Deep sleep slow wave–spindle coupling is selectively linked to plasma amyloid-β levels in older adults in clinical trials

Abstract

Slow wave activity, the signature of deep/slow wave sleep, has consistently been linked to amyloid-beta (Aβ), a biomarker of neurodegeneration. Less is known about how Aβ relates to specific microstructural processes within slow wave sleep, such as the coupling of slow waves and spindles, where better functioning reflects younger age, increased memory, and less brain atrophy. Here, we pooled and re-analyzed data from three clinical trials where participants underwent an adaptation night, a baseline night and a three-night acoustic stimulation intervention to boost slow wave activity. The baseline analysis included 47 older adults (age_mean = 70.5 (0.68)) with varying cognitive functioning, whereas the intervention analysis was conducted on a subsample of 39 older adults (age_mean = 70.5 (0.74)) with varying cognitive functioning. Blood samples post-baseline and post-intervention were analyzed for Aβ 1–42/1-40-ratio. Irrespective of cognitive functioning, slow wave–spindle coupling was the best predictor for baseline Aβ, better than slow wave activity, age or cognitive functioning. Specifically, better Aβ-levels were linked to a coupling physiology resembling a younger brain. While intervention-induced increases in slow wave activity were linked to a beneficial Aβ-response across all cognitive levels, increases in slow wave–spindle coupling benefited Aβ-response exclusively in cognitively impaired individuals. Our results suggest a link between SW–spindle coupling and Aβ going beyond slow wave activity. This hints towards a potential specific function of SW–spindle coupling related to the early pathophysiology of Alzheimer’s disease.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-47886-9.

Introduction

Cognitive decline and dementia are becoming increasingly relevant issues due to our aging population. Amyloid-beta (Aβ)—a prominent pathophysiological marker of Alzheimer’s disease¹—starts accumulating up to 15 years before symptoms appear, with corresponding alterations in brain function becoming evident as early as 10 years prior to symptom onset^2,3. Hence, understanding the underlying trajectories and influences that contribute to the accumulation of Aβ is vital for targeted interventions and dementia prevention. Additionally, investigating whether such targeted interventions can yield favorable outcomes for Aβ is essential.

A growing body of research has linked disturbed sleep to Aβ. For instance, in healthy older adults, shorter self-reported sleep duration and poorer sleep quality were associated with greater cortical Aβ-burden⁴. Furthermore, total sleep deprivation studies have shown to negatively affect Aβ-levels in both CSF⁵ and the brain⁶. Specifically, disturbed slow wave sleep (SWS), the deepest sleep stage, has been shown to interfere with Aβ-levels⁷. Impairments in slow wave activity (SWA)—the power of EEG activity during SWS—have been associated with adverse implications for cortical Aβ-burden⁸ and CSF Aβ-levels⁹. One study showed that disturbed SWA can forecast the rate of cortical Aβ accumulation across subsequent years, even when correcting for age, sex or sleep apnea¹⁰. Sleep, and more specifically SWA, is thought to impact Aβ through its metabolic clearance function, a process that washes out metabolic waste products, such as Aβ^11,12.

Recently, blood-based assessments of Aβ have emerged as a cost-effective and minimally invasive method to monitor Aβ deposition in the brain^13,14. Similarly to CSF- and cortical Aβ, plasma markers of Aβ have been shown to be unfavorably affected by total sleep deprivation^15,16, with SWA specifically identified as a critical factor predicting Aβ-levels¹⁷.

The consistent evidence linking SWA to Aβ invites delving into the more specific phenomena within SWS and investigate how they may relate to Aβ. A distinctive electrophysiological process occurring during SWS is the coupling of slow waves (SWs) and sleep spindles (SW–spindle coupling). It is well established that SW–spindle coupling plays a crucial role in memory functioning¹⁸. It is hypothesized that the hierarchical orchestration of neocortical SWs (< 1.25 Hz), thalamocortical spindles (12–16 Hz) and hippocampal ripples (80–250 Hz) enables a temporally precise coordination between cortical and hippocampal networks^18–20. The hypothesis suggests that these electrophysiological events are organized in a phase-amplitude coupling hierarchy, where ripples are nested within spindle troughs, and spindles are further nested within SW peaks. The resulting synchronization allows memory traces to be selectively reactivated during sleep and gradually integrated into long-term cortical networks—a key component of system level memory consolidation¹⁸.

Research suggests that in the aging brain, spindles become uncoupled from SWs^21,22, a process occurring gradually across the human life span^23,24. Whereas there is a clear cross-frequency directionality of SWs driving spindles in younger individuals (i.e., spindles following the SW peak), in older individuals, a disruption of this hierarchical temporal structure occurs, characterized by the spindle shifting to precede the SW peak^21,24. Importantly, this shift is associated with both decreased memory function and atrophy in the medial frontal cortex^21,22. Regarding biomarkers of neurodegeneration, only few studies have investigated its relationship with SW–spindle coupling in healthy older adults. Some reports showed that SW–spindle coupling was associated with tau burden²⁵ or plasma glial fibrillary acidic protein concentration²⁴. Only two studies indicated that detrimental SW–spindle coupling was associated with cortical Aβ-burden²⁶ or unfavorable plasma Aβ-levels²⁷, respectively. An explanation for the sparsity of results may be that in healthy older adults, Aβ-levels are not elevated enough to be linked to microstructural features like SW–spindle coupling^10,24.

With both SWA and SWS’s microstructural features (like SW–spindle coupling) having been linked to Aβ, the question arises whether SWA alone might merely represent a proxy for something more specific occurring during SWS, like SW–spindle coupling. Hence, our first aim was to explore whether SWA or SW–spindle coupling serves as predictor for Aβ. Using a sample of older adults ranging from cognitively healthy to -impaired allowed us to investigate whether predictive power of coupling on Aβ-dynamics depends on the level of cognitive functioning. Additionally, we examined whether SWA or SW–spindle coupling could—in terms of their predictive value—outperform other variables known to be linked to Aβ—such as cognitive functioning, age or sleep quality^28,29.

In the second part of this paper, we aim to re-explore whether phase-locked acoustic stimulation (PLAS)—a targeted intervention previously shown to enhance both SWA and SW–spindle coupling—might be associated with beneficial changes in Aβ. PLAS operates by presenting short acoustic stimuli during SWS, time-locked to the rising up-phase of a SW. A growing body of evidence shows that PLAS is associated with increased SWA and SW–spindle coupling—both in young^30–33 and older adults^27,34–36. Importantly, stronger PLAS-induced increases in SWA have been associated with more beneficial changes in Aβ across a three-night intervention, an effect that was more pronounced in cognitively impaired vs. healthy older adults^27,36. Here, we revisit these effects, specifically focusing on the previously unexplored question of whether improvements in SW–spindle coupling are also linked to a beneficial Aβ-response and whether this depends on the level of cognitive functioning level. To address these aims, we re-analyzed datasets from previously conducted clinical trials.

Our results imply a unique predictive strength of SW–spindle coupling for baseline Aβ-levels, irrespective of cognitive functioning levels. Moreover, our results suggest that PLAS-induced increases in SWA benefit Aβ-response regardless of cognitive functioning levels, whereas PLAS-induced increases in SW–spindle coupling benefit Aβ-response exclusively in the cognitively impaired group. This hints towards a connection between SW–spindle coupling and Aβ-levels that goes beyond SWA.

Methods

Sample

The baseline sample consisted of 47 older adults (mean age: 70.5 years, SE: 0.68, range: 59–80 years; 29 females, 18 males) with a range of cognitive functioning as measured via the Montreal Cognitive Assessment (MOCA³⁷; mean MOCA score: 26.2, SE: 0.40, range: 20–30). For the purpose of this study, MOCA scores below 26 indicate cognitive impairment (n = 17, mean MOCA score: 23.2, SE: 0.37), while scores from 26 to 30 (n = 30, mean MOCA score: 27.9, SE: 0.27) indicate cognitive health³⁸. A subset of the baseline sample of 39 older adults (mean age: 70.5 years, SE: 0.74, range: 59–79 years; 23 females, 16 males) completed a three-night PLAS protocol (see design & procedure section). The PLAS protocol sample ranged from cognitively impaired (n = 16, mean MOCA score: 23.2, SE: 0.39) to cognitively healthy (n = 23; mean MOCA score: 27.8, SE: 0.30; total sample mean MOCA score: 25.9, SE: 0.44, range: 20–30). The remaining eight participants from the baseline sample were not included in the PLAS protocol sample due to following a different protocol (n = 4) or due to unavailability of post-measurement blood samples (n = 4). See Table 1 for the characteristics of both samples.

Table 1.

Characteristics of study participants. Means and standard errors are provided for both the baseline sample (= one night of natural sleep without intervention), as well as the PLAS protocol sample (= subset of the baseline sample with an additional three consecutive nights of phase-locked acoustic stimulation, see design and procedures). ¹⁾categorical variable to determine the highest academic qualification, ranging from 1 (primary school) to 4 (University). ²⁾Montreal Cognitive Assessment, maximum score: 30, ≥ 26 cognitively healthy. ³⁾Pittsburgh Sleep Quality Index, < 5 normal sleep. ⁴⁾Apnea-Hypopnea Index < 5: normal sleep; 5 ≤ AHI < 15: mild sleep apnea. ⁵⁾Plasma amyloid-beta 42/40 ratio, lower scores are indicative of a higher risk for dementia.

	Baseline Sample (n = 47)		Subset: PLAS protocol Sample (n = 39)
	mean	se	mean	se
Age	70.5	0.68	70.5	0.74
Sex	29 f, 18 m	-	23 f, 16 m	-
Education1	3.1	0.15	3.1	0.16
MOCA2	26.2	0.40	25.9	0.44
PSQI3	4.6	0.35	4.5	0.38
AHI4	8.3	1.18	9.2	1.48
Baseline plasma Aβ42/Aβ405	0.067	0.001	0.066	0.002
Baseline plasma Aβ42	6.629	0.194	6.449	0.210
Baseline plasma Aβ40	98.677	1.984	97.516	2.323

All participants underwent an extensive screening procedure to ensure that inclusion criteria were met. Exclusion criteria were sleep disorders and irregular sleep patterns as assessed via the Pittsburg Sleep Quality Index (PSQI³⁹, and via a screening night in the sleep laboratory to identify sleep apnea or restless leg syndrome, impaired hearing, non-fluency in German, current or previous neurological or psychiatric conditions, substance abuse and use of medication acting on the CNS. The study was conducted in accordance with the Declaration of Helsinki and approved by the local ethics committee of the canton of Bern, Switzerland (2018 − 01979). Written informed consent was obtained from all study participants. The current analyses represent an exploratory re-analysis of a preregistered study available under ClinicalTrials.gov (NCT04277104, first registration: 17/02/2020).

Design & procedure

The full study design contained five nights in the sleep laboratory: an adaptation night, a baseline night (with sham-PLAS), and three experimental nights (with real-PLAS; see Fig. 1). The adaptation night served as an acclimatization to sleeping under laboratory conditions and as a screening night to screen for sleep-based exclusion criteria. After a recovery night at home, participants returned to the sleep laboratory for a baseline night. During the baseline night, sham-PLAS was administered, where a PLAS algorithm detected SW peaks in the online EEG signal and set time markers of these detections, but did not transmit any sound to the sleeping participant. After the baseline night, three consecutive experimental nights ensued, containing real-PLAS where upon detection of a SW peak, an acoustic stimulus (50ms of pink noise) was transmitted to the sleeping participant. Individual hearing thresholds were determined through hearing tests, and used as target stimulus intensity during PLAS. Acoustic stimuli were transmitted via sleepphones^® (AcousticSheep LLC) at a mean volume of 69.2 dB(A). The PLAS algorithm is described elsewhere in more detail^40,41 and its application has previously been documented^27,35,36. In brief, this algorithm analyzes the most recent 120 ms of data and computes correlations between the empirical topographical voltage distribution and a template topography of a typical slow wave peak. If the correlation is rising in > 75% of samples, a SW peak is predicted, and a stimulation is triggered. Hence, the algorithm is not dependent on the absolute amplitude of the signal—a metric that decreases with age⁴².

Fig. 1.

study design. The full study design entailed a total of five nights in the sleep laboratory. All participants were subjected to an adaptation night to get acclimated to the sleep laboratory and to be screened for potential sleep disorders. All participants further completed a baseline night, where sleep was assessed without any intervention. During the baseline night, an algorithm detected SW peaks and markers were set time-locked to the peaks, however, no actual sounds were transmitted to the sleeping person (= sham PLAS). Following the baseline night, three consecutive nights of real-PLAS—with the goal to boost SWA—ensued. During the three experimental nights (E1-E3), acoustic stimuli were transmitted to the sleeping person, time-locked to the detection of a SW peak. Blood samples to assess plasma Aβ 42/40 ratios were drawn in the morning after the baseline night, as well as in the morning after the last experimental night (night E3). Baseline episodic memory performance was assessed before the baseline night. Baseline analyses were confined to the baseline night; PLAS-protocol analyses focused on relative changes during the three experimental nights compared to that baseline.

Measures of cognitive functioning, including episodic memory performance, were assessed throughout the intervention. We previously demonstrated in these data that PLAS increases SWA, SW–spindle coupling and episodic memory performance, and is associated with beneficial Aβ-responses. For detailed descriptions of the tasks and results, see^27,35,36. Here, we were selectively interested in a baseline measure of episodic memory performance as a potential explanatory variable in our statistical models (see statistical analysis section). Baseline episodic memory performance was assessed via a Face-Occupation Association (FOA) task, where participants encoded 40 faces that were each associated with one of 20 occupations (e.g. doctor or gardener). Baseline performance entailed the number of correctly recalled associations (0–40) after two encoding blocks.

Sleep was recorded using a high-density EEG system (128-channel MicroCel Geodesic Sensor Net, Physio16 input box, 400 Series Geodesic EEG System, Magstim EGI, Eugene, OR, USA), at a sampling rate of 500 Hz, referenced to Cz. Polysomnographic scoring of sleep stages was performed according to the criteria of the American Academy of Sleep Medicine⁴³ by an experienced and certified somnologist. See Table 2 for participant’s baseline sleep architecture. Note that this sleep duration is comparatively low relative to population averages in older adults⁴⁴ and may reflect laboratory effects on sleep^45,46 or characteristics of our sample (see also discussion).

Table 2.

Sleep Architecture. Means and standard errors for polysomnographic variables of the baseline night are displayed. ¹Total Sleep Time in minutes. ²Percentages of non-rapid eye movement (NREM) sleep stages N1-N3 and rapid eye movement (REM) sleep. ³Minutes spent awake. ⁴Wake time after sleep onset in minutes. ⁵Sleep onset latency, i.e. the minutes in bed until sleep onset. ⁶Sleep efficiency, i.e. the percentage of minutes spent asleep in relation to minutes lying in bed. Note: see table S4 for the three experimental nights in the PLAS protocol sample.

	Baseline Sample (n = 47)		PLAS protocol Sample (n = 39)
	mean	se	mean	se
TST1	334.6	9.46	342.3	9.12
% N12	30.3	1.71	30.6	1.90
% N22	47.9	1.23	47.2	1.13
% N32	8.7	1.12	9.0	1.32
% REM2	13.1	0.73	13.2	0.82
Wake3	136.3	7.89	140.0	7.43
WASO4	121.2	7.37	124.1	6.83
SL5	15.6	1.78	16.5	2.06
SE6	71.7	1.52	71.1	1.61

Blood samples were drawn after the baseline night and after the last experimental night (E3, see Fig. 1). The samples were taken in the morning approximately an hour after waking and immediately centrifuged and stored in a −80 °C freezer. Plasma samples were sent to the Neurochemistry Laboratory, Amsterdam University Medical Center (Amsterdam, The Netherlands), where Plasma Aβ 1–42 and 1–40 peptides were identified by means of single molecule array immunoassays (IA-N4PE¹³;. Aβ 1–42/1–40 ratios were calculated to account for inter-individual and pre-analytical variability. Lower Aβ 42/40 ratios indicate a greater risk for amyloid deposition^14,47.

EEG processing

EEG processing was performed in MATLAB R2022b (MathWorks) using the toolboxes EEGLAB⁴⁸, FieldTrip⁴⁹ and CircStat (Berens⁵⁰, as well as the phase-amplitude coupling analysis framework by Jiang et al.⁵¹. The data was down-sampled to 200 Hz, re-referenced to a global average and preprocessed via the PREP pipeline for EEGLAB⁵² and FieldTrip’s automatic artifact rejection pipeline. All analyses were performed on sleep containing slow wave activity (N2/N3 sleep) only.

Discrete SW, spindle and coupling event detection

The procedure for the detection of SWs, spindles and coupling events followed previously published work^20,21,53. Note that we use the term slow waves rather than slow oscillations, as the underlying phenomenon reflects alternating cortical up- and down-states that propagate as traveling waves across the cortex rather than a strictly periodic oscillation^54,55. All events were detected in channel Fz. For the detection of SWs, all zero-crossings within the SW-band (0.16–1.25 Hz) filtered data (using a linear-phase FIR filter (order = 500) were marked. SW peaks and SW troughs were defined as the highest and lowest values between two successive positive-to-negative zero-crossings, provided they met a duration (0.8–2 s apart) and amplitude (75th percentile) criterion⁵³. The mean SW amplitude was 16 µV (0.7) for the baseline sample and 15.9 µV (0.8) for the PLAS protocol sample. For the detection of spindles, the signal was first filtered between 12 and 16 Hz using a linear-phase FIR filter (order = 500). Next, the instantaneous amplitude (envelope) of the filtered signal was calculated using a Hilbert transform and smoothed using a 200ms moving average. Data segments exceeding the 75th amplitude percentile for a duration of 0.5–3 s were defined as spindles at their maximum value²⁰. The mean spindle amplitude was 4.1 µV (0.2) for both the baseline sample and the PLAS protocol sample. Finally, SW–spindle coupling events were defined as spindles whose maximum amplitude (spindle peak) occurred within a window of ± 2.5 s around the SW trough, as previously described²¹. Thus, coupling was determined based on the spindle peak relative to the slow wave. For a graphical illustration of slow waves, spindles and their coupling, see Fig. 2A.

Fig. 2.

Illustration of slow waves, spindles and their coupling. (A) For the detection of slow waves and spindles, established duration and amplitude criteria were applied after the original data was filtered in the slow wave and spindle frequency range. Coupling relates to the phase of the slow wave at the spindle’s maximum (blue). (B) Coupling strength is measured using the resultant vector length (red). For this the phase distribution of all spindles is plotted and the average preferred phase is calculated (dark blue). The resultant vector length indicates the amount of circular spread. A short resultant vector (low coupling strength) results from widely spread spindle phases, while a long vector (high coupling strength) indicates tightly clustered phases. (C) Coupling directionality is measured using the phase slope index, which measures the consistency of phase lag or lead between the spindle and slow wave signal. A phase slope index significantly different from 0 suggests the leading signal drives the lagging signal. A positive phase slope index indicates that slow waves drive spindles, while a negative phase slope index indicates that spindles drive slow waves.

SW–spindle coupling

In addition to the quantity of SW–spindle coupling events, two derived measures capturing the quality, i.e. the exact synchronization of SWs and spindles were calculated. Importantly, these measures reflect coupling strength and coupling directionality and thus convey information about the state of the coupling hierarchy beyond the raw phase of spindle occurrence along the slow wave. See Fig. 2 for an overview of all coupling measures.

SW–spindle coupling strength

coupling strength was measured via the resultant vector length (RVL) (Berens⁵⁰. The RVL quantifies how consistently spindles are coupled with SWs (see Fig. 2B). A longer RVL suggests that most spindles are closely clustered around their mean SW phase, indicating stronger coupling. A shorter RVL is indicative of greater variance of spindles within the SW phase, which suggests weaker coupling. To calculate RVL, the data from channel Fz was first filtered within the SW-band (0.16–1.25 Hz). Next, the instantaneous phase angle of the filtered signal was calculated using a Hilbert transform. For the above-defined SW–spindle coupling events (centered spindle events on the peak of the Hilbert-envelope over spindles), the individual phase angles were extracted and used to calculate the mean RVL using CircStat’s circ_r function.

SW–spindle coupling directionality

coupling directionality was measured using the Phase Slope Index (PSI)⁵¹. Here, PSI was calculated between the phase of the SW frequency (0.5–2 Hz; in steps of 0.5 Hz) and the amplitude of the spindle frequency (12–16 Hz; in steps of 1 Hz). The PSI provides a measure of strength and consistency of influence of one signal over the other. A PSI of 0 means that no information flows in either direction. A low magnitude PSI indicates some influence of one signal over the other, and a high magnitude PSI indicates strong and consistent influence. Futhermore, PSI is a measure of lag/lead between two frequencies, where 0 suggests no consistent directional influence of one over the other frequency. A positive value suggests a forward interaction, indicative of the SW leading the spindle, and a negative value suggests a reverse influence with the spindle leading the SW (see Fig. 2C). The former (a positive value) is indicative of a coupling hierarchy toward a “younger” and less atrophic brain^21,24.

To calculate PSI, the EEG data in Fz was epoched ± 2.5s around the previously detected SW troughs. Frequency power was estimated analyzing five cycles of each frequency within a sliding window of 2 s, moving in steps of 1 s. Finally, for each participant, average PSI values were calculated over the frequency subbands.

Statistical analysis

Baseline analyses

To analyze which variables best explain Aβ 42/40 ratio, an optimized regression model was calculated. Although the main dependent variable of interest was Aβ 42/40, all analyses were additionally repeated for the single peptides Aβ 42 and 40. As a starting point, an a priori regression model was defined: Aβ variable ~ MOCA score + number of coupled spindles + SW amplitude + age + coupling strength + coupling directionality. Next, a stepwise regression (max. steps: 10) was performed using R-squared as the optimization criterion to iteratively refine the model. In this process, other potential predictors (spindle amplitude, percentage of sleep stages N1, N2, N3, REM, total sleep time, sleep efficiency, baseline memory performance and sex) were considered. At each step, the model examined the effect of adding a new predictor or removing a predictor already in the model on the overall R-squared. A predictor was added to the model if its inclusion increased R-squared by at least 0.1, and an existing predictor was removed if its exclusion reduced R-squared by no more than 0.05.

To strengthen the robustness of our results, the optimized regression models were calculated both conservatively (excluding potential outliers as determined via Cook’s distances > 4/n⁵⁶ and liberally (including all data points). Additional analyses to ensure the robustness of our results entailed recalculating all models without a specific a priori model. An advantage of not using an a priori model is to reduce potential bias stemming from the pre-selection of predictors. If all procedures (liberal, conservative, informed (with a priori model), and empty (without a priori model) yielded similar results, we considered the results to be robust.

PLAS protocol analyses

In addition to the baseline analysis, we investigated relative changes during the three experimental nights (where PLAS was applied) compared to the baseline night without stimulation (see Fig. 1). We will refer to these relative changes as PLAS-induced changes. To analyze how PLAS-induced changes in sleep markers of interest (SW amplitude and SW–spindle coupling) interact with changes in Aβ 42/40 ratios from pre to post intervention, linear mixed effects models (LMM) were calculated. SW amplitude and SW–spindle coupling were used as sleep markers of interest because (A) they are specific markers that PLAS has been shown to increase^27,31, and (B) they are linked to Aβ, as demonstrated by previous research^8,9,26 as well as our baseline analyses.

SW amplitude and SW–spindle coupling were predicted on the single trial level. LMMs on the single trial level increase statistical power by considering all available observations rather than using aggregated values where information can be lost⁵⁷. Importantly, LMMs are robust to unequal numbers of observations⁵⁷ and have been reported to reduce multiple comparison concerns⁵⁸.

Regarding SW amplitude, this meant that each detected SW event within a night was treated as a separate trial, with each event containing an amplitude value. Regarding SW–spindle coupling, each SW–spindle coupling event within a night served as a separate trial. Here, we first calculated the mean phase angle per night using CircStat’s circ_mean function. Next, the absolute distance of each event from the individual mean phase angle was calculated and divided by 180 (the maximum value), to get the relative distance of each event. The relative distance was then subtracted from 1 to provide a “closeness” estimate instead of a distance estimate per trial, where 1 means maximum “closeness” and 0 minimum “closeness”. Hence, higher values indicate that the spindle is closely aligned with the individual average angle for the night, while lower values indicate that the spindle is further from the individual average angle. This measure can be considered an approximation of the RVL, hence coupling strength, at the single trial level.

We further explored an approximation to the PSI (as a measure for coupling directionality) on the single trial level, by calculating each spindle’s phase angle difference from zero. 0° can be seen as the reversal-point of cross-frequency directionality, with numbers below zero indicating that the spindle lies before the SW peak and numbers above 0 indicating that the spindle lies after the SW peak. The distances both below and above 0 in degrees were rescaled and mapped from their minimum to maximum values onto a range from − 1 to 1, with zero corresponding to 0°. Note that this approach does not measure precisely the same aspect as PSI. While a consistent phase angle difference indicates temporal precedence, the PSI additionally reflects a directional influence by quantifying the consistency of the phase difference as a function of frequency, which our surrogate metric cannot evaluate on single-trial level. Nonetheless, we consider the measure described to be a conceptually close approximation, albeit with some inherent limitations.

Three separate LMMs were calculated for the prediction of SW amplitude, SW–spindle coupling strength, and coupling directionality. The model included the categorical variable nights (BL, E1, E2, E3, see Fig. 1), the change score in Aβ 42/40 from pre to post intervention, as well as their interaction as fixed effects. We controlled for age and MOCA score and included a random intercept for each participant. Maximum likelihood estimation was used to fit the model. Note that the factor night is coded with the first night (pre-intervention, baseline) as reference. Effects on SW amplitude and SW–spindle coupling can therefore be interpreted as contrasts against the baseline night, hence PLAS-induced changes.

Results

Slow wave–spindle coupling as best predictor for plasma amyloid-beta

To examine what variables best explain differences in Aβ-values at baseline, optimized regression analyses were conducted. The conservative optimized regression analysis showed that the model best explaining differences in Aβ 42/40 was the model incorporating both coupling strength (β = 0.052, t(40) = 4.07, p < 0.001) and coupling directionality (β = 1.373, t(40) = 3.74, p < 0.001), but none of the other potential predictors (F(2, 40) = 13.7, adj. R² = 0.38, p < 0.001, see Fig. 3A). The same relationship was shown in the more liberal model, where potential outliers were not removed (F(2, 44) = 7.43, adj. R² = 0.22, p = 0.002; coupling strength: β = 0.036, t(44) = 3.03, p = 0.004; coupling directionality: β = 1.045, t(44) = 2.69, p = 0.01). For both the initial and conservative model, results did not change when using an empty (without a priori) model as baseline for the optimized regression instead of the informed (with a priori) model. For a visualization of the relationship between Aβ 42/40 and coupling parameters, see the left panels in Fig. 3C and D.

Fig. 3.

Association between Aβ and SW–spindle coupling at baseline. (A) A stepwise self-optimized regression approach showed that Aβ 42/40 ratio is best explained by coupling strength (measured via resultant vector length (RVL)) and coupling directionality (measured via phase slope index (PSI)), as represented by black solid elements and green arrows. Other factors that were offered for consideration, but not included in the model, are shown as gray rectangles. (B) An additional model incorporating both the 42 and 40 peptides individually revealed that effects were mainly carried by the Aβ 42 peptide, which is best explained by coupling strength and coupling directionality. For the Aβ 40 peptide, there was only a trend-level effect for the association with coupling strength, and no association with coupling directionality. C./D. Scatter plots showing simple regression plots between Aβ 42/40 ratio (left panels), Aβ 42 (middle panels), Aβ 40 (right panels) and coupling strength, as measured via RVL (C) or coupling directionality, as measured via PSI (D). For illustrative purposes, the plots in (C) and (D) are shown for all participants. The reported t and p values correspond to the values of the individual regressors within the full model under (A) and (B) Regressions fits are depicted for the entire sample (black), as well as separately for healthy (MOCA score ≥ 26, blue) and cognitively impaired (MOCA score < 26, red) participants. Note that the regression results under C for the single amyloid peptides (middle and right panel) are mostly driven by healthy participants (blue), since cognitively impaired participants don’t cover the entire range of the data. While fitting into the overall data cloud, they appear to show an inverse trend on their own. Note also that the full models reflect conservative models, where outlier subjects were excluded from the analysis. For both Aβ 42/40 ratio and Aβ 42, effects remained stable even when all outliers were included. However, for Aβ 40, results were inconsistent with results depending on model choice. Legend: ^*** p < 0.001, ^** p < 0.01, ^(*) p = 0.07, n.s. not significant.

Predicting the Aβ 42 and 40 peptides individually revealed that effects were mainly carried by the Aβ 42 peptide (see Fig. 3B). Coupling strength and coupling directionality were consistently the best predictors for Aβ 42, both in the conservative (F(2, 42) = 9.81, adj. R² = 0.29, p < 0.001; coupling strength: β = 6.078, t(42) = 3.13, p = 0.003; coupling directionality: β = 192.853, t(42) = 3.36, p = 0.002) and more liberal models (F(2, 44) = 8.58, adj. R² = 0.25, p < 0.001; coupling strength: β = 5.719, t(44) = 3.24, p = 0.002; coupling directionality: β = 166.09, t(44) = 2.91, p = 0.006) and results did not change with an empty model as opposed to the informed model. See the middle panels in Fig. 3C and D for a visualization of the relationship between Aβ 42 and coupling parameters.

For the Aβ 40 peptide, results showed no consistent effect. In the conservative model, coupling strength predicted Aβ 40 at trend levels (β = 36.484, t(43) = 1.844, p = 0.072; F(1, 43) = 3.4, adj. R² = 0.05, p = 0.072, see Fig. 3B), but this was not seen in the more liberal model, where MOCA score and SW amplitude were the best predictors (MOCA: β=−1.495, t(44)=−2.046, p = 0.047; SW amplitude: β = 0.81, t(44) = 1.984, p = 0.054; F(2, 44) = 3.23, adj. R² = 0.09, p = 0.049). Furthermore, when using an empty instead of an informed model, both conservative and liberal procedures revealed no significant effect of any variable. The relationship between Aβ 40 and coupling is visualized in Fig. 3C and D, right panels.

To investigate whether effects differed depending on the level of cognitive functioning, the analyses were repeated, separately for cognitively healthy (n = 30) and cognitively impaired (n = 17) participants (see suppl. table S1A for between-group characteristics). The only variable consistently appearing in all models (liberal, conservative, empty, informed) as a significant predictor for Aβ 42/40 and Aβ 42 was coupling strength for healthy older adults, and coupling directionality for cognitively impaired older adults (see supplementary material, table S2). For Aβ 40, no consistency across models was observed. These results suggest that, irrespective of cognitive functioning level, slow wave–spindle coupling was the best predictor for Aβ 42/40 and Aβ 42. The results further suggest that distinct mechanisms of slow wave–spindle coupling may be relevant at different levels of cognitive functioning.

Interestingly, MOCA score was neither significantly correlated with coupling strength (p > 0.9), coupling directionality (p > 0.8), nor Aβ 42/40 (p > 0.4). However, coupling measures were the best predictors for Aβ-levels. This indicates that coupling may be more sensitive to earlier alterations in Aβ-levels compared to MOCA score.

In sum, the results show that coupling strength and coupling directionality are better predictors for Aβ 42/40, and Aβ 42 than any other variable in the model, such as age, MOCA score, or SW amplitude—all variables expected to be linked to Aβ (e.g. Rodrigue²⁹, Rosenblum¹⁷,. Specifically, more favorable Aβ-levels were linked to a more consistent clustering of spindles within the SW phase as opposed to a more chaotic placement (coupling strength). This was particularly seen in healthy older adults. Additionally, more beneficial Aβ-levels were linked to the SW leading the spindle as opposed to the spindle leading the SW (coupling directionality), which was particularly pronounced in the cognitively impaired group. Both these measures were associated with plasma Aβ in a way that resembles the following statement: The more the brain resembled a younger state from a coupling-physiology perspective, the more beneficial the plasma Aβ profile.

PLAS-induced gains in SW amplitude/SW–spindle coupling strength link to beneficial Aβ-response

As our baseline analyses indicated a strong link between SW–spindle coupling and Aβ, we analyzed in a next step whether Aβ-levels could profit from an intervention-induced increase in coupling quality. An additional variable of interest was SW amplitude, as previous reports showed that improvements in SW amplitude were linked to a beneficial Aβ-response^27,36.

Different LMMs were conducted to examine PLAS effects on sleep parameters (SW amplitude and SW–spindle coupling) across three experimental nights (E1, E2 and E3), as well as how these effects interact with changes in Aβ 42/40 ratio from pre to post intervention. Table 3A summarizes the main results for the full model incorporating all participants. There were significant main effects for SW amplitude in all three experimental nights (E1: β = 0.17, p = 0.008; E2: β = 0.81, p < 0.001; E3: β = 0.27, p < 0.001, Table 3A, left panel) as well as SW–spindle coupling strength (E1: β = 0.012, p < 0.001; E2: β = 0.01, p = 0.001; E3: β = 0.007, p = 0.021, Table 3A, right panel), but not SW–spindle coupling directionality (see suppl. material table S3). These findings suggest an overall increase in SW amplitude and SW–spindle coupling strength, but no enhancement in coupling directionality, during each of the three experimental nights in relation to the baseline night. In other words, PLAS increased SW amplitude and SW–spindle coupling strength, but not coupling directionality. For the increase in SW amplitude but not coupling strength nor directionality, a trend-level effect for MOCA score was observed (β = 0.61, p = 0.058), indicating that higher cognitive functioning levels were associated with higher SW-amplitude but not increased SW–spindle coupling quality.

Table 3.

Results of linear mixed effect models (LMMs). The models tested whether PLAS-induced changes in SW amplitude and SW–spindle coupling strength (as measured via an approximation to the resultant vector length on the single trial level) interact with changes in Aβ 42/40 ratios from pre to post intervention. (A) results for the full model containing all participants. (B) results for the cognitively healthy (MOCA score ≥ 26) subgroup. (C) results for the cognitively impaired (Moca Score < 26) subgroup. The models both investigated increases in SW amplitude (left panels) and SW–spindle coupling strength (right panels). Age and MOCA score were included as control variables, with each participant assigned a random intercept. In each model, interactions are highlighted in grey, with p-values for significant interaction and main effects (p < 0.05) shown in bold.

	Estimate	Std. Error	t	p	Estimate	Std. Error	t	p
SW amplitude					SW–spindle coupling strength
A. Full sample models
(Intercept)	−11.134	16.241	−0.686	0.497	0.336	0.191	1.762	0.086
Night E1	0.169	0.063	2.661	0.008	0.012	0.003	3.848	< 0.001
Night E2	0.81	0.063	12.914	< 0.001	0.01	0.003	3.201	0.001
Night E3	0.266	0.062	4.323	< 0.001	0.007	0.003	2.317	0.021
Aβ change	−71.572	250.514	−0.286	0.777	1.561	2.992	0.522	0.605
MOCA score	0.613	0.304	2.02	0.05	0.005	0.004	1.322	0.194
age	0.157	0.183	0.859	0.396	0.001	0.002	0.473	0.639
Night E1 : Aβ change	166.104	18.861	8.807	< 0.001	1.323	0.923	1.434	0.152
Night E2 : Aβ change	87.633	18.935	4.628	< 0.001	1.528	0.917	1.666	0.096
Night E3 : Aβ change	150.576	18.712	8.047	< 0.001	2.539	0.919	2.762	0.006
B. Cognitively healthy sample models
SW amplitude					SW–spindle coupling strength
(Intercept)	−25.036	24.82	−1.009	0.324	0.331	0.296	1.121	0.274
Night E1	0.236	0.085	2.77	0.006	0.008	0.004	2.136	0.033
Night E2	1.091	0.084	12.979	< 0.001	0.01	0.004	2.733	0.006
Night E3	0.571	0.082	6.929	< 0.001	0.001	0.004	0.369	0.712
Aβ change	−104.585	311.864	−0.335	0.74	5.94	3.767	1.577	0.128
MOCA score	1.372	0.745	1.842	0.078	0	0.009	0.043	0.966
age	0.054	0.219	0.247	0.807	0.003	0.003	1.115	0.277
Night E1 : Aβ change	161.65	23.71	6.818	< 0.001	0.338	1.08	0.313	0.754
Night E2 : Aβ change	76.374	24.136	3.164	0.002	−1.075	1.084	−0.991	0.322
Night E3 : Aβ change	62.656	23.541	2.662	0.008	1.47	1.074	1.369	0.171
C. Cognitively impaired sample models
SW amplitude					SW–spindle coupling strength
(Intercept)	0.209	25.286	0.008	0.993	0.715	0.314	2.28	0.037
Night E1	0.043	0.093	0.459	0.647	0.018	0.005	3.419	0.001
Night E2	0.385	0.093	4.133	< 0.001	0.008	0.005	1.518	0.129
Night E3	−0.246	0.091	−2.694	0.007	0.016	0.005	3.11	0.002
Aβ change	−151.533	381.783	−0.397	0.697	−6.41	4.841	−1.324	0.202
MOCA score	−0.95	0.744	−1.278	0.22	−0.003	0.009	−0.316	0.756
age	0.504	0.301	1.675	0.113	−0.002	0.004	−0.506	0.62
Night E1 : Aβ change	163.823	31.546	5.193	< 0.001	3.264	1.721	1.897	0.058
Night E2 : Aβ change	113.477	30.76	3.689	< 0.001	7.015	1.676	4.184	< 0.001
Night E3 : Aβ change	355.484	31.254	11.374	< 0.001	4.565	1.721	2.653	0.008

Notably, highly significant interaction effects were found between Aβ 42/40 ratio change from pre to post intervention and the increase in SW amplitude during all three nights (E1 x Aβ 42/40 change: β = 166.1, p < 0.001; E2 x Aβ 42/40 change: β = 87.6, p < 0.001; E3 x Aβ 42/40 change: β = 150.6, p < 0.001, see Table 3A, left panel). These results indicate that PLAS-induced increases in SW amplitude were strongly and consistently associated with a beneficial Aβ-response across the intervention. Regarding SW–spindle coupling strength, only the third experimental night showed a significant interaction with Aβ 42/40 change (E3 x Aβ 42/40 change: β = 2.5, p = 0.006, see Table 3A, right panel), indicating that increases in SW–spindle coupling strength in the last experimental night are linked to a beneficial Aβ-response. The left panels in Fig. 4A and B visualize these associations, showing the relationship between Aβ 42/40 change scores and increases in SW amplitude and SW–spindle coupling strength across the three nights, respectively. Removing one outlier subject identified by visual inspection (see Fig. 4B, left panel) eliminated the previously significant interaction between Aβ change and increases in coupling strength in the third night. For coupling directionality, no significant interaction effects were observed (see suppl. material, table S3). To sum up, in the full sample, only increases in SW amplitude, but not SW–spindle coupling strength or directionality seem to be associated with beneficial Aβ-response.

Fig. 4.

Relationship between PLAS-induced improvements in sleep parameters and changes in Aβ 42/40 ratio. In each plot, the x-axis represents the change in Aβ 42/40 ratio from pre-intervention to post-intervention, where higher values are indicative of a more favorable change. The y-axis depicts the PLAS-induced increase in (A) SW-amplitude and (B) SW–spindle coupling strength, as measured via a single trial level approximation to the resultant vector length, relative to the baseline night without stimulation. A higher value reflects a beneficial shift in sleep parameters. The left panels show mean values per night for all participants along with their regression fit, while the right panels separately depict means and fit for healthy (MOCA score ≥ 26) and cognitively impaired (MOCA score < 26) participants. While the PLAS-induced increase in SW amplitude is associated with more beneficial changes in Aβ 42/40 ratio in all three nights and irrespective of cognitive health, effects for PLAS-induced increases in SW–spindle coupling strength are evident only in the cognitively impaired group. Note that means and regression fits are presented for illustrative purposes only; the statistical model (LMM, see Table 3) is based on the single trial level of all detected SWs and spindle-matched SWs, respectively. Note also that change in Aβ42/Aβ40 ratio (x-axes) and coupling strength (y-axis in B) are unitless measures. Legend: ^*** p < 0.001, ^** p < 0.01, ^(*) p = 0.058, n.s. not significant.

To investigate whether effects were dependent on the level of cognitive functioning, we conducted separate analyses for cognitively impaired and healthy participants (see suppl. table S1 B for between group characteristics). While interaction effects between increases in SW amplitude and Aβ 42/40 change were seen in both the cognitively impaired and the healthy subgroup (see Fig. 4A, right panel and Table 3B and C, left panels), this was not the case for the increase in SW–spindle coupling strength. Here, in the cognitively healthy subgroup, no significant interaction effects were observed, regardless of whether the outlier participant was included or excluded (see Table 3B, right panel and Fig. 4B, right panel). In the cognitively impaired subgroup however, there was a trend-level interaction effect for the first experimental night (E1 x Aβ 42/40 change: β = 3.3, p = 0.058), and significant interaction effects for the second and third night (E2 x Aβ 42/40 change: β = 7.0, p < 0.001; E3 x Aβ 42/40 change: β = 4.6, p = 0.008, see Table 3C, right panel, Fig. 4B, right panel). For SW–spindle coupling directionality, neither group showed significant interaction effects (see suppl. material, table S3).

Our post hoc analyses investigated whether effects were primarily driven by either the 42 or the 40 peptide. Positive interactions between changes in Aβ and changes in sleep parameters were only observed for the 42 peptide, although inconsistently (SW amplitude: p_E1 < 0.001; p_E2 > 0.3; p_E3 < 0.001; coupling strength: p_E1 = 0.012; p_E2 > 0.6; p_E3 > 0.4), and not for the 40 peptide. This suggests a potential tendency for the observed effects to be somewhat more carried by the 42 peptide. However, due to the lack of consistency, our observed effect is more clearly attributable to the Aβ 42/40 ratio rather than to the 42 peptide alone.

Together these results suggest that PLAS-induced increases in SW amplitude are associated with a beneficial Aβ-response regardless of cognitive functioning level. In contrast, PLAS-induced increases in SW–spindle coupling strength appear to correlate with a beneficial Aβ-response only within the cognitively impaired group. Changes in coupling directionality were not associated with changes in Aβ, irrespective of cognitive functioning levels.

Discussion

Sleep, and in particular SWS has been consistently connected to Aβ-dynamics^7–10,17. Less is known about the link between Aβ and specific microstructural aspects within SWS, such as the coupling of SWs and sleep spindles. Here we show that in older adults cognitive functioning levels, baseline SW–spindle coupling strength (i.e., how consistent the coupling is) and SW–spindle coupling directionality (i.e., whether the spindle or the SW leads) are the best predictors for plasma Aβ-dynamics. Specifically, more favorable Aβ-levels were best predicted by a more consistent clustering of spindles within the SW phase and a shift of the coupling hierarchy toward a “younger” status where the SW leads. We further explored how a three-night PLAS intervention—a non-invasive tool known to boost both SWA and SW–spindle coupling—interacts with Aβ-response. Results showed that PLAS-induced increases in SW amplitude were associated with a more beneficial Aβ-response from pre- to post-intervention, irrespective of cognitive functioning levels. Interestingly, PLAS-induced increases in SW–spindle coupling strength (but not -directionality) were only associated with a more favorable Aβ-response in the cognitively impaired, but not the healthy subgroup. This suggests that PLAS-induced increases in SW–spindle coupling might be selectively linked to beneficial Aβ-response in cognitively impaired older adults, where these dynamics are arguably deteriorating and may thus exhibit more room for improvement.

SW–spindle coupling measures were the best predictors for plasma Aβ 42/40 ratio and Aβ 42 cross-sectionally, better than any other sleep quality measure, such as SW and spindle amplitudes, sleep duration, sleep efficiency, or the percentage of different sleep stages. SW–spindle coupling measures were also better predictors for Aβ 42/40 ratio and Aβ 42 than age, sex or cognitive functioning, as assessed via an episodic memory task and the MOCA score³⁷. This unique predictive value, specifically related to the more pathogenic 42 peptide⁵⁹, suggests that SW–spindle coupling is a critical physiological process related to the early pathophysiology of Alzheimer’s disease.

One study conducted in healthy older adults showed that precise SW–spindle coupling was significantly associated with Aβ-burden over the medial prefrontal cortex²⁶. This result was specific for SW–spindle coupling: SWA was not associated with Aβ-burden, contrasting previous reports⁸. Like our results, this suggests an important and potentially specific role of SW–spindle coupling beyond SWA in predicting Aβ-burden. SW–spindle coupling is known to involve an interplay of neocortical SWs, thalamocortical spindles and hippocampal ripples, facilitating efficient information transfer across widespread brain regions and playing a critical role in memory consolidation^18,20,60. In aging, spindles become uncoupled from SWs, a change that is linked to both brain atrophy and decreased memory functions^21,22. Arguably, early unfavorable Aβ-dynamics might specifically disturb the thalamocortical interplay driving the co-occurrence of spindles and SWs before SWA itself is affected. While SWA may remain intact, the finer, micro-oscillatory dynamics like SW–spindle coupling might already be disturbed. However, the causal direction of this relationship is not resolved.

SW–spindle coupling might also have a specific function in driving Aβ-dynamics. SWA has been shown to be involved in glymphatic clearance, where neurotoxins such as Aβ are washed out of the brain^11,12. Well-functioning coupling-dynamics might therefore be a sign of a highly functioning glymphatic clearance system. Arguably, SW–spindle coupling may serve a similar or at least supporting function in glymphatic clearance as has been shown for SWA¹¹. To answer this question, however, more research is needed, specifically using animal models or human intracranial recordings allowing for more direct physiological measurements and manipulations.

Our results showed that both coupling strength and coupling directionality were important predictors of Aβ. When separately analyzing participants categorized based on their cognitive functioning, we found that in cognitively impaired older adults, Aβ was specifically associated with coupling directionality whereas in cognitively healthy older adults, Aβ was specifically linked to coupling strength. These different SW–spindle coupling metrics may reflect different levels of functional priority. Coupling strength indicates the circular spread of spindles across the SW, i.e. how tightly spindles are clustered around their preferred phase. Coupling directionality, on the other hand, measures a more fine-grained interaction between the two oscillatory signals, reflecting the consistency of phase lag or lead, i.e., where on the SW the spindle occurs. Using an orchestra analogy, coupling strength could represent the alignment of musicians, such as how precisely they synchronize with each other in terms of timing and expression. Coupling directionality might represent how closely the musicians follow the conductor, ensuring the overall coordination of the piece. If the musicians fail to follow the conductor, the performance can fall apart completely—they might even play different sections or fall out of sync entirely. However, if the musicians’ alignment with each other is imperfect, the performance can still work—albeit with some declines in quality.

We propose that coupling directionality may represent the necessary, foundational basis for optimized functionality, whereas coupling strength might rather serve a secondary, facilitating mechanism. Hence, our results suggest that in cognitively impaired older adults, more detrimental Aβ-levels are associated with the breakdown of the hierarchy of SWs and spindles, which may be so fundamental that it leaves no room for the clustering of SWs and spindles to be of predictive value. In healthy older adults, more detrimental Aβ-levels are associated with a deterioration in the clustering of SWs and spindles, but not with the underlying hierarchy. Hence, in healthy participants, irrespective of Aβ-levels, the fundamental mechanism is still intact, while the more supplementary mechanism could be optimized in those with more detrimental Aβ-levels. Although our results suggest that distinct mechanisms of phase-amplitude coupling may be relevant at different levels of cognitive functioning, the neurophysiological correlates of these specific SW–spindle coupling measures have yet to be elucidated. It is noteworthy that MOCA score was neither correlated with coupling strength, coupling directionality, nor Aβ-levels. As Aβ starts accumulating up to 15 years before symptoms appear², our results might indicate that coupling may reflect subtle processes of cognitive decline that precede detectable changes in MOCA score. After identifying SW–spindle coupling as a strong predictor for plasma Aβ ratios, we aimed to investigate whether intervention-induced increases in SW–spindle coupling might be associated with more beneficial Aβ-responses. Our results showed that a three-night PLAS intervention led to both increases in SWA as well as increases in SW–spindle coupling, paralleling previous reports^27,30,31,34. The pooled analysis revealed that only PLAS-induced increases in SW amplitude, but not PLAS-induced increases in SW–spindle coupling strength were associated with a beneficial Aβ-response from pre- to post-intervention. When separately analyzing healthy and cognitively impaired older adults, PLAS-induced increases in SW amplitude were consistently associated with a beneficial Aβ-response. Hence there seems to be a clear relationship between increases in SWA and more favorable changes in Aβ-levels, irrespective of cognitive functioning levels. This is in line with the hypothesis that SWA serves an important role in glymphatic clearance mechanisms potentially allowing for better washout of neurotoxins^11,12. Strikingly, PLAS-induced increases in SW–spindle coupling strength were also associated with a beneficial Aβ-response, albeit only in the cognitively impaired, but not the healthy subgroup. Previously, we reported an association between a detrimental coupling shift and a blood-based biomarker for astrocyte activation (GFAP), indicating a metabolic change associated with the coupling shift²⁴. Hence, in the cognitively impaired subgroup, it could be that increases in SW–spindle coupling may also play a part in Aβ-dynamics in a way that SW–spindle coupling might serve a similar or at least supporting clearance function as is the case for SWA. However, this interpretation remains speculative. Nevertheless, the PLAS-induced increase in SWA and coupling and the subsequently observed direct association between this increase and improved Aβ-response suggests that PLAS could be a valuable tool for enhancing Aβ-dynamics in individuals at risk for dementia. PLAS has been shown to evoke two distinct electrophysiological outcomes - global KC-like responses and local, phase-specific cortico-cortical responses - that likely reflect the engagement of different neural circuits⁶¹. Distinguishing between different PLAS-evoked event types may provide further insight into the mechanisms linking SW–spindle coupling and Aβ and represents an interesting direction for future research.

The reason why PLAS-induced increases in SW–spindle coupling were associated with a more beneficial Aβ-response in the cognitively impaired, but not the healthy group, is a matter of speculation. Arguably, as PLAS targets SWs, PLAS-induced increases in SWs/SWA may represent the primary effect of PLAS, while increases in SW–spindle coupling likely emerge as a downstream, secondary effect: the increased SWA may create more windows of opportunity for coupling to occur. Because of a potentially disadvantaged starting position, PLAS-induced increases in SW amplitude may more effectively drive downstream improvements in SW–spindle coupling in the cognitively impaired group. Support for the notion of a more disadvantaged starting position in cognitively impaired older adults comes from previous research suggesting that atrophy and worse memory are associated with worse SW–spindle coupling^21,22. However, although in our sample, baseline SW–spindle coupling was decreased in cognitively impaired compared to healthy individuals, these differences were not significant (see suppl. table S1). An alternative explanation for linking PLAS-induced increases to more beneficial Aβ-responses could be that instead of a causal effect of the stimulation, the network response to PLAS might simply serve as a predictor of how Aβ-levels progress, without the stimulation itself playing a causal role. However, if more responsive brains were to explain our results, we would expect to find effects either in both groups, or in the healthy group only, where brain responsiveness is arguably higher than in the cognitively impaired group.

Two limitations should be considered with regard to our coupling measures. First, we did not examine potential coupling-phase-dependent effects on Aβ dynamics. As coupling phase represents a circular variable, investigating potential phase-dependent effects would require analytical approaches tailored to circular data and was beyond the scope of the present analyses. Furthermore, although we used single-trial approaches to analyze both coupling strength and coupling directionality, we only found effects for the former. We acknowledge that our approach to coupling directionality on the single trial level does not measure precisely the same aspect as coupling directionality on the aggregated level, the phase slope index. Hence it is possible that PLAS effects were truly limited to coupling strength rather than coupling directionality; however, it is also conceivable that our single trial approach for coupling directionality was not sensitive enough to detect such effects. In line with our previous argument, the absence of an effect for coupling directionality is at least plausible. Our PLAS intervention—particularly given its short-term nature—can enhance SWA as intended, leading to increased clustering of spindles and SWs, but cannot affect the more fine-grained underlying process. It is possible that a longer intervention would be required to impact this finer process.

A limitation of this study is the absence of tau biomarkers in this cohort. Although we observed associations between slow wave–spindle coupling and plasma Aβ measures, additional information on tau markers would have provided important context for interpreting the specificity of these findings. Contemporary biomarker frameworks emphasize both Aβ and tau as core pathological markers of Alzheimer’s disease⁶². Future studies integrating both Aβ and tau biomarkers will therefore be important to further clarify the role of slow wave–spindle coupling in the early pathophysiology of dementia. In addition, although it would be informative to classify participants according to abnormal vs. normal Aβ42/40 levels, recent work suggests that plasma Aβ measures alone do not reliably distinguish Aβ-positive from Aβ-negative individuals without additional biomarkers, particularly p-tau⁶³. As p-tau assays were not yet reliably established at the time of data collection (2019), such a classification was not feasible in the present dataset.

We acknowledge that the total sleep time observed in our sample (5.5 to 5.7 h) appears relatively short. Sleep was recorded in a laboratory setting, which has been shown to differ from habitual sleep at home, with studies reporting shorter total sleep time and lower sleep efficiency^45,46. However, population-based studies indicate that average sleep duration in older adults is typically longer. One study reported a mean sleep duration of 7.6 h (SD = 1.7 h) in older adults, highlighting substantial interindividual variability⁴⁴. Accordingly, the sleep duration in our sample was about 1 SD below the reported mean, which constitutes a limitation of this study.

In this paper, we provide evidence connecting SW–spindle coupling and Aβ-dynamics—both in unmodulated and PLAS-modulated sleep. The unique predictive strength for Aβ, even surpassing SWA and cognitive functioning levels, suggests that SW–spindle coupling is a critical physiological process related to the early pathophysiology of dementia. Newly developed targeted interventions could therefore prioritize older adults with detrimental coupling- dynamics, focusing on those who would potentially benefit the most. Furthermore, our results suggest that PLAS-induced increases in sleep markers are closely linked to favorable Aβ-responses—particularly in cognitively impaired older adults. Overall, the results suggest that PLAS is a useful tool that could yield favorable outcomes for Aβ levels and therefore help in fighting increasing incidence rates of dementia.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: MAZ, MWStudy Design: MAZ, MWData Collection: MWData Analysis: MW, MAZ, CETVisualization: MW, MAZWriting—original draft: MWWriting—review: MAZ, CET, KWWriting—editing: MW, MAZ.

Funding

Declaration.

This work was supported by Dementia Research: Synapsis Foundation Switzerland, in collaboration with the Peter Bockhoff Foundation, the Heidi Seiler Foundation, and the Kurt Fries Foundation: grants 2018-PI02 & 2021-CDA03. This work was further supported by the Swiss National Science Foundation (SNSF): grant number 215333.

Data availability

The datasets generated and analyzed during this study involve sensitive clinical and neurophysiological data from cognitively impaired participants and cannot be made publicly available in accordance with Swiss Human Research Act (HFG) protections. However, a controlled-access dataset may be made available upon reasonable request to the corresponding author, subject to: 1) Submission of a research plan and appropriate ethics approval, or an equivalent document outlining the purpose and justification for the request (e.g., manuscript review, replication study) 2) Signing of a data use agreement to ensure privacy protections and appropriate use.

Declarations

Competing interests

The authors declare no competing interests.