Proteostasis failure exacerbates neuronal circuit dysfunction and sleep impairments in Alzheimer’s disease

Both major pathological protein species in AD exert deleterious effects on sleep function [70], and Aβ oligomers interfere with sleep/wake patterns in mice in a dose-dependent manner [71]. Recently, Özcan and colleagues described a direct effect of Aβ on sleep, via injection of oligomers of differential length into zebrafish [17]. Sleep regulation is Aβ oligomer size-dependent, in which short oligomers increase hypothalamic neuronal activity and induce wakefulness via adrenergic and progesteronergic receptor signalling, long oligomers reduce neuronal activity and induce sleep via the prion protein pathway, and very long oligomers have no effect [17]. Conversely, Aβ oligomer injection in mice induced sleep fragmentation (reduced NREM and REM time and increased sleep stage transitions) but this was not observed in prion protein-deficient mouse strains [72], suggesting a complex mechanism for Aβ-sleep regulation. These results may partially explain the broad spectrum of AD-related sleep disturbances, including reduced total sleep time and increased daytime sleep. Furthermore, sleep and circadian rhythm disruptions in AD may trigger this bi-directional control of sleep, in which impaired clearance and metabolism of Aβ causes alterations in diurnal fluctuations [27, 73].

Multiple recent studies document the relationship of tau with sleep disturbances and electroencephalogram (EEG) abnormalities. In cognitively normal older adults (mean age ~ 73–75), low frequency EEG signal during NREM sleep (1–2 Hz; indicative of delta waves prominent in SWS) exhibit an inverse relationship with AD pathology most prominently with AV-1451 tau positron emission tomography (PET) signal [60, 74], and in AD patients (mild-moderate) sleep–wake disturbances have been shown to correlate with CSF levels of phosphorylated tau [75]. Through an increased neuronal tau release, sleep deprivation in mice and in healthy 30–60-year-old adults compounds this effect with greater brain (in mice) and CSF (> 50% increase in humans) tau [29, 76], in addition to an increase in major AD biomarkers. Higher levels of CSF pT181 and pT217, but not pS202, were observed, with increased levels of non-phosphorylated tau forms at those epitopes [76].

Winer and colleagues assessed associations of tau (¹⁸F-flortaucipir) and Aβ (¹¹C-PIB) PET levels with sleep in older adults (mean age ~ 78), comparing objective (wristwatch actigraphy) and subjective (Pittsburgh sleep quality index: PSQI) measures over 1 week. Objective sleep impairment correlated with greater tau PET in early Braak-related stages, in the EC as well as the medial temporal lobe more broadly, but not with cortical Aβ PET. However, both tau and Aβ are significantly associated with self-reported sleep disturbances, indicating that individuals with more Aβ reported worse sleep quality than they actually had with significant changes in PSQI global and efficiency scores, but not sleep duration. This effect was potentiated by loss of executive function [25]. Whether a subjective underestimation of sleep quality can impact progression to future objective sleep disturbances is an interesting topic for future research; critically, these results indicate the need for EEG, polysomnography (PSG), and actigraphic and accelerometric devices as biomarkers for NDDs (discussed in Sect. "Neurodegenerative disease biomarkers and their relationship to sleep").

Elevated tau has been observed in young individuals with OSA [77], and brain-derived exosomes contain higher levels of total tau, pT181, and Aβ in those with OSA and MCI, compared to OSA alone (age range: 35–65) [78], indicating potentiation of NDD exosome-mediated spread of proteinopathy with sleep disturbances [79–84]. Taken together, these reports indicate an association of tau with sleep that likely contributes to the proteinopathy-sleep bidirectional relationship observed in AD.

Conversely, other studies have demonstrated no changes in total or phosphorylated tau (or Aβ) with 5 days of partial sleep disruption in healthy 20–40-year-olds; total and REM sleep were lost, yet SWS was unimpaired in this paradigm [85], which may have preserved clearance mechanisms such as glymphatics. Overnight interventions in healthy volunteers (aged 35–65), to prevent SWS for one night, also did not show any significant changes in the levels of CSF tau but had elevated Aβ. Poorer actigraphic measures of sleep at home over 6 days associated with higher CSF tau levels in these participants [61]. It is important to note that these two experiments were conducted in young and middle-aged healthy participants, and with increasing age and in NDD, endogenous clearance mechanisms lose efficacy [1], which will contribute to higher protein accumulation after sleep loss.

EEG alpha waves (~ 7–12 Hz) are signatures of rest and quiet wakefulness [74], and have been shown to relate to AD-related tau pathology. In subjective cognitive decline, MCI and AD participants in their 60 s, dampened EEG alpha power and synchronization correlated with increasing levels of CSF total and phosphorylated tau [86], and of CSF phosphorylated tau with a lower peak alpha frequency in the power spectrum of older adults (mean age ~ 70) [87]. When stratified by Aβ (¹¹C-PIB) and by tau PET (¹⁸F-MK-6240) positivity, the peak alpha frequency slowed from ~ 9.5 Hz to ~ 8 Hz in positive groups [87]. In cognitively normal older adults (mean age ~ 75), CSF total and phosphorylated tau, and most notably in those with high p-tau:Aβ ratio, was correlated with an increasing proportion of theta (analyzed in 4–8 Hz) waves, but no changes in delta, alpha or beta frequency bins [88], which may be due to overall EEG slowing. Proteinopathy in the rest-active ‘default mode network’ [89, 90] and tau mediated neuronal suppression [5] may be contributing factors to EEG slowing and impairments in alpha waves. Furthermore, alpha wave effects may stratify by sex: alpha power and total tau negatively correlated in male but not female participants with MCI (mean age ~ 75) [91], and greater resting state alpha EEG activity was reported in female vs. male healthy older adults, as well as in MCI and AD (mean age ~ 69–70) [92]. These results indicate the potential for EEG alpha as a non-invasive AD biomarker, and further research may investigate if impaired alpha wave activity in AD can attenuate the beneficial effects of quiet wakefulness.

The connection of tau with sleep loss and EEG alterations are supported by preclinical models. In tau knockout mice, there is reduced delta power, NREM and total sleep time, and increased state transitions [93]. In P301S (at advanced stages) and rTg4510 tauopathy mice there is reduced delta and theta EEG power associating with sleep alterations [94, 95], and in an FTD-tauopathy mouse model EEG alpha power during the wake-state is decreased [96] indicative of an overall EEG slowing related to tau. Electrophysiological slowing has been shown in rTg4510 cortical neurons, with impairments in NREM sleep UP and DOWN states: prolonging of latency and intervals in UP state and of total DOWN state activity [9]; critically, balance of UP and DOWN states is related to the memory consolidation benefits of sleep [65].

In tauopathy and AD models, sleep deprivation increases tau deposition into paired helical filaments in 3xTg AD mice [97], and tau spread in P301S mice, with hippocampal injection of human tau fibrils, including to LC, involved in arousal in the sleep–wake cycle [29, 98]. Finally, Tg4510 tauopathy model mice exhibit tau inclusions in the suprachiasmatic nucleus (SCN) with arrhythmic expression of PER2 and BMAL1 clock proteins [99]. Tau-deficient Drosophila also exhibit circadian rhythm disruption with abnormal activity patterns, impaired neuronal remodeling in pacemaker neurons, and increased circadian clock proteins [35], and Drosophila expressing 4R tau have circadian arrhythmicity and disrupted sleep [100]. In sum, tau pathology and loss of function in AD is exacerbated by sleep disruptions, and in turn impairs sleep via circadian arrhythmicity and impairments in sleep-regulating neuronal populations.

Plasma biomarkers are a potential method for early detection of sleep-related neurodegeneration, but gold standards for disease diagnosis remain to be established. Headway has been made in AD, where Aβ42, Aβ40, Tau-181, Tau-217 and Tau-231 are all showing promise, along with the inflammatory marker glial fibrillary acidic protein (GFAP) and neurodegenerative marker neurofilament light chain (NfL) [101]. The first three track with sleep disorder changes in CSF levels of Abeta and Tau-181, indicating a direct relationship between sleep disorder and neurodegeneration [102]. Apnea is also associated with increased plasma Aβ42/Aβ40 ratio and phosphorylated-tau [102–104]. Biomarkers of neurodegeneration may represent altered sleep regulation including increased plasma orexin A in plasma in AD [105] and reduced plasma melatonin in Huntington’s disease (HD) [106]. Taken together with the aforementioned increase in AQP4 in CSF [43], this suggests dysfunction of sleep-related pathways in NDDs.

While the majority of biomarkers in relation to sleep dysfunction focus on Alzheimer’s-related proteinopathies, there are other sleep-related biomarkers of disease, including plasma metabolomic [107] and lipidomic [108] profiles. Plasma TNF-α and IL-10 were significantly elevated in REM sleep behavior disorder (RBD) (prodromal to PD) relative to age-matched controls and to decreased IL-6/IL-10 and IL-8/IL-10 levels [109]. TAR DNA-binding protein 43 (TDP-43) and chromosome 9 open reading frame 72 (C9orf72) aggregates are also seen in hypothalamic and SCN neurons that regulate the sleep–wake cycle [110–112]. Development of TDP-43 fluid and PET biomarkers is an active area of research [113–116], and may prove beneficial in amyotrophic lateral sclerosis (ALS)/FTD, limbic-predominant age-related TDP-43 encephalopathy (LATE), and in AD. It is expected that the identification of TDP-43 will take prominence alongside synuclein given that NDDs often have multiple proteinopathies [117–121].

Table 1 expands our discussion of the role of NDD proteinopathies in sleep disturbances to α-synuclein, TDP-43, fused in sarcoma (FUS) and Huntingtin (Htt). TDP-43, along with tau, is the predominant proteinopathy in FTD and ALS, both of which present with sleep disturbances [122, 123]. Loss of orexinergic neurons and detection of TDP-43 inclusions has been reported in the hypothalamus of ALS patients; though these inclusions likely occur at later stages (III and IV) when there is widespread pathology [111, 112, 124]. Dipeptide repeat inclusions from expansion in C9orf72, a common genetic cause of ALS, have been observed in pinealocytes and SCN vasoactive intestinal polypeptide (VIP) neurons. These neurons regulate circadian rhythms and did not contain phosphorylated TDP-43 inclusions [110], suggesting TDP-43 may not be a significant driver of ALS-associated sleep deficits; whereas C9orf72, ALS-associated motor and breathing impairments could be better indicators of sleep loss in patients [122]. For behavioral variant FTD, there is evidence for a potential relationship of orexin dysregulation and the sleep disturbances in these patients (reviewed in [123]), yet similar to ALS, further work is needed to delineate the roles of TDP-43, tau and FUS pathology, and their homeostasis (see Table 1), on hypothalamic function and sleep.

LATE and DLB are common mixed pathologies in people with AD and present in 1/3 to 1/2 of patients [117–119]. Interestingly, Lewy body pathology but not TDP-43 associates with sleep impairments in AD patients [119]. This is not surprising given the prevalence of RBD in synucleinopathies such as PD and DLB ([126]; Table 1). The presence of these pathologies may confound grouping AD phenotypes as specific Aβ- and/or tau-driven pathologies, but present an exciting avenue for elucidating predictive factors of patient progression or resilience [151], as biomarkers are developed.

Biomarkers of sleep disruption and risk of NDDs like AD and PD remain underappreciated, despite the potential benefit for earlier intervention and reduction of sleep disruption. The most reliable biomarker of sleep disruption is EEG and identification of sleep stage, resting and wake cycles during a sleep cycle. Further daytime sleeping, including episodes of quiet wakefulness may indicate poor sleep quality. Brain regions with highest soluble and deposited Aβ levels, such as ‘default mode network’, exhibit high neuronal activity during quiet wakefulness [89, 90]. Critically, the usage of EEG in NDD and in preclinical research is a promising approach to define predictive biomarkers of sleep and cognitive dysfunction in an array of NDDs, including AD, FTD and PD [152–158].

Newer wearables have advanced sleep detection and may be useful in monitoring RBD, including in individuals prodromal to PD or DLB [159], with wearables providing reasonable measures of I < O index (comparing nocturnal and diurnal motor activity) specificity (89%) and sensitivity (63%), and of wake bouts (sensitivity = 96%), while EEG identified micro-sleep instability better (sensitivity, specificity > 75%) [159]. New wearables have also been compared to PSG/EEG recordings with accuracies of 0.51 to 0.53 in detecting REM sleep, 0.52 in detecting light sleep, and 0.79 to 0.83 in detecting deep sleep [160], indicating that while there is promise, the algorithms on wearables like the Oura ring still require greater accuracy for use as a diagnostic tool. Nevertheless, such instruments hold promise in identification of diseases and health related activity, including PD [161]. The use of wearables shows promise as the algorithm utilizes additional information from the wearer to determine REM and other sleep stages. For example, during REM, autonomic changes include surges in heart rate and blood pressure, irregular breathing and loss of thermoregulation [162], data that can be measured from the device. The development of digital health and wearables are currently in their early stages but show promise for AD, DLB and PD [163, 164], especially as e-biomarkers of sleep quality. The potential is evident, though a possible limitation in such devices is compliance and use by individuals with cognitive impairment.

One of the earliest regions affected by hyperphosphorylated tau and neurofibrillary tangles, preceding EC, and hippocampal accumulation, is the LC. LC accumulates tau rapidly between Braak stage 0 and I, with almost all remaining neurons containing tau by Braak stage VI [3, 173–175]. Aβ is seen in later stages in the LC [15], suggesting tau as a selective driver of impairments in the LC. A recent report on the localization and morphology of hyperphosphorylated tau (AT8 +) LC neurons indicates the potential of dendritic spread of tau to LC-connected regions from as early as Braak stage 0, especially from the dorsal LC to neocortex and hippocampus [176]. Through connections with the thalamus, cerebral cortex, basal forebrain (BF), hippocampus and hypothalamus, including inhibition of sleep-promoting GABAergic neurons, LC noradrenergic neurons promote arousal and wakefulness and regulate memory [18, 168, 171, 175, 177], concomitant with the early onset of AD sleep deficits.

Noradrenergic firing is low during NREM and quiescent during REM sleep, which is mediated via GABAergic inhibition from hypothalamic ventrolateral preoptic area (VLPO) and median preoptic nucleus, and hypothalamic galanin- (preoptic area (POA)) and melanin-concentrating hormone (MCH)-neuronal inhibition [168, 170, 171]. Other hypothalamic inputs include orexin, indirect activation from wake-promoting histaminergic neurons, and from the SCN via the dorsomedial hypothalamic nucleus [168, 170]. Noradrenergic outputs include inhibition of VLPO neurons, facilitation of orexinergic-mediated wakefulness, and increased cortical pyramidal neuron excitability [168, 170, 171] (Fig. 2). In sum, LC noradrenergic neurons confer a neuromodulatory tone on sleep-circuitry, promoting arousal and shifting the balance towards the wake state. Noradrenergic vulnerability to tau pathology and autophagic failure [178], as well as its importance in sleep and circadian impairments in AD, is an emerging area of NDD research.

Serotonergic signalling has long been known to control sleep–wake circuitry, with broad, neuromodulatory effects throughout a variety of brain regions and neuronal populations (reviewed in [169, 171, 179]; serotonin circuits are of interest in AD (reviewed in [179, 180]. Serotonin neurons are mainly in the dorsal raphe nucleus (DRN) and receive inhibitory afferents which regulate sleep–wake control, including from MCH hypothalamic neurons, and via reciprocal POA(GABA, galanin)-DRN(serotonin) inhibitory activity [168]. Serotonergic neuronal efferents include the thalamus, hypothalamus, cerebral cortex, BF, other brain stem nuclei, and, notably, inhibit REM generating cholinergic neurons in pedunculopontine and laterodorsal tegmental nuclei (PPT/LDT) (Fig. 2). This activity shifts the sleep–wake balance primarily towards wake-state with REM inhibition, though serotonin does exert sleep-promoting effects dependent on the 5-HT receptor subtype [171]. Critically, serotonin activity is reduced through NREM and more so REM sleep [168, 171], similar to what is observed for noradrenaline. In sum, serotonin is a neuromodulator and sleep regulator with intimate linkage to the hypothalamus and cholinergic sleep–wake circuits.

The BF is one of the major hubs of cholinergic neurons which exert broad, neuromodulatory effects. This population of neurons and their efferents are lost in AD forming the core tenet of the cholinergic hypothesis of AD and the therapeutic usage of acetylcholinesterase inhibitors [181]. Critically, BF cholinergic neurons are sleep–wake regulators (as well as regulating other behaviors including memory and attention), with higher activity linked to wakefulness and to REM sleep, and lower activity during NREM [168, 169]. Extensive cholinergic innervation of the cortex indirectly excites pyramidal neurons, with closely linked cortical-BF oscillatory activity especially during wake and REM sleep [169]. This promotes high frequency cortical neuronal activity, and suppresses low frequency, slow delta waves [169]. Notable inputs to cholinergic BF neurons include orexinergic [170], serotoninergic activity with which depolarization or hyperpolarization depends on the 5-HT receptor subtype [171], as well as innervation from glutamatergic neurons in the parabrachial nucleus [166, 168] (Fig. 2). Parabrachial glutamatergic neurons are wake-promoting and provide a major source of arousal from the brain stem [166, 168], and can mediate interoception-related arousal [169]. Other glutamatergic populations have been implicated in the sleep–wake cycle (reviewed in [166, 168, 169]. The parabrachial nucleus-BF-cortex circuit is critical in promoting the wake-state (Fig. 2).

Within the BF, parvalbumin GABAergic and glutamatergic neurons are also wake- and REM-active and interconnected with local cholinergic neurons [168, 182]. Glutamatergic neurons synapse on cholinergic and parvalbumin neurons, and cholinergic connects directly to parvalbumin [168]. Those three neuronal populations are also each inhibited by the neighboring, sleep-promoting somatostatin (SST) GABAergic neurons [168, 182], indicating the complexity of the BF circuitry as a sleep-regulator (Fig. 2).

PPT/LDT cholinergic neurons function similarly to those in the BF, with excitatory efferents on thalamocortical neurons yet notably low cortical innervation [168, 169]. PPT glutamatergic neurons also innervate the BF and have been shown to cause extensive wakefulness upon chemogenetic activation in mice, and more NREM sleep when inhibited [183] (Fig. 2). Furthermore, activation of PPT cholinergic neurons results in reduced EEG slow waves in NREM with an increased light:deep NREM ratio, and activation of GABAergic neurons reduced REM [183].

In the hypothalamus, histaminergic (tuberomammillary nucleus), orexinergic and MCH (lateral hypothalamus) neurons are impacted in AD and accumulate tau pathology [18]. These neuronal networks are critical regulators of the sleep–wake cycle balancing arousal (histaminergic), wakefulness (including orexinergic-based reduced REM and SWS), with induction of sleep from REM-active MCH neurons [18, 184, 185], and receive input from neuromodulatory brain stem nuclei (noradrenaline, serotonin and acetylcholine; [168, 169, 171]. Histaminergic and orexinergic wake-promoting neurons are lost in AD patients, yet MCH neurons seem to be preserved and resistant to tau accumulation [18, 186], indicating a misbalance in hypothalamic sleep–wake control, especially considering orexin-MCH reciprocal inhibition [168] (Fig. 2). In AD patients, orexinergic neurons are reduced while CSF orexin is often reported to be increased, suggesting a complex, perhaps compensatory, mechanism for dysregulated sleep–wake signals (loss of nighttime sleep and increased daytime napping) [184, 187]. Notably, orexinergic neurons decrease with age in rodents [188, 189], with hippocampal- and LC-projecting orexinergic innervation depleted in aged rats and macaques, respectively [190–192]. Further implications of sleep- and wake controlling neuronal impairments in AD with regards to influence of tau are reviewed in [18].

Sleep restriction in rodents leads to robust cell loss: LC, hypothalamus, medial prefrontal cortex, and CA1 and dentate gyrus hippocampal layers exhibit significant loss ranging from a ~ 1/4–1/2 the number of neurons [193–196]. Interestingly, LC noradrenergic, and hypothalamic orexinergic neurons decreased with chronic intermittent sleep loss, but not neighboring MCH neurons, and all 3 neuronal populations exhibited reduced density of axonal projections [196]. These effects were sustained after a 4-week recovery period, with the exception of a restoration of MCH projections to baseline levels, indicating a long-lasting, chronic effect that can impact balance of sleep–wake cycles and potentially be further exacerbated in individuals presenting with AD neuropathology [196].

The SCN is a structure in the anterior hypothalamus which generates behavioral rhythms via afferents on other hypothalamic nuclei, such as the sub-paraventricular zone and dorsomedial hypothalamic nucleus which then relays to the LC to impact arousal [168, 170] (Fig. 2). The majority of SCN neurons are GABAergic with co-expression of hormones including VIP and vasopressin [168, 169]. The molecular clock and light stimuli modulate the activity of SCN neurons which are more active during day and dampened during nighttime [168]. Critically, SCN neuronal loss [197] and prominent tangle formation with minimal plaque pathology [198], has been documented in AD patients, and the SCN has been linked to circadian disruptions in NDDs including AD, PD and HD [199–201]. Given the vulnerability of SCN VIP neurons to inclusions from C9orf72 expansion [110], ALS and FTD may be considered on this list as well [202].

Recent work suggests that beyond the modulatory tone from monoaminergic, cholinergic and orexinergic neurons, the pivotal framework for the sleep–wake cycle arises from fast neurotransmitters such as glutamate and GABA [166, 167]. GABAergic interneurons promote sleep via inhibition throughout the brain [74] but are impaired across AD progression [203]. The VLPO and median preoptic nucleus of POA contain GABAergic neurons that inhibit wake-promoting neurons, including those in the hypothalamus (orexinergic and histaminergic), DRN, LC and parabrachial nucleus. Galaninergic neuronal release from the POA can also inhibit histaminergic and noradrenergic neurons (reviewed in [166, 168]). Input into the VLPO includes inhibition of sleep-promoting neurons from cholinergic, noradrenergic and, less so, serotonergic neurons [168], as well as histaminergic innervation [204] (Fig. 2). Critically, the POA and VLPO in particular, strongly initiate sleep, with lesions of the VLPO in rodents contributing to ~ 40% loss of sleep time [166]. The intermediate hypothalamic nucleus is the likely human homologue of the VLPO, in which AD patients exhibit a loss of galaninergic neurons with the number of neurons significantly associating with the degree of sleep impairment [205]. The parafacial zone of the medulla has also been shown in mice to be a key promotor of SWS and delta wave EEG via GABAergic-mediated inhibition of parabrachial glutamatergic neurons [206]. There are also non-sleep-promoting GABAergic neurons, including the aforementioned BF parvalbumin neurons, as well as lateral hypothalamic GABAergic neurons that can be wake- and REM-active and inhibit sleep-promoting thalamic and POA/VLPO neurons [166, 207–209] (Fig. 2).

Loss of cortical GABAergic tone has been implicated in the impairment of sleep-dominant slow wave oscillatory activity (reduced power without alterations in the oscillatory frequency) in Aβ-driven mouse models, and acute administration of Aβ induces the same electrophysiological impairment [64]. Optogenetic activation of cortical excitatory neurons rescues slow wave oscillations, restores GABA receptor levels and, interestingly, prevents continued accumulation of Aβ plaque [64]. This data demonstrates the dynamic interaction between neuronal circuitry, sleep and neuropathology in AD. During sleep deprivation in mice, hippocampal SST inhibitory interneurons are activated by cholinergic and potentially orexinergic inputs, and thereby suppress local excitatory activity and impair memory consolidation [210]. Nevertheless, inhibitory interneurons demonstrate higher levels of the autophagy activator BAG3, shown to be protective against tau accumulation [211]. It remains to be elucidated how SST autophagy resilience, yet vulnerability to Aβ and tau in AD [211–213], connects to enhanced inhibitory gating which would occur during extended wakefulness [19, 210].

Overall, these sleep-regulating regions and neurons are important to consider in the sleep-proteostasis interaction as their differential vulnerability or resilience to proteinopathy can impact regulation of the sleep–wake balance and may therefore potentiate AD progression and protein accumulation. Future work can examine the connection of these brain regions with bulk protein clearance and these neuronal populations with their susceptibility to failure of cellular proteostasis.

CSF Aβ and tau are lower in the morning after normal sleep; however, after even one-night of sleep restriction clearance is impaired in middle-aged adults (age range: 40–60) without cognitive impairment, noted by increased morning CSF levels for Aβ42, but not Aβ40 or tau [219]. Conversely, morning total tau in plasma increases by 1.8% from evening levels, and, strikingly, by 17.8% after a night of sleep deprivation in young men (~ 22 years old), with no Aβ plasma changes [220]. Furthermore, Aβ ¹⁸F-florbetaben PET signal increases after a night of sleep loss in healthy adults (age range: 22–72), especially in the hippocampus [221], indicating relevance to driving AD progression and memory impairments. PET imaging, especially with tau tracers, may prove critical to examining effects of acute sleep loss on proteinopathy. Work from Holth [29] and Lucey [222] and colleagues, demonstrated ~ 30% increased CSF Aβ and > 50% increased CSF tau in healthy adults (30–60 years old) with one-night of sleep deprivation.

After five consecutive nights of partial sleep deprivation in healthy adults (age range: 20–40), a 27% increase in CSF orexin concentrations was observed, without changes in amyloid, astroglial, or neurodegeneration biomarkers. Notably, there was reduced time spent in all sleep stages except for SWS [85], and we speculate that residual SWS when sleep is only partially deprived promotes protein clearance, protecting against increased amyloid which has been observed after total acute deprivation [219, 221]; though the healthy young and middle-aged adults in these studies likely exhibit greater protein clearance efficiency than aged individuals and AD patients.

Brain clearance of a magnetic resonance imaging tracer was reduced immediately following one-night total sleep deprivation in adults (average age ~ 42), and persisted even after sleep was restored [223]. This deprivation likely reduced glymphatic brain clearance which is regulated by neuronal activity, and enhanced during SWS [27, 224, 225]. Glymphatic clearance is one of two mechanisms of interstitial fluid (ISF) clearance from the brain which likely work in consort, the other being the periarterial drainage pathway [226, 227]; though the relationship of glymphatics with sleep is undeniable and therefore is a critical mechanism to consider in the sleep-proteostasis axis and for the clearance of Aβ and tau [27, 29, 225, 228, 229].

Glymphatics is a pathway of bulk fluid exchange in which CSF is pumped into the brain from the subarachnoid space, first along the cortical pial arteries. Arterial vasomotive forces move CSF into deeper brain regions in the periarterial Virchow-Robin space, and its transport across the blood–brain barrier (BBB) is mediated by AQP4 channels on the endfeet of astrocytes which encapsulate the brain vasculature. Following CSF-ISF exchange, fluid efflux occurs along the perivenous space to the dural lymphatic system, facilitating the brain clearance of extracellular metabolites and solutes [224]. During SWS, the interstitial space increases by 60% supporting higher rates of glymphatics-mediated Aβ clearance [27], yet during wake or sleep deprivation clearance is reduced with 90–100% higher ISF tau in mice [29] (Fig. 3a). Glymphatic CSF influx significantly correlates with neuronal signals of SWS: low frequency, high-amplitude delta waves and a lower heart rate; whereas a negative correlation was observed with high-frequency beta waves in mice, common in wake state, and no significant relationships with alpha or theta waves, present in wake-state, REM and NREM stage 1 sleep [74, 230].

These results demonstrate the intimate relationship of glymphatic-mediated clearance of Aβ and tau with sleep, and the positive-feedback-loop that can occur as sleep is lost in AD. Mounting vascular impairments occur in AD, including cerebral amyloid angiopathy, vessel tortuosity and rigidity, reduction in cerebral blood flow, and enlargement of periarterial and perivenous spaces; all of which impair glymphatic clearance, and are further impaired by the failure of glymphatic clearance with an abundance of Aβ and tau [226, 227] (Fig. 3a).

Interestingly, diurnal fluctuations in glymphatic clearance show regional differences in the rat brain, with large sleep-associated increases in brain regions involved in circadian rhythm regulation, including the SCN and lateral hypothalamus [231]. Critically, acute sleep deprivation significantly enhances Aβ and tau which can further impair sleep, and lead to persistently impaired protein clearance [219–221, 223].

One of the earliest features of disrupted autophagic flux in AD is an abundance of autophagosomes, which accumulate in neuronal cytoplasm, axons, and most prominently in dystrophic neurites. There is a failure to clear autophagosomes, preventing degradation of Aβ, tau and other proteins within [235]. As autophagic deficits mount, so does neuronal damage, contributing to AD progression [1, 34, 236]. Furthermore, uncleared autophagosomes and autolysosomes accelerate the cell-to-cell spread of Aβ and tau via exosomal release [79–84]. Interestingly, disrupting sleep recapitulates AD-related autophagic dysfunction [237–239], and knocking out autophagic function is sufficient to recapitulate a neurodegenerative phenotype [240, 241].

Genetic manipulation of autophagic flux in drosophila has demonstrated the bidirectional link between sleep and autophagy. During sleep, autophagosomes drop compared to wake periods. Blocking autophagosome formation increases sleep, whereas blocking autolysosomal degradation decreases sleep [239]. In mice, acute sleep deprivation causes autophagosomes to accumulate in hippocampal neurons, with increased expression of LC3B, Beclin-1 and p62, indicative of recruitment of autophagic processes [1, 237]. Autophagic flux is disrupted in mice with chronic sleep fragmentation, with increased autophagosomes, endosomes, and number and size of intracellular lysosomes [238]. This is accompanied by spatial learning and memory impairments after acute and chronic sleep deprivation [237, 238]. Therefore, autophagic recruitment without lysosomal fusion and protein degradation occurs with mounting sleep impairments, furthering proteinopathy and disease progression (Fig. 3b, ALP).

Interestingly, Xie and colleagues report cortical and hippocampal Aβ intracellular accumulations in chronic sleep fragmented wild-type mice, akin to observations in AD models [29, 238, 242]. The authors attribute this to a failure of normal amyloid precursor protein (APP) processing via the endosome-autophagosome-lysosome (EAL) pathway in which reduced flux through the EAL pathway decreases APP trafficking and clearance, facilitating amyloidogenic processing and Aβ accumulation [238, 243, 244]. These results are exciting as they demonstrate a potential mechanistic linkage between sleep loss and AD proteinopathy.

Circadian rhythms and the sleep–wake cycle impact gene expression [217] and are linked to AD pathology. AD patients exhibit circadian arrhythmicity, measured in the clinic by activity/rest cycles, exhibiting delayed cycle phases and lower-amplitude peaks (reviewed in [245]). In aged mice with a knock-out of the biological clock gene Per1, autophagy is impaired, and Aβ42 and presenilin levels are higher [246, 247]. Per genes (including Per1 and Per2) encode the PER1 and PER2 proteins, often referred to collectively as PERIOD protein, which are essential regulators of the timing of circadian oscillations [248–251]. Critically, the circadian clock regulates expression of autophagic genes [26, 252–254], including transcription factors Nr1d1 and C/EBPβ which promote rhythmic expression of autophagy activators, including Beclin-1 and atg4a [26, 252], triggering autophagosome formation. In the mouse hippocampus, autophagosome-related LC3-II, but not cytoplasmic LC3-I, exhibits a circadian rhythmicity, indicative of higher autophagosome formation and autophagic flux in the sleep-associated light phase as opposed to the dark phase [28]. This differs from autophagosome over-accumulation in AD, in which there is a failure of autophagic flux and protein clearance. Notably, the ALP is regulated on a circadian cycle, and arrhythmicity in AD can further the already present autophagic deficits, akin to what occurs with sleep disturbances (Fig. 3b, ALP).

After chronic sleep fragmentation in mice, overall light phase peaks and dark phase lows of hippocampal autophagic flux remain, but the rhythmicity is impaired with abnormal changes within each phase [28]. Acute recovery sleep is insufficient to reverse these changes, and triggers increased Beclin-1 expression during the sleep-associated light phase [28]. Two conclusions can be drawn from this work to frame our understanding of the sleep-proteostasis interaction in AD. Firstly, sleep loss disrupts circadian rhythmic-regulation of autophagic flux, leading to increased Aβ and tau aggregation. Secondly, these deficits persist after recovery, including the continued, aberrant activation of an already overwhelmed autophagic-lysosomal system in AD by Beclin-1, indicating chronic complications for AD progression.

Normally tasked to clear misfolded protein, the UPR is overwhelmed in NDDs, and further exacerbated when sleep is disrupted [1, 33]. Protein kinase RNA-like ER kinase (PERK), inositol-requiring enzyme 1α (IRE1), and activating transcription factor-6 (ATF6α and β) are three crucial proteins that help initiate UPR and from which the binding immunoglobulin protein (BiP) is unbound in response to misfolded proteins. Chronic activation of UPR, especially on the PERK branch, occurs in the brains of AD patients and other tauopathies. AD brains have a heightened expression of UPR activation markers, including phosphorylated PERK, eukaryotic initiation factor 2α (eIF2α), and IRE1, as well as BiP, which correlate with Braak stages [255–258], suggesting that UPR has a bearing on AD pathology at an early stage. There is an increase in UPR in tauopathies; and UPR associates with early hippocampal tau pathology in these disorders [259]. Furthermore, the same neurons and glia that display abnormal tau phosphorylation levels show a corresponding increase in markers of UPR activation as well, corroborating their linkage [259].

UPR chaperones are upregulated during wakefulness, or when sleep is deprived, to help mitigate endoplasmic reticulum (ER) stress by clearing misfolded proteins and reducing protein translation [21, 33]. The ER chaperone BiP binds to misfolded proteins to prevent aggregation and promote re-folding. In the mouse cortex, BiP levels increase progressively as sleep is deprived [216], as a compensation to accumulation of uncleared protein. This phenomenon has been observed in drosophila as well, in which BiP levels rise during sleep loss and fall towards baseline throughout recovery sleep [218]. Increasing normal BiP expression, or of a dominant negative form which decreases BiP function, prolonged or reduced sleep recovery after sleep deprivation, respectively [218]. BiP overexpression slows UPR function [218, 260], which is suggestive of how chronic sleep disruption in AD patients can lead to a state in which the UPR is overwhelmed by abundant misfolded protein (Fig. 3b, UPR). These genetic manipulations had no effect on baseline sleep in the flies [218]. These data are indicative of a parallel UPR activation to mitigate protein accumulation during sleep impairment that is intimately linked to sleep behavior [216, 218].

Interestingly, acute sleep deprivation in young mice promoted protein clearance and reduced translation via UPR, but in aged mice led to pro-apoptotic signalling [261]. Sleep and protein quality control are both impaired with aging [33, 262] resulting in reduced efficiency of refolding aspects in UPR [261, 263–265]. This is evident from the undersupply in chaperone proteins in age related diseases and in aged wild-type rodents [261, 266, 267], corroborating the impact that sleep quality and protein homeostasis have on each other [33]. UPR is activated in orexinergic and noradrenergic wake active neurons with increases in phosphorylated-PERK. This occurs to a greater degree in aged mice correlating with a decline in orexinergic and noradrenergic neuronal activity, with nuclear translocation/activation of CCAAT/enhancer binding protein homologous protein (CHOP) [268]. CHOP is known to signal apoptosis in response to ER stress [269, 270] and to mediate sleep apnea/hypoxia-related cellular stress and injury [271, 272].

The stress response and quality control of the protein homeostatic system becomes dysfunctional with almost all tissues of aged candidates [261, 263, 265], reviewed in [273], demonstrating a complex interaction in AD and other neurodegenerative disorders wherein age, impaired proteostasis and sleep exert individual and synergistic impacts on disease progression. Importantly, UPR chaperone-mediated treatment restores aging related sleep and cognitive impairments in mice [274]. Another UPR therapeutic target for NDDs is eIF2α which attenuates global protein synthesis rates critical for memory and neuronal function, when it is phosphorylated by phosphorylated-PERK; phosphorylated-eIF2α levels are elevated in Alzheimer’s disease and other NDDs [275, 276]. In vitro screening assays of clinically suitable therapies identified that the anti-depressant trazodone hydrochloride, and one other compound (dibenzoylmethane), reversed the effects of phosphorylated-eIF2α on protein synthesis. In vivo treatment of either compound normalized translation and was neuroprotective in prion and tauopathy models [276]. Critically, trazodone can be used to treat insomnia and improve sleep maintenance [277], further indicating the therapeutic potential of the sleep-proteostasis interaction (reviewed in [278]).

Persistent upregulation of UPR is detrimental after extensive proteinopathy and sleep loss, interrupting protein synthesis, promoting neurodegeneration and exacerbating defective sleep-proteostasis positive-feedback-loop [1, 21]. If proteins cannot be appropriately folded in the ER lumen, site of protein synthesis and packaging, proteins are directed to the proteasome to avoid aggregation [279, 280]. Finally, mild ER stress can precondition a neuroprotective mechanism for UPR via recruitment of autophagic processes [1, 281]. However, with prolonged stress like chronic sleep deprivation, UPS, and ALP recruitment to compensate for UPR failure may further overwhelm cellular proteostasis.

The interaction between UPS and sleep disruptions is not as well defined as with autophagy and the UPR. There is evidence that OSA reduces proteasomal activity via intermittent hypoxia [282], which contributes to proteasomal dysfunction after neurodegeneration [1, 283]. With mounting UPS deficits, ubiquitination and degradation of PER1 and PER2 proteins, biological clock proteins and circadian length-regulators, become unbalanced, lengthening circadian rhythms [284] (Fig. 3b, UPS). Tau accumulation is intimately linked with proteasomal dysfunction [285]; although research is needed to elucidate if there is a defined tau-UPS-sleep component contributing to proteinopathy and sleep loss in AD. Importantly, dysfunction in UPS causes compensatory activation of autophagic processes, which are already aberrant in AD [1, 286, 287], leading to chronic deficits in the sleep-proteostasis axis.

Trazodone hydrochloride is a unique therapy on this list due to hitting multiple targets on the sleep-proteostasis axis. Trazodone is an antidepressant therapy approved for use in major depressive disorder, and is commonly used for off-label treatment of patients with sleep disorders, and in AD. Multiple neurotransmitter systems are affected by trazodone treatment, including selective agonism and antagonism of serotonin receptors, serotonin transporter inhibition and reuptake inhibition, antagonism of histaminergic and α1 and α2 adrenergic receptors, and, to a lesser degree, trazodone exhibits anticholinergic activity (reviewed in [346]). In sum, the neuromodulatory effects of trazodone can shift neuronal control of the sleep–wake balance towards sleep (Figs. 2 and 4).

In a phase 3 clinical trial, AD patients with sleep disturbances were treated with 50 mg of trazodone once per day before bed over a 2-week period. Treatment significantly increased sleep time and trended to reduced arousals and wake periods throughout the night, without exacerbating daytime napping or EDS. With this short-term treatment, cognition was unaffected ([335]; Table 2). However, evidence suggests long-term recorded trazodone use (median dose: 50 mg/day) in AD patients with sleep disturbances slows cognitive decline on mini-mental state examination, over a 4-year follow-up [347]. At a higher dose of 200 mg, trazodone is safe and feasible in AD patients, and is within the predicted range for regulating UPR (reviewed in [276, 348]). Although the mechanism of the trazodone-UPR interaction is through normalizing protein synthesis rates [276], Aβ and tau degradation and clearance may be indirectly improved through mitigating neurodegenerative burden, and reducing UPR overactivation which will improve cellular homeostasis (and proteostasis).

Further work is necessary to identify how trazodone can impact AD progression, and at what dose, including potential benefits on protein clearance and neuronal function. Currently, a phase 2 clinical trial of trazodone in MCI is recruiting (as of February 2023) to test the effect of a 4-week, 50 mg treatment on SWS quantity and quality, sleep onset and fragmentation, cognition, as well as hippocampal neuronal activity by functional magnetic resonance imaging (ClinicalTrials.gov Identifier: NCT05282550; Table 2).

Agonists of adrenergic α receptors impact the sleep–wake cycle via a reduction in locus coeruleus noradrenergic activity, decreasing arousal and promoting sleep [349]. Dexmedetomidine is an α2 adrenergic receptor agonist that has demonstrated benefit as a therapeutic in preclinical in vitro and in vivo models including anti-inflammatory, pro-cognitive and pro-neurotrophic effects [350–354], despite association with increased tau phosphorylation (i.e., at pS202, pT205, pT231, pS396) in cells and rodents lasting up to 6-h post treatment [355, 356]. Dexmedetomidine is used clinically as an analgesic and sedative, due to an induction of NREM stage 2-to-3 sleep, including increased slow wave delta oscillations and spindles in EEG [321–323]. One potential benefit to the sleep-proteostasis interaction is that dexmedetomidine acutely enhances glymphatics [323] (see Table 2), but its usage as a primary therapeutic for AD remains limited and the potential for exacerbating tau pathology is of concern.

Currently, the most promising sleep modulating therapy in AD is the orexin antagonist suvorexant, which is used in the treatment of insomnia [304] and was approved in February 2020 for treatment of sleep disorder symptoms in AD patients. Orexin signaling is also reported to regulate cognition, and in AD there is orexinergic dysfunction that occurs with elevated CSF orexin [184, 187]. Suvorexant competes with orexin for binding to receptors OXR1 and OXR2, promoting sleep [304, 357, 358]. In a phase 3 trial to treat insomnia with mild-to-moderate AD, 4 weeks of daily oral treatment with suvorexant significantly improved total sleep time by 73 min more than baseline, (45 min in the placebo group); and treatment did not worsen cognition and was well-tolerated [308]. Recently, Lucey and colleagues demonstrated that acute suvorexant treatment in healthy middle-aged adults decreased CSF levels of Aβ by 10–20% and p-tau (T181) by 10–15%, but not at other p-tau isoforms – S202 or T217 [359].

Suvorexant may prove beneficial in combination therapies, potentially with the recently approved aducanumab or lecanemab, and to probe the sleep-proteostasis axis. Promoting sleep with suvorexant in preclinical AD models can be utilized to assess alterations in specific neuronal vulnerability to tau and Aβ, and potential improvements in glymphatic clearance and autophagic flux. Modulation of orexinergic function, although limited to a small cellular population, may exert beneficial network effects throughout the brain by promoting sleep, helping to balance circadian arrhythmicity, and therefore mitigating deleterious changes in proteostasis. Rebalancing specific neuronal population function (i.e., orexinergic) can potentially correct a network-wide dysfunction seen in AD; for example, suvorexant treatment enhances hippocampal long term potentiation in AD model mice [360].

Characterizing proteostasis improvement from suvorexant treatment in sleep- and AD-associated neurons and regions, in hypothalamic orexin neurons, as well as EC-hippocampal cognitive circuits (Fig. 4), will help define the sleep and proteostasis interaction across disease progression, and inform efficacy of treatment based off disease staging.