Wakefulness enables animals to interface adaptively with the environment. Paradoxically, in insects to humans, the efficacy of wakefulness depends on daily sleep, a mysterious, usually quiescent state of reduced environmental awareness. However, several birds fly non-stop for days, weeks or months without landing, questioning whether and how they sleep. It is commonly assumed that such birds sleep with one cerebral hemisphere at a time (i.e. unihemispherically) and with only the corresponding eye closed, as observed in swimming dolphins. However, the discovery that birds on land can perform adaptively despite sleeping very little raised the possibility that birds forgo sleep during long flights. In the first study to measure the brain state of birds during long flights, great frigatebirds (Fregata minor) slept, but only during soaring and gliding flight. Although sleep was more unihemispheric in flight than on land, sleep also occurred with both brain hemispheres, indicating that having at least one hemisphere awake is not required to maintain the aerodynamic control of flight. Nonetheless, soaring frigatebirds appeared to use unihemispheric sleep to watch where they were going while circling in rising air currents. Despite being able to engage in all types of sleep in flight, the birds only slept for 0.7 h d−1 during flights lasting up to 10 days. By contrast, once back on land they slept 12.8 h d−1. This suggests that the ecological demands for attention usually exceeded that afforded by sleeping unihemispherically. The ability to interface adaptively with the environment despite sleeping very little challenges commonly held views regarding sleep, and therefore serves as a powerful system for examining the functions of sleep and the consequences of its loss.
For over a century, people have wondered whether and how birds sleep in flight . Initially, the idea that birds might sleep on the wing stemmed from the lack of observations of certain species resting on land or water outside the breeding season. The adverse effects that sleep deprivation has on our ability to interact adaptively with the environment also probably contributed to the idea. Over time, evidence for long, non-stop flights in certain species increased and the importance of sleep across the animal kingdom became more apparent [2,3], strengthening the notion that such birds must sleep on the wing. Moreover, an explanation for how birds could (theoretically) sleep in flight was provided by the discovery that dolphins can swim while sleeping with only half their brain at a time (i.e. unihemispherically) , and our subsequent discovery that birds on land can switch from sleeping with both halves simultaneously to sleeping with only one at a time in response to increased ecological demands for wakefulness . By keeping one half of their brain awake and the corresponding eye open, flying birds could maintain aerodynamic control while watching where they are going. Collectively, this research provided such a compelling story that it is commonly assumed (or stated as an established fact) that flying birds fulfil their daily need for sleep by sleeping unihemispherically. However, evidence of long flights is not by default evidence of sleep in flight—recordings of sleep-related changes in brain activity are needed to determine whether birds sleep on the wing . Moreover, the seemingly untenable alternative—birds stay awake during long flights—was made more tenable by our recent discovery that despite sleeping very little pectoral sandpipers (Calidris melanotos) can perform adaptively under demanding real-world ecological circumstances . Consequently, until very recently, the answer to the question, do birds sleep in flight, remained up in the air.
Previously, the lack of tools to record brain activity in flying birds prevented researchers from determining whether and, if so, how birds sleep in flight. Using a data logger developed for recording brain activity in homing pigeons , we recently recorded the electroencephalograms (EEGs) of great frigatebirds (Fregata minor) during long flights . Unexpectedly, the frigatebirds slept both unihemispherically and bihemispherically in flight. In addition, as anticipated by the study on pectoral sandpipers, and contrary to prior expectations, frigatebirds slept remarkably little on the wing. Consequently, it is possible that other birds assumed to sleep in flight actually greatly reduce the time spent sleeping, or dispense with it altogether, during long flights. In this article, I summarize the evidence for non-stop flights in birds and sleep in flying frigatebirds. I also provide perspectives on future research, including the novel opportunity to gain insight into the functions of sleep provided by flight-related sleep loss in birds.
2. Non-stop fliers
The number of species thought to engage in non-stop flights is growing rapidly as the use of small tracking devices increases. Satellite and GPS tags are shrinking in size and weight making it possible to track smaller species with this technology . Accelerometers are providing insight into the flight behaviour of both large [8,10] and small birds [11,12]. Geolocators, small devices that derive a bird's location on the globe by detecting changes in the timing of sunrise and sunset, are detecting long-distance flights in birds as small as 12 g . Although geolocators do not provide a direct measure of non-stop flight, when combined with other lines of evidence, including energy stores, flight speed, wind conditions and the presence of ecological barriers thought to preclude landing (e.g. oceans), continuous flight can be inferred. In addition, in songbirds, variability in daytime light levels has been used to estimate when the birds are flying (low variability) and when they are on land foraging in foliage (high variability) [14,15]. The interpretation of flight behaviour from geolocator data can be strengthened further by the addition of devices that detect water immersion  or flapping (accelerometry) [11,12,17]. In this section, I review the evidence for continuous flight (more than 24 h) in birds.
2.1. Anseriformes (waterfowl)
Among waterfowl, the longest non-stop flight occurs in the Pacific black brant (Branta bernicula nigricans). After staging in Izembek Lagoon in the Alaskan Aleutian islands most of the population flies south to Baja California, Mexico for the winter . Although migration data on individual birds have not been reported, most of the population departs from Alaska under favourable wind conditions and arrives in Mexico en masse. By tracking departure and arrival times, this migratory flight spanning 5300 km has been estimated to last on average 2.3 days. Although brant can stop and rest or sleep on the ocean, the known flight speed and wind conditions during migration, suggest that this flight is performed without stopping. Optimizing the use of favourable winds and the absence of suitable foraging conditions at sea may make stopping disadvantageous.
2.2. Apodiformes (swifts and hummingbirds)
As early as the 1800s, people have speculated whether common swifts (Apus apus) fly throughout the day and night . Initially, this was based on observations of swifts ascending into the sky at dusk and descending in the morning . Also, other than breeding adults, who usually spend the night in the nest, swifts are rarely observed on land or in vegetation during the day or night . Indeed, the few observations of swifts roosting terrestrially are exceptional and typically involve recently fledged juveniles during periods of unusually poor weather conditions . In addition to this indirect evidence, swifts have been detected flying at night via direct observations, radar and radio telemetry.
Common swifts flying at night have been observed from airplanes on at least two occasions. During the First World War a French pilot gliding behind enemy lines at 3000 m with the engine off observed the following: ‘ … we suddenly found ourselves among a strange flight of birds which seemed to be motionless, or at least showed no noticeable reaction.’ (; translated from French in ). Two of the birds hit the airplane and one that landed in the cockpit was identified as a common swift. Based on this encounter, Guérin questioned whether swifts sleep in flight. Weitnauer  followed up on this account, and was also able to observe swifts flying at night from an airplane. These observations clearly demonstrated that some common swifts spend at least part of the night in flight. Also, the observation that the birds seemed unresponsive to and collided with the airplane suggests that they might have been sleeping. However, a large flying object at night may have been so unnatural that even an awake swift might not have responded to the sight of the airplane in a manner detectable by the pilot. For example, animals living on islands without natural predators do not exhibit evasive behaviours when first encountering introduced predators. Furthermore, the collisions may have resulted from the airplane having a higher airspeed than the swifts. Finally, neither observer reported whether the birds' eyes were closed, probably due to the challenges of observing eye state under these conditions. Consequently, these intriguing observations alone do not establish that the swifts were sleeping in flight.
In addition to direct observations, the nocturnal flight behaviour of common swifts has been investigated with radar. Bruderer & Weitnauer  demonstrated that swifts flying at night alternate between periods of flapping (6–8 Hz) lasting 1–6 s and gliding lasting 0.5–5 s. More recently, Bäckman & Alerstam  tracked individual swifts for up to 1 h and observed that they often showed a harmonic oscillation (period 1–16 min) in their orientation relative to the wind direction. Interestingly, there was no relationship between wind speed and the period of this oscillation, suggesting that it does not reflect a strategy to maintain their position over the colony. Instead, the authors speculate that the swifts might drift off the optimal course when they drift to sleep.
The altitude of common swifts flying at night has also been studied with radar. Interestingly, around dusk swifts ascend to 2.5 km and then gradually descend during the middle of the night . Alone this might suggest that high altitudes are attained to provide an altitudinal buffer for the safe occurrence of sleep. However, swifts also return to similarly high altitudes around dawn suggesting that ascents serve other functions, such as assessing approaching weather conditions .
Although these radar studies clearly demonstrate that swifts are flying at all hours of the night, individual swifts have not been tracked throughout the night using this method. By contrast, using radio tags, Tarburton & Kaiser  were able to follow individual swifts and confirmed that they do fly throughout the night. Remarkably, even juveniles spend their first night outside the nest-box on the wing.
Collectively, the direct observations and the radar and radio telemetry studies demonstrate that common swifts usually spend the summer nights flying when not raising young. Moreover, the absence of observations of swifts roosting terrestrially during migration or while wintering in sub-Saharan Africa  raised speculation as to whether they fly non-stop during this time [20,28].
Flight behaviour during the non-breeding season has been examined in Alpine swifts (Tachymarptis melba), and more recently in common swifts. Unlike common swifts, Alpine swifts are not known to spend the night flying on their breeding grounds. Nonetheless, a study using combined geolocator–accelerometer tags suggests that they fly continuously for up to 200 days during migration and their winter in sub-Saharan Africa . The accelerometer measured activity and pitch at 4-min intervals. Periods with low activity and pitch likely corresponding to the bird hanging vertically and, possibly, sleeping, occurred at night during the breeding season and portions of the migration (figure 1). During the rest of the time, the birds were in a horizontal position consistent with flight. Compared with the night during the breeding season, activity was higher during the day and night at this time. Nonetheless, there were periods of relatively low activity at night, especially during the second half of the winter. Assuming that the birds never spent the night horizontal on land, these periods of low activity might reflect an increase in the time spent soaring and gliding, perhaps related to seasonal changes in atmospheric conditions. In a similar study, Hedenström et al.  found that common swifts fly throughout more than 99% of the non-breeding period, with some individuals remaining in flight for up to 300 days. Interestingly, during the rare periods when the birds landed, they roosted vertically or horizontally. It is unclear whether the presence of horizontal roosting in common swifts, but not Alpine swifts, reflects a real species difference or a difference in data acquisition, analysis and interpretation. As such, a comparison of the two swifts using the same methods would be informative. Finally, although these studies provide evidence for extended periods of flight in Alpine and common swifts, as with similar studies demonstrating non-stop flights in other species, they provide no information on whether swifts sleep in flight.
2.2.2. Ruby-throated hummingbird
The hummingbirds of western North America migrate to and from their southern winter range over land . By contrast, in the only eastern species—the ruby-throated hummingbird (Archilochus colubris)—at least some individuals are thought to migrate across the Gulf of Mexico [30,31]. This is based on observations of birds in the spring arriving on the northern coast from over the water  and estimates of flight duration derived from fat levels in birds caught on the northern coast in the autumn . Based on their flight speed, ruby-throated hummingbirds are thought to make the trans-Gulf flight in approximately 20 h . However, studies of radio tagged birds are needed to confirm that they make such flights and to determine their duration .
2.3. Charadriiformes (shorebirds and terns)
2.3.1. Shorebirds (sandpipers and plovers)
Many shorebirds engage in long, multiday flights during migration . Pacific golden-plovers (Pluvialis fulva) make 4-day, 4900 km flights from Alaska to Hawaii without touching the Pacific Ocean . Red knots (Calidris canutus rufa) make flights across the Atlantic Ocean between South and North America lasting up to 6 days and spanning 8000 km . However, the current record holder is a sub-species of the bar-tailed godwit (Limosa lapponica baueri) which migrates from Alaska to New Zealand, a flight spanning 11 690 km and lasting 8.1 days [35,36] (figure 2). In addition, their return migration to Alaska is completed in two long flights via a stopover in China.
Although many long flights involve crossing large bodies of water, some shorebirds appear to fly non-stop, even when passing over suitable habitat. During the autumn, great snipe (Gallinago media) migrate from Sweden to sub-Saharan Africa in one non-stop flight spanning 4300–6800 km and lasting 2–4 days . In addition to flying over unsuitable habitats (Mediterranean Sea and Sahara Desert), this flight includes passage over suitable habitats in Europe. Similarly, ruddy turnstones (Arenaria interpres) flying for up to 7600 km and 7 days pass over suitable habitat on the northern coast of Australia when migrating from southern Australia to Taiwan . Consequently, long-distance flights are not exclusively utilized to pass ecological barriers in shorebirds.
Outside the breeding season sooty terns (Sterna fuscata) are thought to fly non-stop, leading Ashmole  to suggest that they sleep in flight. The evidence for continuous flight is based primarily on the fact that sooty terns are rarely observed floating on the water . Also, probably due to the low lipid content of their preen oil , sooty terns quickly become waterlogged when placed in water , making taking off difficult . Although these observations suggest that sooty terns usually avoid landing on the water, direct measurements are needed to confirm that they do not rest on the surface, particularly at night when they are difficult to observe. Finally, despite often being touted as a bird known to sleep in flight, it is unknown whether they sleep on the wing.
The Arctic tern (Sterna paradisaea) is well known for its nearly pole-to-pole migrations, and therefore might be a candidate for sleep in flight. However, based on the terns' known flight speed and the average daily distance travelled, as determined with geolocators [44,45], the birds have more than 7 h per day that could be used for sleeping on the ocean surface. Whether they actually spend this time on the ocean surface is unknown. Additional monitoring methods, such as water immersion loggers , are needed to further clarify the behaviour of migrating Arctic terns.
2.4. Pelecaniformes (frigatebirds)
Frigatebirds are large tropical seabirds with the lowest wing loading among birds . Early explorers, such as Burton  and later Darwin , noted that frigatebirds never rest on the water. Walt Whitman even wrote a poem about frigatebirds in which he envisioned them sleeping in flight (Thou who hast slept all night upon the storm, 1878). Frigatebirds are thought to avoid landing on the water because limited feather waterproofing and their small poorly webbed feet make taking off difficult or impossible following more than momentary contact . Frigatebirds obtain most of their food by catching flying fish and squid when they jump out of the water to evade predatory fish and cetaceans . Owing to the patchy and unpredictable distribution of this resource, frigatebirds spend long periods searching the ocean for food . Interestingly, when incubating eggs or brooding small chicks at least one parent has to be on the nest 24/7 to prevent predation from other frigatebirds. As a result, the parents alternate between periods on the nest and periods foraging at sea.
Flight behaviour has been studied in two of the five species of frigatebirds. Flights lasting 3.9 days were confirmed in the magnificent frigatebird (Fregata magnificens) using altimeters . In great frigatebirds (Fregata minor) incubating eggs on Europa Island in the Mozambique Channel, flights last 3–12 days , whereas when brooding chicks flights last 1–3 days. By contrast, in non-breeding adults and independent juveniles from the same island, flights can last up to 2.1 months as they circle the Indian Ocean  (figure 3). During flight, frigatebirds gain altitude by soaring in rising air currents under cumulus clouds and then glide down to the next cloud. By relying on this soaring–gliding strategy, frigatebirds minimize the need for flapping and thereby reduce the energetic costs of flight. Although these studies clearly show that frigatebirds engage in remarkable transoceanic flights, they do not establish that frigatebirds sleep in flight. The first evidence for sleep in flying frigatebirds will be discussed below.
2.5. Falconiformes (hawks and falcons)
Hawks and falcons engage in long-distance migrations between their breeding and winter range. During migration, they rely on a soaring–gliding strategy to conserve energy. As a result, they are thought to avoid crossing large bodies of water where thermals are weaker and less reliable. While over land, hawks and falcons also tend to migrate more during the day when thermals are the strongest . Consequently, for most hawks and falcons, there may be plenty of opportunities to sleep at night while on land. Nonetheless, satellite tracking has revealed that two falcon species are known to engage in long over water flights. Eleonora's falcon (Falco eleonorae) crosses the Mozambique Channel between Africa and Madagascar, a flight spanning more than 400 km and lasting 24 h . The Amur falcon (Falco amurensis) flies from the southwest coast of India to the Somalian coast of Africa, a transoceanic flight spanning up to 5600 km and lasting up to 5.4 days . Interestingly, dragonflies (Pantala flavescens), a major component of the falcon's diet, are thought to take the same non-stop migration route at the same time of year  raising the possibility that Amur falcons are able to feed during migration over the ocean. It is unknown whether Amur falcons or the dragonflies sleep during these flights.
2.6. Passeriformes (songbirds and swallows)
Many passerines thought to sleep at night during the non-migratory seasons fly at night during their spring and autumn migrations . When passing over suitable habitat, songbirds land each day to forage. Although sleep has not been recorded in the wild, two studies which examined sleep in songbirds exhibiting migratory behaviour in captivity suggest that they forgo large amounts of sleep while migrating at night, but may recover at least some of the lost sleep while on land during the day. At the time of year when their wild conspecifics are migrating, captive songbirds exhibit migratory restlessness, or Zugunruhe, consisting of hopping around their cage and fluttering their wings at night . This behaviour is thought to reflect the birds' endogenous urge to migrate at certain times of the year , and therefore provides an opportunity to examine seasonal changes in sleep in captivity. When compared with the non-migratory season, white-crowned sparrows (Zonotrichia leucophrys gambelii) and Swainson's thrushes (Catharus ustulatus) exhibiting migratory restlessness reduced the time spent sleeping at night by two-thirds [60–62]. While exhibiting migratory restlessness no behavioural (e.g. eye closure) or EEG signs of sleep were observed. Despite sleeping less at night, both species spent more time drowsy or napping in the day, suggesting that they were compensating in part for sleep lost at night. If these sleep patterns exhibited in captivity reflect those occurring in birds migrating in the wild , songbirds may not need to sleep in flight given the daily opportunity to sleep on land.
In contrast with passage over suitable habitat, several recent studies have shown that some songbirds engage in flights lasting up to several days when crossing ecological barriers. Many North American songbirds engage in non-stop flights when crossing the Gulf of Mexico. During their autumn migration, Swainson's and Wood thrushes (Hylocichla mustelina) fly south from the Alabama coast to the Yucatan peninsula in Mexico, a flight spanning > 1000 km of water and lasting just over 24 h . Although some red-eyed vireos (Vireo olivaceus) engage in similar flights, most take a longer, apparently largely land-based route to the Yucatan. Tree swallows (Tachycineta bicolor) make similar autumn migratory flights across the Gulf lasting 12–36 h . Interestingly, instead of crossing the Gulf of Mexico, blackpoll warblers (Setophaga striata) take an even longer over water route on their way to South America . After departing from the northeast coast of North America they fly non-stop to the Caribbean Islands, a flight spanning up to 2770 km and lasting up to 3 days  (figure 4). By heading southeast over the Atlantic Ocean, blackpoll warblers catch westerly trade winds that aid them in reaching the Caribbean. In doing so, they bypass large portions of the land-based route through the United States that they used in the spring. Interestingly, species closely related to blackpoll warblers sharing similar breeding and winter ranges seem to follow the more land-based route during the spring and autumn migration. Like blackpoll warblers, northern wheatears (Oenanthe oenanthe) migrating from northeastern Canada to northwestern Africa also appear to cross large portions of the Atlantic Ocean. However, whereas modelling suggests that wheatears could make the non-stop 4000–5000 km flight in 31–68 h , geolocator tracking data from one bird suggests that the flights are made with a stop in Europe and possibly Greenland . Finally, several European songbirds engage in non-stop flights lasting 1–3 days that include crossing the Mediterranean Sea and Sahara Desert [14,15].
Although nothing is known about the sleep patterns of songbirds migrating in the wild, a couple observations warrant consideration. First, birds migrating at night often collide with man-made structures . However, this appears to reflect an attraction to light under cloudy conditions, rather than reduced environmental awareness related to sleep. Second, songbirds migrating at night emit flight calls and presumably listen for those emitted by other birds . Although this suggests that the birds are awake, it is unknown how often individual birds call. Consequently, it is conceivable that they sleep between calling and listen for calls while sleeping unihemispherically.
For centuries, people have stated that albatrosses sleep in flight. This idea, perpetuated in the popular press, presumably derives from the fact that they cross vast expanses of the ocean, as confirmed by several tracking studies [71,72]. However, tracking studies have also shown that unlike frigatebirds, several species of albatross typically stop and float on the water for several hours each night . As albatrosses usually do not feed at night when on the surface [74–76], they might use this time to sleep. As long as rough seas do not interfere with sleep, albatrosses may therefore have little need for sleep in flight.
2.8. Extinct birds and their relatives
2.8.1. Pelagornis sandersi
Pelagornis sandersi is an extinct pelagornithid species with the largest known wingspan (6.4 m) among birds. Although P. sandersi is thought to be most closely related to waterfowl, the shape of its wings resembles that of albatrosses, suggesting that it flew in a similar manner . This and the presence of tooth-like structures on its bill suggest that it fed on fish during long-range flights over the ocean. However, it is unclear whether P. sandersi could take off from the water . Consequently, while at sea it may have flown continuously catching fish near the surface, as in frigatebirds.
Pterosaurs were volant archosaurs closely related to dinosaurs, including birds. Although many pterosaurs are thought to have soared over the ocean in search of fish, recent computer modelling suggests that some were unable to take off from the surface . Consequently, if they did not return to land on a daily basis, they may have resorted to sleeping in flight .
Although the strength of the evidence varies across the species examined (based in part on the technology available for use in birds of a particular size), species from six avian orders are thought to engage in long, non-stop flights. In some groups, such as shorebirds, long flights appear to be common, whereas in others only one or a few species are known to engage in long flights (falcons). In many species, long flights only occur during migration (brant, shorebirds, falcons and songbirds), whereas in others long flights are a regular part of life in juveniles or non-breeding adults (common and alpine swifts, great frigatebirds and possibly sooty terns). Even while breeding, frigatebirds alternate between long periods on the nest and long periods in flight. Some long flights are used to traverse unsuitable habitats, such as large bodies of water in species thought to be incapable of safely landing on the surface (songbirds and possibly hummingbirds). Even in species that regularly float on the water for extended periods of time, such as brant, landing on the ocean may be avoided due to the lack of suitable food. In this case, birds may fly non-stop, not because they cannot land, but simply because landing is energetically unprofitable. Although many species do not feed during long flights, and therefore need to accumulate sufficient energy stores prior to departing, some can feed during long flights (common and alpine swifts, frigatebirds, sooty terns and possibly Amur falcons). Finally, some species even bypass (blackpoll warbler) or fly over (some sandpipers) seemingly suitable habits during migration, perhaps reflecting benefits of arriving quickly on the wintering or breeding grounds.
3. How might birds sleep in flight?
Assessing whether and how birds might sleep in flight is complicated by the fact that birds exhibit more than one type of sleep. As in mammals, avian sleep consists of two distinct types of sleep, slow wave sleep (SWS, or non-REM sleep) and rapid eye movement (REM) sleep . The two states are distinguished from one another and wakefulness in part by changes in brain activity measured with the EEG. During wakefulness, the EEG shows low-amplitude, high-frequency activity, whereas during SWS the EEG is dominated by high-amplitude slow waves. The level of slow waves (0.75–4.5 Hz) is typically quantified using a fast Fourier transform algorithm, and expressed as slow wave activity (SWA). As in mammals, SWA during SWS increases with prior time spent awake and decreases with time spent asleep, and is therefore thought to reflect the intensity of homeostatically regulated sleep processes [80,81].
The depth of SWS (based on SWA) often varies between the two hemispheres. Using criteria established in northern fur seals (Callorhinus ursinus) , SWS can be categorized as bihemispheric SWS (BSWS) or asymmetric SWS (ASWS) based on the degree of interhemispheric asymmetry in SWA. ASWS can be further categorized as unihemispheric SWS (USWS) when the asymmetry is particularly pronounced . During ASWS, including USWS, the eye opposite to the hemisphere that is sleeping more deeply is usually closed; whereas the other eye is usually open [5,62,83] (figure 5a). Importantly, birds have the capacity to switch between BSWS and ASWS in response to changing ecological demands. Mallards (Anas platyrhynchos) sleeping in a group on land (figure 5b) switch from BSWS when safely flanked by other birds to ASWS when positioned at the edge of the group . During ASWS at the edge, mallards direct the open eye away from the other birds, as if watching for approaching threats (figure 5c). Moreover, when threatening visual stimuli are presented to the open eye, mallards rapidly initiate escape behaviours. Collectively, this demonstrates that mallards, and perhaps other birds, can resort to ASWS in response to ecological circumstances that require increased environmental awareness.
REM sleep is characterized by EEG activity remarkably similar to that occurring during wakefulness . However, in contrast with wakefulness, the bird has its eyes closed and muscle tone reaches its lowest level, often causing the head to drop . Thermoregulatory behaviours, such as shivering, which can occur during SWS, are suppressed during REM sleep, in part due to the reduction in muscle tone . Despite this reduction in muscle tone, birds are able to maintain their balance while standing during REM sleep . By contrast, mammals capable of standing during BSWS, such as ungulates, have to lay down for REM sleep . As in mammals, REM sleep is associated with REMs and other forms of twitching. Unlike episodes of SWS, which can last several minutes, episodes of REM sleep typically do not last longer than 10 s in birds. Also in contrast with SWS, REM sleep always involves both cerebral hemispheres.
It is commonly assumed that birds sleep unihemispherically during long flights. As in swimming dolphins, by keeping one half of the brain awake and the opposite eye open, birds could maintain aerodynamic control of flight and visually monitor where they are going. However, as previously suggested , BSWS, or even REM sleep, might also be possible during flight, as birds are able to balance while standing in these states. This would simply require an ability to hold the wings in an aerodynamic soaring or gliding position. Also, closure of both eyes during BSWS might not pose a large problem, if the episodes are sufficiently short, as is already the case for episodes of REM sleep in birds sleeping on land. Flapping flight might also be possible, at least during USWS, as sleep in one hemisphere does not appear to interfere with a dolphin's ability to swim in a symmetrical manner. However, it seems less likely that birds could engage in REM sleep while flapping as other motor behaviours, such as shivering, are suspended during REM sleep. In this case, birds could simply glide during the several seconds it takes to complete an episode of REM sleep. Finally, in birds that fly in groups, the flight formation may also influence whether sleep is possible in flight. For example, if flapping in V-formations requires full attention to maximize the aerodynamic benefits  and avoid collisions, then even USWS might not be possible in birds that rely on this flight strategy.
Although sleep in flight seems feasible, the assumption that birds must sleep during long flights was recently challenged by our discovery that at least some birds can function adaptively on very little sleep . Pectoral sandpipers are a polygynous species that breeds under the constant light of the Arctic summer. During an intense three-week period of male–male competition for territories and females, some males sleep very little. Maintaining a territory and convincing choosey females to mate with them requires defending their territory from other males via displays and fights, attracting females via courtship displays, remaining vigilant for intruding males, available females and predators (figure 6a–f), and foraging to obtain the energy needed to support these behaviours. Interestingly, the time spent sleeping varied considerably across males participating in this competition. In the most extreme case, a male was awake more than 95% of the time for 19 days straight. Surprisingly, paternity testing showed that the males who slept the least were more likely to sire young than longer sleeping males (figure 6g). As siring young integrates performance on a variety of tasks, this study demonstrated that reduced performance is not an evolutionarily inescapable outcome of sleep loss. It also raised the possibility that birds engaging in long flights might also forgo large amounts of sleep, or sleep altogether.
4. Sleep in flying frigatebirds
We recently conducted the first study of sleep in flying great frigatebirds . This species is ideal for investigating sleep in flight for several reasons. First, the evidence for long, non-stop flights is particularly strong in great frigatebirds [10,50]. Second, unlike the other birds thought to engage in long flights, frigatebirds are large enough to carry a data logger on their head; albatrosses are larger, but as noted above there is good reason to think that they sleep on the water at night. Third, unlike migratory birds which are difficult to re-locate, breeding frigatebirds return to their nest after long foraging trips, facilitating the retrieval of data loggers. Fourth, frigatebirds primarily rely on a soaring–gliding strategy, which reduces the chances of detecting EEG artefacts related to flapping and, arguably, may be more conducive to sleeping on the wing.
We instrumented 15 female great frigatebirds caring for small chicks on the Genovesa Island, in the Galápagos (figure 7a). Sensors were surgically implanted to record the EEG from both hemispheres. A data logger glued to the head recorded the EEGs and triaxial acceleration of the head for up to 10 days. A GPS data logger taped to the feathers on the back recorded the birds' position and altitude. Following recovery on the nest, the birds departed on foraging trips. After returning to the nest and completing a period of post-flight recovery, we removed the loggers and returned the birds to their chick. Data were retrieved from 14 of the 15 birds.
The foraging trips lasted on average 5.8 days and 1988 km, with the longest encompassing 10 days and 3000 km (figure 7b). During daytime flight, the EEG usually showed waking activity in both hemispheres. However, shortly after sunset, this pattern was replaced by periods with high-amplitude slow waves while soaring or gliding similar to those observed during SWS on land. Interestingly, although SWS was usually ASWS or USWS, as commonly predicted, it sometimes occurred bihemispherically in flight. BSWS did not appear to adversely affect the birds' flight performance; i.e. they did not drop from the sky. This demonstrates that having one hemisphere awake is not essential for maintaining the aerodynamic control of flight. Nonetheless, when compared with sleep on land, a greater proportion of SWS was ASWS or USWS in flight. A careful analysis of the accelerometry data revealed clues to the function of sleeping in this manner. Sleep usually occurred while the birds were soaring in a circle on rising air currents, and never while flapping. Interestingly, there was a strong relationship between the turning direction and the interhemispheric asymmetry in sleep intensity based on EEG SWA. When circling to the left, the left hemisphere was usually sleeping deeper, and when circling to the right the right hemisphere was usually sleeping deeper (figure 8a,b). In addition, we detected an inverse relationship between the circling direction and the interhemispheric asymmetry in EEG gamma activity (30–80 Hz; figure 8c)—a frequency implicated in attention and visual processing . Collectively, this suggests that the eye facing the direction of the turn and connected to the awake hemisphere was open and watching where the bird was going (figure 8d).
We currently do not know exactly why frigatebirds watch where they are going. Mallards sleeping at the edge of a group on land direct the open eye away from other birds, apparently to keep an eye out for approaching predators . However, frigatebirds have no aerial predators. Colliding with the sea might pose a problem for frigatebirds, but sleep usually occurs more than 150 m above the surface and while ascending. Although frigatebirds might watch where they are going to avoid collisions with other frigatebirds using the same air mass (particularly those circling in the opposite direction), the density of birds is thought to be low far from land. Another possibility is that they are monitoring navigational cues or oceanic conditions .
Contrary to most expectations, we also detected brief (less than 5 s) episodes of REM sleep in flight. As on land, REM sleep was characterized by wake-like EEG activity in both hemispheres and, as revealed by the accelerometer recordings, dropping of the head and twitching. Episodes of REM sleep lasted 4.9 s in flight and 5.9 s on land. REM sleep as a per cent of sleep time was also lower in flight than on land (3.5 versus 8.2%). Other than the head drop, we did not observe any other changes in flight behaviour. This is similar to the situation in birds that can stand during REM sleep on land; even when the head drops, indicating reduced neck muscle tone, the birds remain standing .
In addition to sleeping more asymmetrically, there were other pronounced differences between sleep in flight and sleep on land. In flight, sleep occurred primarily during the first few hours of the night, whereas on land it occurred throughout the day and night. Episodes of sleep lasted 12 s in flight, but 52 s on land; the longest episode of sleep was 348 s in flight and 1212 s on land. Perhaps, the most remarkable difference was that despite being able to engage in all types of sleep in flight, frigatebirds slept for only 0.7 h d−1 in flight, whereas after returning to land they slept for 12.8 h d−1. Even at night when 80.5% of the time was spent soaring or gliding and sleep was possible, 79.7% of the time was spent fully awake. Despite sleeping much less in flight, the intensity of SWS based on EEG SWA was lower during all forms of SWS in flight, even towards the end of the flight when the homeostatic pressure for sleep should have been the greatest. This indicates that frigatebirds are able to forestall the recovery of lost sleep (even while sleeping in flight) until they return to land. Once back on land, sleep intensity was the greatest shortly after landing and gradually declined thereafter, suggesting that homeostatic processes proceeded unhindered on land. However, due to limits on the recording duration, we are currently unable to determine whether frigatebirds compensate fully for sleep lost in flight by sleeping deeper and longer on land.
The low amount of sleep in flight suggests that the ecological demands for attention usually exceed that provided by sleeping unihemispherically. In this regard, in future studies, it would be interesting to assess how unihemispheric sleep affects the birds' ability to respond to environmental factors. It will also be important to determine what the frigatebirds are actually paying attention to while flying at night. It is known that frigatebirds follow ocean eddies predictive of favourable foraging conditions during the night . Even though they are thought to rarely feed at night, this may ensure that they are near foraging sites at daybreak, and under favourable conditions (calm sea and full moon) it might also allow them to feed at night . The low amount of sleep at night suggests that even unihemispheric sleep has an adverse impact on the sensory systems used to follow eddies, and perhaps perform other tasks.
The pressure to find food, and therefore stay fully awake, may be particularly strong when frigatebirds are raising chicks, as they need to find sufficient food to feed the chick and sustain themselves during their turn fasting on the nest. As the time available for sleep while confined to the nest is great, this may also contribute to the largely sleepless strategy employed while at sea; i.e. frigatebirds can afford to forgo large amounts of sleep when foraging, because they will have many opportunities to sleep when back on land. If the low amount of sleep in flight is linked to foraging demands, then outside the breeding season, when flights can last over two months , frigatebirds might sleep more during flight as they only need to forage to sustain themselves. The lack of opportunities to sleep on land would also favour more sleep in flight. Testing these predictions in non-breeding birds will be challenging given the duration of the flights and the need to re-locate the birds.
5. Implications for other birds
Obviously, the extent to which the findings in great frigatebirds extend to other species is unknown. Nonetheless, in addition to demonstrating that all types of sleep are possible during soaring–gliding flight, the frigatebirds also revealed that forgoing large amounts of sleep is also possible, a finding anticipated by the discovery of adaptive sleeplessness in pectoral sandpipers . Observations of other species sleeping after completing migratory flights suggest that, as in frigatebirds, they might be compensating for sleep lost during flight . As in frigatebirds, investment in sleep during flight is likely to be determined by the poorly understood ecological demands for wakefulness during flight. In addition, if for some reason sleep is not possible during flapping flight then species that primarily rely on this flight mode (e.g. brant, shorebirds, and songbirds) might not sleep on the wing. Based on these considerations, it seems likely that at least some of the other species that engage in long flights also forgo large amounts of sleep or sleep entirely.
6. General implications
In great frigatebirds, their ability to perform adaptively on little sleep is arguably more remarkable than their ability to sleep in flight. Indeed, even when sleeping in flight, sleep was more fragmented and less deep than sleep on land. In contrast with frigatebirds, in animals ranging from insects  to humans , far shorter periods of sleep restriction and fragmentation  have profound adverse effects on their ability to maintain wakefulness and interact adaptively with the environment. People fall asleep while driving a car after losing just a few hours of sleep, even when aware of the dangers and trying to keep themselves awake. Similarly, pigeons (Columba livia) forced to stay awake under laboratory conditions require nearly continuous stimulation to keep them awake after losing just a few hours of sleep . Currently, it is unknown how frigatebirds sustain adaptive performance despite sleeping very little. Real-world ecological demands and associated motivational states might play a role [96,97]. Indeed, sleep duration in the wild appears to be far more ecologically flexible than commonly recognized from laboratory studies [98–100]. Natural selection in great frigatebirds and sexual selection in pectoral sandpipers  may have also favoured an ability to resist the adverse effects of sleep loss under challenging ecological circumstances. Nonetheless, in both frigatebirds and pectoral sandpipers small amounts of sleep persisted, suggesting that there may be limits on the degree to which animals can curtail sleep. Ultimately, determining how frigatebirds are able to forestall most sleep until returning to land could enhance our understanding of the processes contributing to reduced performance in other animals, including ourselves.
Although this may seem far-fetched to researchers studying sleep in humans or their animal models, a prime example of how comparative work with an ecological perspective can influence our understanding of sleep in humans was recently published. It is well known that people tend to sleep poorly in a new environment. In laboratory studies, this ‘first night effect’ is so prominent that data from the first night is usually ignored. However, based on the observation that mallards shift to sleeping asymmetrically when exposed to increased predation risk , Tamaki and co-workers  wondered whether a similar phenomenon might account for the first night effect in humans. Although we do not engage in unihemispheric sleep, the researchers discovered something similar. On the first night, sleep intensity based on SWA and responsiveness to sounds was lower in the left than the right hemisphere. On the second night, when the subjects slept better, the asymmetry was gone. Apparently, as in mallards, we have an ability to modulate sleep intensity in different parts of the brain in response to changing ecological circumstances. Presumably, this is an adaptive response that allows us to monitor risk when sleeping in a new environment. More generally, these findings demonstrate how basic, ecologically motivated research, seemingly irrelevant to humans, can influence our understanding of human sleep. Consequently, investigations into flight-related sleeplessness may also lead to new perspectives on sleep in humans and sleep in general.
I declare I have no competing interests.
This work was supported by the Max Planck Society.
I thank Dolores Martinez-Gonzalez and Bart Kempenaers for thoughtful comments on the manuscript.
One contribution of 19 to a theme issue ‘Coevolving advances in animal flight and aerial robotics’.
- © 2016 The Author(s)
Published by the Royal Society. All rights reserved.