extent to which sleeping in groups affects sleep; and (3) by correlating behavioral and physiological data, show that sleep can be accurately quantified using solely behavioral indicators.

2. MATERIALS AND METHODS

2.1. Study site and species

Data were collected from A. jamaicensis at the Smithsonian Tropical Research Institute (STRI) from January to June 2019. Twenty‐two adult A. jamaicensis were captured using mist nets (6 x3 m) in forested areas of Soberanía National Park in Panama (Hemingway et al., 2021). Each bat was assigned a unique ID number and weighed, and a forearm measurement was recorded, which was later used to identify bats before filming. See Supporting Information for details on capture and captive keeping conditions.

To avoid a potential first‐night effect, which can result in changes in sleep behavior and sleep onset (Tamaki et al., 2005), all individuals were given a minimum of 30 h to acclimate to captivity before being used for EMG data collection or for social sleep filming. Thirty hours exceeds the typical time used for acclimating rodent models to experimental cages for sleep studies (Gulia, 2024), and also exceeds acclimation times reported by other studies investigating wild bat behavior (e.g., Gomes et al., 2017; Rhebergen et al., 2015). Although it is possible that bat behavior was still impacted by the process of capturing and handling after ~30 h, these concerns are necessary limitations that exist for most work with wild‐caught animals. Once EMG data collection and filming were completed, bats were released at their capture location. To ensure the long‐term welfare of the bats, individuals were not kept for longer than 10 days. All research was licensed and approved both by the STRI (IACUC protocols: 2014‐0101‐2017, 2017‐0102‐2020, 2019‐0101‐2022) and by the Ministry of the Environment of Panamá (MiAmbiente protocols: SE/A 69−15, SE/AH‐2‐16, SE/AH‐2‐17, SE/A‐100‐18, SE/A‐70‐19).

2.2. Social sleep

2.2.1. Sleep behavior filming

Social sleep filming was conducted in an enclosed observation “roost” structure within a large flight cage. The roost consisted of a rectangular box on stilts ~1 m from ground level (L x W x H: ~1.2 m x 0.6 x 0.6 m) with three walls made of a black mesh, allowing the bats to move and roost freely within the enclosure. The fourth side was fixed to a glass window, which allowed for real‐time observations and night filming without disrupting the natural behavior of the bats. Sleep was filmed from 07:00 to 19:00, with two Sony Handycams (Sony DCR‐SR45) in NightShot Mode and four Security Cameras (Circuit CR‐AHD816), illuminated with an infrared light (Figure 1). All cameras were synchronized at the start of each filming night with a flash from a smartphone camera or headlamp. When filming was not taking place, bats continued to be housed together in the observation roost. Bats were also acclimatized in the artificial roost with bats from the sleep experiment before filming commenced. All bats were housed in small holding chambers in the flight cage before being used in our study.

Figure 1.

Setup of social and EMG bat sleep filming at STRI laboratories. EMG filming was conducted inside the laboratory to protect equipment, and social sleep filming was conducted in outdoor flight cages, with (a) window into the flight cage, and (b) observation roost where (c) the bats were filmed. (d) Sony Handycams ~4.5 m from bats and (e) security cameras ~ 0.6 m (fitted with infrared light) were used to film the bats, and food and water were provided beneath the cameras (f). Security camera footage was recorded onto a Windows computer for analysis and allowed researchers to monitor the bats in real time (g). Cameras were setup approximately 1.5 m from the bats. Illustration by Hannah B. Tilley. See Figure S1 for photographs of this setup. EMG, electromyography; STRI, Smithsonian Tropical Research Institute.

We filmed nine groups of bats in four social configurations: three single male bats, two pairs (“dyads,” both males), one group of three (“triads,” 2 females, 1 male), and three groups of four (“tetrads,” two groups of 4 males, one group of 3 females and 1 male). Bats captured on the same night were placed into these social configurations and acclimated to the observation roost for a 24‐h period before sleep filming commenced. To ensure bats in groups were identifiable during sleep filming and in data collection from videos, a unique sticker was placed on the head and/or ears of each bat.

2.2.2. Sleep behavioral data collection and video scoring

We quantified the duration of total sleep time from videos. For each bat, we recorded the “social configuration” (i.e., single, dyad, triad, and tetrad), and when a sleep bout started and ended. We considered the start of a sleep bout (sustained quiescence) to be when the bat stopped moving and stopped interacting with its environment (i.e., the individual was not grooming, feeding, or interacting with another bat) and if visible, when its eyes were closed (as described in Downs et al., 2015). When entering a sleeping state, limb joints, including the shoulders, and knees would relax, moving from a bent posture to a straightened position (pers. obs. by the authors; Table 1). When sleeping, the bat's wings were relaxed and positioned lower and closer to the face than when merely in an awake state of quiescence or “rest.” In addition to wing position differences, during a resting state, the bat's eyes remained fully open and head/pinna would be intermittently rotating. Conversely, if the bat was in a mobile awake state it was usually engaging in an activity such as grooming, feeding or, when in group situations, interacting with another individual. The eyes did not have to be fully closed for the bat to be recorded as sleeping whereas resting bats' eyes were fully open (Downs et al., 2015) (see Supporting Information videos of bats engaging in different behaviors: awake and resting, awake and active, in REM‐like sleep, and asleep). We relied primarily on the Sony Handycam videos for scoring behavior and would use the security camera footage to validate behaviors when they were difficult to identify with the Sony Handycam videos.

Table 1.

 Description of the behaviors used to distinguish between bats that were asleep, exhibiting REM‐like twitches, and awake.

Sleep state	Description of behavior
Asleep	Eyes closed or partially open
	Straightened body posture
	Wings held close to the face
REM‐like	Above criteria for “asleep”
	Twitches of the whole body, wings, or ears
Awake	Eyes fully or partially open
	Less relaxed wings and body posture compared to “asleep”
	Possibly exhibiting active behaviors (e.g., grooming, feeding, looking around, or interacting with conspecifics)

Abbreviation: REM, rapid eye movement.

We observed a twitching behavior akin to the kind of twitches often described during REM sleep, so for each bat we recorded when REM‐like behavior started and ended. We considered a REM‐like sleep bout to have begun once the bat started to exhibit twitching (as described in Sánchez‐López and Escudero, 2011). These twitches ranged from minor wing or nose twitches to full body twitches that could last for several minutes (see Supporting Information Video of A. jamaicensis in REM‐like sleep).

We filmed all bats over a 12‐h period. For one bat from each group (selected at random), we collected data continuously for the 12 h, whereas for the other bats from that same social configuration, we collected data at 15‐min intervals throughout the 12‐h period (at every 15‐min interval after filming started; e.g., 0, 15, 30, 45 min). If the bat was sleeping at the start of the 15‐min interval, we rewound the video to the start of the sleep bout and began collecting data from the point where the bat entered a sleep state.

2.2.3. EMGs

Three additional single male bats (which did not participate in the social sleeping trials described above) were fitted with devices to record EMG during sleep. Bats were filmed indoors in laboratory conditions, with controlled 12‐h light/dark cycles (06:00−18:00). These controls allowed for the temperature and relative humidity to be kept constant (approximately 25°C and 60−70% humidity), minimizing the effect of the environment on the EMG equipment. The bats were kept in the same observation roost structure with the same camera setup (as described in Sleep behavioral data collection and video scoring) that was used for the social sleep filming. The filming occurred for 12 h, between 07:00−19:00 and behavioral sleep data were collected for bats using the same methods that were used in the social sleep trials (see Sleep behavioral data collection and video scoring).

Electrophysiological recordings were acquired using a VESPER wildlife logger (Alexander Schwartz Developments) featuring a 16‐bit triaxial accelerometer sensor (MPU‐9150) with a ±2 g dynamic range and a 16‐bit analog‐to‐digital converter for measuring biopotentials (ADS1192) with a ±0.1 mV. Acceleration and EMG were recorded at a sampling rate of 242 Hz and 50 Hz, respectively, with analogue high‐pass filtering at 1.5 Hz. Data were saved locally to a microSD card in consecutive 325‐s files. To attach the VESPER logger to the bats, we anesthetized the bats using Ketamine (15 mg/kg), Domitor (0.06 mg/kg) and Midazolam (0.2 mg/kg). We then inserted the electrodes through each of the trapezius muscles until the tips of the electrodes were touching the surface of the scapulae. Collodion adhesive, latex with wood glue were used to adhere the electrodes, and a ball chain necklace was used to attach the VESPER board with the accelerometer, and the battery to the back of the bat. These devices were bound together using parafilm and athletic tape, to further ensure that they were secure and unlikely to be displaced by the bat through grooming or excessive movement over the filming period. Filming began 12 h following the procedure, allowing for overnight recovery. Bats were recorded for at least 2 days to ensure that we had sufficient data in the event of a technical error (e.g., an electrode becoming dislodged). For each bat, a 12‐h period identified to have consistent electrophysiological recordings was chosen for behavioral video scoring as well as EMG analysis. For one bat, data were collected 12 h after the operation, for another the data were collected 36 h after, and for the final bat data were collected 60 h after. Although the differing recovery periods could impact behavior, the objective of using EMGs was not to quantify undisturbed sleeping behavior, but rather to confirm the behavioral scoring corresponds to the physiological measurements. Our decision to begin filming the day following the procedures was motivated by the ethical consideration of minimizing the amount of time that bats were required to carry the equipment. Upon completion of filming, the recorders were removed, and the bats were monitored for at least 2 days to ensure that they were healthy before being released at the location where they were captured. To ensure bat welfare protocols were observed, no bats carried EMG recorders for longer than 3 days.

2.3. Data analysis

2.3.1. Visual sleep analysis

All visual sleep analyses were performed in R version 4.0.4 (R Core Team, 2020). For this, we scored the behavior of eight bats where data were collected continuously over 12 h, providing a baseline of their behavior. An additional 14 bats were scored in 15‐min intervals (bats were randomly selected to be scored fully or in intervals). As previously mentioned, some bats were fully scored (i.e., data were collected continuously over 12 h), whereas for others data were collected at 15‐min intervals, so we started by creating two datasets. The first data set (the “full data set") only included bats that were fully scored. The second data set (the “subset data set") included bats that were not fully scored, along with bats that were fully scored; only sleep bouts that overlapped with the 15‐min intervals (i.e., 7:00, 7:15, 7:30, 7:45… 19:00) were included in this data set. Binning data this way ensured that data were consistent across bats. Based on the sleeping start and end times, we calculated the duration of each sleeping bout and each REM bout.

To characterize the sleep architecture of A. jamaicensis, we calculated the mean and standard deviation of total sleeping duration, irrespective of group size. Given that quantification of REM sleep duration from behavioral criteria is not standard practice, descriptive statistics focused only on total sleep. To compare sleep between the sexes (irrespective of group size), we conducted a Wilcoxon signed rank test on the total time spent sleeping for male (n = 16) and female (n = 6) bats, and calculated Wilcoxon's effect size. We analyzed both total sleep time and time spent in REM‐like sleep here, because our goal in comparing the sexes is to compare sleep in one group relative to the other rather than quantifying durations. Similarly, we used the full data set to calculate descriptive statistics here, but used the subset data set to compare the sexes because the objective was to compare the magnitude of sleep differences between males and females rather than to describe sleep across a full 12‐h period.

To assess the impacts of group size on sleep metrics, we built separate generalized linear mixed‐effects models (GLMMs) using the lme4 R package (Bates et al., 2015). The dependent variable in these models was either total sleep duration or REM‐like sleep duration, and the independent variable was group size. We included a random effect of bat ID nested within group ID to account for multiple records of the same individual within each group and for within‐group variation. Both GLMMs were modeled with Poisson error distributions (both sleep duration and REM‐like sleep duration are count data) and models were fit using the subset data set.

2.3.2. Nonvisual sleep analyses

All nonvisual sleep analyses were performed using MATLAB (version R2020b, The Mathworks Inc., 2020). Acceleration in each axis was filtered offline with a band‐pass Butterworth filter with cut‐off frequencies 0.5 and 15 Hz (Van Hees et al., 2014). To validate sleep, sessions were scored using both actigraphy and visual criteria. To estimate sleep time using actigraphy, we derived two thresholds: an activity level threshold for immobility and an immobility time threshold for sleep. Instantaneous root mean square (RMS) was calculated along each accelerometer axis using a 10‐s sliding window. Times at which the instantaneous RMS was <12 mg in all three axes were classified as immobile and all other times as active. This threshold was set at 3 times the RMS noise level of the accelerometer. As sleep latency (the time between the end of activity and the start of sleep) has not been characterized in this species, we tested both a 5‐ and 2‐min immobility threshold for defining periods of sleep. A binary time series of sleep or wake times was then generated at the same sampling rate as the accelerometer. Visual scoring was performed using the method described previously (see Sleep behavioral data collection and video scoring). To compare with actigraphy, visually scored time series were re‐expressed at the sampling rate of the accelerometer and simplified to a binary series in which active and grooming behaviors were designated as wake and all other times designated as sleep. We multiplied the sleep duration during the 12‐h sleep session by 2 to estimate the consistency between methods in scoring total sleep time over a full day without accounting for diurnal variation in sleep propensity. Accuracy, recall and precision of the visual scoring method were calculated against the actigraphy method for both sleep length thresholds.

To determine whether the visually identified REM‐like sleep state represented a unique physiological state from general sleep, we extracted the following parameters from the periods ±2 min from the onset of each REM‐like sleep episode: muscle tone, movement intensity, heart rate and heart rate variability. By matching periods of high intensity activity, we estimated a synchronization error of ±30 s between our visual scoring and physiological measures due to dropped frames accrued over the 12‐h recording period. We therefore chose to only analyze periods >2‐min to ensure that >50% of extracted data was the correct state. The remaining REM‐like episodes <2 min, which accounted for 41% of all REM‐like bouts but only 20% of the total time spent in REM‐like sleep, were not included in the analysis of REM physiology. Time points in the pre/post‐REM‐like onset period in which animals were active according to the accelerometry definition were also rejected.

Muscle tone was assessed as power in the EMG signal using Welch's method (Welch, 1967) with 2‐s Hamming windows (50% overlap) and expressed relative to the pre/post‐mean power. Movement intensity was estimated from changes in acceleration. First, the three acceleration axes were reduced to a single axis by taking the Euclidean norm or magnitude at each sampling point. Then, the standard deviation of the accelerometer magnitude during the pre/post‐REM‐like onset periods was calculated. In one animal in which the EMG was contaminated with ECG, the following metrics were also extracted: mean heart rate and heart rate variability (standard deviation in inter‐beat interval). Heart beats were detected using the RQRS detection algorithm in the PhysioZoo toolbox (Behar et al., 2018). All physiological metrics were averaged across pre/post‐ REM‐like onset periods.

3. RESULTS

3.1. Sleep architecture and social sleeping

On average, ignoring group size, bats spent 6 h 32 min ± 2 h 20 min sleeping across the 12‐h period. Sleep duration was highly variable across individuals, ranging from 2 h 53 min to 9 h 39 min. Each sleep bout lasted 3 ± 1 min. Females spent less time asleep and in REM‐like sleep than males (females: 4 h 47 min ± 26 min; males: 7 h 8 min ± 2 h 26 min; note that these values were calculated with the fully scored bats; n = 2 females, n = 6 males). However, these differences were not statistically significant (Figure 2; mean sleep: W = 30, p = 0.20; mean REM‐like sleep: W = 46, p = 0.91), and effect sizes were low to moderate (r = 0.28) for mean sleep duration, and low (r = 0.03) for REM‐like sleep duration.

Figure 2.

Mean times spent (a) sleeping and (b) in REM‐like sleep for male and female bats. The midline of the boxplots denotes the median, and the data points represent individual bats. These plots were created using the subset data set, so the times displayed are not across a continuous 12‐h period. REM, rapid eye movement.

Group size did not affect either of the sleep metrics that we investigated (mean sleep: χ ² ₃ = 0.84, p = 0.84; mean REM‐like sleep: χ ² ₃ = 1.22, p = 0.75). However, despite lack of statistical significance, the trend in our data suggests that individuals sleeping alone may sleep less than individuals sleeping in small groups (Figure 3). Both mean sleep duration and mean REM‐like sleep duration were highly variable within and across groups of the same and different sizes (Figure 4), and sleep bouts were highly variable across the 12‐h period (Figure 5).

Figure 3.

Mean total time spent sleeping and in REM‐like sleep for bats sleeping alone, in dyads, triads, and tetrads. The midline of the boxplots denotes the median, and the data points represent the overall means of each group (e.g., a single bat, or a single dyad, triad, or tetrad). These plots were created using the subset data set, so the times displayed are not across a continuous 12‐h period. REM, rapid eye movement.

Figure 4.

Time each individual bat spent (a) asleep and (b) in REM‐like sleep, colored by group size. A single data point represents the duration of a sleep bout or REM‐like sleep bout; the sum of all sleep bouts for a single bat thus reflects the total sleep duration or REM‐like sleep duration. The midline of the boxplots denotes the median. These plots were created using the subset data set, so the times displayed are not across a continuous 12‐h period. REM, rapid eye movement.

Figure 5.

Timing and duration of sleep bouts across the full 12‐h subjective day period (data were only collected during these 12 h), for each of the fully scored bats. Showing bats sleeping (a) alone, and in (b) dyads, (c) triads, and (d) tetrads. Data points represent the total duration of individual sleep bouts.

3.2. Validation of visual scoring

To assess the validity of our visual scoring method, we compared our behavioral results with actigraphy‐derived sleep measures. Using an immobility threshold of 5 min, which provides optimal evaluation of sleep time in humans compared to polysomnography (Chae et al., 2009), there was a 76.6 ± 12.6% agreement between methods (i.e., the proportion of sleep bouts detected both visually and using actigraphy). Although the majority of actigraphy‐derived sleep periods were also detected visually (recall of the actigraphy model, 93.8 ± 3.5%), a considerable proportion of visually‐scored sleep periods did not coincide with actigraphic sleep (precision of the actigraphy model, 69.5 ± 24.2%). Using a shorter immobility threshold of 2 min improved the agreement between methods to 89.1 ± 4.3% (recall, 92.5 ± 3.9%; precision, 91.3 ± 6.1%) and led to estimates of sleep per hour during the dark period across animals that were statistically indistinguishable (mean values: actigraphy, 40.9 ± 8.2 min; visual, 41.3 ± 6.0 min).

We found consistent differences in physiology measured before and after the onset of REM‐like sleep (Figure 6). The transition was associated with a decrease in muscle tone (mean values: post−pre,—0.09 μV²/Hz) indicative of muscle atonia during REM‐like sleep concomitant with an increase in movement (mean values: post‐pre, 0.92 mg), which from cross‐referencing with video recordings appears to be the result of twitching, causing the suspended animals to rock (Figure 6). In the single animal for which heart rate metrics could be extracted, the transition to REM‐like sleep was associated with a slight decrease in heart rate (post−pre, –23.72 bpm) and a large increase in heart rate variability (post−pre, 29.45 bpm) (Figure 6).

Figure 6.

Physiological changes associated with the transition to REM‐like sleep in a solitary housed bat. (a) Bar charts in clockwise order from top‐left show muscle tone (EMG power 30−100 Hz relative to the mean of the pre/post‐REM onset periods), movement intensity (standard deviation in accelerometer magnitude), heart rate (mean inter‐beat interval) and heart rate variability (standard deviation in interbeat interval) in the 2‐min periods before and following the onset of REM‐like sleep. Standard error of the mean (SEM) bars are shown for metrics in which multiple animals were measured. Muscle tone is represented relative to the mean muscle tone between time periods for each animal. The black lines represent the individual values for each animal. (b) Muscle tone is decreased but interspersed with high amplitude muscle twitches that initiate whole body movements detectable in accelerometer recordings. Heart rate is decreased whilst heart rate variability is increased during the REM event. The REM event is followed by a brief awakening characterized by sustained movement and recovery of muscle tone and cardiac behavior. EMG, electromyography; REM, rapid eye movement.

4. DISCUSSION

Our results show that a high degree of sleep variation exists among individual bats; a result that is consistent with other recent investigations into bat sleep (Harding et al., 2022). Sleeping in small groups does not appear to be an important determinant of sleep variation, because we did not find statistically significant evidence that sleeping alone differs from sleeping in a small group. However, this result may be owing to some methodological and biological factors, some of which could be investigated in future work (see below). Behavioral scoring of sleep generally aligned with actigraphy, particularly when shorter (2‐min) immobility thresholds were used. Behavioral indicators could therefore be used to measure some sleep metrics in wild bat populations.

4.1. A. jamaicensis sleep architecture

Irrespective of group size, bats only slept for approximately half ( ~ 55%) of the 12‐h filming period (6 h 32 min ± le 2 h 20 min). These data, and those taken from the subset of bats used in the physiology experiments suggest that A. jamaicensis are not “extreme sleepers” as has been suggested of bats generally (Siegel, 2005). Rather, these bats exhibit a typical, overall sleep time compared to other mammals.

Substantial sleep variation existed among individuals (range: 2 h 53 min—9 h 33 min). Many factors could contribute to this high degree of among‐individual sleep variation. In humans and other animals, sleep can vary based on stress (Pawlyk et al., 2008), hunger or food intake before sleep (Van Der Vinne et al., 2014), age (Ancoli‐Israel, 2009), social status (Smeltzer et al., 2022) and personality (Kim et al., 2015). Sex has also been identified as an important predictor of sleep in mammals (Swift et al., 2020), including other bat species.

Although we found no statistically significant evidence to suggest that sex differences exist for A. jamaicensis, preliminary trends suggest that males and females could differ in total sleep duration. Although no other research exists comparing sex differences in A. jamaicensis sleep architecture, the absence of statistical significance here contrasts with research findings in other species of bats. For instance, differences in roost behavior have been observed in a species from the same genus (A. lituratus), where both sexes averaged 20‐min continuous sleep bouts, but males spent more time awake than females after each sleep bout (3 vs. 1 min; Muñoz‐Romo, 2006). Similar differences have been observed in time budget analyses where females spend more time grooming within roosts (Muñoz‐Romo, 2006) or engage in different “awake” behaviors than males (García‐Rawlins et al., 2021). Our limited sample size (n = 6 females) could provide an explanation for the lack of statistical significance in sex‐specific sleep differences we observed, although future work will be required to verify the direction and magnitude of this suggestive trend.

In addition to factors underlying individual‐level variation, future work could investigate how sleep varies among populations or species of bats. In the present study, we sampled bats opportunistically and were unaware of the population of origin for the bats in our study. In the wild, A. jamaicensis can travel up to 10 km (8 ± 2 km) from roosting sites to forage (Morrison, 1979), and, as this species is extremely abundant in the forests sampled, bats from different populations were likely represented in our study. Researchers of future work could deliberately sample roosts, so the geographic origin of a given individual is known, and the formation of social groupings is deliberate with respect to native roosts. Sampling bats from known populations that differ in ecological properties could be used to assess population‐ or species‐level drivers of sleep variation. For instance, different populations or species of bats have diverse diets (Altringham et al., 1996), host‐parasite (van Schaik & Kerth, 2016) or predator‐prey dynamics (see Lima and O'Keefe, 2013), and behaviors relating to roost selection, habitat use (Jung et al., 1999) and sociality (Kerth, 2008). Given the known consequences of sleep variation for performance and fitness (Krueger et al., 2016), understanding the ecological drivers of sleep variation could have evolutionary implications. Future studies with A. jamaicensis could focus on a more highly controlled approach in common garden settings where individuals from known populations could be used to understand the extent to which sleep variation in this species is plastic or genetically based.

4.2. The effects of social sleeping

We found that sleeping in small groups has no effect on sleep in A. jamaicensis, although trends in our data suggest that individuals sleeping in groups may sleep longer than individuals sleeping alone. We suggest that more work is needed to compare how effects of group size on sleep and fitness‐relevant wakeful states affected by sleep differ for larger groups of A. jamaicensis or other bat species before a conclusion can be reached on extent to which sleeping in groups is beneficial, costly, or neither, for sleep itself.

There are several reasons why our results may not be generalizable to other groups of A. jamaicensis or other species of bats. For instance, A. jamaicensis have been recorded sleeping in groups of up to 20 in nature and sometimes sleep in larger interspecific colonies (Ortega & Arita, 1999). Our group sizes (ranging from 2 to 4) were therefore small relative to many natural A. jamaicensis roosts, and stronger trends might have emerged had we compared individuals sleeping alone to those sleeping in groups that push the naturally observed upper limits of A. jamaicensis group sizes (e.g., 15+). Future research could further investigate the impact of intraspecific or interspecific proximity on sleep architecture. Additionally, although we did not observe statistically significant sleep differences for groups of A. jamaicensis, many species of bats roost in much larger colony sizes, with some groups consisting of hundreds of thousands of individuals (e.g., Tadarida brasiliensis; McCracken, O'Shea and Bogan, 2003). In other bat species with different social roosting configurations, social sleep behavior may prove more critical. Finally, because we artificially created social groups, some natural features of social roosting that could impact sleep architecture are not considered in the present study. For instance, individuals may be more likely to socially sleep with kin or familiar individuals, and their sleeping arrangements could differ depending on established dominance hierarchies. Future studies could consider how these within‐roost social dynamics could impact social sleeping.

4.3. Consistency across behavior and physiology

EEG‐based polysomnography remains the gold‐standard for the detection and quantification of sleep due to its strong correlation with behavioral sleep criteria, its consistency across mammals and its ability to differentiate NREM and REM states at a high temporal resolution. Despite ever more sophisticated means of recording EEG in freely behaving animals outside of the laboratory environment continuing to be developed (Kendall‐Bar et al., 2023; van Hasselt et al., 2020), there remain settings in which full EEG studies are not yet practical. For example, whilst neuronal measures are becoming increasingly adopted (e.g., Pophale et al., 2023; Tainton‐Heap et al., 2021), behavioral sleep measures remain the standard for sleep studies conducted in invertebrates (e.g., Kaiser, 1988; Kanaya et al., 2020; Nath et al., 2017). Given the paucity of wild sleep data in the literature, alternatives to EEG that are easier to implement are still necessary.

We found that videographic sleep scoring was highly consistent with accelerometer‐based actigraphy in a small sample of bats recorded overnight under laboratory conditions (92% for total sleep duration). The optimal immobility threshold for actigraphy varies depending on species‐specific latency to sleep onset, the time between becoming immobile and falling asleep, with reports ranging from 40 s in mice (Fisher et al., 2012) to 5 min in humans (Chae et al., 2009). As the 2‐min immobility criterion used in this study falls within this range, we suggest that video is a suitable alternative to actigraphy for scoring total sleep time in this species, as long as the appropriate temporal criterion (in this case, 2 min of immobility) is used. Nevertheless, this methodology would benefit from EEG‐based characterization of latency to sleep onset in this species to evaluate an optimal immobility threshold that ensures shorter bouts are not being missed.

Behavioral scoring of REM sleep is highly desirable but rarely attempted. One reason for this is that behavioral correlates of REM sleep can vary considerably among species, including even the presence of REMs (Blumberg et al., 2020), in contrast to electroencephalographic criteria, which are relatively consistent (e.g., Levendowski et al., 2017). Generally, REM sleep is characterized behaviorally by increased ocular activity and the presence of twitching movements against a background of muscle atonia (Blumberg et al., 2020). Furthermore, heart rate variability has been found to differ between NREM and REM (Campen et al., 2003; Chouchou and Desseilles, 2014). Finally, REM sleep bouts are interspersed with NREM sleep bouts during the sleep period in a characteristic alternating pattern, with a full NREM‐REM cycle lasting anywhere from a few minutes to a few hours, depending on the species (Carskadon and Dement, 2005; Elgar et al., 1988).

We found that video recordings have the potential to track REM bouts in A. jamaicensis, although further work providing EEG confirmation is required. Bats displayed regular bouts of twitching interspersed with non‐twitching sleep throughout the night. In a subset of three individuals, we found that these bouts of twitching were observable in accelerometery data and were associated with a reduced EMG muscle tone compared to prior sleep. In addition, the one animal in which ECG could be extracted showed a 70% increase in the standard deviation of interbeat intervals following the transition to twitching sleep indicative of previously reported changes in heart rate variability. Despite these compelling results, EEG confirmation is nevertheless required because twitching is neither necessary nor sufficient for REM. Twitching is less common in NREM sleep than REM sleep, but does occur regularly in some populations such as infants during quiet sleep or patients with narcolepsy (Geisler et al., 1987; Sokoloff et al., 2021). Therefore, some video‐scored REM bouts could in fact be NREM, leading to an overestimation of total REM sleep duration. Human REM sleep is characterized by the alternation of two microstates: phasic REM periods in which twitching occurs and tonic REM period in which it does not (Simor et al., 2020). If A. jamaicensis do exhibit tonic REM sleep, these periods would be missed by video recordings leading to an underestimation of total REM sleep duration. Phasic REM bouts could also be missed if twitches are of low intensity. This error may be mitigated in bats due their habit of sleeping suspended only by their legs, minimizing the friction component of the inertia that must be overcome to cause them to move. This was observed in the accelerometer recordings as a pendulum‐like oscillation. Hence bats may represent a good target for future studies interested in validating video recordings for REM detection.

Future studies aiming to use EEG to confirm the utility of videos for REM detection in bats could also extend on the work here by attempting to leverage recent advancements in automated video scoring. Indeed, programs such as Deep Lab Cut (Mathis et al., 2018) or SLEAP (Pereira et al., 2022) could allow for greater resolution in the behavioral data to be paired with the physiological data, and could speed up the scoring process, allowing for much larger sample sizes. These automated tools may also be able to track other behaviors, beyond twitches and changes to body posture, that can be indicative of different sleep stages. For instance, in the present study we relied primarily on body posture to define sleep versus rest, which could risk encompassing intermediate sleep states within NREM. Automated video scoring may allow researchers to record finer scale behaviors, such as eye movements if eyes are clearly visible (they were not always visible in our study), providing even deeper insight into the behavioral components of bat sleep.

5. CONCLUSIONS

We aimed to describe sleep architecture in A. jamaicensis, to assess the impacts of sleeping in groups on sleep itself, and to investigate the extent to which physiological and behavioral indicators of sleep are aligned. We found that sleep duration in A. jamaicensis is highly variable among individuals but may not be affected by sleeping in a small group—although more work is needed to corroborate this trend. For comparing physiology, total sleep time was appropriately measured using behavioral indicators, and physiological and behavioral criteria suggested that REM sleep may be detectable in this species without EEG, although future work will again be required to validate this suggestion. These results pave the way for several avenues of future research. To start, much value would be gained by investigating drivers of sleep variation in bats, both intrinsic (e.g., age and sex) and extrinsic (e.g., temperature and predation risk), at the individual, population, and species levels. When considering sociality as one such factor, an emphasis should be placed on studying larger groupings of A. jamaicensis or other bat species, and in groups with differing compositions to what we investigated here (e.g., sex ratios, ages, relatedness, and dominance hierarchies). Many of these studies could be conducted in laboratory settings, such as those aiming to quantify the sleep stages, to validate video recording more precisely for REM detection, or to quantify the genetic versus plastic basis of intraspecific sleep variation. However, an especially exciting avenue for future work will be conducting bat sleep research in natural settings. In showing that some sleep metrics (i.e., total sleep time) can be quantified behaviorally, our study paves the way for this in‐situ research on sleep in the wild.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflict of interest.

ETHICS STATEMENT

All research was licensed and approved both by the Smithsonian Tropical Research Institute (IACUC protocols: 2014‐0101‐2017, 2017‐0102‐2020, 2019‐0101‐2022) and by the Ministry of the Environment of Panamá (MiAmbiente protocols: SE/A 69‐15, SE/AH‐2‐16, SE/AH‐2‐17, SE/A‐100‐18, SE/A‐70‐19).

Supporting information

Supporting information.

Supporting information.

Supporting information.

Supporting information.

Supporting information.

Heckley, A. M. , Harding, C. D. , Page, R. A. , Klein, B. A. , Yovel, Y. , Diebold, C. A. , & Tilley, H. B. (2024). The effect of group size on sleep in a neotropical bat, Artibeus jamaicensis . Journal of Experimental Zoology Part A: Ecological and Integrative Physiology, 341, 1097–1110. 10.1002/jez.2860

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.