Latent mechanisms of plasticity are upregulated during sleep

Abstract

Sleep is thought to serve an important role in learning and memory, but the mechanisms by which sleep promotes plasticity remain unclear. Even in the absence of plastic changes in neuronal function, many molecular, cellular, and physiological processes linked to plasticity are upregulated during sleep. Therefore, sleep may be a state in which latent plasticity mechanisms are poised to respond following novel experiences during prior wake. Many of these plasticity-related processes can promote both synaptic strengthening and weakening. Signaling pathways activated during sleep may interact with complements of proteins, determined by the content of prior waking experience, to establish the polarity of plasticity. Furthermore, precise reactivation of neuronal spiking patterns during sleep may interact with ongoing neuromodulatory, dendritic, and network activity to strengthen and weaken synapses. In this review, we will discuss the idea that sleep elevates latent plasticity mechanisms, which drive bidirectional plasticity depending on prior waking experience.

INTRODUCTION

Time of day is frequently not considered in rodent studies of learning and memory [1]. It is crucial to take time of day into account because sleep/wake states co-vary with circadian time, and sleep is thought to play an important role in learning. Disrupting sleep interferes with learning and memory, and recent studies using technological advances to enhance physiological sleep further support a role for sleep in memory processing and recall [2–4]. Neuronal plasticity, which we define as persistent functional changes in response to altered activity, is the cellular substrate of learning and memory. However, the mechanisms by which sleep promotes plasticity to enhance learning and memory remain unclear.

One fundamental conundrum surrounding the role of sleep in plasticity is that sleep is required even in the absence of a remodeling stimulus. Laboratory animals under constant standard housing conditions are not exposed to novel experience that would alter neuronal activity during wake. Accordingly, Hebbian long-term potentiation (LTP), which underlies many forms of learning, is not engaged during wake under these conditions: LTP is expressed postsynaptically, but physiological studies rarely report differences in postsynaptic measures (miniature excitatory postsynaptic current (mEPSC) amplitude, alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid/N-methyl-D-aspartate receptor (AMPAR/NMDAR) ratio) between sleep and wake [5–10]. Nevertheless, animals robustly cycle through arousal states (wake, non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep), and respond to sleep deprivation with a homeostatic buildup of sleep pressure (for examples, see [11,12]).

In this review, we will discuss cellular, molecular, and physiological changes that have been linked to plasticity, but that occur over sleep/wake states in the absence of plasticity induction. These changes are often at odds with functional measures of postsynaptic strength [5–9], suggesting that they are not a readout of plasticity. Instead, we will explore the idea that these changes are latent mechanisms constitutively upregulated during sleep, such that neurons are poised to undergo plasticity following a remodeling stimulus during prior wake. Many of these latent mechanisms can promote either synaptic strengthening or weakening, making sleep a state during which the brain is primed for bidirectional changes in synaptic strength.

PLASTICITY-RELATED SIGNALING PATHWAYS MODULATED BY SLEEP/WAKE STATES

Building upon earlier microarray studies, several recent reports have taken -omics approaches to identify the molecular changes that are regulated by basal sleep at the transcriptional, translational, and post-translational levels. RNA sequencing has revealed that sleep upregulates transcripts enriched for gene ontology terms related to cell adhesion, RNA processing/transcriptional regulation, dendritic localization, synaptic organization, postsynaptic membrane components, signaling pathways, and the cytoskeleton [11,13–15], pathways that influence structural and functional synaptic plasticity. Sleep-dependent regulation of protein expression also occurs at the levels of mRNA trafficking to the synapse and translation of new and pre-existing mRNAs [11,15,16], and proteomic analysis of forebrain synaptoneurosomes revealed that this results in the synaptic enrichment of proteins involved in cellular signaling [11]. Furthermore, phosphoproteomic analysis demonstrated that sleep/wake states control the phosphorylation of approximately half of all synaptic phosphoproteins in the forebrain, affecting pathways involved in synaptic transmission, cytoskeletal reorganization, and excitatory/inhibitory balance [12]. Together, these studies demonstrate that sleep regulates the expression and function of many synaptic proteins with the capacity to promote synaptic plasticity.

One consistent finding across studies is that transcription of activity-regulated cytoskeleton-associated protein (Arc) is upregulated during wake and downregulated during sleep in forebrain neurons [11,13,15–21]. On the other hand, Arc translation is elevated during sleep, although reports disagree, possibly because this increase is transient [11,19–21]. Arc protein degradation machinery may also decrease during sleep, creating a permissive environment to sustain Arc when it is elevated by plasticity induction during wake [16,22]. The presence of Arc during sleep may promote both synaptic strengthening and weakening. During long-term depression (LTD) and homeostatic downscaling, Arc weakens synapses via endocytosis of AMPARs [23,24]. On the other hand, when synapses are potentiated, Arc increases their relative strength by heterosynaptically weakening nonpotentiated synapses. One underlying mechanism is the presence of inactive CaMKIIβ at less-active synapses, causing Arc accumulation and AMPAR endocytosis [25]. Another mechanism, shown in a recent preprint, is the IRSp53-dependent packaging and release of Arc mRNA and protein in extracellular vesicles after LTP induction, which results in AMPAR endocytosis at neighboring dendrites [26]. Finally, Arc may directly promote synaptic strengthening. Following LTP, Arc stimulates and stabilizes actin polymerization via cofilin and drebrin A at potentiated synapses, although this role is controversial, as others have found that Arc is not required for LTP [27–29].

Homer1a is also regulated over basal sleep/wake states, with increased trafficking of Homer1a protein to the synapse following sleep-rich periods [30]. While long Homer isoforms cross-link synaptic proteins, the short Homer1a isoform acts as a dominant-negative, disrupting interactions between receptors, scaffolding proteins, and second messengers. Among other effects, Homer1a acts on group I metabotropic glutamate receptors (mGluRs) to modulate ionotropic glutamate receptor currents. Homer1a activates mGluRs independent of agonist binding, which in turn decreases AMPAR phosphorylation and surface expression [31]. In addition, Homer1a releases mGluRs from their attachment to the perisynaptic region, allowing translocation to the postsynaptic density where they physically inhibit N-methyl-D-aspartate receptor (NMDAR) activity [32]. While these and other functions have caused Homer1a to be predominantly associated with synaptic weakening via LTD and homeostatic synaptic downscaling, Homer1a also strengthens individual synapses in an input-dependent manner. Homer1a enables interactions between phosphorylated mGluR and the prolyl isomerase Pin1, which potentiates NMDAR-mediated currents [33]. This can reverse the polarity of plasticity depending on the presence of Pin1 and mGluR phosphorylation at individual synapses, and accounts for the input specificity of metaplastic synaptic depression [34].

MODULATION OF SYNAPTIC FUNCTION OVER SLEEP/WAKE STATES

Although a preponderance of evidence indicates that excitatory postsynaptic strength is unaltered by basal sleep [5–9], many reports show that arousal state impacts other aspects of synaptic transmission (Table 1), and these changes could have important consequences following plasticity induction. Excitatory synapses onto excitatory neurons are the best studied. In acute brain slices collected following sleep-rich periods, mEPSC frequency decreases in primary visual (V1), prefrontal (PFC), and anterior cingulate cortices, as well as hippocampal area CA1, compared to wake-rich periods [5–7,9] (but see [8]). These findings are consistent with sleep-dependent regulation of the number of presynaptic release sites. While anatomical changes do not necessarily indicate functional differences (for example, see [35]), structural evidence is generally in agreement with decreased excitation over sleep. Spine density is decreased in sensorimotor cortex and lateral/central amygdala [36,37], and axon-spine interface area in primary somatosensory (S1) and motor (M1) cortices is lower, following sleep [38–40]. These changes are consistent with altered presynaptic release, although some studies have also found decreased AMPAR content in sensorimotor cortex and spine size in the central amygdala following sleep [30,37]. In contrast, increased spine density after sleep has also been reported in CA1, dentate gyrus, and basolateral amygdala [37,41,42]. These differences may reflect dendritic branch- and spine-type-specific effects of sleep on excitatory synapses; indeed, many studies reporting decreased spine measures show differential impacts based on spine size or morphology [30,37,38,40]. Nevertheless, these studies provide evidence for an overall decrease in excitatory transmission in the brain over periods of sleep.

Table 1.

Evidence for altered excitatory and inhibitory synaptic transmission during sleep. Arrows indicate differences observed following spontaneous sleep (measurements taken during the normal rest phase) relative to spontaneous wake (measurements taken during the normal active phase) and/or enforced wakefulness during the normal rest phase, as indicated.

Ref.	Species, age	Methodology	Brain Region(s), Cell Type(s)	Wake Manipulation	Excitation	Inhibition	Circuit Output
Liu et al., 2010 [5]	rat 4–8 weeks mouse 3–4 weeks	Whole-cell patch clamp in acute slice	Frontal cortex L2/3 Pyramidal	Spontaneous 4h enforced	↓ mEPSC frequency (rat) ↓ mEPSC frequency (mouse, vs enforced) ↓ mEPSC amplitude (rat, vs enforced)	-	-
Bridi et al.2020 [6]	mouse 5–10 weeks	Whole-cell patch clamp in acute slice	V1 L2/3, PFC L2/3, CA1 Pyramidal	Spontaneous 4h enforced	↓ mEPSC frequency (vs spontaneous)	↑ mIPSC frequency (vs spontaneous) ↑ sIPSC charge (V1, vs spontaneous and enforced)	↓ E/I ratio of the L2/3–2/3 circuit (V1, vs spontaneous) ↓ ability of AMPAR current to evoke spikes (V1, vs spontaneous)
Bjorness et al., 2020 [7]	mouse 8–12 weeks	Whole-cell patch clamp in acute slice	ACC L2/3 Pyramidal	6h enforced	↓ mEPSC frequency	-	-
Cary and Turrigiano 2021 [8]	Rat 25–31 days	Whole-cell patch clamp in acute slice Single unit recording in vivo	V1 L2/3 V1 L4 PFC L2/3 Pyramidal	Spontaneous	No change, mEPSC frequency, amplitude	-	No change, thalamocortical-evoked spiking
Bridi et al., 2025 [9]	mouse 5–8 weeks	Whole-cell patch clamp in acute slice	V1 L2/3 Pyramidal	Spontaneous	↓ mEPSC frequency	↑ mIPSC frequency ↑ sIPSC charge	↓ E/I ratio of the L2/3–2/3 circuit
Wu et al., 2022 [43]	mouse 8–12 weeks	Whole-cell patch clamp in acute slice	CA1 Pyramidal	Spontaneous	-	↑ mIPSC frequency and amplitude ↓ tonic inhibitory current	-
Zong et al., 2023 [46]	mouse 5–10 weeks	Whole-cell patch clamp in acute slice 2P Ca2+ imaging in vivo	V1 L2/3 PV, Pyramidal	Spontaneous 4h enforced	-	↑ evoked IPSC in slice (Pyramidal, vs spontaneous) ↑ synaptic drive onto PV in slice (↑ mEPSC & ↓ mIPSC frequency vs spontaneous; ↑ sEPSC & ↓ sIPSC charge vs enforced; ↑ sEPSC charge vs spontaneous)	↓ spontaneous and evoked Ca2+ activity in vivo (Pyramidal, vs spontaneous)
Alfonsa et al., 2023 [44]	mouse 4–12 weeks	patch clamp in acute slice and in vivo	S1 L5, A1 L2/3 Pyramidal	Spontaneous 3h enforced	-	↑ hyperpolarization by GABAergic signaling in slice (S1, A1 vs spontaneous; S1 vs enforced) ↑ hyperpolarization by GABAergic signaling in vivo (L2/3, vs enforced)	↓ synaptic integration in slice (S1, vs spontaneous)
Pracucci et al., 2023 [45]	mouse >1 month	2P imaging in vivo	V1 L2/3 Pyramidal	Spontaneous	-	↑ Cl− export by KCC2	-
Untiet et al., 2023 [53]	mouse 7–14 weeks	Fiber photometry in vivo	S1 Astrocytes	Spontaneous	-	↑ astrocytic [Cl−]I (↑ [Cl]o during GABAergic transmission)
Diering et al. 2017 [30]	mouse 8–10 weeks	2P imaging in vivo Western (PSD fraction)	M1 L5 pyramidal Whole forebrain	Spontaneous	↓ surface GluAl in spines with high GluA1 content (M1) ↓ PSD AMPAR content (forebrain)	-	-
Maret et al. 2011 [36]	mouse ~3–6 weeks	2P imaging in vivo	sensorimotor L5 pyramidal	Spontaneous 6–7h enforced	↓ spine density	-	-
De Vivo et al., 2017 [38]	mouse 1 month	SEM	M1, S1 L2 pyramidal	Spontaneous 7h enforced	↓ ASI (small/medium spines, spines with endosomes)	-	-
De Vivo et al. 2019 [39]	mouse 2 weeks	SEM	M1 L2 pyramidal	4.5–6h enforced	↓ ASI	-	-
Nagai et al., 2021 [40]	mouse 2 weeks, 1 month	SEM	M1 L2 pyramidal	4 & 15h enforced 4d CSR Spontaneous	↓ ASI (vs 4h enforced, 2 weeks) ↑ ASI (vs 15h enforced, 2 weeks) ↓ ASI in spines with endosomes (vs CSR, 1 month) ↑ density of spines lacking synapses (vs spontaneous, 1 month)	-	-
Rexrode et al. 2023 [37]	mouse 4.5–5.5 months	Immunocyto-chemistry	Amygdala	5h enforced	↓ overall/stubby spine density (LA) ↓ overall spine density (CeA) ↑ overall/mushroom spine density (BLA) spine subtype- and region-specific morphological changes	-	-
Raven et al., 2019 [41]	mouse 3 months	Golgi staining	DG granule cells	5h enforced	↑ thin and branched spine density (proximal dendritic branches only)	-	-
Bolsius et al. 2022 [42]	mouse 3 months	Golgi staining	CA1 pyramidal	5h enforced	↑ spine density (apical dendritic branches 3–9, basal dendritic branches 3–6 only)	-	-

2P: 2-photon; SEM: serial electron microscopy; ASI: axon-spine interface; ACC: Anterior Cingulate Cortex; LA, CeA, BLA: lateral, central, and basolateral amygdala; TRN: thalamic reticular nucleus; DG: dentate gyrus; CSR: chronic sleep restriction; KCC2: Potassium Chloride Cotransporter 2

Recent studies examining the effects of sleep on inhibitory synaptic transmission show the opposite trend compared to excitation (Table 1). In V1, PFC, and CA1, inhibitory synaptic transmission onto pyramidal neurons is increased following sleep-rich periods, as measured by miniature inhibitory postsynaptic current (mIPSC) frequency [6,9,43]. These changes, which are consistent with increased numbers of presynaptic GABAergic release sites, may act synergistically with postsynaptic mechanisms. For instance, decreased intracellular chloride concentration ([Cl⁻]_i) following sleep-rich periods, which increases hyperpolarization during GABA_AR opening, has been observed in S1 and V1 pyramidal neurons [44,45]. In addition to enhanced inhibition onto pyramidal cells, synaptic drive onto parvalbumin-positive (PV) inhibitory interneurons is elevated following sleep: whole-cell patch clamp recordings of V1 PV cells show that mEPSC frequency and spontaneous (s)EPSC charge is higher, and mIPSC frequency/sIPSC charge is lower, following sleep-rich periods [46]. Interestingly, alterations in inhibitory transmission, which build up over several hours [6], may interact with faster, state-dependent and cell-type-specific changes in inhibitory neuron activity. PV cell activity acutely increases with transitions into NREM and REM sleep in S1 and PFC in vivo [47,48], which in conjunction with elevated inhibitory transmission mechanisms would further enhance inhibitory tone. As a whole, these studies are consistent with an overall increase in inhibitory transmission over periods of sleep.

Taken together, studies demonstrating decreased excitation and increased inhibition during sleep are indicative of a decrease in the excitation/inhibition (E/I) ratio across many brain regions, consistent with direct measurements of the E/I ratio in V1 [6,9]. Importantly, these sleep-dependent changes in synaptic transmission result in altered circuit output. In pyramidal neurons, more AMPAR current is required to reach spike threshold, spike probability is lower, and spontaneous calcium activity in vivo is reduced following sleep-rich periods [6,44,46]. Collectively, these findings demonstrate that sleep decreases the E/I ratio and circuit excitability (Table 1); the consequent changes in neuronal firing rates and patterns may influence activity-dependent plasticity in multiple ways (discussed below).

GLIAL FUNCTION OVER SLEEP/WAKE STATES

Glial cells are dynamically modulated across arousal states [49–53] and are likely to emerge as important players in sleep-dependent plasticity. However, glia are less well-studied than neurons, making their contributions more nebulous. In this section, we review the structural and functional changes that glia constitutively undergo across sleep/wake states, and potential implications for neuronal activity and plasticity.

Both astrocytes and microglia are structurally dynamic across arousal states. Astrocytic processes are further from the synaptic cleft after sleep [49]. This may decrease glutamate clearance, favoring compensatory mGluR-dependent presynaptic depression of glutamate release and/or enhanced postsynaptic LTP expression due to elevated synaptic glutamate [54,55]. Microglial morphodynamics are less clear, as studies have reported both increased and decreased surveillance during sleep [51,52]. Nevertheless, microglial processes contact active spines more than inactive spines during wake without affecting their activity [51]. On the other hand, they are less likely to contact spines during sleep, but when contact is made, it elevates spine Ca²⁺ transients [51]. The dissociation between contact and activity regulation suggests that microglia play distinct roles during waking experience and subsequent sleep. Microglia also digest perineuronal nets by releasing cathepsin S across many brain regions during sleep [56]. Loss of perineuronal nets is thought to create a permissive environment for plasticity and impact PV cell function (reviewed in [57]), further altering the E/I ratio as discussed above.

Astrocytes also differentially regulate their intracellular ion concentrations across arousal states, which may influence multiple plasticity-related processes. For instance, astrocytic Ca²⁺ signaling is decreased, becomes less synchronized, and displays different spatiotemporal dynamics within processes during sleep [50]. While the exact relationship with plasticity is unclear, a recent study showed that the presence or absence of astrocytic Ca²⁺ can determine the polarity of NMDAR-dependent presynaptic plasticity, suggesting that Ca²⁺ in astrocytic subdomains during sleep could influence the direction of plasticity at individual synapses [58]. In addition, astrocytic [Cl⁻]_i increases during sleep, creating a reservoir to boost extracellular [Cl⁻] and neuronal hyperpolarization during GABAergic transmission [53], further decreasing the synaptic E/I ratio.

HOW MIGHT PLASTICITY INDUCED DURING WAKE BE AFFECTED BY SLEEP?

Molecular mechanisms

One proposed mechanism for sleep-dependent plasticity is that learning during wake sets molecular ‘tags’ at remodeling synapses which are ‘captured’ during subsequent sleep, leading to long-term modification of synaptic weights [59]. In this scenario, tags can be positive or negative, determining the polarity of plastic changes that are captured during sleep. The molecular, cellular, and physiological changes during basal sleep described above may provide a milieu favoring bidirectional plasticity, such that the brain is primed to capture either positive or negative tags set during prior waking (Figure 1). This is consistent with the concomitant strengthening and weakening observed during sleep following plasticity induction [60,61].

Figure 1.

Latent bidirectional plasticity mechanisms that are upregulated during basal sleep, and their downstream effectors. Input-specific polarity of STDP (1) depends on coincidence of neuronal reactivation patterns with neuromodulatory signaling. Neuronal reactivation may also induce potentiation or depression based on the magnitude of postsynaptic Ca²⁺ elevation (2) at individual synapses. Incoming activity also interacts with molecular signaling pathways that favor synaptic strengthening or weakening. Increased synaptic Homer1a (H1a) (3) decreases NMDAR signaling and promotes AMPAR endocytosis, but in synapses containing phosphorylated mGluR and Pin1 can enhance NMDAR currents to promote potentiation. Sustained Arc protein levels during sleep (4) may increase synaptic strength by stabilizing actin following LTP. Arc also promotes AMPAR endocytosis, which could heterosynaptically weaken synapses via CaMKIIβ or IRSp53 (not shown) to increase the relative strength of potentiated synapses, or could directly weaken synapses targeted for downregulation.

Several plasticity-related mechanisms upregulated during sleep could capture synaptic tags set during prior wake. For example, Homer1a is elevated at synapses during sleep and favors net synaptic weakening [30]. However, individual synapses could be tagged with mGluR phosphorylation and Pin1 during plasticity induction, protecting them from synaptic weakening during sleep. Furthermore, sleep may upregulate Arc synthesis and/or reduce Arc turnover, raising the possibility that Arc and the molecules it interacts with could also act as tags [16,20,22]. For example, Arc mRNA may tag synapses for weakening via local Arc translation [24]. In addition, LTP induction during wake could set Arc-dependent tags for strengthening (cofilin, drebrin A) and heterosynaptic weakening (inactive CaMKIIβ, IRSp53) during subsequent sleep [25–28]. Intriguingly, microglia are required for synaptic tagging and capture during hippocampal LTP [62], and while the underlying mechanisms are far from clear, this could occur via Arc. Microglia may set tags when they contact active spines during wake by releasing brain-derived neurotrophic factor (BDNF) to phosphorylate neuronal tropomyosin-related kinase receptor B (TrkB), which increases Arc expression [51,63,64]. Ultimately, the polarity and magnitude of synaptic changes that occur at individual synapses during sleep likely depends on the specific complement of synaptic tags set during prior waking experience, which have yet to be conclusively identified, in combination with ongoing neuronal activity.

Reactivation of neuronal firing patterns

Neuronal activity patterns that occur during waking experience are reactivated during subsequent NREM hippocampal sharp wave ripples, cortical slow oscillations, and spindles, as well as during REM sleep and periods of wake [65]. During sleep, inhibiting neuronal populations active during prior learning prevents memory consolidation, pointing to an instructive role of sleep-dependent reactivation in plasticity [66]. Reactivation of inputs likely generates different circuit outputs during sleep and wake states, in part because arousal state-dependent changes to the E/I ratio (due to synaptic and glial changes; Table 1) impact circuit excitability. Furthermore, sleep-dependent changes in the E/I ratio occur in the highly plastic layer 2/3–2/3, but not the less plastic layer 4–2/3, circuit in V1, supporting the idea that sleep-dependent E/I modulation promotes plasticity [6,34]. Accordingly, the fidelity of reactivated spiking patterns is higher during sleep than wake [67], which may be due to improved spike timing precision when the E/I ratio is low. Depending on specific firing patterns and their interaction with ongoing network activity, precise reactivation during sleep could favor either synaptic strengthening or weakening (Figure 1).

High-fidelity reactivation of neuronal activity patterns during sleep may interact with ongoing network dynamics to promote bidirectional synaptic plasticity via spike timing-dependent plasticity (STDP). STDP is induced when pre- and postsynaptic neurons fire within a precise temporal window of one another. The polarity of STDP is determined by the order of firing: LTP is produced when the presynaptic cell fires first, and LTD occurs when the postsynaptic cell fires first (reviewed in [68]). It is plausible that STDP is engaged by reactivation during sleep, as Dickey and colleagues showed that ripples (brief, high-frequency oscillations associated with neuronal reactivation) occur in many human cortical regions during both NREM slow waves and spindles, and unit cofiring during ripples occurs with short delays optimal for STDP [69].

In rodent V1, norepinephrine (NE) and serotonin (5-HT) signaling are necessary for timing-dependent LTP and LTD, respectively, within 5–10 seconds following spike pairing [70]. Both NE and 5-HT are high during wake and minimal during REM sleep, but are released at intermediate levels and undergo infraslow oscillatory activity in cortical regions during NREM sleep [71,72]. Intriguingly, spindles are generated during the downslope of the NE oscillation and terminated by increased NE; this may lead to a scenario in which neuronal reactivation of pre-post pairings occurs during spindle activity, and can then be converted into LTP by the subsequent increase in NE. While less is known about how the 5-HT infraslow oscillation coordinates with ongoing network activity, timing-dependent LTD may also be induced in a similar manner.

Neuronal reactivation could also promote bidirectional plasticity by pairing reactivated presynaptic spiking patterns with postsynaptic depolarization during sleep. In brain slice preparations, the polarity of pairing-induced plasticity depends on the magnitude of postsynaptic depolarization: LTP or LTD is produced when presynaptic spiking is paired with strong or weak depolarization, respectively [73]. Multiple mechanisms control the magnitude of cortical postsynaptic depolarization during sleep. For instance, microglial contact with spines during NREM sleep (but not wake) elevates the frequency of local Ca²⁺ transients [51]. Furthermore, during spindles, Ca²⁺ activity increases specifically in the dendritic compartment of layer 5 neurons, and elevated VIP cell activity causes dendritic disinhibition at the end of delta waves in layer 2/3 [48,74]. This may cause brief but temporally precise windows during which reactivated presynaptic spiking patterns are paired with dendritic depolarization to promote plasticity. Intriguingly, VIP-dependent dendritic disinhibition is also high during REM sleep [47]. While reactivation has been observed during REM sleep, its connection with synaptic plasticity is less clear [65]; nevertheless, dendritic Ca²⁺ spikes are required during REM sleep for both new spine elimination and strengthening following motor learning [60], suggesting a role for dendritic activity during REM sleep in bidirectional plasticity.

CONCLUDING REMARKS

Under baseline conditions in animal models, sleep engages many mechanisms capable of promoting bidirectional plasticity, which may be differentially utilized depending on the content of prior waking. This is compatible with multiple theories of sleep function, including the synaptic consolidation, systems consolidation, and synaptic homeostasis hypotheses (for review, see [65]). The combination of synaptic strengthening and weakening induced during waking experience may be solidified locally during sleep (synaptic consolidation), and/or transferred to novel neural substrates in the brain (systems consolidation). In support of these ideas, sleep is associated with a mixture of synaptic strengthening and weakening, even when there is a net change in one direction [30,36,60,61]. On the other hand, while the mechanisms discussed above can promote plasticity bidirectionally, they may only operate in one direction during sleep. In this scenario, following potentiation during waking experience, synapses undergo compensatory weakening during subsequent sleep (synaptic homeostasis hypothesis).

Developmental age is an important factor when considering synaptic modifications during sleep. For example, decreased spine density during sleep has been reported at ages when synaptic pruning is prominent [36,60], and transcriptional, translational, and post-translational regulation of many synaptic proteins over the sleep/wake cycle changes with animal age [14,19]. Moreover, animals show differential vulnerability to sleep deprivation across postnatal development and adulthood [19], such that discrepancies between studies due to age may be amplified by the use of different sleep deprivation paradigms. Nevertheless, many sleep-dependent changes are largely consistent across age (e.g. mEPSC frequency [5–7,9], Arc expression [11,13,15,16,18–21], Wnt pathway gene expression [14]). Therefore, while sleep may promote some plasticity mechanisms only at specific developmental time points, others appear conserved.

Finally, many of the studies discussed in this review do not differentiate between NREM and REM sleep, making their respective contributions unclear. Several mechanisms discussed above influence NREM slow wave activity, which is locally elevated as a function of activity during prior waking [18,44,50], while REM sleep is required for bidirectional plasticity in multiple paradigms in vivo [60,61]. Whether NREM and REM sleep serve independent or sequential functions in plasticity remains to be determined.

On the surface, it is perplexing why the brain would devote resources to plasticity in the absence of learning. However, from an evolutionary standpoint, animals face ever-changing conditions in their environments (compared to constant conditions in the laboratory) and must rapidly adapt to survive. In natural settings, sleep has a high probability of following novel experiences during prior waking, and latent plasticity mechanisms will be frequently utilized. Therefore, based on the evidence presented in this review, we propose the latent mechanisms of plasticity (LaMP) theory of sleep function, which could explain the evolutionary conservation of sleep as a behavioral state.