well be that all parts of the sleep circuitry dis-
tributed throughout the brain can implement
homeostasis. Hypothalamic preoptic galanin
neurons ( 27 , 28 ), basal forebrain cholinergic
neurons ( 29 ), midbrain ventral tegmental area
(VTA) GABA neurons ( 30 ), raphe serotonin neu-
rons ( 31 ), neocortical layer 5 pyramidal neurons
( 10 ), and astrocytes ( 32 )havesofarbeenpro-
posed as mediators of sleep homeostasis. It
is unclear what“tiredness signals”are being
sensed by this distributed circuitry (see below).
Because NREM sleep can be dispensed with
temporarily, for example, if we stay up all night,
this implies that any restorative function of
NREM sleep need not be immediate (e.g., mice
and humans catch up on lost sleep over many
hours, if not days). Similarly, frigate birds typ-
ically sleep very little (0.7 hours/day) during
10-day periods on the wing yet sleep nearly
13 hours per day when on land ( 33 ). Thus,
whatever is being tracked during sleep depri-
vation and leading to the drive to sleep can be
partially overridden ( 34 ); therefore, this is not
the same as faster breathing and heart rate
after exertion to immediately resupply oxygen.
Hunger and thirst are also drives that can be
similarly put on hold by the executive control
centers in the brain, at least for a while. A grad-
ual process of recovering lost sleep over many
hours suggests anabolic or catabolic housekeep-
ing processes, i.e., synthesis or degradation.
A better understanding of the sleep drive
andhowtheneedforsleepis“clocked”might
reveal the functions of sleep. One way that
sleep drive might track wakefulness has been
discovered in fruit flies and involves a changing
redox state with time spent awake and a change
in the activity of potassium channels, resulting
in an increased firing rate of sleep-promoting
neurons ( 35 ). In principle, such a mechanism
could work in mammals too. In vertebrate spe-
cies, however, the actual biochemical processes
that track the time spent awake and lead to
increased NREM sleep pressure remain unclear.
The simple and compelling idea that there is a
substance that accumulates during sleep, tracks
time awake, provides feedback to induce sleep,
and is then degraded during sleep has been pro-
posed many times, starting with Rosenbaum in
1892 ( 36 ). Adenosine is the best-known candi-
date molecule: It accumulates with time spent
awake in a few select brain areas, such as the
basal forebrain ( 37 ), where it (or adenosine tris-
phosphate) could be secreted from cholinergic
neurons in the basal forebrain and astrocytes
( 29 , 38 ). In addition to its many metabolic roles,
adenosine can act through metabotropic recep-
tors to induce NREM sleep ( 36 ), but appealing
though this idea is, definitive proof that aden-
osine is the primary cause of the sleep drive is
lacking, especially because the adenosine in-
creases that correlate with prolonged waking
seem to happen in only a few brain regions, yet
the sleep homeostasis circuitry is widely distrib-
uted. Many other possible sleep-inducing factors
have been identified, including interleukin 1bandtumor necrosis factor–a, the levels of which in-
crease in the neocortex during sleep deprivation
( 6 ). The brain is not the whole story. Certainly,
peripheral tissue can signal sleep; unidentified
systemic factors from skeletal muscle contribute
to the sleep drive ( 39 ), perhaps explaining why
sleepiness can often be induced by exercise ( 2 ).
Phosphorylation tracks sleep need. In mice, a
gain-of-function mutation in the salt-inducible
kinase 3 (Sik3) gene reduces NREM sleep ho-
meostasis as defined by delta power and in-
creases the amount of sleep ( 7 ). This particular
kinase is widely expressed, including in pe-
ripheral organs, and could also act throughout
the brain, in keeping with the distributed sleep
homeostasis–promoting circuitry. It is not known
how wakefulness activates Sik3, but the targets
of this kinase are numerous. In mice, as wakeful-
ness progresses, there are steady increases in
phosphorylation (and dephosphorylation) of
many proteins in the forebrain, including ion
channels ( 40 , 41 ), which might increase or de-
crease the firing rate of sleep-promoting neu-
rons or inhibit wake-promoting neurons.
One generally ignored question is, what de-
termines when we wake up? Although this is
partly a circadian influence, waking from a re-
freshing night’s sleep suggests that the ho-
meostatic process of sleep has been completed.
Therefore, presumably, the restorative process of
sleep is continuously monitored to determine
that the brain is now ready to wake. This could
be related to the NREM-REM cycle discussed
above because we tend to wake from REM sleep.Can drugs provide the restorative
benefits of natural sleep?
The drive to sleep after a certain amount of sleep
deprivation is so intense that it resembles drug-
induced sedation.“I'm just going to give you
something to make you sleep”is an almost uni-
versally used and comforting metaphor when
we undergo an investigative procedure or are
being prepared for surgery. This is probably
more than a metaphor. Although deep anesthe-
sia suitable for surgery might be best described
as a reversible coma, lower concentrations of an-
esthetics produce a sedative state. The overall
pattern of brain activity in humans, as measured
using functional magnetic resonance imaging, is
markedly similar during NREM sleep and after
sedation with dexmedetomidine or propofol
(Fig. 3B) ( 42 ). Humans given thea2 adrenergic
agonist dexmedetomidine enter a state resem-
bling stage 3 NREM sleep, the deepest type of
sleep ( 43 ). We now know that this is because
sedatives and anesthetic agents can act in the
sleep-wake circuitry ( 18 , 44 – 46 ), especially in the
midbrain circuitry comprising the lateral haben-
ula, VTA, and hypothalamus, to mimic NREM
sleep (although there are no drugs known that
mimic REM sleep). For example, the preoptic hy-
pothalamic galanin neurons that contribute to
NREM sleep homeostasis ( 27 , 28 ) partly mediate558 29 OCTOBER 2021•VOL 374 ISSUE 6567 science.orgSCIENCE
-4-3-2-10Cumulative sleep loss (h)LPO-ΔGalLPO-Gal0 6 12 18 2400.511.522.5Normalized delta powerTime (h)Sleep
deprivationDelta power
reboundControl delta
powerSleep deprivation Catching up on lost sleepControl
EEGSD Delta reboundFig. 2. After sleep deprivation with novel objects, delta oscillations in mice rebound rapidly, but
recovering lost sleep takes much longer.After 6 hours of sleep deprivation using novel objects, lost sleep
is recovered over the following 24 hours, but the rebound in delta oscillations comes back to baseline after
only about 6 hours. Selective genetic ablation of certain neuronal populations can greatly decrease sleep
homeostatic responses. For example, if galanin neurons are selectively ablated in the lateral preoptic
hypothalamus, both measures of sleep homeostasis are significantly blunted. Graphs are replotted from data
in ( 27 ) with permission under the Creative Commons CC-BY license.
SLEEP