Introduction

Sleep disruption is associated with increased risk of Alzheimer’s disease (AD) in epidemiological studies1,2,3, while biomarker studies in cognitively-intact participants suggest that short sleep duration and poor sleep quality are associated with greater AD-related amyloid β (Aβ) and tau pathology4,5,6,7,8,9, even before the onset of clinical symptoms of cognitive and functional decline. Yet the mechanistic link between sleep disruption and AD risk remains undefined. The glymphatic system is a brain-wide network of perivascular pathways along which cerebrospinal fluid (CSF) and interstitial fluid (ISF) exchange, supporting the clearance of interstitial solutes from the brain into the CSF, and thence into the blood. Studies in mice demonstrated that glymphatic function contributes to the clearance of Aβ and tau, is more rapid during sleep, and is impaired by acute sleep deprivation10,11,12,13,14,15. While more rapid glymphatic clearance of solutes during sleep has now been validated in the human brain16,17, whether sleep-active glymphatic activity contributes to the clearance of Aβ and tau in a healthy human population is unknown.

Plasma biomarkers have emerged as a potentially critical tool enabling the non-invasive diagnosis of AD at a large scale within the wider population. Immuno-assay and mass spectrometry-based measures of different Aβ and tau species permit the assessment of AD-related Aβ plaque burden and tau pathology, even during preclinical stages of disease in the years prior to patients’ clinical progression to cognitive impairment and dementia18,19. Although generally regarded as steady-state indicators of brain AD pathological burden, plasma AD biomarker levels are impacted acutely by sleep disruption4,20. While it is possible that these effects reflect reduced rates of cellular Aβ and tau release during sleep, they may also reflect the role of sleep-active glymphatic clearance in the transport of Aβ and tau from the brain to the blood. In a crossover study including 5 healthy participants, plasma Aβ40, Aβ42, non-phosphorylated tau181 (np-tau181), non-phosphorylated tau217 (np-tau217) and phosphorylated tau181 (p-tau181) were each reduced during the overnight period by acute sleep deprivation4. Noting an increased ratio of CSF:plasma Aβ and tau levels, the authors inferred that the processes governing the clearance of Aβ and tau from the brain to the CSF, and thence from the CSF to the plasma must be impaired by acute sleep deprivation. These findings are consistent with the sleep-active glymphatic clearance of Aβ and tau from the brain, yet they do not provide causal support. A second study showed that plasma Aβ and tau levels were reduced in participants that exhibited MRI evidence of impaired glymphatic and CSF-to-plasma clearance21. These findings established a correlation between steady-state AD biomarker plasma levels and impaired CSF circulation, but did not test whether sleep-active glymphatic function directly contributes to the clearance of these peptides during sleep in otherwise healthy individuals.

Here, we designed a study to directly test if sleep-active glymphatic function contributes to the overnight clearance of Aβ and tau from the human brain. We first developed a compartmental pharmacokinetic model to predict the effect that sleep-associated changes in Aβ and tau release and sleep-active glymphatic clearance, reflected by transport between the brain ISF and CSF, would have on morning plasma levels of Aβ and tau species. We then carried out a clinical randomized crossover study where participants underwent overnight normal sleep and sleep deprivation. Prior studies in rodents and humans demonstrate that glymphatic function increases with increasing sleep electroencephalography (EEG) delta and theta power, reduced EEG beta power, reduced heart rate (HR), reduced central noradrenergic tone, and reduced brain parenchymal resistance to fluid flow (RP), and vasomotor oscillations13,17,22,23,24,25. Measuring these key determinants of glymphatic exchange with an investigational device17, we defined the relationship between sleep features measured by EEG, RP measured by dynamic impedance spectroscopy, and HRV and cerebrovascular compliance measured by impedance plethysmography (IPG) and overnight changes in plasma AD biomarker levels. Using this approach, we directly tested the hypothesis that sleep-active glymphatic clearance elevates morning plasma AD biomarker levels, despite a sleep-associated reduction in their production.

Method

All studies were performed between October 2022—June 2023 and were reviewed and approved by University of Florida Institutional Review Board (IRB No. 202201364, Villages study) and Western Institutional Review Board (IRB No. 20225818, UW study). The studies have been registered at ClinicalTrials.gov (https://clinicaltrials.gov/study/NCT06060054 and https://clinicaltrials.gov/study/NCT06222385). Written informed consent was obtained from all study participants during a screening visit, prior to any study activities. Studies were carried out in accordance with the principles of the Belmont Report. The Villages Study enrolled 34 healthy participants 56–66 years of age. The UW Study enrolled 14 healthy participants 49–63 years of age. Participants were excluded if they had cognitive impairment or clinical depression. Cognitive impairment was assessed using the Montreal Cognitive Assessment (MoCA <26) and depression was evaluated using the 15-item Geriatric Depression Scale (GDS > 4). Participants with a self-reported history of diabetes, hypertension, coronary artery disease, pulmonary disease, neurological disease, depression, or anxiety were excluded from the study. Exclusion also applied to participants planning travel to alternate time zones within two weeks of study participation, as well as those with a formal diagnosis of any sleep disorder (e.g., sleep apnea requiring positive airway pressure [PAP] therapy, insomnia, restless leg syndrome, circadian rhythm sleep disorder, or parasomnia). Participants were excluded if they had taken any prescribed or over-the-counter stimulants, sleep aids, or psychiatric medications, including antidepressants, within the past 30 days. During intake, participants were further instructed to abstain from these sleep-altering substances throughout the study period and to avoid caffeine consumption after 12:00 p.m. on the day of the study visit. Compliance with both instructions was confirmed upon arrival at the study visit. Exclusion also applied to individuals who consumed more than 400 mg of caffeine per day; female participants who consumed more than 3 alcoholic drinks on any day or more than 7 drinks per week; and male participants who consumed more than 4 drinks on any day or more than 14 drinks per week. To minimize preparation bias (i.e., altering sleep schedules or taking naps in anticipation of sleep deprivation visits), participants were not informed of their initial visit assignment until 4:00 PM on the day of arrival. Sleep quality prior to and between study visits was subjectively evaluated based on investigator assessment.

Compartmental pharmacokinetic model relating glymphatic exchange to plasma AD biomarker levels

We first developed a compartmental pharmacokinetic model to define the effects that sleep-related changes in interstitial solute release and clearance would have on overnight changes in plasma levels of Aβ and tau. The model (shown in Fig. 1A) is a simplification of prior models26 and includes the minimal number of compartments, transport and elimination pathways needed to define the effect that a change in overnight interstitial solute production/release and glymphatic clearance between the ISF and CSF would have on plasma Aβ and tau levels. The model included a 16-hour period of wake, and either an 8-hour period of sleep or an 8-hour period of sleep deprivation. It includes 6 compartments: Cellular Site of Production/Release (1), Free ISF (2), CSF (3), Plasma22(4), Cellular Site of Uptake/Degradation (5), and Non-Monomeric Pools (6). Solute transport processes and their rate constants for Aβ and tau species based on literature values26,27 are provided in Table 1.

Fig. 1: Six-compartment model of brain to plasma solute exchange.

A Amyloid β (Aβ) and tau are produced in neurons (1) and released into the ISF compartment (2) where they are cleared from the ISF via local cellular uptake (2 → 5) and degradation, blood-brain barrier efflux (2 → 4), or glymphatic efflux to the CSF (2 → 3). Aβ42 and phosphorylated tau species are prone to non-monomeric aggregation, unlike Aβ40 and non-phosphorylated tau. Their monomeric forms can also be cleared from the ISF through further aggregation into non-monomeric structures (2 → 6). CSF solutes may recirculate back into the brain interstitium (3 → 2) or be cleared by CSF efflux pathways to the plasma (3 → 4) from whence peripheral degradation occurs. B–C Compartment concentrations of monomeric amyloid β and tau at steady state in the null model and the neuro-glymphatic model following changes in glymphatic efflux/influx or synaptic and metabolic release. Release of cellular Aβ and tau species from neurons (1) into the ISF (2) in the model is controlled by the cellular solute release rate constant ({k}_{{cell_rel}}) and was kept constant during the 16 hours of wake and was reduced by 30% during the 8 h of sleep occurring in the shaded window between 16 and 24 h. Solid lines show compartment concentrations under the null model, with time-invariant release rate constant ({k}_{{cell_rel}}) and time-invariant exchange between the CSF (3) and ISF (2) compartments. Long dashed lines show change in compartment concentrations when glymphatic efflux*/*influx rate constants ({k}_{{glymph_out}}) and ({k}_{{glymph_in}}), were increased by a factor of 1.5 (B) or decreased by a factor 0.5 (C) during the 8-hour sleep window. Dotted lines show change in compartment concentrations when synaptic and metabolic release rate constant ({k}_{{cell_rel}}) was increased by a factor of 1.25 (B) or decreased by a factor of 0.5 (C) during the 8-h sleep window. Note that increasing (decreasing) either glymphatic efflux/influx or synaptic and metabolic release during sleep increases (decreases) morning plasma level of Aβ and tau species relative to the null model.

The 24-h cellular synthesis of amyloid precursor protein (APP) and tau is invariant to sleep-wake state and was modeled at a constant rate of 10 arbitrary units (a.u.) per hour28,29. Release of cellular Aβ and tau species into the ISF in the model is controlled by the cellular solute release rate constant ({k}_{{cell_rel}}) (Table 1) and was kept constant during the 16 h of wake. During the 8 h of sleep, ({k}_{{cell_rel}}) was reduced by 30% consistent with reported 30% sleep reduction in ISF Aβ and tau species28,29, and during the 8 h of sleep deprivation it was reduced by 10% reflecting a circadian time-of-day effect.

This model was evaluated under two conditions. In the null model, sleep-active glymphatic exchange was posited to not occur; therefore solute transport between the Free ISF compartment (2) and the CSF compartment (3), reflected by the rate constants ({k}_{{glymph_out}}) and ({k}_{{glymph_in}}), were held constant across both sleep and waking (or sleep deprivation) states. In this model, sleep-related decreases in synaptic and metabolic activity, which impact cellular solute release, were not permitted to vary beyond the 30% reduction in the cellular solute release rate constant ({k}_{{cell_rel}}) during the overnight 8-hour sleep period and the 10% circadian-related reduction during the overnight 8-hour sleep-deprivation period. In the neuro-glymphatic model, ({k}_{{glymph_out}}) and ({k}_{{glymph_in}}) increased during sleep, reflecting the sleep-active CSF solute influx and interstitial solute efflux observed in both rodents and humans13,16,17. Additionally, ({k}_{{cell_rel}}) was allowed to vary with sleep from its 30% reduced value relative to wake, representing changes in synaptic and metabolic activity. These changes model variations in solute release into the brain’s interstitial space, which are associated with differences in sleep quality—such as the proportion of non-REM sleep relative to wake-after-sleep onset29,30.

The full derivation of the compartmental pharmacokinetic model is provided in Supplementary Methods and Results.

Clinical study design and participant demographic data

As described previously17, we conducted two cross-over clinical studies in which participants underwent one night of normal sleep and one night of sleep deprivation, in randomized order and separated by two or more weeks (Fig. 2A). One study was conducted in The Villages® community in Central Florida where the University of Florida maintains a satellite academic research center, The UF Health Precision Health Research Center (UF Health PHRC). The second study was carried out at the University of Washington (UW) in Seattle. Study participants underwent peripheral blood draws at 1900 hrs and 0700 hrs, prior to and following the overnight sleep and sleep deprivation periods. Plasma AD biomarkers, including Aβ40, Aβ42, np-tau181, np-tau217, p-tau181, and phosphorylated tau217 (p-tau217) were quantified using C2N Diagnostics’ immunoprecipitation liquid chromatography-tandem mass spectrometry platforms31,32,33. Details of specimen collection and processing are provided in Supplementary Methods and Results. During the overnight period, participants were instrumented with an investigational in-ear wearable device from Applied Cognition17 that measured key determinants of glymphatic function, including sleep features (hypnogram and spectral band power) by electroencephalography (EEG), heart rate variability (HRV) by photoplethysmography (PPG), cerebrovascular pulse transit time (PTT) by impedance plethysmography (IPG), and brain parenchymal resistance (RP) by dynamic impedance spectroscopy (Fig. 2B). A detailed description of this device, the validation of its sleep EEG measures against gold-standard overnight polysomnography, and of the measures of RP against contrast-enhanced magnetic resonance imaging (CE-MRI)-based measures of glymphatic function has recently been reported17. The details of EEG processing, as well as hypnogram and spectral band power computations, are provided for completeness in the Supplemental Methods and Results (Signal Processing – Electroencephalography). Cerebrovascular PTT measurements obtained using impedance plethysmography were validated against contrast-enhanced MRI-based measures of cerebrovascular function, as detailed in the Supplemental Methods and Results (Supplementary Tables 2–6).

Fig. 2: Study schematic and CONSORT diagram.

The Benchmarking Study conducted at The Villages® and the Replication study conducted at the University of Washington were (A) randomized cross-over assignment of overnight sleep opportunity and overnight sleep deprivation designed to define the relationship between parenchymal resistance (RP) and glymphatic function. B Reported here are the overnight investigational device recordings of RP, EEG and HR, and blood analysis of amyloid β and tau levels (Aβ40, Aβ42, np-tau181, np-tau217 and p-tau181). C The Benchmarking Study enrolled 34 participants of which 30 completed both visits. Three were censored due to changes in device data collection and sensor locations. One withdrew following the first MRI scan. Of the 30 that completed the study, 5 overnight sleep visits and 8 overnight wake visits failed the data quality control (QC) criteria to provide sufficient artifact free data to yield results. This resulted in 25 sleep and 22 wake complete data sets. The Replication Study enrolled 14 participants. All 14 completed the study, of which 3 wake visits failed the data QC criteria and one sleep visit failed specific to the analysis of these data because of missing EEG powerband data during REM sleep caused by artifacts. The remaining 13 participants had complete data for analysis.

Testing for Sequence-Related Effects We investigated potential sequence-related effects in the randomized cross-over study by comparing evening-minus-morning plasma biomarker levels following sleep and/or sleep deprivation between the first and second visits. This analysis was conducted separately for amyloid-positive, amyloid-negative, and combined cohorts, resulting in 15 comparisons per group and a total of 45 distinct comparisons.

Development of a multivariate model to relate release and glymphatic clearance features to plasma Aβ and tau concentrations

We developed a series of multivariate mixed models to define the effects that continuous features of glymphatic physiology and synaptic-metabolic activity have on overnight release and clearance of brain interstitial Aβ and tau to the plasma. Within these models, the multiple dependent variable measures for each participant were the morning plasma Aβ and tau (Aβ40, Aβ42, np-tau181, np-tau217, p-tau181) biomarker levels. Plasma p-tau217 levels were measured but excluded from analysis because a large proportion (9 out of 49) of these cognitively-intact individuals exhibited plasma p-tau217 levels below the limit of detection for the assay, consistent with prior study findings34,35. Data from amyloid-negative and amyloid-positive individuals were analyzed separately.

Because of the large number of measured outcomes, we used dimensionality reducing single-index regression36 to combine sleep-related factors into single ‘predictors’. A detailed explanation of this single predictor development is provided in Supplementary Methods and Results. Within these models, two distinct groups of predictors were analyzed: neurophysiological variables (Physio) and hypnographic sleep stages (Hypno). Separating predictors into these groups enabled comparison of both neurophysiological measures and sleep stage durations in explaining observed differences in morning plasma Aβ and tau levels, while mitigating multicollinearity among EEG power bands and sleep stages in the regression models, as previously documented17. The first group of predictors (Physio) included EEG non-rapid eye movement (NREM) delta (0.5–4 Hz) and theta power bands, REM sleep theta and beta power bands, heart rate variability (HRV), and pulse transit time (PTT) during NREM, and parenchymal resistance RP. HRV and PTT values during REM sleep were highly correlated with those during NREM and were excluded. The NREM and REM power bands selected represent the majority of the spectral power during these sleep stages29,30, thus contributing to synaptic–metabolic release, and because EEG delta, beta and theta power bands have established associations with glymphatic clearance13,17,22,23. The second group (Hypno) comprised hypnographic sleep stages, including the durations of REM, N1, NREM (N2 + N3), and wake after sleep onset (WASO). Each group of individual predictors was combined using linear combinations into three single-index predictors for amyloid-negative and amyloid-positive participants separately: the neurophysiological predictor under the sleeping condition (Physio**S) and under the sleep deprivation/wake condition (Physio**W), and the sleep hypnogram predictor under the sleeping condition (Hypno**S).

We developed parallel multivariate linear mixed models representing the null model and the neuro-glymphatic model, as described in detail in Supplemental Methods and Results. Under the null model, morning plasma levels of Aβ and tau species were not dependent upon sleep-active glymphatic exchange and synaptic-metabolic release, and thus were not assumed to be influenced by sleep neurophysiological (Physio) and sleep stage (Hypno) predictors. Rather the morning plasma levels of AD biomarkers (Conc**AM) were regressed on the evening plasma levels (Conc**PM) for Aβ and tau biomarkers, separately for amyloid positive and amyloid negative participants. Potentially confounding variables age, sex, APOE-ε4 status, and study site were included in the model. A circadian confounder was also included in the model, reflecting the interval between the evening AD biomarker sample time and sleep-onset measured by EEG. Multivariate mixed models used participant ID as a random intercept and the categorical biomarker variable as a vector of random slopes. The neuro-glymphatic model shares the features of the null model but included the effects of the single index predictors (Physio**S, Physio**W or Hypno**S), and their respective interaction terms with evening levels of plasma AD biomarkers (Physio**S * Conc**PM, Hypno**S * Conc**PM, Physio**W * Conc**PM). The likelihood ratio test (LRT) of the neuro-glymphatic model versus the null model was used to determine which model performed better at predicting the morning plasma Aβ and tau biomarker levels. The conditional variances of each model, i.e., the residual variation conditional on a participant’s random effects, were used to define the amount of variance in morning plasma Aβ and tau levels explained by the neuro-glymphatic model over that explained by the null model. This is described in detail in Supplementary Methods and Results.

Results

Statistical significance tests presented in the results have not been adjusted for multiple comparisons.

A Consolidated Standards of Reporting Trials (CONSORT) diagram for the Villages Study and UW Study is provided in Fig. 2C. Within the Villages Study, the first three participants were removed from analysis because of a sensor position change in the investigational device. One participant was unable to complete the first MRI session and withdrew from the study. Of the remaining 30 participants (61.8 ± 2.7 years of age; 14 female, 16 male) that completed the Villages Study, five overnight sleep studies and eight overnight wake studies failed data quality control due to excessive artifacts in the recordings, leaving 25 sleep studies and 22 wake studies with analyzable device and biomarker data in the Villages Study. Of the participants enrolled in the UW Study, one was not compliant with the enforced wake protocol and was removed from analysis. A second participant was also excluded because of missing EEG powerband data during REM sleep caused by artifacts, which was required for the current analysis. The remaining 13 participants (55.9 ± 4.6 years of age; 6 female, 7 male) all completed the protocol. All overnight sleep data were usable, but two overnight wake studies were removed because of excessive artifact in the UW Study. Participant demographics, MoCA and GDS scores are listed for each study site and for the combined dataset in Table 2.

We measured plasma Aβ40, Aβ42, np-tau181, np-tau217, p-tau181, and p-tau217 at evening and morning timepoints in each participant prior to and following overnight sleep or sleep deprivation. Summary plasma AD biomarker levels are provided in Table 3. Of 38 participants, a total of 11 were assessed as ‘amyloid-positive’ by the C2N mass spectrometry with a Aβ42/ Aβ40 cutoff value of 0.08931,33,37 (Table 2). Plasma AD biomarker levels shown in Table 3 are stratified by participant amyloid status. Overall, the measured plasma levels and overnight changes with sleep agree with prior reported values for Aβ40, Aβ42 and p-tau18138. Furthermore, the overnight changes in measured plasma levels following sleep and sleep-deprivation did not differ significantly, consistent with the possibility of competing clearance or production effects during sleep and sleep-deprivation.

Of the 45 comparisons conducted to assess sequence effects on overnight changes in plasma biomarkers, only two showed statistically significant differences between the first and second visits: (i) In the amyloid-negative group under sleep deprivation, np-tau217 differed significantly (P = 0.031); (ii) In the combined group under sleep deprivation, np-tau181 showed a significant difference (P = 0.043).

Predictions of the compartmental pharmacokinetic model relating glymphatic exchange to plasma AD biomarker levels

Figure 1B presents the steady-state compartmental concentrations of Aβ and tau over a 48-hour period under the null model (solid lines), as well as under neuro-glymphatic model scenarios during sleep. These include increased glymphatic clearance relative to its sleep baseline, representing improved sleep-related drivers of clearance (dashed lines); and elevated synaptic-metabolic release relative to its sleep baseline, reflecting poor quality, fragmented sleep (dotted lines). We observe that elevations in Aβ and tau concentrations in plasma or CSF above baseline do not distinguish between an increase in sleep-related clearance (desirable) and an increase in production (undesirable).

Figure 1C depicts alternative neuro-glymphatic model scenarios during sleep involving reduced glymphatic clearance relative to baseline, as would occur with a deterioration in sleep-related clearance drivers (dashed lines) and decreased synaptic-metabolic release relative to baseline, as might result from improved, less fragmentated sleep (dotted lines). Both are compared against the null model (solid lines). Across all scenarios, both Aβ and tau concentrations in the plasma increase or decrease relative to the null model, depending on whether glymphatic clearance or synaptic-metabolic release is enhanced or diminished, respectively. These plots demonstrate that changes in Aβ and tau concentrations within the plasma and CSF compartments cannot independently distinguish between decreased glymphatic clearance and decreased solute release relative to baseline, nor between increased glymphatic clearance and increased solute release. An additional derivation of the model, described in detail in the Supplementary Methods and Results, demonstrates that in the null model, morning plasma levels will be linearly dependent on evening plasma levels only, whereas the neuro-glymphatic model has a contribution from the interaction term between the evening plasma levels and the variable rate constants ({k}_{{glymph_out}}), ({k}_{{glymph_in}}) and ({k}_{{cell_rel}}).

We next compared changes in plasma levels of species that remained primarily monomeric in form (Aβ40, np-tau), and species that are prone to form non-monomeric aggregates (Aβ42, p-tau) following increases and decreases in glymphatic clearance and synaptic-metabolic release. If this tendency to form aggregates is ignored, and rate constants for monomer-to-aggregate conversion is set to zero (({k}_{{on}-{rate}}=0)), changes in cellular release ({k}_{{cell_rel}}) or clearance ({k}_{{glymph_out}}) have no effect on the Aβ42/Aβ40 or p-tau/np-tau ratios (Supplementary Table 1, top row). In contrast, when the tendency for Aβ42 and p-tau to form non-monomeric aggregates is included in the model (({k}_{{on}-{rate}} > 0)), increased and decreased solute release and glymphatic clearance shift plasma Aβ42/Aβ40 or p-tau/np-tau ratios (Supplementary Table 1). Both the Aβ42/Aβ40 or p-tau/np-tau ratios increase in response to decreased production or increased glymphatic clearance, and decrease in response to increased production or reduced clearance. This suggests that non-monomeric aggregates act both as a secondary source of monomeric Aβ42 and p-tau species under conditions of reduced production and as an additional sink for these species under conditions of reduced clearance. Under conditions of increased glymphatic clearance, the enhanced removal of Aβ42 and p-tau from the ISF reduces their opportunity to aggregate, resulting in a higher proportion reaching the plasma relative to Aβ40 and np-tau—thus increasing their respective plasma ratios. Thus, unlike Aβ and tau concentrations in plasma and CSF - which cannot independently distinguish between decreased glymphatic clearance and decreased solute release relative to baseline, nor between increased clearance and increased release - changes in plasma Aβ42/Aβ40 or p-tau/np-tau ratios enable us to infer the primary mechanism driving changes in morning plasma AD biomarker levels under sleep or sleep-deprivation conditions.

These pharmacokinetic modeling results provide three concrete predictions that permit us to test the hypothesis that sleep-active glymphatic exchange contributes to overnight changes in plasma AD biomarker levels independent of sleep-dependent changes in solute release:

1. 1)

Under the neuro-glymphatic model, compared to normal sleep, decreased clearance during overnight sleep deprivation would reduce morning plasma AD biomarker levels while increased production would increase levels.

1. 2)

Under the neuro-glymphatic model, under conditions of sleep deprivation, morning plasma AD biomarker levels would increase by the interaction term between evening plasma levels and features of synaptic-metabolic activity contributing to greater production; under normal sleep conditions morning plasma AD biomarker levels will be increased by the interaction term between evening plasma levels and features of sleep-active glymphatic function contributing to greater clearance.

1. 3)

Under the neuro-glymphatic model, under normal sleep conditions morning plasma Aβ42/Aβ40 and p-tau/np-tau ratios will increase with greater sleep-dependent clearance; overnight sleep deprivation will decrease these ratios from greater wake-dependent production.

Effects of sleep-related neurophysiological features on overnight Aβ and tau release and clearance

The predictor Physio**S that led to the best fit of the data based on maximum likelihood for the neuro-glymphatic model in sleep for both amyloid-negative and -positive individuals showed that RP was the overwhelming contributor to Physio**S (Table 4). The output of the null model and neuro-glymphatic model with the predictor Physio**S for both amyloid-positive and amyloid-negative individuals are shown in Table 5. Estimates for single predictor coefficients within the neuro-glymphatic model for each plasma AD biomarker, evaluated at the mean evening biomarker level, are provided in Supplementary Table 7.

Within the null model, evening plasma Aβ and tau levels (Conc**PM) were the main predictor of morning plasma levels in both amyloid-positive and -negative participants, reflecting between-participant evening consistency in AD biomarker levels (Table 5). Within the neuro-glymphatic model, in addition to evening plasma biomarker levels, participant age, sex, and circadian alignment each significantly contributed to morning plasma Aβ and tau levels in both amyloid-positive and -negative participants. APOE-e4 status was significantly associated with lower morning AD biomarker levels in amyloid-positive, but not amyloid-negative individuals. Age was significantly associated with higher morning AD biomarker levels in both groups. Across both amyloid-positive and -negative participants, the Physio**S predictor significantly impacted plasma Aβ and tau levels, and the Conc**PM*Physio**S interaction terms were consistently significant across Aβ and tau analytes (Table 5).

By including overnight RP, PTT, HRV, and EEG power bands in the single predictor Physio**S, the neuro-glymphatic model led to greater predictive performance over the null model (Table 6, LRT, p < 0.001). When comparing the percent variance of morning plasma AD biomarker levels explained by the neuro-glymphatic model compared to the null model, inclusion of overnight Physio**S in the neuro-glymphatic model explained between 49.1% (Aβ40) to 56.0% (Aβ42) additional variance in amyloid-positive individuals, and between 71.3% (p-tau181) to 97.8% (Aβ40) in amyloid-negative individuals (Table 7).

As detailed above, the Physio**S single-index predictor includes several sleep-related physiological features that could contribute variously to changes in glymphatic clearance to the plasma, or to synaptic-metabolic release of Aβ and tau. We carried out a sensitivity analysis to explore the relative contributions of these individual neurophysiological predictors, and of glymphatic clearance versus synaptic-metabolic release to changes in morning plasma Aβ and tau analyte levels. This sensitivity analysis is provided in detail in Supplementary Methods and Results. The analysis suggests that parenchymal resistance RP, cerebrovascular compliance (measured by PTT) and NREM EEG delta power during sleep, which are each associated with enhanced glymphatic exchange in rodents and humans12,13,[17](htt