Beyond RBD: covert REM sleep abnormalities in Parkinson’s disease

Abstract

REM sleep behavior disorder (RBD), involving REM sleep without atonia (RSWA), is a prodromal marker of Parkinson’s disease (PD). We investigated whether RBD may represent the “tip of the iceberg” of pathology driving wider REM sleep disruptions. In this study, 25 early-stage, drug-naïve PD patients and 23 age-matched healthy controls underwent overnight polysomnography with high-density EEG (256 channels) to assess how sleep patterns relate to generalized motor learning (GML). PD patients showed no significant differences in overall sleep architecture but exhibited elevated RSWA levels, with 9 individuals (36% of the PD group) meeting criteria for excessive RSWA (PD^+RSWA). Spectral analysis revealed increased occipital-parietal theta activity during REM sleep in PD^+RSWA patients, which correlated with disease progression. In both healthy participants and PD patients, sleep benefited consolidation of overnight GML. However, only in PD patients, and particularly in PD^+RSWA, REM sleep duration was negatively correlated with behavioral benefits in the GML task. These findings suggest that subtle REM sleep abnormalities extend beyond clinically defined RBD and may contribute to impaired sleep-dependent memory consolidation in early PD. REM sleep disruptions could serve as an early neurophysiological marker of disease progression and a potential target for therapeutic intervention.

Introduction

Sleep is a fundamental physiological necessity, supporting both cognitive and physical functioning. Its distinct stages, including rapid eye movement (REM) sleep and non-REM (NREM) stages (N1–N3), are classified based on characteristic electrophysiological signals recorded via polysomnography (PSG), including electroencephalogram (EEG) activity, electrooculogram (EOG), and electromyogram (EMG)^1,2. NREM sleep progresses through increasingly deeper stages: from N1, the lightest stage, characterized by the disappearance of alpha waves, across N2, which introduces sleep spindles and K-complexes, to the deepest stage, N3 (also called slow-wave sleep; SWS), dominated by high-amplitude slow (<4HZ, “delta”) waves, and NREM stages N1-N3 are also associated with progressively reduced muscle tone. REM sleep, by contrast, is marked by desynchronized EEG activity with prominent theta waves, REMs, and muscle atonia.

A growing body of evidence demonstrates that neurodegenerative diseases, including PD, are associated with sleep disruptions. While PD is primarily defined as a neurodegenerative motor disorder, accumulating evidence reveals that sleep disturbances and cognitive impairments are fundamental non-motor hallmarks of the disease. Moreover, sleep disruptions serve as an early biomarker of PD and may contribute to its underlying pathophysiological mechanisms^3,4. Among the spectrum of sleep alterations linked to PD, REM sleep behavior disorder (RBD) stands out as its most striking manifestation. The defining feature of RBD is REM sleep without atonia (RSWA), characterized by inadequate suppression of muscle tone during REM sleep^1,5. RBD serves as a robust and highly specific predictor of neurodegeneration with over 90% of individuals pheno-converting to PD or other alpha-synucleinopathies within 14 years, surpassing the predictive power of all other prodromal biomarkers⁶. RBD represents a marker of REM-sleep abnormalities that is relatively easy to identify (i.e., one that presents in behavior), and its prevalence among PD patients escalates with disease progression^7–9. Even in the absence of overt RBD among PD patients, the underlying pathology manifests as elevated muscle tone during REM sleep¹⁰. The intricate relationship between RBD and disease progression is further evident by findings that PD patients with RBD endure more severe motor deficits^11,12, and exhibit accelerated motor decline associated with RSWA severity¹³. Similarly, cognitive impairment and dementia in PD strongly correlate with the degree of RSWA¹⁴ and RBD severity¹⁵.

Previous research has identified EEG alterations during REM sleep associated with cognitive decline in PD patients with and without RBD^16–18, as well as in iRBD patients^19,20. It was further suggested that these altered EEG patterns represent heightened cortical arousal and diminished neural plasticity, which may account for the cognitive deficits associated with RBD^18,21. It is plausible that a shared underlying neuronal deficit during sleep disrupts both EEG patterns and cognition, resulting in attenuated post-sleep gains among PD patients^22,23.

Among its myriad functions, sleep is particularly important for learning and memory consolidation^24–26. Within the motor domain, motor sequence learning paradigms serve as classic tools for investigating these processes, wherein discrete, repetitive movements are progressively integrated into coherent sequences²⁷. Following an initial acquisition period of the sequential motor skill, an additional late phase can occur in which the acquired knowledge undergoes further modification during the process of consolidation^28,29. Importantly, specific aspects of consolidation are sleep-dependent, with both REM sleep and NREM sleep implicated as critical mediators²⁴. In our recent work, we leveraged an extended version of the traditional sequential task and demonstrated that sleep benefits performance accuracy in a generalized motor learning (GML) in both PD patients and in healthy participants, whereas equivalent wake periods confer no such benefit³⁰. However, given the evidence linking REM sleep abnormalities to neurodegeneration, it remains unclear whether sleep disruptions are associated with sleep-dependent motor learning in PD.

We hypothesized that RBD represents merely the “tip of the iceberg” within a broader spectrum of covert REM-sleep abnormalities which may further influence cognitive processes. To this end, the present study aims to broaden the understanding of PD-related REM sleep alterations and explore their relationship with sleep-dependent consolidation of motor learning. We conducted full PSG sleep recording in a laboratory setting equipped with high-density (256-channel) EEG (hd-EEG) in early-stage, drug-naïve PD patients and healthy adults of similar age, investigating REM sleep abnormalities beyond RBD, and whether such abnormalities may be associated with behavioral measures of overnight motor learning in the GML task.

Results

Participant characteristics

A total of 34 drug-naïve individuals with recently diagnosed PD and 29 healthy, age-matched controls were recruited. As shown in Fig. 1, all participants performed the GML task before sleep and after waking, alongside overnight PSG recording. Four subjects were excluded from the study after scoring below 24 on the MoCA (1 control, 3 PD), and an additional six subjects (4 control, 2 PD) were excluded due to moderate-severe sleep apnea. Five subjects (1 control, 4 PD) were excluded due to the absence of REM sleep or insufficient overall sleep time (≤2 h) in the night monitored at the lab. The final sample comprised of 25 participants with PD and 23 healthy, age-matched controls.

Fig. 1. Study design overview.

Participants from both the control and PD groups completed the General Motor Learning (GML) task prior to sleep and immediately after waking. Overnight, high-density polysomnography (hdPSG) was employed to record detailed sleep architecture, including REM sleep, NREM stages, and potential abnormalities such as REM sleep without atonia. Behavioral performance on the GML task was assessed pre- and post-sleep to evaluate overnight improvement. Simultaneously, physiological signals (e.g., respiratory measurements, muscle activity) alongside spectral EEG measures were collected and analyzed to explore their relationships with sleep-dependent motor learning. The figure was generated using BioRender (https://BioRender.com).

As summarized in Fig. 2, PD patients had a mean Hoehn & Yahr score of 1.4 ± 0.6, showing significantly impaired motor function compared to controls (UPDRS part III, p = 1.010e-9), while no significant differences were found in age, global cognitive function, or memory between groups (p > 0.05). Memory-specific differences were also not observed between groups, as shown in Supplementary Table 1.

Fig. 2. Summary of study variables and corresponding forest plot illustrating correlation statistics with post-sleep improvement.

Between-group comparisons are presented as mean ± standard deviation along with associated p-values. The forest plot depicts 95% confidence intervals for Spearman correlations between study variables and post-sleep improvement for the control group (blue) and PD group (red). Significant correlations (p < 0.05) with positive (Spearman’s r > 0) or negative (Spearman’s r < 0) associations are marked with triangles, color-coded by group. MoCA Montreal Cognitive Assessment, DGS digit span task, UPDRS-III unified Parkinson’s disease rating scale—Part III (Motor Examination), H&Y Hoehn and Yahr scale, TST total sleep time, REM rapid eye movement sleep N1 non-rapid eye movement sleep stage 1, N2 non-rapid eye movement sleep stage 2, SWS slow-wave sleep, RSWAi REM sleep without atonia index, ERSWA excessive REM sleep without atonia, RBD REM sleep behavior disorder.

Sleep characteristics and REM sleep without atonia

The two groups did not exhibit significant differences in terms of sleep architecture parameters or sleep spindle characteristics. However, distinct differences were evident with respect to RSWA, and 9 PD patients (36%) met the criteria for excessive RSWA(PD^+RSWA), whereas the remaining 16 PD patients (PD^−RSWA) and all 23 control subjects were classified as having no excessive RSWA (Fig. 2 and Supplementary Fig. 1). Notably, of the 9 PD^+RSWA patients, only 6 exhibited motor events accompanied by dream-enactment behavior, thus meeting the clinical criteria for an RBD diagnosis. These individuals also showed a trend toward more advanced clinical characteristics (Supplementary Table 2), reaching marginal significance for Hoehn & Yahr scores (p = 0.049).

EEG spectral abnormalities during REM sleep

To further explore the electrophysiological signatures of abnormal REM sleep, we examined spectral and topographical aspects of the EEG during epochs of REM sleep (Methods), separately for individuals categorized as PD^−RSWA and PD^+RSWA. Scalp topographies across EEG frequency bands (Fig. 3A) revealed a significant difference between groups in the theta frequency band (4–8 Hz; Z = 3.44, p = 0.041). Cluster-based permutation tests further revealed elevated theta activity in the PD^+RSWA group compared to healthy controls, predominantly over occipital and parietal regions (Fig. 3B, 64 electrodes, corrected p < 0.025). A post-hoc ROI analysis focusing on these electrodes (Fig. 3C) pinpointed the effect to the higher theta frequency bins (6–8 Hz, t = 3.78, p < 0.000). Interestingly, no significant differences were observed between the PD^−RSWA and control groups, suggesting a distinct electrophysiological profile associated with the presence of excessive RSWA.

Fig. 3. Spectral and topographical differences during REM sleep.

A Relative power across frequency bands (delta: 1–4 Hz; theta: 4–8 Hz; alpha: 8–12 Hz; sigma: 12–15 Hz; beta: 15–25 Hz; and gamma: 25–40 Hz) is shown for control participants (left column), PD-RSWA participants (PD subjects without excessive RSWA, 12 participants, middle column), and PD^+RSWA participants (PD subjects without excessive RSWA, 9 participants, right column). B Topographical t-maps depict the results of unpaired t-tests comparing control to PD^-RSWA participants (left column), control to PD^+RSWA participants (middle column), and PD-RSWA vs. PD^+RSWA participants (right column). Black crosses denote electrodes with uncorrected p-values < 0.025, while white crosses indicate electrodes that belong to statistically significant clusters (p < 0.025) after correcting for multiple comparisons. C The ROI analysis illustrates the distribution of theta power in control participants (blue), PD^−RSWA (red), and PD^+RSWA (yellow). Violin plots, alongside box plots, show the average of significant nodes across low (4–6 Hz) and high (6–8 Hz) theta frequencies. Group differences were analyzed using a two-way ANOVA, followed by post-hoc independent t-tests, with corrections for multiple comparisons using the false discovery rate (FDR) method. Statistically significant differences are denoted by ****p < 0.0005.

A correlation analysis was conducted within the PD group to examine the relationships between theta power and specific variables. Importantly, no significant correlation was found between theta power and RSWAi at the individual subject level (r = 0.15, P = 0.478), suggesting that elevated theta activity is not directly linked to RSWA events and is not driven by non-neuronal muscle artifacts. Nevertheless, theta power exhibited significant correlations with both the H&Y disease stage (r = 0.24, P = 0.005; Supplementary Fig. 2) and motor symptom severity, as measured by the UPDRS-III scale (r = 0.83, P = 0.039; Supplementary Fig. 2). These findings suggest that, similar to RSWA, theta power alterations reflect intrinsic neuropathological changes associated with PD, emphasizing its potential as a biomarker for tracking disease progression.

Behavioral performance on the GML task

To explore possible behavioral manifestations and characterize sleep-dependent consolidation of newly acquired motor learning, participants engaged in the GML task³⁰ prior to sleep and immediately upon awakening (Fig. 1). Initial performance assessments revealed no significant differences between the two groups in speed (Z = 1.06, p = 0.290, Supplementary Fig. 3) or in accuracy, measured as the number of correct responses within a five-element sequence (Z = 1.06, p = 0.290, Supplementary Fig. 3).

To evaluate the potential contribution of overnight sleep to motor skill improvement, we quantified the changes in task performance between the two sessions, focusing on accuracy, which was previously identified as the primary sleep-modulated component of the GML task³⁰. As illustrated in Fig. 4A, healthy control subjects demonstrated an average post-sleep improvement of 4.8 ± 4.0% (Z = 4.06, p = 0.000), whereas participants from the PD group exhibited an average increase of 3.4 ± 4.7% (Fig. 4A; Z = 2.96, p = 0.003). No significant difference was detected between the PD and control groups (Z = 1.32, p = 0.186) or between PD subgroups (Supplementary Fig. 4, PD^−RSWA vs. PD^+RSWA; Z = 0.40, p = 0.692).

Fig. 4. Impact of REM sleep on overnight task performance and group-specific correlations.

A Violin plots with accompanying box plots show mean pre- and post-sleep GML task performance for control participants (blue) and PD participants (red). Individual performance scores, calculated as the average number of correct responses, are represented by black circles, with gray lines connecting each participant’s pre- and post-sleep scores. While no significant differences were found between the two groups for pre- or post-sleep performance, both groups exhibited significant post-sleep improvement. Statistically significant differences are denoted as **p < 0.005 and ***p < 0.0005. B Correlation plots illustrate the relationship between GML task improvement and REM sleep duration in control (blue triangles) and PD participants (red circles). A significant negative correlation was identified in the PD group (right panel), while no significant correlation was observed in the control group (left panel). Marker size reflects RSWAi levels, with individuals displaying excessive RSWA (PD^+RSWA) highlighted by yellow edges. C The PD group was further subdivided into PD^−RSWA (without excessive RSWA, n = 16, left) and PD^+RSWA (with excessive ^RSWA, n = 9, right). A strong negative correlation between REM sleep duration and task improvement was found in the PD^+RSWA subgroup, while no significant correlation emerged in the PD^−RSWA subgroup. Improvement indices were calculated as the difference in correct responses between pre- and post-sleep sessions. Black lines represent Spearman regression fits, with shaded areas indicating 95% confidence intervals.

Sleep-dependent consolidation and REM sleep disruption

To further determine whether specific sleep components could significantly contribute to GML improvement, we conducted correlation analyses between individual post-sleep gains and sleep parameters. In line with previous literature^24,31, healthy controls showed a positive correlation between post-sleep task improvement and both N2 sleep duration and sleep spindles, while these correlations were absent in the PD group (Fig. 2 and Supplementary Fig. 5).

Next, we focused on REM sleep to test our hypothesis that REM sleep disruptions may show additional signs of pathology and may be associated with behavioral impairments. In line with this hypothesis, we observed a significant negative correlation between REM sleep duration and post-sleep improvement, exclusively in PD patients (Figs. 2 and 4B). This association was even more prevalent in PD patients with excessive RSWA (PD^+RSWA). Indeed, as illustrated in Fig. 4C, separate analyses revealed a robust negative correlation within the PD^+RSWA subgroup (r = −0.83, p = 0.008), whereas the negative correlation in the PD^-RSWA subgroup was not significant (r = −0.35, p = 0.187). Re-analyzing sleep-disturbance measures as predictors across the entire PD cohort via a generalized linear regression model also confirmed that, when combined with excessive RSWA, REM sleep duration shows a strengthened negative association with task improvement. Importantly, RSWAi was determined to be a modulator uniquely relevant to REM sleep, exerting no influence on the correlation between N2 sleep and overnight task improvement (r = 0.18, p = 0.644). Together, these findings suggest that the disrupted overnight consolidation observed in PD patients, particularly those exhibiting excessive RSWA, represents a behavioral manifestation of the underlying REM sleep abnormalities, further reinforcing the critical role of REM sleep integrity in motor memory processes.

Discussion

This study unveils covert REM sleep abnormalities in early-stage PD, offering a window into the nascent stages of neurodegeneration. By comparing REM sleep characteristics and their interplay with motor learning consolidation in drug-naïve PD patients and healthy controls, we uncovered distinct pathological features of REM sleep. These included increased EEG theta activity and elevated RSWA, which was associated with disruption of the delicate scaffolding of sleep-dependent memory stabilization. Together, these findings suggest that sleep pathologies in PD extend beyond the overt manifestations of RBD, uncovering a more insidious neurophysiological landscape that could serve as a promising diagnostic avenue for early-stage PD neurodegeneration.

Importantly, our cohort consisted of drug-naïve, early-stage PD patients, helping mitigate confounding factors such as advanced disease progression or medication impacts, enabling a clearer examination and interpretation of the effects. The absence of significant group differences in broad sleep architecture parameters and sleep spindles contrasts with previous studies^32,33. We believe this largely stems from focusing on early-stage, drug-naïve patients. In this regard, our study overcomes these confounders and highlights the need to look beyond global sleep architecture (i.e., sleep stages) in early-stage patients to uncover more subtle alterations in sleep activities such as those present in REM sleep.

Our approach to examining RSWAi, as opposed to a binary RBD classification, enabled us to move beyond the robust behavioral manifestations of RBD and probe deeper into subtle neurophysiological measures. While our findings on RBD incidence align with established statistics³⁴, RSWAi emerged as a more sensitive indicator, identifying additional patients with EMG abnormalities during REM sleep that might otherwise have remained undetected. Our results confirm and extend previous research showing that increased EMG activity can serve as a reliable biomarker, reflecting an underlying neurodegenerative cascade and associated with disease progression in PD, including cognitive impairments^34–36. Notably, we identified a significant negative correlation between REM sleep duration and subsequent task improvement, modulated by RSWA levels and pronounced in the PD^+RSWA subgroup. Unlike the sleep-supported gains observed in healthy individuals, where sleep typically fortifies motor learning^24,25, REM sleep in PD^+RSWA paradoxically seems to undermine task performance gains, acting as a double-edged sword. The negative effect of abnormal REM sleep may partially counteract the benefits of sleep. This unexpected reversal in REM sleep’s role may be yet another echo of the neurodegenerative trajectory, reflecting the complex interplay between neuropathological disruptions and sleep.

REM sleep in PD^+RSWA patients was marked by elevated theta activity over occipital and parietal regions. This finding aligns with previous studies that identified heightened theta rhythms during REM sleep in both PD^16,17 and iRBD populations^18,19. Furthermore, elevated theta activity in isolated RBD (iRBD) has been associated with an increased risk of progressing to a defined alpha-synucleinopathy³⁷ and the emergence of mild cognitive impairment²⁰. Consistent with our findings, previous research has highlighted higher theta frequencies (6–8 Hz) as a hallmark of pathological REM sleep in PD^16,18. These spectral abnormalities, in conjunction with the distinct behavioral profiles observed in our study, underscore a clear differentiation between PD^−RSWA and PD^+RSWA groups, with only the latter exhibiting significant deviations from healthy controls. Moreover, theta power showed significant correlations with PD progression and motor symptom severity, demonstrating the detrimental impact of dysregulated REM sleep on the disease’s trajectory. The accumulation of short-term pathological indicators associated with RSWA raises the possibility that noxious processes may co-occur along these motor events, possibly accelerating neurodegeneration. Clinical guidelines for RBD treatment primarily focus on behavioral manifestations³⁸, and could neglect the broader neurophysiological disruptions that may contribute to disease progression.

In principle, any observed EEG alterations in the PD^+RSWA subgroup could potentially reflect muscle contamination from RSWA-related events. However, the lack of significant correlation between these factors, and the specificity of elevated power to the theta band (rather than broadband), render this explanation unlikely. This finding also challenges the idea that theta oscillations directly contribute to motor events during REM sleep. Alternatively, we propose that, throughout the PD neurodegenerative cascade, a pathological process unfolds within REM sleep. This process involves a complex interplay of neuronal networks that progressively disrupt sleep-related functions. As the process intensifies and spreads across relevant neural infrastructures, it gives rise to different phenotypic expressions, ranging from impaired consolidation of motor learning and elevated theta oscillations to the eventual full manifestation of RBD. We speculate that the severity of RSWA is interconnected with the degree of the underlying pathological process, with RBD representing more advanced neurodegeneration³⁹. Moreover, the risk is not restricted to PD patients with RBD, as the progression of REM sleep pathology appears to parallel the overall disease progression, including increased RSWA levels and heightened risk for the development of RBD^34–36. Thus, both RSWA and cortical EEG abnormalities may signify a more pervasive subcortical dysfunction intrinsic to PD, manifesting prominently during REM sleep.

This may also point towards a shared neurophysiological disruption that orchestrates the spectrum of abnormalities observed in the PD^+RSWA subgroup. Accordingly, pathological damage may not be confined to neural circuits regulating REM sleep atonia. Postmortem studies of idiopathic iRBD patients provide further support to this notion: α-synuclein aggregates that initially accumulate in brainstem regions controlling muscle tone, progressively spread throughout the brain, mirroring the clinical trajectory of the disease and ultimately leading to overt α-synucleinopathies⁴⁰. Thus, damage to additional regions involved in REM sleep regulation may account for both the observed EEG alterations and the impaired motor learning consolidation seen in the PD^+RSWA subgroup. Such damage could provide an explanation for both the observed EEG alterations and the impaired motor learning consolidation in the PD^+RSWA subgroup. Of particular interest are the ponto-geniculo-occipital (PGO) waves, which originate in the pontine reticular formation and travel through the lateral geniculate nucleus to the occipital cortex. These waves, a defining feature of REM sleep, are more prominent during phasic REM sleep and are associated with sleep-dependent brain functions such as memory consolidation⁴¹. Animal studies have shown that PGO waves are closely linked to immediate increases in theta amplitude within the hippocampus and occipital cortex⁴². In humans, sharply contoured EEG theta waves are proposed to serve as electrophysiological correlates of PGO waves⁴¹. Disruption of this system may thus contribute to elevated cortical theta activity, impaired memory consolidation, and the broader disturbances observed in REM sleep among PD^+RSWA patients.

Over the past decade, there has been growing recognition of the bidirectional relationship between sleep and PD pathophysiology. Although it remains to be fully elucidated whether REM sleep disturbances actively contribute to neurodegeneration or represent secondary manifestations of disease progression, our findings add to the accumulating evidence implicating disrupted sleep architecture as a potential driver of PD pathology. Alterations in the sleep–wake cycle have been associated with increased DNA damage⁴³, impaired metabolic clearance⁴⁴, heightened oxidative stress⁴⁵, and the subsequent accumulation of neurotoxic proteins such as alpha-synuclein in brain cells⁴⁴, all of which are thought to promote neuronal dysfunction and cell loss.

These findings may also carry important clinical relevance. Currently, RBD is not typically regarded as harmful in itself, and management often focuses on safety measures for patients and their bed partners to prevent injury. Although pharmacological treatments (e.g., clonazepam, melatonin) are sometimes prescribed to suppress dream-enactment behaviors, these interventions are generally reserved for patients at risk of harm³⁸. However, our findings cautiously raise the possibility that REM sleep abnormalities may warrant more proactive management. If validated in future studies, strategies aimed at preserving REM sleep integrity could hold therapeutic potential for attenuating or even preventing neurodegenerative cascades in PD. Furthermore, RSWAi may represent a more sensitive and scalable biomarker than the categorical diagnosis of RBD, enabling earlier identification and possibly treatment of at-risk individuals prior to the onset of clinical symptoms.

In conclusion, our results extended the link between REM sleep pathology and PD and point toward a common unifying mechanism giving rise to a wide array of abnormalities beyond RBD. Future studies, including longitudinal follow-up studies and investigations of at-risk populations for PD, could help expand these observations and assess the predictive validity of REM sleep pathology signatures in PD progression. Gaining a deeper understanding of REM sleep characteristics in PD holds promise for identifying and refining early-stage neurophysiological changes that could improve diagnostic precision and could guide future research and therapeutic strategies.

Several limitations of the current study should be explicitly acknowledged. As part of a data-driven approach, PD patients were included in the cohort without prior knowledge of the presence or severity of RSWA. Consequently, the sample size was relatively small, and the groups were unevenly matched, particularly with the PD^+RSWA subgroup that comprises only nine individuals, leading to imbalances in group comparisons. This limited sample size reduces the statistical power for subgroup comparisons and increases the potential for variability to influence the findings. To mitigate these concerns, we employed appropriate non-parametric statistical tests designed to account for such disparities.

Second, RSWA quantification was based solely on EMG recordings from the chin muscle. Although additional limb EMG channels could provide complementary information, this approach has been shown to offer excellent diagnostic accuracy for RSWA assessment⁴⁶.

Finally, our sleep analysis relied on a single-night laboratory recording, which is susceptible to the “first-night effect.” Sleep parameters, including RSWA levels, may vary across nights⁴⁷, potentially introducing additional variability. Nevertheless, any significant group differences observed in this study are likely to reflect robust effects that extend beyond these methodological considerations.

Methods

Participants

Thirty-four drug-naïve individuals with a clinical diagnosis of idiopathic Parkinson’s disease (PD; Hoehn & Yahr stage 1–2.5) were recruited from the Movement Disorders Clinics of Tel Aviv Sourasky Medical Center (TASMC). An additional 29 healthy individuals of similar age were recruited as a control group. The study was approved by the TASMC internal review board, and all participants provided written informed consent.

Participants were 50–80 years of age and right-hand dominant. They had no prior history or presence of neuropsychiatric disorders (with the exception of PD) and were not receiving dopamine agonists, levodopa, antidepressant drugs, or sleep medications. Screening and initial evaluation included a cognitive assessment, with inclusion criteria of a Montreal Cognitive Assessment (MoCA) score ≥24 (previously shown to serve as a sensitive measure to detect cognitive impairment^48,49) and a Digit Span Memory Test score ≥10. In addition, participants with PD also underwent a clinical evaluation by a movement disorders specialist to exclude patients with clinically detectable cognitive decline. Motor evaluation was carried out using the Unified Parkinson’s disease rating scale (UPDRS-III) with inclusion criteria for healthy individuals ≤2. Participants with moderate-severe obstructive sleep apnea were excluded from further analysis.

Experimental procedure and behavioral paradigm

Participants were instructed to refrain from alcohol and caffeine consumption and to avoid taking naps during daytime in the hours prior to the initiation of the experiment. Subjects underwent a full-night sleep recording at the TASMC, with the GML task performed prior to sleep and once again upon awakening. Sleep began no more than 2 h after the subject’s habitual sleep time.

The present study utilizes a subset of behavioral data previously published in Lanir-Azaria et al.³⁰. Briefly, memory consolidation of GML was assessed using a within-subject design, where participants were trained on a modified version of motor sequence learning task. The task requires pressing four numeric keys using the non-dominant hand, repeating a five-element sequence as quickly and accurately as possible. The task consists of eight blocks each containing a different five-element sequence presented in a random order. Unintended double key presses were corrected post-hoc and did not exceed 5% of the data.

Sleep recordings

all participants underwent an overnight laboratory hd-EEG recording (256 channels; Electrical Geodesics, Inc. [EGI]) coupled with standard PSG monitoring that included EOG, EMG recordings from the chin as well as from limb muscles in a contralateral configuration, electrocardiogram (ECG); thoracic and abdominal wall motion; oronasal airflow and oxygen saturation (SpO₂) and video. EEG Signals were referred to Cz, amplified using an antialiasing filter, amplified via an AC-coupled high input impedance amplifier (NetAmps 300, EGI), and digitized at 1000 Hz. Before the recording began, electrode impedance of all sensors was confirmed to be below 50 kΩ. Timings of recordings were based on participants’ typical sleep schedule.

Sleep scoring

Sleep stages, arousals, and respiratory events were manually scored offline according to the American Association of Sleep Medicine (AASM) guidelines (AASM; Troester et al.¹). Sleep scoring was reviewed by a board-certified sleep physician.

Automatic detection of sleep spindles was implemented in EEGLAB⁵⁰ compatible with MATLAB R2022a (The MathWorks Inc.). The methodology employed a complex demodulation transformation of the EEG signal, using a 5 Hz bandwidth centered at 13.5 Hz (spanning 11–16 Hz) to capture spindle events lasting between 0.5 and 2 s. Spindle analysis was conducted on a subset of 16 manually inspected, artifact-free central electrodes containing sleep epochs N2/N3, resampled to 250 Hz, and averaged together. Detected spindles were subsequently verified by a trained expert. Key spindle parameters extracted from this method included spindle amplitude, duration, and density (number of spindles/minute of NREM sleep) for each subject.

For quantification of REM sleep without atonia index (RSWAi),all included participants had a minimum of 5 min of continuous REM sleep data, meeting the reliable standard required for RSWA assessment⁵¹. The scoring process was conducted manually by two independent scorers who were blinded to the subject’s group. To ensure consistency, a coherence test was performed on four subjects scored by both raters (two controls, two PD patients), resulting in an average agreement of 92%.

Although limb EMG was collected, these channels were less reliable due to variable signal quality and occasional disconnections. We therefore based RSWA quantification on the chin EMG, following established methods validated for high diagnostic accuracy (AUC = 0.990) using mentalis EMG alone⁴⁶. RSWAi was calculated based on EMG activity recorded from electrodes placed near the mentalis muscles, and motor event scoring followed the SINBAR method⁴⁶. Each 30-s epoch was subdivided into ten 3-s mini-epochs. Artifacts and EMG activity associated with respiratory events were excluded. Phasic EMG activity was defined as bursts lasting 0.1–5.0 s with amplitude exceeding twice the baseline. The end of each phasic event was marked by a return to baseline or an inter-burst interval longer than 250 ms. A 30-s epoch was classified as phasic if it contained 5 or more 3-s mini-epochs with phasic events. Similarly, epochs were classified as tonic if more than 50% of the epoch showed a sustained elevation in EMG amplitude, at least twice the baseline muscle tone or greater than 10 μV above the minimal EMG amplitude. Epochs were classified as “any” EMG activity if they contained phasic, tonic, or mixed EMG activity.

To calculate the SINBAR index, the number of epochs classified within each EMG activity subtype (phasic, tonic, or any) was summed and expressed as a percentage of REM sleep. Following previous studies⁴⁶, participants were classified as exhibiting excessive RSWA if they exceeded the threshold in one or more of the following criteria: (1) phasic RSWAi greater than 10.6%, (2) tonic RSWAi greater than 8.7%, or (3) any RSWAi greater than 14.5%⁴⁶.

Motor events were also used as indices to verify the presence of dream-enhanced behavior, and RBD diagnosis was made in accordance with the AASM guidelines¹.

Data analysis

EEG preprocessing was conducted in MATLAB (The MathWorks Inc.) using a combination of custom scripts and functions from the FieldTrip toolbox⁵². The signals were first high-pass filtered at 0.1 Hz and band-stop filtered at 50 Hz to eliminate powerline interference. The data were then down-sampled to 250 Hz and subjected to a bandpass filter to retain frequencies between 0.5 and 45 Hz. Epochs of REM sleep were extracted and processed separately for each participant.

To ensure data quality, artifacts were removed through a multi-step process. First, poor-quality channels and epochs were manually identified and excluded (less than 10% of channels and epochs). Discarded channels were interpolated using a weighted average of their neighboring electrodes. Independent component analysis (ICA) was then applied using FieldTrip’s runica algorithm to eliminate physiological noise, such as eye movements, cardiac signals, and muscle activity (with less than 25% of components being removed). After component rejection, the data were restricted to 163 scalp channels (excluding channels from the face and neck) and re-referenced to the average of these channels.

The data of two control participants were excluded from further analysis due to suboptimal data quality. Their behavioral data were retained in the analysis, whereas the usage of their EEG data was refined for sleep staging, spindle detection, and RSWA scoring.

Spectral analysis was conducted for each channel by applying a Fourier transform on clean 2-s REM epochs (Welch-averaged modified periodogram with a Hamming window). This analysis was performed for all participants across each frequency band: delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), sigma (12–15 Hz), beta (15–25 Hz), and gamma (25–40 Hz). Each frequency band was normalized by calculating the power sum within each band divided by the total power summed across all bands. The normalized power was then averaged across all REM epochs for each subject, followed by averaging across subjects to assess group differences.

Data analysis was conducted using MATLAB (The MathWorks Inc.) and figures were further edited with BioRender.com. The Shapiro–Wilk test was employed to verify the normality of the data, and results are expressed as mean ± standard deviation (STD). Between-group differences in behavioral, demographic, clinical, and sleep-related measurements were assessed using the Mann–Whitney U test for continuous variables and the Chi-squared test for categorical variables. Within-group differences in task performance (pre- vs. post-sleep) were evaluated using the Wilcoxon Signed-Ranks test. Spearman’s rank correlation test was used to examine the relationship between post-sleep task improvements and experiment variables. Fisher’s z transformation was employed to calculate 95% confidence values. Statistical significance was defined as p-values < 0.05.

To assess differences in EEG spectral power during REM sleep, we subdivided the PD group into those with excessive RSWA (PD^+RSWA) and those without (PD^−RSWA), and conducted a three-group comparison. We aimed to explore spectral differences without a priori assumptions about specific frequency bands, beginning with a two-way ANOVA to compare global (averaged across 163 electrodes) relative power across all frequency bands. Significant differences (detected in the theta band) were further analysed for topographical spectral trends. To account for multiple comparisons, we employed nonparametric, cluster-based permutation statistics. This method involved pairwise group comparisons (Control vs. PD^-RSWA, Control vs. PD^+RSWA, PD^−RSWA vs. PD^+RSWA) using channel-wise independent-sample t-tests, testing the null hypothesis that both groups originate from the same probability distribution. Channels with a p-value < 0.025 and at least two significant neighboring channels were considered part of a cluster. The test statistics for the cluster permutation test were based on the largest absolute summed t-values within a cluster. A total of 1500 permutations were used to calculate Monte Carlo p-values, and significant topographical differences were identified with a p-value below 0.025 (two-sided test).

We also conducted a region-of-interest (ROI) analysis, averaging significant nodes across low (4–6 Hz) and high (6–8 Hz) theta frequencies. Group differences were tested using a two-way ANOVA, followed by post-hoc independent t-tests, corrected for multiple comparisons with the false discovery rate (FDR) method. All raw p-values are reported, with significance considered only after correction.

Author contributions

S.L.A. and N.G. conceived and designed research; N.G. secured funding and supervised the project; N.G led patient recruitment and clinical diagnosis together with J.Z; S.L.A. designed the motor learning task and performed data collection; S.L.A. analyzed the data with supervision from Y.N.; R.T. led breathing assessment and supervised clinical sleep aspects together with J.Z; J.Z. led motor symptom and atonia/RBD assessments; S.L.A. and Y.N. wrote the manuscript, with feedback from all authors.

Data availability

The raw data supporting the findings of this study will be made available by the authors in accordance with IRB data-sharing guidelines. All code used for data analysis in this study is openly available in a dedicated GitHub repository: https://github.com/NirLab-TAU/RSWAi_eeg_analysis.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41531-026-01295-x.