NATURE|Vol 437|27 October 2005|doi:10.1038/nature04284
Hypothalamic regulation of sleep and
Clifford B. Saper1, Thomas E. Scammell1 & Jun Lu1
A series of findings over the past decade has begun to identify the brain circuitry and neurotransmitters
that regulate our daily cycles of sleep and wakefulness. The latter depends on a network of cell groups that
activate the thalamus and the cerebral cortex. A key switch in the hypothalamus shuts off this arousal
system during sleep. Other hypothalamic neurons stabilize the switch, and their absence results in
inappropriate switching of behavioural states, such as occurs in narcolepsy. These findings explain how
various drugs affect sleep and wakefulness, and provide the basis for a wide range of environmental
influences to shape wake–sleep cycles into the optimal pattern for survival.
In 1916, Baron Constantin von Economo, a Viennese neurologist,
began to see patients with a new type of encephalitis that specifically
attacked regions of the brain that regulate sleep and wakefulness1. This
disorder, which was eventually called encephalitis lethargica or von
Economo’s sleeping sickness, swept through Europe and North America during the second decade of the twentieth century; by the end of
the following decade it had apparently disappeared, as only sporadic
and unconvincing reports have appeared since. Although the virus
that caused it was never identified, von Economo was able to identify
the areas of the brain in which lesions caused specific alterations of
wake–sleep regulation (Fig. 1). Only during the past decade has the
accuracy of his observations come to be appreciated, as key components of the wake–sleep-regulatory system were found to reside at the
sites that von Economo first identified.
In this review, we cover recent advances in understanding the brain
circuitry that regulates sleep and produces wakefulness, including cell
groups in the brainstem, hypothalamus and basal forebrain (BF) that
are crucial for arousing the cerebral cortex and thalamus. These neurons are inhibited during sleep by a system of -aminobutyric acid
(GABA)-containing neurons, in which the ventrolateral preoptic
nucleus (VLPO) seems to have a key role. Mutual inhibition between
the arousal- and sleep-producing circuitry results in switching properties that define discrete wake and sleep states, with sharp transitions
between them. We examine how this switch is stabilized by orexin
neurons in the lateral hypothalamus (LHA), and why loss of these neurons results in narcolepsy. We also explore how basic drives affect this
system, including the homeostatic need to sleep that accumulates as a
function of prolonged wakefulness, the circadian drive that moulds the
daily cycles of sleep–wakefulness, and allostatic influences that adapt
sleep and wakefulness to external behavioural events. Interactions
with this circuitry might explain the effects of a range of drugs on sleep
The ascending arousal system promotes wakefulness
The majority of patients with von Economo’s encephalitis lethargica
slept excessively. Many slept for 20 or more hours per day, arising only
briefly to eat and drink. Their cognitive function was reasonably
intact, but they would soon return to sleep — a cycle that lasted for
Figure 1 | A drawing of the human brainstem, taken from von Economo’s
original work. It illustrates the site of the lesion (diagonal hatching) at the
junction of the brainstem and forebrain that caused prolonged sleepiness,
and the site of the lesion (horizontal hatching) in the anterior
hypothalamus that caused prolonged insomnia. The arrow points to a
region between the two, including the posterior lateral hypothalamus. Von
Economo suggested that narcolepsy was caused by lesions at this site.
many weeks before recovery1. Von Economo found that these patients
invariably had lesions at the junction of the midbrain and the diencephalon. He therefore proposed that there was an ascending arousal
system originating in the brainstem that kept the forebrain awake
(Fig. 1). During the years after the Second World War, investigators,
such as Moruzzi and Magoun and their many colleagues, showed that
this influence was mediated by an ascending arousal pathway that
begins in the rostral pons and runs through the midbrain reticular for-
Department of Neurology and Program in Neuroscience, Harvard Medical School, Beth Israel Deaconess Medical Center, Boston, Massachusetts, 02215, USA.
©2005 Nature Publishing Group
NATURE|Vol 437|27 October 2005
Figure 2 | A schematic drawing showing some key components of the
ascending arousal system. A major input to the relay and reticular nuclei
of the thalamus (yellow pathway) originates from cholinergic (ACh) cell
groups in the upper pons, the pedunculopontine (PPT) and laterodorsal
tegmental nuclei (LDT). These inputs facilitate thalamocortical
transmission. A second pathway (red) activates the cerebral cortex to
facilitate the processing of inputs from the thalamus. This arises from
neurons in the monoaminergic cell groups, including the
tuberomammillary nucleus (TMN) containing histamine (His), the A10
cell group containing dopamine (DA), the dorsal and median raphe nuclei
containing serotonin (5-HT), and the locus coeruleus (LC) containing
noradrenaline (NA) . This pathway also receives contributions from
peptidergic neurons in the lateral hypothalamus (LHA) containing orexin
(ORX) or melanin-concentrating hormone (MCH), and from basal
forebrain (BF) neurons that contain -aminobutyric acid (GABA) or ACh.
Note that all of these ascending pathways traverse the region at the junction
of the brainstem and forebrain where von Economo noted that lesions
caused profound sleepiness.
mation2; hence, the concept of the ‘ascending reticular activating system’ achieved wide currency2,3.
Studies in the 1970s and 1980s clarified the nature of this pathway
(Fig. 2). A key finding was that the ascending arousal system largely
originates from a series of well-defined cell groups with identified neurotransmitters4. This pathway has two major branches. The first
branch is an ascending pathway to the thalamus that activates the thalamic relay neurons that are crucial for transmission of information to
the cerebral cortex. The major source of upper brainstem input to the
thalamic-relay nuclei, as well as to the reticular nucleus of the thalamus, is a pair of acetylcholine-producing cell groups: the pedunculopontine and laterodorsal tegmental nuclei (PPT/LDT)5. The neurons
in the PPT/LDT fire most rapidly during wakefulness and rapid eye
movement (REM) sleep, which is the stage accompanied by cortical
activation, loss of muscle tone in the body and active dreams6. These
cells are much less active during non-REM (NREM) sleep, when cortical activity is slow. Their input to the reticular nucleus is crucial, as it
sits between the thalamic-relay nuclei and the cerebral cortex, acting
as a gating mechanism that can block transmission between the thalamus and cerebral cortex, which is important for wakefulness7. Other
inputs to the thalamic midline and intralaminar nuclei originate more
broadly in the upper brainstem, including the reticular formation, the
PPT/LDT, the monoaminergic systems (see below) and the
parabrachial nucleus8. The intralaminar and midline nuclei are also
believed to have a role in cortical arousal.
Figure 3 | A schematic drawing to show the key projections of the
ventrolateral preoptic nucleus (VLPO) to the main components of the
ascending arousal system. It includes the monoaminergic cell groups (red)
such as the tuberomammillary nucleus (TMN), the A10 cell group, the
raphe cell groups and the locus coeruleus (LC). It also innervates neurons
in the lateral hypothalamus (LHA; green), including the perifornical (PeF)
orexin (ORX) neurons, and interneurons in the cholinergic (ACh) cell
groups (yellow), the pedunculopontine (PPT) and laterodorsal tegmental
nuclei (LDT). Note that the VLPO neurons lie within the region outlined
by von Economo for the anterior hypothalamic lesion that caused
insomnia. 5-HT, serotonin; GABA, -aminobutyric acid; gal, galanin; NA,
noradrenaline; His, histamine.
The second branch of the ascending arousal system bypasses the
thalamus, instead activating neurons in the lateral hypothalamic area
and BF, and throughout the cerebral cortex4,9,10. This pathway originates from monoaminergic neurons in the upper brainstem and caudal hypothalamus, including the noradrenergic locus coeruleus (LC),
serotoninergic dorsal (DR) and median raphe nuclei, dopaminergic
ventral periaqueductal grey matter and histaminergic tuberomammillary neurons. The input to the cerebral cortex is augmented by
lateral hypothalamic peptidergic neurons (containing melanin-concentrating hormone (MCH) or orexin/hypocretin), and BF neurons
(containing acetylcholine or GABA). Lesions along this pathway, particularly in the LHA and rostral midbrain, produce the most profound
and long-lasting forms of sleepiness or even coma11,12. Neurons in each
of the monoaminergic nuclei that contribute to this pathway have the
property of firing fastest during wakefulness, slowing down during
NREM sleep and stopping altogether during REM sleep13–15. Orexin
neurons in the LHA are, similarly, most active during wakefulness16–18,
whereas MCH neurons are active during REM sleep19. Many BF neurons, including most cholinergic neurons, are active during both wake
and REM sleep20.
Given the anatomy and functions of these systems, it is not surprising that von Economo found that lesions at the junction of the midbrain and forebrain, which block both ascending pathways, produce
profound and long-lasting impairment of arousal.
The VLPO promotes sleep
Von Economo’s second major observation was an opposite reponse in
a small percentage of the victims of encephalitis lethargica1. Rather
than being sleepy, they became insomniac and slept for only a few
hours each day. Typically, they were extremely tired, but found it diffi-
©2005 Nature Publishing Group
NATURE|Vol 437|27 October 2005
cult to fall asleep, could sleep for only a short time, and then awoke and
were unable to fall back asleep. These patients had lesions involving the
basal ganglia and adjacent anterior hypothalamus. Later experiments
in animals identified a hypothalamic site involving the lateral preoptic area where lesions caused similar insomnia21,22.
During the 1980s and 1990s, investigators began to examine the
inputs to the monoaminergic cell groups that might be responsible for
their remarkable, stereotyped and coordinated changes in firing patterns associated with sleep. One such cell group, the VLPO, was found
to send outputs to all of the major cell groups in the hypothalamus and
brainstem that participate in arousal23 (Fig. 3). The VLPO neurons are
primarily active during sleep, and contain the inhibitory neurotransmitters, galanin and GABA24–26. These neurons form a dense cluster, as
well as a more diffuse extended part of the nucleus24,25.
These observations suggested that damage to the VLPO might have
caused the insomnia in von Economo’s patients. Experiments showed
that cell-specific lesions of the VLPO in animals reduced both NREM
and REM sleep by more than 50% (refs 27, 28). Lesions of the VLPO
cluster primarily reduced NREM sleep, whereas lesions of the
extended VLPO mainly disrupted REM sleep. The extended VLPO
neurons provide the main output from the VLPO to the LC and DR,
which are thought to be important in gating REM sleep27,29. By contrast, the VLPO cluster more heavily innervates the histaminergic neurons, which are closely linked to transitions between arousal and
The VLPO also receives afferents from each of the major
monoaminergic systems32. Both noradrenaline and serotonin inhibit
VLPO neurons33. The latter do not have histamine receptors, but the
tuberomammillary neurons also contain GABA34, which is inhibitory
to VLPO neurons35, as well as several other potentially inhibitory peptides, such as galanin and endomorphin36,37. Therefore, the VLPO can
be inhibited by the very arousal systems that it inhibits during sleep
loss of neurons in the human equivalent of the VLPO24 with ageing40–42.
Orexin/hypocretin neurons and state stability
In the years that followed the epidemic of encephalitis lethargica, several well-known clinicians, such as Wilson in London and Spiller in
Philadelphia, recorded a rash of cases of narcolepsy, a mysterious illness in which the patients had attacks of irresistible sleepiness, as well
as episodes of cataplexy during which emotional situations would
cause them to lose muscle tone and collapse to the floor43. Wilson
believed that at least some of the cases were post-encephalitic, and von
Economo proposed that the responsible pathology involved the posterior hypothalamus1.
In 1998, two groups of investigators simultaneously discovered a
pair of closely related neuropeptides, named orexins by one group and
hypocretins by the other44–46. These peptides are produced exclusively
by a cluster of neurons in the posterior half of the LHA. One year later,
two groups of investigators again simultaneously found that a lack of
orexins or their type 2 receptor can cause symptoms of narcolepsy in
experimental animals46,47. The following year, it was reported that
humans who have narcolepsy with cataplexy have few orexin neurons
The flip-flop switch
A circuit containing mutually inhibitory elements sets up a selfreinforcing loop, where activity in one of the competing sides shuts
down inhibitory inputs from the other side, and therefore disinhibits
its own action. Such a circuit is called a ‘flip-flop switch’ by electrical
engineers, who design them to produce discrete states with sharp transitions4. Flip-flop circuits tend to avoid transitional states, because
when either side begins to overcome the other, the switch ‘flips’ into
the alternative state. This flip-flop circuit model might explain why
wake–sleep transitions are often relatively abrupt (one ‘falls’ asleep and
suddenly wakens), and both humans and animals spend only a small
part of each day (typically 1–2%) in transitional states (Fig. 4). There
are obvious adaptive advantages to a wake–sleep system that makes
rapid and complete transitions, as it would be dangerous for animals
to have impaired alertness while engaging in waking behaviours, and
inefficient for them to spend their sleep periods half-awake.
However, flip-flop switches can also make unwanted transitions
with little warning. When a small perturbation gives one side a sudden
advantage, it can turn off the alternative state relatively abruptly (such
as falling asleep during a momentary lapse of attention while driving).
Interestingly, mathematical models show that when either side of a
flip-flop neural circuit is weakened, homeostatic forces cause the
switch to ride closer to its transition point during both states38. As a
result, there is an increase in transitions, both during the wake and the
sleep periods, regardless of which side is weakened. This is certainly
seen in animals with VLPO lesions, which fall asleep about twice as
often as normal animals, wake up much more often during their sleep
cycle and, on the whole, only sleep for about one-quarter as long per
bout28 — in other words, they wake up and are unable to fall back
asleep during the sleep cycle, but also are chronically tired, falling
asleep briefly and fitfully during the wake cycle, much like von
Economo’s patients. Interestingly, elderly people, who have similar
problems although perhaps on a lesser scale39, have been found to have
Figure 4 | A schematic diagram of the flip-flop switch model. During
wakefulness (a), the monoaminergic nuclei (red) inhibit the ventrolateral
preoptic nucleus (VLPO; purple), thereby relieving the inhibition of the
monoaminergic cells, and that of the orexin (ORX) neurons (green), and the
cholinergic pedunculopontine (PPT) and laterodorsal tegmental nuclei
(LDT; yellow). Because the VLPO neurons do not have orexin receptors, the
orexin neurons serve primarily to reinforce the monoaminergic tone, rather
than directly inhibiting the VLPO on their own. During sleep (b), the firing
of the VLPO neurons inhibits the monoaminergic cell groups, thereby
relieving their own inhibition. This also allows it to inhibit the orexin
neurons, further preventing monoaminergic activation that might
interrupt sleep. The direct mutual inhibition between the VLPO and the
monoaminergic cell groups forms a classic flip-flop switch, which produces
sharp transitions in state, but is relatively unstable. The addition of the
orexin neurons stabilizes the switch. 5-HT, serotonin; ACh, cholinergic;
eVLPO, extended ventrolateral preoptic nucleus; GABA, -aminobutyric
acid; gal, galanin; LC, locus coeruleus; NA, noradrenaline; PeF, perifornical;
REM, rapid eye movement; TMN, tuberomammillary nucleus.
©2005 Nature Publishing Group
NATURE|Vol 437|27 October 2005
in the LHA and low orexin levels in the cerebrospinal fluid48–50. So,
within a remarkably short period of time, the molecular basis of this
mysterious illness was identified.
Interestingly, few people with narcolepsy have mutations in either
the orexin ligand or receptor genes49,51. In most narcoleptics, the disease begins in the second or third decade of life, and the loss of orexin
neurons is remarkably specific (producing no injury to the adjacent
neurons that produce MCH)48,49. The cause is believed to be autoimmune, although convincing evidence for this hypothesis is still lacking, and it might be a neurodegenerative condition51. Other patients
with lesions of the posterior LHA due to a range of causes, such as
those first reported by von Economo and Wilson, also acquired narcolepsy.
The orexin neurons are mainly active during wakefulness and especially during motor activity when animals actively explore their environment16–18. They have ascending projections to the cerebral cortex,
as well as descending projections to all the monoaminergic and cholinergic cell groups of the arousal systems47,52. There are mutual projections between the VLPO neurons and the orexin neurons32,53,54, but the
former do not express orexin receptors55. Hence, the orexin neurons
reinforce the arousal systems, but probably do not directly inhibit the
VLPO (Fig. 4). This asymmetric relationship could help stabilize the
flip-flop switch, like a ‘finger’ on the switch that might prevent
unwanted transitions into sleep. The increase in homeostatic sleep
drive due to consolidated wakefulness might, in turn, help produce
consolidated sleep. Narcoleptic people and animals lack this influence
and behave as if their sleep flip-flop switch has been destabilized51,56.
They do not sleep more than normal individuals, but easily doze off
during the day and wake more often from sleep at night53, as the flipflop model would predict.
Homeostatic regulation of sleep
Although the purpose of sleep remains unknown, it clearly has a
restorative effect on the brain. Like other homeostatic systems, such as
that governing body temperature, which seek to recover to a set point
when perturbed, sleep deprivation is followed by extra recovery sleep
that is proportional to the sleep loss. An influential model of sleep regulation, proposed by Borbely and colleagues, includes both homeostatic and circadian drives for sleep57,58. The homeostatic influence is
believed to be due to some structure or substance that accumulates
‘need to sleep’ during prolonged wakefulness, and discharges this
homeostatic need during sleep. NREM and REM sleep probably have
separate homeostatic mechanisms, and, after a period of sleep deprivation, NREM sleep is typically repleted first.
The mechanisms for this homeostatic determinant remain unclear.
The VLPO neurons, for example, do not accumulate a need to sleep, as
they remain at a low level of activity after a prolonged period of sleep
deprivation, until the animal actually falls asleep23,27. At that point,
VLPO neurons fire about twice as fast as they do during normal
sleep27, implying that they are under the influence of, but distinct from,
the homeostatic factors that reflect sleep need.
Adenosine has been proposed as a homeostatic accumulator of the
need to sleep6,59,60. During prolonged wakefulness, the energy-producing systems in the brain run down, with exhaustion of brain glycogen reserves and depletion of ATP levels61,62. During prolonged
wakefulness, as ATP is degraded to ADP, AMP and eventually adenosine, extracellular adenosine levels rise in some parts of the brain,
including the BF63. Injection of adenosine or an adenosine A1 receptor agonist into the BF of cats6, or an adenosine A2a receptor agonist
near the VLPO in rats, causes sleep; the latter also results in expression
of Fos (a marker of neuronal activity) in VLPO neurons64. In addition,
adenosine might disinhibit the VLPO via presynaptic A1 receptors by
reducing inhibitory GABAergic inputs35. Hence, at least one mechanism for homeostatic sleep drive might be an accumulation of a sleeppromoting substance that enhances the activity of sleep-promoting
cells and reduces the activity of wake-promoting neurons. In this way,
adenosine, and perhaps other somnogens, can allow the activation of
the VLPO that is necessary to trigger a sleep episode.
Circadian regulation of sleep
Borbely also proposed a circadian influence on sleep and wakefulness
(process C) that is distinct from the homeostatic drive (process S) for
sleep57,58. Careful studies of humans in a forced desynchrony protocol,
where they experience a 28-hour ‘day’ without any cues as to external
time, have confirmed a strong 24-hour circadian rhythm in sleep
drive65. Recent studies in experimental animals have clarified how this
drive is maintained.
The suprachiasmatic nucleus (SCN) serves as the brain’s ‘master
clock’. Neurons in the SCN fire in a 24-hour cycle that is driven by a
transcriptional–translational loop, which persists even when the neurons are dissociated in cell culture66,67. Loss of the SCN abolishes the
circadian rhythms of a range of behaviours and physiological
Box 1 | Drugs that affect sleep and wakefulness
Many of the drugs that are used to regulate sleep and wakefulness are now
thought to act directly on the wake and sleep systems that have been
Wake-promoting drugs, such as amphetamines, methylphenidate
(Ritalin) and modafinil (Provigil), all act on the dopamine (DA)-reuptake
transporter. Their potency is roughly proportional to their ability to block
DA reuptake, and mice in which the DA-transporter gene has been deleted
fail to respond to this entire class of medications94. Modafinil might also
interfere with noradrenaline reuptake, and this dual mode of action might
explain why it can work without inducing other dopaminergic effects, such
Some sleep-promoting drugs interfere with the actions of the ascending
arousal system. For example, antihistamines that cross the blood–brain
barrier, such as diphenhydramine (Benadryl), cause drowsiness by blocking
the arousing influence of the central histaminergic system. This side effect
accounts for the popularity of antihistamines that do not cross the bloodbrain barrier and, hence, do not cause drowsiness. Centrally acting DA
antagonists also cause drowsiness96. These drugs are used primarily to
treat psychotic disorders, and this side effect is often a limiting factor in
drug acceptability. Paradoxically, D2-agonist drugs can also cause
drowsiness at low dosages96. The D2-presynaptic autoreceptor inhibits the
firing of dopaminergic neurons, and therefore results in reduced
dopaminergic activity. Similarly, dexmedetomidine, an -2 adrenergic
agonist, reduces firing of noradrenergic neurons, and is thought to induce
loss of wakefulness by disinhibiting the ventrolateral preoptic nucleus
(VLPO), which is ordinarily under noradrenergic inhibition during
wakefulness97 (Fig. 4).
The largest class of sleep-promoting drugs consists of those that
enhance the activity of -aminobutyric acid-A (GABA-A) receptors,
including barbiturates, benzodiazepines, chloral hydrate, ethanol and
most gaseous anaesthetics98. Few of these drugs are direct agonists
(that is, few compete with GABA for binding); rather, most bind to other
components of the GABA-A receptor and enhance the action of GABA at
the receptor. At low dosages, these drugs might act on the targets of the
VLPO, which also uses GABA as a neurotransmitter to silence the
arousal system. At higher dosages, they suppress firing in much of the
central nervous system. Benzodiazepines, such as diazepam (Valium),
act on GABA-A receptors that contain a -subunit and subunits of the
-1, -2, -3 or -5 class98. Newer ‘non-benzodiazepines’, such as
zolpidem (Ambien) or eszopiclone (Lunesta), also bind at - sites but
do not have the classic benzodiazepine structure and are more selective
for the 1-containing receptors, which are important in sedation but not
reduction of anxiety98. Gaboxadol binds to GABA-A receptors that
contain the -4 and (but not ) subunits. These receptors, which are at
highest concentration in the thalamus, limbic system and cerebral
cortex, are not accessible to benzodiazepines, and hence represent a
distinct drug target99. This might reduce the arousal drive from the
medial temporal and prefrontal cortex that is seen during insomnia, as
gaboxadol can activate Fos expression by neurons in the VLPO, and
increases the amount of slow-wave sleep100.
©2005 Nature Publishing Group
NATURE|Vol 437|27 October 2005
Figure 5 | A schematic diagram to illustrate the three-stage
integrator for circadian rhythms. The suprachiasmatic
nucleus (SCN) serves as a biological clock, but has few
outputs to sleep-regulatory systems. Most of its output
goes into the region in light brown, which includes the
ventral (vSPZ) and dorsal (dSPZ) subparaventricular
zone, and the dorsomedial nucleus of the hypothalamus
(DMH). Neurons in the vSPZ relay information necessary
for organizing daily cycles of wake–sleep, whereas dSPZ
neurons are crucial for rhythms of body temperature.
Outputs from the SPZ are integrated in the DMH with
other inputs, and DMH neurons drive circadian cycles of
sleep, activity, feeding and corticosteroid secretion. Cycles
of body temperature are maintained by dSPZ projections
back to the medial preoptic area (MPO), whereas the
DMN is the origin of projections to the VLPO for sleep
cycles, to the corticotropin-releasing hormone (CRH)
neurons of the paraventricular nucleus (PVH) for
corticosteroid cycles, and to the lateral hypothalamic
(LHA) orexin and melanin-concentrating hormone
neurons for wakefulness and feeding cycles. The
integrative steps in the SPZ and DMH allow circadian
rhythms to adapt to environmental stimuli, such as food
availability (for example, leptin and ghrelin action via the
ventromedial (VMH) and arcuate (ARC) nuclei), as well
as visceral sensory inputs, cognitive influences from the
prefrontal cortex and emotional inputs from the limbic
processes, including sleep, if the animals are not given other external
timing cues68. Under normal circumstances, the SCN is reset on a daily
basis by light inputs from the retina during the day, and by melatonin
secretion from the pineal gland during the dark cycle69,70. The light signal is received from a specialized set of retinal ganglion cells that contain the photopigment melanopsin71. These timing signals keep the
clock in synchrony with the external day–night cycle.
The link between the SCN and the sleep system has been the subject of much recent investigation (Fig. 5). The SCN has relatively modest projections to the VLPO or the orexin neurons32,53,54,72. However,
the bulk of its output is directed toward the adjacent subparaventricular zone (SPZ) and the dorsomedial nucleus of the hypothalamus
(DMH). The SPZ contains a ventral part, just above the SCN, and a
dorsal part, just below the paraventricular nucleus73. Cell-specific
lesions of the ventral SPZ disrupt the circadian rhythms of sleep and
wakefulness, as well as locomotor activity, but have minimal effects on
body-temperature rhythms. Conversely, lesions of the dorsal SPZ
severely impair circadian rhythms of body temperature, but not
wake–sleep or locomotor activity73. Therefore, direct projections from
the SCN to sleep or thermoregulatory regions are not sufficiently
strong to maintain the circadian rhythms of these functions, and the
relay neurons in the SPZ are required.
The SPZ also has relatively limited projections to the VLPO, orexin
neurons and other components of the wake-sleep-regulatory system32,53,54,74. However, a major target is the DMH73,74. This region
receives inputs from many more neurons in the SPZ than the SCN, so
the SPZ is in a position to amplify the output of the SCN. Cell-specific
lesions of the DMH also profoundly diminish circadian rhythms of
sleep and wakefulness, as well as locomotor activity, corticosteroid
secretion and feeding75. Interestingly, animals with DMH lesions sleep
about one hour more each day and have much less locomotor activity,
implying that the output of the DMH is mainly activating. This theme
is also reflected in the corticosteroid levels, which, in animals with
DMH lesions, remain at the lowest basal levels throughout the day.
Body temperature retains a normal circadian variation, but is about
0.5 °C lower than in control animals75.
The DMH is one of the largest sources of input to the VLPO and
orexin neurons32,54,75,76, and is crucial for conveying SCN influence to
the wake–sleep-regulatory system75,77. The DMH projection to the
VLPO comes largely from GABA-containing neurons (that is, those
that promote wakefulness by inhibiting sleep), and the projection to
the LHA originates from neurons containing glutamate and thyrotropin-releasing hormone (which should presumably be excitatory
and promote wakefulness)75. The DMH has relatively few direct outputs to the brainstem components of the ascending arousal system, but
the orexin neurons have extensive projections to these targets47,52,76.
Examination of Fos patterns in the DMH show that it contains many
more active neurons during wakefulness than during sleep78.
Why might the brain need such a complex three-stage pathway for
circadian control of sleep and other behaviours? The answer might lie
in the observation that, in a wide range of both nocturnal and diurnal
animals, the SCN is always active during the light cycle and the VLPO
is always active during the sleep cycle23,24. Therefore, in animals that
are nocturnal versus diurnal, there must be an intervening set of circuitry that allows the circadian cycle to be set at opposite phases,
despite an identical clock input and sleep-control system. In fact, the
circadian rhythms in many animals are not entirely fixed. For example, bats are often considered to be the quintessential nocturnal animal. This is true for bats in Finland during the summer, when there are
insects flying around at night for them to eat and predatory birds flying around during the day to discourage them. However, during the
cooler weather of the spring and autumn, few insects fly at night and
predatory birds have migrated to other climes. At these times of year,
the activity cycles of bats shift so that they can take advantage of the
food source that is available during the day, and they shift their activity pattern entirely to the daylight hours78,79.
Similarly, it is possible to invert many of the circadian rhythms of
laboratory rats by restricting their access to food to the daylight hours.
For example, if animals are only permitted to eat during the latter half
of the light period, they awaken, become active and have an increase
in body temperature a few hours before the food is due to be presented80. This behaviour persists even if the food is withheld
completely for 2 days; hence, this circadian pattern is not dependent
on external cues. The ability to alter the wake–sleep, activity, feeding,
body temperature and corticosteroid rhythms of these animals correlates with a shift in activation of the DMH (but not the SCN, which
©2005 Nature Publishing Group
NATURE|Vol 437|27 October 2005
stays locked to the original light–dark cycle) to the new wake cycle,
and lesions of the DMH prevent these behavioural shifts78 (J. J. Gooley
and C. B. Saper, unpublished observations). So, the DMH seems to
integrate clock information from the SCN and SPZ with feeding, temperature, social and other cues, providing animals with the flexibility
to adapt their behavioural and physiological cycles to the environment,
thereby maximizing their chances of survival (Fig. 5).
Allostatic regulation of sleep
The rapid expansion of our knowledge about the substrates of the regulation of sleep–wakefulness and circadian rhythms highlights the
lack of knowledge about how these systems are controlled in a complex
and ever-changing environment. Although we are beginning to understand the homeostatic drive for sleep, it is clear that sleep patterns and
the circadian cycle can be modified by external events, such as availability of food or external temperatures. Similarly, homeostatic and circadian drives for sleep can be overcome for brief periods when
external events demand an emergency response.
McEwen and Stellar, in 1993, proposed the term ‘allostatic’ drive81
for situations where “rather than maintaining constancy, the physiologic systems within the body fluctuate to meet demands from
external forces.” We know remarkably little about how these external
forces overcome the homeostatic and circadian systems or reset them
(Fig. 5). There are clearly cues from visceral sensory systems and
feeding regulatory systems to the arousal systems54,82–85. Visceral
inputs relayed through the nucleus of the solitary tract, such as gastric stretch, have a synchronizing sleep-inducing influence86,
whereas absence of sufficient food causes an arousing influence87.
However, we know little about the mechanisms by which cognitive
and emotional systems can act on the sleep or circadian control systems. Recent studies have emphasized the inputs to the SCN82,
VLPO32 and orexin neurons53,54 from corticolimbic sites, such as the
infralimbic cortex, ventral subiculum, lateral septum and bed
nucleus of the stria terminalis, as possible routes by which this influence might be expressed.
Positron-emission tomography studies during sleep in humans
with insomnia also show increased activity in corticolimbic sites,
including the medial prefrontal cortex and medial temporal lobe, compared with sleeping subjects without insomnia88. These inputs might
maintain a hyperaroused state, which can be essential in an emergency
situation, such as when a doctor must stay awake to care for a sick
patient overnight. However, when the arousal systems override the
homeostatic and circadian regulation of sleep during periods of behavioural stress or depression, the result might be unwanted and debilitating insomnia.
Understanding the interactions of these allostatic influences with the
sleep and circadian systems is a major challenge for the next decade.
However, learning to control our wake–sleep systems holds the
promise of improved health and cognitive performance (Box 1).
Individuals who lose even small amounts of sleep on a daily basis
show progressive impairment of cognitive performance and elevation of C-reactive protein, which predicts cardiovascular risk89–91.
Because older individuals sleep about half an hour less per day, it is
possible that at least some of their cognitive decline and increase in
cardiovascular disease might be explained by sleep restriction92. Similarly, sleep loss might impair performance among adolescents who
arise early for school, shift workers, overnight long-haul truckers and
even medical personnel working in hospitals93. The public-health
implications of sleep loss indicate that there is a great deal at stake in
working out the mechanisms that regulate our daily cycles of sleep
1. Von Economo, C. Sleep as a problem of localization. J. Nerv. Ment. Dis. 71, 249–259
2. Moruzzi, G. & Magoun, H. W. Brain stem reticular formation and activation of the EEG.
Electroencephalogr. Clin. Neurol. 1, 455–473 (1949).
3. Starzl, T. E., Taylor, C. W. & Magoun, H. W. Ascending conduction in reticular activating
system, with special reference to the diencephalon. J. Neurophysiol. 14, 461–477 (1951).
4. Saper, C. B., Chou, T. C. & Scammell, T. E. The sleep switch: hypothalamic control of sleep
and wakefulness. Trends Neurosci. 24, 726–731 (2001).
5. Hallanger, A.H., Levey, A. I., Lee, H. J., Rye, D. B. & Wainer, B. H. The origins of cholinergic
and other subcortical afferents to the thalamus in the rat. J. Comp. Neurol. 262, 104–124
6. Strecker, R. E. et al. Adenosinergic modulation of basal forebrain and preoptic/anterior
hypothalamic neuronal activity in the control of behavioral state. Behav. Brain Res. 115,
7. McCormick, D. A. Cholinergic and noradrenergic modulation of thalamocortical
processing. Trends Neurosci. 12, 215–220 (1989).
8. Krout, K. E., Belzer, R. E. & Loewy, A. D. Brainstem projections to midline and intralaminar
thalamic nuclei of the rat. J. Comp. Neurol. 448, 53–101 (2002).
9. Saper, C. B. Organization of cerebral cortical afferent systems in the rat. II.
Hypothalamocortical projections. J. Comp. Neurol. 237, 21 (1985).
10. Jones, B. E. Arousal systems. Front. Biosci. 8, S438–S451 (2003).
11. Ranson, S. W. Somnolence caused by hypothalamic lesions in monkeys. Arch Neurol.
Psychiatr. 41, 1–23 (1939).
12. Gerashchenko, D., Blanco-Centurion, C., Greco, M. A. & Shiromani, P. J. Effects of lateral
hypothalamic lesion with the neurotoxin hypocretin-2-saporin on sleep in Long-Evans
rats. Neuroscience 116, 223–235 (2003).
13. Aston-Jones, G. & Bloom, F. E. Activity of norepinephrine-containing locus coeruleus
neurons in behaving rats anticipates fluctuations in the sleep-waking cycle. J. Neurosci. 1,
14. Fornal, C., Auerbach, S. & Jacobs, B. L. Activity of serotonin-containing neurons in
nucleus raphe magnus in freely moving cats. Exp. Neurol. 88, 590–608 (1985).
15. Steininger, T. L., Alam, M. N., Gong, H., Szymusiak, R. & McGinty, D. Sleep-waking
discharge of neurons in the posterior lateral hypothalamus of the albino rat. Brain Res.
840, 138–147 (1999).
16. Estabrooke, I. V. et al. Fos expression in orexin neurons varies with behavioral state.
J. Neurosci. 21, 1656–1662 (2001).
17. Mileykovskiy, B. Y., Kiyashchenko, L. I. & Siegel, J. M. Behavioral correlates of activity in
identified hypocretin/orexin neurons. Neuron 46, 787–798 (2005).
18. Lee, M. G, Hassani, O. K. & Jones, B. E. Discharge of identified orexin/hypocretin neurons
across the wake-sleep cycle. J. Neurosci. 25, 6716–6720 (2005).
19. Verret, L. et al. A role of melanin-concentrating hormone producing neurons in the central
regulation of paradoxical sleep. BMC Neurosci. 4, 19 (2003).
20. Lee, M. G., Hassani, O. K., Alonso, A. & Jones, B. E. Cholinergic basal forebrain neurons
burst with theta during waking and paradoxical sleep. J. Neurosci. 25, 4365–4369 (2005).
21. Nauta, W. J. H. Hypothalamic regulation of sleep in rats. An experimental study.
J. Neurophysiol. 9, 285–314 (1946).
22. McGinty, D. J. & Sterman, M. B. Sleep suppression after basal forebrain lesions in the cat.
Science 160, 1253–1255 (1968).
23. Sherin, J. E., Shiromani, P. J., McCarley, R. W. & Saper, C. B. Activation of ventrolateral
preoptic neurons during sleep. Science 271, 216–219 (1996).
24. Gaus, S. E., Strecker, R. E., Tate, B. A., Parker, R. A. & Saper, C. B. Ventrolateral preoptic
nucleus contains sleep-active, galaninergic neurons in multiple mammalian species.
Neuroscience 115, 285–294 (2002).
25. Sherin, J. E., Elmquist, J. K., Torrealba, F. & Saper, C. B. Innervation of histaminergic
tuberomammillary neurons by GABAergic and galaninergic neurons in the ventrolateral
preoptic nucleus of the rat. J. Neurosci. 18, 4705–4721 (1998).
26. Szymusiak, R., Alam, N., Steininger, T. L. & McGinty, D. Sleep-waking discharge patterns
of ventrolateral preoptic/anterior hypothalamic neurons in rats. Brain Res. 803, 178–188
27. Lu, J. et al. Selective activation of the extended ventrolateral preoptic nucleus during rapid
eye movement sleep. J. Neurosci. 22, 4568–4576 (2002).
28. Lu, J., Greco, M. A., Shiromani, P. & Saper, C. B. Effect of lesions of the ventrolateral
preoptic nucleus on NREM and REM sleep. J. Neurosci. 20, 3830–3842 (2000).
29. Verret, L. et al. Localization of the neurones potentially responsible for the inhibition of
locus coeruleus noradrenergic neurones during paradoxical sleep in the rat. J. Comp.
Neurol. (in the press).
30. Ko, E. M., Estabrooke, I. V., McCarthy, M. & Scammell, T. E. Wake-related activity of
tuberomammillary neurons in rats. Brain Res. 992, 220–226 (2003).
31. John, J., Wu, M. F., Boehmer, L. N. & Siegel, J. M. Cataplexy-active neurons in the
hypothalamus: implications for the role of histamine in sleep and waking behavior.
Neuron 42, 619–634 (2004).
32. Chou, T. C. et al. Afferents to the ventrolateral preoptic nucleus. J. Neurosci. 22, 977–990
33. Gallopin, T. et al. Identification of sleep-promoting neurons in vitro. Nature 404, 992–995
34. Vincent, S. R., Hokfelt, T. & Wu, J.-Y. GABA neuron systems in hypothalamus and the
pituitary gland. Neuroendocrinology 34, 117–125 (1982).
35. Chamberlin, N. L. et al. Effects of adenosine on GABAergic synaptic inputs to identified
ventrolateral preoptic neurons. Neuroscience 119, 913–918 (2003).
36. Martin-Schild, S., Gerall, A. A., Kastin, A. J. & Zadina, J. E. Differential distribution of
endomorphin 1- and endomorphin 2-like immunoreactivities in the CNS of the rodent.
J. Comp Neurol. 405, 450–471 (1999).
37. Kohler, C. et al. Galanin immunoreactivity in hypothalamic neurons: further evidence for
multiple chemical messengers in the tuberomammillary nucleus. J. Comp. Neurol. 250,
38. Chou, T. C. Regulation of Wake-Sleep Timing: Circadian Rhythms and Bistability of SleepWake States, 82-99. Thesis, Harvard Univ. (2003).
39. Ancoli-Israel, S. & Kripke, D. F. Prevalent sleep problems in the aged. Biofeedback Self
Regul. 16, 349-359 (1991).
40.Hofman, M. A. & Swaab, D. F. The sexually dimorphic nucleus of the preoptic area in the
human brain: a comparative morphometric study. J. Anat. 164, 55–72 (1989).
©2005 Nature Publishing Group
NATURE|Vol 437|27 October 2005
41. Allen, L. S., Hines, M., Shryne, J. E. & Gorski, R. A. Two sexually dimorphic cell groups in
the human brain. J. Neurosci. 9, 497–506 (1989).
42. Byne, W. et al. The interstitial nuclei of the human anterior hypothalamus: an
investigation of sexual variation in volume and cell size, number and density. Brain Res.
856, 254–258 (2000).
43. Wilson, S. A. K. Modern Problems in Neurology (Edward Arnold, London, 1928).
44.Sakurai, T. et al. Orexins and orexin receptors: a family of hypothalamic neuropeptides
and G protein-coupled receptors that regulate feeding behavior. Cell 92, 573–585 (1998).
45. de Lecea, L. et al. The hypocretins: hypothalamus-specific peptides with neuroexcitatory
activity. Proc. Natl Acad. Sci. USA 95, 322–327 (1998).
46. Lin, L. et al. The sleep disorder canine narcolepsy is caused by a mutation in the
hypocretin (orexin) receptor 2 gene. Cell 98, 365–376 (1999).
47. Chemelli, R. M. et al. Narcolepsy in orexin knockout mice: molecular genetics of sleep
regulation. Cell 98, 437–451 (1999).
48. Thannickal, T. C. et al. Reduced number of hypocretin neurons in human narcolepsy.
Neuron 27, 469–474 (2000).
49. Peyron, C. et al. A mutation in a case of early onset narcolepsy and a generalized absence
of hypocretin peptides in human narcoleptic brains. Nature Med. 6, 991–997 (2000).
50. Ripley, B. et al. CSF hypocretin/orexin levels in narcolepsy and other neurological
conditions. Neurology 57, 2253–2258 (2001).
51. Scammell, T. E. The neurobiology, diagnosis, and treatment of narcolepsy. Ann. Neurol.
53, 154–166 (2003).
52. Peyron, C. et al. Neurons containing hypocretin (orexin) project to multiple neuronal
systems. J. Neurosci. 18, 9996–10015 (1998).
53. Sakurai, T. et al. Input of orexin/hypocretin neurons revealed by a genetically encoded
tracer in mice. Neuron 46, 297–308 (2005).
54. Yoshida, K., McCormack, S., Espana, R. A., Crocker, A. D. & Scammell, T. E. Afferents to
the orexin neurons. J. Comp. Neurol. (in the press).
55. Marcus, J. N. et al. Differential expression of orexin receptors 1 and 2 in the rat brain.
J. Comp. Neurol. 435, 6–25 (2001).
56 Mochizuki, T. et al. Behavioral state instability in orexin knock-out mice. J. Neurosci. 24,
57. Borbely, A. A. & Tobler, I. Brain Mechanisms of Sleep (eds McGinty, D. J. et al.) 35–44
(Raven, New York, 1985).
58. Achermann, P. & Borbely, A. A. Mathematical models of sleep regulation. Front. Biosci. 8,
59. Radulovacki, M., Virus, R. M., Djuricic-Nedelson, M. & Green, R. D. Adenosine analogs
and sleep in rats. J. Pharmacol. Exp. Ther. 228, 268–274 (1984).
60. Benington, J. H. & Heller, H. C. Restoration of brain energy metabolism as the function of
sleep. Prog. Neurobiol. 45, 347–360 (1995).
61. Kong, J. et al. Brain glycogen decreases with increased periods of wakefulness:
implications for homeostatic drive to sleep. J. Neurosci. 22, 5581–5587 (2002).
62. Shepel, P. N., Ramonet, D., Stevens, P. & Geiger, J. D. Purine level regulation during energy
depletion associated with graded excitatory stimulation in brain. Neurol. Res. 27, 139–148
63. Porkka-Heiskanen, T., Strecker, R. E. & McCarley, R. W. Brain site-specificity of
extracellular adenosine concentration changes during sleep deprivation and
spontaneous sleep: an in vivo microdialysis study. Neuroscience 99, 507–517 (2000).
64.Scammell, T. E. et al. An adenosine A2a agonist increases sleep and induces Fos in
ventrolateral preoptic neurons. Neuroscience 107, 653–663 (2001).
65. Dijk, D. J. & Czeisler, C. A. Contribution of the circadian pacemaker and the sleep
homeostat to sleep propensity, sleep structure, electroencephalographic slow waves,
and sleep spindle activity in humans. J. Neurosci. 15, 3526–3538 (1995).
66. Reppert, S. M. & Weaver, D. R. Coordination of circadian timing in mammals. Nature 418,
67 Jin, X. et al. A molecular mechanism regulating rhythmic output from the
suprachiasmatic circadian clock. Cell 96, 1–20 (1999).
68. Moore, R. Y. & Eichler, V. B. Loss of a circadian adrenal corticosterone rhythm following
suprachiasmatic lesions in the rat. Brain Res. 42, 201–206 (1972).
69. Johnson, R. F., Moore, R. Y. & Morin, L. P. Loss of entrainment and anatomical plasticity
after lesions of the hamster retinohypothalamic tract. Brain Res. 460, 297–313 (1988).
70. Cassone, V. M., Chesworth, M. J. & Armstrong, S. M. Entrainment of rat circadian
rhythms by daily injection of melatonin depends upon the hypothalamic suprachiasmatic
nuclei. Physiol. Behav. 36, 1111–1121 (1986).
71 Gooley, J. J., Lu, J., Fischer, D. & Saper, C. B. A broad role for melanopsin in nonvisual
photoreception. J. Neurosci. 23, 7093–7106 (2003).
72. Watts, A. G., Swanson, L. W. & Sanchez-Watts, G. Efferent projections of the
suprachiasmatic nucleus: I. Studies using anterograde transport of Phaseolus vulgaris
leucoagglutinin in the rat. J. Comp. Neurol. 258, 204–229 (1987).
73. Lu, J. et al. Contrasting effects of ibotenate lesions of the paraventricular nucleus and
subparaventricular zone on sleep-wake cycle and temperature regulation. J. Neurosci. 21,
74 Deurveilher, S. & Semba, K. Indirect projections from the suprachiasmatic nucleus to
major arousal-promoting cell groups in rat: implications for the circadian control of
behavioural state. Neuroscience 130,165–183 (2005).
75. Chou, T. C. et al. Critical role of dorsomedial hypothalamic nucleus in a wide range of
behavioral circadian rhythms. J. Neurosci. 23, 10691–10702 (2003).
76. Thompson, R., Swanson, L. W. & Canteras, N. Organization of projections from the
dorsomedial nucleus of the hypothalamus: a PHA-L study in the rat. J. Comp. Neurol. 376,
77. Aston-Jones, G., Chen, S., Zhu, Y. & Oshinsky, M. L. A neural circuit for circadian
regulation of arousal. Nature Neurosci. 4, 732–738 (2001).
78. Saper, C. B., Lu, J., Chou, T. C. & Gooley, J. The hypothalamic integrator for circadian
rhythms. Trends Neurosci. 28, 152–157 (2005).
79. Nyholm, H. Zur okologie von Myotis mystacinus (Leisl.) und M. daubentoni (Leisl.)
(Chiroptera). Ann. Zool. Fenn. 2, 77–123 (1955).
80.Stephan, F. K. The “other” circadian system: food as a Zeitgeber. J. Biol. Rhythms 17,
81. McEwen, B. S. & Stellar, E. Stress and the individual. Mechanisms leading to disease. Arch
Intern. Med. 153, 2093–2101 (1993).
82. Krout, K. E., Kawano, J., Mettenleiter, T. C. & Loewy, A. D. CNS inputs to the
suprachiasmatic nucleus of the rat. Neuroscience 110, 73–92 (2002).
83. Elmquist, J. K., Elias, C. F. & Saper, C. B. From lesions to leptin: hypothalamic control of
food intake and body weight. Neuron 22, 221–232 (1999).
84. Elmquist, J. K., Ahima, R. S., Elias, C. F., Flier, J. S. & Saper, C. B. Leptin activates distinct
projections from the dorsomedial and ventromedial hypothalamic nuclei. Proc. Natl Acad.
Sci. USA 95, 741–746 (1998).
85. Saper, C. B. in The Rat Nervous System (ed. Paxinos, G.) 761–796 (Elsevier Academic, San
86. Schachter, S. C. & Saper, C. B. Vagus nerve stimulation. Epilepsia 39, 677–686 (1998).
87. Yamanaka, A. et al. Hypothalamic orexin neurons regulate arousal according to energy
balance in mice. Neuron 38, 701–713 (2003).
88. Nofzinger, E. A. et al. Functional neuroimaging evidence for hyperarousal in insomnia. Am.
J. Psychiatry 161, 2126-2128 (2004).
89. Meier-Ewert, H. K. et al. Effect of sleep loss on C-reactive protein, an inflammatory
marker of cardiovascular risk. J. Am. Coll. Cardiol. 43, 678–683 (2004).
90. Van Dongen, H. P., Maislin, G., Mullington, J. M. & Dinges, D. F. The cumulative cost of
additional wakefulness: dose-response effects on neurobehavioral functions and sleep
physiology from chronic sleep restriction and total sleep deprivation. Sleep 26, 117–126
91. Doran, S. M., Van Dongen, H. P. & Dinges, D. F. Sustained attention performance during
sleep deprivation: evidence of state instability. Arch. Ital. Biol. 139, 253–267 (2001).
92. Prinz, P. N. Age impairments in sleep, metabolic, and immune functions. Exp. Gerontol. 39,
93. Landrigan, C. P. et al. Effect of reducing interns’ work hours on serious medical errors in
intensive care units. N. Engl. J. Med. 351, 1838–1848 (2004).
94. Wisor, J. et al. Dopaminergic role in stimulant-induced wakefulness. J. Neurosci. 21,
95. Saper, C. B. & Scammell, T. E. Modafinil: a drug in search of a mechanism. Sleep 27, 11–12
96. Roth, T. et al. Assessment of sleepiness and unintended sleep in Parkinson's disease
patients taking dopamine agonists. Sleep Med. 4, 275–280 (2003).
97. Nelson, L. E. et al. The ±-2 adrenoreceptor agonist dexmedetomidine converges on an
endogenous sleep-promoting pathway to exert its sedative effects. Anaesthesiology 98,
98. Rudolf, U. & Mohler, H. Analysis of GABAA receptor function and dissection of the
pharmacology of benzodiazepines and general anesthetics through mouse genetics.
Annu. Rev. Pharmacol. Toxicol. 44, 475–498 (2004).
99. Wisden, W., Laurie, D. J., Monyer, H. & Seeburg, P. The distribution of 13 GABAA receptor
subunit mRNAs in the rat brain. I. Telencephalon, diencephalons, mesencephalon.
J. Neurosci. 12, 1040–1062 (1992).
100. Lancel, M., Wetter, T. C., Steiger, A. & Mathias, S. Effect of the GABAA agonist
gaboxadol on nocturnal sleep and hormone secretion in healthy elderly subjects. Am. J.
Physiol. Endocrinol. Metab. 281, E130–E137 (2001).
Acknowledgements This work was supported by United States Public Health
Author Information Reprints and permissions information is available at
npg.nature.com/reprintsandpermissions. The authors declare
competing financial interests: details accompany the paper on
www.nature.com/nature. Correspondence should be addressed to C.F.B.
©2005 Nature Publishing Group