Enhancing sleep slow waves with auditory stimulation in people with Parkinson’s disease: a randomized proof-of-concept trial

Abstract

Background: Deep sleep supports restorative functions and has emerged as a potential therapeutic target in neurodegenerative diseases, such as Parkinson’s (PD). Phase-targeted auditory stimulation (PTAS) non-invasively enhances slow waves, the electrophysiological hallmark of deep sleep. Objectives: To investigate, for the first time, whether PTAS can be successfully applied in PD, using a wearable device at home. Setting and design: Single-center, randomized, sham-controlled, double-blind, cross-over study. Participants: Of 43 screened PD patients, 33 had subjective sleep disturbance and qualified for home screening with the device; of these, eight were excluded due to non‑feasibility of PTAS (procedural difficulties n = 1, worsened insomnia n = 2, limited neurophysiological target engagement n = 5) and two withdrew. Twenty‑three participants with confirmed feasibility were randomized.Intervention: Three nights of PTAS compared to sham stimulation using wearable devices at participants’ homes. PTAS was applied in ON and OFF windows for the entire night in 14 participants and the first 2.5 h after the first stimulus (part-night) in nine.  Main outcome: Slow-wave enhancement measured by slow-wave activity (SWA). Results: PTAS increased low-frequency SWA by almost 30% during ON windows, with corresponding increases in slow-wave amplitudes, in 21 analyzed participants (two posthoc excluded). Whole-night PTAS – but not part-night PTAS - showed a trend toward longer wake after sleep onset and lower sleep efficiency. Participants receiving part-night PTAS reported improved subjective daytime sleepiness after the third night. No relevant adverse events occurred. Conclusions: This proof-of-concept study demonstrates successful enhancement of slow waves with PTAS in PD, in participants with confirmed feasibility during screening. Our observation of the potential advantages of part-night PTAS warrants further exploration. Our data encourage longer and larger trials to determine the therapeutic potential of PTAS in PD and other neurodegenerative diseases.

Trail registration: ClinicalTrials.gov: NCT04589182 (29/09/2020).

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-53443-1.

Introduction

Parkinson’s disease (PD) is the second most common neurodegenerative disease after Alzheimer’s disease (AD). Although 1–2% of older adults aged 65 and above already have PD¹, its incidence and related economic burden are expected to further rise exponentially^2,3. PD is slowly progressive and currently incurable, often resulting in complications and loss of independence. Although PD is well-known for its cardinal motor symptoms, such as bradykinesia and tremor, nonmotor symptoms are key components of PD too⁴. Sleep–wake disturbances are among the most common and bothersome nonmotor symptoms of PD, with a significant impact on patients’ quality of life⁴. Sleep maintenance insomnia affects up to 90% of patients, excessive daytime sleepiness around 30%, and rapid eye movement (REM) sleep behavior disorder (RBD) between 30% and 80%, depending on age⁵. However, treatment options for nonmotor symptoms, especially sleep–wake disturbances, are scarce, and disease-modifying therapies remain a major unmet need.

Growing evidence supports a bidirectional relationship between sleep and neurodegeneration: Neurodegeneration causes sleep–wake disturbances, and disturbed sleep may in turn accelerate neurodegeneration⁶. Consequently, interventions that improve sleep offer unique opportunities for targeting neurodegeneration⁷. Deep sleep has emerged as an especially promising therapeutic target because several related functions may exert beneficial effects on neurodegeneration^8–10. Deep sleep is defined by electroencephalographic (EEG) slow waves, which are large-amplitude oscillations in the delta frequency range (0.75–4.5 Hz) that result from synchronous neuronal firing in thalamocortical networks¹¹. In PD, translational evidence suggests that deep sleep worsens progressively from preclinical to late stages^12–16. Deep-sleep deficiency in PD correlates with daytime sleepiness¹⁷, disease milestones such as dyskinesia and cognitive impairment^18–20, putative imaging markers of altered brain clearance²¹, pathological beta-oscillations^22,23, and clinical progression^24–27. Importantly, enhancing deep sleep pharmacologically alleviates sleep–wake disturbances in PD patients²⁸, and even reduces alpha-synuclein accumulation, the pathological hallmark of PD, in mice²⁹. Mechanistically, sleep could regulate accumulation of alpha-synuclein and other pathological proteins, such as ß-amyloid, by reducing their production or increasing clearance^30–32. Critically, modulation of deep sleep, rather than sleep per se, influences pathological proteins^33,34. Pharmacological enhancement of deep sleep is limited by side effects and tolerance issues and may alter sleep macrostructure. Thus, nonpharmacological methods that are well tolerated and allow targeted modulation of specific sleep oscillations are particularly promising, especially in older clinical populations^8,9,28,35.

Phase-targeted auditory stimulation (PTAS) is a nonpharmacological method of modulating slow waves, the EEG hallmark of deep sleep. Real-time or closed-loop acoustic stimulation are alternative names for PTAS. This method works by presenting non-arousing acoustic stimuli synchronized to a specific phase of endogenous slow waves³⁶. PTAS has been shown to enhance slow wave amplitudes when targeting the up-phase of slow waves, which is reflected in increased EEG spectral power in the corresponding frequency range, slow wave activity (SWA, 0.75–4.5 Hz)^36–38. Importantly, enhancing slow waves with PTAS has been shown to benefit deep sleep-related functions, such as episodic and motor memory consolidation³⁶, immunological processes³⁹, and even levels of pathological proteins^40,41. Novel wearable devices enable self-applied PTAS over multiple nights at home, which is an important prerequisite for translating enhancement of slow waves with PTAS into a clinical application for therapeutic purposes^42–44.

However, PTAS is less efficient in healthy older adults and those with mild cognitive impairment than in healthy young adults^43,45–47. Moreover, it is unknown whether PTAS can be used to enhance slow waves in people with PD, who in addition to older age show complex sleep–wake disturbances and altered sleep physiological processes^5,48,49. In particular, PD patients show deep-sleep deficiency compared to controls, with reduced SWA, lower slow wave density, and shorter deep-sleep duration^14,50,51.

In the current study, we translated a novel approach for self-applied, at-home PTAS, recently successfully performed in healthy older adults⁴³, into an application in PD patients with subjectively disturbed sleep (Fig. 1). We chose this specific target population because PD patients could benefit from slow wave enhancement, based on the evidence described above, and we reasoned that successful implementation would support the potential of applying PTAS in clinical populations with subjectively and objectively disturbed sleep, given that prior studies of PTAS/auditory stimulation focused on the inclusion of healthy sleepers. Translation included the personalization and optimization of sleep detection and PTAS algorithm settings to account for the high variability in sleep EEG characteristics and ensure neurophysiological target engagement. We aimed to evaluate the potential of PTAS to enhance slow waves, and to investigate possible effects of PTAS on sleep macrostructure and the subjective sleep–wake experience (e.g., sleep quality, daytime sleepiness).

Fig. 1.

Study design. (a) This is a randomized, counterbalanced, double-blind, sham-controlled, crossover study comparing the effects of three nights of phase-targeted auditory stimulation (PTAS) with three nights of sham stimulation on sleep in people with Parkinson’s disease and subjective sleep disturbance, who were randomized 1:1 after successful screening. Outcomes included sleep EEG features and subjective sleep–wake experience. (b) PTAS is a nonpharmacological method for enhancing slow waves, the EEG hallmark of deep sleep. This noninvasive method works through administration of brief (50 ms) nonarousing stimuli of multifrequency pink noise that precisely target the up-phase of putative endogenous slow waves. Participants used the Mobile Health Systems Lab SleepBand v3 (MHSL-SB), a wearable device for at-home sleep recording and PTAS, based on real-time EEG processing and automated detection of sleep and estimation of slow-wave presence^42,43. During one intervention period, the device applied PTAS using a windowed ON–OFF approach (STIM). During the other intervention period, the muted device recorded EEG along with potential stimuli (SHAM). (c) PTAS was active for the entire night in the first 14 participants (whole-night PTAS) and in the first part of the night only (part-night PTAS) in the subsequent nine participants (see methods for more details). Created with biorender.com.

We hypothesized that the effects of PTAS observed in previous studies, with increased SWA as a main effect, can be replicated in PD patients when PTAS settings are tailored to the deep-sleep alterations seen in this clinical population. We further hypothesized that PTAS may improve subjective outcomes of sleep–wake experience.

Results

Patient characteristics

Patients had a median age of 58 years, an equal sex distribution, and mild-to-moderate disease severity according to Hoehn & Yahr Stages (Table 1). Disturbed sleep was present according to the inclusion criteria based on the PDSS-2 subscale, although its severity was still moderate according to published cutoffs⁵² (Table 1). Depressive symptoms were absent or mild (Table S1). All participants were on dopaminergic medication, at modest dosages, and three on antidepressants and one on melatonin (Table 1); all medications were kept stable prior to and throughout the study in accordance with the study protocol.

Table 1.

Patient characteristics.

Characteristics	n = 21
Age (years)	58 (53, 60)
Females	10 (48%)
Disease duration (years)	5 (4, 7)
Hoehn & Yahr Stage
II	20 (95%)
III	1 (5%)
MDS-UPDRS III	22 (17, 30)
Levodopa equivalent dose (mg/d)	650 (510, 831)
Sleep-active medication and antidepressants	4 (melatonin n = 1, trazodone n = 1, sertraline n = 1, venlafaxine n = 1)
Body Mass Index (kg/m2)	26.1 (24.3, 27.1)
Parkinson’s Disease Sleep Scale (PDSS-2)	17 (15, 25)
PDSS-2 disturbed sleep subscale	9 (8, 11)
Epworth Sleepiness Scale	11 (8, 14)

We randomized a total of 23 participants (Fig. 2). After completing the study in the first 14 participants, we adapted the PTAS protocol to restrict PTAS to the first part of the night for the following nine participants in the STIM condition (see Methods and Discussion for more details), resulting in two subgroups (Table S1). Two individuals had to be excluded from analysis, resulting in a total of 21 participants in the final analysis, of which 13 had received whole-night and eight part-night PTAS. The absence of neurophysiological target engagement during screening was an exclusion criterion that applied to five out of 33 participants (15%) that had undergone a screening night (Fig. 2, see Methods for more details).

Fig. 2.

CONSORT flow chart.

Table shows median values (interquartile range).

Of 43 screened subjects, 23 were randomized, 18 excluded due to not meeting the criteria, and two decided against participation after successful screening. See Methods section for more details.

Slow-wave activity and other slow-wave features

To analyze the immediate effects of PTAS on non-rapid eye movement (NREM) sleep EEG features, we analyzed the common stimulation time in all participants, which is the first part of the night. We found significant differences in NREM sleep EEG power between STIM and SHAM, with an increase in the low-frequency range of SWA (1–1.25 Hz) similar to that observed in healthy older adults⁴³. We also found a decrease in the adjacent mid-frequency range of SWA (1.5–2.25 Hz) (Fig. 3a). Given these mixed differences in SWA when analyzing all NREM sleep windows, we next compared EEG power between STIM and SHAM for ON windows, during which stimulation was applied, and for subsequent OFF windows, when it was not, separately. Here, we found that PTAS increased SWA by up to 29.52 ± 16.82% between 0.75 Hz and 1.5 Hz during ON windows (Fig. 3b). Conversely, SWA decreased by up to 11.85 ± 6.48% between 0.75 Hz and 3.5 Hz and at 4.5 Hz during subsequent OFF windows (Fig. 3b). Therefore, the mixed differences in SWA across all NREM sleep windows corresponded to the ON–OFF-related dynamics of SWA modulation, with strong increases in the low-frequency range when stimulation was ON and mild decreases in the entire SWA frequency range after stimulation when stimulation was OFF. This pattern of SWA redistribution between ON and OFF windows is in line with previous PTAS studies in older adults and patients with mild cognitive impairment^45,46.

Fig. 3.

Effects of phase-targeted auditory stimulation (PTAS) on NREM sleep EEG power. (a) Frequency bin-wise EEG power changes across all NREM windows. The y-axis displays relative differences between STIM and SHAM in normalized EEG power across frequency bins between 0.5 Hz and 20 Hz for all windows of NREM sleep during the first part of the night, the common stimulation time across all participants. The p-values above the dotted red line indicate p < 0.05 (robust linear mixed model (RLMM), fixed factor condition) within each frequency bin (see methods). (b) Frequency bin-wise EEG power changes across ON and OFF windows separately. The y-axis displays the equivalent differences in relative NREM sleep EEG power as in Fig. 3a but for ON windows with at least one stimulus (light blue color) and subsequent OFF windows (beige color) separately. Note that the values are relative to the corresponding ON and OFF windows during SHAM. The p-values above or below the dotted lines indicate p < 0.05 (RLMM, fixed factor condition) within each frequency bin (see methods). (c) Plots show the dose-dependent relationship between number of stimuli (stim#) and changes in SWA 0.75–4.5 Hz, for differences between ON and OFF windows (left), mean SWA during ON windows (middle), and mean SWA during OFF windows after stimulation (right). The increasing levels of SWA along the number of stimuli during ON and OFF windows in SHAM indicate that more stimuli can be applied when more putative slow waves are present and SWA is higher as a consequence. The effect of the intervention is seen in the difference between the STIM and SHAM trajectories. In the left and middle plots, values are categorized and matched for the number of stimuli during STIM (blue color) or potential stimuli during SHAM (red color). In the right plot, values are categorized and matched for the number of stimuli during preceding ON windows. Asterisks indicate p < 0.05 (*), p < 0.01 (**), and p < 0.001 (***) from the RLMM analysis within each category of number of stimuli. Parentheses indicate results from exploratory analyses that followed an insignificant interaction. All N = 21. Note that y-axes are condensed for illustrative purposes.

Another study in healthy older adults had suggested a dose-dependent effect of PTAS, with more stimuli applied during ON windows leading to greater SWA increases⁴³. We confirmed this dose dependence with a robust linear mixed model (RLMM) analysis that showed a significant interaction between the condition (STIM vs. SHAM) and number of stimuli (#nstim) factors (see Methods section Statistical analysis) on SWA (Fig. 3c, left and middle). The RLMM suggested that the decrease in SWA in subsequent OFF windows was not dose-dependent (Fig. 3c, right). However, exploratory analyses suggested that differences between STIM and SHAM in SWA during OFF windows were only significant when the preceding ON windows contained four or more stimuli (Fig. 3c, right). There was no difference in the response in SWA effects in ON and OFF across the three nights (ON windows: F _{condition x night} (1,69.26) = 0.77, p = 0.38; OFF windows: F _{condition x night}  (1,69.28) = 0.82, p = 0.37; ON-OFF difference: F _{condition x night} (1,69.99) = 0.04, p = 0.85, RLMM).

Next, we explored which changes in slow-wave features underly the observed increase in SWA. We found higher peak-to-peak amplitudes of slow waves during ON windows in STIM compared to SHAM nights, whereas amplitudes during OFF window were comparable between conditions (Table 2). Slow-wave duration and the number of slow waves did not differ between sham and verum nights, neither in ON nor OFF windows.

Table 2.

Condition main effects on slow-wave characteristics by window type.

Slow-wave feature	ON windows		F(dnum, dfden), p	OFF windows		F(dnum, dfden), p
	Sham (M, SE)	Verum (M, SE)		Sham (M, SE)	Verum (M, SE)
Peak-to-peak amplitude (µV)	70.642 (2.193)	74.021 (2.193)	F(1,18.98) = 6.187, p = 0.022	63.986 (2.286)	62.396 (2.286)	F(1,18.98) = 0.967, p = 0.337
Duration (s)	0.891 (0.012)	0.899 (0.012)	F(1,18.98) = 1.826, p = 0.192	0.874 (0.011)	0.873 (0.011)	F(1,18.26) = 0.026, p = 0.870
Number of slow waves (#)	3.824 (0.087)	3.914 (0.087)	F(1,18.98) = 1.575, p = 0.225	3.809 (0.102)	3.798 (0.102)	F(1,18.98) = 0.012, p = 0.912

Peak-to-peak amplitude and slow-wave duration values represent model-estimated marginal means (± SE). #The number of offline-detected slow waves was averaged separately for ON windows and their subsequent OFF windows. Estimates and p-values were obtained using robust linear mixed models (RLMM, see methods section).

Sleep macrostructure

We found no effects of PTAS on sleep macrostructure when analyzing the first part of the night, the common stimulation time in all participants (Table S2). We then ran separate whole-night analyses on participants who had received PTAS for the entire night (whole-night subgroup, n = 13 for analysis) or only the first part of the night (part-night subgroup, n = 8 for analysis). Here, we observed trends towards more whole-night wake after sleep onset (WASO; RLMM: F _condition (1,8.9) = 4.55, p = 0.062) and lower whole-night sleep efficiency (SE; RLMM: F _condition (1,8.9) = 5.11, p = 0.05) during STIM than SHAM, but only in the whole-night subgroup (Table S3). This trend was evident for the time spent in WASO (min) but not for the number of awakenings (RLMM: F _condition (1,9.6) = 0.71, p = 0.419), indicating a tendency for longer WASO but not more frequent awakenings during STIM than SHAM in participants who had received PTAS for the entire night. Notably, this trend was not evident when analyzing only the first part of the night in these participants (results not shown), which may indicate potential disadvantages for sleep continuity when PTAS is applied for the entire night, including the later part of the night. In line with this assumption, we observed no effects of STIM compared to SHAM on whole-night sleep macrostructure in participants receiving PTAS only during the first part of the night (Table S3).

Subjective outcomes

We tested the effects of PTAS on subjective outcomes, including sleep quality, daytime sleepiness, and mood, assessed by daily visual analogue scales (VAS), daily momentary sleepiness assessed with the Karolinska Sleepiness Scale (KSS), and PD-related sleep disturbances assessed with the 15-item Parkinson’s Disease Sleep Scale, version 2 (PDSS-2) and its disturbed sleep subscale pre- and post intervention⁵³. Previous studies suggested that any beneficial effects of PTAS might require several nights of intervention^40,41. Therefore, we also tested for interaction effects between the condition and night factors. We found a significant interaction of condition and night on subjective daytime sleepiness assessed by VAS (F _{condition x night} (1,24.55) = 6.95, p = 0.014). Investigating this effect further, we found that PTAS was associated with a reduction in daytime sleepiness, but only after the third and last nights and not after the first two nights, and only in the part-night subgroup but not the whole-night subgroup (Fig. 4); This effect remained significant in a sensitivity analysis after excluding one statistical outlier in the part‑night subgroup (identified using Tukey’s > 1.5 × IQR criterion; F _{condition × night} (1, 21.23) = 7.44, p = 0.013). We did not observe other changes in subjective ratings as a function of condition or its interaction with night (Table S4).

Fig. 4.

Effects of phase-targeted auditory stimulation (PTAS) on subjective daytime sleepiness. The Y-axis shows changes from baseline (BL, means, and SEM bars) in VAS ratings (0–100) of daytime sleepiness across the three intervention nights in the two subgroups that received whole-night PTAS (left, n = 13) or part-night PTAS (right, n = 8). Only in the part-night subgroup was the interaction of condition (STIM, SHAM) and night (Night 1, Night 2, Night 3) significant (robust linear mixed model (RLMM): F _{condition x night} (1,24.55) = 6.95, p = 0.014), with significant differences after Night 3. Results remained significant after excluding one outlier. Asterisk (*) indicates p < 0.05 resulting from RLMM analyses performed for each night separately.

Because daytime sleepiness did not improve and sleep continuity (longer WASO, lower SE) showed worsening trends in the whole-night subgroup, we assessed whether changes in sleep continuity due to whole-night PTAS may have negatively influenced daytime ratings. We performed exploratory correlation analyses between changes in WASO and SE and differences in KSS and VAS ratings on sleep quality, daytime sleepiness, and mood. We averaged daily measures of WASO, SE, and subjective outcomes during each intervention and calculated relative differences in WASO and SE and differences in KSS and VAS ratings between STIM and SHAM. This analysis revealed no associations of differences in WASO or SE with differences in the KSS or VAS ratings (all p > 0.47).

We did not observe serious adverse events, and the intervention was tolerated well.

Discussion

This study is the first to establish proof of the concept that phase-targeted auditory stimulation (PTAS) can enhance slow waves, the EEG hallmark of deep sleep, in Parkinson’s disease (PD) patients with subjectively disturbed sleep. We administered PTAS over three nights using a wearable device at home, for the entire night in 14 and the first part of the night in nine participants. Our main finding is that PTAS increases slow-wave activity (SWA) in PD patients, in a dose-dependent manner. We further found that part-night PTAS improved subjective daytime sleepiness without significantly influencing sleep macrostructure. These effects were not observed with whole-night PTAS. Overall, these encouraging findings underscore the potential of PTAS as a non-invasive, nonpharmacological neurostimulation method for boosting deep sleep and related functions, even in clinical populations with neurodegenerative diseases and disturbed sleep such as PD.

Effects of PTAS on sleep in PD

Although the present study population largely differs from the mostly healthy participants in previous studies, we show that the effects of PTAS on SWA in such a clinical population are comparable to the effects in previous studies.

First, we showed that PTAS increases low-frequency SWA during NREM sleep, with corresponding increases in slow-wave amplitudes, with no evidence that the effect varied across three consecutive nights. Interestingly, PTAS showed mixed effects within the SWA frequency range during and immediately after stimulation: whereas SWA in the low frequency range increased strongly when stimulations were applied during ON windows (around + 30% in our study), SWA in the entire delta frequency range decreased mildly after stimulations during subsequent OFF windows (around − 10% in our study). This pattern resembles previous observations on SWA redistribution or reorganization, between ON and OFF windows described in healthy older adults and patients with mild cognitive impairment^45,46. The relevance of the PTAS-induced SWA redistribution remains unclear, but it could reflect a positive reorganization of SWA: First, Papalambros et al. reported that the degree of SWA redistribution between ON and OFF windows related to improvements in overnight memory consolidation^45,46. Second, Winer et al. reported strikingly different effects of low and high-frequency SWA on AD pathology, with SWA < 1 Hz relating to less and SWA > 1 Hz to more future accumulation of ß-amyloid in older adults⁵⁴. Other studies confirmed the relevance of low-SWA in the context of neurodegeneration, by showing that low-SWA, rather than full-SWA, relates to cognitive impairment in PD patients²⁰, and AD pathology in early Alzheimer diseas⁵⁵. Accordingly, the potential of PTAS for increasing SWA in the low frequency range has been proposed as a promising therapeutic approach to neurodegeneration⁵⁶. Future studies should further explore the relevance of PTAS-induced changes in SWA and how SWA needs to be modulated exactly to promote deep sleep-related functions with beneficial effects in the context of neurodegeneration.

Previous studies reported variable, strongly diminished responsiveness to PTAS in older individuals than younger ones, without SWA redistribution^43,47. It is worth mentioning that our study differs from these studies in that we pesonalized stimulation settings to establish neurophysiological target engagement with PTAS during screening and excluded participants in whom this was not achieved. This enriches our final sample with participants more likely to respond to PTAS and limits direct comparisons with prior studies that did not personalize stimulation settings to achieve a response.

Second, we confirmed that PTAS may increase SWA in a dose-dependent manner⁴³. We found that higher numbers of stimuli lead to greater increases in SWA during ON windows. Noteworthy, however, the association between higher numbers of stimuli and greater increases in SWA could also result from periods of intrinsically higher SWA by itself favoring larger stimulation-induced increases in SWA, and more stimuli being applied by the algorithm at the same time. Interestingly, the decrease in SWA during OFF windows did not show a similar dose dependence. This observation suggests that the decrease in SWA during OFF windows is not directly proportional to the intensity of the preceding stimulation. However, exploratory analyses showed that the OFF decrease reached significance only after maximum amounts of stimuli during preceding ON windows.

Moreover, we found new, albeit indirect, evidence that PTAS applied only for the first 2–3 h of the night may have advantages over whole-night PTAS. Most evidence to date consistently shows that PTAS modulates microstructural features of NREM sleep, such as SWA, but does not impact sleep macrostructure^35,57. However, a recent study in older adults showed tendencies for reductions in REM sleep, which correlated with worsening mood experience during PTAS compared to the sham condition⁴³. A reduction in REM sleep due to PTAS may appear to be a plausible explanation for such an effect on mood, especially when PTAS is applied during the second half of the night, when REM sleep and low-amplitude NREM sleep predominate. It is conceivable that increasing SWA during late NREM sleep alters the dynamics between NREM and REM sleep. If so, approaches that respect the temporal dominance of NREM and REM sleep may seem a beneficial choice, particularly for patients with neurodegenerative diseases who already show altered and often reduced REM sleep^50,58. Therefore, after completing the study in the first 14 participants, we decided to restrict PTAS to the first part of the night, during which deep sleep physiologically predominates, for the following nine participants. Our aim was to achieve effects of PTAS on NREM sleep while avoiding potential impact on REM sleep or overnight dynamics of SWA with physiologically low levels of SWA during late sleep. Interestingly, we observed that the intervention led to decreased sleep continuity when PTAS was applied for the entire night, at statistical trend level. In contrast, we observed no effects on sleep macrostructure when we applied PTAS only for the first part of the night. However, we did not observe any effects of PTAS on REM sleep, regardless of how long PTAS was applied during the night. One explanation for this difference from previous findings on the impact of whole-night PTAS on REM sleep in healthy older adults is that the amount of REM sleep is reduced in patients with neurodegenerative diseases, such as PD⁵⁰. Importantly, we did not find evidence that the subtle changes in sleep macrostructure during whole-night PTAS had an impact on subjective daytime ratings. Moreover, it should be stressed that our findings indicated longer wake time after falling asleep but not more frequent awakenings during STIM compared to SHAM nights. One possible explanation is that potential awakenings during late NREM sleep might be associated with difficulties falling asleep again, especially in a clinical population with sleep disturbance and eventually after having received SWA enhancement in the previous hours of sleep. Moreover, it is possible that the changes in sleep macrostructure during STIM were not primarily related to PTAS but to other reasons uncontrolled by the experimental design, for example, fluctuations in chronic sleep disturbance.

Importantly, the current study was not designed to systematically compare protocols with different overnight durations of PTAS, because we changed our protocol halfway through the study due to new evidence at that time⁴³. Consequently, participants who received the protocols with different PTAS durations were not matched, but even differed, by chance, in demographic and clinical characteristics, which prevents direct comparisons and definitive conclusions on this matter (Table S1). For example, the proportion of females was higher among participants with whole-night PTAS than among those with part-night PTAS. Given that sex differences exist in sleep and neurodegenerative disease, our observations could also be driven by variable responses to PTAS as a function of sex. Future studies should compare the effects of whole-night PTAS with part-night PTAS more systematically while controlling for additional confounders such as the group differences in the present study, because aligning PTAS to the homeostatic overnight dynamics of SWA could have certain advantages.

Clinical effects of PTAS in PD

We found evidence that PTAS potentially provides delayed clinical benefits. Specifically, we observed a delayed improvement in daytime sleepiness due to PTAS, which became detectable after two nights and statistically significant after three nights of PTAS in the part-night subgroup. Excessive daytime sleepiness in PD is multifactorial and largely driven by PD pathology because it is independent from nocturnal sleep according to most studies⁵⁹. However, recent evidence connects daytime sleepiness in PD to NREM sleep deficiency at the microstructural level¹⁷. Our observation of delayed improvement in daytime sleepiness is reminiscent of the delayed improvement in daytime sleepiness in patients with PD or narcolepsy treated with pharmacological slow-wave enhancement, which only fully sets in after several weeks of treatment^28,60. Therefore, our study adds more evidence suggesting that slow-wave enhancement is a potential strategy for treating daytime sleepiness in clinical populations, such as PD. Nevertheless, we consider this finding preliminary because of the small sample size in the part-night PTAS subgroup. Moreover, we did not observe improvements in other subjective outcomes. To the best of our knowledge, previous PTAS studies did not include clinical populations with sleep disturbances and did not systematically assess the effects of PTAS on subjective sleep–wake experience. However, previous work has suggested that PTAS may have delayed effects on clinical outcomes, as shown in the improvement of memory performance in older adults^40,41. Therefore, it is conceivable that prolonged administration of PTAS could be necessary, and that the intervention of three nights applied in the present study was too short, to elicit more robust clinical benefits. Accordingly, longer clinical trials applying PTAS over several weeks or months might be needed to fully assess its clinical efficacy. The availability of wearable devices for at-home PTAS offers exciting opportunities for exploring long-term effects of PTAS in humans^44,61.

Limitations and further implications

Our study has several limitations. First, due to the remote study design, it was impossible to control for the potential impact of several factors that may have influenced the outcome of this study, even with the chosen within-patient design. These factors include all aspects of behavior that can impact on sleep, including various kinds of daytime activities such as sports and napping, light exposure, bedtime routines, and dietary factors. Consequently, study participants were instructed to adhere to their regular routines as closely as possible, and for example, subjects with highly irregular sleep–wake patterns were excluded from the study. Nevertheless, a certain degree of freedom is an unavoidable part of any study that aims at translating PTAS from controlled single-night in-lab experiments to real-world at-home applications. Three of 33 participants were excluded for inability to apply the wearable device at home, including two who reported worsened insomnia. Although we consider this rate as rather low given the complexity of the present target population, this observation highlights important challenges of wearable-based interventions in clinical subpopulations that need to be considered for future applications and design of clinical trials.

Second, as mentioned above, we adjusted our protocol in the overnight duration of PTAS, without previously intending to compare PTAS protocols. This may have reduced statistical power for detecting certain effects of PTAS on clinical outcomes. Although we were able to pool the data for the immediate stimulation effect on NREM sleep EEG by restricting the analysis to the common stimulation time during the first part of the night, next-day ratings are always influenced by the whole sleep period, resulting in different conditions for the two subgroups. Our observations on potential advantagesof restricting PTAS to the deep-sleep-rich first part of the night are certainly of interest but need to be considered as anecdotal at this point, especially since participants in the two subgroups were not matched and showed demographical and clinical differences that could influence the effects of PTAS as discussed earlier.

Third, we included patients with subjective sleep disturbances as part of the study design, but objective sleep assessment, for example with polysomnography, was not included in the screening. Therefore, the risk of unknown objective sleep disorders, such as covert sleep apnea, is possibly higher among our study sample than among participants identifying as good sleepers. Nevertheless, the recruitment via our outpatient clinic and neurological practices as well as the thorough history taking and continuous sleep EEG monitoring during screening and throughout the study provided additional information and allowed to exclude participants with sleep apnea or restless legs syndrome, previously known or evident during screening, which applied to five participants. Symptoms of restless legs syndrome are frequent in PD patients, fluctuate naturally, and are difficult to distinguish from other nocturnal symptoms, especially nocturnal wearing off, i.e. the worsening of motor symptoms when the effect of dopaminergic medication wears off during the night. Therefore, depending on their intention, future studies may consider polysomnography or home screening for sleep apnea or periodic limb movements during sleep during screening. We acknowledge that the characterization of the participants’ sleep phenotype remains incomplete without polysomnography recording at study entry. Moreover, the inclusion criterion of subjectively disturbed sleep may limit generalization to all PD patients. However, our rational for this criterion was to assess the short-term effects of the intervention on disturbed sleep. Moreover, disturbed sleep is highly prevalent in PD, and the severity of disturbed sleep in our sample was moderate – not severe – based on the median PDSS-2 sleep disturbance subscore of 9 (range 0 to 20) and total score of 17 (range 0 to 60), which is just around - not well above - published cutoffs for relevant sleep disturbances and within the range reported in many unselected PD patients^52,62. Therefore, we argue that although generalization is limited, patients in our study are still comparable to a large portion of PD patients. Further, our eligibility criteria allowed us to assess the feasibility and principle functioning of wearable-based PTAS/auditory stimulation in a clinical population with disturbed sleep, which will help advance this novel noninvasive stimulation method that has been established primarily in selected healthy sleepers.

Fourth, we excluded five out of 33 patients who showed no evidence of neurophysiological target engagement with PTAS during the screening nights, because we aimed to characterize the effects of PTAS after confirming its feasibility, including evidence for appropriate delivery of stimuli and neurophysiological engagement. Therefore, our study should not be used to judge the efficacy of PTAS in PD patients. The resulting rate of 15% participants without neurophysiological engagement, i.e. probable nonresponders, is low in comparison with the range of responses in previous studies in older adults and a recent report in people with AD⁶³. Notably, we may have optimized the rate of neurophysiological engagement by individual adaptations of volume and algorithm thresholds for NREM sleep and slow wave estimation, which could optimize the application of PTAS in populations with low-amplitude slow waves, such as older adults and patients with neurodegenerative disease^64,65. Understanding individual differences in the responsiveness to PTAS is crucial for advancing the field of auditory stimulation but was not explored in the present study due to the small numbers of probable nonresponders and overall modest sample size. Therefore, future studies with larger or pooled samples, should address individual differences in PTAS response, especially in the context of aging and neurodegeneration. We excluded patients with deep brain stimulation; however, implanted deep brain stimulation devices show synergistic potential with PTAS applications⁶⁶.

Fifth, medication may have influenced slow-wave sleep; however, medication regimens were stable, strong sedatives were excluded, and the crossover design minimizes the risk of systematic confounding in one particular condition.

Outlook and perspectives

The present study contributes to translational research on the therapeutic potential of sleep interventions in neurodegenerative disease. Evidence supporting a neuroprotective role of deep sleep in PD and other neurodegenerative diseases is growing, and our study encourages further applications of PTAS to explore the potential of nonpharmacological slow-wave enhancement for targeting neurodegeneration in greater detail. Some aspects of our study may extend to clinical populations with chronic sleep–wake disturbances due to other causes and encourage further pilot studies on this behalf. Moreover, future studies should explore the potential of PTAS for targeting other symptoms of PD that may be related to the restorative function of sleep, including cognitive impairment.

Furthermore, we have demonstrated the successful application of a wearable device for at-home PTAS in a clinical population. Through remote monitoring and adaptation of settings, such wearable systems may help tailor PTAS to the specific needs of some clinical populations, for example, those characterized by low-amplitude slow waves or high variability in sleep EEG characteristics. Indeed, personalized sweet-spot approaches could be a prerequisite for translating sleep modulation, such as PTAS, into large-scale clinical applications, especially in patients with altered sleep–wake functioning.

Conclusion

This proof-of-concept study demonstrates the potential of PTAS as a nonpharmacological method for boosting sleep slow waves - the EEG hallmark of deep sleep - in people with PD. Noteworthy, stimulation settings were personalized during screening with the wearable PTAS device at home for neurophysiological target engagement, which was achieved in most participants. Future studies should apply PTAS over longer periods to fully explore its therapeutic potential for symptomatic benefits and disease-modifying properties in PD and other neurodegenerative diseases.

Methods

Ethics

This study was conducted according to the Declaration of Helsinki principles and approved by the Cantonal Ethics Committee of Zurich (reference: BASEC 2020 − 00431) and Swissmedic, the Swiss agency for drugs and therapeutic products (reference: 10000681), and registered at ClinicalTrials.gov (NCT04589182, date of first registration: 29/09/2020). All participants gave written informed consent prior to inclusion in the study. Participants received 200 Swiss francs to reimburse expenses (e.g. travel to the study site).

Study design and intervention

This was a single-center, 1:1 randomized, counterbalanced, double-blind, sham-controlled, crossover study conducted at the Department of Neurology, University Hospital Zurich, Switzerland. The study took place between Oct 2020 and May 2022, (whole-night subgroup: Oct 2020–Jul 2021; part-night subgroup: Sep 2021–May 2022). Due to the SARS-CoV-2 pandemic, the study design minimized personal contacts and visits to the study site and embraced remote assessments whenever possible to adhere to regulatory restrictions in pandemic times.

Figure 1 summarizes the study design and intervention. Participants used a wearable device for PTAS application at home, the Mobile Health Systems Lab (MHSL) SleepBand v3 (MHSL-SB)^42,43. The study technician (MS, RS) who was not involved in recruitment or data collection performed block randomization (n = 12), to assign eligible participants to equal groups starting with STIM or SHAM, using a random allocation sequence based on a sequential alphabetical list generated with R code prior to the start of the study. Personnel involved in enrolling eligible participants and assessing outcomes (SJS, JH) as well as participants were blinded. Participants were switched from three consecutive nights of PTAS (STIM) to three nights of sham stimulation (SHAM), or vice versa, after a washout period of four nights (Fig. 1a). Thus, both intervention periods took place on the same weekdays. During interventions, participants wore the MHSL-SB the entire night at their regular bedtimes in their regular sleeping environment. The experimenter instructed all participants to maintain a regular sleep–wake rhythm and maintain their habitual routines before and throughout the entire study. Adherence was assessed by regular telephone calls with the experimenter scheduled at least before, after, and once during each intervention period; daily questionnaires; and wrist actigraphy worn by the participants seven days before and throughout the entire study.

With two participants, we had to repeat the STIM intervention because more than one night was not recorded correctly due to technical failures in the initial intervention period. We had to exclude two participants - one from each subgroup - after completing both interventions due to restless legs syndrome and sleep apnea. In the final analysis, we included 21 patients, with data from 117 intervention nights available out of 126. Nine nights were unavailable for analyses owing to technical failures: six in SHAM and three in STIM.

Participants

Participants were recruited via advertisements in institutional websites, printed magazines of the Parkinson Schweiz organization, patient support groups, the movement disorder outpatient clinic of the Department of Neurology, University Hospital Zurich, and selected neurological practices in Switzerland.

Screening included a visit followed by up to four screening nights with the MHSL-SB. The consultation took place at the study site or online and at the participants’ homes and included detailed history taking and medical and neurological examination by a study physician with experience in PD and sleep medicine (SJS or JH). Motor examination was performed in the medical ON state after intake of regular dopaminergic medication for documentation of the motor subtype and Movement Disorders Society Unified Parkinson’s Disease Rating Scale Part III (UPDRS III)⁶⁷. Participants completed standardized questionnaires, including PDSS-2, Epworth Sleepiness Scale (ESS), RBD-Single-Question Screen (RBD-SQ)⁶⁸, and the Hamilton Depression Scale (HADS) and provided information about symptoms of restless legs syndrome and risk factors of obstructive sleep apnea.

Eligible participants had to have a diagnosis of PD according to international criteria⁶⁹, with mild to moderate disease severity (Hoehn & Yahr Stages II-III), and self-reported sleep disturbance defined by ≥ 7 points on questions 1–3 and 14 of the Parkinson’s Disease Sleep Scale revised version (PDSS-2)^53,70. The PDSS-2 is a self-report questionnaire composed of three factors covering [1] motor symptoms at night, [2] PD symptoms at night, and [3] disturbed sleep, from which we summed items 1–3 and 14 (bad sleep quality, difficulties falling asleep, difficulties staying asleep, tired and sleepy after waking in the morning)⁵³. We chose this approach to select participants with the most common phenotype of disturbed sleep in PD^71,72. Likewise, we excluded patients with known sleep disorders that can significantly disrupt sleep and are common also outside PD, such as sleep apnea with apnea-hypopnea index (AHI) > 15/h and restless legs syndrome, but not with sleep disturbances more closely related to PD, such as REM sleep behavior disorder and sleep disturbances rated with the PDSS-2^71,72. In case the study physician suspected sleep apnea (clinical assessment, STOP-BANG), respiratory polygraphy was performed⁷³. The most important further additional inclusion criteria were the ability to follow study procedures and stable dosing of Parkinsonian medication for at least 14 days prior to the first intervention and during the entire study. The most important exclusion criteria were known presence of neurologic (other than PD), psychiatric, or severe medical conditions, atypical or poorly levodopa-responsive Parkinsonism, cognitive impairment (Montréal Cognitive Assessment, MoCA < 24), regular use of central nervous system depressant or sleep-inducing substances, working night shifts, travelling more than two time zones in the last month before or during intervention, alcohol abuse (i.e. >0.5 l wine or 1 l beer per day), substance abuse, high caffeine consumption (> 5 servings per day, including coffee and energy drinks), implants for deep brain stimulation, inability to hear the tones of the MHSL-SB, and absent neurophysiological engagement with the PTAS during screening (see next section or more details).

The total numbers number of screened and included subjects can be seen in Fig. 2. Although subjectively disturbed sleep, with symptoms of insomnia, was an inclusion criterion, we had to exclude two participants due to worsening of sleep-onset insomnia when testing the device at home. Two participants, one from each subgroup, had to be excluded from the analysis after completing the study, because all criteria for restless legs syndrome were met (n = 1), and positional sleep apnea was suspected and later confirmed (n = 1) only towards the end of the intervention.

Screening nights took place at the participants’ homes. The experimenters (SJS, JH, LK) provided participants with the MHSL-SB, a tablet, and instructions on usage. From the data collected during screening nights, we determined eligibility for applying the device at home and the occurrence of PTAS-related neurophysiological engagement, and titrated the volume of the MHSL-SB if needed.

Phase-targeted auditory stimulation

,Phase-targeted auditory stimulation was applied with the MHSL-SB as previously described in detail⁴³. In brief, the MHSL-SB is a configurable wearable system for sleep recording and modulation that enables real-time EEG processing and enhancement of slow waves with PTAS^37,42,43, which can be optimized for use in PD patients⁶⁵.

The system records single channel EEG at the forehead (Fpz) referenced to the mastoid for real-time signal processing, eye movements (two electrooculography, EOG, channels), and muscle activity (two electromyography, EMG, channels) at the chin for post-processing purposes.

As in previous studies, we used alternating 6-s windows during which stimulation is, in principle, enabled (ON) or disabled (OFF)⁴³. The system delivers nonarousing 50 ms stimuli of multifrequency pink noise targeting the up-phase of putative slow waves whenever a slow wave is suspected during the 6s-ON windows, thus a stimulation logic is met. The stimulation logic involves four criteria that need to be fulfilled in a cascade of subalgorithms: (1) Sleep classification: NREM sleep (N2, N3) is classified based on delta and high-beta power as well as their ratio; When NREM sleep is classified and stable (i.e., continuously present for 10 min for the first NREM sleep event and for three minutes for any following event), subsequent algorithms are executed: 2) SWA classification: The presence of ongoing slow waves is estimated based on delta power; (3) EEG phase estimation: A phase-locked loop (PLL) method continuously estimates the instantaneous phase of the EEG signal, allowing to deliver auditory stimuli at a predefined target phase around 45°; (4) Beta power detection: High beta power, a proxy of awake and arousal, prevents stimulation. These algorithms have configurable thresholds, which can be tailored to individual sleep EEG characteristics. The accuracy and performance of these algorithms have been validated^42,65and remained high even when sleep slow waves show morphological alterations, such as low amplitudes⁶⁴. We assessed the accuracy of the phase-targeting by the PLL algorithm, by extracting the phases of offline detected slow waves at stimulus onset which we reported relative to the instantaneous slow-wave phase. Phase angles were averaged across all stimulation events for each condition. This analysis showed accurate online phase detection of estimated EEG slow waves with the PLL method (Figure S1).

To achieve sufficiently strong stimuli and avoid arousal induced by the stimuli at the same time, the system automatically adapts the volume during PTAS within a predefined dB range: the volume increases within predefined limits during deep sleep in the absence of arousals or decreases or pauses in the presence of arousals detected by the system. In this study, we set 52 dB as default, tested this volume during wakefulness, and titrated up to a range of 70–80 dB if needed to surpass individual hearing thresholds. In addition, we lowered the preset volume if participants reported hearing tones during the night. Applied volumes ranged between minimal 39.4 to 57.7 (median 42.8) dB and maximal 57.6 to 68.6 (median 64.9) dB.

During SHAM, all procedures were identical to the STIM condition, but the device volume was muted while recording the signal along with timing and number of theoretical stimuli that would have been applied under STIM conditions.

Previous studies applying PTAS showed that the degree of SWA enhancement varies greatly and can be absent in old-aged adults^46,47,65. This variable response could be at least partially driven by insufficient automated NREM sleep detection or insufficient volumes of acoustic stimuli. Therefore, we individualized the algorithm‘s thresholds and volume settings during screening to achieve neurophysiological target engagement, defined by a significant within-night difference (p < 0.05) in SWA and/or a visible auditory evoked potential (AEP) induced by stimuli during ON compared to OFF windows, in at least one of the screening nights. Failure to reach such an engagement was used as an exclusion criterion, which applied to five out of 33 home-screened participants (15%) (Fig. 2).

Finally, we implemented two protocols with different durations of PTAS: Whole-night PTAS: PTAS was applied for the entire night, as performed in the study using MHSL-SB in healthy older adults⁴³. Part-night PTAS: While wearing the device for the entire night PTAS was restricted to the deep sleep-rich, first part of the night: 2.5 h after the first stimulus in all but one participant who had low numbers of stimuli due to sleep fragmentation and thus received 3 h after the first stimulus. We switched between these protocols halfway after completing the first 14 participants with whole-night PTAS (posthoc defined as the whole-night subgroup) to complete the following nine participants with part-night PTAS (posthoc defined as the part-night subgroup). This switch was prompted by new findings that whole-night PTAS may affect mood through its impact on REM sleep⁴³. Accordingly, the part-night PTAS intends to align with the physiological predominance of deep sleep during the first part of the night and avoids potential impact on REM sleep, which physiologically predominates during the second part of the night. In the whole-night PTAS protocol, on average (± SD), 2037.92 ± 776.41 stimuli were applied per night; in the part-night protocol, 1589.96 ± 967.71 stimuli.

Sleep macrostructure

The MHSL-SB single-channel EEG (FpZ-A2), EOG, and chin EMG were used for scoring sleep stages for all intervention nights. Experienced sleep specialists (SF, PV, RP, SJS) scored according to the AASM criteria⁷⁴. In the whole-night subgroup, initial automated scoring used a deep learning classification⁴³ that had been retrained on a PD data set, followed by expert control, which took precedent in case of disagreement with the algorithm. In the part-night subgroup, only expert scoring was performed. From scored data, we calculated absolute and relative amounts of sleep stages (N1, N2, N3, R), duration and incidences of wake after sleep onset (WASO), number of awakenings, and sleep efficiency (SE, % TST of sleep period time).

Sleep EEG analysis

Spectral power in the sleep EEG was analyzed as previously described⁴³. In brief, EEG signals were high- and low-pass filtered at 0.5 Hz and 40 Hz, respectively, with the EEGLAB toolbox for Matlab (The MathWorks, Inc., Natick, MA)^75[,76. Then we used the pwelch function to calculate EEG spectral power between 0.5 and 30 Hz for all 6-s ON and OFF windows. We only included windows that were classified as NREM sleep without artifacts by the online algorithm and were offline scored as N2 or N3. Furthermore, we applied a semi-automated artifact rejection and automated outlier detection using the isoutlier function of Matlab^77,78.

To be consistent with Lustenberger et al. (2022), we normalized the spectral power in each 0.25 Hz frequency bin to the summed power up to 30 Hz for the spectral density analysis (Fig. 3a and b)⁴³. In line with their approach, we calculated the absolute power in the SWA range in ON and OFF windows that were grouped by the number of stimuli (#stimuli) during ON and the corresponding OFF windows (1–2, 3–4, > 4) to assess a potential dose-dependent response to PTAS (Fig. 3c). However, because spectral density analysis had revealed effects in a broader SWA range, we did not restrict this analysis to the low-frequency range of SWA but included the full SWA band: 0.75–4.5 Hz. Furthermore, we also categorized OFF windows by the number of stimuli during preceding ON windows to assess whether preceding stimulation influenced subsequent OFF windows in a dose-dependent manner.

We additionally analyzed slow-wave features other than SWA including slow-wave amplitude, duration, and numbers per ON and OFF windows during STIM and SHAM. Slow waves were identified as positive deflections between consecutive negative peaks in band-pass filtered (0.5–4 Hz, stopband 0.1 and 10 Hz, Chebyshev Type II filter) artifact free N2 and N3 epochs⁷⁹. Waves were selected if the half-wave duration between two consecutive zero crossings was 0.25–1 s. Positive amplitude was defined as the absolute voltage at the positive peak, and peak-to-peak amplitude as the absolute difference between the positive peak and the most negative peak of the same wave. Positive amplitudes were restricted to the 1st–99th percentile range, and waves outside this range were discarded⁸⁰.

Subjective outcomes

Participants answered visual analogue scales (VAS) for several symptoms and the PDSS-2 on a tablet. The PDSS-2 rates the severity of 15 typical nocturnal symptoms in PD by asking about their frequencies over one week (never, 1d/week, 2-3d/week, 4-5d/week, >5d/week)⁵³. We adapted the scale to assess symptoms over three nights, the duration of our intervention, instead of one week (0/3 nights, 1/3 nights, 2/3 nights, 3/3 nights). This adapted PDSS-2 was collected before and after each intervention (Pre, Post), to assess change during intervention (Fig. 1a). We considered the total PDSSS-2 score and items 1–3 and 14 from the disturbed sleep subscale⁵³.

Daily subjective ratings included VAS (0–100) for subjective sleep quality (how refreshing was your sleep? How good was your sleep quality? every morning), mood (every evening), and average daytime sleepiness (every evening), and the Karolinska Sleepiness Scale for assessing momentary sleepiness (every morning and evening). These outcomes were collected every day during interventions, and on that day on which the intervention started as a baseline (pre).

Sample size

In accordance with power analyses from previous studies targeting sleep-wake disturbances in PD through pharmacological enhancement of deep sleep a sample size of 12 patients is needed²⁸. However, since no studies using PTAS in this population had been performed before, we added a safety margin of up to 24 patients depending on an interim analysis after the first 12 patients. The interim analysis was omitted due to a change in the PTAS protocol (see section “Phase-targeted auditory stimulation”), and data analysis was performed when the preset study end was reached.

Statistical analysis

The statistical analysis was conducted using R Studio software (R Foundation for Statistical Computing, Vienna, Austria). We ran a robust linear mixed model (RLMM) applying the Kenword-Roger approximation method⁸¹ to compare the effects of STIM to SHAM on sleep (NREM sleep EEG power, slow wave features [amplitude, duration, numbers per ON and OFF windows], sleep macrostructure) and subjective outcomes. In these models, we included condition (STIM vs. SHAM), period (1 vs. 2), night (1, 2, 3), sequence (STIM–SHAM vs. SHAM–STIM), and in case of combined analyses, subgroup (whole-night vs. part-night) as fixed factors and subject and condition as nested random factor. For the dose–response analysis of SWA in ON and OFF windows (Fig. 3c), we averaged values across the nights and included #stimuli (1–2, 3–4, > 4) and its interaction with the condition factor as fixed factors. We repeated these analyses with data for all three nights across all windows with at least one stimulus or flag, for testing the interaction between night and condition.

For the PDSS-2 analysis, we included the time point factor (pre- vs. post intervention) and its interaction with the condition factor to assess the change induced by the intervention. For all other analyses, we included night (1, 2, 3) as fixed factor in the models. Because previous studies suggested that the beneficial effects of PTAS might require several nights of intervention^40,41, we also tested for interaction effects between condition and night in subjective outcomes. Following any significant interaction, we ran separate analyses for categories with different numbers of stimuli, time points, or nights. In the daily assessed subjective ratings, we further included the corresponding baseline value (pre) as a predictor. Any p-values < 0.05 were considered significant.

Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

SJS, RH, CRB, and AM conceptualized and designed the study. SJS, JH, and LK prepared and conducted the study under supervision of AM and CRB. RH, WK, GD, LF, CL, RS, VJ, and MS provided methodological and/or technical support and guidance. LB provided the retrained automatic scoring algorithm. SJS, SF, RP, and PV visually scored the sleep EEG data. SJS, JH, SF, KR, BM, and AM analyzed data. SJS wrote and AM revised the main manuscript text, all authors reviewed the manuscript and provided feedback and approved the submitted version of the manuscript.

Funding

Swiss National Science Foundation, Parkinson Schweiz, Swiss Neurological Society, Schweizerische Hirnstiftung.

Data availability

Data can be requested from the corresponding author upon reasonable request and in accordance with regulatory rules.

Declarations

Competing interests

The authors SJS, AM, CL, SF, VJ, KL, LB, LK, MS, RP, PV, BM, and RS declare no competing interests. The authors CRB, RH, and WK are founders and shareholders, GDP is board member, and LF is employee of Tosoo AG that has licensed the PTAS technology used in this work.