signature=bb63a14efc37a6db4583ed5a104b1872,Neurophysiological signature of gamma-hydroxybutyrate aug...

Permission

The study was approved by SwissMedic and Ethics Committee of the Canton of Zurich and registered at ClinicalTrials.gov (NCT02342366). All participants provided written informed consent according to the declaration of Helsinki.

Study design

The study followed a randomized, placebo-controlled, balanced, double-blind, crossover design. All study participants completed a screening night in the sleep laboratory, to exclude sleep-related disorders such as sleep apnea (>5 apneas/hypopneas per hour of sleep), restless legs syndrome (>5 periodic limb movements per hour of sleep), occurrence of sleep onset REM sleep episodes, and insufficient sleep efficiency (<80%) before definite enrollment into the study. The study protocol consisted of two randomized, experimental nights (GHB and placebo) separated by a washout phase of 7 days. Each experimental night was preceded by an adaptation night for habituation to the laboratory environment.

Participants

Twenty healthy, male volunteers (mean age: 25.8 ± 5.1 years) completed the study. Following criteria were required for inclusion: (i) male sex (to avoid unknown pregnancy and a potential impact of menstrual cycle on primary outcome variables); (ii) age within the range of 18–40 years; (iii) absence of somatic or psychiatric disorders; (iv) no first-degree relatives with a history of heritable psychiatric disorders such as schizophrenia, bipolar disorder, autism, and attention deficit hyperactivity disorder; (v) non-smoker; (vi) no history of regular drug use (lifetime use <5 occasions of each drug, except occasional cannabis use). No participant reported previous experiences with GHB in their life. Participants had to refrain from illegal drugs for 2 weeks and from caffeine for 1 week before the first experimental night and throughout the study. No alcohol was allowed 24 h before each study night. Participants were instructed to keep a regular sleep–wake rhythm with 8 h in bed from 23:00 to 07:00 during 1 week before the first experimental night and in the week between the two experimental nights. To ensure compliance with this requirement, participants wore an actimeter on the non-dominant arm and kept a sleep–wake diary. Both actigraphic recordings and sleep–wake diaries were qualitatively inspected by an experienced member of the study team before the experimental nights, to ensure adherence of the subject to the imposed sleep–wake schedule. No subject had to be excluded because of violation of the instructed sleep schedule. Moreover, sleep efficiency in all adaptation nights preceding each experimental night was above 80%.

All participants received a monetary compensation for study participation.

Urine immunoassay

On each test night, urine samples were taken upon arrival in the laboratory, to ensure that all participants abstained from illegal drug use (Drug-Screen Multi 12-AE, Nal von Minden GmbH, Regensburg, DE).

Drug administration

Study volunteers sleeping in the sleep laboratory were awoken at 2:30 a.m. to receive 50 mg/kg of GHB (Xyrem®; Cantonal Pharmacy, Zurich, Switzerland) dissolved in 2 dl of orange juice or placebo, matched in appearance and taste. The administered dose represents the maximal therapeutic starting dose in narcolepsy. After GHB/placebo intake, volunteers where allowed to immediately return to sleep.

EEG data acquisition

Sleep was quantified by all-night polysomnography with Rembrandt® Datalab (Version 8; Embla Systems, Planegg, Germany) from 23:00 (lights-off) to 07:00 (lights-on). The recording setup consisted of 19 EEG electrodes (Fp1, Fp2, F3, F4, F7, F8, Fz, T3, T4, T5, T6, C3, C4, Cz, P3, P4, Pz, O1, 02) according to the 10–20 system [26], a bipolar electrooculogram, a submental electromyogram (EMG), and an electrocardiogram (ECG). The individual EEG electrode coordinates were marked by cutting the subjects’ hair at the electrode position, to ensure that the electrodes were placed at the very same place in both experimental conditions.

All data were recorded with dedicated polygraphic amplifiers (Artisan®, Micromed, Mogliano Veneto, Italy). As in previous studies, the analog signals were conditioned by a high-pass filter (EEG: −3 dB at 0.15 Hz; EMG: 10 Hz; ECG: 1 Hz) and an antialiazing low-pass filter (−3 dB at 67.2 Hz), digitized and stored with a resolution of 256 Hz (sampling frequency of 256 Hz) [27, 28].

Sleep stage scoring

For sleep scoring, the C3-A2 derivation was used. Sleep variables were visually scored based on 30 s epochs according to the criteria of the American Academy of Sleep Medicine [29]. Movement- and arousal-related artifacts were visually identified and excluded from analyses. The following sleep variables were computed: (i) duration of sleep stages (N1, N2, N3, and REM sleep, and wakefulness); (ii) duration of NREM sleep (time spent in stages N2 and N3); (iii) total sleep time (TST; time spent in N1, N2, N3, and REM sleep); (iv) time in bed (TIB; time between lights-off and lights-on); and (v) sleep efficiency index (SEI = [TST/TIB] × 100). Sleep variables were computed for the first and second halves of the sleep episodes, as well as for the entire night.

Spectral analysis

Power spectra were computed by a Fast-Fourier transform based on 4 s epochs (Hanning window, linear detrending, 50% overlap), resulting in a frequency resolution of 0.25 Hz. Spectra between 0.5–20 Hz were investigated. The average spectral power was computed across all epochs of a given sleep stage (N2, N3, NREM, REM), separately for the entire night, the first half, and second half of the night.

The power spectra were computed for the first and second halves of the sleep episodes, as well as for the entire night, separately for NREM sleep, stages N2 and N3, and REM sleep.

Current source density analysis

To investigate how GHB affected physiological EEG oscillations in NREM sleep, 10 min segments around the maximum of slow-wave activity (SWA) values in the first NREM sleep episode following GHB/placebo administration were extracted for further analyses. All used segments were located in the peak phase of drug action (t45–t75) [30]. The procedure is depicted in Fig. 1, indicating the time point of GHB (red triangle) and placebo (black triangle) administration, and segment extraction for current source density (CSD) analysis (marked with a green arrow) in a representative individual. Extracted segments were preprocessed using Brain Vision Analyzer 2 software (Brain Products GmbH). First, EEG data were re-referenced to the average of all scalp electrodes. Second, a bandpass filter from 0.5 to 40 Hz was applied to the EEG data, to attenuate channel drifts and satisfy the assumption of stationarity necessary for computing independent component analysis. Third, movement-related artifacts were visually identified and excluded. Fourth, the preprocessed segment was segmented into 12 s segments and exported for further analysis. The minimal segment number used for CSD analyses was 45. Two subjects had to be excluded from CSD analysis because of insufficient data quality. The CSD was analyzed using exact low-resolution brain electromagnetic topography (eLORETA, http://www.uzh.ch/keyinst/loreta.htm), a widely used mathematical approach to estimate the sources of electrical currents measured with scalp EEG. eLORETA computes a three-dimensional EEG CSD map by applying a three-spherical shell model restricting the solution space to the gray matter and hippocampus, resulting in 6239 different voxels of 5 × 5 × 5 mm each. The anatomical brain model used for the computation of intracerebral CSD values was registered on a digitized average magnetic resonance imaging (MRI) brain of the Talairach and Tournoux atlas (Brain Imaging Centre, Montreal Neurological Institute). To compute spectral density (μA/mm2), the signal was split into delta (0.5–4.5 Hz), theta (4.5–8 Hz), alpha (8–12 Hz), sigma (12–16 Hz), and beta (16–20 Hz) frequency bands.

Fig. 1

Hypnogram (top row), time frequency analysis (middle row), and course of slow-wave activity (bottom row; SWA, power density within 0.75–4.5 Hz; C3A2 derivation) of a single subject (N = 1) after intake of a placebo (left panel) and GHB (right panel). The triangles indicate the time point of placebo (black) GHB (red) administration. Arrows indicate time point of EEG segment extraction for eLORETA analysis (CSD and LPS). N1 stage 1 sleep; N2 stage 2 sleep, N3 stage 3 sleep; R, REMS; W, wake

Lagged phase synchronization analysis

The lagged-phase synchronization (LPS) analysis implemented in the eLORETA software was used to characterize differences in functional connectivity among brain areas in GHB-augmented sleep when compared with placebo. Phase synchronization of neuronal oscillations has been repeatedly shown to reflect the coordination of activity among distinct brain regions [31, 32]. Quantification of LPS thus offers a valuable tool to investigate physiological processes underlying functional connectivity between distinct neuronal assemblies [33].

This method has undergone cross-modal validation from both diffusion tensor imaging [34] and functional MRI [35]. LPS allows calculating the functional connectivity among regions of interest (ROI) for all defined frequency bands. Six brain regions, which were significantly affected by GHB (as revealed by the CSD analysis), were chosen for ROI analyses: anterior cingulate cortex (ACC; X = 1, Y = 29, Z = 14), posterior cingulate cortex (PCC; X = 1, Y = −51, Z = 8), left dorsolateral prefrontal cortex (ldlPFC; X = −39, Y = 40, Z = 28), right dorsolateral prefrontal cortex (rdlPFC; X = 39, Y = 40, Z = 28), left parahippocampal gyrus (lPHG; X = −27, Y = −40, Z = −14), and right parahippocampal gyrus (rPHG; X = 27, Y = −40, Z = −14).

Statistical analyses

A linear mixed-effects model, with condition (GHB vs. placebo) as within-subject factor and subject ID as random effect was employed on R (RStudio Version 1.0.136; RStudio, Inc.) for the analyses of the sleep variables (R-package “lme4,” Version 1.1–15). For all applied models, normal Q–Q plots were applied, demonstrating normality of the residuals. Moreover, the assumption of homoscedasticity and linearity was verified using a Tukey-Anscombe (residuals vs. fitted) plot. Furthermore, autocorrelation plots did not reveal any autocorrelation among the residuals. P-values of post-hoc tests (R-package “emmeans,” Version 1.2.1) were corrected for multiple comparison using Benjamini–Hochberg correction of the false discovery rate [36]. Spectral data were statistically analyzed with a linear mixed-effects model (R-package “nlme;” Version 3.1), while controlling for inter-frequency bin correlations, by an autoregressive moving average model and comparing bin-wise with general linear hypothesis tests. Statistical differences in CSD between the conditions were calculated using a nonparametric mapping approach as implemented in the eLORETA software [37]. A voxel-by-voxel-dependent t-test was used, whereby the contrast between the conditions for all frequency bands were calculated separately. Corrections for multiple comparisons were performed across voxels and frequency bands, applying a randomization strategy with 5000 permutations. Moreover, statistical regularization (variance smoothing parameter for t-statistic at 5%) was applied.

Statistical differences in LPS between the conditions were calculated using a nonparametric mapping approach (as implemented in the eLORETA software) [37]. A paired t-test was used, calculating the contrasts between the conditions for all frequency bands. Corrections for multiple comparisons were performed across ROIs and frequency bands, applying a randomization strategy with 5000 permutations.