A few hours of wakefulness or sleep can modify the molecular composition of excitatory synapses, change their efficacy and make synapses grow or shrink. Before discussing sleep-dependent regulation of long-term synaptic plasticity, it is opportune to review the major mechanisms by which synapses are strengthened or weakened, reshaped, and eventually stabilized at the molecular level.

Synaptic plasticity is the ability of a synapse to change in strength in response to use or disuse. Diverse forms of synaptic plasticity exist at excitatory, glutamatergic synapses in the mammalian brain. Among them are long-term potentiation (LTP), long-term depression (LTD), and homeostatic scaling of synaptic strength. LTP and LTD are sustained increases and decreases, respectively, in synaptic efficacy induced by patterned synaptic activity. Homeostatic scaling refers to the ability of a neuron to modulate its firing rate by globally increasing or decreasing synaptic efficacy on all inputs to the dendrite. In contrast to LTP and LTD, which are input-specific (Hebbian plasticity), homeostatic scaling does not affect the relative difference in strength between inputs (non-Hebbian plasticity).

Importantly, LTP and LTD share a common set of mechanisms even if they represent opposite changes in synaptic strength. Both forms of long-term synaptic plasticity require actin cytoskeletal remodeling within dendritic spines and changes in spine morphology. Many studies report enlargement of spines in LTP and shrinkage or loss of spines in LTD (Okamoto et al., 2004; Bourne and Harris, 2008). Similarly, both LTP and LTD require de novo protein synthesis, including local regulation of protein translation in dendrites (Bramham and Wells, 2007). Below, we outline some of the canonical mechanisms linked to LTP and LTD.

LTP is divided into early (E-LTP) and late (L-LTP) phases which are mechanistically distinct. E-LTP typically lasts 1–2 h after LTP induction. This phase depends on the post-translational modification and trafficking of pre-existing proteins. It does not require new gene expression or protein synthesis. In brief, activated N-methyl-D-aspartate receptor (NMDAR)-type glutamate receptors trigger rapid entry of calcium into spines. Calcium influx impacts myriad signal transduction pathways, many of which are present within the postsynaptic spine itself. Initial signaling events include activation of numerous calcium-responsive protein kinases (calcium and calmodulin-dependent protein kinase II (CaMKII), extracellular signal-regulated protein kinase (ERK), and protein kinases A (PKA) and C (PKC) see Figure 1A). Activation of these pathways regulates both endosomal trafficking of AMPA receptors and modulation of actin cytoskeletal dynamics in spines, leading to enhanced postsynaptic membrane expression of AMPA-type glutamate receptors and a transient enlargement of dendritic spines (see Figure 1B; Malinow and Malenka, 2002; Bosch and Hayashi, 2012; Lisman et al., 2012).

The formation of stable L-LTP, which lasts many hours and days, requires new gene expression and protein synthesis (Stanton et al., 1984; Matthies et al., 1990; Costa-Mattioli et al., 2009; Sossin and Lacaille, 2010; Gal-Ben-Ari et al., 2012). The first period of protein synthesis occurs within the first 2 h following LTP induction. L-LTP is associated with stable enlargement and remodeling of the postsynaptic density (a large multi-protein complex attached to the membrane), enlargement of pre-existing dendritic spines, as well as de novo synapse formation (see Figure 1C; Lisman and Raghavachari, 2006; Bourne and Harris, 2008). Inhibition of protein synthesis prevents maintenance of the change in synaptic efficacy initiated during E-LTP. Additionally, protein synthesis inhibitors prevent stable increases in actin filaments (F-actin) associated with L-LTP without affecting actin cytoskeletal formation during E-LTP (Bourne et al., 2007; Bramham, 2008; Murakoshi and Yasuda, 2012).

LTD may be induced following activation of metabotropic glutamate receptors (mGluRs) and NMDARs. Unlike LTP, LTD expression relies on activation of phosphatases (i.e., PP1 and calcineurin) and is accompanied by removal of AMPARs from the postsynaptic membrane, thus lowering synaptic efficacy. Again, stabilization of the change in synaptic efficacy requires protein synthesis (but not necessarily new mRNA expression; Malenka and Bear, 2004). This is paralleled by a net decrease in spine F-actin and a shrinkage or retraction of dendritic spines (Tada and Sheng, 2006; Bosch and Hayashi, 2012).

Finally, it should be noted that LTP mechanisms probably differ between brain regions, and different input patterns can generate distinct forms of LTP (For detailed accounts see Ho et al., 2011; Panja and Bramham, 2014).

The impact of sleep loss on long-term synaptic plasticity has been investigated in recent decades. The majority of studies have employed sleep deprivation or sleep restriction to assess the benefits of sleep. Several protocols have been used to induce sleep loss.

Studies on how sleep and sleep loss affect the regulation of gene expression have yielded insights into the molecular mechanisms at play during sleep. In addition, a handful of studies have explored sleep stage-specific regulation of gene expression.

The advent of genome-wide expression profiling (i.e., blood, brain tissue) allowing the screening of thousands of transcripts has given researchers the opportunity to look at specific effects of sleep vs. sleep loss in the brain (Cirelli et al., 2004; Mackiewicz et al., 2007). Sleep specific changes in the mouse cortex involve as many as 2090 mRNAs which increase or decrease in their steady-state expression (Mackiewicz et al., 2007). Most of the genes increased during sleep are linked to macromolecular biosynthesis and transport, supporting a restorative function of sleep at the cellular level. In healthy humans, just 1 week of insufficient sleep (6 h per day) affects the expression of 711 different mRNAs in whole blood relative to subjects getting sufficient sleep (8.5 h per day; Möller-Levet et al., 2013). Genes altered by sleep restriction in this study were involved in sleep homeostasis, circadian rhythms, oxidative stress, and metabolism. Moreover, the insufficient sleep was associated with poor cognitive performance in a vigilance test. These studies support the idea that biosynthetic pathways are “recharged” during sleep for optimal function during wakefulness. However, it is also possible that metabolic changes are required to support bursts of protein synthesis or other energy-expensive processes during sleep.

Only a few IEGs have been causally linked to late LTP and long-term memory. These genes include the activity-dependent cytoskeletal-associated protein (Arc, a.k.a Arg3.1), the transcription factor zif268 (a.k.a. egr-1, krox24, Ngfi-A), and the neurotrophin brain-derived neurotrophic factor (BDNF; Guzowski et al., 2000; Jones et al., 2001; Plath et al., 2006; Messaoudi et al., 2007; Bekinschtein et al., 2008; Penke et al., 2014). Given their essential role in consolidation mechanisms, these genes could be expected to be induced at one point or another during the NREM-REM sleep cycle.

A recent global quantification of gene expression in mammalian cells concludes that the cellular abundance of proteins is predominantly determined at the level of translation (Schwanhäusser et al., 2011). Translation proceeds in three phases: initiation, elongation, and termination. Translation initiation is the process whereby the mRNA is recruited to the ribosome. The translation factor, eukaryotic initiation factor 4E (eIF4E), is required for translation of most mRNAs. eIF4E binds to the 5′ terminal m⁷GpppN cap structure on mRNA and serves to recruit the scaffolding protein, eIF4G, and other factors to form the translation initiation complex. The critical interaction between eIF4E and eIF4G is regulated by eIF4E-binding proteins (4E-BPs). In its unphosphorylated state, 4E-BP is bound to eIF4E and translation is inhibited. Phosphorylation of 4E-BP catalyzed by the mammalian target of rapamycin complex 1 (mTORC1) triggers the release of 4E-BP and enhances translation (Gingras et al., 2001; Proud, 2007). ERK signaling to mitogen-activated protein kinase-interacting kinases (MNKs) is associated with enhancement of translation though the mechanisms are not fully understood. MNKs bind directly to eIF4G and catalyze the phosphorylation of eIF4E at Ser209. Phosphorylated eIF4E has decreased affinity for mRNA binding, which, in theory, could facilitate protein synthesis by recruiting initiation complexes and therefore more ribosomes to the RNA (Buxade et al., 2008). During translation elongation the polypeptide chain is formed as the ribosome moves along the mRNA. Eukaryotic elongation factor 2 (eEF2) plays a key role in catalyzing the translocation of peptidyl-tRNAs from the A-site to the P-site on the ribosome. When phosphorylated, eEF2 does not bind the ribosome and global translation is slowed down. By mechanisms that are not fully understood, translation of certain synaptic proteins (Arc, αCaMKII) is maintained or enhanced under conditions of eEF2 phosphorylation (Scheetz et al., 2000; Chotiner et al., 2003; Soulé et al., 2006; Park et al., 2008; Gal-Ben-Ari et al., 2012). eEF2 kinase, the only known kinase for eEF2, is regulated by calcium/calmodulin, mTORC1, and ERK signaling (Proud, 2007).

Several lines of evidence implicated sleep in the regulation of the protein synthesis. Two early studies involving in vivo incorporation of radioactive leucine in the brain revealed that global rates of protein synthesis were regulated during sleep (Ramm and Smith, 1990; Nakanishi et al., 1997). Both studies concluded that rates of protein synthesis correlate positively with the amount of NREM sleep. More recently, Vazquez et al. (2008) performed a proteomics screen of spontaneous sleep-wake state dependent changes in cortical protein expression and demonstrated rapid changes on the order of minutes. Sleep is associated with upregulation of numerous genes in the rodent cortex, including genes encoding translation initiation factors (eIF4b, eIF5, eIF3 subunits 3, 8 and 12), and eEF2 (Cirelli et al., 2004; Mackiewicz et al., 2007, 2008). Not surprisingly, proteomic studies indicate that changes in protein expression patterns depend on the duration of sleep deprivation. Short periods (6 h) of sleep deprivation altered expression of 11 proteins associated with synaptic function or cytoskeletal regulation in the basal forebrain cholinergic region (Basheer et al., 2005), while 7 days of sleep deprivation was associated with enhanced cortical expression of cytochrome C, the latter possibly indicative of metabolic stress (Cirelli et al., 2009).

In addition, a small subset of transcripts involved in tRNA activation is upregulated during sleep (Mackiewicz et al., 2007). Microarray analysis of mouse hippocampal tissue obtained after 5 h of sleep deprivation identified a decrease in the expression of mRNAs associated with protein translation (Vecsey et al., 2012). Independent validation confirmed decreased expression of total and phosphorylated mTOR following sleep deprivation. This difference was absent in mice permitted to sleep for 2.5 h after the sleep deprivation (rebound sleep). This upregulation of translation-related genes has been taken in support of active protein synthesis during sleep. Alternatively, translation factor synthesis in sleep may be restorative in nature, preparing the translational machinery for waking protein synthesis.

A recent study examined the relationship between sleep quality and quantity of home cage housed rats with the activity-state (phosphorylation) of translation factors eIF4E and eEF2 (Grønli et al., 2012). In the hippocampus, no association was found between sleep and translation factor activity. In the prefrontal cortex, more NREM sleep was associated with higher eIF4E and eEF2 phosphorylation. eEF2 phosphorylation correlated positively with sleep quality (total time spent in SWS) and negatively with poor sleep quantity (number of waking episodes). Levels of phosphorylated eIF4E correlated positively with the number of SWS and REM sleep episodes. Taken together, this suggests that sleep quality (based on the amount and number SWS episodes) correlates positively with phospho-eEF2 and phospho-eIF4E levels. These changes provide only an indirect measure of (enhanced) translational activity and more work is needed to profile the impact on translation and protein expression. However, dual eIF4E/eEF2 phosphorylation is mechanistically linked to protein synthesis-dependent forms of LTP in the dentate gyrus and plasticity (ODP) in the visual cortex (Kanhema et al., 2006; Panja et al., 2009; Seibt et al., 2012; Dumoulin et al., 2013).

Following 8 h of sleep deprivation, phosphorylation of eIF4E decreased in the dentate gyrus, but not in the CA region (Grønli et al., 2012). In contrast, eEF2 phosphorylation was elevated in both hippocampal regions and the prefrontal cortex. Thus, sleep deprivation has brain region-specific effects on translation initiation and elongation activity. Surprisingly, sleep deprivation increased Arc mRNA levels in the rat prefrontal cortex without affecting Arc protein expression. This dissociation between Arc mRNA and protein expression in sleep-deprived rats might be explained by enhanced ubiquitination and proteasomal degradation of Arc protein (Soulé et al., 2012). When Arc transcription is persistently stimulated, protein degradation imposes a powerful brake on protein expression.

Interestingly, sleep quality and quantity prior to sleep deprivation predicted the effects of sleep on translational factor activity in the prefrontal cortex, but not in the hippocampal regions. Phosphorylation of eEF2 was associated with previous SWS (positive) and waking episodes (negative), while levels of phosphorylated eIF4E were associated with prior episodes of SWS and REM sleep (positive). The implication may be that a good nights’ sleep prior to sleep loss diminishes the impact of sleep deprivation on protein synthesis (Grønli et al., 2012).

Direct evidence for protein synthesis-dependent consolidation of synaptic plasticity has come from studies of ODP in the cat visual cortex (Aton et al., 2009; Seibt et al., 2012; Seibt and Frank, 2012). Sleep consolidates ODP by strengthening cortical responses to non-deprived eye stimulation (Aton et al., 2009). In a recent study (Seibt et al., 2012), cats were given monocular deprivation for 6 h in either wakefulness or sleep, combined with 6 h of intracortical infusion of the mTORC1 inhibitor, rapamycin. Vehicle-infused controls exhibited enhanced phosphorylation of 4E-BP1 and enhanced cortical expression of Arc and BDNF (and other proteins). Rapamycin blocked the sleep-related protein expression and consolidation of ODP, but did not affect plasticity induced during wakefulness. Dumoulin et al. (2013) further showed that consolidation of ODP requires ERK-MNK signaling leading to eIF4E phosphorylation (Dumoulin et al., 2013), as was found during LTP consolidation in the dentate gyrus (Panja et al., 2009). Taken together these results suggest that mTORC1 and ERK-MNK signaling are both required for sleep-dependent protein synthesis and consolidation of ODP.

In sum, gene and protein expression during sleep is likely important for changes in synaptic efficacy and consolidation of waking experience during sleep. The next section discusses recent insights into how functional and structural plasticity are regulated during sleep.

A decade ago, Tononi and Cirelli (2003) proposed a theory for a function of sleep in synaptic processes linked to cognitive functioning. Binding together a significant part of current knowledge of sleep, the synaptic homeostasis hypothesis highlights the role of sleep in the downscaling of synaptic strength after prolonged wakefulness. In this view, synaptic strengthening during wakefulness occurs via LTP-like mechanisms, while NREM SWA induces mechanisms of LTD or depotentiation throughout the cerebral cortex (Tononi and Cirelli, 2003, 2006, 2012). According to this model, SWA results in a global homeostatic downscaling of synaptic weights in which the synapses enlarged by LTP during wakefulness are reduced in size during sleep and the weakest synapses are eliminated. Such scaling may enhance signal-to-noise ratios for information encoded during waking. By preventing saturation of input strength, homeostatic downscaling may serve to retain the information encoding capacity of networks. In addition, morphological scaling of spines would offset the metabolic expense of maintaining large synapses.

Electrophysiological recording of miniature excitatory postsynaptic currents (mEPSCs; currents which reflect spontaneous release of neurotransmitters from single vesicles) of layer II/III pyramidal neurons in the frontal cortex of mice and rats demonstrates wake-related increases and sleep-related decreases in synaptic efficacy (Liu et al., 2010). The frequency and amplitude of mEPSCs was enhanced after the dark period (wakefulness) and decreased after the sleep period. Matching these electrophysiological changes, the abundance of GluA1-containing AMPARs in biochemically fractionated synaptosomes was 40% higher after wakefulness than after sleep (Vyazovskiy et al., 2008; Hinard et al., 2012). Dephosphorylation of GluA1 on Ser845 is associated with decreases in channel open probability and decreased surface expression of AMPARs. Hinard et al. (2012) show that Ser845 phosphorylation is enhanced according to time spent awake, which appears compatible with the lack of synaptic depression at the level of AMPAR regulation during wakefulness. The fact that these changes are detected in synaptosomes from whole cortex and hippocampus is consistent with global scaling at the synaptic level.

Two recent studies provided evidence for distinct roles for NREM and REM sleep in modulation of synaptic plasticity. Chauvette et al. (2012) measured local field potentials in the rat somatosensory cortex of head-restrained cats during wake, comparing responses obtained before (wake 1) and immediately after (wake 2) a period of NREM. The responses were enhanced in wake 2, and longer periods of NREM were associated with larger evoked responses. A large transient increase in the response was observed in wake 2 but not after additional periods of NREM or REM sleep. The fact that delta power was increased in wake 2 compared to wake 1 is indicative of sleep inertia (“sleepiness”). Hence, it is possible that state-dependent modulation of synaptic efficacy contributes to the transient enhancement of the response (Winson and Abzug, 1978; Bramham and Srebro, 1989). However, a smaller stable increase in the evoked response was present in wakefulness after several sleep cycles, and this stable increase was mimicked by in vitro stimulation and intracellular hyperpolarization designed to mimic cortical slow wave “downstates” of NREM sleep (Chauvette et al., 2012). In sum, the work suggests that rapid upscaling (potentiation of the evoked responses) can occur in a cortical network during NREM sleep.

Grosmark et al. (2012) reported a prominent role of REM sleep in sleep-related neuronal plasticity. They show that overall firing rates of hippocampal pyramidal cells and interneurons increase moderately during NREM sleep periods, but decrease more during REM sleep, giving an overall net decrease in global firing from the neuronal population across a sleep cycle. Of major significance is the observed difference in pyramidal neuron firing during and between the intermittently occurring phasic events of NREM sleep known as sharp wave-ripples (SWRs). SWRs are irregular, synchronized bursts of neuronal activity in the hippocampus which are synchronized with thalamo-cortical spindle activity (Buzsáki et al., 2013). Grosmark et al. (2012) observed overall firing decreases in the periods between SWRs, but the synchrony and mean firing rate during ripple events increased across sleep in correlation with the power of the REM sleep theta rhythm. These findings are compatible with global downscaling of neuronal firing between SWRs and upscaling during SWRs. It is currently unclear whether these changes in neuronal firing and synchrony are mediated by LTP/LTD-type events on hippocampal pyramidal and interneurons. For further discussion see Born and Feld (2012), Tononi and Cirelli (2012), Cirelli (2013), Frank (2013), and Rasch and Born (2013).

At the anatomical level, two-photon microscopy has been used to visualize changes in dendritic spines of cortical neurons during sleep. Maret et al. (2011) showed that wakefulness is associated with a net increase in dendritic spines while sleep is associated with net spine loss. Yang and Gan (2012) ascribed the loss of spines during sleep to higher rates of turnover. However, it is not known whether spines size and density varies in a sleep-stage specific manner as shown for synaptic field potentials and neuronal firing activity.

The evidence reviewed suggests that sleep-stage specific changes in synaptic efficacy and plasticity, firing activity, and network synchrony develop over the course of sleep. However, no consensus exists on how synaptic efficacy is regulated across the sleep cycle. In Figure 1, we offer a scenario for how synaptic plasticity is regulated during sleep. The working hypothesis is an attempt to integrate the cell biology of synaptic plasticity with the electrophysiological data.

We propose that specific cell biological events underlying homeostatic scaling and synaptic potentiation are parsed to specific stages of sleep and stage-specific population events. Downscaling during NREM sleep is supported by electrophysiological, biochemical and morphological data (Figure 1D; Grosmark et al., 2012; Cirelli, 2013). During REM sleep, immediate early genes such as Arc and zif268 are triggered by PGO-waves (Figure 1E; Ribeiro et al., 2002; Ulloor and Datta, 2005). Pharmacological cholinergic activity mimicking phasic REM sleep epoch also drives Arc expression in glutamatergic neurons (Soulé et al., 2012). Hence one function of REM sleep in this model is to provide immediate early gene induction in a broad population of cortical and hippocampal project neurons. A large body of work suggests that neuronal ensemble activity representing recent learning during the wake state is replayed in a time-compressed format during the SWRs of NREM sleep (Lee and Wilson, 2002; Skaggs et al., 2007; O’Neill et al., 2010). Grosmark et al. (2012) showed that firing synchrony during SWRs develops gradually over successive NREM-REM sleep cycles. It follows that sparse, but synchronous synaptic firing, is repetitively replayed during NREM SWRs (Figure 1F). Thus, in NREM sleep an interplay may exist between synaptic potentiation and homeostatic scaling, with synaptic potentiation occurring during SWRs and scaling during the inter-ripple periods. Restorative macromolecular synthesis of the translational machinery (ribosomal proteins, translation factors, tRNA) occurs during sleep and may function to support bursts of synaptic protein synthesis.

As a multifunctional dendritically translated protein, Arc could play a role in coordinating diverse forms of plasticity during sleep. In REM sleep, Arc mRNA would be synthesized and transported to dendritic processes (Figure 1E). During NREM sleep, the synaptic activity of SWR events is proposed to drive local translation of Arc and other dendritically localized mRNAs. Repetitive bursts of translation during the night would ensure synapse-specific, protein synthesis-dependent potentiation (Figure 1F). Extrapolating from LTP studies, local Arc synthesis would consolidate synaptic potentiation through regulation of actin cytoskeletal dynamics and enlargement of dendritic spines (Fukazawa et al., 2003; Messaoudi et al., 2007). The extremely rapid rates of Arc mRNA and protein degradation are well-suited for mediating bursts of protein expression during SWRs. Arc mRNA is subject to rapid translation-dependent decay (perhaps limiting synthesis to translation by a single ribosome), while Arc protein is rapidly ubiquitinated and targeted for degradation in the proteasome (Rao et al., 2006; Giorgi et al., 2007; Soulé et al., 2012). In the same neurons, Arc protein could function to mediate homeostatic scaling and LTD. Dendrite-wide downscaling might be achieved through nuclear import of Arc leading to downregulation of GluA1 transcription (Korb et al., 2013), or through selective targeting of Arc to inactive or weakly activated synapses resulting in Arc-dependent endocytosis of AMPARs (Shepherd et al., 2006; Beique et al., 2010; Okuno et al., 2012). Clearly, it will be important to elucidate the time-dependent functions of Arc, and the possible role of post-translational modifications of the protein in dictating its localization and function (Bramham et al., 2010; Craig et al., 2012).

As summarized in the above sections, sleep loss can actively affect synaptic plasticity, synaptic efficacy and cognitive functioning. Such outcomes are sensitive to the various protocols used to induce sleep loss. Enforcing wakefulness when the brain is programmed to sleep may induce effects unrelated to sleep loss per se. Sleep restriction or sleep loss are often associated with an increase (but temporary) in the activity of the neuroendocrine stress systems by altering the state or function of the hypothalamo-pituitary-adrenal (HPA) axis. Most of the studies investigating the impacts of sleep loss on synaptic plasticity are performed in brain regions sensitive to stress. Among them is the hippocampus, which is involved in the negative feedback response to stress and helps to determine whether stress is ongoing. Since sleep deprivation can be stressful, it is important that studies aim to control for such non-specific effects. Several studies discussed in this review controlled for hormonal stress response following sleep deprivation (Palchykova et al., 2006; Hagewoud et al., 2010; Süer et al., 2011; Grønli et al., 2012), signifying that sleep loss rather than stress perturbs the changes in synaptic plasticity. From this emerges the question of how stress impacts sleep and synaptic plasticity, to which we now turn our attention.

Acute stress like social defeat, tail suspension, restraint, forced swim, or foot shock are developed as tools to mimic an immediate threat resulting in despair-like behavior. The behavioral effects (e.g., sleep changes) are often transient, typically gone within 1–3 days after termination of the stressor (Meerlo et al., 1997; Kinn et al., 2008).

Exposure to social defeat induces changes in NREM sleep but leaves REM sleep unaffected. An immediate increase in deep NREM sleep which dissipates during the following 12 h has been reported (Meerlo et al., 1997). Such an increase re-occurs 4 days after defeat and dissipates again within 14 days after social defeat (Kinn et al., 2008; Kinn Rød et al., 2014, in press). Inescapable foot-shock has been shown to increase wakefulness and then decrease REM sleep (Sanford et al., 2010; O’Malley et al., 2013). Additionally, Philbert et al. (2011) report long-lasting increases in sleep fragmentation (21 days after the stress exposure; Philbert et al., 2011). Similarly, wakefulness is also reported to increase after 1–2 h of restraint or forced swimming, while REM sleep increases during the sleep rebound (Cespuglio et al., 1995; Dewasmes et al., 2004). Hence, both NREM and REM sleep are differently affected by the nature of the stressor.

To the best of our knowledge, no study has directly focused on the acute stress–sleep interaction at the synaptic level. Studies of stress alone show that restraint suppresses maintenance of hippocampal LTP, enhances LTD in vitro and in vivo, and impairs cognitive functions like spatial memory (Foy et al., 1987; Bodnoff et al., 1995; Kim et al., 1996; Xu et al., 1997; Conrad et al., 2004; Krugers et al., 2006; Chen et al., 2010). Foot shock also facilitates LTD induction and slightly impairs learning of a spatial task directly after stress exposure, but enhances memory retrieval 5 days later (Xiong et al., 2003). Despite the transient sleep changes reported to occur after social defeat, this stressor is shown to produce long-lasting effects on hippocampal LTP and LTD. Artola et al. (2006) showed that the threshold for LTP induction is still raised and that for LTD lowered 7–9 months after defeat and individual housing (Artola et al., 2006).

Following a short stressful experience, de novo gene expression and protein synthesis, which are crucial to long-term synaptic changes, are rapidly but transiently altered. Activity of the ERK pathway and its downstream targets (including zif268) are increased after restraint and forced swimming (Gutièrrez-Mecinas et al., 2011), and ERK activation mediates the effects of restraint tail shock stress on hippocampal LTP (Yang et al., 2004). Other IEGs necessary for the stabilization of activity-dependent synaptic plasticity are also affected by acute stress. Arc and BDNF are among those and their expression varies in a region-specific manner. Cortical upregulation of both Arc mRNA and protein expression is indeed detected after restraint (Mikkelsen and Larsen, 2006). Restraint also induces a rapid, transient modification of BDNF expression across several brain regions. Importantly, the impact of stress depends on the individual’s age. Defeat during adolescence and adulthood differentially regulates expression of several plasticity-related IEGs. A recent study shows that mRNA levels for Arc and BDNF (among others) are elevated following social defeat in adolescence, but not in adulthood (Coppens et al., 2011).

Compiled evidence suggests that acute stressful events have the capacity to induce sleep disturbances and alter long-term synaptic plasticity. Unfortunately, there is no data available on the effects of stress-sleep interactions on LTP, LTD, de novo gene expression or protein synthesis after acute stress.

When stress becomes chronic the physiological changes are more profound and long-lasting. Importantly, the changes vary accordingly with the intensity, frequency, and particularly the unpredictability of the stressor. The unpredictability is of importance to overcome stress habituation that occurs if the stressors are given repeatedly in a controllable manner.

The various protocols of repeated stress have been shown to affect sleep differently. Animals remain awake throughout a 6 h recording period after exposure to 2 days of inescapable foot shock; in contrast, 3–5 days exposure decreases wakefulness and REM sleep (Papale et al., 2005; O’Malley et al., 2013). Four days of forced swimming or restraint both decrease NREM sleep, and restraint additionally decreases REM sleep compared to baseline (Papale et al., 2005).

Repeated exposure to stressors may constitute an environmental risk factor for the development of anxiety and depression. When restraint is given repeatedly, 2 h for 10 days in rats, REM sleep is altered for at least 21 days after termination of the stressor (Hegde et al., 2011). Importantly, the impact of restraint stress on REM sleep was bimodally distributed. One group of rats manifested an increase in REM sleep and anxiety-like behavior, while the other group showed reduced REM sleep and no anxiety-like behavior (Hegde et al., 2011). One animal model that was developed to mimic minor daily hassles is chronic mild stress (Willner, 2005). Various mild stressors unpredictably given for 4 weeks decrease deep NREM sleep and increase time in REM sleep and wakefulness (Cheeta et al., 1997; Grønli et al., 2004, 2012). These sleep changes parallel those found in human depression. Although sleep alterations are one of the hallmark symptoms of depression and anxiety, there is limited research in rodents on the role of sleep in stress related depression and anxiety. Further research on sleep, depression- and anxiety related behaviors is an interesting direction for future investigation.

Hippocampal LTP maintenance is suppressed, and LTD enhanced in vitro and in vivo after acute and chronic restraint. This is also found when restraint is given in combination with tail shock as well as chronic corticosterone or chronic stress exposure (Foy et al., 1987; Bodnoff et al., 1995; Kim et al., 1996; Xu et al., 1997; Krugers et al., 2006; Chen et al., 2010). Brief exposure to mild stress also affects synaptic plasticity, as induction of LTP is blocked and LTD is facilitated (Xu et al., 1997). Moreover, this facilitation of LTD is abolished by acclimatization to, or removal from the mild stressors (Xu et al., 1997).

Chronic stress, in general, is associated with changes at the transcriptional and translational levels. At the transcriptional level, chronic stress has been shown to both increase (defeat and novel cage; Pardon et al., 2005) and impair CREB activity (glucocorticoid treatment; Föcking et al., 2003, chronic mild stress; Grønli et al., 2006). These contrasting results may relate to differences in glucocorticoid concentration release due to different intensity or chronicity of the stressor. At the translational level, chronic stress enhances phosphorylation of the translational regulators eIF4E and eEF2 in prefrontal cortex but not in the hippocampus or dentate gyrus (Grønli et al., 2012). This upregulation of translational activity may be taken as evidence in support for active protein synthesis after stress in the cortical areas.

Recently, sleep-stress interaction was examined at the translational level using the chronic mild stress model. Being chronically stressed abolishes associations between an individuals’ sleep quality/quantity and translational activity. The sleep parameters are no longer predictive for cortical activity of initiation factor eIF4E and elongation factor eEF2 (Grønli et al., 2012). Similarly, chronic stress abolishes associations between sleep parameters measured prior to 8 h sleep deprivation and cortical translational activity as assessed after sleep deprivation. Given that persons experiencing chronic stress and depressed patients complain of non-restorative sleep, it is tempting to speculate that such lack of association between sleep quality and optimal rates of protein synthesis may be one of the underlying causes. Moreover, the effect of 8 h of sleep loss is modulated after chronic stress. In stressed rats, decreased activity of cortical eEF2 was found, whereas increased eEF2 activity occurs in non-stressed animals. No change of these translational regulators was observed in hippocampus. This may suggest that sleep deprivation counteracts the effect of chronic stress on eEF2 activity, in a region-specific manner. Interestingly, acute sleep deprivation has been reported to have antidepressant effects in humans (Wu and Bunney, 1990). The findings from animal studies raise the possibility that sleep deprivation may serve to restore or optimize rates of cortical protein synthesis in depressed patients.

Recent work shows that circadian changes in glucocorticoids are necessary for the formation and stabilization of dendritic spines in cortex after motor learning, and chronic and excessive exposure to glucocorticoids destabilizes learning-associated spines and impairs memory retention (Liston et al., 2013).

The brain is in constant change across the lifespan, starting from the early stages of life in utero. Early life (pre- and postnatal, as well as childhood and adolescence) hosts important developmental phases which allows the brain to mature. Being exposed to early life stress such as prenatal stress, maternal separation, low maternal care, or stress during adolescence has consistently been found to alter stress sensitivity in adulthood (Lupien et al., 2009).

Mammals show large amounts of active sleep (that parallels adult REM sleep) during early postnatal brain development. The predominance of REM sleep during early life is often taken in support of a role for REM sleep in processes of brain maturation and plasticity (Frank, 2011). The studies on sleep-related changes are scarce and the findings are divergent. Exposure of stress in utero may result in a prolonged first REM sleep episode and less NREM sleep in adulthood, compared to non-stressed controls (Rao et al., 1999). Long maternal separation (typically 3 h per day in the first 2 postnatal weeks) is reported to diminish the quality of deep NREM sleep, to alter total sleep time (decrease or increase), and to increase wakefulness compared to non-handled, handled, and brief maternally separated offspring. Moreover, the negative feedback regulation of the HPA axis in long maternal-separated offspring is suggested to be impaired and corticosterone level is elevated in long compared to brief maternal-separated offspring (Mrdalj et al., 2013).

Altered stress sensitivity in adulthood is also reflected in sleep changes. Exposure to later life stressor(s) affects sleep differently according to early life experience. Adult exposure to acute stress (2 h of cold) is followed by decreased REM sleep and elevated corticosterone levels, both in long maternal-separated offspring and handled controls (Tiba et al., 2004). Adult experience of chronic unpredictable mild stressors induces more time in sleep, more REM sleep episodes and more NREM sleep episodes ending in REM sleep in long, compared to brief, maternally separated offspring (Mrdalj et al., 2013). REM sleep deprivation in adult long maternal-separated offspring seems not to potentiate a present memory deficit (Garcia et al., 2013).

Independently of wakefulness, NREM or REM sleep, early life stress reduces brain activity measured by EEG, an effect potentiated by exposure to chronic stress as adults (Mrdalj et al., 2013). Offspring that receive low maternal care show poor LTP when they are adult, as opposed to those that were given high maternal care (Champagne et al., 2008). Single (short and prolonged), or repeated maternal separation can affect LTP expression in hippocampus and prefrontal cortex (Cao et al., 2013), without any change in the number of neurons and astrocytes (Baudin et al., 2012).

Activation of synaptic plasticity-related genes is assumed to represent an early step in the adaptation of neuronal networks to a stressful environment. Maternal separation after the first 2 postnatal weeks, at day 14–16, induces rapid increase in hippocampal Arc and zif268 mRNAs, accompanied by morphological changes such as an increase in spine number on CA3 dendrites (Xie et al., 2013). In rats exposed to isolation rearing, cortical upregulation of Arc mRNA and increase in both Arc and BDNF proteins is observed (Wall et al., 2012). Note that single (short and prolonged), or repeated maternal separation alter hippocampal BDNF expression (Roceri et al., 2002, 2004; Koo et al., 2003; Fumagalli et al., 2004; Nair et al., 2007). Notably, the prior history of maternal separation impacts the effect of adult stress on BDNF transcripts via modulation the upstream transcriptional activator CREB. An impact on neuronal progenitor proliferation is also reported, suggesting that alterations in CREB/BDNF may contribute to individual differences in hippocampal networks (Nair et al., 2007). In the cortex, the length of early life manipulations appears to be more important for these changes than their timing. Repeated early life stress induces a clear reduction of cortical BDNF levels in adult animals (Koo et al., 2003; Fumagalli et al., 2004; Roceri et al., 2004), whereas a single maternal deprivation does change BDNF expression (Roceri et al., 2002).

Stressful events early in life induce long-term sleep disturbances and alter long-term synaptic plasticity. Unfortunately, as for acute stress, there is no available data regarding the effects of stress-sleep interactions on LTP, LTD, de novo gene expression or protein synthesis after early life stress.

The findings on sleep changes after stress discussed above, raise an important issue that different stress modalities result in distinct sleep responses. Moreover, stress responses are mediated through the concerted activity of many brain areas and induce structural changes in neuronal networks. Changes can be short or long-lasting (Fuchs et al., 2006).

Brain areas involved in the stress response include areas important for sleep and wakefulness; the hypothalamus (including deep NREM sleep active neurons in the ventral lateral preoptic area), amygdala (activity is depotentiated during REM sleep), hippocampus (generating theta activity in REM sleep), prefrontal cortex (generating the highest voltage and the slowest NREM sleep waves compared to other cortical regions) and numerous brainstem regions promoting wakefulness like the locus coeruleus and raphe. Stress-induced changes in the activity of one or several of these brain regions may explain the different sleep changes.

The recovery from sleep loss is sensitive to stress. Likewise, recovery from stress is sensitive to sleep disturbances. If the individual has been exposed to stress prior to the sleep loss, the sleep recovery may be altered. Little data is available on how sleep recovery is affected by stress experience prior to sleep loss. One study has shown that exposure of rats to social defeat prior to 6 h of sleep deprivation potentiated changes in the recovery sleep by an increase in deep NREM sleep (Meerlo et al., 2001).

The available data on sleep disturbances and drive for sleep after acute stress suggests that stress accelerates the buildup of sleep need. NREM and REM sleep are differently affected by the nature of the stressor. Restraint increases REM sleep while social defeat increases deep NREM sleep. Cognitive functioning is considered to be potentiated and LTP-like changes facilitated after transient stress (Luine et al., 1996; Shors, 2001). However, more studies are needed to define the selective role of NREM and REM sleep rebound after stress. Moreover, knowledge on how prior stress may impair the sleep recovery after sleep loss is limited.

The recovery of stress may result in a long-lasting disruption of normal circadian sleep pattern by decreased or increased sleep throughout the 24 h period. Again, changes in specific sleep stage depend on the type of stressor. In the rats’ active phase restraint decreases sleep efficiency, NREM and REM sleep, whereas foot shock, swimming, cold as well as chronic controllable stress increase REM sleep 4 days after the stress exposure (Kant et al., 1995; Papale et al., 2005).

Behavioral factors appear to be important for the understanding of the variations in sleep changes brought by stress. During a stress situation, the coping strategy may play a significant role. Animals fighting back during a social conflict before being defeated show fragmented NREM sleep, an effect becoming more robust in the long-term (day 21 post defeat) compared to animals showing quick submission and passivity (Kinn Rød et al., 2014, in press). Importantly, increases in REM sleep have been observed if the organism controls the stressor (e.g., escapable foot shock; Kant et al., 1995; Sanford et al., 2010) and LTP is impaired following inescapable, but not escapable, shock in a shuttle box avoidance task (Shors et al., 1989).

In summary, when an individual is subjected to environmental stressors in any phase of life, sleep is affected. Sleep disturbances after stress are modulated by several factors among which are the brain areas activated by stress, the ability to recover from stress, the behavioral coping strategy and ability to control the stressor.