Multi-night cortico-basal recordings reveal mechanisms of NREM slow-wave suppression and spontaneous awakenings in Parkinson's disease

Abstract

Sleep disturbance is a prevalent and disabling comorbidity in Parkinson's disease (PD). We performed multi-night (n = 57) at-home intracranial recordings from electrocorticography and subcortical electrodes using sensing-enabled Deep Brain Stimulation (DBS), paired with portable polysomnography in four PD participants and one with cervical dystonia (clinical trial: NCT03582891). Cortico-basal activity in delta increased and in beta decreased during NREM (N2 + N3) versus wakefulness in PD. DBS caused further elevation in cortical delta and decrease in alpha and low-beta compared to DBS OFF state. Our primary outcome demonstrated an inverse interaction between subcortical beta and cortical slow-wave during NREM. Our secondary outcome revealed subcortical beta increases prior to spontaneous awakenings in PD. We classified NREM vs. wakefulness with high accuracy in both traditional (30 s: 92.6 ± 1.7%) and rapid (5 s: 88.3 ± 2.1%) data epochs of intracranial signals. Our findings elucidate sleep neurophysiology and impacts of DBS on sleep in PD informing adaptive DBS for sleep dysfunction.

Fig. 1. Methodology, data collection and analysis procedures.

A Schematic of the RC + S system setup for recording intracranial cortical Field Potentials (FP) in participants. B Illustrations of the placement of RC + S sensing depth electrodes in subcortex (middle and right) for both Subthalamic Nucleus (STN) and Globus pallidus internal (GPi) and cortical ECoG locations (left). Example image from PD2 and PD3 participants. C Setup of the Dreem2, portable headband for recording in-home overnight polysomnography. D Illustration of a single night of sleep recording in a PD participant (DBS ON) with polysomnography (purple) showing sleep stages (right y-axis; AW: awake; RM: REM; [N1, N2, N3]: NREM) and simultaneous cortical (top 2 panels) and subcortical (bottom 2 panels) spectrogram of FPs from both hemispheres showing multi-frequency changes across sleep stages where the x-axis is time and y-axis (left) is frequency (Hz). FP was recorded bilaterally from cortical and subcortical regions. E Flowchart of data analysis and preprocessing procedures for multi-night sleep dataset of all participants (n = 5; ~10 nights per participant) and ON/OFF dataset (2 nights per participant) of PD participants (n = 4). F Representative traces of the RC + S FP time series (5 s epochs) in all sleep stages from cortex (left column) and subcortex (right column; STN). Columns share scale bars and rows share color legends (Wake, REM, N1, N2, and N3). Data from one PD participant (PD2) with ON stimulation from the left hemisphere. G Comparisons of spectral powers of intracranial FPs among sleep stages in cortex (left) and subcortex (right) for a single participant across multiple nights (n = 12; PD2; DBS ON; 5 s epochs; averaged across each night; data pooled from both hemispheres; shares color legend with F). Data are presented as mean ± SEM. Spectral comparisons for all participants are provided in Supplementary Fig. 4. Source data are provided as a Source Data file.

Fig. 2. Dynamic changes in power spectra and functional connectivity between cortical and subcortical regions during N2/N3 NREM sleep.

A Power spectrum changes during N2/N3 NREM with wake as baseline for all participants (n = 5) during ON stimulation in cortical (top) and subcortical (bottom) areas. y-axis shows the difference in power spectra (mean ± SEM; dB) between N2/N3 NREM and wakefulness. B Power in delta increases (top) while beta decreases (bottom) during N2/N3 NREM compared to wakefulness during ON stimulation in cortical (left) and subcortical (right) areas. Bar plots show the difference in spectral power (mean ± SEM; each data point shows average difference in dB across one night). C During OFF stimulation, delta power increases while beta decreases in N2/N3 NREM compared to the wakefulness in PD participants (n = 4) in cortical (top) and subcortical (bottom) areas (difference in power spectra in dB; mean ± SEM). D Difference in cortical spectral power between ON and OFF stimulation (ON power-OFF power; each colored line for one participant; mean ± SEM in gray) in 4 PD participants in N2/N3 NREM (top), showing increased delta and decreased alpha and sigma activities (8–15 Hz) while ON stimulation. The spectral power in subcortical regions didn’t show any statistically significant difference (bottom). E Changes in cortical-subcortical spectral coherence (mean ± SEM) during N2/N3 NREM with wakefulness as baseline for all participants (n = 5) during ON stimulation. y-axis shows the difference in spectral coherence between N2/N3 NREM and wakefulness. F Total difference in spectral coherence in delta (left) and beta (right) during N2/N3 NREM compared to wake during ON stimulation. Barplots show difference in spectral coherence (mean ± SEM; each point shows average difference in spectral coherence across one night). G During OFF stimulation, delta coherence increases while beta coherence decreases in N2/N3 NREM compared to the wakefulness in PD participants (n = 4; mean ± SEM). Data from both hemispheres were pooled for all panels. Baseline is shown as horizontal line at 0 for A, C, D, E, and G. For B and F: n = 12 (PD2); n = 11 (PD3); n = 11 (PD7); n = 10 (PD9); n = 9 (Dystonia). Source data are provided as a Source Data file.

Fig. 3. Inverse relationship between subcortical beta and cortical delta activities during N2/N3 NREM sleep.

A Example of subcortical beta (purple) and cortical delta (green) power (PD3; single night; ON stimulation; smoothed with 20-point Gaussian kernel) depicting the inverse relationship. B Average Spearman’s rho correlation between subcortical beta and cortical delta power for 4 PD participants in ON (left; mean ± SEM; each point shows overnight correlation) and OFF (right; stem plots; n = 4; single night per participant) stimulation. C Scatter plots depicting the correlation between subcortical beta (STN: brown, red; GPi: blue, light blue) and cortical delta power in 4 PD participants (ON stimulation; 5 s epochs; each plot is single night data pooled from both hemispheres). LME models constructed for cortical delta with subcortical beta as fixed and hemisphere as random effect (PD3: β = −0.58, p value = 0; PD9: β = −0.87, p value = 1.2e−144; PD2: β = −0.41, p value = 1.8e−129; PD7: β = −0.41, p value = 1.3e−139). D Normalized cross-correlation between subcortical beta and cortical delta power (mean ± SEM) showing the subcortical beta preceding cortical delta in PD participants (ON stimulation). The bar plot (left; each point shows overnight lag) shows lags in subcortical beta compared to cortical delta. Example of cross-correlation showing the lag in subcortical beta with normalized cross-correlation vs. lag time (s) for subcortical beta with cortical delta as reference (right; PD2; single night; ON stimulation; dashed vertical line is zero-lag). E Interactions between cortical delta and beta. The bar plot (left) shows average Spearman’s rho correlation between cortical delta and beta power (mean ± SEM; 4 PD participants across multiple nights; each point shows overnight correlation; ON stimulation). The scatter plots (middle and right) show cortical delta and beta power (ON stimulation; 5 s epoch; single nights for PD2 and PD9). LME models were similar to C (PD2: β = 0.18, p value = 4.7e−12; PD9: β = −0.36, p value = 1.7e−60). For barplots in B (ON stimulation), D and E: data pooled from both hemispheres with n = 12(PD2), n = 11(PD3), n = 11(PD7), and n = 10(PD9). Data from all panels are from N2/N3 NREM. All p values were two-sided. Source data are provided as a Source Data file.

Fig. 4. Changes in N2/N3 NREM spectral power before spontaneous awakenings.

Subcortical beta increases before spontaneous awakening. A Subcortical beta power (mean ± SEM; 5 s epochs; 4 PD participants; ON stimulation; data pooled from both hemispheres) during N2/N3 NREM to wakefulness transitions (left). Vertical dashed-line (purple) shows awakening time. x-axis shows time since N2/N3 NREM sleep onset (left) and time since awakening (middle). The black line (top; Norm of RC + S accelerometry; mean ± SEM; rescaled with min-max normalization) shows across-participant movement for all N2/N3 NREM to wakefulness transitions. The bar plots (mean ± SEM) show change in subcortical beta power during immediate pre-awakening N2/N3 NREM (−7.5 s, top) and early post-awakening (+12.5 s, bottom) compared to average subcortical beta power in deep N2/N3 NREM (N2/N3 NREM data after 40 s from N2/N3 NREM onset to 40 s before awakening). Data pooled from both hemispheres. The average early post-awakening (+12.5 s; p value = 0.003) and immediate pre-awakening N2/N3 NREM subcortical beta powers (−7.5 s; p value = 0.0003; inset zoomed plot shows the rise of beta; black arrow shows −7.5 s) are higher compared to deep N2/N3 NREM. B Same as A, for cortical beta showing no significant trend across participants for both pre and post-awakenings. C Same as A, for subcortical delta showing a significant reduction across participants for post-awakenings (+12.5 s) compared to deep N2/N3 NREM (p value = 0.03). D Same as A, for cortical delta power which gradually increases as sleep deepens. The average early post-awakening (+12.5 s) delta powers are lower than those during deep N2/N3 NREM (p value = 2.4e−20). The average early post-awakening (+12.5 s) cortical delta power is also lower than the immediate pre-awakening N2/N3 NREM delta power (−7.5 s). For barplots in all panels, LME models were constructed for bandpower with deep N2/N3 NREM vs. pre/post-awakening and disease states as fixed and participants as random effects (n = 1022; 5 participants) with two-sided p value < 0.05*, <0.01** and <0.001***. Source data are provided as a Source Data file.

Fig. 5. Classification of N2/N3 NREM vs. wakefulness with cortical data.

A Flowchart describing the machine learning (ML) model generation using support vector machine (SVM) and performance evaluation. B Performance of participant-specific ML models for N2/N3 NREM vs. wakefulness classification for all PD participants (n = 4) with classical 30 s epoch window in terms of confusion matrices (left) and receiver operating characteristic (ROC) performance (right). C Same as B, for 5 s epoch window. D Bandpower feature importance and ranking where x-axis represents 6 bandpower features and y-axis shows average mutual information between bandpower and N2/N3 NREM and wake state across all PD participants (mean ± SEM; n = 4; each dot is one participant). 5 s data epochs were utilized. E Depiction of the top three bandpower features (delta, beta and gamma) in a scatter plot for data from N2/N3 NREM to wake transitions. Data points represent 5 s epochs from a single PD participant (PD2). Color bar (left) shows the time around awakening in seconds. F Performance of the ML models trained on 5 s epochs shown in C during N2/N3 NREM to wake transitions. The x-axis represents time in seconds around awakening and y-axis is wake classification by the ML models across all transitions of the participant (mean ± SEM) with n = 86(PD2), n = 104(PD3), n = 163(PD7), and n = 59(PD9). The vertical black dashed line shows awakening time and the horizontal green dashed line represents 50% average wake detection by the models. For all panels, left and right side data were pooled. For ground truth of 5 s epochs, actual awakening events within the classical 30 s sleep epochs were determined with EEG and accelerometry data by a board certified sleep physician and then segmented into N2/N3 NREM and Wake 5 s segments (see “Methods”). Source data are provided as a Source Data file.