Sleep disturbances are very common in Alzheimer’s disease (AD) and represent a major clinical problem (Pollak and Perlick, 1991; Moran et al., 2005). These symptoms appear early, and individuals with prodromal AD frequently have pronounced changes in sleep architecture and reduced sleep time (D’Rozario et al., 2020). Moreover, increasing evidence supports that sleep changes might be present in preclinical AD (Yaffe et al., 2014; Leng et al., 2019). A recent meta-analysis showed that sleep duration was the most consistent non-cognitive predictor of cognitive deterioration (Hudon et al., 2020).

A bidirectional link between chronic sleep disturbances and AD has been formulated (Ju et al., 2014). Animal and human experiments demonstrate the accumulation of amyloid-β (Aβ) after sleep deprivation (Lucey and Holtzman, 2015; Lucey et al., 2018). On the other hand, Aβ deposition perturbs the sleep-wake cycle (Lucey and Holtzman, 2015). Therefore, disrupted sleep would lead to the accumulation of amyloid in the brain which in turn would affect sleep, establishing a vicious circle that enhances the amyloid accumulation, increasing the risk of dementia (Ju et al., 2014). However, sleep decline is almost universal in elders (Ohayon et al., 2004), and to clarify its role as a clinical biomarker it is important to understand to what extent sleep disturbance is related to early signs of neurodegeneration. One way to address this issue is to correlate sleep parameters with pathophysiology proxies, such as Aβ or tau protein, the two hallmarks of AD, in cognitively healthy individuals. Several studies in elders have found associations between sleep and Aβ, determined by CSF levels or more frequently by PET scan (Ju et al., 2013; Spira et al., 2014; Sprecher et al., 2015; Branger et al., 2016; Brown et al., 2016; Varga et al., 2016). More recently, studies have explored the relationship between tau and sleep (Kam et al., 2019; Lucey et al., 2019). The majority of these studies have used self-reported questionnaires to evaluate sleep (Spira et al., 2014; Sprecher et al., 2015; Branger et al., 2016; Brown et al., 2016). Less frequently, sleep was assessed by one-night in-lab polysomnography (PSG) (Varga et al., 2016).

We are still far from incorporating sleep to AD risk prediction. First, we need more studies to clarify the relationship between the AD core biomarkers and sleep. In such studies, it is important to use accurate and objective sleep measurements but at the same time, these measurements should be scalable to allow for broad clinical use. Another essential point is to take into consideration possible effect modifiers or confounders, as the underlying mechanisms are poorly understood and might be modified by such factors as sex, depression symptoms, medication use, and very importantly, by the APOE ε4 allele. Several publications have reported that APOE ε4 affects sleep in cognitively healthy individuals (Drogos et al., 2016; Kahya et al., 2017) and that it might also play a modulatory effect between sleep and AD pathology (Lim et al., 2013; Burke et al., 2016; Hwang et al., 2018).

We hypothesized that sleep is associated with early neurodegenerative changes in cognitively healthy elderly individuals and that APOE might modulate this relationship. Therefore, we aimed to test whether total sleep time, measured by actigraphy during 1 week, was related to CSF AD core biomarkers (Aβ42/40, phosphorylated-tau-181, and total-tau) in elderly individuals, considering possible confounders and effect modifiers. We specifically aimed to explore the potential interaction effect between the APOE ε4 allele and sleep on AD pathology.

Out of the 127 participants, there were 70.1% females, and the mean age was 65.5 ± 6.3. The distribution of education levels was 34.5% tertiary, 38.8% secondary, and 26.7% primary. The average MMSE was 28.8 ± 1.5, and the FCSRT delayed cued recall score 14.8 ± 2 out of 16. The proportion of APOE ε4 carriers was 30.7%, and 37.8% had at least one positive AD core CSF biomarker. The average total sleep was 442.24 ± 73 min (7.37 h). According to the “ATN” classification, 62.2% of the individuals had all three biomarkers negative and 29.1% were in the Alzheimer’s continuum: 12.6% have amyloid positive biomarkers and p-tau and t-tau negative, and 16.5% had all three biomarkers positive. Three individuals were amyloid positive, p-tau positive and t-tau negatives, and only one individual was amyloid positive, p-tau negative and t-tau positive; all of them had borderline values for both tau markers, thus, these four individuals were pooled with those with all three positive markers. Finally, 8.7% were t-tau positive in the absence of Aβ (Table 1).

The most important finding of our study was the inverse correlation between total sleep time estimated by a commercial actigraphy watch and tau pathology in a cognitively healthy population. Our results are in line with previous observational studies linking sleep duration with cognitive impairment and dementia risk (Yaffe et al., 2014). These studies have been recently reviewed in a meta-analysis showing that both short and long sleep duration was consistently associated with worse cognitive performances (Hudon et al., 2020). Therefore, it has been proposed that the association between sleep time and the risk of cognitive deterioration is V-shaped. However, the fact that most of these studies employed self-reported assessments of sleep time, and evaluated individuals at different stages in the AD continuum, may have contributed to some inconsistency in the results (Tworoger et al., 2006; Schmutte et al., 2007; Faubel et al., 2009; Blackwell et al., 2011; Ramos et al., 2013; Yaffe et al., 2014). Our analysis using objective estimation of total sleep and core AD CSF biomarkers as a proxy for dementia risk, strongly supports that those cognitively healthy elderly individuals with lower sleep time are potentially at greater risk of future neurodegenerative diseases based on higher t-tau and p-tau levels in the CSF.

Despite the evidence of the role of Aβ as a potential link between sleep and AD, we were not able to find an association with Aβ CSF levels. Most studies analyzing sleep and Aβ have used amyloid PET and employed self-reported sleep assessments (Spira et al., 2014; Sprecher et al., 2015; Branger et al., 2016; Brown et al., 2016; Carvalho et al., 2018); finding statistical associations with a variety of sleep-related variables like sleep latency (Branger et al., 2016; Brown et al., 2016), daytime sleepiness (Carvalho et al., 2018), or self-reported less adequate sleep (Sprecher et al., 2015, 2017); only Spira et al. (2014) found an association with sleep time, but was self-reported as well. The use of CSF Aβ analysis instead of amyloid PET might also explain differences in results. Even though it has been shown that CSF amyloid levels (Aβ40 and Aβ42) increase in humans after sleep deprivation (Lucey et al., 2018), we must consider that in preclinical AD Aβ42 levels decreases in comparison to Aβ40. To our knowledge, our study includes the largest sample assessing the relationship between an objective estimation of sleep time and Aβ in CSF in cognitively healthy individuals. In line with our results, those studies that used actigraphy to determine sleep time objectively did not find an association with Aβ (Ju et al., 2014; Lucey et al., 2019). On the other hand, perhaps a determination of sleep components, such as slow-wave activity (SWA), might have increased our chances to detect an association with Aβ, as reported before (Varga et al., 2016; Kam et al., 2019; Liguori et al., 2020), although a recent study, with a sample size similar to ours, did not find an association between SWA and Aβ levels (Lucey et al., 2019).

Our data would suggest an early association between tau and sleep independent of Aβ in cognitively healthy individuals. Prospective studies using PET support that tau and amyloid, initially have separate trajectories, although at some point they interact and amyloid drives tau extension beyond the medial temporal lobes (Schöll et al., 2016). Therefore, in AD the deposits of these two proteins follow different patterns, and contrary to the Aβ pathology that usually starts in the cortex, tau accumulates first in the brainstem and subcortical regions (Serrano-Pozo et al., 2011). Recent neuropathological studies have shown that subcortical brain areas implicated in sleep-wake regulation, are very vulnerable and early affected by AD tau neurodegeneration (Oh et al., 2019). In a longitudinal study, it was found that uninterrupted sleep was associated with less neurofibrillary tangles load in the APOE ε4 carriers, but there was no association with Aβ pathology (Lim et al., 2013). Finally, our study is in line with recent publications showing the association between tau pathology and several sleep neurophysiological parameters like SWA (Lucey et al., 2019) and the density of spindles at non-rapid eye movements (NREM) stage 2 (Kam et al., 2019). Therefore, sleep changes in individuals at preclinical stages of AD might correlate with early tau brainstem pathology, and as a sort of “canary in a coal mine,” sleep disturbances might be an early AD biomarker (Iranzo et al., 2009; Bombois et al., 2010). Alternatively, chronic poor sleep might be causing higher tau levels and potentially increasing the risk of neurodegeneration. The same research group that has pioneered studies linking amyloid dynamics to sleep has recently discovered that tau levels are also strongly influenced by the sleep-wake cycle. In mice, tau levels in interstitial fluid and CSF increased around 90% during wakefulness and 100% during sleep deprivation. In human CSF tau increased 50% after sleep deprivation. And these findings appeared to have pathological significance, as in a tau seeding and spreading mouse model chronic sleep deprivation increased tau pathology spreading (Holth et al., 2019). Nevertheless, both mechanisms are compatible, and there could also be a two-way association between tau and sleep.

Our second main finding was the significant interaction between sleep and APOE ε4 carriers, which together with the stratified analysis indicates that the relationship between sleep and tau was predominantly due to the APOE ε4 individuals. This is in line with recent publications indicating that ApoE plays a fundamental role in tau pathology. It has been demonstrated that ApoE affects tau pathogenesis, neuroinflammation, and tau-mediated neurodegeneration independently of Aβ pathology (Shi et al., 2017). Additionally, Koch and colleagues have shown that only in APOE ε4 AD patients there is a strong association between cognitive decline, impaired cortical plasticity and CSF t-tau values, suggesting a relative susceptibility to APOE ε4 patients to the amount of tau-related pathological burden (Koch et al., 2017). Our results are in line with previous studies suggesting the role of APOE as a modulator of the relationship between sleep and AD pathology. In individuals with the APOE ε4 allele, less sleep consolidation was related to higher AD incidence and tau pathology load in the autopsy (Lim et al., 2013). We also found that, as reported before (Drogos et al., 2016), carriers of APOE ε4 slept significantly less time than non-carriers. As a potential explanation, it has been proposed that APOE ε4 individuals might have more respiratory problems during sleep (Gottlieb et al., 2004), supporting this hypothesis, several studies have found higher levels of tau pathology in sleep apnea patients (Liguori et al., 2017; Carvalho et al., 2020). Additionally, cognitively healthy APOE ε4 carriers had lower gray matter volume if they suffered from insomnia (Grau-Rivera et al., 2020). In our study, we did not find an association between CSF biomarkers and sleep apnea; however, we only assessed sleep respiratory problems based on clinical records.

Our study has explored the clinical applicability of sleep changes as potential early biomarkers of neurodegeneration. Our data suggest that a simple sleep parameter, such as total sleep time estimated by an actigraphy watch, is a valid predictor of tau pathology with potential clinical use. In contrast, self-assessment with a sleep questionnaire did not yield any significant association with AD CSF biomarkers. These devices allow complex and variable data to be monitored over long periods, enabling more ecological observation than laboratory assessments, as PSG, which can compensate for the lower accuracy of the estimate. Importantly, this is an inexpensive and scalable technology that could be incorporated to risk prediction algorithms together with other readily available information. However, studies tracking sleep changes during more prolonged periods, and the use of more accurate portable devices that provide information about the architecture of sleep, could significantly improve the diagnostic performance of sleep assessment as an AD screening method. Longitudinal studies may also help to understand better the direction of this association and the potential value of preventive measures.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author/s.

The studies involving human participants were reviewed and approved by the Comité de Ética de Investigación Clínica de Cantabria (CEIC). The patients/participants provided their written informed consent to participate in this study.

SL-G: collection of the clinical data and drafting of the manuscript. CL: collection of the clinical data and critical revision of the manuscript. AP and MG-M: neuropsychological assessments. MK, AF-R, and MB: collection of the clinical data. LR-G: collection of the sleep data. JI and ML-H: CSF analytical procedures. ER-R: critical revision of the manuscript. PS-J: study concept and design, data analysis, drafting of the manuscript, and funding acquisition. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.