INSIGHT REVIEW NATURE|Vol 437|27 October 2005|doi:10.1038/nature04286 1272 Sleep-dependent memory consolidation Robert Stickgold1 The concept of ���sleeping on a problem��� is familiar to most of us. But with myriad stages of sleep, forms of memory and processes of memory encoding and consolidation, sorting out how sleep contributes to memory has been anything but straightforward. Nevertheless, converging evidence, from the molecular to the phenomenological, leaves little doubt that offline memory reprocessing during sleep is an important component of how our memories are formed and ultimately shaped. The question of how sleep might contribute to learning and memory consolidation is an old one. In the first century AD, the Roman rhetorician Quintilian, commenting on the benefits of sleep, noted that, ���what could not be repeated at first is readily put together on the following day and the very time which is generally thought to cause forgetfulness is found to strengthen the memory���1. Although this may have been obvious to him, it has been less obvious to the research com- munity, and, until a seminal paper by Karni, Sagi and colleagues in 1994 (ref. 2), the topic received relatively little attention within either the sleep or memory research communities. But over the past 10 years, the rate of publication of research papers on sleep-dependent learning and memory consolidation has increased fivefold3. Evidence support- ing sleep-dependent memory consolidation has come from a range of molecular, cellular, physiological and behavioural studies (for a review, see ref. 4 for an opposing view, see ref. 5). One of the major problems facing this area of research is that the terms sleep, memory and memory consolidation all refer to complex phenomena, none of which can be treated as a singular event. I begin this review by clarifying my use of these terms, and then present some of the more convincing evidence from studies of procedural learning in humans. I then review more broadly the behavioural evidence for sleep-dependent consolidation of perceptual and motor skill proce- dural memories, declarative memories and complex cognitive proce- dural memories. I follow this by outlining converging evidence from molecular, cellular, neurophysiological, brain-imaging and dream studies, all of which support an important, and sometimes essential, role for sleep in memory consolidation. Sleep and memory There is more than one type of memory. Consider, for example, the capital of France, what you had for dinner last night and how to ride a bicycle. All three of these recollections require information that you have learned and stored, but they are very different types of memory. Multiple memory systems store different classes of memory in differ- ent brain regions and, quite probably, in different formats. Memories are most commonly divided into declarative memories, which are those that a person can call to mind (for example, the capi- tal of France or last night���s dinner), and non-declarative memories, which are those that are normally used without conscious recollection (for example, how to ride a bicycle or how to talk your way out of a parking ticket) (Fig. 1)6. Declarative memories are further divided into episodic memories, that is, memories of specific events (such as what you had for dinner last night), and semantic memories, in other words memories of general information (such as the capital of France). Non- declarative memories are also divided into several subcategories, such as procedural skills. Others in the field use the term memory in a more restricted manner, such as limiting it to episodic memories (see the commentary by Hobson in this issue, p. 1254). There is no consensus on what processes should be covered by the term ���memory consolidation���. The term memory consolidation origi- nally referred to a process of memory stabilization, through which memories become resistant to interference7,8. But after the initial encoding of a sensorimotor experience, a series of cellular, molecular and systems-level alterations develop over time, automatically and out- side of awareness, that stabilize and enhance the initial memory rep- resentation, converting it into a long-lasting and optimally integrated memory. These include not only cellular and molecular processes occurring at the local synaptic level, but systems-level reorganizations of individual memories as well (see refs 9, 10, for example). These additional memory-consolidation processes show the greatest evi- dence of sleep dependence, and I include all of them under the umbrella term memory consolidation. It is important to note that there is no consensus on how many distinct post-encoding processes exist, how they should be defined, and which should be considered under the rubric of memory consolidation. For example, memory sta- bilization cannot be considered absolute, because that would mean that other consolidation processes, such as enhancement and reorga- nization of memories, would not be possible. 1Department of Psychiatry, Harvard Medical School, and Centre for Sleep and Cognition, Beth Israel Deaconess Medical Centre, 330 Brookline Avenue/FD-861, Boston, Massachusetts 02215, USA. Non-declarative Non- associative Memory Episodic Semantic Declarative Conditioning Priming Procedural skills Figure 1 | Categorization of memory systems. Awake Sleep stages REM NREM-1 NREM-2 NREM-3 NREM-4 00:00 01:00 02:00 03:00 04:00 Time 05:00 06:00 07:00 Figure 2 | Categorization of sleep stages. Nature Publishing Group �� 2005

NATURE|Vol 437|27 October 2005 INSIGHT REVIEW 1273 sequence test14,15 and a motor adaptation test16, demonstrated that all subjects show post-training improvement after a night���s sleep but not during an equivalent period of being awake. The results also demon- strated that the amount of overnight improvement correlates with the amount of specific sleep stages or sleep events, and that total or partial sleep deprivation can prevent the normal overnight improvement. These results leave little doubt that sleep contributes to the consol- idation of memories, especially their enhancement. But the details of these processes remain unclear. Although improvement on the visual texture discrimination task correlates with the levels of REM and SWS, improvements in the motor sequence task correlate with lighter stage 2 NREM. With the motor adaptation task, improvement was linked, using EEG patterns, to SWS. Thus, even for these procedural skill tasks, sleep-dependent consolidation does not seem to depend consistently on one specific aspect of sleep. Instead, each stage of sleep seems to contribute differently to these processes, and we have pro- posed that the multiplicity of sleep stages has evolved, in part, to pro- vide optimal brain states for a range of distinct memory consolidation processes. I will now look at each of the tests in more detail. Visual texture discrimination Several studies have supported the sleep-dependent enhancement of a foreground���background discrimination task (Fig. 3a���c). In this task, subjects must identify the orientation of an array of diagonal bars against a background grid of horizontal bars. Although an initial study found overnight improvement blocked by REM deprivation, but not When we speak of consolidation processes as being ���sleep depen- dent���, we are hypothesizing that they occur primarily during sleep. But most sleep-related processes can occur during periods of wakefulness and vice versa (for example, waking hallucinations and sleep walking). Nonetheless, in most cases, experimental studies have found sleep- dependent processes occurring only during sleep. Unlike the nature of memory and memory consolidation, the struc- ture of human sleep is clear. A night of sleep is composed of ~90- minute cycles divided into periods of rapid eye-movement sleep (REM) and non-REM sleep (NREM), with NREM further divided into stages 1 to 4 (Fig. 2). Stages 3 and 4 are the deepest stages of sleep, and are referred to collectively as slow-wave sleep (SWS) on the basis of the patterns of large, slow (0.5���4-Hz) oscillations observed on the elec- troencephalogram (EEG). Sleep stages differ not only in depth of sleep but also in the frequency and intensity of dreaming, EEG oscillations, eye movements, muscle tone, neuromodulation of cortical circuits, regional brain activation and communication between memory sys- tems11. REM, stage 2 and SWS have all been implicated in sleep-depen- dent memory processing, as have specific patterns of synchronous cortical neuronal activation associated with these stages, including ponto-occipitogeniculate waves and theta rhythms in REM, sleep spindles in stage 2 and the slow-wave activity of SWS. Sleep-dependent memory enhancement Results from three laboratories who asked volunteers to perform three different tasks, a visual texture discrimination test12,13, a motor Retest AM PM AM AM AM PM PM PM AM Train Retest Train Retest Train 28 26 24 22 20 Sequences /30 s Sequences /30 s 85 Day Night 48 h 65 45 25 % Stage 2 NREM sleep in fourth quarter 6 4 2 0 Improvement in rate Visual texture discrimination task Motor sequence learning task Motor adaption learning task Days Improvement (ms) SWS1 �� REM4 (% �� %) 30 20 10 0 Improvement (ms) Test interval (h) Daytime Overnight Improvement (ms) 20 10 0 * * * a Sleep versus wake d Sleep versus wake g Sleep versus wake b Sleep stage correlation e Sleep stage correlation h Slow-wave activity correlation j Localization of slow-wave activity increase 6 12 18 24 50 0 1 2 3 4 7 100 150 200 c Sleep deprivation f Sleep deprivation Sleep Wakefulness Performance change (%) (mean directional error) 12 8 4 0 ���4 30 20 10 0 0 40 Improvement (mean directional error) Rotation/no rotation 50 132 185 20 40 30 20 10 0 Sleep Sleep Sleep % Slow-wave activity increase % 15 10 5 0 ���5 ���10 ���15 40 30 20 10 0 Figure 3 | Sleep-dependent consolidation of procedural memories. a���c, Participants in a visual texture discrimination task show improvement only after post- training sleep, even when finger movement was suppressed with mittens during wake periods (d). Improvement correlated with early-night slow-wave sleep and late-night REM sleep. d���f, Participants in a motor sequence finger-tapping task show similar sleep-dependent improvement, correlated with late-night stage 2 non-REM sleep. g���j, Participants in a motor adaptation task also show sleep-dependent improvement, correlated with EEG slow-wave activity in task-related regions of the cortex. All error bars represent s.e.m. See the text for further details. Green bars, performance without intervening sleep or without sleep on the first post-training night. Dark red bars, performance after normal sleep. See the text for further details. Panel J is reproduced with permission from ref. 21. Nature Publishing Group �� 2005