A molecular basis for interactions between sleep and memory

Synopsis

The electrophysiological properties of the sleeping brain profoundly influence memory function in various species, yet the molecular nature by which sleep and memory interact remains unclear. We summarize work that has established the cAMP-PKA-CREB intracellular signaling pathway as a major mechanism involved in the wakeful consolidation of memory in many organisms while highlighting newer evidence that this pathway has a role in sleep regulation, sleep deprivation and potentially sleep-memory interactions. We explore the possibility that sleep might influence memory processing by reactivating the same molecular cascades first recruited during learning during a sort of “molecular replay”. Lastly, we discuss how new approaches together with established techniques will aid in our understanding of the nature of sleep-memory interactions.

I. Introduction

The “sleep-for-memory” hypothesis has gained considerable momentum in response to a growing body of literature detailing the beneficial effects of sleep on memory function. Indeed, memory function from humans to flies appears to benefit from, if not depend on, some core property of sleep. Over time, sleep has been proposed to serve a variety of functions ¹²⁰. However, it has been argued that the core function of sleep is likely cellular since the function of proteins from membrane receptors to transcription factors are remarkably consistent across phylogenetic lines, as is the need for sleep, regardless of differences in sleep architecture and brain complexity ²¹. The sleep-for-memory hypothesis posits that a primary cellular function of sleep is memory consolidation, the process by which newly acquired information is stabilized and stored. Like the need for sleep, the molecular mechanisms that mediate memory consolidation are also conserved across species. Although little is known about the molecular mechanisms mediating sleep-dependent memory processing, there is evidence to suggest that wakeful experience is recapitulated during sleep, suggesting, by extension, that the same molecular processes that promote memory consolidation might be re-engaged during sleep to further stabilize and improve memory.

Here we present several key concepts and molecular processes that are known to underlie learning and memory consolidation. We then give an overview of some the electrophysiological events specific to various sleep states that might serve to modulate activity within these signaling pathways. We address recent progress, questions still unanswered, and the challenges and limitations associated with the methodologies used to study sleep-memory interactions on a molecular level.

II. Stages of memory processing

Memory storage is often differentiated into pharmacologically dissociable stages ¹. During the encoding or acquisition phase, perceived information is transformed into synaptic activity through the propagation of electric and chemical signals between neurons at synapses. Changes in synaptic activity resulting from the encoding process must be stabilized or consolidated if they are to become part of an enduring memory. The consolidation process can be further distinguished into three forms: synaptic consolidation, which occurs soon after acquisition or encoding; systems consolidation, whereby memory is reorganized throughout the brain over weeks to years; and reconsolidation which is a stabilization process that can occur after the retrieval of a memory. Often, multiple sessions or trials are required to form a particular association and related memory. Therefore, memory of those associations might involve multiple iterations of acquisition, consolidation and reconsolidation. In this way, sleep could potentially influence multiple stages of memory processing (see Figure 1) ¹⁷⁰. Here we review some of the key molecular events involved in each stage of learning and memory formation with the idea that some or all of the processes described below could potentially be re-engaged during sleep-dependent memory processing. The reader is also referred to Figure 2 for a schematic of the molecular pathways involved in memory consolidation.

Fig. 1. Permissive and inhibitory effects of sleep and sleep deprivation on the stages of memory formation.

Encoding allows relevant environmental information to be converted into a neural construct that can be stored within the brain and recalled at a later time. Newly learned information must be stabilized by strengthening and stabilizing synapses that were activated during learning. Once consolidated, memory can be retrieved and used by the organism. Under certain conditions, retrieved memories render them sensitive to disruption, requiring them to undergo the process of reconsolidation. Reconsolidated and previously learned memories may also be consolidated on a systems level. Systems consolidation typically refers to the reorganization or transference of memory traces out of the hippocampus to areas in the neocortex. Sleep-dependent processing of memory has been shown in impact all stages of memory formation. Neural oscillations during NREM and REM sleep have permissive, enhancing effects on each stage of memory formation (depicted with blue arrows), whereas sleep-deprivation or the prevention of sleep-related oscillations impairs memory formation (shown with red lines). The role of sleep and sleep deprivation in reconsolidation is less understood and is denoted with question marks.

Fig. 2. Signaling pathways involved in the encoding and consolidation of learning and memory.

During encoding, presynaptic release of glutamate activates NMDA and AMPA-Rs in the postsynaptic cell (synapse shown on the left). Ca²⁺ influx through these receptors and through voltage-gated Ca²⁺ channels activates Ca²⁺-calmodulin which can activate CaMKII, PKA and others, which can then increase AMPA and NMDA receptors through the phosphorylation of various receptor subunits. Long-term memory formation requires transcription and translation dependent mechanisms in order to strengthen and stabilize synaptic connectivity. Ca²⁺ influx during encoding can activate adenylyl cyclase (AC) which increases intracellular levels of the second messenger cAMP. Increased cAMP levels upregulate PKA-dependent phosphorylation of other protein kinases (e.g., the MAP kinases) which, together with PKA, can translocate to the nucleus to activate the transcription factor CREB. cAMP also activates Epac⁹⁸, although the role of Epac in sleep-related plasticity has not been investigated. Other signaling proteins such as CAMKIV and CAMKII²,¹⁹,³⁷,⁷⁵,⁸⁶,¹⁰⁴,¹⁷³,¹⁷⁵ also play a role in CREB activation. CREB together with its coactivator CBP (and others) regulate gene transcription by promoting the binding of transcriptional machinery to target genes and by making target genes more accessible to transcriptional machinery by altering chromatin structure⁹,¹²,¹⁸,⁴⁵,⁵⁷,⁷⁷,¹³⁴,¹⁶⁵. Proteins synthesized in the soma or locally in dendrites ¹⁵² are transported to synapses that have been tagged by prior activity. AMPA-R trafficking into the post-synaptic membrane (shown in the synapse on the right) plays an important role in the long-term maintenance of synaptic strength. Importantly, cAMP signaling can be regulated by the actions of neurotransmitters that bind to G protein-coupled receptors. Levels of modulatory neurotransmitters, many of which are coupled to both G_s- and G_i-coupled receptors, change dramatically during the sleep-wake cycle. Activation of G_s- and G_i-coupled receptors stimulate and inhibit AC activity, respectively, which alters cAMP production accordingly ¹. Sleep deprivation and circadian rhythms can also alter intracellular levels of cAMP, MAPK and other plasticity-associated molecules ³⁵,¹⁶⁴.

Acquisition/Encoding

Many neurotransmitters systems are involved in the encoding process, but the study of glutamate-mediated transmission of information through α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPA-R) and N-methyl-D-aspartate receptors (NMDA-R), has occupied center stage in the study of learning and memory and its cellular correlate, long-term potentiation (LTP, discussed below) ^87,119. Transduction of environmental information through glutamate and other modulatory receptors triggers transcriptional and translational mechanisms that mediate gene expression, synaptic plasticity and ultimately behavior. Specifically, Ca²⁺ influx through glutamate receptors can trigger the covalent modification (e.g., phosphorylation) of pre-existing proteins at active synapses ^67,135 resulting in short-term changes in synaptic strength. These short-term changes, along with changes in the presynaptic neuron, alter the probability of the post-synaptic neuron firing to future stimuli. As shown in Figure 2, activity-dependent increases in intracellular Ca²⁺ levels can activate Ca²⁺-calmodulin which can then activate calmodulin dependent protein kinase, (CaMKII) and protein kinase A (PKA). CaMKII and PKA can then phosphorylate AMPA and NMDA-R subunits thereby altering synaptic transmission. This increase in receptor function is temporary and allows information to be conveyed in the short term. Short-term memory can then be committed to long-term memory (LTM) or forgotten as the synaptic connections eventually weaken.

Two main theories have been proposed to explain how sleep might influence learning and memory formation. The “active systems” theory proposes that cellular activity observed during learning is replayed or recapitulated in some form during post-learning sleep ^82,174. Cellular replay during sleep could mimic the encoding process or help to link new memories to previously consolidated memory networks. According to the “synaptic homeostasis” hypothesis ¹⁵⁷, the net strength of synapses increase during wake and are downscaled during sleep. Indeed, AMPA-R subunit phosphorylation has been shown to be higher during wake and lower after sleep in the cortex and hippocampus of rats ¹⁶⁶ (but see 60). The elimination of synapses that have only undergone weak encoding, and are not needed for LTM formation, could reduce the “signal to noise” ratio in the brain and allow the encoding of new information to proceed more efficiently. Both theories cite mechanisms that could impact encoding and memory processing and are not necessarily mutually exclusive.

Synaptic consolidation

Gene transcription and translation is required for the consolidation of LTM (reviewed in ⁶⁷) and for the maintenance of LTP ¹³¹, the long-lasting enhancement of signal transmission between neurons at synapses that have been repeatedly stimulated. Like STM and LTM, LTP can be distinguished into at least two forms: early (E-LTP) and late (L-LTP) where L-LTP requires transcription and translation. An important aspect of L-LTP is the maintenance of increased synaptic strength brought about by the trafficking of AMPA-Rs containing the GluR1 subunit to the appropriate synapses. Conversely, the strength of synapses can also be decreased during synaptic depression by the removal of AMPA-Rs from the synapse ^22,99. This balance between potentiation and depression, or plasticity, is generally thought of as the basis of memory formation ¹¹³. Indeed, the study of LTP has contributed a great deal to our understanding of the cellular and molecular biology of learning and memory and is becoming an important tool in understanding sleep-memory interactions ^10,97,131.

Increases in intracellular Ca²⁺ also play a critical role in plasticity-associated transcription and translation. Ca²⁺ influx stimulates the production of the second messenger 3′-5′-cyclic adenosine monophosphate (cAMP) by adenylyl cyclase. cAMP activates at least three targets important in memory processing: PKA, exchange protein directly activated by cAMP (Epac), and hyperpolarization-activated cyclic nucleotide-gated channels (HCN) channels (see ⁴ for review). Once activated, PKA and other plasticity-associated kinases (e.g., MAP kinase, extracellular signal-regulated kinase (ERK), CaMKII and CaMKIV ^37,104) can phosphorylate and activate the transcription factor cAMP response element binding protein (CREB) which, together with its co-activators regulate, plasticity-associated gene transcription ^{2,124,132,133} cAMP can also activate Epac, which through Rap1 and other targets, can promote memory formation possibly through CREB activation ^52,98. Plasticity-related gene products (PRPs) are translated and targeted to previously activated synapses that require stabilization. Protein synthesis occurring locally in the dendrites can also be regulated by calcium influx through NMDA receptors, and has been shown to play an important role in memory formation ¹⁷.

At least one round of transcription- and translation-dependent consolidation is required to stabilize a new memory. However, Bourtchouladze, et. al. ¹² and others ^18,45,57, have shown that, depending on the training protocol used, two or more waves of PKA activation and protein synthesis are required after training ^9,77,134. The exact nature of these waves of activity is unknown but their existence suggests that some sort of feedback or cyclical activity occurs during the post-training time – a time when the experimental subjects were likely sleeping.

Fluctuations in the levels of excitatory and inhibitory neurotransmitters that occur during sleep and extended wakefulness can have important influences on memory consolidation (see ^58,96 for recent reviews). For example, sleep-dependent fluctuations of neurotransmitters that bind to G protein-coupled receptors (GPCRs) can stimulate or inhibit adenylyl cyclase to modulate cAMP levels (see Figure 2). Dopamine and acetylcholine levels are known to vary dramatically during different sleep stages and can activate both stimulatory and inhibitory GPCRs. Thus, the overall neuromodulatory effect of sleep-related neurotransmission may also depend on brain region-specific expression patterns of a particular GPCR. Extended wakefulness and sleep deprivation can also increase levels of adenosine which decreases cAMP by binding G_i-coupled GPCRs. In addition, sleep deprivation can increase levels of phosphodiesterase, the enzyme responsible for degrading cAMP, although the mechanism of this increase is still under investigation ¹⁶⁴.

So how might neurons keep a “record” of which synapses out of thousands should be strengthened and maintained? This could be an even bigger problem during sleep when we are unable to intentionally rehearse information. Restated from a molecular level perspective, how does encoding at a particular synapse instruct the cell to deliver the required PRPs resulting from gene transcription and translation to a particular synapse? To solve this problem, it was proposed that encoding triggers the production or delivery of molecular “tags” at synapses activated during learning, providing an “address” to facilitate the delivery of PRPs ^46,114. Evidence of synaptic tagging was first demonstrated in various neuronal preparations ^{47,48,102,103} and has been tied to CREB- and transcription-dependent mechanisms ^103,125 and to A kinase-anchoring proteins (AKAPs), that serve to compartmentalize PKA in functionally distinct regions of the neuron ⁷¹. The identity of the tag(s) and specific PRPs remain, for the most part, unknown. However, Homer1a has been proposed to fulfill most criteria of a potential tag ¹¹⁷ and is upregulated in mice after sleep deprivation suggesting there may be a connection between the overall level of activated synapses and tag expression with extended wakefulness ¹⁰⁰. PKMzeta is also thought to be a PRP that plays a role in the maintenance of L-LTP and memory ^139,177. Indeed, any of the cellular machinery involved in local translation might serve as a tag. Importantly, a behavioral correlate of tagging has recently been described ¹⁰⁹ where weak inhibitory avoidance training in rats, which does not normally produce LTM for the event, can be consolidated into LTM by exposure to a novel environment. This conversion requires protein synthesis and dopamine (D1/D5) receptor activity. In light of these findings, tagging might be a method by which PRPs mobilized during post-learning wake and sleep are targeted to the appropriate synapses.

Because bouts of sleep occurring soon after learning enhance memory (reviewed in ¹⁶⁹), it appears that sleep benefits synaptic consolidation in particular. Sleep can improve speed and reduce errors in motor performance tasks or in word recall tasks ^50,106. Whether these benefits are conferred by sleep-specific molecular pathways or boost activity in molecular pathways already activated during wakeful consolidation is not fully understood. The fact that sleep-specific gene expression has been demonstrated could argue for former possibility, but it does not exclude the latter. That is, sleep-dependent gene expression in addition to sleep-dependent modulation of gene expression initiated during learning could occur in parallel.

Systems consolidation

Once consolidated on a synaptic level, newly formed memories undergo a process of reorganization on a broader, systems level. Systems consolidation traditionally refers to the slow transference of memory out of the hippocampus to the neocortex for permanent storage ^25,44,151, but more current views include mechanisms by which new memories are incorporated into distributed networks of previously consolidated memories ¹²⁸. Further, it has been cited that some memories always remain hippocampus dependent and that others have never resided there, stressing the notion that the neocortex is likely more involved during early consolidation than previously appreciated ^161,172. These theories are beginning to bridge the events that occur during synaptic consolidation to those classically defined as systems consolidation ²⁵, incorporating a role of sleep in both processes.

The sleep-for-memory hypothesis also posits that sleep, in addition to enhancing encoding and synaptic consolidation, promotes the reorganization of memory during systems consolidation. Pharmacological studies of the function of sleep in systems consolidation are difficult to perform due to the long time-course of memory reorganization and the distributed nature of memory traces, but genetic studies in mice ¹¹² and human imaging studies appear to be consistent with a role for sleep in systems consolidation. Using a declarative word-pair learning task, Gais et al. ⁴⁹ showed that sleep after learning increased functional connectivity between the medial prefrontal cortex and hippocampus during retrieval tests 48 hrs later and enhanced activity in prefrontal cortex during retrieval 6 months later. Sleep-dependent shifts to neocortical-based memory representations could result in more efficient retrieval ^128,155. A recent functional MRI study conducted by Orban et al. ¹¹⁸ also demonstrates that sleep promotes the reorganization of brain activity over long periods of time. The authors trained human subjects in a place-finding navigational task; those that slept after training tended to use an extended hippocampo-neocortical network to perform the task in early retrieval sessions and striatal regions in later sessions. However, subjects that were sleep deprived showed significantly less striatal activity during later retrieval tests, suggesting sleep deprivation altered the normal course of memory reorganization over time. Interestingly, performance between the two groups was unaltered, demonstrating that the reorganization of memory between systems does not always enhance performance in learning and memory tasks but can reflect the transference of well-learned information to systems designed to process automated behaviors.

Retrieval and Reconsolidation

Retrieval refers to the reactivation of memory traces. Under certain conditions, retrieval of a memory trace renders it sensitive to disruption by amnesic treatments for a short period of time ^55,107,137. Such sensitized memories require “reconsolidation,” during which the retrieved memories undergo further consolidation to be restabilized and stored. Reconsolidation could be a mechanism by which older memories are update and crosslinked with newly formed memories ⁹². Although, much less is known about the molecular events underlying retrieval and reconsolidation, they appear to have unique molecular signatures involving specific molecular pathways and brain regions ^93,111,127. However, PKA activity appears to be a consistent requirement for the retrieval and reconsolidation of many forms of memory, at least in some brain regions ^105,110. For example, retrieval of 1-trial inhibitory avoidance memory requires the activation of AMPA and NMDA-Rs in addition to PKA and ERK in the hippocampus and cortex. However, in the amygdala, only AMPA-R activity is required for retrieval; NMDA-R activity, PKA and ERK are unnecessary ^8,80,81. Although reconsolidation has been demonstrated in humans ¹⁶⁷ there has been much debate on the boundary conditions that dictate whether reconsolidation of a retrieved memory occurs ¹⁶⁰ and if amnesia related to the blockade of reconsolidation reflects memory erasure or impaired retrieval ¹¹¹. The effect of sleep on retrieval and reconsolidation has been less studied, but sleep may affect the reorganization of memories that undergo reconsolidation (see ^128,148 for more detailed reviews on the role of sleep in memory reactivation and reconsolidation) and REM sleep deprivation impairs the retrieval of discriminative avoidance training in rats ³.

Summary of Part I

Memory traces instantiated during wake appear to be altered in a sleep-dependent fashion. Sleep occurring within several hours of training, can improve performance in a variety of memory tasks and induce changes in retrieval-associated brain activity ^40,171, suggesting that the neuronal substrates of memory are redistributed during sleep ³⁰. Initial studies suggest that sleep-dependent plasticity, much like the initial encoding and consolidation of memory, requires the activation of NMDA-Rs, AMPA-Rs, and PKA ^6,42,43,50. Synaptic tagging and multiple waves of intracellular activity might guide and induce the sleep-related strengthening of memories, respectively. However, it is not entirely clear whether sleep-dependent plasticity is controlled by cellular and molecular events that precede sleep, in a sense priming the trace for further modification during sleep, or whether sleep acts more or less independently on memory. Thus, a primary goal from a molecular standpoint is to determine how the unique properties of the sleeping brain might tap into the molecular mechanisms known to instantiate memories during wake. Below we examine how sleep might interact with the molecular pathways involved in memory consolidation.

II. Sleep states and memory

In many organisms, sleep can be broadly characterized as fluctuating between two states: rapid eye movement (REM) sleep and non-rapid-eye movement (NREM) sleep ⁸⁹ and each of these states may contribute differentially to memory consolidation ¹³⁰. Several fundamental oscillatory patterns can be seen in the EEG during wake and sleep, including delta (< 1.0-4 Hz), theta (4-9 Hz), beta (15-35 Hz) and gamma (>40 Hz) oscillations ⁷⁶. During wake, low-amplitude, high frequency gamma oscillations can be observed throughout the cortex^69,146 but as the brain transitions from wake to NREM sleep (which has four stages in humans), gamma oscillations give way to the characteristic high amplitude, low-frequency slow waves (<4 Hz) of NREM sleep ¹⁴⁴. During REM sleep, rapid desynchronous gamma oscillations are obvious in the cortical EEG as are oscillations in the theta frequency ^15,156,162 which are similar to theta rhythms observed during exploratory movements performed during wake ^{7,13,59,74,116}. REM sleep is further characterized by the presence of muscle atonia except for eye and whisker movements in humans and rodents, respectively ^5,27,28,84.

Several approaches have been traditionally used to study sleep-memory interactions. As illustrated above, sleep deprivation has been a useful tool but can produce unintended effects on learning and memory as it can be quite stressful on the organism. A second approach correlates sleep-related oscillations (e.g., their duration, frequency, phase and power) to gains in task performance. A third approach examines how learning alters subsequent sleep architecture and EEG patterns. Newer approaches attempt to alter or prevent the occurrence of specific oscillatory rhythms during sleep to provide more direct evidence of sleep-memory interactions. In the following sections, we further examine how oscillatory patterns characteristic to NREM and REM sleep might influence plasticity-associated signaling pathways which are summarized in Figure 2. We also discuss how homeostatic and circadian control of sleep states might influence sleep-memory interactions on a molecular level.

NREM and memory

During NREM sleep, slow-wave oscillations in the delta frequency and specialized oscillations known as sleep spindles (11-15 Hz) ^115,136 and sharp wave-ripple (SPW-R) complexes (140-200 Hz) can be observed in the hippocampus and neocortex ^14,116,178. Spindle activity during stage II NREM appears to correlate well with improvements in memory tasks (see ²⁶ for a detailed review on sleep spindles). Stages III and IV constitute deep sleep and are collectively referred to as slow wave sleep (SWS). Slow waves reflect the synchronous transitions between “up states” during which neurons of the cortex and hippocampus are collectively depolarized and fire together, and “down states” during which they are hyperpolarized and relatively silent ⁷⁹. Indeed, evidence suggests that spindles and SPW-Rs play a particularly important role in sleep-memory interactions^29,70 and tend to occur together during SWS ^{38,108,140,141} This temporal coherence is thought to reflect the coordinated transfer of information between the hippocampus and neocortex ^126,140, however the exact nature of this hippocampal-neocortical “dialogue” is not fully understood on a cellular or molecular level ^158,29,147. Interestingly, the sequence in which specific neurons (e.g., place cells in the hippocampus) are activated during wake appears to be recapitulated, or “replayed,” in various forms during subsequent periods of wake ⁹⁰, NREM (in particular during SPW-R events ^90,174), and REM sleep ¹⁷⁴. Furthermore, blockade of hippocampal output in genetically modified mice impaired systems consolidation and reduced the frequency of hippocampal ripples and experience-dependent, ripple-associated reactivation of hippocampal cell pairs ¹¹².

In light of evidence demonstrating that sequences of neuronal activation seen during learning reoccur during post-learning periods of wake and sleep, it is tempting to hypothesize that there is an ensuing period of “molecular replay” when molecular events engaged during learning are re-engaged during sleep. Several studies support the notion that signaling pathways mediating hippocampal long-term plasticity must be reactivated to consolidate memory, perhaps even for its maintenance as a ‘remote’, cortex-dependent memory trace ^9,24,138,159.

Perhaps the most compelling evidence that sleep-associated oscillations influence memory processing has been obtained from studies in which oscillations have been either ehanced or inhibited after learning. Marshall et al. used transcranial electrical stimulation in humans to stimulate SWA during NREM sleep after learning in two different hippocampus-dependent tasks or two procedural motor skill tasks ¹⁰¹. Stimulation with slow, 0.75 Hz oscillations during early night NREM sleep selectively enhanced memory retention in the hippocampus-dependent tasks and increased slow wave sleep, as well as endogenous cortical slow oscillations and spindle activity in the frontal cortex, providing the most direct evidence of a role for sleep-associated oscillatory rhythms in memory. Stimulation at 5 Hz or at 0.75 Hz late in the night had no enhancing effect relative to sham stimulation. Interestingly, although memory for the procedural tasks benefitted from sleep in general, no further gains were observed with slow-wave stimulation. Another recent study examined the effect of suppression of SPW-R complexes in rats trained in a hippocampus-dependent spatial-reference memory task using the radial arm maze ⁵⁴. SPW-R events blocked by stimulation of the ventral hippocampal commissure during post-learning rest and sleep significantly impaired performance relative to rats that received the same number of stimulations but that had intact SPW-R events. Lastly, it has been demonstrated that re-exposure to olfactory cues during SWS, that were part of a training context in a prior learning task, was found to enhance subsequent memory ¹²⁹. Thus, the type and timing of oscillations during NREM sleep appear to play an important role in sleep-dependent memory consolidation. Despite advances in our understanding of sleep-memory interactions on an electrophysiological level and of the molecular biology of LTM formation, little direct evidence exists to link sleep-related oscillations with memory processing on a molecular level.

REM and memory

During REM sleep, brain activity in the gamma and theta frequency range is similar to that during wake ^94,145, but the body is paralyzed due to atonia; for this reason, REM is often referred to as paradoxical sleep. Interestingly, REM sleep deprivation can result in memory impairments; however the degree of impairment appears to be determined by the complexity of the task. Indeed, simple tasks (e.g., passive avoidance, one-way active avoidance and simple mazes), which are quickly learned, are less affected by REM sleep deprivation. Conversely, tasks which require more complex associations to be learned (e.g., discriminative and probabilistic learning, complex maze learning and instrumental conditioning) are particularly sensitive to REM sleep deprivation ^66,123. Furthermore, REM sleep deprivation is effective only if the animal has reached a certain level of learning ^34,66. In humans, the amount of REM sleep late in sleep cycle (and NREM sleep early in the night) correlates with improvements on a visual discrimination tasks ¹⁴⁹. Similar results have been found in a number of tasks and species and have been recently reviewed in detail elsewhere ^29,70.

On a molecular level, little is known of the effects of REM sleep on memory processing. REM sleep deprivation, however, has been linked to lower binding of noradrenaline to alpha-and beta-adrenergic receptors throughout the brain ⁶⁸ which could result in memory impairments by lowering cAMP via decreased G_s stimulation of adenylyl cyclase. REM sleep deprivation has also be been linked to lower levels of phosphorylated CREB and deprivation of both NREM and REM for 5 hrs has recently been shown to produce decreased cAMP signaling in the hippocampus accompanied by impairments in the consolidation of contextual fear memory in mice ¹⁶⁴. Thus, cAMP-related signaling appears to also play a role in the effects of REM sleep deprivation on memory consolidation.

III. Homeostatic and circadian control of sleep-dependent memory processing

To understand sleep-memory interactions, we must also understand how sleep and synaptic plasticity are regulated by homeostatic and circadian processes. The homeostatic regulation of sleep can be seen in increased levels of adenosine during sleep deprivation, changes in gene expression, the post-translational modification of proteins, and increases in SWS during periods of sleep rebound after sleep deprivation ¹⁶³. Importantly, wakeful experience and learning have been shown to increase the length of time spent in NREM and REM sleep ⁶². Increased sleep duration due to learning, for example, could provide a longer window of time for oscillatory events including spindles and SPW-Rs to occur, thereby increasing the overall beneficial effects of sleep on memory consolidation. Alternatively, increased sleep duration could accompany learning-dependent alterations in the electrophysiological properties (e.g., power) of sleep-associated oscillatory activity. Indeed, learning-dependent increases in sleep need accompanied by increased SWA during NREM sleep and higher numbers of spindles in brain regions activated during learning have been demonstrated and correlated with levels of sleep-dependent performance enhancements ^61,72,73. Thus, sleep need, at least under certain circumstances, appears to be regulated at local and global levels, warranting further investigation ¹⁵³. In contrast, Walker et al. ¹⁶⁸ has shown that doubling the quantity of initial training does not alter the amount of sleep-dependent learning in a finger tapping motor task. However, it has been shown that more difficult motor sequences benefit more from sleep than easy motor sequences ^32,91. This could be because weakly encoded procedural memories experience a greater benefit from sleep than strongly encoded memories, simply because there is more room for improvement. If task performance in well-learned memories is near maximal, as far as accuracy and speed are concerned, then little benefit from sleep might be expected.

In associative learning paradigms (e.g., fear conditioning), task difficulty might predict the magnitude of post-learning changes in sleep architecture and oscillatory patterns. Indeed, Steenland et al. ¹⁴³ recently examined the effects of trace fear conditioning where the associative strength between a conditioned stimulus (a tone) and the unconditioned stimulus (a footshock) was altered by changing the difficulty of the task. That is the task can be made more difficult to learn by lengthening the duration of trace time - the delay between termination of the CS and presentation of the US. Mice conditioning using a short, 15 sec trace resulted in significant enhancements in delta power during subsequent NREM sleep relative to baseline, whereas mice conditioned with a more difficult 30 sec trace showed no significant enhancements in delta power relative to baseline. Together, these data suggest learning and plasticity affects sleep need and the electrophysiological properties of subsequent sleep, which, in turn, can modulate sleep-dependent memory enhancements.

Circadian rhythms, in addition to homeostatic mechanisms, can influence memory processing and post-learning sleep ¹¹ through the regulation of transcription and translation within various brain regions ^53,154. For example, hippocampal gene expression and enzyme activity have been shown to vary over the course of the day ^20,31,35. In fact, up to 10% of the mammalian transcriptome may be under the control of circadian rhythms ^19,33,95,121 Relevant to the role of cAMP-PKA-CREB pathway in sleep-memory interactions, Eckel-Mahan et al. have recently shown that ERK, MAPK and CREB phosphorylation, in addition to cAMP levels, undergo circadian oscillations in the hippocampus. Interference with these oscillations with physiological and pharmacological agents impaired the maintenance of hippocampal-dependent LTM suggesting reactivation of the cAMP-CREB pathway is influenced in part by circadian rhythms ³⁵.

Clock genes (including Clock, Bmal1, Period, Cryptochrome, and Timeless in Drosophila) genes are also intimately tied to circadian rhythms in mammals and Drosophila and are thought to modulate learning and memory but there role in sleep-memory interactions have been largely unexplored ^36,65. Clock genes could provide a means of binding sleep and memory consolidation. For example, clock genes may play a role in the temporally restricted expression of the effects of learning on sleep (i.e., increased slow wave activity confined to the first few hours of sleep) or to explain why sleep can be delayed until later in the same day yet still benefit memory ^{39,88,150,168} Certainly, genetic models targeting clock genes will prove useful in determining their role in sleep-memory interactions.

IV. Molecular replay and sleep-dependent memory processing

The replay of hippocampal cellular activity, during ripple events for example, is thought to reflect cellular activity that occurred during prior bouts of wake ^{41,85,122,142,174} Therefore, it may be possible that cellular replay during sleep could trigger a similar replay of the molecular events that occurred during encoding and early consolidation ⁵¹. Although the notion of molecular replay is far from conclusive, many of the signaling pathways shown in Figure 2 function in synaptic plasticity ⁷⁵, synaptic homeostasis ⁷⁵ learning and memory ^86,173,175 and sleep ^19,63. Indeed, the cAMP-PKA-CREB pathway plays a role in the transduction of environmental information during memory formation and is modulated by sleep deprivation, circadian rhythms and neurotransmitter systems involved in sleep-wake regulation.^{56,63,64,83,164,176}. Thus, the cAMP-PKA-CREB signaling pathway is well-suited to play an integrative role in sleep-memory interactions.

V. The use of genetic models to study sleep-memory interactions

The ability to manipulate gene expression in mice and other organisms is an important tool in understanding sleep ²³. Regional and even cell type-specific expression of altered genetic material can assess the role of those genes within precise loci in the brain. While the temporal control of gene expression has improved ⁷⁸, the required temporal resolution to study sleep-memory interactions in mice, for example, must be on the order of minutes, since most studies are conducted while test subjects are otherwise sleeping. Moreover, rapid regulation is required because mice return to sleep shortly after training. Without this level of temporal control, genetic modifications could produce non-specific effects on memory and sleep or alter sleep-wake regulation preventing a direct analysis of sleep-memory interactions. Approaches similar to those used in optogenetic systems, whereby specific cell types can be optically stimulated in vivo via expression of light-activated channel rhodopsins ¹⁶, could be engineered to precisely modulate oscillatory patterns that occur during sleep to provide a direct measure of their effects on memory.

VI. Conclusion

Mounting evidence demonstrates that sleep plays a beneficial role in memory consolidation. Multifunctional signaling pathways involved in memory, sleep and sleep-wake regulation may have evolved to integrate the effects of sleep with memory processing. However, in order to better understand the role of sleep in memory processing, we need to establish testable, falsifiable hypotheses using a combination of new genetic approaches and methods that regulate the electrophysiological properties of the brain during sleep. These approaches will provide more direct evidence of sleep-memory interactions that extend beyond the correlational confines of standard methodologies toward a deeper understanding of the molecular basis of sleep-memory interactions.