Sleep Medicine Reviews (2006) 10, 49–62

www.elsevier.com/locate/smrv

THEORETICAL REVIEW

Sleep function and synaptic homeostasis

Giulio Tononi*, Chiara Cirelli

Department of Psychiatry, University of Wisconsin, 6001 Research Park Blvd., Madison, WI 53719, USA

KEYWORDS Summary This paper reviews a novel hypothesis about the functions of slow wave
Long-term depression; sleep—the synaptic homeostasis hypothesis. According to the hypothesis, plastic
Synaptic scaling; processes occurring during wakefulness result in a net increase in synaptic strength in
Learning; many brain circuits. The role of sleep is to downscale synaptic strength to a baseline
Consolidation; level that is energetically sustainable, makes efficient use of gray matter space, and
Delta sleep; is beneficial for learning and memory. Thus, sleep is the price we have to pay for
Slow waves; plasticity, and its goal is the homeostatic regulation of the total synaptic weight
Slow oscillation impinging on neurons. The hypothesis accounts for a large number of experimental
facts, makes several specific predictions, and has implications for both sleep and
mood disorders.
Q 2005 Elsevier Ltd. All rights reserved.

Introduction best-established models of sleep regulation—the

two-process model. 2 The model distinguishes
This paper discusses a novel hypothesis—the synap- between the circadian and the homeostatic regu-
tic homeostasis hypothesis—which claims that lation of sleep propensity. The circadian com-
sleep plays a role in the regulation of synaptic ponent (process C) describes how sleep propensity
weight in the brain.1 The synaptic homeostasis changes during the 24 h. Process C is well under-
hypothesis can account for several aspects of sleep stood, both in its mechanisms, centered in the
and its regulation, and makes several specific suprachiasmatic nucleus, and in its function, which
predictions. In brief, the hypothesis is as follows: is to restrict sleep to a time of day that is
(1) Wakefulness is associated with synaptic poten- ecologically appropriate. The homeostatic com-
tiation in several cortical circuits; (2) Synaptic ponent (Process S) accumulates exponentially
potentiation is tied to the homeostatic regulation during wakefulness and is discharged when we
of slow wave activity; (3) Slow wave activity is sleep, also exponentially but with a faster time
associated with synaptic downscaling; (4) Synaptic course (Fig. 1). The time course of Process S was
downscaling is tied to the beneficial effects of sleep derived from a physiological variable, EEG slow-
on neural function and, indirectly, on performance. wave activity (SWA) in the electroencephalogram
A useful way of introducing the synaptic homeo- (EEG) of non rapid eye movement (NREM) sleep. The
stasis hypothesis is to relate it to one of the homeostatic regulation of SWA suggests that it may
reflect some restorative aspect of sleep, but what
* Corresponding author. Tel.: C1 608 263 6063; fax: C1 608 this aspect may be remains unknown.
263 9340. According to the present hypothesis, Process S
E-mail address: [email protected] (G. Tononi). describes the process of synaptic homeostasis.
1087-0792/$ - see front matter Q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.smrv.2005.05.002
50 G. Tononi, C. Cirelli

sleep ends. In addition to claiming a correspon-

dence between the homeostatic Process S and total
synaptic strength, the hypothesis proposes specific
mechanisms, whereby synaptic strength would
increase during wakefulness and decrease during
sleep, and suggests why the tight regulation of
synaptic strength would be of great importance for
the brain.

Synaptic homeostasis: a schematic

diagram

Fig. 1 The two-process model involving the circadian The diagram in Fig. 2 presents a simplified version
component (process C) and the homeostatic component of the main points of the hypothesis. During
(process S). wakefulness (yellow background), we interact
with the environment and acquire information
Specifically, the curve in Fig. 1 can be interpreted about it. The EEG is activated, and the neuromodu-
as reflecting how the total amount of synaptic latory milieu (for example, high levels of nor-
strength in the cerebral cortex (and possibly other adrenaline, NA) favors the storage of information,
brain structures) changes as a function of wakeful- which occurs largely through long-term poten-
ness and sleep. Thus, the hypothesis claims that, tiation of synaptic strength. This potentiation
under normal conditions, total synaptic strength occurs when the firing of a presynaptic neuron is
increases during wakefulness and reaches a maxi- followed by the depolarization or firing of a
mum just before going to sleep. Then, as soon as postsynaptic neuron, and the neuromodulatory
sleep ensues, total synaptic strength begins to milieu signals the occurrence of salient events.
decrease, and reaches a baseline level by the time Strengthened synapses are indicated in red, with

Fig. 2 The synaptic homeostasis hypothesis.

Sleep function and synaptic homeostasis 51

their strength given by a number. Note that one recalibrated level of synaptic strength, and the
synapse grows to a strength of 150, while another cycle can begin again.
synapse does not change and stays at 100. Note also
the appearance of a new synapse with a strength of
five. Due to the net increase in synaptic strength,
The main claims of the synaptic homeo-
waking plasticity has a cost in terms of energy
requirements, space requirements, and progress- stasis hypothesis
ively saturates our capacity to learn.
When we go to sleep (blue background), we After this schematic depiction of the hypothesis,
become virtually disconnected from the environ- we now turn to describing its main points in more
ment. Changes in the neuromodulatory milieu detail, and to discussing some supporting evidence.
trigger the occurrence of slow oscillations in
membrane potential, comprising depolarized and Wakefulness and synaptic potentiation
hyperpolarized phases, which affect every neuron in
the cortex, and which are reflected in the EEG as The synaptic homeostasis hypothesis states that
SWA. The changed neuromodulatory milieu (for wakefulness is accompanied by synaptic poten-
example, low levels of noradrenaline) ensures that tiation in a large fraction of cortical circuits,
synaptic activity is not followed by synaptic poten- resulting in a net increase in synaptic weight.
tiation, which makes adaptive sense given that According to the hypothesis, plastic changes
synaptic activity during sleep is not driven by would occur through much of waking life, whenever
interactions with the environment. Because average we are alert and make behavioral choices, whether
synaptic strength at the end of the waking period is or not we are specifically engaged in experimental
high, neurons undergoing sleep slow oscillations are learning paradigms. After all, synapses and neurons
highly synchronized. As a result, the EEG of early do not know whether they are engaged in a learning
sleep shows slow waves of high amplitude. paradigm, but only whether strong presynaptic
The slow waves, however, are not just an firing is accompanied by postsynaptic depolariz-
epiphenomenon of the increased synaptic strength, ation or firing in the presence of appropriate levels
but have a role to play. The repeated sequences of of neuromodulators, which should be a frequent
depolarization—hyperpolarization cause the down- occurrence during alert wakefulness. Also, accord-
scaling of the synapses impinging on each neuron, ing to the hypothesis, plastic changes occurring
which means that they all decrease in strength during wakefulness, at least in the adult, would
proportionally, here by 20%. Thus, a synapse that result more often in long-term potentiation (LTP)
after wakefulness had strength of 100 is downscaled than in long-term depression (LTD), thus resulting
to 80, another synapse, which had been potentiated in a net potentiation of synaptic strength.
to 150, is downscaled to 120 (green color). The
synapse with a strength of 5, having being downscaled Evidence
below a minimum strength, has been ‘downselected’ Direct evidence supporting this part of the hypothesis
or removed altogether. The reduced synaptic comes from anatomical work demonstrating a net and
strength reduces the amplitude and synchronization diffuse increase in synaptic density in animals
of the slow oscillations in membrane potential, which exposed to enriched environments likely to induce
is reflected in a reduced SWA in the sleep EEG. LTP-like molecular changes.3 Local increases in
Because of the dampening of the slow waves, synaptic density have also been observed. For
downscaling is progressively reduced, making the example, stimulating a whisker for 24 h produces a
process self-limiting when synaptic strength reaches selective net increase of synaptic density (by 35%) on
an appropriate baseline level. cortical neurons in the corresponding barrel field.4
Indeed, total synaptic strength, which had Strongly suggestive evidence for synaptic poten-
increased from 200 (100C100) at the beginning of tiation comes from the finding that spontaneous
wakefulness to 255 (100C150C5) at the end of wakefulness is regularly associated with the diffuse
wakefulness, is downscaled back to 200 (120C80) induction of molecular changes usually associated
by the end of sleep. By returning total synaptic with LTP5,6, including the phosphorylation of cAMPÿ
weight to an appropriate baseline level, sleep responsive elementÿbinding protein (CREB) and the
enforces synaptic homeostasis. This has benefits induction of genes such as Arc, brain-derived
in terms of energy requirements, space require- neurotrophic factor (BDNF), nerve growth factorÿ
ments, and learning and memory. Thus, when we induced gene A (NGFI-A), Homer, and neuronal
wake up, neural circuits do preserve a trace of the activityÿregulated pentraxin (Narp) (e.g.7–9). This
previous experiences, but are kept efficient at a induction of LTP-related genes during spontaneous
52 G. Tononi, C. Cirelli

wakefulness can increase further if animals are kept Evidence

awake longer by gentle handling, or if they engage in This portion of the hypothesis relies on evidence
extensive exploration of their environment (Huber from both animal and human studies. Consistent
et al., in preparation). During sleep, by contrast, with the hypothesis, increasing the duration of
the expression of LTP-related genes is severely wakefulness by a few hours (by gentle handling)
reduced or abolished.5,6,10 Support for the notion produces an increase in the expression of markers
that synaptic strength may increase during wakeful- of synaptic potentiation,10 followed by an increase
ness also comes from experiments in humans11 and in SWA. The level of induction of LTP-related
mice12,27 showing that brain metabolism, which is genes22 and the amount of SWA during sleep23
mostly due to synaptic activity, increases from early should depend not just on the duration, but also on
to late wakefulness (see below). the quality of wakefulness. Indeed, recent exper-
iments have shown that, for the same duration of
wakefulness, if animals spend more time exploring
Mechanisms
From an evolutionary point of view, it makes their environment, markers of synaptic poten-
tiation are further upregulated, and SWA during
sense that the potentiation of neural circuits
subsequent sleep is further increased (Huber et al.,
should occur during wakefulness, when an animal
in preparation). On the other hand, if wakefulness is
is active and exposed to the environment, and not
not accompanied by LTP-like changes in synaptic
during sleep, when neural activity is unrelated to
strength, the homeostatic increase in SWA should
external events.13 However, given that spon-
be eliminated. This prediction was confirmed by
taneous mean firing rates of cortical neurons in
examining animals with a lesioned noradrenergic
wakefulness and sleep are comparable,14 how is
the induction of LTP-related genes restricted to system, which have a greatly reduced expression of
LTP-related molecules in the cerebral cortex after
wakefulness? One reason may be that sensory,
periods of wakefulness.5,20 Although in these
motor, or cognitive activities that occur during
animals the amount and timing of sleep were
active wakefulness are often associated, in a
unchanged, the peak in SWA that is normally seen
small subset of neurons, with high peak firing
in the morning hours after the nocturnal activity
rates that are likely to give rise to LTP-related
phase was markedly dampened, and so was the SWA
plastic changes.15 Another mechanism could be
response to sleep deprivation.112 Thus, the hypoth-
the state-dependent firing of certain neuromodu-
latory systems.16 For example, the firing of the esis suggests that it is not wakefulness as such, but
the induction of LTP-related molecules normally
noradrenergic system is high during wakefulness,
associated with wakefulness, that is responsible for
especially during salient events, while it is very
driving the homeostatic increase in SWA.
low or absent during sleep.17 Noradrenaline is
An intriguing prediction of the hypothesis is that,
important for the induction of LTP,18 and nor-
to the extent that synaptic potentiation is particu-
adrenergic lesions impair at least some forms of
larly strong in specific brain areas, SWA during
learning.19 Indeed, if the noradrenergic inner-
subsequent sleep should increase disproportio-
vation of the cerebral cortex is destroyed,
P-CREB, Arc, BDNF, NGFI-A, Homer and Narp are nately in that area—a kind of local intensification
down close to the levels seen in sleep even when of sleep. We have searched for signs of local slow
wave homeostasis using high definition EEG to
the animal is awake and behaving, and even if the
investigate sleep after learning a visuomotor
waking EEG is essentially unchanged.5,20 The
task.24 In this task, performed shortly before
effect of noradrenaline is specific, since seroto-
bedtime, subjects reached for visual targets using
nergic lesions have no effect on the expression of
a handheld cursor while unconsciously adapting to
such genes.21
systematic rotations imposed to the perceived
cursor trajectory.25 One week earlier or later,
Synaptic potentiation and slow wave subjects performed a control task that was
homeostasis subjectively indistinguishable and kinematically
identical, but in which the cursor trajectory was
The synaptic homeostasis hypothesis states that the not rotated. Thus, the only difference between the
homeostatic regulation of SWA is tied to the amount two tasks was that the rotation adaptation task
of synaptic potentiation that has occurred during involved (implicit) learning of a compensatory
previous wakefulness. Specifically, the higher the rotation, presumably mediated by synaptic poten-
amount of synaptic potentiation in cortical circuits tiation in specific brain areas, whereas the control
during wakefulness, the higher the increase in SWA task did not. Indeed, previous PET work had shown
during subsequent sleep. that rotation learning involves a circumscribed
Sleep function and synaptic homeostasis 53

region in right parietal cortex.25 As predicted by the studies, stronger cortico-cortical connections
hypothesis, when we compared rotation adaptation cause a stronger activation of the sodium-depen-
to control tasks, we found a local increase of SWA dent potassium current, which leads in turn to a
(27%) extending over a small cluster of electrodes. longer and more hyperpolarized down-phase, and
Thus, the presumed induction of local plastic thus to slow oscillations of increased amplitude.
changes associated with practicing a visuomotor Furthermore, stronger cortico-cortical connections
task is associated with a local induction of SWA in increase the degree of synchronization among
subsequent sleep.24 The increase in power was populations of neurons.42 Both effects, the
mostly in the slow wave frequency range, and it increased amplitude of the slow oscillation of
declined over time, just like the global homeostatic individual cells, and the increased synchronization
response of SWA. Moreover, the increase of SWA of slow oscillations among populations of cells, are
was localized exactly at the predicted spot in right reflected in slow waves of larger amplitude at the
parietal cortex. These new results are in line with EEG level.42–44 Increased neuronal synchronization
previous evidence for local SWA homeostasis in due to stronger cortico-cortical synapses would also
humans26 and rats,27 and fit as well with other explain why the increase in power after wakeful-
proposals suggesting that sleep may be regulated ness extends to other frequency bands besides the
locally.28,29 Well-documented topographic differ- slow wave or delta band,45–50 although it remains to
ences in slow wave homeostasis, with frontal be explained why the time course of the increase
regions showing an especially strong response to varies for different frequency bands.51
sleep deprivation,30,31 may also be related to
topographic differences in the susceptibility to
plastic changes.32,33 Slow wave homeostasis and synaptic
Evidence for a relationship between synaptic downscaling
strength or density and SWA also comes from
developmental studies. SWA changes during the We have assumed that LTP-related changes occur-
lifespan in a way that seems to follow cortical ring in the cortex during wakefulness lead to a net
synaptic density,34,35 as indicated directly by increase in synaptic weight onto neurons, and that
electron microscopy on post-mortem tissue36 and such increase is reflected in increased amplitude of
by MRI estimates of the amount of gray matter.37 sleep slow waves. According to the hypothesis, such
Thus, both synaptic density and SWA reach a peak slow waves are not a mere epiphenomenon, but have
in adolescence, after which they decline rapidly, a function to perform: to promote a generalized
and continue a slower decline into old age. depression or downscaling of synaptic strength.
Pathological decreases in synaptic density, as Downscaling refers to a proportional reduction in
observed in neurodegenerative disorders and the strength of all synapses converging onto the
schizophrenia, are also associated with reductions same neuron: if all synapses shed the same
in SWA.38,39 Moreover, after visual deprivation proportion of their weight, total synaptic weight
during the critical period—a procedure associated can be reduced while preserving relative differences
with synaptic depression,40 slow waves are reduced in synaptic strength and thereby memory traces.
by 40% in the absence of changes in sleep Down- as well as up-scaling of cortical and
architecture.41 hippocampal synapses have been observed both
in vitro and in vivo.52,53 In these experiments, a
Mechanisms proportional reduction of the strength of all
What could be the mechanism linking local synaptic synapses impinging on a neuron could be produced
potentiation during wakefulness with increased by artificially elevating synaptic input to that
slow waves during sleep? Underlying the SWA neuron for several hours, whereas a prolonged
recorded in the EEG are oscillations in neuronal block of neural activity produced the opposite
membrane potential, the most important of which effect. Such activity-dependent mechanisms of
is a slow oscillation that is generated by cortical synaptic scaling ensure that neurons maintain a
cells and synchronized by cortico-cortical connec- regulated firing level in the face of uncontrollable
tions.14 The slow oscillation comprises a depolar- changes in their inputs.
ized up-phase, during which neurons fire at According to the hypothesis, sleep-dependent
relatively high rates, followed by a hyperpolarized synaptic scaling would ensure primarily the homeo-
down-phase, during which neurons are silent. The static control of synaptic weight, and only
down-phase is probably brought about by a sodium- indirectly of neuronal firing levels. Also, since
dependent potassium current that is activated as a wakefulness is assumed to result in net synaptic
function of neuronal firing. According to modeling potentiation, sleep would serve primarily to scale
54 G. Tononi, C. Cirelli

synapses down, rather than up. Like activity- phosphatase that dephosphorylates AMPA receptors
dependent synaptic scaling, however, sleep-depen- strengthened during long-term potentiation, pro-
dent downscaling would affect most or all of a tein phosphatase I, metabotropic glutamate
neuron’s synapses. In this respect, downscaling is receptor subunit 2, which is required for synaptic
conceptually different from long-term depression, depression, and several proteins involved in vesicle
which affects select groups of synapses, or depo- recycling, such as Nÿethylmaleimide-sensitive
tentiation, which affects only recently potentiated factor (NSF). Also, NREM sleep is associated with
ones.54 Nevertheless, downscaling is likely to use higher levels of insulin,60 which promotes the
many of the same molecular mechanisms involved internalization of AMPA receptors and LTD.61
in depression/depotentiation. Substantial evidence Thus, at least at the molecular level, sleep may
indicates that these forms of plasticity depend on not just be unfavorable to synaptic potentiation,
the dephosphorylation and subsequent internaliz- but specifically conducive to synaptic downscaling.
ation of alphaÿaminoÿ3ÿhydroxyÿ5ÿmethylÿ4ÿ More direct tests of this prediction can be
isoxazolepropionate (AMPA) receptors that ulti- envisaged. It is already known that sleep altogether
mately leads to a reduction in synaptic efficacy.55,56 favors dephosphorylation in the brain.62 It now
Whichever the specific mechanism, the hypothesis is appears that sleep is associated with the selective
that a generalized synaptic downscaling during sleep, dephosphorylation of AMPA channel residues
including possibly the downselection or pruning of involved in synaptic depression (Cirelli et. al., in
certain synapses, serves to ensure the maintenance preparation).
of balanced synaptic input to cortical neurons. Possible anatomical evidence for downscaling
during sleep comes from studies showing that
Evidence increases in synaptic density triggered by learning
We have seen above that the amplitude and are evident for just a few hours, after which synaptic
synchronization of sleep oscillations appear to density returns to baseline. For example, rats who
reflect the strength of cortico-cortical connections. had learned the position of a hidden platform
To the extent that this is true, the well-established showed an increase in hippocampal synaptic density
exponential decrease of EEG SWA during sleep2 after 9 h of training (three training blocks at 3 h
represents strong electrophysiological evidence for intervals). If left alone for 12 h followed by one last
downscaling during sleep. Especially relevant is the training block, rats remembered the location of the
finding that, if SWA is suppressed through acoustic platform even better. However, synaptic density
stimuli during the first part of sleep, SWA increases was down to control levels, as if the intervening 12 h
greatly in the second part of sleep.57 Since slow with ad libitum sleep had reestablished synaptic
waves must occur for sleep SWA to decline, they homeostasis.63 It is not yet known, however,
must be more than a mere epiphenomenon of sleep. whether the return of synaptic density to baseline
A decrease in SWA as soon as we enter sleep has also levels necessarily requires sleep as opposed to
been observed after learning tasks that increase merely the passage of time.
SWA locally.24 Further evidence comes from multi- Finally, functional evidence that NREM sleep may
unit recordings where neurons that are coactivated be associated with synaptic downscaling or with
during waking display correlated firing during sleep, synaptic downselection comes from studies of
presumably reflecting stronger synaptic links monocular visual deprivation in kittens, a well-
between them. Interestingly, the strength of the known model of cortical plasticity. During a critical
correlation, and presumably the strength of under- period of brain development, occluding one eye
lying synapses, decays rapidly (within 30 min) during when the animal is awake in the light for 6 h greatly
the sleep episode,58 in line with downscaling (see reduces the ability of cortical cells to respond to
also59). Neuroimaging data showing a decrease in the occluded eye. It is now thought that such plastic
absolute levels of brain metabolism after sleep are reduction is due to depression of cortical connec-
also consistent with the hypothesis that synaptic tions related to the deprived eye.40 The plastic
strength is downscaled during sleep,11 (see below). depression of responses to the occluded eye can be
A role for sleep in downscaling is compatible with increased if the animal remains awake in the light,
recent molecular evidence. We have seen that but not in the dark, for six more hours. Remarkably,
during sleep the expression of LTP-related mol- an equivalent increase in depression can be seen if
ecules reaches a low level.5 On the other hand, the animal is allowed to sleep for 6 h in the dark.64
sleep is associated with the upregulation of This result has been interpreted in terms of sleep-
molecules implicated in depotentiation/ mediated ‘consolidation’, but it could be due as
depression.6 Such molecules include calcineurin, a well to sleep-dependent downscaling.
Sleep function and synaptic homeostasis 55

Mechanisms The functional advantages of synaptic

Why would SWA lead to synaptic downscaling? As we downscaling during sleep
have seen, the fundamental cellular phenomenon
underlying NREM sleep rhythms is the slow oscil- According to the hypothesis, synaptic downscaling
lation, which is seen in virtually every cortical cell during sleep would offer several benefits. The most
recorded intracellularly.14 Remarkably, the slow important benefits have to do with basic commod-
oscillation occurs at a frequency—less than 1 Hz— ities in the brain, such as energy and space, but
that is ideally suited to induce depotentiation/ downscaling may also benefit learning and memory.
depression in stimulation paradigms.54 Low fre-
quency oscillations during sleep may promote
Energy savings
depression through changes in calcium dynamics,
About 40% of the energy requirements of the
which are crucial for depression.54 The unique
cerebral cortex—by far the most metabolically
neuromodulatory milieu of NREM sleep—low acetyl-
expensive tissue in the body—are due to neuronal
choline, noradrenaline, serotonin, and histamine—
repolarization following postsynaptic potentials.67
may also be important, as well as the fact that
The higher the synaptic weight impinging on a
BDNF, which prevents depression, is low in sleep.65
neuron, the higher this portion of the energy
The most significant factor promoting downscaling,
budget. Moreover, increased synaptic weight is
however, could be the very sequence of depolariz-
thought to lead to increased average firing
ation (up-phase) and hyperpolarization (down-
rates,68 and spikes in turn are responsible for
phase) that characterizes slow oscillations at the
another 40% of the gray matter energy budget.67
cellular level. 14 The close temporal pairing
Therefore, it would seem energetically prohibitive
between generalized spiking at the end of the up-
for the brain to let synaptic weight grow without
phase and generalized hyperpolarization at the
checks as a result of waking plasticity. Indeed, if
beginning of the down-phase may indicate to
PET data11 offer any indication, after just one
synapses that presynaptic input was not effective
waking day energy expenditure may grow by as
in driving postsynaptic activity, a key requirement
much as 18%. Sleep, and the accompanying down-
for depression.54 Alternatively, depolarization–
scaling of synapses, would then be needed to
hyperpolarization sequences might be sufficient to
interrupt the growth of synaptic strength associ-
trigger downscaling. Yet another possibility is that
ated with waking and prevent synaptic overload.
depression may be powerfully triggered by the
The recalibration of cortical circuits during sleep
temporal pairing between generalized hyperpolar-
would yield a brain that remains energetically
ization at the end of the down-phase and general-
efficient despite keeping trace of previous waking
ized spiking at the beginning of the up-phase.
experiences.
Whatever the precise mechanisms of SWA-
dependent downscaling, an appealing feature of
this entire process is that it could be self-limiting. As Space savings
shown by computer simulations,42 the progressive Another benefit of synaptic downscaling/downse-
reduction of synaptic strength due to SWA-depen- lection during sleep would be in terms of space
dent downscaling reduces postsynaptic depolariz- requirements. Synaptic strengthening is thought to
ation, an effect further amplified by the reduced be accompanied by morphological changes, includ-
synchronization of slow oscillations among different ing increased size of terminal boutons and spines,
cells. As a consequence, sodium dependent potass- and synapses may even grow in number (e.g.3,4,63).
ium currents that bring about the hyperpolarized But space is a precious commodity in the brain, and
phase are less activated. Eventually, cortical cells even minuscule increases in volume are extremely
stop alternating between crisp up- and down- dangerous. For example, neocortical gray matter is
phases, and hover instead around an intermediate tightly packed, with wiring (axons and dendrites)
membrane potential that prevents further down- taking up w60% of the space, synaptic contacts
scaling. When this endpoint is reached, SWA has (boutons and spines) w20%, and the rest (cell
decreased exponentially to a minimum, and synaptic bodies, vessels, extracellular space) the remaining
weight has returned to baseline. This notion is 20%.69 Thus, sleep would be important not just to
consistent with experimental studies showing that keep in check the metabolic cost of strengthened
after 4–5 NREM sleep episodes SWA reaches an synapses, but also to curb their demands on brain
asymptote and remains at a fairly constant level.66 real estate.
56 G. Tononi, C. Cirelli

Benefits for learning and memory disappearance during sleep of the red synapse
Sleep-dependent downscaling may have additional with a weight of five). Indeed, just as predicted,
benefits for learning and memory. For instance, it is when subjects were tested after sleep following the
likely that, due to the combined energy and space rotation adaptation task, they showed a significant
costs of uninterrupted synaptic plasticity, the enhancement of their performance, which was
ability of the brain to acquire new information absent in subjects who trained in the morning and
would rapidly grind to a halt in the absence of were retested after 8 h of wakefulness. Moreover,
downscaling. In this sense, sleep would not only be performance enhancement after sleep was strongly
the price we have to pay for plasticity the previous correlated with the increase in SWA in the right
day, but also an investment to allow the organism to parietal areas involved in the task. Finally, the
learn afresh the next day. Indeed, in certain brain strongest correlation (rZ0.9) was with the increase
areas, such as the hippocampus, radical synaptic of signal-to-noise ratios during learning.
downscaling may be necessary to clean the slate Other groups have found that sleep can indeed
and rapidly adapt to a new environment. Another enhance performance in certain tasks.72–79 These
benefit of downscaling/downselection would be to studies generally assume that sleep may enhance
promote synaptic competition, which may be performance by ‘replaying’ patterns of neural
especially important during development, a time activity obtained during training in wakefulness. It
of exuberant synaptic growth. For example, con- is frequently suggested that such replay may
nections between strongly correlated neurons actually potentiate synapses (e.g.80). The synaptic
would survive, while others may be eliminated.70 homeostasis hypothesis, by contrast, predicts that
In the adult, downscaling could benefit learning in sleep may enhance performance by global down-
yet another way by increasing signal-to-noise ratios scaling, thanks to the postulated increase in signal-
in the relevant brain circuits. To illustrate, consider to-noise ratios. This possibility is more economical
again the visuomotor task discussed in connection (and energy efficient), and has the important
with local slow wave homeostasis.24 The neural advantage of not requiring great fidelity in sleep
substrates of many forms of visuomotor learning are ‘replays’ (see below). Nevertheless, downscaling
thought to be changes in synaptic strength within and the relative potentiation of recently tagged
circuits in motor and parietal areas. PET studies synapses are not mutually exclusive , especially
indicate that, during visuomotor learning, brain because downscaling should promote competition.
activation is at first diffuse and bilateral,71 and
only after further practice does it converge upon Is sleep necessary for synaptic homeostasis?
more restricted foci of cortical activation.25 This Any proposal about the function of sleep should be
pattern is not surprising, since visuomotor learning is able to provide a convincing explanation of why the
an incremental process, during which early proposed function can only be fulfilled by sleep and
executions are inaccurate, and only slowly converge not by quiet wakefulness.13 Otherwise, why would
upon correct trajectories. Thus, it is likely that early sleep—a potentially dangerous behavior character-
on some synapses contributing to erroneous or ized by loss of contact with the environment—be so
imperfect movements may be potentiated (noise), universal, and why would sleep pressure be so
although later on synapses contributing to a correct overwhelming? While downscaling during wakeful-
movement will become progressively more effica- ness cannot be ruled out a priori, there are several
cious (signal). In Fig. 2, this is indicated by the reasons why sleep might be necessary. Perhaps the
appearance, in addition to the appropriately most important reason is that, in order to
strengthened red synapse with a weight of 150, of determine how much downscaling is needed to
a small red synapse with a weight of five. maintain synaptic homeostasis, a neuron should be
It is here that synaptic downscaling during sleep able to assess its total synaptic input in an unbiased
can play a role. According to the hypothesis, during manner, which is to say off-line, independent of
sleep the strength of each synapse would decrease behavioral requirements. This is difficult to do
by a proportional amount, until the total amount of during wakefulness. Suppose, the waking day is
synaptic weight impinging on each neuron returns spent in reiterating certain behavioral tasks, so that
to a baseline level. Provided there is a threshold certain neural circuits are strongly and repeatedly
below which synapses become ineffective, silent, activated. Based on high average synaptic input,
or disappear, synapses contributing to the noise, neurons partaking in such circuits would engage in a
being on average much weaker than those con- much heavier dose of downscaling that they
tributing to the signal, would cease to interfere in actually need. During sleep, by contrast, neural
the execution, and the signal-to-noise ratios would activity occurs spontaneously and off-line, virtually
increase (in Fig. 2, this is indicated by the disconnected from behavioral requirements. This
Sleep function and synaptic homeostasis 57

spontaneous activity is likely to reflect synaptic arterial pCO2, relates to cerebral oxygen utilization
strength rather than outside influences. In this way, and therefore to metabolic rate. Quite unexpect-
a neuron’s synaptic input would represent an edly, absolute blood flow values were 18% higher at
unbiased estimate of the synaptic strength imping- the end of the waking day than after a night of
ing on it, and the neuron could downscale sleep, and this was the case almost everywhere in
appropriately. Another reason why downscaling the brain. A change of this magnitude is not usually
might best occur during sleep is that, at the seen in PET studies, even less so when comparing
molecular level, generalized changes in synaptic two identical ‘resting’ conditions, but it would be
strength may be incompatible with the need to consistent with a net increase in synaptic strength
selectively increase the strength of certain during wakefulness. The deoxyglucose study in rats
synapses, as is the case during learning. And of is also consistent with this picture, in that glucose
course, to the extent that downscaling is promoted utilization was considerably higher in waking before
by repetitive depolarization—hyperpolarization sleep than in waking after sleep.83 Also consistent is
sequences, these are perfectly compatible with a study using transcranial Doppler ultrasonography,
sleep but would seriously interfere with behavior if where waking cerebral blood flow velocity in the
they were to occur during wakefulness. middle cerebral artery was 6.6% lower post-sleep
compared to pre-sleep.84 Another study85 showed
that cerebral blood flow velocities are significantly
higher during the first NREM sleep episode than
Some implications of the synaptic
during the second or the last. Even for the same
homeostasis hypothesis sleep stage, such as stage 2, values were always
lower later in the night. Indeed, the average
To the extent that the main claims of the synaptic decrease of blood flow velocity during a night of
homeostasis hypothesis are justified by the avail- sleep parallels closely the decrease of SWA. Further
able evidence, they offer a fresh perspective on studies comparing absolute values of brain glucose
several aspects of sleep and sleep medicine. In what and oxygen consumption (and other metabolic
follows, we will consider some intriguing impli- parameters) before and after sleep in a standar-
cations of the hypothesis for neuroimaging studies, dized waking condition, should help establish
and briefly discuss the possibility that a dysregula- whether normal wakefulness is indeed
tion of synaptic homeostasis may be implicated in accompanied by a generalized increase in brain
disorders such as insomnia and depression. energy requirements. Equally important, they
should determine to what extent changes in blood
The synaptic homeostasis hypothesis, flow and metabolism between evening and morning
neuroimaging, and reactivation are due to sleep rather than, for example, to a
circadian modulation of arousal.
Some of the most intriguing corollaries of the The hypothesis also predicts that intense
synaptic homeostasis hypothesis concern neuroima- learning tasks triggering local synaptic changes in
ging. As was mentioned above, it has been specific brain regions may lead to a local hyperme-
calculated that nearly 80% of cortical grey matter tabolic ‘trace’ under resting conditions (or during
metabolism is related to neural activity,67,81 half of the performance of unrelated tasks). This is
it to support action potentials and half to support because, to the extent that learning the task
postsynaptic potentials. Thus, synaptic strength leads to an increase in synaptic strength in specific
may control w40% of the cortex energy needs brain regions, those regions should show an increase
directly, and potentially more because of indirect in metabolic need even at rest, when they are
effects on firing rates.68,82 If, as predicted by the undergoing spontaneous activity. On the other
hypothesis, synaptic weight increases in the course hand, such hypermetabolic traces should be pro-
of normal wakefulness, brain metabolism should gressively normalized over a sufficiently long period
also increase. Support for this prediction comes of sleep. According to the hypothesis, several
from PET experiments in humans11 and deoxyglu- imaging experiments showing a ‘reactivation’
cose studies in mice.83 The human study found a during stages of sleep (both NREM and REM) after
remarkable difference in the absolute value of learning a task,86–88 could then be interpreted as
cerebral blood flow at rest between an awake scan showing a metabolic trace of the induction of
in the evening, after a variable schedule of partial synaptic potentiation during previous wakefulness.
sleep deprivation, and an awake scan in the The hypothesis predicts that such traces should be
morning, after several hours of sleep. Blood flow visible also in subsequent wakefulness, but should
was measured with O15, and its value, corrected for decrease in intensity in the course of sleep. By the
58 G. Tononi, C. Cirelli

same token, the hypothesis suggests that the insomnia and syndromes characterized by the
electrophysiological instances of ‘replay’58,89–93 subjective feeling of non-restorative sleep, as
can also be interpreted as a ‘trace’ of synaptic well as in psychiatric disorders characterized by
strengthening induced during previous wakefulness, significant sleep disturbances, such as depression.
expressed in this case as correlated spontaneous Primary insomnia is a 24-h disorder in which
firing. Consistent with this idea, electrophysiologi- subjective feeling of non-restorative sleep is
cal experiments indicate that the so-called replay associated with fatigue, difficulty concentrating,
also occurs during quiet wakefulness,58 and that its cognitive impairment, irritability and mood
intensity decreases in the course of sleep, at least changes,94–96 all features that may result from
in the hippocampus. Multielectrode recordings in impaired synaptic homeostasis. Primary insomnia
multiple forebrain structures also indicate that, is often associated with hyperarousal,97,98 which is
while multiunit activity patterns observed in a sleep likely to impair synaptic homeostasis during sleep.
episode do bear a statistical relationship with the Recent neuroimaging studies have shown that
activity patterns triggered by previous waking insomniacs have globally increased brain metab-
experiences, the similarity is extremely low-fide- olism during both waking and sleep, possibly a
lity.93 If neural activity during sleep resulted in reflection of hyperarousal, but show relative
synaptic potentiation, as is sometimes claimed, reductions of glucose metabolism in prefrontal
what would be learned during sleep would be cortex, possibly due to insufficient sleep restor-
mostly noise, rather than signal. ation. Indeed, during sleep insomniac patients fail
to deactivate medial prefrontal cortex, anterior
cingulate cortex, and parts of the thalamus,
Dysregulations of synaptic homeostasis and suggesting a disruption of homeostatic mechan-
cognitive impairment isms.98 In line with this suggestion, recent studies
suggest that enhancing sleep homeostasis using
From the clinical perspective, one of the most sleep restriction can result in the improvement of
relevant predictions of the synaptic homeostasis insomnia.99 According to the hypothesis, then,
hypothesis is that, if the process of synaptic some of the symptoms of primary insomnia may
homeostasis is prevented by sleep deprivation or be due, at least in part, to synaptic overload and to
sleep restriction, symptoms related to synaptic compensatory changes in neuronal excitability.
overload of neocortical and limbic circuits should Major depression shares many of the same
follow. These could include cognitive impairment, symptoms with primary insomnia. Sleep disturbances
loss of sleep-related performance enhancement, such as insomnia or hypersomnia are defining features
emotional impairment, difficulty concentrating, of the disorder. Major depression and insomnia are
and fatigue. Such symptoms are expected to occur epidemiologically related, and individuals with
because synaptic overload should cause metabolic insomnia are more likely to develop depression than
overload, neuropil crowding, a decrease in neuronal normal sleepers.100 Abnormalities of sleep structure,
signal-to-noise ratios, and a reduction of the such as disruption of slow wave sleep and increased
capacity for plastic change. Metabolic overload, in REM sleep latency, are common features of
turn, may lead to compensatory reactions, such as a depression, as are changes in SWA during baseline
homeostatic reduction in neuronal excitability or sleep.101,102 Imaging studies have pointed to local
increased synaptic failure. Thus, two kinds of brain abnormalities, such as hypofrontality during
symptoms should be expected after sleep depri- both wakefulness and sleep, and smaller sleep-
vation and sleep restriction: (i) ‘sleepiness’ related decrements in activity in fronto-parietal
(increased propensity to fall asleep) and related areas103 Moreover, several studies indicate that
symptoms, due to globally increased sleep pressure sleep deprivation, which provides an acute challenge
and mediated by central homeostatic mechanisms; to sleep homeostasis, can ameliorate many of the
and (ii) ‘non-restorative symptoms’ (cognitive symptoms of depression.104 Also, acute sleep depri-
impairment, fatigue etc.) due to the local effects vation and chronic antidepressant treatment result in
of sleep loss on cortical and limbic circuits, and the induction of a similar set of genes involved in
mediated by synaptic overload or homeostatic synaptic potentiation, such as BDNF and P-CREB.105
changes in excitability. The hypothesis also The synaptic homeostasis hypothesis suggests
suggests that a dysregulation of synaptic homeo- that, at least in some cases, a local reduction of
stasis may be implicated in certain neuropsychiatric slow wave homeostasis may reflect an insufficient
disorders. Specifically, symptoms resulting from level of synaptic strength during wakefulness. For
dysregulation of synaptic homeostasis are likely to example, in a proportion of depressed subjects, in
be prominent in sleep disorders such as primary line with neuroimaging reports of reduced
Sleep function and synaptic homeostasis 59

prefrontal activation (hypofrontality), baseline cerebral cortex where sleep rhythms are different,
synaptic strength in certain cortico-limbic circuits such as the hippocampus? Or to other species, such as
may be lower than normal. As predicted by the the fruit fly? Can the changes in SWA homeostasis also
hypothesis, this local reduction of synaptic strength be affected by changes in the balance between
should be evident as a local reduction of SWA during excitatory and inhibitory circuits? Moreover, what is
sleep. According to the hypothesis, then, these the role of REM sleep? And what is the role of sleep
subjects would be expected to improve with sleep spindles, given that their time course and homeostatic
deprivation because extended wakefulness would regulation differ from those of SWA?107,108 Another
potentiate those very circuits to reach normal levels issue concerns the effects of sleep deprivation and
of synaptic strength. In fact, depressed subjects who sleep restriction. Specifically, what happens if
respond best to sleep deprivation are those who synaptic downscaling is prevented from occurring or
report a diurnal improvement in mood.106 On the is incomplete? Do other mechanisms intervene to
other hand, the occurrence of synaptic downscaling reduce neuronal excitability and thereby metabolic
during sleep would explain why the antidepressant needs?109 This question is particularly important since
effects of sleep deprivation are typically short- during chronic sleep restriction SWA reaches a steady
lasting. The hypothesis also suggests that, in such
state,110 whereas performance deficits are cumulat-
patients, therapeutic approaches aimed at locally
ive.111 Finally, what are the implications of the
potentiating synaptic strength, such as targeted
hypothesis concerning pharmacological treatments?
transcranial magnetic stimulation, or specific
Do drugs that reduce SWA, such as benzodiazepines,
learning tasks, ideally augmented by noradrenergic
negatively affect downscaling and thereby sleep
agonists, would be most effective if systematically
applied every morning. restoration? Conversely, can synaptic downscaling
At the present stage, these suggestions are be potentiated by pharmacological agents that can
necessarily tentative and imprecise. Nevertheless, exaggerate the slow oscillations of NREM sleep?
the notion that disturbances in sleep-related Altogether, the main claims of the synaptic
synaptic homeostasis may play a significant role in homeostatic hypothesis are consistent with a large
the pathogenesis of psychiatric and neurological body of evidence at the behavioral, molecular, and
disorders is worth considering whenever there is neurophysiological level, and with results obtained
evidence for an imbalance of plastic or metabolic with techniques ranging from computer simulations
neuronal processes.

Research agenda
To validate the synaptic homeostasis hypoth-
Conclusion esis, we need to test some of its key
predictions:
In summary, the synaptic homeostasis hypothesis
† waking behaviors associated with synaptic
makes four main claims: (1) Wakefulness is associ-
potentiation should be followed by an
ated with synaptic potentiation in several cortical
increased SWA homeostatic response during
circuits; (2) Synaptic potentiation is tied to the
sleep. This can be tested for example using
homeostatic regulation of slow wave activity; (3)
learning tasks, enriched environment, or high-
Slow wave activity is associated with synaptic
frequency electrophysiological stimulation
downscaling; (4) Synaptic downscaling is tied to
† waking behaviors associated with synaptic
the beneficial effects of sleep on neural function
depression should be followed by a blunted
and, indirectly, on performance. From these claims
SWA homeostatic response during sleep. This
derive several intriguing possibilities, including the
can be tested for example by using sensory
possibility that a dysregulation of synaptic homeo-
deprivation paradigms or low-frequency elec-
stasis may be implicated in disorders such as
trophysiological stimulation
insomnia and depression. Moreover, the hypothesis
† brain metabolism should increase during
has relevant implications for neuroimaging studies.
wakefulness and decrease after sleep. This can
The hypothesis also triggers some further questions
be tested using deoxyglucose studies in animals
that have not been addressed here. For example,
and various neuroimaging approaches in
what are the complex relationships between the local
humans
regulation of sleep as mediated by synaptic homeo-
† learning a task should leave a metabolic
stasis, and the global regulation of sleep as mediated
trace in the involved area that is visible under
by hypothalamic and brainstem centers? How does the
resting conditions and is reduced after sleep.
hypothesis apply to brain structures other than the
60 G. Tononi, C. Cirelli

to human neuroimaging. The very fact that the 16. Tononi G, Cirelli C, Pompeiano M. Changes in gene expression
hypothesis accounts for such disparate results during the sleep–waking cycle: a new view of activating
systems. Archives Italiennes De Biologie 1995;134(1):21–37.
represents one of its more appealing features. 17. Aston-Jones G, Bloom F. Activity of norepinephrine-
However, it should be made clear that several containing locus coeruleus neurons in behaving rats
aspects of the hypothesis are still speculative. Most anticipates fluctuations in the sleep–waking cycle.
of these aspects, fortunately, lend themselves to J Neurosci 1981;1:876–86.
stringent experimental tests. 18. Walling SG, Harley CW. Locus ceruleus activation initiates
delayed synaptic potentiation of perforant path input to
the dentate gyrus in awake rats: a novel beta-adrenergic-
and protein synthesis-dependent mammalian plasticity
mechanism. J Neurosci 2004;24(3):598–604.
Acknowledgements 19. Robbins TW, Everitt BJ. Central norepinephrine neurons
and behavior. In: Bloom FE, Kupfer DJ, editors. Psycho-
pharmacology: the fourth generation of progress. New
Supported by RO1-MH65135 York: Raven Press; 1995. p. 363.
20. Cirelli C, Pompeiano M, Tononi G. Neuronal gene expression
in the waking state: a role for the locus coeruleus. Science
1996;274(5290):1211–5.
References 21. Tononi G, Cirelli C, Shaw PJ. Molecular correlates of sleep,
the waking state, and sleep deprivation. In: Borbély A,
1. Tononi G, Cirelli C. Sleep and synaptic homeostasis: a Hayaishi O, Sejnowski TJ, Altman JS, editors. The
hypothesis. Brain Res Bull 2003;62(2):143–50. regulation of sleep. Strasbourg: HSFP; 2000. p. 155–68.
*2. Borbely AA, Achermann P. Sleep homeostasis and models of 22. Kelly MP, Deadwyler SA. Experience-dependent regulation
sleep regulation. J Biol Rhythms 1999;14(6):557–68. of the immediate-early gene arc differs across brain
3. Klintsova AY, Greenough WT. Synaptic plasticity in cortical regions. J Neurosci 2003;23(16):6443–51.
systems. Curr Opin Neurobiol 1999;9(2):203–8. 23. Meerlo P, de Bruin EA, Strijkstra AM, Daan S. A social
4. Knott GW, Quairiaux C, Genoud C, Welker E. Formation of conflict increases EEG slow-wave activity during sub-
dendritic spines with GABAergic synapses induced by whisker sequent sleep. Physiol Behav 2001;73(3):331–5.
stimulation in adult mice. Neuron 2002;34(2):265–73. *24. Huber R, Ghilardi MF, Massimini M, Tononi G. Local sleep
*5. Cirelli C, Tononi G. Differential expression of plasticity- and learning. Nature 2004;430(6995):78–81.
related genes in waking and sleep and their regulation by 25. Ghilardi M, Ghez C, Dhawan V, Moeller J, Mentis M,
the noradrenergic system. J Neurosci 2000;20(24):9187–94. Nakamura T, et al. Patterns of regional brain activation
*6. Cirelli C, Gutierrez CM, Tononi G. Extensive and divergent associated with different forms of motor learning. Brain Res
effects of sleep and wakefulness on brain gene expression. 2000;871(1):127–45.
Neuron 2004;41(1):35–43. 26. Kattler H, Dijk DJ, Borbely AA. Effect of unilateral
7. Ying SW, Futter M, Rosenblum K, Webber MJ, Hunt SP, somatosensory stimulation prior to sleep on the sleep EEG
Bliss TV, et al. Brain-derived neurotrophic factor induces in humans. J Sleep Res 1994;3(3):159–64.
long-term potentiation in intact adult hippocampus: require- 27. Vyazovskiy V, Borbely AA, Tobler I. Unilateral vibrissae
ment for ERK activation coupled to CREB and upregulation of stimulation during waking induces interhemispheric EEG
Arc synthesis. J Neurosci 2002;22(5):1532–40. asymmetry during subsequent sleep in the rat. J Sleep Res
8. Silva AJ. Molecular and cellular cognitive studies of the role of 2000;9(4):367–71.
synaptic plasticity in memory. J Neurobiol 2003;54(1):224–37. 28. Krueger JM, Obal F. A neuronal group theory of sleep
9. Wallace CS, Withers GS, Weiler IJ, George JM, Clayton DF, function. J Sleep Res 1993;2(2):63–9.
Greenough WT. Correspondence between sites of NGFI-A 29. Krueger JM, Obal Jr F, Kapas L, Fang J. Brain organization
induction and sites of morphological plasticity following and sleep function. Behav Brain Res 1995;69(1-2):177–85.
exposure to environmental complexity. Brain Res Mol Brain 30. Finelli LA, Borbely AA, Achermann P. Functional topography
Res 1995;32(2):211–20. of the human nonREM sleep electroencephalogram. Eur
10. Cirelli C, Tononi G. Gene expression in the brain across the J Neurosci 2001;13(12):2282–90.
sleep–waking cycle. Brain Res 2000;885(2):303–21. 31. Huber R, Deboer T, Tobler I. Topography of EEG dynamics
*11. Braun AR, Balkin TJ, Wesenten NJ, Carson RE, Varga M, after sleep deprivation in mice. J Neurophysiol 2000;84(4):
Baldwin P, et al. Regional cerebral blood flow throughout 1888–93.
the sleep–wake cycle. An H2(15)O PET study. Brain 1997; 32. Grutzendler J, Kasthuri N, Gan WB. Long-term dendritic spine
120(Pt 7):1173–97. stability in the adult cortex. Nature 2002;420(6917):812–6.
12. Vyazovskiy VV, Achermann P, Borbely AA, Tobler I. The *33. Trachtenberg JT, Chen BE, Knott GW, Feng G, Sanes JR,
dynamics of spindles and EEG slow-wave activity in NREM Welker E, et al. Long-term in vivo imaging of experience-
sleep in mice. Arch Ital Biol 2004;142(4):511–23. dependent synaptic plasticity in adult cortex. Nature 2002;
13. Tononi G, Cirelli C. Some considerations on sleep and 420(6917):788–94.
neural plasticity. Arch Ital Biol 2001;139(3):221–41. 34. Feinberg I. Effects of maturation and aging on slow-wave
*14. Steriade M. The corticothalamic system in sleep. Front sleep in man: Implications for neurobiology. In: Wauquier A
Biosci 2003;8:D878–D99. DC, Radulovacki M, editor. Slow Wave Sleep: Physiological,
15. Sjostrom PJ, Turrigiano GG, Nelson SB. Rate, timing, and Pathophysiological and Functional Aspects. New York:
cooperativity jointly determine cortical synaptic plasticity. Raven Press; 1989.
Neuron 2001;32(6):1149–64. 35. Feinberg I, Thode Jr HC, Chugani HT, March JD. Gamma
distribution model describes maturational curves for delta
wave amplitude, cortical metabolic rate and synaptic
* The most important references are denoted by an asterisk. density. J Theor Biol 1990;142(2):149–61.
Sleep function and synaptic homeostasis 61

36. Huttenlocher PR, Dabholkar AS. Regional differences in 57. Dijk DJ, Beersma DG, Daan S, Bloem GM, Van den
synaptogenesis in human cerebral cortex. J Comp Neurol Hoofdakker RH. Quantitative analysis of the effects of
1997;387(2):167–78. slow wave sleep deprivation during the first 3 h of sleep on
37. Sowell ER, Thompson PM, Tessner KD, Toga AW. Mapping subsequent EEG power density. Eur Arch Psychiatry Neurol
continued brain growth and gray matter density reduction in Sci 1987;236(6):323–8.
dorsal frontal cortex: Inverse relationships during postado- 58. Kudrimoti HS, Barnes CA, McNaughton BL. Reactivation of
lescent brain maturation. J Neurosci 2001;21(22):8819–29. hippocampal cell assemblies: effects of behavioral state,
38. Petit D, Gagnon JF, Fantini ML, Ferini-Strambi L, experience, and EEG dynamics. J Neurosci 1999;19(10):
Montplaisir J. Sleep and quantitative EEG in neuro- 4090–101.
degenerative disorders. J Psychosom Res 2004;56(5):487–96. 59. Colgin LL, Kubota D, Jia Y, Rex CS, Lynch G. Long-term
39. Feinberg I. Schizophrenia: caused by a fault in programmed potentiation is impaired in rat hippocampal slices that
synaptic elimination during adolescence? J Psychiatr Res produce spontaneous sharp waves. J Physiol 2004;
1982;17(4):319–34. 558(Pt3):953–61.
*40. Heynen AJ, Yoon BJ, Liu CH, Chung HJ, Huganir RL, 60. Simon C, Brandenberger G, Saini J, Ehrhart J, Follenius M.
Bear MF. Molecular mechanism for loss of visual cortical Slow oscillations of plasma glucose and insulin secretion
responsiveness following brief monocular deprivation. Nat rate are amplified during sleep in humans under continuous
Neurosci 2003;6(8):854–62. enteral nutrition. Sleep 1994;17(4):333–8.
41. Miyamoto H, Katagiri H, Hensch T. Experience-dependent 61. Man HY, Lin JW, Ju WH, Ahmadian G, Liu L, Becker LE, et al.
slow-wave sleep development. Nat Neurosci 2003;6(6):553–4. Regulation of AMPA receptor-mediated synaptic trans-
*42. Hill S, Tononi G. Modeling sleep and wakefulness in the mission by clathrin-dependent receptor internalization.
thalamocortical system. J Neurophysiol 2005;93(3):1671–98. Neuron 2000;25(3):649–62.
43. Bazhenov M, Timofeev I, Steriade M, Sejnowski TJ. Model of 62. Cirelli C, Tononi G. Changes in anti-phosphoserine and anti-
thalamocortical slow-wave sleep oscillations and transitions phosphothreonine antibody binding during the sleep–
to activated States. J Neurosci 2002;22(19):8691–704. waking cycle and after lesions of the locus coeruleus.
44. Compte A, Sanchez-Vives MV, McCormick DA, Wang XJ. Sleep Res Online 1998;1:11–18.
Cellular and network mechanisms of slow oscillatory 63. Eyre MD, Richter-Levin G, Avital A, Stewart MG. Morpho-
activity (!1 Hz) and wave propagations in a cortical logical changes in hippocampal dentate gyrus synapses
following spatial learning in rats are transient. Eur
network model. J Neurophysiol 2003;89(5):2707–25.
J Neurosci 2003;17(9):1973–80.
45. Aeschbach D, Matthews JR, Postolache TT, Jackson MA,
64. Frank MG, Issa NP, Stryker MP. Sleep enhances plasticity in
Giesen HA, Wehr TA. Dynamics of the human EEG during
the developing visual cortex. Neuron 2001;30(1):275–87.
prolonged wakefulness: evidence for frequency-specific
65. Sheng M, Hyoung Lee S. AMPA receptor trafficking and
circadian and homeostatic influences. Neurosci Lett 1997;
synaptic plasticity: major unanswered questions. Neurosci
239(2-3):121–4.
Res 2003;46(2):127–34.
46. Cajochen C, Brunner DP, Krauchi K, Graw P, Wirz-Justice A.
66. Dijk DJ, Brunner DP, Borbely AA. Time course of EEG power
Power density in theta/alpha frequencies of the waking
density during long sleep in humans. Am J Physiol 1990;
EEG progressively increases during sustained wakefulness.
258(3 Pt 2):R650–R61.
Sleep 1995;18(10):890–4.
67. Attwell D, Laughlin SB. An energy budget for signaling in the
47. Cajochen C, Foy R, Dijk DJ. Frontal predominance of a
grey matter of the brain. J Cereb Blood Flow Metab 2001;
relative increase in sleep delta and theta EEG activity after
21(10):1133–45.
sleep loss in humans. Sleep Res Online 1999;2(3):65–9.
68. Buzsaki G, Csicsvari J, Dragoi G, Harris K, Henze D, Hirase H.
48. Cajochen C, Knoblauch V, Krauchi K, Renz C, Wirz- Homeostatic maintenance of neuronal excitability by burst
Justice A. Dynamics of frontal EEG activity, sleepiness discharges in vivo. Cereb Cortex 2002;12(9):893–9.
and body temperature under high and low sleep pressure. 69. Chklovskii DB, Schikorski T, Stevens CF. Wiring optimization
Neuroreport 2001;12(10):2277–81. in cortical circuits. Neuron 2002;34(3):341–7.
49. Borbely AA, Tobler I, Hanagasioglu M. Effect of sleep 70. Cohen-Cory S. The developing synapse: construction and
deprivation on sleep and EEG power spectra in the rat. modulation of synaptic structures and circuits. Science
Behav Brain Res 1984;14(3):171–82. 2002;298(5594):770–6.
50. Huber R, Deboer T, Tobler I. Effects of sleep deprivation on 71. Krakauer JW, Ghilardi MF, Mentis M, Barnes A, Veytsman M,
sleep and sleep EEG in three mouse strains: empirical data Eidelberg D, et al. Differential cortical and subcortical
and simulations. Brain Res 2000;857(1-2):8–19. activations in learning rotations and gains for reaching: a
51. Dijk DJ, Beersma DG, Daan S. EEG power density during nap PET study. J Neurophysiol 2004;91(2):924–33.
sleep: reflection of an hourglass measuring the duration of 72. Fenn KM, Nusbaum HC, Margoliash D. Consolidation during
prior wakefulness. J Biol Rhythms 1987;2(3):207–19. sleep of perceptual learning of spoken language. Nature
*52. Turrigiano GG. Homeostatic plasticity in neuronal net- 2003;425(6958):614–6.
works: the more things change, the more they stay the 73. Fischer S, Hallschmid M, Elsner AL, Born J. Sleep forms
same. Trends Neurosci 1999;22(5):221–7. memory for finger skills. Proc Natl Acad Sci USA 2002;
53. Desai NS, Cudmore RH, Nelson SB, Turrigiano GG. Critical 99(18):11987–91.
periods for experience-dependent synaptic scaling in visual 74. Gais S, Plihal W, Wagner U, Born J. Early sleep triggers
cortex. Nat Neurosci 2002;5(8):783–9. memory for early visual discrimination skills. Nat Neurosci
54. Kemp N, Bashir ZI. Long-term depression: a cascade of 2000;3(12):1335–9.
induction and expression mechanisms. Prog Neurobiol 75. Karni A, Tanne D, Rubenstein BS, Askenasy JJ, Sagi D.
2001;65(4):339–65. Dependence on REM sleep of overnight improvement of a
55. Malinow R, Malenka RC. AMPA receptor trafficking and perceptual skill [see comments]. Science 1994;265(5172):
synaptic plasticity. Annu Rev Neurosci 2002;25:103–26. 679–82.
56. Turrigiano GG. AMPA receptors unbound: membrane 76. Maquet P, Schwartz S, Passingham R, Frith C. Sleep-related
cycling and synaptic plasticity. Neuron 2000;26(1):5–8. consolidation of a visuomotor skill: brain mechanisms as
62 G. Tononi, C. Cirelli

assessed by functional magnetic resonance imaging. 95. Benca RM, Ancoli-Israel S, Moldofsky H. Special consider-
J Neurosci 2003;23(4):1432–40. ations in insomnia diagnosis and management: depressed,
77. Mednick S, Nakayama K, Stickgold R. Sleep-dependent learning: elderly, and chronic pain populations. J Clin Psychiatry
a nap is as good as a night. Nat Neurosci 2003;6(7):697–8. 2004;65(supplement 8):26–35.
78. Stickgold R, James L, Hobson JA. Visual discrimination 96. Drake CL, Roehrs T, Roth T. Insomnia causes, conse-
learning requires sleep after training. Nat Neurosci 2000; quences, and therapeutics: an overview. Depress Anxiety
3(12):1237–8. 2003;18(4):163–76.
79. Walker MP, Brakefield T, Morgan A, Hobson JA, Stickgold R. 97. Bonnet MH, Arand DL. Hyperarousal and insomnia. Sleep
Practice with sleep makes perfect: sleep-dependent motor Med Rev 1997;1(2):97–108.
skill learning. Neuron 2002;35(1):205–11. 98. Nofzinger EA, Buysse DJ, Germain A, Price JC, Miewald JM,
80. Sejnowski TJ, Destexhe A. Why do we sleep? Brain Res 2000; Kupfer DJ. Functional neuroimaging evidence for hyperar-
886(1–2):208–23. ousal in insomnia. Am J Psychiatry 2004;161(11):2126–8.
81. Rothman DL, Behar KL, Hyder F, Shulman RG. In vivo NMR 99. Morin CM, Hauri PJ, Espie CA, Spielman AJ, Buysse DJ,
studies of the glutamate neurotransmitter flux and Bootzin RR. Nonpharmacologic treatment of chronic
neuroenergetics: implications for brain function. Annu insomnia. An American academy of sleep medicine review.
Rev Physiol 2003;65:401–27. Sleep 1999;22(8):1134–56.
82. Pavlides C, Winson J. Influences of hippocampal place cell 100. Riemann D, Voderholzer U. Primary insomnia: a risk factor
firing in the awake state on the activity of these cells during to develop depression? J Affect Disord 2003;76(1–3):255–9.
subsequent sleep episodes. J Neurosci 1989;9(8):2907–18. 101. Riemann D, Berger M, Voderholzer U. Sleep and
83. Vyazovskiy VV, Welker E, Fritschy JM, Tobler I. Regional depression—results from psychobiological studies: an over-
pattern of metabolic activation is reflected in the sleep EEG view. Biol Psychol 2001;57(1–3):67–103.
after sleep deprivation combined with unilateral whisker 102. Armitage R. Microarchitectural findings in sleep EEG in
stimulation in mice. Eur J Neurosci 2004;20(5):1363–70. depression: diagnostic implications. Biol Psychiatry 1995;
84. Kuboyama T, Hori A, Sato T, Mikami T, Yamaki T, Ueda S. 37(2):72–84.
Changes in cerebral blood flow velocity in healthy young 103. Germain A, Nofzinger EA, Kupfer DJ, Buysse DJ. Neurobiol-
men during overnight sleep and while awake. Electro- ogy of non-REM sleep in depression: further evidence for
encephalogr Clin Neurophysiol 1997;102(2):125–31. hypofrontality and thalamic dysregulation. Am J Psychiatry
85. Hajak G, Klingelhofer J, Schulz-Varszegi M, Matzander G, 2004;161(10):1856–63.
Sander D, Conrad B, et al. Relationship between cerebral 104. Giedke H, Schwarzler F. Therapeutic use of sleep depri-
blood flow velocities and cerebral electrical activity in vation in depression. Sleep Med Rev 2002;6(5):361–77.
sleep. Sleep 1994;17(1):11–19. 105. Payne JL, Quiroz JA, Zarate Jr CA, Manji HK. Timing is
86. Maquet P, Peigneux P, Laureys S, Boly M, Dang-Vu T, everything: does the robust upregulation of noradrenergi-
Desseilles M, et al. Memory processing during human sleep cally regulated plasticity genes underlie the rapid anti-
as assessed by functional neuroimaging. Rev Neurol (Paris) depressant effects of sleep deprivation? Biol Psychiatry
2003;159(11):6S27–69. 2002;52(10):921–6.
87. Peigneux P, Laureys S, Fuchs S, Collette F, Perrin F, 106. Reinink E, Bouhuys AL, Gordijn MC, Van Den Hoofdakker RH.
Reggers J, et al. Are spatial memories strengthened in the Prediction of the antidepressant response to total sleep
human hippocampus during slow wave sleep? Neuron 2004; deprivation of depressed patients: longitudinal versus
44(3):535–45. single day assessment of diurnal mood variation. Biol
88. Peigneux P, Laureys S, Fuchs S, Destrebecqz A, Collette F, Psychiatry 1993;34(7):471–81.
Delbeuck X, et al. Learned material content and acquisition 107. Dijk DJ, Czeisler CA. Contribution of the circadian pacemaker
level modulate cerebral reactivation during posttraining and the sleep homeostat to sleep propensity, sleep structure,
rapid-eye-movements sleep. Neuroimage 2003;20(1):125–34. electroencephalographic slow waves, and sleep spindle
89. Wilson MA, McNaughton BL. Reactivation of hippocampal activity in humans. J Neurosci 1995;15(5 Pt 1):3526–38.
ensemble memories during sleep. Science 1994;265(5172): 108. Dijk DJ, Shanahan TL, Duffy JF, Ronda JM, Czeisler CA.
676–9. Variation of electroencephalographic activity during non-
90. Hirase H, Leinekugel X, Czurko A, Csicsvari J, Buzsaki G. rapid eye movement and rapid eye movement sleep with
Firing rates of hippocampal neurons are preserved during phase of circadian melatonin rhythm in humans. J Physiol
subsequent sleep episodes and modified by novel awake 1997;505(Pt 3):851–8.
experience. Proc Natl Acad Sci USA 2001;98(16):9386–90. 109. Davis GW, Bezprozvanny I. Maintaining the stability of
91. Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G. neural function: a homeostatic hypothesis. Annu Rev
Replay and time compression of recurring spike sequences Physiol 2001;63:847–69.
in the hippocampus. J Neurosci 1999;19(21):9497–507. 110. Brunner DP, Dijk DJ, Borbely AA. Repeated partial sleep
92. Pennartz CM, Lee E, Verheul J, Lipa P, Barnes CA, deprivation progressively changes in EEG during sleep and
McNaughton BL. The ventral striatum in off-line processing: wakefulness. Sleep 1993;16(2):100–13.
ensemble reactivation during sleep and modulation by 111. Van Dongen HP, Maislin G, Mullington JM, Dinges DF. The
hippocampal ripples. J Neurosci 2004;24(29):6446–56. cumulative cost of additional wakefulness: dose–response
93. Ribeiro S, Gervasoni D, Soares ES, Zhou Y, Lin SC, Pantoja J, effects on neurobehavioral functions and sleep physiology
et al. Long-lasting novelty-induced neuronal reverberation from chronic sleep restriction and total sleep deprivation.
during slow-wave sleep in multiple forebrain areas. PLoS Sleep 2003;26(2):117–26.
Biol 2004;2(1):E24. 112. Cirelli C, Huber R, Gopalakrishnan A, Southard TL, Tononi G.
94. Roth T, Roehrs T. Insomnia: epidemiology, characteristics, Locus ceruleus control of slow-wave homeostasis. J. Neurosci
and consequences. Clin Cornerstone 2003;5(3):5–15. 2005;25(18):4503–11.