Findings in vivo and in vitro indicate that the plasticity induction process triggers short-lived changes that mainly depend on post-translational modifications of existing proteins and rapid morphological alterations of dendritic spines (Lang et al., 2004; Hernandez and Abel, 2008; Bailey et al., 2015). At the molecular level, many forms of plasticity induction rely on N-methyl-D-aspartate receptors (NMDARs) activation leading to intracellular calcium (Ca²⁺) influx and rapid activation of various protein kinases [e.g., calcium/calmodulin dependent kinases (CaMKs), protein kinases A (PKA), and C (PKC), extracellular signal-regulated protein kinase (ERK)] and phosphatases [e.g., protein phosphatase 1 (PP1), calcineurin] (Ho et al., 2011). Enzymatic activation triggers two main modifications at synapses: changes in α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) number and activity (Hayashi et al., 2000; Esteban et al., 2003; Yang et al., 2008) and morphological changes of spines. Experience and long-term potentiation (LTP) protocols are associated with rapid (within 1–2 h) changes in spine volume and number. These changes are transient and only a fraction of newly formed spines will persist over time (Xu et al., 2009; Yang et al., 2009). This is consistent with the idea that stabilization of structural plasticity induced by experience can take several hours to weeks, and involves both strengthening and weakening of selected synapses (Hofer et al., 2009; Xu et al., 2009; Yang et al., 2009; Sanders et al., 2012; El-Boustani et al., 2018). This initial modification of spine structure is thought to rely mainly on phosphorylation-dependent changes in actin cytoskeleton dynamics rather than new protein synthesis (Lang et al., 2004; Bhatt et al., 2009; Okamoto et al., 2009; Caroni et al., 2014; Bailey et al., 2015).

In addition to these protein synthesis-independent synaptic changes, experience also rapidly induces new gene expression which is the first step toward consolidation of plastic changes in neurons (Flavell and Greenberg, 2008). In particular, gene transcription activation occurs within minutes in specific neuronal ensembles activated by experience (Minatohara et al., 2016; Lisman et al., 2018). The two main signatures of experience-dependent transcription are the phosphorylation of cAMP response element binding protein (CREB) and the transient expression of Immediate Early Genes (IEGs) (Flavell and Greenberg, 2008; Caroni et al., 2014; Minatohara et al., 2016). Collectively, those data suggest that at the molecular and structural levels, early forms of plasticity are initially fragile, require covalent change of pre-existing proteins and rapid IEG transcription, but are largely independent of translation.

Studies in animals and humans show that many forms of memory are labile and susceptible to interference until sleep occurs, after which they are resistant to further disruption (reviewed in Aton et al., 2009b). A very recent study by Wang et al. (2018) demonstrated that, compared to sleep, wakefulness globally increases phosphorylation in the brain. Interestingly, among the proteins that showed increased phosphorylation, most of them were synaptic proteins associated with pre-synaptic short-term plasticity (e.g., Syn1, Rims1), post-synaptic density structure (e.g., scaffolding proteins, actin remodeling), glutamate receptors (e.g., Grin2b/5), or inhibition of translation (e.g., AMPK) (Wang et al., 2018). In vitro measures show that extended wakefulness (i.e., sleep deprivation, SD) impairs hippocampal LTP dependent on cAMP/PKA/CREB signaling, but not LTP that requires new translation (Vecsey et al., 2009). In the developing visual cortex, plasticity in vivo induced during waking is independent of translation, as inhibition of kinases critical for translation activation (i.e., mTOR, ERK, MNK) has no effect on waking plasticity (Seibt et al., 2012; Dumoulin et al., 2015). Finally, the stabilization of learning-induced spine formation and pruning in the rodent cortex does not occur during wakefulness, and is instead only detectable after sleep (Yang et al., 2014; Li et al., 2017). Therefore, plasticity during wakefulness is similar to ‘early’ forms of plasticity in vitro that are also labile and translation independent. In the next section, we will describe how these early molecular changes outline physiological processes involved in cell-wide and synapses-specific priming in neurons and circuits activated by experience.

If wakefulness and sleep are partners in synaptic plasticity, how does the induction of plasticity during wakefulness lead to secondary modifications during sleep? The answer may lie in a phenomenon known as ‘metaplasticity’ (Hulme et al., 2013; Yee et al., 2017). Metaplasticity refers to neural changes that influence the capacity for subsequent synaptic plasticity (“plasticity of synaptic plasticity” (Abraham and Bear, 1996)). Among the various forms of metaplasticity, synaptic tagging (Frey and Frey, 2008; Redondo and Morris, 2011) and changes in neuronal excitability (Benito and Barco, 2010) provide two types of mechanisms that mark or prime synapses and neurons engaged in the learned experience, respectively. Importantly, neuronal circuits are primed according to their activation history (i.e., weak or strong) and thus allow delayed consolidation while maintaining cell- and input-specificity, as well as bidirectionality of plastic changes. Although current metaplasticity mechanisms do not consider sleep and wakefulness and have been mainly integrated into models of memory allocation (Rogerson et al., 2014; Lisman et al., 2018), it is easy to see how a similar process could provide a molecular bridge between plastic changes in one brain state and those in another (Figure 1).

In vivo electrophysiological recordings and dendritic spine imaging in rodents have provided evidence for bi-directional synaptic plasticity changes during sleep (reviewed in Frank and Cantera, 2014; Puentes-Mestril and Aton, 2017; Navarro-Lobato and Genzel, 2018). Recordings of neuronal spiking, an indirect measure of plasticity, shows both increased and decreased firing rates in the hippocampus and cortex during sleep. More specifically, while NREM sleep is accompanied by more heterogeneous and bi-directional changes in spiking rates (Chauvette et al., 2012; Grosmark et al., 2012; Miyawaki and Diba, 2016; Watson et al., 2016; Norimoto et al., 2018), firing rates tend to decrease across individual REM sleep episodes (Grosmark et al., 2012; Miyawaki and Diba, 2016; Watson et al., 2016). However, increased firing rate following exposure to novel perceptual experience has been shown to occur during both sleep stages in sensory cortical areas (Ribeiro et al., 2007; Aton et al., 2014; Clawson et al., 2018).

When measured after sleep, circuits can be strengthened or weakened (reviewed in Frank and Cantera, 2014). For example, sleep-dependent changes in ODP induced by monocular deprivation (as measured by evoked electrophysiological responses) involve a gain of response to the open eye and a weakening of response to the deprived eye (Aton et al., 2009a; Seibt et al., 2012). Although decreases in evoked electrophysiological responses to stimuli are reported after sleep in some rodent studies (Vyazovskiy et al., 2008), this is not always found. For example, in the cat, electrically evoked cortical responses measured during wake are larger after a period of sleep (Chauvette et al., 2012). Electrophysiological recordings also show that sleep is required for an experience-dependent form of LTP (Cooke and Bear, 2010, 2012) in the adult visual cortex in vivo (Aton et al., 2014; Durkin and Aton, 2016). Some of the heterogeneity in firing rates after sleep may also be explained by the class of neurons recorded (i.e., inhibitory vs. excitatory) and the basal level of firing rate (i.e., low vs. high) (Levenstein et al., 2017; Clawson et al., 2018).

Recent in vivo imaging studies, which provide more direct measurements of synaptic plasticity, also show complex changes in spine structure during sleep and suggest a role for sleep in stabilization of both spinogenesis and spine pruning (reviewed in Sigl-Glöckner and Seibt, 2018). Sleep during adolescence, when spine turn-over and pruning are high, seems to favor spine pruning in some cortical areas (Maret et al., 2011; Yang and Gan, 2012). Spine pruning during development is central to the refinement of synaptic connections in response to experience (Zuo et al., 2005; Yu et al., 2013). During adulthood, when spines are much more stable, periods of sleep following experience have been shown to be important for both synaptic strengthening/spinogenesis and synaptic weakening/spine pruning (Yang et al., 2014; Li et al., 2017), consistent with long-term structural modifications at synapses. This effect seems to mainly affect recently formed spines, leaving more stable ones relatively unaffected by sleep (Yang et al., 2014; Li et al., 2017). Furthermore, data in the developing rodents further suggest that sleep may favor the pruning of smaller spines in the cortex (Vivo et al., 2017). A similar trend has been recently described in vitro in the hippocampus in immature mice (Norimoto et al., 2018). While speculative, it is possible that the large and small spines represent the positive and negative tagging status of synapses, respectively. In the context of our model, this would further suggest that sleep might be particularly important for the final stabilization of these bi-directional transient changes in spine morphology induced during plasticity induction.

Specific cellular changes during NREM sleep may also be predictive of plastic change during REM sleep. For example, increased power in specific NREM oscillations associated with plasticity (i.e., spindles) correlate positively with changes in Ca²⁺ activity in cortical dendrites (Seibt et al., 2017) and CaMKII phosphorylation in hippocampal neurons (Blanco et al., 2015) during REM sleep. These results suggest that experience may affect neurons and circuits differently during both sleep stages but that these changes are most likely complementary. This is also in agreement with existing two-stage models emphasizing an important role for NREM-REM sleep in memory consolidation (see the section “Comparisons and Contrasts With Other Theories” for discussion). In the following sections, we discuss how different sleep stages may cooperate for the capture and translation of PRPs on a background of cell-specific increase in intrinsic excitability (Figure 3).

According to current ideas about the STC, the first and necessary step toward stabilization of functional and structural plasticity is the capture of PRPs during the lifetime of the tag (minutes to hours) (Redondo and Morris, 2011). We propose that periodic reactivation of task-specific circuits during NREM oscillations, combined with priming during wake, provides an efficient mechanism for PRPs capture in a neuron- and synapse-specific manner. Localized changes in excitability can determine which dendritic segment or group of synapses are more likely to engage in synchronization patterns within a given circuit (Alarcon et al., 2006). Since reactivation affects both positively and negatively tagged synapses, this would also explain how sleep leads to bi-directional changes in synaptic plasticity. In our view, what matters is the presence of these oscillations, rather than their form which reflects differences in the underlying circuitry. NREM oscillations in general can promote synaptic capture of PRPs because they all involve a reactivation of a network and they increase in number or intensity in a use-dependent fashion.

This idea is supported by several observations. Three main NREM sleep oscillations have been implicated in experience-dependent plasticity and memory consolidation: slow-wave activity (SWA; 0.5–4 Hz) and spindles (sigma: 10–16 Hz) in the cortex, and sharp wave ripples (SPW-R; 100–200 Hz) in the hippocampus (Rasch and Born, 2013; Buzsáki, 2015; Ulrich, 2016). Although these oscillations are generated by different neuronal populations within the thalamocortical (i.e., spindles and SWA) and hippocampal (i.e., SPW-R) circuits, they all share common physiological and functional features. SWA, spindles and SPW-R have been shown to be temporally coupled across brain regions (Siapas and Wilson, 1998) and are all increased (either in number or intensity) following learning and experience (Rasch and Born, 2013; Ulrich, 2016; Norimoto et al., 2018). These use-dependent changes in NREM oscillations have been clearly demonstrated for cortical SWA and spindles in humans and animals (Vyazovskiy et al., 2000; Huber et al., 2004, 2006, 2007a; Eschenko et al., 2006; Mölle et al., 2009; Johnson et al., 2012; Cox et al., 2014). Some of these oscillations have also been shown to include coordinated reactivation of neurons and synapses previously engaged in learning tasks (Bergmann et al., 2012; Chen et al., 2013; Sadowski et al., 2016). The best evidence of this reactivation (or “replay”) occurs in hippocampal cells during SPW-R (Buzsáki, 2015), but NREM replay is also reported in cortical ensembles (Euston et al., 2007; Ji and Wilson, 2007).

Waking experience has also been shown to increase synchronization and coherence in the activity of task-related neurons in the hippocampus and cortex during sleep (Gulati et al., 2014; Durkin et al., 2017; Ognjanovski et al., 2017). This increase in NREM network synchronization also correlates with plasticity measures and memory performance (Gulati et al., 2014; Ognjanovski et al., 2018). There is also evidence that the latter events are linked to neuronal priming and necessary for PRPs capture. Constitutive CREB activation (which increases excitability), via CAMKIV overexpression in mice increases NREM SWA (Steenland et al., 2010).

The direction of plastic change during sleep is likely influenced by the neurons’ recent waking activity history, which in turn determines the setting of different tags. For example, SPW-R during sleep appears to induce synaptic weakening in vivo only in neurons that were not engaged in a previous novel experience (Norimoto et al., 2018). Similarly, only neurons that are highly responsive during a motor task will show increased SWA spiking coherence during subsequent sleep (Gulati et al., 2014). In addition, while spindle activity has been linked to increased plasticity in vitro (Rosanova and Ulrich, 2005) and in vivo (Timofeev et al., 2002), more recent data also show that spindles predict global decreases in firing rate across sleep stages (Miyawaki and Diba, 2016). Consistent with a previous hypothesis (Contreras et al., 1997; Steriade and Timofeev, 2003), Seibt et al. (2017) recently demonstrated using in vivo Ca²⁺ imaging that spindles might be particularly important for synchronized reactivation of dendritic populations.

Given its important role as a sensor of synaptic activation, different levels of intracellular Ca²⁺ in different groups of synapses during NREM oscillations may mediate PRPs capture weakening or strengthening. Using two-photon imaging in anesthetized mice, Chen et al. (2013) demonstrated that spines that displayed increased Ca²⁺ during UP states of slow oscillations represented the same population that showed Ca²⁺ increased during previous sensory evoked responses. Although performed in anesthetized mice, this supports the idea that synaptic tags are set during waking and captured during subsequent NREM sleep. Consistent with bi-directional plasticity changes during NREM sleep, it was recently shown that individual cortical dendrites display both increased and decreased Ca²⁺ activity during spindles (Seibt et al., 2017). This further suggests that local changes in Ca²⁺ levels during spindles may trigger intracellular mechanisms that favor both potentiation and depression in a dendrite-specific manner.

Recent studies using two-photon imaging of single dendritic spines in mice provide direct evidence that sleep stabilizes experience-dependent structural plasticity during development and adulthood (Yang et al., 2014; Li et al., 2017). In both studies, lack of sleep disrupted both motor learning and structural plasticity stabilization. As mentioned before, an important step for the stabilization of labile into more permanent synaptic changes is new protein synthesis. More specifically, localized mRNA translation near/at synapses is a key mechanism neurons use to promote synapse-specific structural modifications during development and adulthood (Holt and Schuman, 2013). Our model proposes that, while experience-dependent PRPs transcription occurs predominantly during wake, the final translation of PRPs transcripts necessary for synapse-specific structural plasticity stabilization occurs during REM sleep (Figure 3B).

Translation activation in the brain is specifically enhanced during sleep (Table 2) and is accompanied by increased expression of PRPs proteins (Figure 2). Among the multiple pathways involved in translation regulation for plasticity and memory, one important step is activation of the eukaryote initiation factor 4E (eIF4E) resulting in translation initiation activation (Richter and Sonenberg, 2005). Two main pathways can lead to eIF4E activation: the mammalian target of rapamycin (mTOR)/4E-BPs and ERK/Mnk1 pathways (Banko and Klann, 2008). In addition to showing increased cortical and hippocampal activation during periods dominated by sleep (Eckel-Mahan et al., 2008; Seibt et al., 2012; Vecsey et al., 2012; Dumoulin et al., 2015; Luo et al., 2013; Tudor et al., 2016; Diering et al., 2017), inhibition of these pathways in the visual cortex of cats during sleep impairs the consolidation of plastic changes during ODP (Seibt et al., 2012; Dumoulin et al., 2015). A similar inhibition during waking had no effect on the labile changes triggered by monocular deprivation (Seibt et al., 2012; Dumoulin et al., 2015). Furthermore, inhibition of mTOR signaling during sleep prevents the enhancement of synaptic potentiation and reverses the depression in visual circuits normally seen during ODP (Seibt et al., 2012). This is consistent with a role for translation in stabilization of bi-directional structural plastic changes. Finally, memory impairment induced by sleep deprivation can be rescued by enhancing 4E-BP2 activity in the hippocampus of sleep deprived mice (Tudor et al., 2016).

These results in mice and cats suggest that the increase in translation initiation during sleep is particularly critical for brain plasticity and memory consolidation. But are PRPs actually translated during sleep? In contrast to their transcripts, protein expression of several PRPs increase or remain at high levels during sleep in the hippocampus and cortex [i.e., BDNF, Arc, PSD95, Glur1; Figure 2 and (Ulloor and Datta, 2005; Seibt et al., 2012; Dumoulin et al., 2015; Del Cid-Pellitero et al., 2017; Renouard et al., 2018)]. Inhibition of mTOR and ERK signaling during sleep also decreases PRPs protein levels (Seibt et al., 2012; Dumoulin et al., 2015) suggesting a regulation of expression at the translational level. Activation of translation initiation and increased protein expression of some PRPs during sleep are also more pronounced at synapses supporting a synaptic-specific mechanism (Seibt et al., 2012).

REM sleep appears to be a preferred time for PRPs translation and structural plasticity. In mice, sleep-dependent structural plasticity is promoted by REM sleep during development and after learning (Li et al., 2017). Enzymes regulating translation and expression of PRPs proteins during sleep seem to also depend specifically on REM sleep. In the mouse hippocampus, the increase in ERK phosphorylation during the light (rest) phase (Eckel-Mahan et al., 2008) is maximal after a period rich in REM sleep (Luo et al., 2013). REM sleep deprivation in cats significantly decreases cortical ERK phosphorylation (Bridi et al., 2015). The cellular mechanisms by which REM sleep promotes translation are not known but may be supported by the specific network activity occurring during that stage. In the hippocampus, increased Arc and BDNF proteins after learning correlates with amounts of REM sleep pontine-geniculate-occipital waves (PGO) (Ulloor and Datta, 2005) (Table 1). Protein synthesis in dendrites of hippocampal CA1 neurons can be elicited in vitro with electrical stimulation in the theta (4–10 Hz) range (Feig and Lipton, 1993). Interestingly, dendritic protein synthesis in this model was dependent on pharmacological activation of muscarinic receptors, suggesting a role for the high cholinergic tone during REM sleep in translation activation at synapses. This observations are consistent with recent in vivo data showing an important role for REM sleep theta oscillations for memory consolidation (Boyce et al., 2016; Ognjanovski et al., 2017). Furthermore, both synaptic potentiation and depression can be elicited depending on the theta oscillatory phase (i.e., peak vs. through) at which electrical stimulation is applied (reviewed in Poe et al., 2010).

Immediate activation of consolidation upon learning is not optimal at the cellular and system levels. As discussed in Redondo and Morris (2011), consolidation of successive experiences at the time of encoding would result in superimposed activation of multiple cellular consolidation cascades within neurons. This would require extremely tight temporal regulation and high energy demands. This would also prevent selection of salient information and experience as all incoming stimuli would be stored (creating interference and a signal-to-noise problem). This problem, albeit expressed in different ways, was recognized many decades ago by scientists interested in the role of sleep in memory consolidation (reviewed in Alger et al., 2015). Sleep theoretically solves the problem as incoming stimuli are dampened and an internal reactivation of primed circuits ensures a very high signal-to-noise ratio during the consolidation process.

An advantage of the STC model is that cellular consolidation mechanisms are triggered within successive check points in time by binding significant experiences within shared circuits and neurons using shared cellular pathways. We propose in our model that these check points occur in sleep, when the brain is partially disengaged from incoming information and experience. This is consistent with the idea that memories are not just consolidated during sleep but are also actively reorganized to help cognitive processes such as generalization (Stickgold and Walker, 2013) or creative thinking and problem solving (Lewis et al., 2018). The ways by which sleep may achieve this is not clear but mechanisms of selective “memory triage” (Stickgold and Walker, 2013) and network association (Cai et al., 2009) have been proposed. Linked to this idea, part of the memory consolidation process involves the integration of new experience into pre-existing memory engrams (Rasch and Born, 2013) which would require the sharing and association of common circuits. While some theories have implicated REM sleep for memory association (Lewis et al., 2018), we propose that this process may particularly benefit from PRPs capture during NREM sleep (Figure 3A). The STC model and CREB-dependent increases in neuronal excitability have been proposed to facilitate associative memory (i.e., memory allocation) through the activation of shared circuits and synapses by experiences close in time (Govindarajan et al., 2006; Silva et al., 2009; Redondo and Morris, 2011; Rogerson et al., 2014; Lisman et al., 2018). If two experiences activate partially overlapping circuits, this leads to the tagging of synapses close in space (e.g., clustered on the same dendrites) and the co-activation of some neurons shared by these experiences (Figure 1A). During subsequent NREM sleep reactivation, different experiences can then be linked throughout the network. Enhanced CREB-dependent excitability will bias these neurons to be reactivated together during sleep and favor their linkage (Lisman et al., 2018). Similarly, synapses activated at close proximity will share the same type of tags and PRPs and will more likely strengthen (or weaken) in clusters, an important mechanism of memory allocation and storage (Rogerson et al., 2014; Kastellakis et al., 2015).

Our model provides a new way to connect waking experience to sleep homeostasis. As discussed by Benington, any proposed function of sleep should describe how the progress of that function determines sleep need (Benington, 2000). In our model, neurons primed during waking experience will express more (or more intense) modulation by NREM sleep oscillations. Over the course of the sleep period, this leads to the capture, translation, and removal of the priming elements, which also reduces the electrophysiological metrics of sleep homeostasis. Thus, in contrast to other hypotheses linking, for example, NREM SWA to plasticity (Tononi and Cirelli, 2003, 2006), heightened SWA is not caused by stronger synapses in wakefulness, nor does it result only in synaptic weakening during sleep. SWA increases are instead caused by the priming of neurons differentially engaged during waking experience. As this priming includes positive and negative tagging, subsequent increases in NREM SWA lead to synaptic weakening and strengthening, depending on the synaptic tag. If this is true, one would predict that changes in priming (i.e., excitability and tags) parallel changes in sleep need. There is evidence to support this idea. Increased synaptic phosphorylation, which is involved in synaptic tagging according to our model (see the section “Tags”), parallels increases in sleep need and decreases during sleep (Wang et al., 2018). PRPs transcript levels generally are maximal during waking and then decline during sleep (Figure 2 and Table 1). Several IEGs show a peak and decline in their cortical expression following SD and recovery sleep that parallels the time constant for the discharge of sleep homeostasis (as measured by SWA) (Gerstner et al., 2016). In addition, some of the more dramatic sleep phenotypes in mutant animals result from changes in membrane excitability [e.g., potassium channels in drosophila (Cirelli et al., 2005; Bushey et al., 2007; Koh et al., 2008; Wu et al., 2014) and mice (Espinosa et al., 2004; Douglas et al., 2007)].

The goal of this article is not to exhaustively review all other theories about sleep and plasticity. However, it is useful to put our theoretical model in context. Other theorists have proposed roles for REM and NREM sleep in synaptic plasticity and memory consolidation. For example, at a systems level, it has been proposed that NREM sleep oscillations transfer and propagate memory traces from the hippocampus to neocortical areas (Ribeiro and Nicolelis, 2004; Rasch and Born, 2013; Navarro-Lobato and Genzel, 2018). REM sleep episodes could then promote consolidation of the transferred engram at the molecular level, mainly via transcriptional events (Ribeiro and Nicolelis, 2004; Rasch and Born, 2013; Grønli et al., 2014; Navarro-Lobato and Genzel, 2018). Some models also include a role for translation during sleep (Grønli et al., 2014; Navarro-Lobato and Genzel, 2018). However, unlike our model, this is primarily linked to NREM sleep and promotes global restorative biogenesis and local synaptic strengthening (Grønli et al., 2014).

Our model is broadly compatible with these ideas, but distinct in important ways. We propose that the bulk of transcriptional events occur during waking and is necessary for PRP synthesis at the transcriptional, not translational level. We focus on synaptic changes that may accompany systems level changes, thus our ideas are not incompatible with the transfer of information from the hippocampus to the cortex. We add to the proposed function of NREM sleep oscillations in that we suggest PRP capture at selected synapses in reactivated circuits (regardless of brain structures). In contrast to other ideas positing different roles for REM and NREM sleep in plasticity (Crick and Mitchison, 1983; Giuditta et al., 1995; Rasch and Born, 2013; Giuditta, 2014; Almeida-Filho et al., 2018; Navarro-Lobato and Genzel, 2018), our ideas are based on current ideas about metaplasticity, rather than uniform processes that drive plasticity in one (net) direction, or transcriptional events only. Our model also differs in the description of priming mechanisms that link plastic changes in waking with subsequent changes in sleep.

Our model also differs from the ‘Synaptic Homeostasis Hypothesis (SHY)’ which proposes that sleep weakens synapses while sparing others (a process referred to as ‘down-selection’), resulting in a ‘net’ weakening of synapses after sleep. This type of plasticity has also been termed ‘downscaling’ (Tononi and Cirelli, 2003) and described in terms similar to homeostatic plasticity (Turrigiano, 2008). Although SHY allows for some synapses to become stronger during sleep, this occurs under special (e.g., non-physiological) conditions, and therefore is not a major aspect of SHY (Tononi and Cirelli, 2014). Finally, according to SHY ‘protection’ from down-selection does not involve an absolute increase in synaptic strength; merely a relative change compared to the net downscaling of other synapses in sleep (Tononi and Cirelli, 2014; Cirelli and Tononi, 2015). Our model instead states that synaptic strengthening is just as likely to occur as weakening during sleep (depending on prior waking experience), is not necessarily relative to other synapses, does not require ‘net’ synaptic downscaling and reflects physiological and adaptive processes.

This does not make our model incompatible with synaptic homeostatic adjustments during sleep. Some evidence cited in support of SHY may be explained by circadian rhythms rather than sleep (Frank and Cantera, 2014). Other findings suggest that some forms of homeostatic plasticity may only occur during wakefulness (Hengen et al., 2016). Some synaptic changes ascribed to ‘net’ downscaling may instead reflect the influence of negative tags and PRPs as we describe. Nevertheless, homeostatic adjustments of synapses during sleep may occur but in a manner that more generally restores set-points of activity and synaptic weight upward or down-ward depending on the neuron (Watson et al., 2016; Levenstein et al., 2017). In addition, as shown recently in a study of inhibitory and excitatory receptors on the soma of cortical principal neurons (Del Cid-Pellitero et al., 2017), homeostatic scaling in sleep could provide a global means of adjusting excitability. In this context, this might reflect a sleep-dependent and bi-directional restorative process, reflecting the neurons’ previous waking history. This homeostatic process would not interfere with Hebbian synaptic plasticity on individual synapses, neurons or circuits, as originally proposed (Turrigiano, 2008).