hunting for meaning after midnight？
Even sound sleepers have restless brains.Your body may be largely motionless onceyour head hits
the pillow, but inside yourskull, millions of neurons are busily firingaway, often in synchronized
bursts that sendwaves of electricity sweeping across the sur-face of your brain.
What all this neural activity accom-plishes, if anything, is a mystery, and part ofthe even larger
puzzle of why we sleep. Oneidea that has gained favor in recent yearsis that during certain stages
of sleep, thebrain replays experiences from the day tostrengthen the memory of what
happened.Support for this notion comes from a varietyof experiments with rodents and
people,including a new study in this issue suggest-ing that boosting such memory 一
relatedactivity in the sleeping brain can improvememory performance in humans.
Some researchers suspect that replayingthe recent past during sleep is more than amemory aid.
This review may also give thebrain a chance to catch important informa-tion it missed the first
time around. "There'smore and more evidence accruing that whatwe're seeing during sleep is not
just astrengthening of memories," says RobertStickgold, a neuroscientist at Harvard Medical School in Boston. "What the brain isreally trying to do is extract meaning."
Such ideas aren't universally accepte 氏 however. One new and controversial hypothe-sis suggests
that memory and other cognitivebenefits are merely side effects of the truefunction of sleep:
dialing down synapses thathave gotten overexcited by daytime activity.And some skeptics aren't
convinced thatsleep has anything to do with memory at all.Given all the uncertainties, researchers
saythe quest to understand the sleeping brain isjust beginning. But they're already
findingfascinating clues about what happens whenwe're off in the Land of Nod.
Let me see that again
The first experimental evidence that the brainreplays recent experiences during sleep camefrom
experiments with rats begun in the 1990sby neuroscientist Bruce McNaughton of theUniversity of
Arizona, Tucson, and his gradu-ate student Matthew Wilson. McNaughtonand Wilson recorded the
electrical activity ofneurons called "place cells" in the rat hippo-campus. These neurons have an
affinity forparticular locations, so that as a rat runsaround its enclosure, a given place cell
fireseach time the rodent passes through that cell's
favorite spot. Because individual place cellsrespond only to a specific location, each timea rat
takes a different route, a different sequenceof place cells fires. Subsequent studies foundthat
sequences of place-cell firing that occuras a rat explores a new environment arereplayed the next
time the rat dozes, as if theanimal retraces its steps during sleep.
Humans may do something similar. In2004, a research team led 勿 Pierre Maquet ofthe University
of Liege, Belgium, usedpositron emission tomography (PET) to mon-itor brain activity in men
playing a virtual-reality game in which they learned to navi-gate through a virtual town (actually a
scenefrom the shoot-'em-up video game DukeNukem). The same regions of the hippo-campus that
revved up when the subjectsexplored the virtual environment also becameactive when the men
slipped into slow-wavesleep that night. This sleep stage is often con-sidered the deepest:
Slow-wave sleepers arehard to rouse. During this sleep stage, electroencephalogram (EEG) traces show waves of 工 electrical activity throughout the brain that 婆 peak
about once a second. PET scans 霎 revealed that the more intense hippocampal’昌 activity a
volunteer had during slow-wave 鑫 sleep, the better he performed the next day
1360
9 MARC 日 2007 VOL 3 巧 SCIENCE
www.sciencemag.org
Published 妙 AAAS
when he sprinted through the virtual town tofind certain objects as quickly as possible,Maquet and
colleagues reported in the28 October 2004 issue of Neuron.
Maquet's findings showed a nice correla-tion between neural activity during sleep andsubsequent
memory performance, says JanBorn, a neuroscientist at the University ofLubeck in Germany. The
natural next step,Born says, was to see whether artificiallyboosting memory-related neural activity
dur-ing sleep could improve memory perform-ance. On page 1426, Born's team describes
anattempt to do just that.
The researchers had volunteers play a videoversion of the card game Memory (also knownas
Concentration), in which they had to learnand remember the locations of pairs of cardsbearing the
same image in a group of 30 cards.Each matched pair flashed on the screen fora few seconds, one
at a time, with all theother cards facing down. After the volunteershad seen the locations of all the
pairs, theresearchers tested the subjects' recall 勿 turn-ing one of the 3 0 cards face up and asking
themto find its match. The researchers then usedEEG electrodes to monitor the volunteers'brain
activity while they slept.
Once the volunteers entered slow-wavesleep, the researchers gave some of them a puffof
rose-scented air. They'd previously givensome of the subjects a whiff of rose duringtheir initial
training session with the cards,reasoning that the odor would reactivate mem-ories of the training
session in these subjectswithout waking them. Indeed, functional mag-netic resonance imaging
(fMRI) scans insleeping subjects revealed that the odor acti-vated the hippocampus in those who
had expe-rienced it previously, even though the EEGshowed no disruptions in the subjects'
slumber.Although they didn't remember smelling rosesin their sleep, the subjects who got the
fragrantprompt remembered the matched pairs betterthe next day, getting 97% correct compared
to86% for subjects who'd received no odor whilesleeping. Subjects who got the rose odor
eitherwhile awake or while in REM sleep, on theother hand, showed no memory boost; nor
didpresenting the odor during slow-wave sleephelp subjects who hadn't been exposed to
roseduring the training session. "It's the first studyto really demonstrate that one can
influencememory with stimuli that explicitly activatethe hippocampus during sleep," says
Wilson,now an associate professor at the Massachu-setts Institute ofTechnology in Cambridge.
Born's findings fit with a popular view ofhow the brain files memories away for long-term storage,
a process neuroscientists callmemory consolidation. According to this
hypothesis, memories are first encoded 勿 the hippocampus and later-perhaps in amatter of hours
or days-transferred forlong-term storage to the cerebral cortex, orneocortex. Several lines of
evidence supportthis scenario, among them the observation 山 at many people who suffer amnesia
follow-ing damage to the hippocampus can stillrecall events and facts learned prior to theirinjury
even though they're unable to formnew memories. In such patients, old memo-ries must reside
somewhere other than thehippocampus. How the brain might transfermemories from the
hippocampus to the neo-cortex isn't known, but it's assumed torequire some kind of
back-and-forth com-munication between the two structures.
ple playing a game that required manipulat-ing strings of numbers were more than twiceas likely
to have a flash of insight thatenabled them to solve the problem morequickly after a night of sleep
than after a sim-ilar time of wakefulness.
More recently, at a sleep research meetinglast month at the Salk Institute for BiologicalStudies,
Stickgold presented findings suggest-ing that people find missing connections whilethey sleep.
His group had volunteers play acard game in which they attempted to predictwhether it would
"rain" based on cards theresearchers had shown them. The game wasrigged so that the card with
the diamonds,for instance, was followed by sunny weather84% of the time. Each card had its own
rule,
A study 勿 Wilson and postdoctoral fellowDaoyun Ji published in the January issue ofNature
Neuroscience supports this idea. Wilsonand Ji found that, much like hippocampalplace cells,
neurons in the visual cortex of ratsreplay firing sequences during slow-wavesleep that match their
activity during the rats'daytime maze running. Moreover, the scien-tists found that the replay in
the visual cortexhappens in lockstep with replay in the hippo-campus. "It's the first time we see
sequencesboth in the hippocampus and the neocortexand their coordination in time," says Maquet.
More than memory
Beyond simply fortifying memories, thebrain may be sifting through recent experi-ences during
sleep to identify rules about causeand effect or other useful patterns, someresearchers suspect. One
of the first hints ofthis phenomenon came from a 2004 Naturepaper 卜 Born's group. They reported
that peobut the volunteers did not know the rules evenexisted. As expecte 氏 they did no better thanchance
at first. But after 200 predictions, theirsuccess rates improved. When the subjectscame back 12
hours later to try again, they hadimproved even more-but those who'd sleptimproved about 10%
more than those whohadn't. Although it's a modest improvement,"there's a growing sense that
there's activelearning during sleep," says Wilson.
There's also increasing evidence that differ-ent stages of sleep are involved in consolidat-ing
different kinds of memory, says MatthewWalker, a neuroscientist at Harvard MedicalSchool.
Spatial memories, like those formedby playing Memory or by navigating through amaze or virtual
town, seem to be consolidatedduring slow-wave sleep. The same appears tobe true for declarative
memories, whichinvolve remembering facts-but not necessar-ily other kinds of memory. Some
studies havefound that the brain processes memories with a
www.sciencemag.org SCIENCE VOL 3 巧 9 MARC 日 2007
1361
Published 妙 AAAS
strong emotional component during rapid-eye-movement (REM) sleep and processesmemory for
motor skills, such as tapping out adifficult sequence on a keyboard, duringstage 2 and REM sleep
(see diagram, p. 1361).Why this division of labor exists is a puzzle,but Walker and others
speculate that the differ-ent physiological and neurochemical milieusassociated with different
sleep stages mayfavor certain kinds of neural plasticity.
Some researchers point out, however, thatthe literature on which sleep stages relate towhich types
of memory is peppered with con-tradictions. "There is inconsistency here, andsomeone has to be
wrong," says Jerome Siegel,a neuroscientist at the University of California,Los Angeles. Stickgold
and other proponentsof a sleep-memory link acknowledge that theyface many unresolved issues
about the role ofsleep in memory consolidation. "There aremassive questions remaining about
how exten-sive it is, how important it is, exactly whichstages of sleep affect which types of
memory,"Stickgold says.
Another wide-open issue is whether there'sa link between dreaming and memory-relatedneural
activity during sleep. The kind of directreplay of recent experience suggested byWilson's work, for
example, doesn't seem tobe the stuff of dreams. Stickgold's group hasfound that only 1% to 2% of
episodes fromdreams reflect events from the previous day.If dreams don't directly reflect the
memory-consolidation process, what do they reflect?"We're in no man's land," says Walker.
also bigger, taking up precious space. Andfinal 玩 too much LTP may saturate synapses,leaving
them unable to get any stronger whenthe brain needs to learn something new.
Sleep restores homeostasis by ratchetingdown synaptic strengths, Tononi argues. It's afar more
important service than providing amodest boost in memory performance, hesays. "Sleep is too
high a price to pay for the15% improvement we see in some things,"Tononi says. "I think it does
something muchmore fundamental to the neuron: It's the pricewe pay for plasticity." How might
sleep resetsynaptic strengths? One clue, Tononi says,comes from a study 勿 Swiss researchers pubjects were unable to learn as well the next day,Tononi says.
Tortoni's view of the role of sleep is "aninteresting and intuitive idea," Walker says.He also thinks
it's not necessarily incompati-ble with the notion that sleep enhancesmemory by strengthening the
underlyingsynapses, as proponents of the memory-replay scenario have generally assumed.
"Ithink they could act not just independentlybut synergistically."
But not everyone is ready to embraceTononi's proposal. Says Stickgold: "It's anelegant hypothesis
that doesn't have a lot ofdata behind it."
lished in the 15 January Journal ofPhysiology. They reported that 一
stimulating slices of rat brain
Bah, humbug
Those advancing the links
to cause once-per-second
bursts of neural firing
induces a type of
between sleep and memory have other hurdles 氰 to overcome. One of
synaptic weakeningcalled long 一 termdemession‘LTD).，lonom mmnxs Ir 一 s
the most disconcert-ing inconsistenno、 coincidence., .,
1｛roar 生 h。 co 甘份‘一气 nated neural tir-ing during
slow
::} % i}, .
瓤旅龚翼
黔璧夔】掣
cies in the sleep-memory liter-ature，Siegelexplains, is that
wave sleep happens at
雌舔沪 studies of totalsleep deprivationhave failed to find a
this same frequency.
Another hypothesis
Not everyone is onboard with the idea thatbrain activity during sleep is primarily aboutreplaying
recent experiences. Giulio Tononi, aneuroscientist at the University of Wisconsin,Madison, has
recently advanced a very differ-ent hypothesis. He proposes that the purposeof sleep, at least as far
as the brain is con-cerned, is to weaken neural connectionsthroughout the brain.
In Tononi's view, the synaptic connectionsbetween neurons get progressively strongerduring the
day as a result of long-term potenti-ation (LTP), a physiological process by whichneurons that fire
at the same time strengthentheir connections with each other. Most neuro-scientists consider LTP a
major mechanism ofneural plasticity-and therefore of learningand memory (Science, 22 December
2006,p. 1854). Yet a day's worth of LTP can be toomuch of a good thing, Tononi
contends.Stronger synapses increase the brain's energyneeds-a serious concern for an
energy-hogging organ that already accounts for 20010 ofa person's metabolism. Stronger synapses
are
The slow waves, one per
secon 氏 could induce LTD todial down synapses that gottoo strong during the day.
Other evidence comesfrom human studies, includ-ing one from Tononi's groupthat found that slow
waves
Hot spot. EEG recordings duringsteep indicate peak stow 一 waveactivity (red) in brain
regionsinvolved in daytime learning.
deleterious effect ondeclarative memory. LikeTononi, Siegel also questionswhether relatively
modestmemory benefits offer enoughof an adaptive advantage tocompensate for being
unre-sponsive for hours a day. "I'mjust not convinced that there is
measured by EEG in sleeping subjects weremost intense in brain regions involved in learn-ing an
arm-movement task the previous day.That's consistent with the idea that the slowwaves happen
where they're needed most torestore synaptic homeostasis. Conversely,immobilizing a person's
arm in a slingdecreases slow-wave intensity in arm-relatedareas of neocortex the following night,
Tononiand colleagues reported in the September 2006issue of Nature Neuroscience.
Tononi also points to a paper Walker'sgroup published in Nature Neuroscience inJanuary. Those
researchers found that under-graduate volunteers deprived of a night's sleepwere less able than
well-rested peers to learnnew word pairs the next day. (fMRI scans sug-gested that the deficit was
specifically relatedto memory-brain regions that modify atten-tion and alertness functioned
normally in thesleep-deprived undergrads.) With no slow-wave sleep to dampen their synapses,
the subany connection at all" between sleep and mem-ory, Siegel says. He favors the idea that
sleepevolved to help animals conserve energy fortheir entire bodies and to prevent them
frombeing active at times when they're less likely tofind food and more likely to be eaten.
To be sure, memory is not the only func-tion of sleep, counters Stiekgold, but the evi-dence that it
is one function of sleep, at least inmammals, is too great to ignore. Once sleepevolved, he says,
evolution figured out how tomake that downtime as productive as possible.Stickgold also argues
that the modestimprovements typically seen in sleep-mem-ory studies are nothing to yawn at. He
is fondof pointing out that a 15% gap in performanceis the difference between winning the
BostonMarathon and coming in 3000th. In competi-tive circumstances, small advantages canmake
a big difference. But whatever you do,try not to lose any sleep over it.
-GREG MILLER
www.sciencemag.org SCIENCE VOL 3 巧 9 MARC 日 2007
1363