Zolpidem and sleep deprivation: Different effect on EEG power spectra


Borbély Dr Institute of Pharmacology and Toxicology, University of Zürich, Winterthurerstrasse 190, 8057 Zürich, Switzerland. Tel.: +41 1 635 5959; fax: +41 1 635 5707; e-mail: borbely@pharma.unizh.ch.


To study the role of GABA-ergic mechanisms in sleep regulation, the combined action of 40 h sleep deprivation and either 20 mg zolpidem or placebo on the sleep electroencephalogram (EEG) were investigated by quantitative EEG analysis in eight young men who participated in a positron emission tomography study. Compared with baseline, sleep deprivation increased low-frequency (1.25–7.0 Hz) EEG power in non-rapid eye movement (NREM) sleep in the placebo night. After administration of zolpidem, power in the 3.75–10.0 Hz range and 14.25–16.0 Hz band was reduced. The largest decrease was observed in the theta band. Comparison with placebo revealed that zolpidem attenuated power in the entire 1.75–11.0 Hz range. The plasma concentration of zolpidem at 4.5 h after intake showed a positive correlation with the drug-induced difference in power from placebo in the 14.25–16.0 Hz band. Regional EEG analysis based on bipolar derivations along the antero-posterior axis disclosed, for NREM sleep, a drug-induced posterior shift of power in the frequency range of 7.75–9.75 Hz. Zolpidem did not affect rapid eye movemnt sleep spectra. We conclude that sleep deprivation and agonistic modulation of GABAA receptors have separate and additive effects on power spectra and that their effects are mediated by different neurophysiological mechanisms.


The homeostatic facet of sleep regulation has been investigated by spectral analysis of the sleep electroencephalogram (EEG). Enhancing ‘sleep pressure’ by sleep deprivation has been shown to augment power in the delta, theta and alpha bands (between 0.5 and 10 Hz), and to reduce power in the frequency range of sleep spindles (12.25–15.0 Hz) in non-rapid eye movement (NREM) sleep ( Borbély et al. 1981 ; Dijk et al. 1990 , 1993; Aeschbach et al. 1996 ). Although it has been demonstrated that the extent of these changes is determined by the duration of prior wakefulness, the physiological significance of this relationship and the underlying mechanisms remain obscure.

It has been hypothesized that increased potassium conductance (gK+), mediated by the neuromodulator adenosine, underlies the sleep–wake-dependent changes in the sleep EEG ( Benington and Heller 1995). In accordance with this view, adenosine agonists were shown to mimic the effects of sleep deprivation in the rat ( Benington et al. 1995 ; Schwierin et al. 1996 ). In humans, the adenosine antagonist caffeine suppressed low-frequency activity and enhanced activity in the frequency range of sleep spindles ( Landolt et al. 1995a , 1995b). Apart from gK+, the only other inhibitory ionic conductance in neurons of the cerebral cortex is gCl (reviewed by Benington and Heller 1995). However, because the reversal potential for Cl is at ≈ –70 mV, increase in gCl was suggested to be unable to hyperpolarize neuronal membranes to the range needed to potentiate EEG slow wave activity, i.e. below –75 mV ( Benington and Heller 1995). Accordingly, increased gCl should rather antagonize the hyperpolarizing effects of gK+ during NREM sleep by dampening hyperpolarization beyond –70 mV. Indeed, pharmacological agents that increase gCl via GABAA receptors, such as benzodiazepines (BDZs), zolpidem and zopiclone (i.e. agonistic GABAA modulators), have long been known to suppress slow waves and increase sleep spindles ( Borbély et al. 1985 ; Dijk et al. 1989 ; Trachsel et al. 1990 ; Brunner et al. 1991 ; Aeschbach et al. 1994 ). However, in contrast to these findings, zolpidem has also been reported to enhance delta power in anesthetized rats ( Depoortere et al. 1986 ) and to promote Stage 4 sleep in humans complaining of insomnia ( Benoit et al. 1994 ). Moreover, recent studies indicated that selective GABAA agonists, such as muscimol and THIP, and the GABA uptake inhibitor tiagabine, in part mimic the effect of sleep deprivation on sleep and the sleep EEG ( Lancel et al. 1997 , 1998; Faulhaber et al. 1997 ). Thus, the role of GABAA receptors in the physiological regulation of sleep is unclear.

Zolpidem is an imidazopyridine hypnotic, which differs in chemical structure from BDZs and zopiclone ( Salvà and Costa 1995). Its sedative properties in doses of 5–20 mg ( Merlotti et al. 1989 ) are thought to result from selective binding to central BDZ receptors of the α1 subtype, located within the GABAA–BDZ receptor complex, which forms a pentameric glycoprotein structure linked to an integral chloride ion channel. Activation of α1-recognition sites potentiates central nervous GABA-ergic inhibition by increasing the frequency of chloride channel openings. Autoradiographic studies in rats revealed that α1 receptors are localized primarily in layer IV of the cerebral cortex, substantia nigra, olfactory bulb and inferior colliculus ( Niddam et al. 1987 ; Benavides et al. 1993 ). However, the presence of GABA was shown to enhance the binding of [3H]-zolpidem in vitro independent of the relative receptor density in various brain structures, including deep layers of the fronto-parietal and cingulate cortex ( Ruano et al. 1993 ).

The main objective of this study was to explore the role of GABA-ergic mechanisms in sleep regulation in humans in a combined EEG and positron emission tomography (PET) study ( Finelli et al., 2000 ). The main interest from the EEG point of view was the mode of interaction of sleep deprivation and the agonistic GABAA receptor modulator zolpidem, whose individual effects had been reported previously ( Borbély et al. 1981 ; Dijk et al. 1990 , 1993; Brunner et al. 1991 ; Aeschbach et al. 1996 ). The study extended to regional aspects of the EEG, which were recently shown to exhibit a characteristic evolution during the course of normal sleep ( Werth et al. 1996 , 1997).


Subjects and study protocol

Male volunteers from the student populations of the University of Zürich and the Swiss Federal Institute of Technology were recruited for participation in a brain imaging study ( Finelli et al., 2000 ). They all reported being good sleepers, being in good health and not taking any medication or consuming any illicit drugs. All subjects were nonsmokers and reported a consumption of fewer than five alcoholic drinks per week, and fewer than three caffeine-containing beverages (coffee, tea, cola drinks) per day. Their sleep diaries and questionnaires revealed regular bedtimes between ≈ 23.00 and 07.00 h and no subjective sleep disturbances. They were screened by polysomnography in the sleep laboratory to exclude sleep disturbances such as sleep apnea and nocturnal myoclonus. The screening night was spent under conditions similar to those in the PET scanner. A molding plastic mattress forced the subjects to sleep on their backs during the initial 4 h of the sleep episode. Subjects with sleep apnea, a nocturnal myoclonus-index of five or more per hour of sleep, and those who were unable to sleep under conditions of restricted mobility, were excluded. Eight healthy, right-handed men with a mean age of 23.9 ± 0.6 y (SEM) (range: 21–27 y) were selected and paid for their participation. Written informed consent was obtained prior to the adaptation night. Subjects were asked to abstain from ethanol and caffeine at least 3 days prior to each sleep recording. During the prestudy week, they were also instructed to keep a regular sleep–wake cycle with sleep scheduled from 23.00 to 07.00 h. Compliance with these instructions was verified on the basis of records from activity monitors worn on the wrist of the nondominant arm, as well as by determination of the breath-ethanol concentration and the level of caffeine in the saliva upon arrival in the sleep laboratory.

The study protocol was approved by the local ethics committee for research on human subjects. It consisted of a baseline night in the sleep laboratory (preceded by an adaptation night) and two experimental nights at a 1-week interval in a PET scanning facility located in the University Hospital. During baseline, subjects slept in a regular bed from 23.00 to 07.00 h. Experimental nights were preceded by an 8-h sleep episode in the sleep laboratory and a subsequent 40-h sleep deprivation period to promote sleep in the PET scanner. During sleep deprivation, the subjects were under continuous supervision by an experimenter. Because of clinical need for the scanner, sleep episodes in the PET facility had to be confined to 23.00 to 06.00 h. Thus, analyses were restricted to the first 401 min after lights-out, i.e. the longest common duration of all sleep periods. After the first 4 h of the experimental nights (≈ 23.00 to 03.00 h) spent on the scanner, subjects were transferred to a bed next to the scanner. Sleep was recorded continuously by polysomnography. Thirty minutes prior to lights-out, subjects received a capsule that contained either zolpidem tartrate (Stilnox®) or placebo. Thus, a single oral dose of 20 mg zolpidem (i.e. the highest recommended hypnotic dose) was administered to each subject according to a randomized, placebo-controlled, double-blind, cross-over design.

Prior to experimental nights, an intravenous line was placed in the subjects’ left or right antecubital vein to inject the H215O bolus ( Finelli et al., 2000 ), and to take blood to determine zolpidem plasma concentrations. Blood samples (≈ 10 mL) were collected 30 min before drug intake (≈ 22.00 h), prior to lights out (≈ 23.00 h) and upon transfer of the subjects from the PET scanner to the bed (≈ 03.00 h). Samples were stored at ≈ 6°C during the night. They were assayed by high-performance liquid chromatography (HPLC) with UV-detection (254 nm) on the following day. Because of technical problems, the plasma samples of one subject could not be analysed.

Polygraphic recordings

The EEG (derivations C3A2, and F3, F4, C3, C4, P3, P4, O1 and O2 referenced against Cz), electro-oculogram (EOG) and submental electromyogram (EMG) were recorded using a polygraphic amplifier (PSA24, Braintronics Inc., Almere, The Netherlands), digitized and transmitted via fibre-optic cables to a personal computer. EEG, EOG and EMG signals were conditioned by the following analogue filters: a high-pass filter (–3 dB at 0.16 Hz), a low-pass filter (–3 dB at 102 Hz, < –40 dB at 256 Hz) and a notch filter (50 Hz). Data were sampled with a frequency of 512 Hz, digitally filtered (EEG: low-pass FIR filter, –3 dB at 49 Hz; EMG: band-pass FIR filter, –3 dB at 15.6 and 54 Hz) and stored on hard-disk with a resolution of 128 Hz. Power spectra for consecutive 4-s epochs, weighted by application of a 10% cosine window, were computed using a Fast-Fourier transform routine, resulting in a frequency resolution of 0.25 Hz. Values above 25 Hz were omitted. In C3A2, adjacent 0.25-Hz bins were averaged into 0.5-Hz (0.25–5.0 Hz) and 1.0-Hz (5.25–25.0 Hz) bins. Sleep stages were scored visually for consecutive 20-s epochs according to the criteria of Rechtschaffen and Kales (1968). Power spectra of five consecutive 4-s epochs were averaged and matched with the sleep scores. Four-second epochs with artefacts were identified visually and eliminated. Regional differences were investigated by comparing the spectra of the bipolar derivations along the antero-posterior axis (F3C3, C3P3, P3O1 and F4C4, C4P4, C4O2). Because of artefacts, the data of two subjects were excluded from topographical analysis.

Statistics and data analysis

For statistical analyses the SAS® General Linear Model procedure (SAS Institute Inc., Cary, NC) with Greenhouse-Geisser correction was used (if a factor had more than two levels). Visually scored sleep variables and EEG power spectra in NREM sleep (Stages 2, 3 and 4) and REM sleep were analysed. Significant effects of the treatment and differences between bipolar derivations were assessed using one- and two-way analyses of variance for repeated measures (r ANOVA). Pairwise comparisons between baseline, placebo and zolpidem nights were performed with one-way r ANOVAs. The significance level was set at P < 0.05. To approximate a normal distribution, values of sleep latency, REM sleep latency, absolute power densities and ratios of power densities between derivations were log-transformed prior to statistical tests.


Concentration of zolpidem in plasma

The plasma concentration of zolpidem was below the limit of detection in all subjects prior to drug intake. At lights-out, i.e. 30 min after zolpidem administration, it was below the detection limit in three subjects, and 194.0, 116.0, 16.6 and 0.77 μg·L–1 in the others [mean concentration: 46.8 ± 29.3 (SEM) μg·L–1; n=7]. Four hours later, after completion of the PET scans, the mean zolpidem concentration was 95.57 ± 9.08 (SEM) μg·L–1 (range: 70.4–142.0 μg·L–1).

Sleep variables derived from visual scoring

To assess the effect of sleep deprivation and zolpidem administration on visually scored sleep variables and EEG power spectra, the baseline sleep recordings were subdivided into the first (≈ 0–4 h after lights-out) and second halves (≈ 4–7 h after lights-out) of the nights. For completion of the first half, an attempt was made to individually match the mean duration in the PET scanner between the zolpidem and the placebo nights.

Sleep variables of the first and second halves of the baseline and experimental nights are presented in Table 1. In comparison with baseline, REM sleep was reduced in both sleep deprivation conditions (i.e. after zolpidem and placebo) during the first 4 h of the sleep episodes. After zolpidem, Stage 4 and slow-wave sleep (SWS) were increased, whereas the time to fall asleep and the duration of Stage 1 were decreased. Wakefulness after sleep onset was prolonged after placebo. Independent of treatment, Stage 1 and the combined arousal variable (W + MT + 1) were lower during the final 3 h of the sleep episodes than in baseline.

Table 1.   Sleep variables during first and second half of baseline and experimental nights Thumbnail image of

When compared with placebo, zolpidem increased total sleep time (TST), SWS and Stage 4 in the PET scanner (Table 1). In contrast, wakefulness after sleep onset, Stage 1 and the combined arousal variable were decreased. With the exception of a shorter sleep latency for zolpidem, no significant differences between treatments were observed in the second half of the experimental nights.

EEG power spectra

The distribution of wakefulness, NREM sleep (Stages 1, 2, 3 and 4) and REM sleep, and the time course of slow-wave activity (SWA, power within 0.75–4.5 Hz) and spindle frequency activity (SFA, power within 12.25–15.0 Hz) are illustrated for one individual in Fig. 1. On all three nights, SWA and SFA exhibited the typical modulation by the NREM–REM sleep cycle, even though the initial REM sleep episode was occasionally skipped. SWA declined during the course of the sleep episodes. Note the prominent enhancement of SWA in the first NREM sleep episode of the sleep deprivation–placebo condition, which was not present in the sleep deprivation–zolpidem condition. Sleeping in the PET scanner caused no major disturbance. However, the occurrence of consolidated REM sleep episodes was delayed.

Figure 1. Sleep profiles, slow‐wave activity (SWA, EEG power within 0.75–4.5 Hz; C3A2‐derivation) and spindle frequency activity (SFA, EEG power within 12.25–15.0 Hz) of a subject during baseline and the two experimental nights. W=wakefulness (including MT); N=NREM sleep (Stages 1.

Figure 1. Sleep profiles, slow-wave activity (SWA, EEG power within 0.75–4.5 Hz; C3A2-derivation) and spindle frequency activity (SFA, EEG power within 12.25–15.0 Hz) of a subject during baseline and the two experimental nights. W=wakefulness (including MT); N=NREM sleep (Stages 1.

–4); R=REM sleep; S=slow-wave sleep, i.e. Stage 3 or 4; 2=Stage 2. Horizontal bars indicate time in the PET scanner. Arrows and vertical dashed lines indicate PET scans during sleep. The curves of SWA and SFA were smoothed by a three-point moving average.

Changes in mean EEG power spectra in NREM sleep (Stages 2, 3 and 4) are shown in Fig. 2. Spectra were computed separately for the first and second halves of the experimental nights, as well as for the entire nights. The all-night sleep deprivation effect in the placebo condition consisted of an increase of power in the delta/theta range (1.25–7.0 Hz) with the largest increase at 1.5 Hz, and a decrease in the 14.25–15.0 Hz bin. These changes were very similar to those observed during the first 4 h of the placebo sleep episode. However, they did not reach statistical significance in the PET scanner. EEG power was enhanced significantly in the 1.25–6.0 Hz range after transfer to the bed. After zolpidem, power was reduced in the upper delta, theta and lower alpha range (3.75–10.0 Hz), as well as in the high sigma band (14.25–16.0 Hz). The largest decrease occurred in the theta band. The all-night changes mimicked closely the changes in the PET scanner. Nevertheless, the reduction in the theta range (5.25–10.0 Hz) persisted during the final 3 h of the zolpidem sleep episode.

Figure 2. Relative EEG power spectra (C3A2.

Figure 2. Relative EEG power spectra (C3A2.

-derivation) in NREM sleep of the placebo and zolpidem sleep episodes. Spectra were computed separately for the first 4 h spent in the PET scanner (left), the final 3 h of the sleep episodes following the transfer to the bed (middle) and the entire nights (right). For each frequency bin, means (n=8, SEM) are expressed as a percentage of the corresponding value in the baseline (upper) or placebo night (lower; indicated as horizontal dashed lines at 100%). Values are plotted at the upper limit of 0.5-Hz or 1-Hz bins. Triangles above the abscissae denote frequency bins for which power differed significantly from the baseline (placebo, ▵; zolpidem, ▴) or placebo condition (one-way r ANOVA with factor ‘treatment’). Orientations indicate the direction of deviation.

Direct comparison of the all-night spectra after zolpidem and placebo ( Fig. 2, lower panels) revealed that the drug reduced power in the entire 1.75–11.0 Hz range. In the PET scanner, EEG power was enhanced in the lowest bin of the spectrum (0.25–0.5 Hz), and reduced in delta, theta and alpha frequencies (2.25–11.0 Hz). The relative spectra computed for the period after transfer to the bed showed a persistent depression of power in the low-frequency range (1.25–10.0 Hz) and an enhancement in the 12.25–13.0 Hz and 20.25–24.0 Hz bands.

Sleep deprivation reduced power in the alpha band in REM sleep after both treatments (placebo, 8.25–10.0 Hz; zolpidem, 8.25–11.0 Hz; data not shown). However, no significant drug-induced changes were observed in REM sleep.

To assess the time course of spectral changes in NREM sleep, the data were analysed for consecutive 2-h intervals. A two-way r ANOVA on log-transformed absolute power values with the factors ‘treatment’ and ‘2-h interval’ revealed significant main effects for ‘treatment’ (2.25–9.0 Hz) and ‘2-h interval’ (0.25–9.0 Hz, 12.25–13.0 Hz and 20.25–22.0 Hz), as well as significant interactions (5.25–9.0 Hz). In all three intervals, power was reduced in the delta and theta frequencies (0–2 h, 2.75–9.0 Hz; 2–4 h, 2.75–9.0 Hz; 4–6 h, 3.25–3.5 Hz and 4.75–8.0 Hz). The deviation from placebo was most pronounced in the 6–9 Hz range, but diminished over consecutive 2-h intervals.

Relationship of zolpidem plasma concentration with spectral changes in NREM sleep

To examine the relationship between the zolpidem plasma concentration ≈ 4.5 h after drug intake (i.e. at the transition from the PET scanner to the bed) and the all-night drug-induced change in power in NREM sleep, Pearson’s product moment correlation coefficients were computed for 1-Hz bins (not shown). The plasma level of zolpidem showed a significant positive correlation with relative power (percentage of placebo) in the high sigma frequency range (14.25–15.0 Hz, 15.25–16.0 Hz; P < 0.05, r2=0.84 and r2=0.76, respectively).

Regional differences of EEG spectra

The three panels of Fig. 3 show the relative NREM sleep spectra of the drug night for three bipolar derivations along the antero-posterior axis. Spectra are averaged over the left and right hemispheres. The pattern of drug-induced changes seen for the referential derivation (i.e. C3A2; Fig. 2) was also present for the bipolar derivations. It consisted of a reduction of power in the delta and theta bands, which was most pronounced in the theta range. A significant decrease of power in most bins between 15 and 21 Hz was observed in the anterior derivation (FC), which was not present in the middle (CP) and posterior derivations (PO), or in the referential derivation.

Figure 3. All‐night EEG power spectra of bipolar derivations along the antero‐posterior axis in NREM sleep (Stages 2, 3.

Figure 3. All-night EEG power spectra of bipolar derivations along the antero-posterior axis in NREM sleep (Stages 2, 3.

and 4) after administration of zolpidem. Power density values for each frequency bin (0.25-Hz resolution) are expressed as a percentage of the corresponding value after placebo (vertical lines represent ± 1 SEM, n=6). Bipolar derivations: FC=frontal-central; CP=central-parietal; PO=parietal-occipital. Mean values over the left and right hemispheres are plotted in the middle of the 0.25-Hz bins. Vertical bars above the abscissae represent significant log F-values: one-way r ANOVA on log-transformed absolute power values with the factor ‘treatment’ (d.f.=1,5).

The regional differences in the bipolar derivations along the antero-posterior axis are shown in Fig. 4, in which the mean values of the central-parietal (CP) and parietal-occipital (PO) derivations are expressed as a percentage of the frontal-central (FC) derivation. The F-values of a two-way r ANOVA with the factors ‘derivation’ (FC, CP, PO) and ‘treatment’ (placebo, zolpidem) confirmed that significant treatment effects prevailed in the 0.25–0.5 Hz, 1.25–10.5 Hz and 14.25–16.5 Hz range. Although the regional differences were similar for the placebo and zolpidem conditions, a significant interaction ‘treatment’ × ‘derivation’ was present in most bins within the 7.75–9.75 Hz band. The latter was due to a drug-induced posterior shift of power in these frequency bins. A significant main effect of ‘derivation’ was present in the spindle frequency range, i.e. 12.25–13.75 Hz.

Figure 4. Regional differences of all‐night power spectra in NREM sleep (Stages 2, 3, and 4.

Figure 4. Regional differences of all-night power spectra in NREM sleep (Stages 2, 3, and 4.

) along the antero-posterior axis after placebo (continuous lines) and zolpidem (dashed lines). (Upper) For each frequency bin, power density values in the central-parietal (CP, thick lines) and parietal-occipital derivations (PO, thin lines) are expressed as a percentage of the corresponding values in the frontal-central derivation (FC, 100%). Mean values over the left and right hemispheres are plotted in the middle of the 0.25-Hz bins (n=6). (Lower) Vertical bars represent significant log F-values: two-way r ANOVA on log-transformed absolute power density values with the factors ‘treatment’ (placebo, zolpidem; d.f.=1,5) and ‘derivation’ (FC, CP, PO; d.f.=2,10).


This study shows that, in comparison with placebo, zolpidem prolonged TST and the duration of SWS and Stage 4 during the first 4 h of recovery sleep after 40 h sleep deprivation. Furthermore, zolpidem reduced the duration of wakefulness after sleep onset, Stage 1 and the combined arousal variable. It also shortened the latency to return to sleep after awakening for the transition from the PET scanner to the bed. In contrast, no effect of zolpidem on REM sleep variables was detected. Whereas only minor modifications of sleep architecture have been found during undisturbed sleep in healthy subjects ( Merlotti et al. 1989 ; Brunner et al. 1991 ; Gillin et al. 1996 ), our findings are in good agreement with the reported effect of various doses of this drug under non-sleep conducive conditions ( Walsh et al. 1990 ; Cluydts et al. 1995 ; Roth et al. 1995 ). Accordingly, in this study, effects on sleep architecture were restricted to the first half of recovery sleep episodes, which were spent in the PET scanner. Although sleep continuity was not severely disrupted ( Fig. 1), 24 min of wakefulness during the first 4 h of the placebo recovery night (Table 1) indicate slightly impaired sleep. The observed changes in visually scored sleep variables suggest a drug-induced improvement in sleep in a mildly uncomfortable environment. Nevertheless, we detected profound sleep deprivation and drug-induced alterations in NREM sleep and REM sleep power spectra, which were strikingly similar to those found under undisturbed conditions. In NREM sleep, the all-night effect of sleep deprivation included enhancement of delta and theta frequencies (1.25–7.0 Hz), with the maximum increase at 1.5 Hz, and a concomitant reduction of power in the spindle frequency range (14.25–15.0 Hz; Fig. 2). Similar changes have been noticed repeatedly after 38–42 h of total sleep deprivation ( Borbély et al. 1981 ; Dijk et al. 1993 ; Aeschbach et al. 1996 ), as well as in studies of partial sleep deprivation, which induced lower sleep deficits ( Brunner et al. 1993 ; Endo et al. 1998 ). In REM sleep, power was reduced in the alpha range after both placebo and zolpidem administration. Significant reduction of alpha power in REM sleep has also been reported previously after total and partial sleep deprivation ( Borbély et al. 1981 ; Brunner et al. 1993 ), and has been attributed to the REM sleep deficit ( Endo et al. 1998 ; Roth et al. 1999 ).

Administration of 20 mg zolpidem 30 min prior to recovery sleep prevented the sleep deprivation-induced increase in delta and theta frequencies in NREM sleep ( Fig. 2). In fact, the hypnotic led to a prominent reduction in power between 3.75 and 10 Hz. Compared with placebo, zolpidem enhanced power in the 0.5-Hz bin and reduced power between 2.25 and 11.0 Hz during sleep in the PET scanner. Whereas the enhancement in the lowest frequency bin is reflected by the increase in SWS, the reduction of power in the higher delta and theta range was not reflected in the scored sleep stages. Opposite changes in slow EEG components below and above 1 Hz have been observed previously in the course of baseline sleep and after administration of the BDZ midazolam ( Trachsel et al. 1990 ; Achermann and Borbély 1997). During the second half of the zolpidem sleep episode, power spectra further disclosed a significant enhancement of EEG activity in the frequency range of sleep spindles (12.25–13.0 Hz) and in the 20.25–24.0 Hz band. These effects correspond to the distinct changes that have been described in non-sleep-deprived subjects for various BDZ receptor agonists ( Borbély et al. 1985 ; Dijk et al. 1989 ; Aeschbach et al. 1994 ), and were referred to as the ‘spectral GABA-BDZ signature’ ( Trachsel et al. 1990 ; Brunner et al. 1991 ). In accordance with a previous study ( Brunner et al. 1991 ), no zolpidem-induced changes in EEG spectra have been found in REM sleep. Sleep state-specific effects may differentiate the imidazopyridine from BDZ hypnotics, which also affect REM sleep spectra ( Borbély et al. 1985 ; Dijk et al. 1989 ; Aeschbach et al. 1994 ). Whether this pharmacodynamic specificity is related to the specific action of zolpidem at α1 receptors remains to be established.

The 20-mg dose of the hypnotic resulted in mean plasma levels of 95.6 μg.L–1 4.5 h after intake. This concentration range is in agreement with the pharmacokinetic properties of the drug. Specifically, in young adults, a single oral 20-mg dose was reported to produce a peak plasma concentration of 192–324 μg.L–1≈ 1–2 h after zolpidem administration, and to be eliminated with a plasma half-life time of 1.5–2 h ( Salvà and Costa 1995). It has previously been suggested that the changes in SFA or sleep spindles correspond to the pharmacokinetics of BDZ hypnotics (see Aeschbach et al. 1994 for discussion). Moreover, the BDZ receptor antagonist flumazenil has been reported to antagonize the increase in SFA but not the decrease in SWA ( Gaillard and Blois 1989). We found a positive correlation between the zolpidem plasma concentration in the middle of the night and the all-night change in the 14.25–16.0 Hz band. Although not significantly related to the drug concentration, a progressively diminishing reduction of power between 5.25 and 9.0 Hz was apparent in NREM sleep ( Fig. 2). This time course might reflect the gradual elimination of the drug.

Topographic analysis along the antero-posterior axis revealed significant interactions between treatment and derivation in most bins between 8 and 10 Hz ( Fig. 4). This indicates that, although the typical drug-induced changes were evident in all bipolar derivations, the compound affected the spatial distribution of EEG power. The reduction of power in the theta/alpha frequency range was ≈ 60% in the frontal-central derivation, and ≈ 40% in the central-parietal and parietal-occipital derivations ( Fig. 3). These regional differences in the effect of zolpidem might be related to particularly strong receptor binding of zolpidem in fronto-parietal structures ( Ruano et al. 1993 ), which may have led to a drug-induced posterior shift of power in NREM sleep. Dennis et al. (1988 ) reported an antero-posterior gradient of BDZ receptor density in primate cortex. Our findings provide further evidence for the usefulness of spatio-temporal analysis in pharmacological studies of the sleep EEG.

To interpret the results, the limitations in the study design need to be mentioned. Thus, the baseline data were recorded in the sleep laboratory under conditions different to the experimental nights, which included a 4-h sleep period in the PET scanner and a subsequent sleep period in a hospital bed. This inevitably entailed some disturbed sleep during the 4-h sleep period in the scanner. Moreover, although the subjects continued to sleep in a hospital bed, the environment differed from baseline. Nevertheless, the main results do not appear to have been seriously compromised. Identical conditions prevailed for the zolpidem and placebo recordings, which can, therefore, be compared directly. Moreover, the comparison of the placebo nights with baseline revealed the typical sleep deprivation effects, which were not masked by the less sleep-conducive conditions in the PET scanner.

The results suggest that 40-h sleep deprivation and 20 mg zolpidem have additive effects on sleep EEG power spectra. Therefore, the typical EEG changes in NREM sleep and REM sleep, which are associated with prolonged wakefulness are not likely to be mediated by agonistic activation of GABAA receptors. The present study further demonstrates that standard scoring criteria are inadequate for representing faithfully the changes in sleep induced by pharmacological agents, and that computer-aided analyses are mandatory in such studies.


We thank T. Berthold, PET Center, Division of Nuclear Medicine, University Hospital, Zürich, for his help during the study, Dr B. Gander, Department of Pharmacy, Swiss Federal Institute of Technology, Zürich, for providing us with placebo granulate and gelatin capsules, and Dr K. Rentsch, Institute of Clinical Chemistry, University Hospital, Zürich, for the determination of the zolpidem plasma concentrations. The comments from Dr I. Tobler are gratefully acknowledged. This research was supported by the Swiss National Science Foundation (grant no. 3100-042500.94 and 3100-053005.97) and the Human Frontiers Science Program RG 81/96.


  1. Present address: University of California, San Diego, Mental Heart Clinical Research Center, Psychiatry Service (116-A), VA San Diego Healthcare System, 3350 La Jolla Village Drive, San Diego, CA 92161, USA.