Bridging the gap between lab and field sleep studies: a proof-of-concept for studying wild rats in semi-captive environments

Abstract

Sleep is a vital and universal behavior distinct from mere inactivity, yet its ecological role remains poorly understood due to methodological limitations in recording sleep in the wild. Using a small, low-power biologger, collecting brain activity, body movements, and physiology, we recorded key sleep parameters in wild black rats (Rattus rattus) under semi-captive conditions. We developed a rapid (<1 h) surgical procedure using a custom subdermal flexible electrode, providing signal quality comparable to standard cortical electrodes. Our validated semi-captive setup allowed animals to remain in their natural environment with ad libitum food and social contact while minimizing interactions. This protocol enables the study of sleep’s ecological role and the influence of environmental factors on sleep expression, offering insights into its evolution. Additionally, it can help clarify sleep’s central role in the context of global environmental change. By monitoring general behavior and sleep patterns in four wild rats for up to 10 days post-surgery, as well as feeding behavior for over a month, we observed no signs of pain or stress, with sleep patterns stabilizing within 2 days. This approach provides a unique tool to assess sleep variability and flexibility, demonstrating its feasibility for studying sleep in small (<200 g) wild animals.

Statement of Significance.

Sleep is vital but poorly understood in ecological contexts due to methodological constraints. We present a novel biologging approach that enables high-quality sleep recordings in small wild animals. By studying wild black rats in semi-captive conditions, we bridge lab-based research and real-world environments. This method allows the investigation of the ecological role of sleep, providing a new tool to assess sleep flexibility and adaptation in natural settings.

Sleep is not simple rest. It is associated with a low level of vigilance that makes it a risky state [1]. It is complex and can be expressed under multiple forms (drowsiness, Non-Rapid Eye Movement (NREM) sleep, REM sleep, unihemispheric sleep, etc.), which likely support different functions and adaptations. Moreover, sleep is a fundamental and universal behavior that has been described in all animals to date [2], suggesting that it may provide important benefits for animal survival and reproductive success (Figure 1). Indeed, much effort has been directed towards understanding its mechanisms [3], and various studies demonstrated a strong association between sleep and developmental [4], cognitive [5, 6], restorative [7], and maintenance [8] processes, providing an indirect benefit to animal fitness through an improvement of the animal’s performance during wake (Figure 1). On the contrary, when animals are sleeping, they are sacrificing their vigilance to avoid predation and their time to forage and reproduce to that state (Figure 1). In addition, if there are some physiological benefits of sleeping, it has also been demonstrated in laboratory species and humans that there is an important physiological cost of not sleeping. Indeed, unlike other resting states, sleep states are also homeostatically regulated, and their disruption, or deprivation, can have deleterious physiological consequences [9, 10], leading to possible negative impacts on animal fitness. Altogether, this demonstrates that a fine balance exists between sleep and wake.

Figure 1.

Schematic representation of the interactions between sleep and wake, showing the theoretical benefits and the costs of sleep on animal fitness (survival and reproduction).

Unfortunately, most of the sleep research has been performed in laboratory environments on laboratory model animals or humans. These efforts failed to inform us on this ecological balance in animal time budgets or on the diversity of sleep adaptations [11–13]. Indeed, only very few studies have been conducted on wild species in laboratory conditions or in natura. A large diversity of sleep expressions was nonetheless reported from those studies on non-conventional animals [2, 14], with elephants sleeping 2 hours a day and armadillos up to 18 hours, marine mammals and birds being able to sleep unihemispherically. The quantity of REM sleep also varies across species. European hedgehogs have 3.5 hours of REM sleep a day, whereas horses can have 0.5 hours, and bottlenose dolphins seem to have no REM sleep [15]. This highlights the central role of sleep and its substates but also suggests a large range of adaptations, likely driven by physiological and environmental factors. Understanding those factors may thus provide important cues to decipher the eco-evolutionary role of sleep. Then, more than describing the diversity of sleep expression across the animal kingdom, its flexibility and variability across and within individuals and populations should be studied in wild species under natural conditions.

Some experiments have, for example, been conducted to show that parental care could have driven some extreme sleep fragmentation in penguins [16], that foraging in frigate birds [17] and sexual competition in sandpiper birds [18] may induce important reductions of sleep with no obvious physiological cost of this sleep loss. Seasonal variations, which often drive food availability or induce thermal challenge, could also impact the sleep distribution and quantity in barnacle geese [19]. As a chronic source of stress, predation pressure [20] may also be an important driver of the evolution of sleep in natural populations and may have driven the observed variations in sleep characteristics across species. Indeed, variability in some sleep characteristics, such as its daily duration [21], the relative amounts of REM and NREM sleep [22], the daily timing [23], the occurrence of unihemispheric sleep [24], and sleep fragmentation [25], have been putatively correlated with levels of predation pressure.

Those studies are unfortunately too rare to widely understand the ecological role of sleep. This is mostly explained by three factors: I. An understanding of sleep and sleep states (not simply inactivity) requires collecting information on brain activity and thus relies on dedicated methodologies such as electroencephalography (EEG). II. Sleep, being a rapidly evolving dynamic state, requires recording techniques with high temporal resolution. III. Sleep is affected by many abiotic (light, temperature, noise, weather, etc.) and biotic (predation, reproduction, parental care, food resources, parasitism, etc.) factors, which must be controlled or integrated in statistical models to estimate causality and thus require long-term recording. Despite those limitations, methodologies have already been deployed in the wild to study sleep, often with a trade-off between the precision needed to disentangle sleep and its substates from inactivity, the recording capacity, the animal size, and the number of individuals [26–29]. Recent studies revealing the adaptive nature of sleep states, for example, have been based on data collected for 10 days maximum (usually less) on species weighing usually more than 4 kg (Table 1) [16–18, 21, 30–34].

Table 1.

Comparative table of sleep recordings using EEG conducted on wild species in natural environments (semi-captive conditions and wild) and associated methodologies.

Species	Three-toed sloth (Bradypus variegatus)	Ostrich (Struthio camelus)	Pectoral Sandpipers (Calidris melanotos)	barn owls (Tyto alba)	Sloths (Bradypus variegatus and Bradypus pygmaeus)	Fregate bird (Fregata minor)	wildebeest (Connochaetes taurinuys)	Northern elephant seal (Mirounga angustirostris)	Chinstrap penguins (Pigoscelis antractica)
Weight (kg)	3.5–4.5	82 ± 4	~ 0.1	0.3–0.4	3–4	1.3 ± 0.024	~250	>30 [26–200]	4.14 ± 0.3
n	3	6	11	66	15	14	2	8	12
Recording duration (days)	3.1–5.1	9.2 ± 2.8 [0.7–18.6]	5.5 ± 0.7 [3.6–9.14]	2.7 [0.8–3.8]	[8.8–10]	5.76 ± 0.67 [0.26–10.02]	3	2.5–5	8.75 ± 0.8 [3–10]
EEG	x	x	x	x		x	x	x	x
EMG	x	x	x				x		x
EOG		x
Accelerometer		x		x		x	x	x	x
Gyroscope								x
Magnetometer								x
GPS						x		x	x
Radio Collar					x
Temperature		x							x
Environment	Wild	Natural enclosures	Wild	Nest boxes	Wild	Wild	Natural enclosures	Wild	Wild
Year	2007	2008	2011	2011	2009	2014	2016	~2020	2019
Reference	Rattenborg et al. 2008	Lesku et al. 2011	Lesku et al, 2012	Scriba et al, 2013	Voirin et al. 2014	Rattenborg et al. 2016	Malungo et al. 2021	Kendall Bar et al. 2023	Libourel et al. 2023

The relative important weight of the animals chosen for these studies, as well as the limited recording duration (relative to the animal’s ecology and life history), is mostly constrained by the capacity of the neurologgers, i.e., biologgers capable of recording brain activity. Even if the dimension and power consumption of such loggers have significantly decreased over the past 20 years, the volume of data and the battery size still remain a limitation, together with the need to recapture the animals (for data retrieval). Additionally, the stress induced by the capture and the natural variability of sleep states could also be a limitation when interpreting sleep data collected in the wild. A trade-off between animal size, recording duration, environmental factors, and recapture rate is thus inevitable.

Moreover, if only large animals are recorded for sleep, this could induce a bias, as they represent only a few percentages of the biodiversity [35–37], and the weight has also been importantly correlated with other morphological, physiological, behavioral, and ecological traits. From a comparative perspective, it is important to develop studies on smaller species (<200 g), particularly birds and mammals, and more specifically rats and mice. Indeed, those species exhibit a wide range of physiological and ecological adaptations without the potential confounding factors associated with larger species. Moreover, most of our knowledge of the neurophysiology of sleep comes from laboratory studies on rats and mice, providing a strong foundation for any ecological study.

In addition, assessing sleep does not only involve estimating its quantity. Sleep is a complex trait, with its different elements having potential reciprocal and/or independent benefits and costs. Sleep fragmentation, circadian distribution, homeostasis, depth, infraslow rhythm, asymmetry, and phasic events constitute some of the important elements that could respond to or be modulated by environmental pressures. The homeostatic regulation of sleep, for example, has been related to seasonality [38] and social pressure [39]. Whereas NREM sleep has been suggested to be mostly involved in restorative [7, 40] and cognitive processes [6], REM sleep is thought to be more involved in stress-related behavior [41, 42], associated with memory and emotional processing [43, 44], thus, potentially responding to predation pressure [45, 46]. Fragmentation/drowsiness or unihemispheric sleep, on the other hand, could be an adaptation to maintain vigilance while fulfilling the need to sleep [16, 17, 47–49].

To bridge the gap between the fine mechanistic approaches used in laboratory studies, often on inbred model species, and the evolutionary and ecological approaches that attempt to understand sleep on a broader scale, we have developed a novel methodology applicable to rats and possibly to any other small animals (>80 g). We demonstrate that this method allows for the characterization of the full phenotype of sleep and its substates over longer durations, at both the individual and population levels. Our methods not only finely characterize the natural variability of the main sleep traits but also allow for the estimation of phenotypic flexibility in wild animals in their natural habitats, limiting the potential bias of confounding factors like social interaction, resource availability, or parental care. We deployed our methods successfully on black rats (Rattus rattus) in New Caledonia in 2023. The use of rats is particularly relevant as their sleep biology is well described in the lab, and they are present in a wide variety of ecological contexts (i.e., varying climatic conditions, with and without predators or competitors, etc.), allowing us to assess the impact of ecological factors on sleep phenotypes.

Methods

Ethical approval

The surgical procedure and methodology were validated and refined on lab rats (APAFIS #33624-2021102712475018 v6). In agreement with the local New Caledonian authority (DAVAR) and the territorial laws (in particular Lp. 240-2 and Lp. 243-4 from the country law n° 2017-12 from August 23, 2017), the experiments were ethically approved and conducted under the supervision of a local veterinarian.

Capture

Animals were captured in the austral winter, June 2023, in New Caledonia, in a tropical forest nearby Port-Laguerre (Païta) agronomic research station (22.1037° S, 166.3189° E). Up to six traps (wire cage rodent live traps with spring doors (Pro-trap, Connovation NZ)) were deployed between 04:00 pm and 05:00 pm (Figure 2). The traps were baited with coconut and peanut butter. The rats were collected the next morning. In total, we equipped four black rats (Rattus rattus) (2 females and 2 males) weighing 180 g, 95 g, 197 g, and 208 g. Individuals were included to balance sex ratio and span various body weights. One 81 g female Pacific rat (Rattus exulans) was also kept as a control for behavior in semi-captive conditions without implantation.

Figure 2.

Black rats (Rattus rattus) trapped in 2023 (left). Mesh cages constructed on different capture sites (middle and right).

Semi-natural enclosures

In order to mitigate the trade-off between ease of monitoring, ease of cage construction, and keeping the animals in a similar environment as their natural habitat, we built a 1 m³ cage (1 m × 1 m × 1 m) in galvanized steel mesh with 1 mm diameter wire, spaced by 1.3 cm (Figure 2). The cages were placed 20 m from the location where the individuals were captured and were buried 10 cm into the ground. Yet, this design should reduce competition over territories that otherwise could have impacted sleep. A thick layer of vegetation and wood sticks taken from the immediate animal’s habitat were placed into the cage to recreate a familiar natural soil. A plastic food dispenser with a 200 g capacity provided food ad libitum and could only be accessed from inside the cage. A flowerpot filled with water was buried in the cage, and a water bottle was fixed on the cage near the food dispenser. Finally, a cylindrical shelter of 10 cm by 20 cm made of PVC was provided and covered by vegetation and was either placed on the ground or elevated. We also placed two autonomous custom-designed cameras equipped with infrared lights during the first 6 days of the recording to monitor the animals’ behavior. We decided to maintain the animal in semi-captive conditions in order to standardize some of the key parameters that could impact it, such as food availability and type. Finally, social interactions were maintained as rats could interact with conspecifics through the mesh.

Surgery

Surgeries were conducted in the field under a tent at ambient temperature (27–28 °C). After capture, the animal was placed into a small transparent Plexiglas box (10 cm × 10 cm × 25 cm), weighed, and a cotton ball soaked with isoflurane was introduced in the box. After 1–2 minutes, when the animal showed a reduction in its breathing rate, it was removed from the box and administered a mixture of 80 mg/kg of ketamine (Ketamil) and 0.5 mg/kg of Medetomidine (Medetor) intramuscularly. The animal was then shaved and positioned into a plastic 3D-printed stereotactic frame (https://hackaday.io/project/163510-3d-printed-rodent-stereotaxic-device) adapted to its size. The ear bars were adapted to be used in pressure on the skull and not into the ear cavity (Figure 3). A surgical field was positioned over the animal, and ophthalmic gel was used to protect its eyes from drying. The skin over the skull was cleaned with physiological serum and Betadine. A local analgesic (lidocaine) was injected and sprayed over the skin. After 1–2 min, a 2.5-cm anteroposterior incision was made into the skin, which was spread apart with small clamps. The skull was cleaned with H₂O₂ and slightly scratched with the scalpel blade after drying. The nuchal muscles (rectus capitis) were slightly spread with thin round-tip forceps to prepare for the insertion of the EMG probe. Using an electrode interface board (EIB) that will later support the logger (Figure 3), the custom flexible probe we developed (Figure 3) was then positioned over the skull using the cerebral suture as a reference to position the electrode at the right place. This flexible probe, made in polyimide with gold-plated conductive sites, was designed to record the electrical activity of the olfactory bulb and the frontal and parietal cortex, and the ocular and muscle activity. The use of this flexible probe significantly reduces the invasive nature by keeping the skull intact and thus the duration of the full procedure (45–60 minutes). We next added some conductive EEG paste (EC2, Natus) on the recording site of our flexible probe (Figure 4). The electrode was then repositioned and maintained over the skull with a toothpick while a primer dental cement (Ivoclar Vivadent, Adhese Universal VivaPen) was applied and dried with UV light. When all the electrodes were glued over the skull with a first layer of primer dental cement, a second layer of dental cement (Ivoclar Vivadent Tetric Evoflow A1) was applied. At this time the logger was inserted into the EIB with its 3D-printed protective casing, and the device was maintained in its final position. This allowed us to fix with dental cement a plastic nut on the skull to maintain the biologger in place with plastic screws (M3, 6 mm long). After the EIB and nut were secured in place, the logger was removed, and the electrode was completely covered in dental cement. The EIB and nut were completely secured with dental cement. The skin was then sutured around the EIB connector, and Betadine was applied over the wound. An anti-inflammatory (0.2 mg/kg, meloxicam) and antibiotic (0.1 mL/kg, oxytetracycline) were injected subcutaneously into the back. The logger in its casing was then started and fixed on the head of the animal. The signal quality was checked for a few minutes with the Bluetooth connection, and then the logger was programmed to record on its embedded memory. After that, the animal anesthesia was reversed using 0.25 mg/kg of atipamezole (Reversor). The animal was then placed into its cage, where we supervised its awakening. All rats were fully awake after 10 to 20 minutes. In accordance with our ethical procedure, all animals were monitored twice a day for 48 hours post-surgery. A monitoring sheet was filled out to assess their general condition, posture, and responsiveness, among other parameters. If necessary, the animal was taken out of its cage to receive appropriate care.

Figure 3.

(a) Anesthetized black rat after the implantation of the logger on the stereotactic frame. (b) Sleep logger connected on the EIB with the flexible electrode. (c) Live recordings of the EEG of a black rat in its cage, collected with Bluetooth on a tablet. (d) A rat carrying the sleep logger in its cage. (e) A rat equipped with the sleep logger over its head, eating in its cage. (f) 10 minutes of recording (muscle tone extracted from the EMG, brain activity from the parietal site represented in time frequency, and head movements extracted from the accelerometer) during NREM sleep, REM sleep, and active wake, showing the typical brain signatures of the different vigilant states (delta, sigma, and theta activity), as well as the muscle atonia typical of REM sleep (right).

Figure 4.

(a) Hypnogram representing the evolution of the sleep states across the recording days. Because not all days were complete and scored, and because we scheduled the logger after the 8/06 to record one day every two days there is some missing values (b) Evolution of the wake and sleep state quantities per hour (wake; NREM sleep; REM sleep) showing that the sleep state quantities stabilized two days after the surgery. (c) Number of wake, NREM, and REM sleep bouts per hour, showing a more fragmented sleep the first days after the surgery. (d) Illustration of the signal collected over 60 seconds during each vigilant state, from the top to the bottom: EMG; parietal EEG represented in normalized color-coded time frequency (blue = low spectral power; red = high spectral power); EEG frontal; 3-axis acceleration in; 3-axis angular speed collected from the gyroscope.

Biologger

In order to record cerebral, physiological, and behavioral activity, we used a Phynitty PNano logger (Manitty SAS) derived from a previously developed custom-made logger [27, 50]. The device itself weighs 1.2 g (1.45 with the SD card) and measures 25 × 12 mm. The logger can be used in live mode (data collected on a tablet, Figure 4) or can be programmed (with Bluetooth) to run custom sequences of recording. The logger can record up to six monopolar channels—six EEGs (electroencephalograms) in our case—and three differential channels. Here we only used two, one for electromyography (EMG) and one for electro-oculography (EOG). The sampling rate (typically 256 Hz for eight channels) can be customized. In addition to these electrophysiological channels, the device records ambient temperature and can have an external channel (light, temperature, etc.). These additional channels are very useful to quantify environmental variation when recording animals in their own environment. Additionally, a 9-axis inertial measurement unit (IMU) can be activated. The IMU contained a 3D accelerometer, a 3D magnetometer, and a 3D gyroscope with customizable acquisition range. The power consumption is proportional to the sampling rate and the number of channels activated. In our case, we recorded six EEGs, one EMG, and one EOG at 256 Hz. In addition, we recorded acceleration and gyroscope data at 64 Hz and ambient temperature at 32 Hz. The power consumption with this configuration was 2.8 mA. We used two battery types: a large battery (EVE EF651625, 700 mAh, 8 g) and a smaller battery (EVE EF651615, 400 mAh, 5 g). We could record up to 10 days continuously. However, as the logger allows scheduling the recordings, these can be spread over weeks or months, as the power consumption is negligible when the system is not recording. To prepare the logger, we soldered the battery, inserted the SD card, tropicalized it with acrylic resin (APL, Electrolube), and wrapped it in Teflon tape before inserting it into its plastic case. The total weight of the logger was 12 g (large battery) for the rats weighing over 150 g and 9 g (small battery) for the lighter individuals. The weight of the logger never exceeded 10% of the weight of the animal (5.8%, 6.1%, 6.7%, 9.5%).

Recordings

Four rats were equipped with our system. We programmed the loggers to record 2–3 days, with all eight electrophysiological channels sampled at 256 Hz, the ambient temperature channel at 32 Hz, and the accelerometer, magnetometer, and gyroscope at 64 Hz. The range of the accelerometer was set at ±2G and the gyroscope at 125 dps. The four rats were set up after the surgery in their cage. We also monitored a non-implanted rat in a cage to evaluate the impact of the logger on animal behavior. After that initial recording session, the animals were recaptured and sedated with a cotton ball of isoflurane to check the data and set up a new set of recordings of 1 day every second/third day. After that last manipulation, the animals were left in their cage, and their food consumption was evaluated by checking the food dispenser every 4–6 days when the food dispenser and bottle of water were refilled. Upon each visit, we also checked the animal’s status and position. The experiment was stopped after 30 days, the animals were weighed without the logger, and the data were collected from the SD card. After the experiments, the animals were euthanized following local regulations on invasive species. The final weight gain/loss was evaluated, as well as the signal quality and the proportion of the different vigilant states over the days of recording. Three states (wake, NREM, and REM sleep) were scored in 5-second epoch bouts based on the electrophysiological signals [27]. Every complete 24-hour recording without artifacts was analyzed. The change of the vigilance state quantities (wake, NREM, and REM sleep), as well as their distribution, was used to estimate the possible disturbances due to the semi-natural conditions. Indeed, we were expected to see sleep perturbations because of the new environment and stress caused by the capture, manipulation, and surgery recovery. But as sleep is highly genetically driven [51, 52], it should then stabilize when the environmental conditions remain stable, which should be the case in our semi-captive conditions if the weather is stable. We also computed the percentage of inactivity to compare with the true proportion of sleep in order to determine to what extent an accelerometer itself can or can’t be a reliable way to estimate sleep. To do this, we extracted the percentage of inactivity based on the norm of the three channels of the acceleration, filtered with a high-pass filter (cutoff frequency 1 Hz, order 10). The mean of the acceleration was computed for each 5-second epoch. We used as a threshold to score inactivity and activity, which was the mean acceleration that separated true sleep from true wake. The videos (when available) were also inspected to look at the animal’s behavior to detect potential signs of stress/anxiety or abnormal behavior.

Results

The flexible probe: surgery

Usually, surgeries to record animal sleep states typically require drilling small holes into the skull to insert microelectrodes, which allows recording a stable EEG over days or even months. In addition, electrodes are implanted into the nuchal muscles to detect the muscle atonia during REM sleep and differentiate sleep from wake. Electro–oculogram (EOG) is further used to monitor the phasic ocular activity occurring during REM sleep. The use of the flexible electrode design employed here enabled us to refine the surgery and refrain from drilling holes. This made it easier and quicker to implant animals in the field and shortened the recovery process. Of the four implanted animals, three recovered well and were successfully recorded throughout the experiment. Unfortunately, one rat did not survive. Despite a rehydration with NaCl and an additional injection of meloxicam 17 hours after the surgery, the animal was found dead the next morning. Video observations showed that this rat was never seen eating or drinking post-surgery. Additionally, heavy rain occurred on the second night after surgery. Given that the shelter was positioned at the bottom of the cage, we suspect that the animal may have succumbed to hypothermia despite the branches and foliage placed in its shelter. Although we can exclude post-surgery infection, multiple factors likely contributed to the rapid decline and death of this animal. Lastly, we validated on three individuals that the procedure used (i.e., gluing the electrodes to the skull) was adequate to record for at least 30 days.

The flexible probe: signal quality

Using a flexible probe was a challenge, as the sensors are not directly in contact with the brain, which could possibly lead to an attenuated signal (decreased signal-to-noise ratio). In addition, the use of a flexible probe could be less stable over time, and the amplitude may change after a few days of recording, which would be a problem to consistently analyze the data. Yet, our recordings showed excellent signal quality, consistent across channels, mostly artifact-free, with a desynchronized EEG in the frontal and parietal areas when the animal is awake. The left and the right hemispheres also display similar signals, which is expected in this species. When the animal was active, a typical theta (5–9 Hz) oscillation occurred in the EEG, allowing us to identify and score that state. During NREM sleep and REM sleep, we also observed the classical electrophysiological features of each state, including spindle activity, slow waves, hippocampal theta, and phasic theta when phasic activity occurs during REM sleep (Figure 3). As opposed to the robust change in the gamma band of intracranial olfactory bulb LFP signals between wake and sleep [53], the surface olfactory bulb activity did not significantly change between wake and sleep in the gamma band (50–70 Hz). EOG was recorded for the first time with a flexible probe, but the signal was not consistent across individuals. This is likely due to the electrode position and the fact that the measure was a differential between the left and the right EOG. However, in some cases the EOG provided a usable signal when ocular activity occurred during wake and REM sleep. The EMG was of sufficient quality to detect muscular bursts of activity but also muscle atonia when the animal switched from NREM sleep to REM sleep (Figure 3F). This signature is very useful to develop automated state scoring, as the muscle tone is one of the rare features that differentiate NREM sleep from quiet wake (including freezing). In addition, to these signals, we were also able to collect information on animal movement thanks to the 3-axis accelerometer, 3-axis gyroscope, and 3-axis magnetometer. Those sensors could be used to identify movement, head position, and head direction (magnetometer), but also grooming or freezing when associated with brain activity. Lastly, thanks to the gyroscope, we were also able to distinguish the breathing rate during the resting phase, which could be of interest in some cases. Altogether, the quality of our recording revealed our ability to identify a wide panel of vigilant states, and in particular sleep states (Video S1). The quality of the signal is also sufficient and stable enough across at least 10 days of recording to develop automated sleep scoring and extract with precision state quantity, distribution, and fragmentation. In addition, some specific features associated with cognition (like spindle activity), sleep depth (delta power), or REM sleep (phasic theta), as well as wake substates (quiet wake, active wake, grooming, freezing, . . . ), could be easily identified over multiple days. When comparing sleep (scored with EEG/EMG electrodes) to the proportion of inactivity based on the acceleration alone, a measure often used as a proxy of sleep [54–56], we revealed an overestimation of 14.6 ± 8.9% of the sleep quantity. This comes from the misclassification of quiet wake as sleep when using only acceleration or actimetry. This error could be even more important if the animal is freezing.

Behavior

Thanks to the cameras we placed around the cages, we were able to monitor the animals’ behavior. Animals adapted quite fast to the semi-captivity. They quickly found the shelter, where they built a nest with vegetation available in the cage (Video S1). They also quickly found food and water. We were able to observe the rats eating and drinking in the 5–10 hours after the surgery. The animals also climbed on the mesh and ran on the branches inside the cage (Video S2). Finally, in one animal, we also observed some social interaction with a conspecific lasting most of the night (Video S3). Stress in rats is often characterized by excessive grooming, time spent in the shelter, or freezing. Except for the first days when we visited the animals (Video S4), those states were not observed during our recordings. The animals were also not observed biting the mesh. After the first session, when the animals were recaptured, none displayed lesions that could be related to repetitive attempts to escape, which could sometimes be observed in trapped animals. The three individuals with a logger did not lose significant weight (respective weight gain: 0 g; 8 g; −1 g) after 2–3 days post-surgery when we retrieved the first data suggesting they recovered quickly from the intervention.

Sleep monitoring

Thanks to the EEG, the EMG, and the acceleration data, we were able to score vigilant states of three rats over 3–10 days. The first couple of days after the animals were manipulated, the quantity of NREM sleep tended to be high, whereas the quantity of REM sleep decreased (Figure 5A and D). The total proportion of sleep (NREM and REM sleep) is variable across individuals the first days after manipulation. However, we observed that the relative quantities of NREM and REM sleep tend to stabilize 2 days after manipulation (Figures 4 and 5). Indeed, the coefficient of variation (CV = std/mean) of the three rats decreased to 20.7% for the total sleep time, 1.4% for NREM, and 6.4% for REM sleep when they were not manipulated the day before, vs. 22.6%, 5.3%, and 29.4% if we include all the recording days. When looking at the total sleep time intra-individual variation, the CV also becomes lower ([7%–17%] vs 20.7%), suggesting a stable individual phenotype. When comparing the intra-individual variation of NREM and REM sleep relative to the total sleep time compared to the global CV, the differences are even smaller, suggesting that these traits are highly conserved across individuals. In addition, the highest variations of REM sleep quantities relative to the total sleep time are found when the manipulation days are not removed (6.4% vs. 29.4%). Lastly, the mean bout duration more than the number of bouts also tends to stabilize after manipulation.

Figure 5.

Evolution of the daily quantity (a), number of episodes (b), and mean episode duration (c) of NREM (a–c) and REM sleep (d–f). The percentage of sleep (NREM plus REM sleep) (g) and the relative quantity of NREM (h) and REM (i) sleep. Each animal is color-coded. The red circles indicate days that the animals were manipulated in the previous 24 hours (surgery; data retrieval).

Long-term monitoring

To evaluate the feasibility of monitoring animals for the long term, our four individuals (three carrying a logger, one without) were followed for a month after the surgery (Figure 5). Animals kept their loggers at all times, even when they were not recording data. The monitoring of the food consumption shows that all individuals were eating 19.6 ± 3.3 g of food every day (around 22 g for the animal weighing more than 150 g and 17 g for the lightest one (80 g)). The animals, during the first 3 wk of recording, were always observed in their shelter when the experimenter came to fill the food dispenser. The rats were also reactive when the experimenter checked the animal’s status by touching the shelter. However, between the 28th of June and the 5th of July (the dates of the two last visits), one animal was found lethargic (weight loss 20%), and two animals were found dead, including the animal without a logger (weight loss 16% and 1.2%), while the remaining animal was still in good health (no weight loss). Although we can rule out natural causes of death, it is likely that the animals died due to the heavy rain and temperature drop observed the day before, as their shelter, placed on the floor, lacked sufficient insulation. The food dispensers were generally found empty, which suggests that the animals were eating more during this second part of the experiment. We also had to change one food dispenser, as intruder rats were chewing on it to get food. The animal that survived was the only one that had a high-mounted shelter.

Discussion

Sleep assessment in the wild

A clear trade-off between the need to sleep and the need to remain awake or interact with the environment exists. However, there is an important pressure to sleep, likely because of physiological needs. Understanding how animals manage this need remains crucial to better understand how animals survive, especially in a changing world. To date, very few studies have been able to investigate sleep in the wild, and most of them have focused on large species, whereas small species constitute the majority of the biodiversity. Additionally, a necessary condition for such studies is to monitor the animal’s brain activity, as sleep strongly differs from mere inactivity [16, 17, 30, 31, 33]. Until now, this was not easy in small animals (approximately 100 g) as it was challenging to design and implement lightweight bio loggers as well as methodologies to perform surgeries in the field. Here, we showed that using inactivity as a proxy of sleep in rats is inaccurate. Specifically, the quantity of sleep is overestimated by around 15%, but this could be higher if the animal displays specific behavior, such as freezing, notably when hiding from a predator (Video S4). Thus, our results show that, to decipher the role of sleep and its substates, recording the cerebral activity is unavoidable. Yet, EEG can provide significant insights into how sleep varies across individuals and populations and adapts to environmental changes or physiological needs. Some limitations, however, remain when trying to record animal brain activity longitudinally. One major limitation is the weight of the logger, primarily driven by the size of the battery. In our experiment, the total logger weight ranged from 5.8% to 9.5% of the animal’s body weight, exceeding the commonly recommended 5% limit typically applied to flying or swimming animals [57]. However, the impact of carrying a logger cannot be reduced to weight alone. Factors such as the device’s form factor, buoyancy, aerodynamics, and attachment method can each differently affect an animal’s normal behavior [58, 59]. In our case, the absence of significant weight loss after 1 month, along with behavioral observations, suggests that although further weight reduction would be beneficial, carrying a head-mounted logger may have had only a limited impact on the rat behavior under semi-captive conditions. Nevertheless, it remains uncertain whether the same device would affect behavior similarly in truly free-ranging animals. Additional experiments are needed to address this question. Another important limitation is the reduced autonomy of EEG loggers and the necessity of recapturing animals to retrieve data. Indeed, it is not feasible to wirelessly transmit the large volume of recorded data without significantly increasing the battery size. As an alternative, small radio tags could be attached to facilitate the recovery of free-ranging individuals, but this approach would also add additional weight to the device. Measuring the low amplitude (few microvolts) of the EEG and minimizing movement artifacts to maintain a stable signal across days are also challenging, especially for small animals. By developing a less invasive methodology relative to typical lab-based methods involving a flexible probe and a small programmable logger, we were able to obtain excellent data on three small animals with a good and stable signal quality. This allowed us to record the complete sleep expression, as well as the complete panel of waking behavior, within a relatively long timeframe, as the recording can be delayed or scheduled intermittently. This opens new avenues to understand the ecological role of sleep in nature and its response to environmental changes. Some limitations, however, still remain, especially in terms of size and weight. In the future, we expect that the battery capacity will help to overcome this limitation, but for now the sampling rate could be decreased to record for longer periods of time, as well as the number of channels, and the recording days could be spread across a longer period. As an example, recording 2 EEG and 1 EMG at 128 Hz 1 accelerometer at 32 Hz and 1 temperature at 8 Hz can be recorded for approximately 9 days with a 400 mAh battery (logger of 7 g) and approximately 16 days with a 700 mAh battery (logger of 12 g).

Semi-captive vs. captive recordings vs. free-moving animals

The vast majority of sleep studies are currently conducted under laboratory conditions, often in inbred strains of mice and rats and under artificial (un-ecological) conditions (stable temperature and light cycle, small cage, often tethered to assess EEG). Although this has resulted in an important body of knowledge on the mechanisms of sleep in these animals, the lack of genetic diversity and environmental complexity prevents us from addressing ecological and evolutionary questions. On the opposite, recording non-conventional wild animals in their habitat can reveal the true flexibility and adaptive nature of sleep, but this remains technically challenging in particular because of the high number of possible confounding factors.

Whether we should bring those animals into the lab, keep them in semi-captive conditions, or record them in the wild remains an open question, as there exist trade-offs between those conditions. In the lab, wild animals may be exposed to intense stress as they will be exposed to a novel environment and removed from their natural habitat, but this would allow us to record their sleep continuously under standardized conditions, reducing the bias of external factors. Unfortunately, in these conditions the sleep phenotype may reflect the stress response on sleep traits and may hinder the assessment of natural individual variability. One key advantage is that all individuals would be recorded in the same conditions. On the other hand, when recording sleep in the wild, we are limited by the need to recapture the animal, and important confounding factors that cannot be controlled or easily quantified exist. This is the case of the social interactions, food type and availability, health condition, predation exposure, etc. Even if, ideally, it would be better to record sleep in completely free and wild animals, a high unexplained variability may exist, especially regarding the limited number of days/weeks for which small animals can be recorded. A good alternative is to record animals in semi-captive conditions in their own environment, as done here. The advantages are partial control over factors like food or social interactions coupled with high-quality recordings under reduced stress conditions. The main disadvantage is the possible stress induced by the semi-captive conditions. This approach, while still requiring some improvements, allowed us to characterize the baseline full sleep expression—in other words, the intrinsic sleep expression of wild populations. In addition, by recording animals in semi-captive conditions, we can record for longer periods of time if batteries are changed, providing insights into the interindividual variability of the sleep expression. Another important concern when recording animals in semi-captivity is the possible stress induced by limiting the freedom of the animal. Our new method, however, demonstrated that after a period of manipulation, when sleep is possibly impacted with modifications (in NREM and REM duration in particular), the stress tends to be minimal, as the animals were displaying normal behavior, including eating, exploring, and sleeping in quantities that tended to stabilize after a few days. We cannot completely reject the fact that some level of stress is present, but because our animals do not show stereotypic behaviors, like excessive grooming or repetitive biting of the mesh, nor display any lesions on the snout, this strongly suggests that our recording conditions are well-adapted to understanding the diversity of sleep expression in different rat populations. Beyond capturing the full expression of sleep in wild animals that have grown in diverse environments, recording the sleep of small animals in semi-captive conditions also opens new perspectives for testing their flexibility in response to various stressors. These include simulating the presence of a predator, altering the diet, or relocating the cage to a new territory with different pressures. This approach provides deeper insights into the adaptability of sleep.

Possible improvements

This experiment was the first one to date, allowing us to record precisely sleep in wild rats in their own environments. Indeed, almost all information collected on small wild mammals and birds was done by bringing the animal into lab conditions [2, 22, 60]. With the exception of sandpipers [18] and the small antechinus recorded in an outdoor enclosure [61], only large species (>1 kg) have been recorded in wild or semi-captive conditions [62]. Despite the fact that our new method and technology intends to fill the gap between the lab and the field in small species, some improvements could be made. To prevent intruder rats from trying to eat the food of our experimental rats, the food dispenser could be metal or installed in the center of the cage, and its capacity could be increased to avoid filling it too regularly. In addition, we lost three of the four rats after heavy rains and temperature drops. This could be prevented by installing a small roof over the shelter and by installing the shelter away from the bottom of the case, for instance, in an elevated position. We should also provide more bedding to the animal to let it construct a more insulating nest, thus reducing hypothermia risk. A more ecological solution, but logistically and practically more difficult, would be to construct larger cages, allowing the animals to have access to more natural shelters, bedding, . . . In addition, testing the reliability of our recording methods, specifically the flexible electrode, in standardized conditions in the lab, is still to be done to ensure that the signal is stable over longer durations (months). The recording of the olfactory bulb is not mandatory, as our data shows that the signature is not useful for sleep scoring in wild rats in semi-captive conditions. EOG also did not provide any consistent information and could thus be removed if the objective is not to quantify the density of eye movements during REM sleep. Alternatively, when EOG would be of interest, measuring its local activity with a differential between two electrodes over one eye instead of in-between the two eyes would provide a more accurate and stable signal. Finally, recording the ECG by connecting two small wires to our EIB could bring information related to energy expenditure or stress. Indeed, heart rate itself and heart rate variability have been used for a long time to assess these two parameters. Finally, it could be of interest to use more precise tracking methods such as RFID microchips to determine how much time the animal spends in its shelter and spends eating or foraging, as using a camera is sometimes difficult in the wild, especially for long-term recordings, though this would add to the total weight of the implant.

Conclusions

Because sleep is a vital and complex state, its ecological role remains poorly understood. It remains thus crucial to develop methods to characterize its variability and flexibility across populations and individuals under different environmental conditions. By providing a fast and minimally invasive surgical procedure and a recording technology adapted to small species, we established the proof-of-concept for sleep studies on small animals in the wild. We demonstrated that rats adapt well to the recording conditions and semi-captivity, making it feasible to assess the complete sleep phenotype in wild populations. However, significant improvements—particularly in shelter positioning—are still needed to ensure a safe and isolated sleeping site for the animal. Sleep is as important as feeding, reproduction, and avoiding predation, but its direct role in species survival remains unknown. The novel approach presented here for small species of mammals opens new perspectives in our understanding of the ecological drivers of the evolution of this enigmatic state.

Funding

The study was funded by the interdisciplinary mission of the CNRS (PEPS: life adaptation to its environment: SLEEPRED).

Disclosure statement

Financial disclosure: none. Non-financial disclosure: none.

Preprint repositories

https://ecoevorxiv.org/repository/view/7576/

Availability of data and materials

Data will be available upon request.

Author contributions

Paul-Antoine Libourel (Conceptualization [Lead], Data curation [Lead], Formal analysis [Lead], Funding acquisition [Lead], Investigation [Lead], Methodology [Lead], Project administration [Lead], Resources [Lead], Software [Lead], Supervision [Lead], Validation [Lead], Visualization [Lead], Writing—original draft [Lead], Writing—review & editing [Lead]), Sébastien Arthaud (Formal analysis [Equal], Methodology [Equal], Validation [Equal], Visualization [Equal], Writing—review & editing [Supporting]), Antoine Bergel (Investigation [Equal], Methodology [Supporting], Validation [Equal], Writing—review & editing [Equal]), Fabrice Brescia (Resources [Supporting], Writing—review & editing [Supporting]), Azzedine Dib (Methodology [Equal]), Bertrand Massot (Methodology [Equal]), Eric Vidal (Resources [Equal], Writing—review & editing [Supporting]), and Wilfreid Weiss (Methodology [Equal], Validation [Equal], Writing—review & editing [Equal])