the modulation of memory by sleep - open computing facilitysanl/reprints/...functional changes in...

This article was originally published in the Encyclopedia of Sleep published by Elsevier, and the attached copy is provided by Elsevier for the author's benefit and for the benefit of the author's institution, for non-commercial research and educational use including without limitation use in

instruction at your institution, sending it to specific colleagues who you know, and providing a copy to your institution’s administrator.

All other uses, reproduction and distribution, including without limitation commercial reprints, selling or licensing copies or access, or posting on open internet sites, your personal or institution’s

website or repository, are prohibited. For exceptions, permission may be sought for such use through Elsevier's permissions site at:

http://www.elsevier.com/locate/permissionusematerial Saletin J.M. and Walker M.P. (2013) The Modulation of Memory by Sleep. In: Kushida C. (ed.) The

Encyclopedia of Sleep, vol. 1, pp. 503-512. Waltham, MA: Academic Press.

© 2013 Elsevier Inc. All rights reserved.

Author's personal copy

The Modulation of Memory by SleepJ M Saletin and M P Walker, University of California, Berkeley, CA, USA

ã 2013 Elsevier Inc. All rights reserved.

GlossaryDeclarative memory: Memory for items that can be verbally

declared, including autobiographical memories (episodic

memories) and knowledge-based facts (semantic

memories), dependent largely upon structures in the medial

temporal lobe.

Functional magnetic resonance imaging (fMRI):

A neuroimaging technique allowing for the measurement of

functional changes in brain activity measured by the

response of regional blood-oxygen to changes in magnetic

field strength.

Long-term potentiation (LTP): One possible cellular

mechanism of long-term memory, indicated by

long-standing change in the excitability of a neuron,

following original stimulation; LTP is modulated by both

sleep and sleep deprivation.

Medial temporal lobe (MTL): Medial surface of the

temporal lobe of the cerebral cortex, consisting of the

hippocampus, entorhinal cortex, parahippocampal gyrus,

and perirhinal cortex, along with the amygdala.

Regions within the MTL are considered crucial for processing

of long-term declarative memory.

Memory consolidation: Period of memory stabilization

resulting in greater stability of the learned memory in the

face of interference, and greater ability to retrieve the

memory at a later date.

Memory encoding: The stage of originally learning the

information constituted in a memory.

Memory recall: Ability to retrieve some or all of the

information stored in memory.

Nonrapid eye movement sleep (NREM): Collective

stages of sleep characterized by increasingly cortical

de-activity, and characteristic slow waves and sleep

spindles.

Positron emission tomography: A neuroimaging

technique using small doses of radioactive isotopes to

measure regional changes in brain metabolism.

Procedural memory: A form of nondeclarative memory

(e.g., motor skills).

Rapid eye movement sleep (REM): The stage of sleep,

commonly associated with dreaming activity, characterized

by acetyl cholinergic desynchronized electrical brain activity,

peripheral muscle atonia and phasic eye movement bursts.

Slow-wave activity (SWA): Quantitative measure of the

power (strength) of slow (�0.5–4Hz) frequencies in the

sleeping EEG.

En
cyclopedia of Sleep http://dx.doi.org/10.1016/B978-0-12-378610-4.00333-8
Encyclopedia of Sleep

Introduction

Despite the vast amount of time sleep takes from our lives,

we still lack consensus on its function. In part, this is perhaps

because sleep, like its counterpart wakefulness, may serve not

one but many functions, for the brain and body alike. Centrally,

sleep is a brain phenomenon, and over the past 20 years, an

exciting revival has taken place within the neurosciences, focus-

ing on thequestionofwhywe sleep, and specifically targeting the

role of sleep in a number of cognitive and emotional processes.

This article aims to provide a synthesis of these recent findings in

humans, with the goal of extracting consistent themes across

domains of brain functions that appear to be regulated by

sleep. Section ‘Memory Processing and Brain Plasticity’ will ex-

plore the role of sleep in memory and brain plasticity, and also

examine competingmodels of sleep-dependent learning. Section

‘Association, Integration, and Creativity’ will address the role of

sleep beyondmemory consolidation, particularly in processes of

association, integration, and creativity.

Memory Processing and Brain Plasticity

When considering the role of sleep in memory processing, it is

pertinent to appreciate that memories evolve over time. Specif-

ically, memories pass through discrete stages in their ‘lifespan.’

The conception of a memory begins with the process of

encoding, resulting in a stored representation of an experience

within the brain. However, it is now understood that a vast

number of postencoding memory processes can take place. For

memories to persist over the longer time course of minutes

to years, an off-line, unconscious operation of consolidation

appears to be necessary, affording memories greater resistance

to decay (a process of stabilization), or even improved recol-

lection (a process of enhancement). Sleep has been implicated

in both the encoding and consolidation of memory.

Sleep and Memory Encoding

Early human studies in the 1960s, that described the impact of

sleep deprivation on declarative memory encoding, indicated

that ‘temporal memory’ (memory for when events occur) was

significantly disrupted by a night of pretraining sleep loss.

Subsequent studies replicated and extended this work, demon-

strating impaired learning and retention of fact-based informa-

tion in subjects deprived of sleep for 36 h, even in subgroups

that received caffeine to overcome nonspecific effects of lower-

ed alertness. Furthermore, the sleep-deprived subjects often

displayed significantly worse insight into their memory encod-

ing performance, resulting in lower predictive ability of their

performance.

Cognitive neuroscientists have examined the neural basis of

such memory impairments using functional magnetic reso-

nance imaging (fMRI). In one of the first such studies, sleep

503

, (2013), vol. 1, pp. 503-512

http://dx.doi.org/10.1016/B978-0-12-378610-4.00333-8

504 Sleep and the Nervous System | The Modulation of Memory by Sleep

deprivation significantly impaired learning activity in the me-
dial temporal lobe (MTL), while the prefrontal cortex actually

expressed greater activation. Most interesting, the parietal

lobes, which were not activated during learning in the control

group that had previously slept, were significantly active in the

deprivation group. Such findings suggest that inadequate sleep

(at least following one night) prior to learning produces bi-

directional changes in episodic encoding activity, involving the

inability of the MTL to engage normally during learning, com-

bined with potential compensation attempts by prefrontal re-

gions, which in turn may facilitate the recruitment of parietal

lobe function.

The impact of sleep deprivation on memory formation may

be especially pronounced for emotional material, depending

on whether it is emotionally negative, positive, or neutral. For

example, a recent report demonstrated that when collapsing

across all types of emotional stimuli, sleep-deprived subjects

exhibited a striking 40% reduction in the ability to form new

human memories under conditions of sleep deprivation.

When these data were separated into the three affective catego-

ries (negative, positive, or neutral), the magnitude of encoding

impairment differed. In those who had slept, both positive and

negative stimuli were associated with superior retention levels

relative to the neutral condition, consonant with the notion

that emotion facilitates memory encoding. However, there was

severe disruption of encoding and hence later retention for

neutral and especially positive emotional memory in the

sleep-deprived group. In contrast, a relative resistance of nega-

tive emotional memory was observed in the deprivation group.

These data suggest that, while the effects of sleep deprivation

are directionally consistent across memory subcategories, the

most profound impact is on the encoding of positive emo-

tional stimuli, and to a lesser degree, emotionally neutral

stimuli. In contrast, the encoding of negative memory appears

to be more resistant to the effects of prior sleep loss, at least

following one night.

The impact of sleep deprivation on the neural dynamics

associated with declarative memory encoding has also been

examined using event-related fMRI. In addition to perfor-

mance impairments under the condition of sleep deprivation,

and relative to a control group that slept, a highly significant

and selective deficit was identified in bilateral regions of the

hippocampus – a structure known to be critical for learning

new episodic information. When taken together, this collec-

tion of findings indicate the critical need for sleep before

learning in preparing key neural structures for efficient next-

day learning. Without adequate sleep, hippocampal function

becomes markedly disrupted, resulting in the decreased ability

for recording new experiences, the extent of which appears to

be further governed by alterations in prefrontal encoding

dynamics.

While these findings describe the detrimental impact of a

lack of sleep, recent work has conversely characterized the

proactive benefit of sleep, and specific sleep physiology, on

restoring episodic memory encoding ability. Episodic learning

capacity was assessed twice across a 6-h waking interval. Fol-

lowing the first learning session at 12.00, half of the subject

remained awake, while the other half obtained a 100min nap

opportunity. Both groups then performed the second learning

session at 18.00. Learning capacity deteriorated in those who

Encyclopedia of Sleep,

remained awake, yet sleep blocked this deterioration and re-

stored episodic encoding ability. Moreover, the extent of learn-

ing restoration in the nap group was correlated with both the

amount of stage-2 nonrapid eye movement sleep (NREM), and

specifically the number of fast sleep spindles over the prefron-

tal cortex. Taken together with the neuroimaging results above,

these demonstrate an important role for sleep, and specific

NREM oscillations, before learning in preparing the hippocam-

pus for efficient episodic encoding ability.

Sleep and Memory Consolidation

Using a variety of behavioral paradigms, evidence for the role

of sleep in memory consolidation has now been reported

across a diverse range of phylogeny. Perhaps the earliest refer-

ence to the beneficial impact of sleep on memory is by the

Roman rhetorician, Quintilian, stating that:

. . .[it] is a curious fact, of which the reason is not obvious, that the

interval of a single night will greatly increase the strength of the

memory. . . Whatever the cause, things which could not be recalled

on the spot are easily coordinated the next day, and time itself,

which is generally accounted one of the causes of forgetfulness,

actually serves to strengthen the memory.

A robust and consistent literature has demonstrated the

need for sleep after learning in the subsequent consolidation

and enhancement of procedural memories. Early work focus-

ing on the role for sleep in declarative memory processing was

somewhat less consistent, but more recent findings have now

begun to reveal a robust beneficial effect of sleep on the con-

solidation of declarative memory, which is our focus here.

Several studies have now demonstrated off-line consolida-

tion on a word-pair associates task following sleep, attributed

to early night sleep, rich in slow-wave sleep (SWS), and more

recently, slow delta waves (0.5–4Hz) and the very slow cortical

oscillation (<1Hz). Following learning of a word-pair list,

a technique called ‘direct current stimulation’ was used to

induce slow oscillation-like field potentials in the prefrontal

cortex (in this case, at 0.75Hz) during early night SWS. Direct

current stimulation not only increased the amount of slow

oscillations during the simulation period (and for some time

after), but also enhanced next day word-pair retention, suggest-

ing a critical role for SWS neurophysiology in the off-line

consolidation of episodic facts.

Rather than simply testing memory recall, others have since

revealed that the extent of sleep has the ability to protect

declarative memories using experimentally induced learning

disruption. Taking advantage of a classic interference tech-

nique called the A-B–A-C paradigm, subjects first learned

unrelated word-paired associates, designated as list A-B (e.g.,

leaf-wheel, etc.). After overnight sleep, or wakefulness during

the day, half of the subjects in each group learned a new,

interfering list containing a new associate paired with the first

word, designated as list A-C (e.g., leaf-nail, etc.), before being

tested on the original A-B list (e.g., leaf-wheel, etc.). In the

groups that did not experience the interfering challenge –

simply being trained and then tested on list A-B – sleep pro-

vided a modest benefit to memory recollection. However,

when testing the groups that were exposed to the interfering

list learning (list A-C) prior to recalling the original list

(2013), vol. 1, pp. 503-512

Sleep and the Nervous System | The Modulation of Memory by Sleep 505

(list A-B), a large and significant protective benefit was seen in
those that slept. Thus, memories tested after a night of sleep

were significantly more resistant to interference, whereas across

a waking day, memories were far more susceptible to this

antagonistic learning challenge. Yet, it was only by using an

interfering challenge, the A-C list, that the true benefit of sleep’s

protection of memory was revealed; a benefit that would not

necessarily have been evident in a standard study-test memory

paradigm.

One mechanism proposed to underlie these effects on

hippocampal-dependent learning tasks (and see next section)

is the reactivation of memory representations at night. A con-

siderable number of reports have investigated the firing

patterns of large networks of individual neurons across the

wake–sleep cycle in animals. The signature firing patterns of

these hippocampal and cortical networks, expressed during

waking performance of spatial tasks and novel experiences,

appear to be ‘replayed’ during subsequent SWS (and in some

studies, also rapid eye movement (REM)). Homologous evi-

dence has been reported in the human brain using a virtual

maze task in combination with positron emission tomography

scanning. Daytime learning was initially associated with hip-

pocampal activity. Then, during posttraining sleep, there was a

reemergence of hippocampal activation, specifically during

SWS. Most compelling, however, the amount of SWS reactiva-

tion in the hippocampus was proportional to the amount of

next-day task improvement, suggesting that this reactivation is

associated with off-line memory improvement.

It is well known that memory can be strongly modulated by

smell; most of us have associated the smell of a certain perfume

or cologne with a particular person, and when we encounter

that same perfume again, it often results in the powerful cued

recall of memories of that particular person. In one such ex-

ample, however, following learning of a spatial memory task

that was paired with the smell of rose, the odor was not re-

presented again at retrieval, but instead, during subsequent

SWS that night – a time when consolidation was presumed to

be occurring. Relative to a control condition where the odor

was not presented again during SWS, the reperfusion of the

rose scent at night resulted in significantly improved recall the

following day. Moreover, the representation of the odor

resulted in greater (re)activation of the hippocampus during

SWS. These findings support the role of SWS in the consolida-

tion of individual declarative memories, and may indicate an

active reprocessing of hippocampal-bound information during

SWS. Extending these findings is the demonstration of similar

sleep-dependent reactivation of memory using auditory rather

than olfactory cues, describing a selective ability to manipulate

individual item memory consolidation during SWS.

Counter to the classical notion of sleep universally benefiting

all previously encoded memories, emerging findings are begin-

ning to describe a more nuanced and selective role for sleep

in off-line memory consolidation. For example, sleep, often

REM sleep, has been demonstrated to preferentially consolidate

negative emotional memories relative to neutral items. Beyond

emotion, sleep has recently been demonstrated to selectively

enhance memories weighted toward monetary rewards.

Interestingly, and in line with the view that sleep may

actively play a role in targeted forgetting as well as remember-

ing, the most recent work has described a differential role for


sleep in remembering and forgetting using a directed forgetting

learning paradigm, where individual items (words) are cued

during initial encoding to either be remembered, or forgotten.

Compared to a wake control group, sleep, and specifically fast

sleep spindles over left parietal cortex, preferentially enhanced

those items previously cued to be remembered during learning,

but did so without strengthening those items previously cued

to be forgotten: indicative of specific preference based on prior

waking instruction (Figure 2). Collectively these studies sug-

gest that sleep does not universally consolidate all episodic

memories learned during the day. Instead, sleep appears to

be more ecologically attuned to qualitative aspects of encoded

experiences (e.g., emotionality, reward potential, or explicit

cue instructions and intentions), resulting in discriminatory

off-line memory processing.

Models of Sleep-Dependent Memory Processing

Elucidating the neural mechanisms that control and promote

sleep-dependent human memory consolidation remains an

active topic of research, and debate. It is perhaps unlikely that

multiple different memory systems, involving diverse cortical

and/or subcortical networks, require the same underlying neu-

ral mechanisms for their modulation. Even if they do, it is not

clear that this process would rely on just one type of sleep-stage

physiology. At present, two intriguing models of sleep-

dependent plasticity, relevant to declarative memory, have

been offered to account for the overnight facilitation of recall,

which build on different aspects of neural activity during sleep:

(1) hippocampal–neocortical dialog; (2) synaptic homeostasis

hypothesis.

Hippocampal–neocortical dialogThere is considerable agreement that structures within the

MTL, most notably the hippocampal complex, are crucial for

the formation and retrieval of new declarative memories. These

structures are believed to guide the reinstatement of recently

formed memories by binding together patterns of cortical ac-

tivation that were present at the time of initial learning.

A classical model of declarative memory consolidation sug-

gests that information initially requires MTL binding, but

over time, and by way of slow off-line processes, is eventually

integrated into neocortical circuits (Figure 3). Neocortical

structures thus become the eventual storage site for consoli-

dated episodic memories through cross-cortical connections,

and as a consequence, the MTL is not necessary for their

retrieval. Therefore, the classical model of memory consolida-

tion holds that neocortical structures become increasingly

important for the retention and retrieval of successfully consol-

idated episodic memories, while the corresponding contribu-

tion of the hippocampus progressively decreases. In addition to

its role in binding distributed corticalmemory components, the

hippocampus plays a critical role in reactivating these networks,

specifically during sleep. This process of reactivation would,

over multiple sleep cycles across a night, and/or multiple oc-

currences of sleep over many nights, gradually strengthen the

initially weak connections between neocortical sites, thereby

reinforcing them (Figure 3). Eventually, this strengthening

would allow the original information to be activated in the

cortex, independent of the hippocampus.

, (2013), vol. 1, pp. 503-512

16

** ***

*

60%

40%

20%

0%

12

8

Num

ber

of w

ord

s re

calle

d

R-F

diff

eren

ce m

easu

re

4

0R-words

(a) (b) (c)

F-words R-F difference

P3

−0.2

P(fast spNo-nap group Nap group

Figure 2 Role of sleep on directed remembering and forgetting. (a–b) Behavprior cue instruction (remember, R-words; forget, F-words) in the nap and nocalculated as the subtraction of these scores (R–F; expressed as a proportionsignificance at: *p<0.05 and **p<0.01. Error bars represent S.E.M. (c–d) Pspindles: (d) correlation topographical plot demonstrating strength of relationactivity at the P3 (left parietal) electrode site showing the strongest relationshincrements). Modified from Saletin JM, Goldstein AN, and Walker MP (2011)memories. Cerebral Cortex 21(11): 2534–2541, http://dx.doi.org/10.1093/cer

100

*

**

80

60

40Per

cent

rec

all

20

0

Condition(no interference) (interference)

Wake Sleep Wake-I Sleep-I

Figure 1 Impact of sleep on the consolidation and stabilization ofdeclarative memory. Percent correct recall for B words from the originalA-B pair after a 12-h retention interval of either wake or sleep following nointerference or interference learning (list A-C). {p<0.10, *p<0.05,**p<0.001; error bars indicate S.E.M. Modified from Ellenbogen et al.(2007).




Interestingly, these models make two predictions about the

impact of sleep on declarative memory. The first is that declar-

ative memories from the day prior should be more resistant to

interference the next day, due to the increased corticocortical

connections formed during overnight consolidation (Figure 1).

This prediction has been demonstrated by the finding of

greater postsleep resistance to interference, using the A-B–A-C

paired-associates paradigm. A second and far less considered

benefit of this sleep-dependent dialog is the encoding capacity

of the hippocampus (Figure 3). If the strengthening of cortico-

cortical connections takes place during sleep, albeit iteratively,

then blocking sleep after hippocampal learning should negate

this off-line transfer, preventing the development of indepen-

dence from (or ‘refreshing’ of) the hippocampus, and by doing

so, decrease the capacity for new hippocampal learning the

next day. This second premise appears to accurately explain

the findings discussed in the memory encoding section above,

which describe a significant impairment of hippocampal

encoding activity when sleep has not taken place (through

deprivation), being associated with a decreased ability to

form new episodic memories.

Recent evidence has provided further support for this sleep-

dependent dialog and neural transformation of declarative

memory. One such study examined the benefit of daytime

naps on episodic declarative memory consolidation. In addi-

tion to a long-term evaluation of memory over 3months, there

was also a short-term evaluation of memory across the first

day, which included an intervening nap period (90min) be-

tween training and testing of the original studied (‘remote’)

stimuli. Interestingly, the duration of NREM SWS during the

intervening nap correlated positively with later recognition

memory performance, yet negatively with retrieval-related

activity in the hippocampus. Expanding on these findings

is the demonstration that one night of posttraining sleep

*

1.2

1.0

0.8

0.6

0.4

0.2

0 0.4 0.8

P3 fast spindle density(#spindles/NREM min)

1.2

r = 0.66P = 0.0006

1.6

R-F

diff

eren

ce m

easu

re

(d)

0.7

earson’s r-valueindle density and R-F)

ioral data. Memory performance: (a) number of words recalled based on-nap groups and (b) the efficiency measure of directed forgetting,of total recall). Between group comparisons (line across bars) reflecthysiological data. Relationship between memory performance and sleepship between fast sleep-spindle density and R–F score. Note spindleip (color bar indicates r value, with noted corresponding critical p valueThe role of sleep in directed forgetting and remembering of humancor/bhr034.

(2013), vol. 1, pp. 503-512

Cortical modules

Hippocampus

Sleep-spindles Slow-waves

Sleep WakeConsolidated original memory

Refreshed encoding capacity

Time

Wake

Figure 3 Model of sleep-dependent hippocampal–neocortical memory consolidation. At encoding the hippocampus rapidly integrates informationwithin distributed cortical modules. Successive sleep-dependent reactivation of this hippocampal–cortical network leads to progressive strengthening ofcorticocortical connections, which over time, allow these memories to become independent of the hippocampus and gradually integrated withpreexisting cortical memories.



deprivation, even following recovery sleep, significantly

impairs the normal modulation of hippocampal activity

associated with episodic memory recollection. Furthermore,

first-night sleep deprivation also prevented an increase in

hippocampal connectivity with the medial prefrontal cortex,

a development that was only observed in those that sleep after

learning.

While no one study has yet demonstrated that the neural

signature of learning during the day is subsequently reactivated

and driven by characteristics of SWS at night, and that the

extent of these properties are consequently proportional to

the degree of next-day recall and memory reorganization; col-

lectively, they offer an empirical foundation on which to

entertain this possibility.

Synaptic homeostasis hypothesisIn recent years, an orthogonal theory of SWS and learning has

emerged, which postulates a role for sleep in regulating the

synaptic connectivity of the brain – principally the neocortex.

The synaptic homeostasis model considers NREM SWS, and

specifically the magnitude of slow-wave activity (SWA) of SWS,

as a brain state that promotes the decrease of synaptic connec-

tions, not their increase. Accordingly, plastic processes, such as

learning and memory occurring during wakefulness, result in a

net increase in synaptic strength in diffuse brain circuits. The

role of SWS, therefore, and the slow oscillation in particular,

is to selectively downscale or ‘de potentiate’ the synaptic

strength back to baseline levels, preventing synaptic over

potentiation, which would result in saturated brain plasticity.

In doing so, this rescaling would leave behind more efficient

and refined memory representations the next day, affording

improved recall. A number of human studies have provided

evidence supporting their model. For example, it has been

shown that learning of a motor-skill adaptation task during

the day subsequently triggers locally specific increases in corti-

cal SWA at night, the extent of which is proportional to both

the amount of initial daytime learning and the degree of next-

day improvement (Figure 4).

How can the two concepts of neural reactivation (such as

increased fMRI activity during SWS) and neural homeostasis

(such as increased slow-wave EEG activity) be interpreted at a


neural synaptic level? It could be argued that both these

reported changes reflect either an increased neural reactivation,

or an increase in SWA associated with homeostasis and synap-

tic downscaling. Furthermore, both hypotheses, while dis-

tinctly different, mechanistically, could offer complementary

benefits at the network level in terms of signal-to-noise ratio

(SNR), andmay account for overnight memory improvements.

Specifically, homeostatic synaptic downscaling could result in

the removal of superfluous neural connections, resulting in

improved SNR. However, neural reactivation and strengthen-

ing of experience-dependent circuits, without removing redun-

dant synaptic connects, may equally improve SNR. Therefore,

both mechanisms, while different, could produce a similar

outcome – enhanced fidelity of the memory representation.

Presumably the combination of both would perhaps produce

the most optimal and efficient memory trace, yet a careful

delineation of these possibilities remain an important goal

for future studies.

Interestingly, the homeostasis model would also predict

that sleep deprivation, specifically the prevention of SWS,

would also negate effective new learning the next day, due to

over potentiation of the synaptic connections. Thus, any region

that exhibits SWA, and is involved in representing memory

(e.g., hippocampus), would display a corresponding inability

to code further information beyond a normal waking duration

(�16h in humans). Such a premise may offer an alternative

explanation to the marked hippocampal encoding deficits

reported under conditions of sleep loss.

A role for sleep spindles?Independent of both these models, which consider the role of

SWS in memory processing, a number of reports have also

described an association between learning and the hallmark

feature of stage-2 NREM – sleep spindles; short (�1 s) synchro-

nous bursts of activity expressed in the EEG in the 10–16Hz

frequency range. For example, following learning of a motor-

skill memory task, posttraining sleep spindles over the motor

cortex were recorded, evaluating the difference in spindle ac-

tivity in the right, learning hemisphere (since subjects perform

the task with their left nondominant hand), relative to the

left, nonlearning hemisphere. Remarkably, when sleep-spindle

, (2013), vol. 1, pp. 503-512

Learning

(a) (b)

(c)(d)

1.6

0.6

%

1530

20

10

0

0 10

Increase in slow-wave activity

20 30 40 50 %

10

5

0 50

132

185

−5

−10

−15

1.0

Difference[learning - no learning]

Imp

rove

men

t in

mem

ory

(mea

n d

irect

iona

l err

or)

No learning

Figure 4 Slow-wave activity and motor-skill memory. Topographical high-density EEG maps of slow-wave activity (SWA) during NREM sleep followingeither (a) motor-skill learning or (b) a nonlearning control condition, and (c) the subtracted difference between SWA in the learning versusnonlearning condition, demonstrating a local homeostatic increase above the learning-related central-parietal brain region. (d) The correlation betweenthe amount of overnight improvement on the task (measured the next day) and the extent of increase in SWA across subjects. Modified fromHuber R, Ghilardi MF, Massimini M, and Tononi G (2004) Local sleep and learning. Nature 430(6995): 78–81.



power at electrode sites above the primary motor cortex of the

nonlearning hemisphere (left) were subtracted from those in

the learning hemisphere (right), representing the within

subject, between-hemisphere difference in spindle activity fol-

lowing learning, a strong predictive relationship with the

amount of memory improvement emerged.

Such findings indicate that the enhancement of specific

memory representations is associated with electrophysiological

events expressed at local, anatomically discrete locations of the

brain. Contrasting with the proposed impact of SWS, the

mechanistic benefit of sleep spindles may be related to their

faster stimulating frequency; a range suggested to facilitate

long-term potentiation (LTP; a foundational principal of syn-

aptic strengthening in the brain), and not synaptic depression.

This increase in spindle activity may represent a local, endog-

enous trigger of intrinsic synaptic plasticity, again correspond-

ing topographically to the underlying memory.

Increases in posttraining spindle activity are not limited to

procedural memory tasks. For example, significantly higher

sleep-spindle density has been reported in subjects that under-

went a daytime episodic learning session (encoding of word-pair

associates) compared to controls that didnot perform the learning

session. Moreover, the spindle density was associated with the

proficiency ofmemory recalled the next day in the learning group.

Reconciling modelsNone of these models necessarily need to be wrong. Instead,

aspects of each may afford complementary and synergistically


beneficial outcomes for memory. Clues to this possibility lie

within the ordered structure of human sleep, with NREM SWS

dominating early in the night, and stage-2 NREM and REM

prevailing later in the night. When placed in this temporal

framework, a progression of events emerges that may be opti-

mal for the neuroplastic modulation of memory representa-

tions. From a reactivation perspective, the predominance of

hippocampal–neocortical interaction would take place in the

early SWS-rich phase of the night, leaving corticocortical con-

nections on offer for later processing during stage-2 NREM and

REM. Similarly, and even in coincidence, SWS may downscale

cortical (and possibly subcortical) plasticity, and do so in a

learning-dependent manner, again leaving only those repre-

sentations (or aspects of these representations) which are

strongest – including those strengthened by hippocampal–

neocortical interplay – for processing during these latter pe-

riods of sleep, dominated by faster frequency oscillations.

This concept is analogous to the art of sculpture. During the

day, through experience, substantial informational ‘clay’ is

acquired on the cortical pedestal; some relevant, some not.

Once accumulated, the next step is to carve out and select the

strongest and most salient memory representations (‘statues’)

for the organisms – a mechanism that SWS, occurring first and

predominately early in the night, may be ideally suited for.

Following such downscaling and/or dynamic selection of

memory through translocation, the remaining cortical repre-

sentations – the rough outline of the sculpted form – may

finally be strengthened by faster frequency oscillations,

(2013), vol. 1, pp. 503-512


including those of sleep spindles (and potentially PGO-wave
burst during REM), more associated with the potentiation of

synaptic connections, not their depotentiation. This final step

is akin to polishing and improving the detailed features of the

memory statue; which in terms of computational modeling

would offer improved signal-to-noise quality within the sys-

tem. Such a cooperative mechanism, which appreciates the

temporal order of the wake–sleep cycle (acquisition, followed

by postprocessing), and within sleep, the ultradian pattern of

sleep-stage progression across a night (selection and removal,

followed by strengthening), would produce a network of

stored information that is not only more efficient, but for

those representations remaining, more enhanced. Both these

processes would predict improved recall of remodeled individ-

ual memories from the prior day, and further afford the syn-

aptic capacity for efficient acquisition of new ‘information clay’

the next day.

Association, Integration, and Creativity

As critical as consolidation may be – an operation classically

concerned with individual memory items – the association and

integration of new experience into preexisting networks of

knowledge is equally, if not more, important. The resulting

creation of associative webs of information offers numerous

and powerful advantages. Indeed, the end goal of sleep-

dependent memory processing may not be the enhancement

of individual memories in isolation, but instead, their integra-

tion into a common schema, and by doing so, facilitate the

development of universal concepts; a process which forms the

basis of generalized knowledge, and even creativity.

Association and Integration

Perhaps the earliest demonstration that sleep may be involved

in a form of memory generalization was described using an

artificial grammar task, where subjects trained and later tested

on the ability to transfer phonological categories across differ-

ent acoustic patterns. The task required forming newmappings

from complex acoustical sounds to preexisting linguistic cate-

gories, which then generalized to new stimuli. As such, it

involved both a declarative process of forming specific memo-

ries associated with the learned stimuli, together with a proce-

dural component involving mapping across the set of learned

sounds that supports generalization to novel stimuli. During

the initial training session there was a significant improvement

in recognition performance on the task. However, when

retested after a 12-h waking interval, performance had decayed.

Yet, if subjects were retested following a night of sleep, this

ability for memory generalization was restored. Supporting

this concept, others have also demonstrated that sleep, and

not equivalent time awake, can integrate related but novel

phonemes into preexisting long-term lexical memory stores

overnight.

Similarly, infants when exposed to ‘phrases’ from an artifi-

cial language during a learning session – for example, phrases

like ‘pel – wadim – jic’ – until the infants became familiar (as

indexed by look responses). These three syllable units had an

embedded rule, which was that the first and last unit formed a


relationship of nonadjacency; in this case, pel predicts jic. The

infants were then retested several hours later, yet some infants

took normally scheduled naps, while others were scheduled at

a time when they would not sleep after learning. At later

testing, infants again heard the recordings, along with novel

phrases in which the predictive relationship between the first

and last word was new. Infants who did not sleep recognized

the phrases they had learned earlier, yet those who had slept

demonstrated a generalization of the predictive relationship to

new phrases, suggesting that the intervening process of sleep

allowed the reinterpretation of prior experience, and supported

the abstraction of commonalities – that is, the ability to detect

a general pattern in new information.

This sleep-dependent hypothesis of integration was recently

tested by examining human relational memory – the ability to

generalize previously acquired associations to novel situations.

Participants initially learned five ‘premise pairs’ (A>B, B>C,

C>D, D>E, E>F). Unknown to subjects, the pairs contained

an embedded hierarchy (A>B>C>D>E>F). Following an

off-line delay of 20min, 12 h across the day or 12 h containing

a night of sleep, knowledge of this hierarchy was tested by

examining relational judgments for novel ‘inference’ pairs,

either separated by one-degree of associative distance (B>D,

C>E pairs), or by two-degrees of associative distance (B>E

pair). Despite all groups achieving near identical premise-pair

retention after the off-line delay (i.e., the building blocks of the

hierarchy), a striking dissociation was evident in the ability to

make relational inference judgments. Subjects that were tested

soon after learning in the 20min group showed no evidence

of inferential ability, performing at chance levels (Figure 5(a)).

In contrast, the 12h groups displayed highly significant rela-

tional memory development. Most remarkable, however, if the

12h period contained a night of sleep, a near 25% advantage

in relational memory was seen for the most distantly con-

nected inferential judgment (the B>E pair; Figure 5(a)).

Together, these findings demonstrate that human memory

integration takes time to develop, requiring slow, off-line

associative processes. Furthermore, sleep appears to preferen-

tially facilitate this integration by enhancing hierarchical mem-

ory binding, biasing the development of the most distant/

weak associative links amongst related yet separate memory

items (Figure 5(b)). It is also interesting to note a further

advantage of this sleep-dependent assimilation process. When

stored as individual premise pairs (top row, Figure 5(b)) the

size/number of items (‘bits’) of information to code is ten (A–B,

B–C, C–D, D–E, E–F). However, when formed into a hierarchy,

the informational load is compressed, reduced by nearly 50% to

just six bits (A–B–C–D–E–F). Therefore, a supplementary bene-

fit of sleep-dependentmemory associationmay be the improved

efficiency of memory storage, in addition to a more generalized

representation.

Thus, the overnight strengthening and consolidation of

individual item memories (reviewed above), may not be the

ultimate objective of sleep-dependent memory processing, es-

pecially when considering that declarative (nonemotional)

memories decay over the long term. It is then interesting to

speculate whether sleep serves to facilitate two complimentary

objectives for declarative memory, which span different time

courses. The first may be an initial process of consolidating

individual item (episodic) memories that are novel, whichmay

, (2013), vol. 1, pp. 503-512

100

90

80

Per

cent

cor

rect

70n.s.

n.s.

60

50

Inference pairs

(a)

(b)

Immediate A B

Wake

1� inference

1� and 2� inference

Sleep

1� 2� 1� 2� 1� 2�(chance)

A B

A B C D E F

B C

B C

C D

C D

D E

D E

E F

E F

20 min 12 h wake 12 h wake

**

*

Figure 5 Sleep-dependent integration of human relationalmemory. (a) Delayed inference (associative) memory performance(% correct) in a relational memory task following different off-linedelays. Immediate testing after just a 20min off-line delay,demonstrated a lack of any inferential ability resulting in chanceperformance on both one-degree (first order) and two-degree andtwo-degree (second order) associative judgments. Following amore extended 12 h delay, across the day (wake group), performancewas significantly above chance across both the one and two-degreeinference judgments. However, following an equivalent 12 h off-linedelay, but containing a night of sleep (sleep group), significantlybetter performance was expressed on the more distant two-degreeinference judgment compared with the one-degree judgment.(b) A conceptual model of the effects of sleep on memoryintegration. Immediately after learning, the representation of eachpremise is constituted as the choice of one item over another(A>B, etc.), and these premises are isolated from one anotherdespite having overlapping elements. After a 12-h period with nosleep, the premise representations are partially integrated by theiroverlapping elements, sufficient to support first-order transitiveinferences. However, following a 12-h off-line period with sleep,the premise representations are fully interleaved, supporting bothfirst- and second-order transitive inferences. *p<0.05; error barsindicate S.E.M.




occur in the relative short term. Over a longer time course,

however, and utilizing these recently consolidated item mem-

ories prior to their fading, sleep may begin the process of

extraction (of meaning) and abstraction (building associa-

tional links with existing information), thereby creating more

adaptive semantic networks. Ultimately, individual item mem-

ories would no longer be necessary for the goal that sleep is

trying to achieve, and only the conceptual meaning of such

experiences would remain. Whether the subsequent loss of

item memories is passive, or whether sleep plays an active

role in this process remains to be examined, but this is a

testable hypothesis, that is, forgetting (individual items) is the

price we pay for remembering (general rules).

Creativity

One potential advantage of testing associative connections and

building cross-linked systems of knowledge is creativity – the

ability to take existing pieces of information and combine

them in novel ways that lead to greater understanding and

offer new advantageous behavioral repertoires. The link be-

tween creativity and sleep, especially dreaming, has long been

a topic of intense speculation. From the dreams of both August

Kekule which led to the conception of a simple structure for

benzene, and Dmitry Mendeleyev that initiated the creation of

the periodic table of elements, to the late night dreaming

of Otto Loewi which inspired the experimental demonstration

of neurochemical transmission, even scientific examples of

creativity occurring during sleep are not uncommon.

Quantitative data have further demonstrated that solution

performance on tests of cognitive flexibility using anagram

word puzzles is more than 30% better following awakenings

from REM sleep compared with NREM awakenings. Similarly,

a study of semantic priming has demonstrated that, in contrast

to the situation in waking, performance following REM sleep

awakenings shows a greater priming effect by weakly related

words than by strong primes, while strong priming exceeds

weak priming in NREM sleep, again indicating the highly

associative properties of the REM sleep brain. Even the study

of mental activity (dreams) from REM sleep indicates that there

is not a concrete episodic replay of daytime experiences, but

instead, a much more associative process of semantic integra-

tion during sleep.

Yet, the most striking experimental evidence of sleep-

inspired insight was elegantly demonstrated when using a

mathematical ‘number reduction task,’ a process of sleep-

dependent creative insight. Subjects analyzed and worked

through a series of eight-digit string problems, using specific

addition rules. Following initial training, after various periods

of wake or sleep, subjects returned for an additional series of

trials. When retested after a night of sleep, subjects solved the

task, using this ‘standard’ procedure, 16.5% faster. In contrast,

subjects who did not sleep prior to retesting averaged less than

a 6% improvement. However, hidden in the construction of

the task was a much simpler way to solve the problem. On

every trial, the last three response digits were the mirror image

of the preceding three. As a result, the second response digit

always provided the answer to the problem, and using such

(2013), vol. 1, pp. 503-512


‘insight,’ subjects could stop after producing the second re-
sponse digit. Most dramatically, nearly 60% of the subjects

who slept for a night between training and retesting discovered

this short cut the following morning. In contrast, no more than

25% of subjects in any of four different control groups that did

not sleep, had this insight. Sleeping after exposure to the

problem, therefore doubled the likelihood of solving it (al-

though it is interesting to note that this insight was not present

immediately following sleep, but took over 100 trials on aver-

age to emerge the next day).

In summary, substantial evidence now suggests that sleep

serves a meta-level role in memory processing that moves far

beyond the consolidation and strengthening of individual

memories, and instead, aims to intelligently assimilate and

generalize these details off-line. In doing so, sleep may offer

the ability to test and build common informational schemas of

knowledge, providing increasingly accurate statistic predic-

tions about the world, and allowing for the discovery of

novel, even creative next-day solution insights.

Conclusions

While not fully complete, we will soon have a new taxonomy

of sleep-dependent memory processing, and one that will su-

persede the polarized all-or-none views of the past. With such

findings, we can come to a revised appreciation of how both

wake and sleep unite in a symbiotic alliance to coordinate the

encoding, consolidation and integration of our memories, the

ultimate aim of which maybe to create a generalized catalog of

stored knowledge that does not rely on the verbose retention of

all previously learned facts.

Acknowledgments

The author wishes to thank Edwin Robertson, Robert

Stickgold, Allison Harvey, Ninad Gujar and Els van der Helm

for thoughtful insights. This work was supported in part by

grants from the National Institutes of Health (NIA AG31164);

and the University of California, Berkeley.

See also: Associations and Consequences of SRBD: Sleep-Related Breathing Disorder (SRBD) – Attention and Vigilance.

Further Reading

Crick F and Mitchison G (1983) The function of dream sleep. Nature304(5922): 111–114.

Dave AS and Margoliash D (2000) Song replay during sleep and computational rulesfor sensorimotor vocal learning. Science 290(5492): 812–816.

Dave AS, Yu AC, and Margoliash D (1998) Behavioral state modulation of auditoryactivity in a vocal motor system. Science 282(5397): 2250–2254.

Diekelmann S and Born J (2010) The memory function of sleep. Nature Reviews.Neuroscience 11(2): 114–126.

Drummond SP and Brown GG (2001) The effects of total sleep deprivation on cerebralresponses to cognitive performance. Neuropsychopharmacology 25(supplement 5):S68–S73.


Drummond SP, Brown GG, Gillin JC, Stricker JL, Wong EC, and Buxton RB (2000)Altered brain response to verbal learning following sleep deprivation. Nature403(6770): 655–657.

Dumay N and Gaskell MG (2007) Sleep-associated changes in the mentalrepresentation of spoken words. Psychological Science 18(1): 35–39.

Ellenbogen JM, Hulbert JC, Stickgold R, Dinges DF, and Thompson-Schill SL (2006)Interfering with theories of sleep and memory: Sleep, declarative memory, andassociative interference. Current Biology 16(13): 1290–1294.

Fenn KM, Nusbaum HC, and Margoliash D (2003) Consolidation during sleep ofperceptual learning of spoken language. Nature 425(6958): 614–616.

Fischer S and Born J (2009) Anticipated reward enhances offline learning during sleep.Journal of Experimental Psychology. Learning, Memory, and Cognition35(6): 1586–1593.

Fosse MJ, Fosse R, Hobson JA, and Stickgold RJ (2003) Dreaming and episodicmemory: A functional dissociation? Journal of Cognitive Neuroscience 15(1): 1–9.

Gais S, Albouy G, Boly M, et al. (2007) Sleep transforms the cerebral trace of declarativememories. Proceedings of the National Academy of Sciences of the United States ofAmerica 104(47): 18778–18783.

Gais S, Molle M, Helms K, and Born J (2002) Learning-dependent increases in sleepspindle density. Journal of Neuroscience 22(15): 6830–6834.

Gomez RL, Bootzin RR, and Nadel L (2006) Naps promote abstraction inlanguage-learning infants. Psychological Science 17(8): 670–674.

Harrison Y and Horne JA (2000) Sleep loss and temporal memory. Quarterly Journal ofExperimental Psychology 53(1): 271–279.

Hu P, Stylos-Allen M, and Walker MP (2006) Sleep facilitates consolidation ofemotionally arousing declarative memory. Psychological Science 17(10): 891–898.

Huber R, Ghilardi MF, Massimini M, and Tononi G (2004) Local sleep and learning.Nature 430(6995): 78–81.

Ji D and Wilson MA (2007) Coordinated memory replay in the visual cortex andhippocampus during sleep. Nature Neuroscience 10(1): 100–107.

Jones MW and Wilson MA (2005) Theta rhythms coordinate hippocampal–prefrontalinteractions in a spatial memory task. PLoS Biology 3(12): e402.

Louie K and Wilson MA (2001) Temporally structured replay of awake hippocampalensemble activity during rapid eye movement sleep. Neuron 29(1): 145–156.

Mander BA, Santhanam S, Saletin JM, and Walker MP (2011) Wake deterioration andsleep restoration of human learning. Current Biology 21(5): R183–R184.

Morris GO, Williams HL, and Lubin A (1960) Misperception and disorientation duringsleep. Archives of General Psychiatry 2: 247–254.

Nishida M, Pearsall J, Buckner RL, and Walker MP (2009) REM sleep, prefrontal theta,and the consolidation of human emotional memory. Cerebral Cortex19(5): 1158–1166.

Nishida M and Walker MP (2007) Daytime naps, motor memory consolidation andregionally specific sleep spindles. PLoS One 2(4): e341.

Payne JD and Kensinger EA (2010) Sleep leads to changes in the emotionalmemory trace: Evidence from fMRI. Journal of Cognitive Neuroscience23(6): 1285–1297.

Peigneux P, Laureys S, Fuchs S, et al. (2004) Are spatial memories strengthened in thehuman hippocampus during slow wave sleep? Neuron 44: 535–545.

Poe GR, Nitz DA, McNaughton BL, and Barnes CA (2000) Experience-dependentphase-reversal of hippocampal neuron firing during REM sleep. Brain Research855(1): 176–180.

Rasch B, Buchel C, Gais S, and Born J (2007) Odor cues during slow-wavesleep prompt declarative memory consolidation. Science 315(5817): 1426–1429.

Ribeiro S, Gervasoni D, Soares ES, et al. (2004) Long-lasting novelty-induced neuronalreverberation during slow-wave sleep in multiple forebrain areas induction ofhippocampal long-term potentiation during waking leads to increased extrahippocampal zif-268 expression during ensuing rapid-eye-movement sleep.PLoS Biology 2(1): E24.

Rudoy JD, Voss JL, Westerberg CE, and Paller KA (2009) Strengthening individualmemories by reactivating them during sleep. Science 326(5956): 1079.

Saletin JM, Goldstein AN, and Walker MP (2011) The role of sleep in directed forgettingand remembering of human memories. Cerebral Cortex 21(11): 2534–2541. http://dx.doi.org/10.1093/cercor/bhr034.

Skaggs WE and McNaughton BL (1996) Replay of neuronal firing sequences in rathippocampus during sleep following spatial experience. Science271(5257): 1870–1873.

Stickgold R, Scott L, Rittenhouse C, and Hobson JA (1999) Sleep-inducedchanges in associative memory. Journal of Cognitive Neuroscience11(2): 182–193.

Takashima A, Petersson KM, Rutters F, et al. (2006) Declarative memory consolidationin humans: A prospective functional magnetic resonance imaging study.

, (2013), vol. 1, pp. 503-512

http://dx.doi.org/10.1093/cercor/bhr034

http://dx.doi.org/10.1093/cercor/bhr034


Proceedings of the National Academy of Sciences of the United States of America103(3): 756–761.
Tononi G and Cirelli C (2003) Sleep and synaptic homeostasis: A hypothesis. BrainResearch Bulletin 62(2): 143–150.

Tononi G and Cirelli C (2006) Sleep function and synaptic homeostasis. Sleep MedicineReviews 10(1): 49–62.

Wagner U, Gais S, and Born J (2001) Emotional memory formation is enhanced acrosssleep intervals with high amounts of rapid eye movement sleep. Learning andMemory 8(2): 112–119.

Wagner U, Gais S, Haider H, Verleger R, and Born J (2004) Sleep inspires insight.Nature 427(6972): 352–355.

Walker MP, Liston C, Hobson JA, and Stickgold R (2002) Cognitive flexibility acrossthe sleep–wake cycle: REM-sleep enhancement of anagram problem solving. BrainResearch. Cognitive Brain Research 14(3): 317–324.


Walker MP and Stickgold R (2006) Sleep, memory and plasticity. Annual Review ofPsychology 10(57): 139–166.

Walker MP and Stickgold R (2010) Overnight alchemy: Sleep-dependent memoryevolution. Nature Reviews. Neuroscience 11(3): 218.

Wilhelm I, Diekelmann S, Molzow I, Ayoub A, Molle M, and Born J (2011) Sleepselectively enhances memory expected to be of future relevance. Journal ofNeuroscience 31(5): 1563–1569.

Wilson MA and McNaughton BL (1994) Reactivation of hippocampal ensemblememories during sleep. Science 265(5172): 676–679.

Yoo SS, Gujar N, Hu P, Jolesz FA, and Walker MP (2007) The human emotional brainwithout sleep – A prefrontal amygdala disconnect. Current Biology17(20): R877–R878.

Yoo S, Hu P, Gujar N, Jolesz F, and Walker M (2007) A deficit in the ability to form newhuman memories without sleep. Nature Neuroscience 10(3): 385–392.

(2013), vol. 1, pp. 503-512

the modulation of memory by sleep - open computing facilitysanl/reprints/...functional changes in...

Documents