To explore the effect of gaboxadol on NREM EEG in transient insomnia using power spectral analysis and evaluate the response between men and women.
This was a randomized, double-blind, 3-way, parallel-group transient insomnia study in 22 sleep laboratories. After a baseline night (N1), subjects underwent a 4-h phase-advance of their habitual sleep time the following night (N2). Healthy subjects aged 18-64 y were given single-blind placebo on N1 followed by double-blind treatment on N2 (gaboxadol 10 mg [n = 271], 15 mg [n = 274], or placebo [n = 277])
At baseline, women showed significantly greater values in low frequency activity (< 10 Hz) and in high spindle/low beta frequency activity (14-18 Hz) compared to men. During the phase advance (placebo N2-baseline N1), there was a significant increase in power within the high spindle/low beta frequency range (15-17 Hz) and a significant reduction in beta activity (20-32 Hz), which was greater in women than men. Gaboxadol induced a significant (dose-related) increase in low frequencies (< 8 Hz) and a significant (dose-related) decrease within the alpha/spindle range (11-12 Hz). The effect was dependent upon sex, with a greater magnitude of effect observed in women than men.
Gaboxadol shows a characteristic NREM EEG spectral profile in a model of transient insomnia. Men and women show clear differences in NREM EEG activity at baseline, to gaboxadol treatment and to phase-shifts in habitual sleep/wake times. The exact mechanisms underlying the sex differences remain unclear, but sex is an important variable in studies evaluating sleep and gaboxadol.
Ma J; Dijk DJ; Svetnik V; Tymofyeyev Y; Ray S; Walsh JK; Deacon S. EEG power spectra response to a 4-h phase advance and gaboxadol treatment in 822 men and women. J Clin Sleep Med 2011;7(5):493-501.
The functional significance of individual differences in SWS and the pharmacological enhancement of SWS is under intense investigation.1–4 The clinical interest in a SWS-enhancing agent, one that augments delta activity, is that such agents may be able to do so in a subject group with diminished SWS/SWA (e.g., aging and males) and thereby provide some form of benefit to the individual in terms of sleep, daytime performance, and ultimately quality of life. Sex (gender) is beginning to emerge as one of the key predictors of individual differences of SWS under normal sleep conditions, but it is not known whether sex can contribute to the individual differences observed in sleep disruption or in specific pharmacological manipulations of SWS. Even though gaboxadol is no longer in clinical development for insomnia, it can be used as a pharmacological research tool with a unique mechanism of action, and specifically one that enhances SWS. Ultimately such investigations may lead to a better understanding of both the functional significance of SWS as well as the profound sex differences in the prevalence of insomnia.
Gaboxadol is a novel selective extrasynaptic GABAA agonist with the α4δ containing receptors currently seeming the most relevant target for mediating its hypnotic activity.5 This receptor subtype occurs abundantly in the thalamus, cerebral cortex, and limbic system6—areas of importance in the regulation of sleep. The mechanism of action of gaboxadol is unique in that it directly (and independently from GABA) opens the chloride channel of the GABAA receptor via interaction with the GABA binding site.
Current Knowledge/Study Rationale: Sex is beginning to emerge as one of the key predictors of SWS and the sleep EEG under normal conditions but it is not known whether sex also predicts individual differences in the EEG observed under sleep disruption or pharmacological manipulation of SWS.
Study Impact: This study highlights that quantitative EEG analysis identifies sex as an important predictors of both individual differences in SWS and the sleep EEG under normal baseline conditions, during sleep/wake manipulations as well as a predictor of the effects of a pharmacological SWS-enhancing agent. Sex is therefore an important variable to consider in clinical trials evaluating sleep and SWS and may also lead to a better understanding of both the functional significance of SWS and the profound sex differences in the prevalence of insomnia. Large scale quantitative EEG analysis is now possible.
The extrasynaptic location of the δ containing receptors has been associated with tonic inhibitory activity, versus the phasic inhibitory activity of γ containing synaptic GABAA receptors, which mediate the effects of benzodiazepine receptor agonists. Several groups have shown that δ containing receptors are crucial for the generation of tonic currents in different parts of the brain.7,8 Indeed, Belleli et al.7 have calculated that tonic inhibition mediated by extrasynaptic GABAA receptors accounts for the great majority of the GABAA receptor-mediated inhibitory charge in a location like the ventral basal thalamus, which exhibits both phasic and tonic modes of inhibition.7 Thus, tonic inhibition exerts profound effects on neuronal excitability. Moreover, the α4β3δ subtype of GABAA receptors are insensitive to allosteric modulation by benzodiazepine receptor agonists.9 The functional consequences of these anatomical and pharmacological differences are yet to be fully characterized. However, studies have shown a consistent pattern of effects on sleep that differentiate gaboxadol from benzodiazepine receptor agonists and most other GABAergic drugs.
In humans, gaboxadol has a peak plasma concentration within approximately 30 min of ingestion and an elimination half-life of 1.5-2 h.10,11 In a recent study by Dijk et al., sex differences in the EEG response to gaboxadol were reported in a Phase II transient insomnia study, where the response to gaboxadol in low-frequency activity was greater in women than men.12 There is also increasing evidence of sex differences under normal conditions with no interventions in both traditional measures of sleep and in the spectral composition of sleep EEG, where women have more low-frequency activity than men.13–17 It has also been shown that spindle activity changes in relation to the menstrual cycle.18 It is not known what the significance of these EEG changes mean or whether the response to induced transient insomnia would be different between men and women. Neurosteroids are known to interact with the GABAA receptor complex and most notably with α4 and δ containing subunits. There is also some preclinical evidence to suggest that interactions between gaboxadol and steroids do exist.19,20
Gaboxadol has been shown to facilitate sleep onset and more consistently sleep maintenance using traditional PSG methods of visual scoring. Small PSG studies in healthy young and elderly subjects have shown that gaboxadol increases sleep continuity and increases SWS without suppressing REM sleep.21–24 These findings have been confirmed in Phase II and III studies in healthy subjects in models of transient insomnia25,26 and in Phase II and III studies in patients with primary insomnia.27–29 From a sleep microarchitecture perspective, gaboxadol consistently enhances low-frequency activity (1-8 Hz) and suppresses spindle activity (11-14 Hz).22,25,30
In the current study, the phase-advance model of transient insomnia was chosen because it is associated with an increase in wakefulness prior to sleep, and thus a reduction of homeostatic sleep pressure, while at the same time the circadian drive for wakefulness is very strong, i.e., sleep is scheduled close to the wake maintenance zone.31 It is of interest to know whether an SWS-enhancing agent such as gaboxadol can still enhance SWS under these circumstances.
The prespecified objectives of the original study were to examine the effects of 10 mg and 15 mg gaboxadol versus placebo on traditional visually scored and patient-reported sleep endpoints in a 4-h phase-advance model of transient insomnia and are presented in a separate published report.26 In that study, the primary prespecified efficacy endpoints were PSG wake after sleep onset (WASO) and latency to persistent sleep (LPS).
The specific objectives of the present post hoc analysis reported in this paper were to explore the effect of gaboxadol on NREM sleep EEG by using high-throughput EEG power spectral analysis and to further investigate whether the response is different between men and women. The analysis used data collected from a large multicenter parallel-group study and is not only the largest evaluation of the effects of gaboxadol on EEG power spectra but, as far as the authors are aware, is also the largest evaluation of the effects of a transient insomnia model per se on EEG power spectra. It is also the first comparison of evaluation of the effects of a transient insomnia model on EEG power spectra in men and women. This paper provides additional information not just on the effects of an SWS-enhancing drug on neurophysiology in both men and women, but also in terms of investigating the neurophysiology of transient insomnia in men and women.
The study was conducted at 22 sleep laboratories in the USA under a common protocol, which was approved by an institutional review board for each site. Prior to initiating study procedures, subjects gave written informed consent.
Subjects and Screening Procedures
Healthy men and women between the ages of 18 and 64 years were recruited by general advertising in this randomized, double-blind, placebo-controlled, parallel group study. Seven days after a medical screening visit, subjects underwent a PSG screening night and multiple sleep latency test (MSLT) the following day. Subjects were excluded if they had a sleep or psychiatric disorder based upon DSM-IV criteria (1994) as assessed by a sleep history and the Mini International Neuropsychiatric Interview,32 or a score > 5 on the Pittsburgh Sleep Questionnaire Inventory,33 or a score > 12 on the Epworth Sleepiness Scale (ESS).34,35 Subjects were required to report (1) bedtime between 21:00 and 24:00 ≥ 4 nights per week; (2) usual sleep duration between 6.5 and 9 h; and (3) typical sleep latency < 30 min. Patients with apnea and/or periodic limb movement disorder were excluded based on the screening PSG night evaluation. An MSLT was conducted 4 times during the day after the PSG screening night (2, 4, 6, and 8 h after waking) to exclude subjects considered to be excessively sleepy. Each subtest was terminated after 1-2 min of unequivocal sleep according to standard experimental procedures36 to exclude subjects with a mean MSLT score < 10 min. After completion of the PSG screen, subjects completed an electronic sleep diary for a 7-day period at home to confirm the previous report of habitual bedtime, sleep duration, and sleep latency.
Subjects who met entry criteria were randomly assigned to treatment group and spent 2 consecutive nights at the sleep laboratory for PSG evaluation, generally 7 days after the screening PSG. Study drug was administered 30 min prior to PSG start time on each night. On night 1 (baseline), subjects received single-blind placebo, and PSG started at the median habitual bedtime from the 7 nights between PSG screening and the baseline night, as determined from the electronic diary. On night 2 (phase advance), subjects received double-blind placebo, gaboxadol 10 mg, or gaboxadol 15 mg. PSG start and stop times on night 2 were 4 h earlier than on night 1. Total PSG recording time on each night was 8 h.
Other and more detailed methods can be found in Walsh et al.26
EEG Power Spectra Measures
This trial involved 22 clinical sites across North America, often with different PSG recording equipment, software, and system settings. However, all sites were required to follow a specific PSG recording procedure detailed in the study PSG operations manual, which specified that the sampling rate of EEG channels should be ≥ 200 Hz, the impedance of each EEG electrode should be < 5 kΩ before the start of recording, and low-pass filter settings should be appropriate to avoid aliasing.
PSG recordings were exported as EDF files. Each PSG recording includes multiple channels of several types of physiological signals, including those generated by EEG from the brain cortex. All EEG signal segments containing severe artifacts were removed using a validated high-throughput procedure. Sleep EEG artifacts are composed of different types, such as muscle artifacts, ocular artifacts, EKG-induced artifacts, and artifacts due to loose electrodes. Because of the different origins of these artifact types, they demonstrate distinguishable signal features that enable the high-throughput software system to detect and remove each type separately using a module specially designed for that artifact type. After the artifact identification and rejection stage, spectral analysis was performed on the C3-A2 derivation using a nonparametric spectra estimation (Welch) algorithm, with a Hamming window of 4 sec, and a 1-sec overlap between 2 adjacent windows. An expert review of the power spectra curves of the EEG recordings from each subject was then performed to determine if any resultant power spectra curves should be rejected due to significant artifacts not identified by the automated process. About 8% of the records were excluded from subsequent statistical analysis after EEG experts, blinded to treatment information, confirmed that the signal quality of the recording was so low that EEG signals were dominated by various artifacts (e.g. muscle or ocular artifacts). After unblinding, these abnormal records were further checked against treatment. They were not related to treatment and hence do not alter the interpretation of results.
Sleep stage power spectra represents the power spectra calculated from EEG recorded at non-wake sleep stages, i.e., sleep stages 1, 2, 3, 4, and REM. Similarly, NREM sleep power spectra are calculated from NREM sleep stages (1, 2, 3, and 4), and REM sleep power spectra are from REM sleep stage. Previous analyses have shown that the effects of sleep manipulations, including sleep deprivation (i.e. for e.g.37) and pharmacological manipulations,38 on EEG power spectra are similar in stages 1, 2, 3, and 4. Thus, not much is learned from presenting spectral data separately for stages 1, 2, 3, and 4, and the effects of these manipulations can be reliably summarized by presenting the effects on NREM sleep. In addition, if we consider all NREM sleep together, a more reliable estimation of the power spectra can be obtained because of the larger amount of data available. Thus, we report power spectra during all NREM sleep epochs (i.e., stages 1, 2, 3, 4 combined).
A power spectra profile was calculated from each recording. The power spectra profile is composed of a sequence of power spectra values with a 1-Hz resolution in the frequency domain ranging from 0.5 to 32.5 Hz, which covers the conventional frequency bands of SWA ([delta] 0.5-4.25 Hz), theta (4.25-8 Hz), alpha (8-12 Hz), sigma (12-15 Hz), and beta (15-32.5 Hz). However, some drug or biological effects can cross the borders of traditional frequency band widths, and there is no standard definition for frequency bands; therefore analysis over the whole power spectra profile (in 1-Hz frequency bins) can save us from analyzing the power spectra of each arbitrary frequency band separately. The inferences on different frequency bands can easily be derived from the inference on the 1-Hz power spectra profile. For graphical purposes, power values are plotted at the midpoint of each frequency range (e.g., at 1 Hz for 0.5-1.5 Hz bin), and frequency band widths are added illustratively.
Absolute power spectra data were log-transformed prior to statistical modeling because the models generally assume that the data follow a Gaussian distribution. Relative power spectra are not used because this type of data obscures differences in overall EEG amplitude between subjects, which may be relevant when we describe drug or sex effects.
To assess sex differences at baseline, data from all treatment groups on Night 1 were used. As previously mentioned, each recording generated a sequence of power spectra values in the frequency domains, thus a linear mixed-effects (LME) model was employed to better model the hierarchical structure of power spectra both within and between subjects. A LME model has both fixed-effect factors, which are equivalent to the independent variables in a linear model, and the random effect part, which explains the known variability in the data.39 In our model, the fixed-effect factors involved frequency, sex, and their interactions. The random effect modeled the between-subject variability at different frequencies. Because the relationship between power spectra and frequency is nonlinear, the frequency cannot be directly included as a numerical variable in the model. Including it as a categorical variable would also be counterintuitive and inappropriate. Instead, and following previous similar statistical approaches,13 power spectra were represented in the model as a linear combination of a set of B-spline functions14 to model the relationship between the power spectra and the frequency. The estimated sex differences at different frequencies, along with their confidence intervals, were derived from the fitted LME model.
For graphical presentation of the power spectra differences between women and men at baseline (night 1), the exponential of the estimated mean difference between women and men (which is equivalent to the ratio of the mean spectra of women to that of men) was assessed (with no-difference = 100%).
The effect upon NREM sleep EEG spectra of the phase-advance procedure was assessed using the phase-advance night (night 2) in the placebo group compared to “baseline” (night 1) overall and then separately in men and women. The study was designed as a randomized controlled trial with 3 treatments, and thus its design could not guarantee that subjects were only subjected to the 4-h phase advance between nights 1 and 2. However, it is reasonable to assume that the difference between night 1 and night 2 due to factors other than phase advance was insignificant. Firstly, the baseline night was conducted under single-blind placebo treatment, and thus the effect of the placebo treatment on power spectra between nights 1 and 2 was assumed to be balanced or at least minimized in the study design. In addition, the subjects all underwent at least a screening PSG night in the laboratory prior to the first baseline night, and subjects had placebo treatment on both the baseline night and the phase-advance night in different periods; thus the habituation effect between nights 1 and 2 was assumed to be minimal.
An LME model similar to the one previously described was employed to evaluate the effect of gaboxadol on EEG power spectra, except that the fixed-effect factors included the frequency-derived B-spline coefficient set, treatment, sex, and all of their interactions in this case. The baseline EEG spectra were incorporated in the model using the change-from-the-baseline approach. The effect of gaboxadol treatment on EEG power spectra was quantified as the ratio of the mean power spectra of gaboxadol treatment group to that of placebo treatment group and was directly obtained from the fitted model using a contrast matrix of gaboxadol treatment versus placebo. The effect of gaboxadol versus placebo was then evaluated separately in men and women following the same approach.
To evaluate how the effect of gaboxadol changed with time, the whole-night PSG recording was divided into 3 equal parts, and the effect of gaboxadol during each third of the night was obtained, using a similar model previously presented. Although arbitrary, dividing the night into thirds is an accepted method to visualize time course effects. This method is preferred over an analysis on the basis of NREM periods, because the latter analysis can be confounded by treatment effects on NREM period, and it often leads to either some data not being included or a different number of subjects contributing to averages of some NREM periods because of differences between individuals.
Whenever necessary, a multiplicity adjustment method,41 was employed to address the multiple comparisons across different frequencies.
In order to identify whether sex effects were caused by body weight, the following analyses were employed. To simplify the modeling, only the mean power spectra in SWA and theta bands, instead of the full power spectra profile used in all previous analyses, were used as the dependent variables in the statistical models. The rationale for this was that data in the SWA and theta bands showed the clearest sex difference response to gaboxadol treatment, and therefore both bands were likely to be more sensitive for assessing the impact of body weight on the effect of sex. Two versions of linear models were considered to explore 2 different possible interactions between treatment and 3 other factors (sex, weight, and age). In the first model, the interactions between treatment and each of the 3 factors were taken into consideration, while in the second model, only the interaction between treatment and weight was taken into consideration. A logarithm transform was applied to the age factor to make it more Gaussian.
Subject Selection and Accounting
Spectral analysis was performed on 822 healthy subjects (N = 277 for placebo, N = 271 for gaboxadol 10 mg, N = 274 for gaboxadol 15 mg).
As shown in Table 1, there were no differences in MSLT, sleep efficiency, body mass index (BMI), or age between the 3 treatment groups, or between men and women within these treatment groups. There were, however, slightly more women (approximately 60%) than men in each treatment group. The median habitual bedtime was 23:00 in each of the 3 treatment groups. The median bedtime for the phase-advance night was 19:00 in each of the 3 treatment groups.
Baseline demographic, sleep efficiency, and MSLT scores (Night 1)
|Characteristic||Placebo (N = 277)||Gaboxadol 10 mg (N = 271)||Gaboxadol 15 mg (N = 274)|
|Mean (SD) age, years||30.9 (10.0)||30.8 (9.8)||30.3 (9.9)|
|Men/women mean (SD) age, years||29.0 (8.3)/32.2 (11.0)||28.7 (7.7)/32.0 (10.7)||28.6 (8.6)/31.1 (10.4)|
|Mean (SD) body mass index ||24.7 (3.7)||25.0 (3.7)||24.4 (3.5)|
|Men/women mean (SD) body mass index ||25.7 (3.3)/24.0 (3.8)||25.6 (3.2)/24.7 (3.9)||25.2 (3.0)/24.0 (3.7)|
|PSG sleep efficiency (%)||86.1 (8.6)||85.7 (9.8)a||86.1 (9.3)|
|Men/women PSG sleep efficiency (%)||85.4 (10.2)/86.6 (9.1)||85.3 (11.3)/86.0 (8.9)b||85.2 (8.8)/86.5 (8.1)|
|MSLT (min)||16.7 (4.4)||16.5 (4.0)||16.2 (3.3)|
|Men/women MSLT (min)||16.8 (4.6)/16.7 (4.3)||16.5 (3.5)/16.5 (4.3)||16.1 (3.1)/16.3 (3.4)|
Baseline demographic, sleep efficiency, and MSLT scores (Night 1)
Comparison of Women versus Men NREM EEG Power Spectra at Baseline
Figure 1 shows the mean power spectral densities of the women relative to the men at baseline. Compared to men, women showed significant (p < 0.05) increases in low-frequency activity (1-10 Hz) and in the high spindle/low beta frequency range (14-18 Hz). Beta activity up to 32 Hz remained higher for women, but this was not significantly different for any 1-Hz bin compared to men.
Mean power spectra densities of women relative to men (men = 100%) for each frequency bin
Mean power spectra densities of women relative to men (men = 100%) for each frequency bin
Effect of the Phase-Advance Procedure on NREM EEG Power Spectra
Examination of night 1 (baseline) and night 2 (phase advance) data from the placebo group demonstrates the effects of the phase advance procedure on EEG power spectra (Figure 2). There was a significant (p < 0.05) increase in power in the high spindle/low beta frequency ranges (15-17 Hz) and a reduction in the high beta range (20-32 Hz). There was a minimal effect on delta and theta frequency activity, except perhaps in the very lowest frequency bin (Figure 2A). There was no significant effect of the model on low-frequency activity in men or women, although women did have a significant (p < 0.05) increase in the lowest frequency bin. Women had significant suppression at the 11-13 Hz and 20-32 Hz and increases at 15-16 Hz (p < 0.05; Figure 2B). Men also had reduced high frequency activity (> 18 Hz) but to a lesser degree, and this was only significant at 31 Hz (p < 0.05). The greatest effect of the model in men tended to be an increase of power density within the high sigma/low beta frequency range (14-17 Hz; p < 0.05).
Effect of a 4h phase advance on EEG power spectra (1Hz bins) (A) overall and (B) women (□) and men (▴) during 8h of NREM sleep
Horizontal bars represent significant differences (p < 0.05) between men and women. Values are plotted at the midpoint of each frequency range (e.g. at 1Hz for 0.5-1.5Hz). Frequency band widths are presented schematically.
All data are presented relative to baseline. Horizontal bars represent significant differences (p < 0.05) between baseline and phase advance night. N = 109 for males, N = 142 for females and N = 251 combined. Values are plotted at the midpoint of each frequency range (e.g. at 1Hz for 0.5-1.5Hz). Frequency band widths are presented schematically.
Effect of a 4h phase advance on EEG power spectra (1Hz bins) (A) overall and (B) women (□) and men (▴) during 8h of NREM sleep
Treatment Effects on Whole-Night NREM EEG Power Spectra
Examination of NREM EEG (Figure 3A) revealed a dose-related increase in low-frequency activity (p < 0.05, up to 8 Hz for 10 mg and 15 mg) and a dose-related decrease in alpha/spindle activity (p < 0.05, 11 Hz for 10 mg, 11-12 Hz for 15 mg). There were no significant effects at any other frequency. This effect appeared to be dependent upon sex (Figure 3B). Gaboxadol enhanced low frequency activity and suppressed spindle activity to a greater extent in women than in men. In women, both doses of gaboxadol significantly enhanced low-frequency activity between 1-8 Hz (p < 0.05), and in a dose-related manner. In men, only 15 mg gaboxadol significantly enhanced power between 1-8 Hz (p < 0.05). The magnitude of effect at the low-frequency range was visibly lower for men than women. Spindle activity (11-14 Hz) was significantly suppressed for 15 mg in women only (p < 0.05). There also appeared to be a different magnitude of power response at this frequency range between men and women. Although not significant, the 15-mg dose in men appeared to enhance higher frequency activity.
Effects of gaboxadol on the NREM sleep EEG power spectral densities (1Hz bins) during 8h of sleep after a 4h phase advance of subjects' habitual bedtimes for (A) combined sex (• 10mg, n = 247; ○ 15mg, n = 253) and (B) men and women separately (▴- - -▴ men 10mg, n = 94; ▴––▴men 15mg, n = 88; □- - -□ female 10mg, n = 153; □––□ female 15mg, n = 165)
All values are expressed relative to the power spectral densities obtained during the placebo condition. Horizontal bars represent significant differences (p < 0.05, adjusted for multiplicity) between gaboxadol vs. placebo. Values are plotted at the midpoint of each frequency range (e.g. at 1Hz for 0.5-1.5Hz). Frequency band widths are presented schematically.
Effects of gaboxadol on the NREM sleep EEG power spectral densities (1Hz bins) during 8h of sleep after a 4h phase advance of subjects' habitual bedtimes for (A) combined sex (• 10mg, n = 247; ○ 15mg, n = 253) and (B) men and women separately (▴- - -▴ men...
There was no significant interaction (p > 0.1) between the treatment effect observed and body weight for the SWA and theta bands with either dose of gaboxadol.
Time Course of Effect of Gaboxadol on NREM EEG Power Spectra
Figure 4 shows the time course of effect of gaboxadol on NREM EEG power spectral density in men and women versus placebo. In women, gaboxadol 10 mg showed significant (p < 0.05) enhancement of low-frequency activity during the first two-thirds of the night (at 1-7 Hz in the first third and at 1-8 Hz for the second third). There was a significant reduction at 10 Hz. In men, only 10 mg gaboxadol achieved a more reliable statistically significant (p < 0.05) increase in low-frequency activity (1 Hz, 6-8 Hz) during the second third of the night. With 15 mg gaboxadol, the magnitude of effect was greater and lasted significantly longer (into the last third of the night) compared with 10 mg for both men and women. In women, 15 mg gaboxadol showed a significant increase in low-frequency activity (1-8 Hz) and a significant reduction in the spindle frequency range (first third: 10-14 Hz, second third: 10-15 Hz). This difference remained in the last third of the night, with a significant increase at 6-7 Hz and a significant reduction at 12-13 Hz. In men, 15 mg gaboxadol showed significant enhancement at 1-8 Hz in the first two-thirds and remained significant at 6-7 Hz in the last third. The greatest difference between men and women was more apparent during the first third of the night and was greater with 15 mg gaboxadol.
Time course effects of gaboxadol across the night (per third) on NREM EEG power spectral densities (1Hz bins) during 8h of sleep after a 4h phase advance of subjects' habitual bedtimes for men and women (▴- - -▴ men 10mg, n = 94; ▴––▴men 15mg, n = 88; □- - -□ women 10mg, n = 153; □––□ women 15mg, n = 165)
All values are expressed relative to the power spectral densities obtained during the placebo condition. Horizontal bars represent significant differences (p < 0.05, adjusted for multiplicity) between gaboxadol vs. placebo for men and for women. Values are plotted at the midpoint of each frequency range (e.g. at 1Hz for 0.5-1.5Hz). Frequency band widths are presented schematically.
Time course effects of gaboxadol across the night (per third) on NREM EEG power spectral densities (1Hz bins) during 8h of sleep after a 4h phase advance of subjects' habitual bedtimes for men and women (▴- - -▴ men 10mg, n = 94; ▴––▴men 15mg, n = 88; □- -...
The prespecified hypotheses of the original published study on which power calculations were based stated that gaboxadol would reduce wake after sleep onset (WASO) and latency to persistent sleep (LPS) versus placebo. WASO and LPS were significantly reduced, but the effects on sleep onset were much less marked. To add some context to the EEG power spectra results, the prespecified sleep continuity and sleep stage measures associated with gaboxadol treatment are presented in Supplemental Table S1 at www.aasmnet.org/jcsm (as described in Walsh et al.)26 In summary, gaboxadol 10 and 15 mg increased total sleep time by 23.9 and 34 min, reduced wake after sleep onset by 8.1 and 15.2 min, and reduced latency to persistent sleep by 2.5 and 3.7 min, respectively, as a difference to placebo. The number of awakenings was not affected by treatment. The major changes on sleep architecture were observed on SWS (50% to 60%) and stage 2. Gaboxadol 10 and 15 mg increased SWS by 11.6 and 21.4 min, reduced stage 1 by 2 and 3.4 min, increased stage 2 by 10.4 and 14 min, and increased REM by 3.6 and 2.6 min, compared with placebo.
In order to put these visually scored sleep and EEG power spectra results into context, the sleep continuity and sleep stage measures associated with the phase-advance procedure are presented in Supplemental Table S2 at www.aasmnet.org/jcsm (as described in Walsh et al.)26 In summary, total sleep time was reduced by a mean of 39 minutes to 374 min, wake after sleep onset increased approximately 15 min to 53 min, and latency to persistent sleep increased by only 3.6 min to 19.1 min. The number of awakenings was not increased as expected, but reduced significantly by 1.3. The major changes on sleep architecture were observed on stage 2 (17.6-min reduction) and REM (15.5-min reduction). There were no significant effects of the model on SWS, with only a small reduction in stage 1 (by 6 min). Thus the effect of the model on sleep disruption was modest; and with no significant effects on SWS, this model may not demonstrate the full potential of an SWS-enhancing agent to be evaluated.
The examination of the treatment effect of gaboxadol and the phase-advance model itself on EEG power spectra was not prespecified in the original study, nor was the evaluation of sex differences. The EEG power spectra results reported in this paper were part of a post hoc analysis to help evaluate the neurophysiology of an SWS-enhancing agent with a large dataset.
In summary, the EEG power spectra results demonstrated that (1) at baseline, women showed significant increases in low-frequency activity (< 10 Hz) and in high spindle/low beta frequency activity compared to men; (2) during the phase-advance (placebo N2 versus baseline N1), there was a significant increase in power in high spindle/low beta activity and a significant reduction in high beta activity with minimal effect in low frequencies, and the magnitude of effect (of frequencies > 11 Hz) was different between sexes during the phase advance; and (3) gaboxadol induced a significant (dose-related) increase in low-frequency activity (< 8 Hz) and a significant (dose-related) decrease in alpha/spindle activity; the effect was dependent upon sex, with a greater magnitude of effect observed in women than men.
Before discussing in greater depth the effect of drug treatment on EEG power spectra in a model of transient insomnia and trying to identify differences in treatment response between sexes, we first discuss the differences between sexes at baseline and the effect of the model per se on EEG power spectra in men and women. Compared to men, women showed significant increases in low-frequency activity < 10 Hz and in the high spindle/low beta frequency range (14-18 Hz) at baseline. Beta activity up to 32 Hz remained higher for women, but this was not significantly different for any 1-Hz bin compared to men. This shows quite remarkable replication of the findings of Dijk et al.15 and Carrier et al.,13 who used much smaller sample sizes. The results of those studies effectively showed a higher total power in all frequency ranges for women than men, except for the high alpha/low spindle range. These results could be related to EEG generating mechanisms,42 rather than sleep regulatory mechanisms, but appear to be frequency specific and therefore related to specific functional effects.
As far as the authors are aware, this is a first large-scale study to evaluate the effect of a transient insomnia model on EEG power spectra, and specifically the responses in men and women separately. This 4-h phase advance model of transient insomnia disrupted objective and subjective sleep maintenance and onset measures in healthy, young adult normal sleepers as described by Walsh et al.26 No significant effects of the model on SWS were observed, with the major changes in sleep stage architecture being reductions in stage 2 and REM sleep. Spectral analysis of the NREM EEG confirms a lack of effect on low-frequency activity that characterizes visually scored stage 3 and 4 sleep. The predominant effect seems to be an increase in the 15-17 Hz range and a decrease in the 20-32 Hz range. The increase in the 15-17 Hz range and the reduction in the 11-13 Hz (women only) represents the well-documented circadian modulation of low- and high-frequency spindle activity.12,43
To evaluate the independent effects of the model, an adequate control group would have consisted of subjects sleeping at their habitual bedtime for another night, because this would control for potential order effects. However, the differences between EEG power spectra in the placebo phase-advance condition versus the baseline night are frequency specific, and, particularly for the effects in the sleep spindle range, are much larger than what we have ever observed in a comparison of two consecutive baseline nights. It should also be noted that subjects received single-blind placebo treatment on the “baseline” night immediately before the phase-advance condition under placebo. We are therefore confident that these effects represent effects of the model rather than effects of placebo treatment per se or period effects.
The reduction in the 20-32 Hz during the phase advance cannot be explained by the circadian modulation of EEG activity in this frequency range, but whether this represents the effects of continued adaptation to the laboratory environment is not known.44
With gaboxadol treatment, low-frequency EEG activity was increased and spindle activity was attenuated in a dose-related manner. This is consistent with the increase in visually scored SWS in the same study.26 The reduction in 10-12 Hz activity and increase in theta activity are further indicators of a shift to deeper sleep. This spectral profile of gaboxadol is in good agreement with previous studies on healthy subjects in a model of transient insomnia, healthy adult and elderly subjects, and patients with primary insomnia.21,22,25,30 Thus despite different study designs and different study populations, gaboxadol clearly shows a characteristic effect on NREM EEG power density, with only the magnitude of effect differing depending on dose and possibly study population.
This effect also seems to be consistent between men and women. However, it is clear that the magnitude of effect is dependent upon sex. In women, both gaboxadol doses produced greater increases in low-frequency activity and greater reductions in spindle activity range (11-15 Hz). This analysis in a large parallel-group study confirms the original finding of Dijk et al.12 describing sex-related differences to gaboxadol in a similar transient insomnia model using a crossover design. In the previous study, gaboxadol enhanced delta and theta activity significantly more in women than men in both NREM and REM sleep.
This difference between men and women in response to gaboxadol may be related to numerous factors, such as the number and distribution of GABAA receptor subtypes, pharmacokinetics, or through interaction between neurosteroids and GABAA receptors. There was no significant interaction between body weight and treatment effect (on SWA and theta), suggesting that body weight was not a confounding factor in the sex differences observed in this study. The α4δ containing GABAA receptors currently seem the most relevant target mediating hypnotic activity of gaboxadol5 and are predominantly extra-synaptically located.45,46 This receptor subtype occurs abundantly in the thalamus, cerebral cortex, and limbic system6—areas of importance in the regulation of sleep. This receptor subtype may also be a focus for neurosteroid action.
Neurosteroids function as effective positive modulators of the GABAA receptor. However, during normal cycles of endogenously circulating steroids across the menstrual cycle, during pregnancy, or following chronic stress, prolonged exposure and withdrawal can occur, which can change GABAA receptor expression (most notably in those receptors with α4 and δ subunits).47 A sex difference in the effect of gaboxadol has in fact been observed in insomniac patients, with a tendency for women to show a larger beneficial effect in self-reported sleep maintenance and onset measures48 and in an objective measure of sleep maintenance (PSG wake after sleep onset).29 Unfortunately, these sex differences in this current paper were observed post hoc and were not part of any a priori analyses, so no details on menstrual cycle or menopausal status were collected. Despite this, robust sex differences were observed even when these factors were not controlled for. Furthermore, effects of the menstrual cycle on SWA are considered to be minor.18
As well as treatment-related differences with sex, there were also some differences compared to baseline between men and women in their response to the 4-h phase advance shift. Only women showed a significant reduction in alpha activity, as well as a reduction in the beta range in response to the model.
In conclusion, this study shows that the NREM EEG spectral profile of gaboxadol in a phase-advance model of transient insomnia is similar to other published studies in healthy subjects and patients; and that there are clear differences between men and women in NREM EEG activity at baseline, to gaboxadol treatment and to phase shifts in habitual sleep/wake times. Our results show that in women, i.e., those who have more SWA than men, we observe the largest effects due to acute changes in a sleep/wake cycle and to gaboxadol treatment. The extent to which the changes in the EEG as well as the sex differences, as induced by gaboxadol treatment, reflect an intensification of the normal sleep process cannot, of course, be derived from these EEG data, but will require assessments of (for example) daytime function.4,49,50 The exact mechanisms underlying the sex differences remain to be elucidated, but this study underlines that sex is an important variable in any clinical study evaluating sleep and SWS-enhancing agents. This study highlights that sex is one of the predictors of both individual differences of SWS at baseline and in phase-advanced sleep, as well as a predictor of the effects of a specific pharmacological manipulation with an SWS-enhancing agent. Ultimately this may lead to better understanding of both the functional significance of SWS and the profound sex differences in the prevalence of insomnia.
The study was involves the investigational use of gaboxadol and was sponsored by Merck Inc. Drs. Walsh and Dijk have consulted for Merck and Lundbeck. Dr. Dijk has received research support from Philips Lighting, H Lundbeck A/S, GlaxoSmithKline, Merck, Takeda, Organon, Biotechnology and Biological Science Research Council (BBSRC), Wellcome Trust, Covance, the US Air Force Office of Scientific Research (AFOSR). He has also served as consultant to/speaker for Philips Lighting, H. Lundbeck A/S, Cephalon, Merck, GlaxoSmithKline, Sanofi Aventis, Pfizer, UCB, Takeda, Wyeth, Lilly and Actelion. Drs. Ma, Svetnik, Tymofyeyev and Ray are employees of Merck. Dr. Deacon was an employee of Lundbeck when the study was carried out.