Science - USA (2021-10-29)

the delta and theta electrical oscillations that
occur during NREM and REM sleep could be
important for rebalancing the network during
sleep ( 19 ), and such oscillations may also be
incompatible with normal cognitive function.
A good example of how oscillations can block
cognition is provided by the anesthetic keta-
mine. This compound elicits a highly selective,
delta-like rhythm (1 to 3 Hz) in layer 5 pyram-
idal neurons of the retrosplenial cortex, thus
inducing a dissociation of stimulus detection
and response ( 20 ), which is obviously inconsis-
tent with a normally functioning waking state.
In contrast to NREM sleep, REM sleep may
not be essential for life. It is possible to genet-
ically ablate all REM sleep in laboratory mice
with no apparent ill effects ( 21 ). On the other
hand, and perhaps unexpectedly, people who
report good and refreshing sleep have good
REM sleep continuity, whereas consolidated
NREM appears to be less important for the per-
ception of a good night's sleep ( 22 ). Theories for
the function of REM sleep are not constrained
by the need for the brain to be offline because
the brain is active during REM ( 5 ) and brain
temperature rises ( 13 ). It could be that because

REM sleep always follows NREM sleep, their

functions are linked. For example, REM could

be an algorithm that tests whether the restor-

ative function of NREM has been completed.

If it has, we wake up. An important question

to solve is, what determines the timing and

sequence of the NREM and REM cycles?

The homeostatic drive to sleep

In many animal species, sleep deprivation in-

creases the drive for both NREM and REM

sleep ( 2 ). This homeostatic sleep drive strength-

ens the idea that some important restorative

process is underway and is the reason that we

must“catch up”on lost sleep. This catching up,

however, is not just about quantity but also

quality, at least for NREM sleep. If we miss one

night’s sleep of 7 hours, say, our next night of

sleep is likely to be subjectively“deeper,”as well

as potentially longer. Although it remains mys-

terious what this deep sleep is providing phys-

iologically, a pervasive concept is that a reliable

marker of NREM sleep depth in both humans

and other terrestrial vertebrates is the power

exhibited in the EEG in the delta range of fre-

quencies (~0.5 to 4 Hz). The EEG delta power,

particularly the higher end of this band of fre-

quencies ( 14 ), markedly increases after sleep

deprivation but rapidly rebounds. However,

many hours more are needed for the actual

sleep loss to be restored, implying that addi-

tional, potentially restorative processes that are

not reflected in EEG delta power per se are un-

derway (Fig. 2). The high delta power immedi-

ately after sleep deprivation might nonetheless

be permissive for the restorative benefits that

follow, perhaps through brain cooling.

We can consciously fight sleepiness to a cer-

tain extent, a top-down control from the neocor-

tex imposing its will on the“sleep centers,”and

we might be determined to stay awake, but even-

tually we give way to sleep. After a certain period

of intense cortical activity, local delta oscillations

“break out”in areas of the neocortex even if ani-

mals are behaving as if they are awake ( 23 ). One

might then ask, if this restricted“sleep”is pro-

viding restorative benefits locally, then why does

the whole animal need to shut down and enter

global sleep? Why could the brain not be contin-

ually having smaller domains entering“sleep”so

that normal function is still possible at the whole-

animal level? This could be because it is necessary

to cool the brain for the restoration to occur, and

this cannot be achieved locally, only centrally. In

our view, the term“sleep”is best restricted to a

whole-animal behavior, and local short periods

of increases in delta power are clearly insuffi-

cient for whatever purpose global sleep serves.

The sleep drive can also be assessed inde-

pendently of EEG delta power by changes in

behavior. When we are strongly sleep deprived,

we become highly motivated to find a way to

sleep, in the same way that strong thirst and

hunger motivate us to drink and eat. If sleep

deprived enough, we will do almost anything to

sleep. Afternoon naps provide additional evi-

dence for an accumulating sleep drive during

the time spent awake and show that this drive

can be partially dissipated by short periods of

acute sleep. An afternoon nap delays the onset

of later NREM sleep, which starts off with a

lower EEG delta power than it would have

done had no nap been taken ( 2 ). If naps are

too long, it can be difficult to get to sleep at all

in the evening. This implies that there must be

some feedback between the process(es) that

provides restorative benefits and the mech-

anism that tracks the time awake.

Classical views of how the hunger drive is or-

ganized implicated hunger centers in the hy-

pothalamus; for the sleep drive, the preoptic

hypothalamus was shown to have a particularly

strong sleep-promoting role ( 24 ). However, it is

now clear that motivational drives for food and

water emerge in a distributed way throughout

the brain ( 25 ). Similarly, the basis for the sleep

drive is likely to be more distributed than orig-

inally anticipated (Fig. 3A). Indeed, in mice, the

circuitry that induces both NREM and REM

sleep is itself widely distributed ( 26 ). It could

SCIENCEscience.org 29 OCTOBER 2021•VOL 374 ISSUE 6567 557

Waking

Waking NREM sleep

Deterioratingbrain function

Sleeping

Time (hours)

Brain temperature

Time (minutes)

Normal

sleep onset

Sleep onset after

SD sleep deprivation (SD)

0 10 20

Frequency (Hz)

200

100

0

EEG Power

WAKE

0 10 20

Frequency (Hz)

200

100

0

EEG Power

NREM

38

37

36

35 GABAergic IPSC

Time (ms)

A

B Waking

NREM sleep

and sedation

37 ̊C 35.5 ̊C

Fig. 1. The most probable reason for sleep is to allow fundamental housekeeping processes to occur.
(A) During waking, there is a progressive deterioration of brain function until a point is reached when sleep is
triggered, consciousness is lost, and repair is initiated. This change in vigilance state is reflected in the EEG power
spectrum, which moves from a low-power broad spectrum to a high-power spectrum that peaks at delta
frequencies (0.5 to 4 Hz) and is characteristic of NREM sleep. With prolonged NREM sleep, the brain cools. If sleep
is delayed, brain deterioration continues until a point is reached when sleep onset is inevitable. (B) During
sustained NREM sleep, the brain cools by ~1.5°C, which would prolong inhibitory postsynaptic currents by about
the same extent as sedative doses of anesthetics.