Sleep, Learning, and Dreams: Off-line Memory Reprocessing

R. Stickgold et al.
Science 294, 1052 (2001);
DOI: 10.1126/science.1063530

This copy is for your personal, non-commercial use only.

If you wish to distribute this article to others, you can order high-quality copies for your
colleagues, clients, or customers by clicking here.
Permission to republish or repurpose articles or portions of articles can be obtained by
following the guidelines here.

The following resources related to this article are available online at

www.sciencemag.org (this information is current as of August 28, 2014 ):

Downloaded from www.sciencemag.org on August 28, 2014

Updated information and services, including high-resolution figures, can be found in the online
version of this article at:
http://www.sciencemag.org/content/294/5544/1052.full.html
This article has been cited by 237 article(s) on the ISI Web of Science
This article has been cited by 62 articles hosted by HighWire Press; see:
http://www.sciencemag.org/content/294/5544/1052.full.html#related-urls
This article appears in the following subject collections:
Neuroscience
http://www.sciencemag.org/cgi/collection/neuroscience

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the
American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright
2001 by the American Association for the Advancement of Science; all rights reserved. The title Science is a
registered trademark of AAAS.
 SLEEP, DREAMS, AND MEMORY
69. T. R. Barrett, B. R. Ekstrand, J. Exp. Psychol. 96, 321 74. S. Gais, W. Plihal, U. Wagner, J. Born, Nature Neuro- National de la Recherche Scientifique (FNRS) (Bel-
(1972). sci. 3, 1335 (2000). gium) and presently is a Research Fellow at the
70. M. J. Fowler, M. J. Sullivan, B. R. Ekstrand, Science 179, 75. L. De Gennaro et al., Electroencephalogr. Clin. Neu- Wellcome Department of Cognitive Neurology, Uni-
302 (1973). rophysiol. 95, 252 (1995). versity College London (UK). The work reported here
71. R. Yaroush, M. J. Sullivan, B. R. Ekstrand, J. Exp. 76. J. H. Herman, H. P. Roffwarg, Science 220, 1074 is supported by the FNRS (Belgium), by the Univer-
Psychol. 88, 361 (1971). (1983). sity of Liège, and by the Queen Elisabeth Medical
72. A. Giuditta, et al., Behav. Brain Res. 69, 157 (1995). 77. L. De Gennaro, M. Ferrara, L. Urbani, M. Bertini, Exp. Foundation. I thank C. Frith for reviewing an earlier
73. R. Stickgold, L. Scott, C. Rittenhouse, J. A. Hobson, J. Brain Res. 130, 105 (2000). version of the manuscript and two anonymous re-
Cognit. Neurosci. 11, 182 (1999). 78. P.M. is a Senior Research Assistant at the Fonds viewers for thoughtful comments.

REVIEW

Sleep, Learning, and Dreams: Off-line

Memory Reprocessing
R. Stickgold,1* J. A. Hobson,1 R. Fosse,1,2 M. Fosse1

Converging evidence and new research methodologies from across the procedural training when rats show increased
neurosciences permit the neuroscientific study of the role of sleep in amounts of REM and during which REM
off-line memory reprocessing, as well as the nature and function of deprivation leads to diminished retention. For
dreaming. Evidence supports a role for sleep in the consolidation of an many of the early REM deprivation studies,
array of learning and memory tasks. In addition, new methodologies allow the apparent decrease in recall after depriva-
the experimental manipulation of dream content at sleep onset, permit- tion may be the consequence of deprivation-
ting an objective and scientific study of this dream formation and a induced stress (2, 4). But other studies (23)
renewed search for the possible functions of dreaming and the biological have demonstrated performance decrements
processes subserving it. 20 hours after REM deprivation, but not 8 to
16 hours after deprivation (24, 25). This is the
It is 200 years since David Hartley (1) first ory of dream interpretation (14), there has opposite of what a stress model would pre-
suggested that dreaming might alter the been a frustrating dearth of scientific evi- dict. Other studies have shown effects as long
strength of associative memories, but the dence concerning the mechanism of dream as a week after REM deprivation (26).
basic proposition that either sleep or construction and its possible functions. One These findings in no way suggest that
dreaming plays a role in the off-line repro- such function might be as part of a multilevel REM is critical for all memory consolidation.
cessing of memories remains hotly debated system of sleep-dependent learning and Substantial memory consolidation occurs
(2– 4 ). Recent developments in molecular memory reprocessing, wherein dreams would during normal waking, and many memory
genetics, neurophysiology, and the cogni- be the conscious manifestation of these pro- tasks are unaffected by subsequent REM de-
tive neurosciences have produced a striking cesses. New approaches described below of- privation (2, 4, 15). Nor is there clear evi-
body of research that provides converging fer a methodology for experimentally ap- dence that REM sleep enhances subsequent
evidence for an important role of sleep in proaching these questions. encoding (27). Furthermore, memory consol-
learning and the reprocessing of memories idation is most likely not the only function of
(5). Behavioral Studies of Learning and REM sleep, not explaining, for example, the
On the basis of patterns of brain electri- Memory in Sleep decrease in REM during the first year of life
cal activity measured in the electroenceph- Behavioral studies of sleep and learning in (2).
alogram (EEG), eye movements, and mus- humans and animals, neurochemical and neu- In humans, posttraining REM deprivation
cle tone (6 ), sleep can be broadly divided rophysiological studies of the brain basis of impairs retention of procedural learning (20,
into rapid eye movement sleep (REM) and possible sleep-dependent memory process- 28). Declarative memory tasks in general
non–rapid eye movement sleep (NREM), ing, and neurocognitive studies of informa- have not shown any sleep dependence [e.g.,
with the human REM-NREM cycle typical- tion processing during sleep provide evidence (29)], although some studies have suggested
ly having a 90-min period. Recent evidence for an interdependence between sleep, learn- that deep, slow-wave sleep (SWS) early in
strengthens the hypothesis that sleep plays ing, and memory. Still, considerable contro- the night may aid in their consolidation (30,
a role in learning and memory processing at versy surrounds the question (2, 4, 15). For 31).
several levels, including the REM-depen- additional discussions of these questions, see REM may also enhance the processing of
dent developmental wiring of binocular the accompanying reviews by Maquet (5) and emotional memories. There is enhanced re-
cells in visual cortex (7, 8), procedural Siegel (16). call for emotionally salient memories after
learning of a visual discrimination task (9 – Research into sleep and memory began in periods of sleep rich in REM (32), and sev-
12), and the development of problem-solv- earnest after the discovery of REM in 1953 eral older studies similarly support a role for
ing skills (13). (17). Since then, a wide range of animal REM in processing emotional memories (27,
In contrast, since Freud proposed his the studies have supported the hypothesis that 33–36). In addition, shortenings of REM la-
REM plays a critical role in learning (18 –21). tencies and increases in REM densities have
A meta-analysis concluded that REM sleep been reported in major depression (37, 38),
1
Laboratory of Neurophysiology and Department of
Psychiatry, Harvard Medical School, Boston, MA
plays a critical role in the consolidation of the state of bereavement (37, 39), war-related
02115, USA. 2Institute of Psychology, University of procedural learning but not of declarative anxiety (40), and, more generally, posttrau-
Oslo, Box 1094 Blindem, N-0317 Oslo, Norway. memory (22). In a synthesis of the animal matic stress disorder (41).
*To whom correspondences should be addressed. E- literature, Smith proposed the existence of Some of the strongest evidence for human
mail: [email protected] “REM windows” (18), periods of time after learning being sleep dependent comes from a

1052 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org

 SLEEP, DREAMS, AND MEMORY
visual texture discrimination task (10, 42, (52) might facilitate information flow from hance LTP, considered critical for hippocam-
43). On this task, improvement is not seen the hippocampus to the neocortex (53). pal memory formation (49). Neural network
until after posttraining sleep (Fig. 1A) (11), In contrast, theta rhythms in REM may simulations (54) have suggested that such an
and sleep deprivation on the night after train- support information transfer from neocortex alternating “hippocampo-neocortical dialog”
ing eliminates all benefits of training, even to hippocampus (53), where theta waves en- (53) could enhance the encoding of hip-
when measured after two full nights of recov-
ery sleep (Fig. 1B) (12). Karni et al. (10)
found no improvement after a night with
selective REM deprivation, but did see im-
provement after selective SWS deprivation.
Other studies suggest that both SWS and
REM are required (Fig. 1, C and D) (11), a
result in keeping with the two-step model
proposed by Giuditta et al. (44) for the con-
solidation of learning in rats, but contrary to
what would have been expected on the basis
of the Karni et al. study (10).
More generally, studies suggest that REM
might modify neocortical networks in gener-
al, rather than simply those involved in pro-
cedural learning, with REM effects reported
for learning of complex logic games (13), for
foreign language acquisition (45), and after
intensive studying (19). The fact that REM
appears to play little or no role in memory
consolidation on simple tests of declarative
memory has led some researchers (2, 4, 46)
to doubt that REM plays any role in memory,
citing studies that conclude that long-term
REM suppression in depressed and narcolep-
tic humans produces “no adverse effects on
cognition/memory” (4, p. 874). Unfortunate-
ly, none of the studies cited in these reviews
looked at performance on either procedural or
complex learning tasks after a night of post- Fig. 1. Sleep-dependent learning of a texture discrimination task: Subjects were trained and then
retested at a later time. Each subject was retested only once, and each point represents a separate
training sleep. Instead, they used almost en- group of subjects. (A) Improvement across a night’s sleep. Subjects were trained and then retested
tirely simple declarative memory tasks retest- either 3 to 12 hours later on the same day (open circles) or after 8 to 24 hours after a night’s sleep
ed within minutes after training, where we (filled circles). All told, n ⫽ 57, with n ⫽ 7 to 9 for individual points. Error bars ⫽ SEM. (B)
would not expect to find any effect of REM Improvement across a week. Solid bars: Subjects were retested the same day as training (day ⫽ 0)
deprivation. A resolution of this question or after 1 to 7 days (n ⫽ 122). Open bar: Subjects were sleep deprived the night after training and
must await the testing of these patient popu- retested after a total of 3 days (n ⫽ 11). Error bars ⫽ SEM. (C) Overnight improvement was
correlated with both the amount of SWS (solid squares) and of REM (open circles) in each quarter
lations with tasks such as the texture discrim- of the night, and the Pearson correlation coefficient was plotted (n ⫽ 12). Significant correlations
ination task, which otherwise appear to be were seen for the percentages of time spent in SWS during the first quarter of the night (SWS1)
REM dependent. and in REM during the last quarter of the night (REM4). (D) SWS1 was multiplied by REM4 for each
subject and plotted against the individual’s overnight improvement. From Stickgold et al. (11, 12).
Sleep Architecture and Physiology
Probing the mechanisms underlying the pos-
Table 1. Brain physiology shifts across sleep states. Human sleep is divided into REM and NREM, with
sible roles of sleep in memory processing NREM further subdivided into sleep onset (stage 1 sleep), light NREM (stage 2), and SWS (stages 3 and
requires knowledge of the complex physiol- 4). The physiological parameters listed here are characterized by robust state-dependent changes that are
ogy of sleep. The electrophysiologically de- thought to be linked to sleep-dependent learning and memory reprocessing. Arrow represents changes in
fined stages of sleep differ along several di- activity relative to waking. See text for explanations.
mensions, some of which are shown in Table
1. Researchers have speculated that many of Stage 2
Physiological correlates of sleep stages REM SWS
these phenomena might contribute to learning NREM
and memory processing in sleep. Synchronous brain electrical activity 4 to 6 Hz 12 to 14 Hz 0.5 to 4 Hz
Synchronous brain activity. Steriade (47, Eye movements _ + +
48) has hypothesized that high rates of ⬃10- Muscle tone + 2 2
Hz firing of neocortical neurons during the External inputs + 2 2
long-lasting depolarization phase of SWS os- Hippocampal-neocortical dialog (HC-NC) NC3 HC ? HC3 NC
cillations might induce long-term potentia- Cholinergic modulation (ACh) _ 2 2
tion (LTP) at cortical synapses (49 –51), Aminergic modulation (NE and 5-HT ) + 2 2
Glucocorticoids (GC) (2) – (1)
which could serve to reorganize or respecify
Frontal activation (DLPFC) + ? 2
connections within neural networks and func- 1 ? 2
Limbic activation (e.g., anterior cingulate cortex)
tionally connect distant cortical regions. Sim- Sensory cortices 1 ? 2
ilarly, sharp wave potentials seen in SWS

www.sciencemag.org SCIENCE VOL 294 2 NOVEMBER 2001 1053

 SLEEP, DREAMS, AND MEMORY
pocampally dependent memories in the neo- all brain regions become less active in SWS sumably reflects the consolidation of learn-
cortex (3). compared with waking. But although many re- ing. These and related findings led to the
Phasic ponto-geniculo-occipital (PGO) gions remain relatively inactive in REM (75), suggestion that the PKA signaling pathway
waves, which activate visual and motor cor- dorsolateral prefrontal cortex (DLPFC), which mediates sleep-dependent learning and mem-
tices as well as the amygdala and hippocam- is involved in decision making and memory, ory processes (79).
pus (55), are seen during the transition from becomes further inactivated in this state (75). At A qualitatively different form of sleep-
NREM to REM and throughout REM. These the same time, several midline limbic struc- dependent synaptic plasticity has been dem-
waves may play an important role in memory tures, including both the anterior cingulate and onstrated during early postnatal development
consolidation in the rat (56) and have been medial orbitofrontal cortices and the amygdala, of the cat visual system (8, 84). Studies com-
proposed to reactivate memory traces during (76) become reactivated to levels at or above bining monocular visual deprivation with
REM dreaming (57). waking levels (77). sleep deprivation (7) suggest that sleep con-
It has been hypothesized that gamma Taken together, these studies of sleep tributes as much to developmental changes in
waves (⬃40 Hz) mediate the binding of sen- physiology provide considerable circumstan- synaptic connectivity as does visual experi-
sory features in both waking consciousness tial evidence for both REM and NREM play- ence, presumably by consolidating or en-
(58) and REM dreaming (59), although there ing important roles in memory consolidation. hancing the changes that occurred during the
are no data linking them to learning or mem- Direct evidence for such roles comes from prior period of monocular deprivation.
ory consolidation in sleep. studies that link physiological processes with Similarly, molecular changes accompany
Neuronal replay. Stronger evidence of the behavioral outcomes. the reorganization of the receptive fields of
possible role of these processes in learning barrel cortex neurons after trimming of mys-
and memory comes from analyses of neuro- Sleep Physiology, Memory Processing, tacial whiskers in the rat. These fluctuations,
nal activity in the rat hippocampus. During and Behavior in nerve growth factor (85) as well as levels
sleep, replay of recent waking patterns of Hennevin et al. investigated the ability of the of mRNA for the GABA-synthetic enzyme
neuronal activity is seen within the CA1 layer brain to encode and consolidate memories glutamate decarboxylase (86), are modulated
of the hippocampus. This reactivation is seen during REM through direct brain stimulation by sleep deprivation. Stimulation of whiskers
during SWS for about half an hour after (24). Their results indicate that both the con- also affects subsequent sleep EEG patterns
learning (60, 61) and in REM after 24 hours solidation of learning and the formation of (87).
(62, 63). Although this replay in SWS may new associations can be mediated by pontine
simply reflect continued activity unrelated to reticular formation (PRF) activation during Cognitive Processing in REM and
sleep (64), the presence of replay in REM sleep (78). In addition, a correlation between NREM
only after 24 hours demonstrates that these an increased density of PRF-generated PGO Studies of dreaming often investigate aspects
patterns of neuronal activation are specifical- waves during posttraining REM and subse- of this conscious experience that are poten-
ly reactivated during REM. Neuronal replay quent improved task performance has been tially relevant to our understanding of learn-
during REM can be synchronized with theta reported (56), suggesting that PGO waves ing and memory. One approach, which seeks
wave activity, shifting from in-phase (i.e., facilitate learning in the rat. to identify isomorphisms between the basic
coincident with peaks of theta waves) to out- Biochemical aspects of memory consoli- neurophysiological features of REM and the
of-phase (coincident with theta troughs) over dation have recently been reviewed (79, 80). formal properties of REM dreams (57), led to
4 to 7 days (65), a time course similar to that Both protein synthesis and phosphokinase A the first physiologically based model of
over which initially hippocampally depen- (PKA) are required for hippocampally medi- dream construction (57, 88) and has yielded
dent memories become independent of the ated learning, and their inhibition during rich and novel data. We have recently re-
hippocampus (66, 67). Such a shift might REM windows produces effects similar to viewed this literature in detail (89).
produce a switch from LTP and memory those produced by REM deprivation (81, 82). An alternate approach is to actually mea-
consolidation to long-term depression (LTD) In addition, exposure to an enriched learning sure cognitive processes within minutes of
(50, 68) and memory erasure (69) after effec- environment induces the immediate early awakenings from REM and NREM. During
tive transfer of the memory to the neocortex. gene zif-268 during subsequent REM (83). this period of “sleep inertia” (90), the brain is
A similar replay during sleep has been found Zif-268 expression normally coincides with thought to remain in a state similar to that of
for neurons mediating vocal learning in song synaptic modification and, during REM, pre- the prior sleep period (91).
birds (70), and this may mediate the consol-
idation of this learning.
Neuromodulators. The REM-NREM cy- Fig. 2. The ultradian cycle and
information processing. Changes
cle also displays marked shifts in levels of in cholinergic neuromodulation
neuromodulators in the brain. Brainstem sys- and hippocampo-neocortical
tems that control the REM-NREM cycle in- communication are superim-
clude the noradrenergic (NE) locus coer- posed on the 90-min human
uleus, the serotonergic (5-HT) dorsal Raphe REM-NREM cycle across the
nucleus, and the cholinergic (ACh) nuclei of night. The slow shift from SWS
domination to REM domination
the dorsolateral pons (71). Whereas NREM is across the night is seen in
characterized by decreases in all three neuro- amounts of SWS and REM, as
modulators compared with waking, ACh lev- well as in the duration of REM
els in REM are equal to or higher (72) than periods and in both the ampli-
during wake (Fig. 2), and levels of NE and tude and frequency of rapid eye
5-HT drop to near zero (73). movements within REM. The
cholinergic neuromodulation is
Regional brain activation. Finally, positron presumed to follow this pattern
emission tomography (PET) studies have dem- because rapid eye movements
onstrated unique patterns of regional brain ac- parallel the activity of brainstem
tivation across wake-sleep states (74). Almost cholinergic neurons. From Stickgold et al. (3).

1054 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org

 SLEEP, DREAMS, AND MEMORY
Using this technique, we have shown (92) from verbal reports of dreaming. Such reports memory of an event, recalled as an integrated
that semantic priming favors weaker associ- indicate, for example, that as the brain state whole, with the actual waking event (or “ep-
ations after REM awakenings than after progresses from quiet waking to sleep onset, isode”) replayed in one’s mind. Episodic
NREM awakenings (Fig. 3) and that solving NREM, and finally REM, hallucinations in- memories are thought to consist of multiple
anagrams is similarly enhanced after REM crease sharply in frequency, whereas directed hippocampally linked memory traces located
compared with NREM awakenings (93). thinking gradually decreases (Fig. 4) (98 – within neocortical regions and dependent on
Both of these findings support the contention 101). Along with the increase in hallucina- the hippocampus for their integrated recall
that REM favors more “fluid thinking” than tions, REM dreams show a parallel increase (105). In contrast, dream researchers normal-
NREM, perhaps as a result of the decreased in bizarre, hyperassociative elements (89) ly ask if the source of a dream element is a
aminergic and increased cholinergic modula- and emotions (102, 103). waking event, independent of how memories
tion found in this state (94). Changes in sleep states appear to be ac- of the event are stored in the brain. Thus, if a
More generally, the cognitive changes companied by a shift in the sources of mem- subject has a phone conversation with an old
seen during REM may be the combined result ories incorporated into the dreams as well. In friend in the evening and subsequently
of three physiological characteristics of a recent meta-analysis, it was found that dreams of mountain climbing with the friend,
REM: (i) the shift in neuromodulatory bal- when subjects were asked to identify the the dream element is judged to have an epi-
ance from aminergic to cholinergic, (ii) the waking sources for dream elements, episodic sodic memory source, even though the dream
decreased activity in DLPFC and increased memory sources were less frequently identi- element shows almost no similarity to the
activity in both the anterior cingulate cortex fied in REM than in NREM or at sleep onset, event and clearly is not, itself, a replay of an
and amygdala (75–77), and (iii) the de- paralleling the decline in directed thinking episodic memory.
creased outflow of information from hip- across these stages (104) (Fig. 4). In fact, when subjects identify waking
pocampus to neocortex (53). Taken together, Dream elements often appear to arise events as the sources of dream elements, the
these findings suggest that the brain in REM from memories of waking events. But the fact dreams themselves rarely replay episodic
is tuned more for the processing of associa- that an element can be traced to a specific memories. When 364 dream elements from
tive memories than for the simple consolida- waking event does not necessarily mean that 299 dream reports with identified origins in
tion of recent memory traces and may ex- an episodic memory was used for dream con- prior waking were analyzed, only 1 to 2%
plain, in part, various features of REM struction. Episodic memories are defined as a were found to have these properties of epi-
dreams, including their bizarre, hyperassocia-
tive quality (95) and minimal incorporation
Fig. 4. Memory sources,
of episodic memories (96, 97). thoughts, and halluci-
nations. Percentage of
Dreams and Memory dream elements (“the-
Sources of dream elements in waking and matic units”) with iden-
sleep. Dreams presumably reflect the activa- tified episodic memory
tion and recombination of memories, and sources and percentage
of dreams containing
both these memories and associations to them directed thinking and
may be altered in some ways in the process. hallucinating. Data for
But which memory systems are activated dur- episodic memory
ing dreaming remains uncertain. sources are taken from
Evidence of memory activation comes Baylor and Cavallero
(104) and for thinking
and hallucinating from
Fosse et al. (101). For
episodic memory
sources, sleep onset
(SO): n ⫽ 27; REM and
NREM: n ⫽ 93. For
thinking and hallucinating, n ⫽ 16.

Table 2. Memory systems and dream elements. Subjects recorded 299 dream reports (criterion A)
containing 364 dream elements (criterion B) indicated by the subjects to have identifiable sources in
waking thoughts and events of the preceding 2 weeks. Although subjects ascribed the sources of 147 of
these dream elements to prior events (criterion C), the remaining 217 elements were ascribed not to
events, but to prior waking thoughts. Of the 147 dream elements that could conceivably represent
replays of episodic memories, only 38 (10% of all elements) occurred in the same location in both the
dream and waking event (criterion D), and only 12 (3% of all elements) conserved at least two other
aspects of the waking event (e.g., characters and actions). When these 12 dream elements were compared
by external judges with their ascribed waking sources, only five were rated as probable replays of episodic
memories and only three additional elements were judged as possible replays, representing a total of five
Fig. 3. Semantic priming across wake-sleep to eight instances (1 to 2%) of possible episodic replay. Data from Fosse et al. (96, 97).
states. Priming is defined as the decrease in
reaction times when identifying a target word Criterion Subjects Reports Elements
that is preceded by a semantically related
“prime” word compared with identifying a tar- A All reports with content 29 299 –
get preceded by an unrelated prime. Priming B Elements with waking sources 27 194 364 (100%)
(ms) ⫽ RTunrelated ⫺ RTrelated. Within-state C Elements with episodic sources 22 104 147 (40%)
comparison p values are shown in bars below D ⫹Conserved location 17 31 38 (10%)
the graph. From Stickgold et al. (92). PM: wake E ⫹Additional conserved features 9 11 12 (3%)
subjects, tested in the afternoon, n ⫽ 20; REM F ⫹Judged episodic 4 to 6 5 to 8 5 to 8 (1 to 2%)
and NREM: n ⫽ 44.

www.sciencemag.org SCIENCE VOL 294 2 NOVEMBER 2001 1055

 SLEEP, DREAMS, AND MEMORY
sodic memories. Instead, the dream elements amnesic subjects (113). Nevertheless, the neo- play an important role in the selection of
normally reflect only one or two aspects of cortical sources of these images were not sim- memories for incorporation into dreams, the
the waking experience (Table 2). ply stored sensory representations of recent dreams themselves often show little or no
This suggests that the brain sources for stimuli, as Tetris players occasionally reported emotional content (114). They thus seem to
dream elements are not hippocampally medi- images from past versions of Tetris and Alpine access and integrate memories and emotions
ated episodic memories, but cortical traces of Racers reported images from actual skiing in a manner uniquely different from that seen,
discrete components of the episodic memo- (115). for example, in REM. The nature of NREM
ries, which then presumably are combined These initial experimental studies have dreaming and how it does and does not differ
with associated semantic memories. With demonstrated that hypnagogic dreaming in- from REM dreaming remains a confusing
dorsolateral prefrontal cortex deactivated in volves (i) a high rate of incorporation of field (46, 89) and is beyond the scope of this
both REM and NREM (75, 76, 106, 107) and memories of events from the day or (ii) from review.
the hippocampal formation producing only older related memories, with (iii) a preference Because REM dreams are normally the
minimal cortical output in REM (53), actual for emotionally salient material but (iv) with- longest and the most visually intense, bizarre,
episodic memories may be inaccessible and out high dream affect (114, 115) and (v) and emotional, researchers have proposed
hence irrelevant to the dream construction without hippocampal or medial temporal lobe that the unique physiology of REM must
process. Such a conclusion would profoundly involvement. Although the sleep onset period contribute to this more intense dream produc-
alter the standard conceptualization of dream- differs from normal NREM and REM sleep tion. In 1977, Hobson and McCarley (57)
ing and would strongly constrain any models both in its characteristic dream features (117, proposed that REM dreams are initiated by
of dream construction and function. Further 118) and polysomnographic features (6), chaotic brainstem activity that then activates
evidence in support of this conclusion comes these findings nevertheless further constrain cortical regions, where the “brain/mind”
from studies of sleep onset dreaming, re- the shape of any general theory explaining makes the best sense it can out of a noisy
viewed below. the nature and function of dreaming. signal, bringing forth, rather than hiding,
Emotions and dreams. Emotions may play meaning. Almost 25 years later, our increased
a central role in the functioning of the brain- Modeling Dream Theory understanding of brain systems makes a more
mind during dreaming. In REM, the central A comprehensive theory of dreaming must detailed picture possible.
nucleus of the amygdala plays a crucial role address two questions: how dreams are con- During REM, limbic forebrain struc-
in the activation of medial prefrontal cortical structed and what purpose that construction tures and the amygdala are activated while
structures associated with the highest order process might serve. To answer the first ques- both DLPFC and the locus coeruleus be-
regulation of emotions (76, 108, 109). This tion, we need to show (i) how neurophysiol- come less active. This presumably inhibits
adds to the deactivation of DLPFC, normally ogy sets the stage for memory selection, (ii) the ability of DLPFC to allocate attentional
associated with higher cognitive functions how it favors associative processes that pro- resources (and the dreaming brain classi-
(110), in REM. Thus, the brain appears to be duce bizarre and unrecognizable representa- cally pays little attention to bizarre incon-
biased toward emotional processing in this tions of memories, and (iii) how these ele- gruities in dreams). At the same time, the
state. ments are combined with others into a narra- inhibition of hippocampal outflow would
As noted earlier, there is evidence for both tive, often with high emotional content. To prevent the reactivation of episodic memo-
emotion-enhanced REM and REM-facilitated answer the second question, we need to know ries (53). Dreams would thus be construct-
retention of emotionally salient memories. if, how, and why the dream construction pro- ed largely from those primarily weak neo-
Moreover, both depression (111) and the cess is behaviorally, psychologically, and cortical associations available during REM
presleep viewing of unpleasant films (112) psychosocially useful. Theories of dreaming (92). Although the process of incorporation
correlate with reports of negative emotion in continue to appear unabated (46, 57, 89, of these weak associates is unknown, we
early night REM dreaming. How specific 119 –121), but each theory addresses only a predict that associated emotions, mediated
aspects of emotional events affect dream con- subset of the questions necessary for a com- by both the amygdala and medial orbito-
struction remains obscure, as it has been dif- prehensive theory, and all contain enormous frontal cortex, play an important role. Thus,
ficult to reliably induce the incorporation of explanatory gaps. the resulting dreams would appear to be not
waking events or emotions into subsequent In the context of a multilevel system of only unpredictable and bizarre but highly
dreams. sleep-dependent memory reprocessing, emotional as well.
Sleep onset dreaming. Many of the prob- dreams represent the conscious awareness of We hypothesize that these features reflect
lems associated with identifying dream sources complex brain systems involved in the repro- an attempt, on the part of the brain, to identify
can be eliminated by studying hypnagogic cessing of emotions and memories during and evaluate novel cortical associations in the
dreams, which occur at sleep onset. These trun- sleep. But in discussing the functions of sleep light of emotions mediated by limbic struc-
cated dreams show robust incorporation of day- and dreaming, it is important to remember tures activated during REM. This would be in
time experiences and are experimentally con- that neither is a uniform process. Sleep onset, keeping with the proposed role in waking of
trollable. We have manipulated hypnagogic NREM, and REM each are characterized by a these structures in the identification of mis-
dream content by having subjects play the vid- unique physiology and normative dream matches between expected and actual behav-
eo game Tetris (113) or the arcade style down- structure. As such, there is a need to discuss ioral outcomes (122–125) and would also
hill skiing simulator, Alpine Racer II (114, each of them separately. explain the similarities seen between cholin-
115). Reports were collected from subjects in Hypnagogic dreams normally lack the bi- ergic and PGO activity in the amygdala dur-
their own homes, with their sleep monitored by zarreness, self-representation, emotions, and ing REM on the one hand and during alerting
the Nightcap sleep monitoring system, rather narrative complexities common to REM and orienting responses in awake animals on
than standard polysomnography (116, 117). dreams. Although they are often tightly the other (126 –128). Such evaluations could
Using these games, we obtained sleep onset linked to prior waking activities, they can then lead to the strengthening or weakening
reports of images of Tetris or downhill skiing in also display associated memories from the of specific activated associations, providing
up to 89% of subjects and 42% of first-night distant past. Here, as in REM, it appears that the functional consequence of REM dream-
reports (114, 115), with no difference in fre- the hippocampal episodic memory system is ing. Although this model is highly specula-
quency or content between normal and densely inactive. And although emotions appear to tive, it is only through such integration of the

1056 2 NOVEMBER 2001 VOL 294 SCIENCE www.sciencemag.org

 SLEEP, DREAMS, AND MEMORY
converging neuroscientific and psychological 49. T. Otto, H. Eichenbaum, S. Wiener, C. Wible, Hip- 92. R. Stickgold, L. Scott, C. Rittenhouse, J. A. Hobson, J.
pocampus 1, 181 (1991). Cognit. Neurosci. 11, 182 (1999).
data that we can hope to construct a new
50. C. Pavlides, Y. J. Greenstein, M. Grudman, J. Winson, 93. M. P. Walker, C. Liston, J. A. Hobson, R. Stickgold, in
cognitive neuroscience of dreaming. Brain Res. 439, 383 (1988). preparation.
51. P. T. Huerta, J. E. Lisman, Neuron 15, 1053 (1995). 94. D. Q. Beversdorf, J. D. Hughes, B. A. Steinberg, L. D.
References and Notes 52. J. J. Chrobak, G. Buzsáki, J. Neurosci. 14, 1660 Lewis, K. M. Heilman, Neuroreport 10, 2763 (1999).
1. D. Hartley, Observations on Man, His Frame, His (1994). 95. A. N. Mamelak, J. A. Hobson, J. Cognit. Neurosci. 1,
Duty and His Expectations ( Johnson, London, 1791). 53. G. Buzsáki, Cerebr. Cortex 6, 81 (1996). 201 (1989).
2. J. A. Horne, Neurosci. Biobehav. Rev. 24, 777 (2000). 54. G. E. Hinton, P. Dayan, B. J. Frey, R. M. Neal, Science 96. M. J. Fosse, R. Stickgold, R. Fosse, J. A. Hobson, Sleep
3. R. Stickgold, Trends Cognit. Sci. 2, 484 (1998). 268, 1158 (1995). 24, A179 (2001).
4. R. P. Vertes, K. E. Eastman, Behav. Brain Sci. 23, 867 97. M. J. Fosse, thesis, Harvard University, Cambridge,
55. J. M. Calvo, A. Fernandez-Guardiola, Sleep 7, 202
MA (2001).
(2000). (1984).
5. P. Maquet, Science 294, 1048 (2001).
6. A. Rechtschaffen, A. Kales, A Manual of Standardized
56. S. Datta, J. Neurosci. 20, 8607 (2000).
57. J. A. Hobson, R. W. McCarley, Am. J. Psychiatry 134,
99. 㛬㛬㛬㛬
98. D. Foulkes, J. Abnorm. Social Psychol. 65, 14 (1962).
, P. S. Spear, J. D. Symonds, J. Abnorm.
Psychol. 71, 280 (1966).
Terminology, Techniques and Scoring System for 1335 (1977). 100. A. Rechtschaffen, P. Verdone, J. Wheaton, Can.
Sleep Stages of Human Subjects (Brain Information 58. A. K. Engel, P. Fries, P. Konig, M. Brecht, W. Singer, J. Psychiatry 8, 409 (1963).
Service, Univ. of California, Los Angeles, 1968). Conscious. Cognit. 8, 128 (1999). 101. R. Fosse, R. Stickgold, J. A. Hobson, Psychol. Sci. 12,
7. M. G. Frank, N. P. Issa, M. P. Stryker, Neuron 30, 275 59. R. Llinas, U. Ribary, Proc. Natl. Acad. Sci. U.S.A. 90, 30 (2001).
(2001). 2078 (1993). 102. I. Strauch, B. Meier, In Search of Dreams: Results of
8. J. P. Shaffery, H. P. Roffwarg, S. G. Speciale, G. A. 60. C. Pavlides, J. Winson, J. Neurosci. 9, 2907 (1989). Experimental Dream Research (State University of
Marks, Brain Res. 114, 109 (1999). 61. M. A. Wilson, B. L. McNaughton, Science 265, 676 New York Press, Albany, NY, 1996).
9. S. Gais, W. Plihal, U. Wagner, J. Born, Nature Neu- (1994). 103. R. Fosse, R. Stickgold, J. A. Hobson, Sleep, in press.
rosci. 3, 1335 (2000). 62. K. Louie, M. A. Wilson, Neuron 29, 145 (2001). 104. G. W. Baylor, C. Cavallero, Sleep 24, 165 (2001).
10. A. Karni, D. Tanne, B. S. Rubenstein, J. J. M. Askenasy, 63. H. Hirase, X. Leinekugel, J. Csicsvari, A. Czurko, G. 105. D. L. Schacter, E. Tulving, Memory Systems 1994
D. Sagi, Science 265, 679 (1994). Buzsaki, J. Neurosci. 21, RC145 (2001). (MIT Press, Cambridge, MA, 1994).
11. R. Stickgold, D. Whidbee, B. Schirmer, V. Patel, J. A. 64. J. A. Hobson, R. A. Stickgold, Curr. Biol. 5, 35 (1995). 106. P. Maquet et al., J. Neurosci. 17, 2807 (1997).
Hobson, J. Cognit. Neurosci. 12, 246 (2000). 65. G. R. Poe, D. A. Nitz, B. L. McNaughton, C. A. Barnes, 107. E. A. Nofzinger, M. A. Mintun, M. B. Wiseman, D. J.
12. R. Stickgold, L. James, J. A. Hobson, Nature Neurosci. Brain Res. 855, 176 (2000). Kupfer, R. Y. Moore, Brain Res. 770, 192 (1997).
2, 1237 (2000). 66. J. J. Kim, M. S. Fanselow, Science 256, 675 (1992). 108. E. A. Nofzinger et al., Soc. Neurosci. Abstr. 22, 27
13. C. Smith, in The Functions of Dreaming, M. K. A. 67. G. Winocur, Behav. Brain Res. 38, 154 (1990). (1996).
Moffitt, R. Hoffman, Ed. (SUNY Press, New York, 68. C. Holscher, R. Anwyl, M. J. Rowan, J. Neurosci. 17, 109. A. R. Damasio et al., Nature Neurosci. 3, 1049
1993), pp. 341–361. 6470 (1997). (2000).
14. S. Freud, The Interpretation of Dreams, J. Strachey, 69. F. Crick, G. Mitchison, Nature 304, 111 (1983). 110. P. S. Goldman-Rakic, Philos. Trans. R. Soc. London
Ed. (Basic Books, New York, 1900). 70. A. S. Dave, D. Magoliash, Science 290, 812 (2000). Ser. B Biol. Sci. 351, 1445 (1996).
15. J. A. Horne, M. J. McGrath, Biol. Psychol. 18, 165 71. J. A. Hobson, R. W. McCarley, P. W. Wyzinski, Sci- 111. R. Cartwright, A. Luten, M. Young, P. Mercer, M.
(1984). ence 189, 55 (1975). Bears, Psychiatry Res. 81, 1 (1998).
16. J. Siegel, Science 294, 1058 (2001). 72. H. Kametani, H. Kawamura, Life Sci. 47, 421 (1990). 112. C. Lauer, D. Riemann, R. Lund, M. Berger, Psycho-
17. E. Aserinsky, N. Kleitman, Science 118, 361 (1953). 73. Changes in both ACh and 5-HT have been proposed physiology 24, 263 (1987).
as mediators of sleep-dependent memory consoli-
19. 㛬㛬㛬㛬
18. C. Smith, Neurosci. Biobehav. Rev. 9, 157 (1985).
, L. Lapp, Sleep Res. 16, 211 (1987). dation (79, 129). Cortisol levels also vary across the
night and appear to affect memory consolidation
113. R. Stickgold et al., Science 290, 350 (2000).
114. K. M. Emberger, thesis, Harvard University, Cam-
21. 㛬㛬㛬㛬
20. C. Smith, Behav. Brain Res. 69, 137 (1995).
, G. M. Rose, Physiol. Behav. 59, 93 (1996). during sleep (130).
74. J. A. Hobson, E. F. Pace-Schott, R. Stickgold, D. Kahn,
bridge, MA (2001).
115. R. Stickgold, J. A. Hobson, K. Emberger, Soc. Neuro-
23. 㛬㛬㛬㛬
22. C. Smith, Behav. Brain Res. 78, 49 (1996).
, J. M. Conway, G. M. Rose, Neurobiol. Learn. Curr. Opin. Neurobiol. 8, 239 (1998).
75. A. R. Braun et al., Brain 120, 1173 (1997).
sci. Abstr., in press.
116. J. A. Hobson, Stickgold, Conscious. Cognit. 3, 1
Mem. 69, 211 (1998).
24. E. Hennevin, B. Hars, C. Maho, V. Bloch, Behav. Brain 76. P. Maquet et al., Nature 383, 163 (1996). (1994).
Res. 69, 125 (1995). 77. J. A. Hobson, R. Stickgold, E. F. Pace-Schott, Neuro- 117. J. T. Rowley, R. Stickgold, J. A. Hobson, Conscious.
25. C. Smith, S. Butler, Physiol. Behav. 29, 469 (1982). report 9, R1 (1998). Cognit. 7, 67 (1998).
26. C. Smith, C. MacNeill, Psychobiology 21, 109 (1993). 78. Her findings indicate that during REM, the rat brain 118. D. Foulkes, G. Vogel, J. Abnorm. Psychol. 70, 231
27. M. H. McGrath, D. B. Cohen, Psychiatr. Bull. 85, 24 can (i) respond at the neuronal level to associations (1965).
(1978). learned during prior waking, (ii) form new associa- 119. M. Jouvet, The Paradox of Sleep: The Story of
28. C. Smith, M. Whittaker, Sleep Res. 16, 536 (1987). tions after simultaneous stimulation of the central Dreaming (MIT Press, Cambridge, MA, 1999).
29. V. Castaldo, V. Krynicki, J. Goldstein, Percept. Mot. gray and the medial geniculate nucleus, and (iii) 120. A. Revonsuo, Behav. Brain Sci. 23, 877 (2000).
Skills 39, 1023 (1974). enhance the benefits of posttraining REM through 121. M. Solms, The Neuropsychology of Dreams: A

31. 㛬㛬㛬㛬
30. W. Plihal, J. Born, J. Cognit. Neurosci. 9, 534 (1997).
, J. Born, Psychophysiology 36, 571 (1999).
direct stimulation of the mesencephalic reticular
formation (MRF), presumably through augmenta-
Clinico-Anatomical Study (Lawrence Erlbaum Asso-
ciates, Mahwah, NJ, 1997).
32. U. Wagner, S. Gais, J. Born, Learn. Mem. 8, 112 tion of processes mediated by the pontine reticular 122. W. J. Gehring, R. T. Knight, Nature Neurosci. 3, 516
(2001). formation. In addition, posttraining MRF stimulation (2000).
33. C. Grieser, R. Greenberg, R. H. Harrison, J. Abnorm. decreases the normal posttraining REM increase as 123. W. J. Gehring, M. G. H. Coles, D. E. Meyer, E. A.
Psychol. 80, 280 (1972). well as the deleterious effects of posttraining REM Donchin, in Perspectives of Event-Related Potentials
34. R. D. Cartwright et al., Psychophysiology 12, 561 deprivation. Research, G. Karmos et al., Eds. (Elsevier, Amster-
(1975). 79. L. Graves, A. Pack, T. Abel, Trends Neurosci. 24, 237 dam, 1995), pp. 261–272.
35. R. Greenberg, R. Pillard, C. Pearlman, Psychosom. (2001). 124. M. Falkenstein, J. Hohnsbein, J. Hoorman, in Per-
Med. 34, 257 (1972). 80. G. Tononi, C. Cirelli, Arch. Ital. Biol. 139, 221 (2001). spectives of Event-Related Potentials Research, G.
36. I. Lewin, D. Gombosh, in Sleep: Physiology, Bio- 81. B. M. Gutwein, P. J. Shiromani, W. Fishbein, Pharm. Karmos et al., Eds. (Elsevier, Amsterdam, 1995), pp.
chemistry, Psychology, Pharmacology, Clinical Im- Biochem. Behav. 12, 377 (1980). 287–296.
plications, W. P. Koella, P. Levin, Eds. (Karger, Basel, 82. R. Bourtchouladze et al., Learn. Mem. 5, 365 (1998). 125. C. S. Carter et al., Science 280, 747 (1998).
Switzerland, 1973), pp. 399 – 403. 83. S. Ribeiro, V. Goyal, C. Mello, C. Pavlides, Learn. 126. A. R. Morrison, L. D. Sanford, R. J. Ross, in Rapid Eye
37. R. Cartwright, Arch. Gen. Psychiatry 40, 197 (1983). Mem. 6, 500 (1999). Movement Sleep, B. N. Mallick, S. Inoue, Eds. (Dek-
38. D. J. Kupfer, F. G. Foster, Lancet 2, 648 (1972). 84. J. P. Shaffery et al., Dev. Brain Res. 97, 51 (1996). ker, New York, 1999), pp. 51– 68.
39. C. I. Reynolds et al., Biol. Psychiatry 34, 791 (1993). 85. J. A. Brandt et al., Brain Res. 898, 105 (2001). 127. P. J. Whalen, Curr. Dir. Psychol. Sci. 7, 177 (1998).
40. R. Greenberg, C. A. Pearlman, D. Gampel, Br. J. Med. 86. L. Churchill et al., Sleep 24, 261 (2001). 128. J. M. Calvo, K. Simon-Arceo, in Handbook of Behav-
Psychol. 45, 27 (1972). 87. V. Vyazovskiy, A. A. Borbely, I. Tobler, Sleep Res. 9, ioral State Control: Molecular and Cellular Mecha-
41. R. J. Ross, W. A. Ball, K. A. Sullivan, S. N. Caroff, 367 (2000). nisms, R. Lydic, H. A. Baghdoyan, Eds. (CRC Press,
Am. J. Psychiatry 146, 697 (1989). 88. J. A. Hobson, in Sleep and Cognition, R. Bootzin, J. Boca Raton, Fl, 1999), pp. 391– 406.
42. A. Karni, D. Sagi, Proc. Natl. Acad. Sci. U.S.A. 88, Kihlstrom, D. Schacter, Eds. (American Psychological 129. M. E. Hasselmo, Trends Cognit. Sci. 3, 351 (1999).

43. 㛬㛬㛬㛬
4966 (1991).
, Nature 365, 250 (1993).
Association, Washington, DC, 1990), pp. 25– 40.
89. J. A. Hobson, E. F. Pace-Schott, R. Stickgold, Behav.
130. W. Plihal, R. Pietrowsky, J. Born, Psychoneuroendo-
crinology 24, 313 (1999).
44. A. Giuditta et al., Behav. Brain Res. 69, 157 (1995). Brain Sci. 23, 793 (2000). 131. This research was supported by grants from NIH,
45. J. DeKoninck, D. Lorrain, G. Christ, G. Proulx, D. 90. A. Lubin, D. Hord, M. L. Tracy, L. C. Johnson, Psycho- the Norwegian Research Council (MH 48,832 and
Coulombe, Int. J. Psychophysiol. 8, 43 (1989). physiology 13, 334 (1976). DA 11,744), and the MacArthur Foundation’s
46. M. Solms, Behav. Brain Sci. 23, 843 (2000). 91. D. F. Dinges, in Sleep and Cognition, R. Bootzin, J. Mind-Body Network. The authors thank M. Walk-

48. 㛬㛬㛬㛬
47. M. Steriade, Neuroscience 101, 243 (2000).
, Trends Neurosci. 22, 337 (1999).
Kihlstrom, D. Schacter, Eds. (American Psychological
Association, Washington, DC, 1990), pp. 159 –178.
er and E. Pace-Schott for valuable comments on
the manuscript.

www.sciencemag.org SCIENCE VOL 294 2 NOVEMBER 2001 1057