|Additional related information may be found at:|
|Neuropsychopharmacology: The Fifth Generation of Progress|
Processes Underlying Sleep Regulation
Processes Underlying Sleep Regulation
Alexander A. Borbély, Peter Achermann, Beat Geering, and Irene Tobler
Sleep and waking are the two major functional states which in homeotherms can be unambiguously identified by both behavioral and electrophysiological criteria. It is in particular the intensity dimension of sleep which can be viewed as allowing sleep to fulfill its putative need-dependent functions without disrupting the circadian sleep-wake rhythm. There is evidence that the circadian and sleep-wake dependent aspects of sleep regulation are based on separate mechanisms.
Three distinct processes can be identified which underlie the regulation of the sleep-wake cycle in humans and other mammals (Figure 1): (1) A homeostatic process determined by prior sleep and waking; (2) a clock-like circadian process that is largely independent of sleep and waking; * and (3) an ultradian process that is characterized by the alternation of the two basic sleep states non-rapid-eye-movement (nonREM) sleep and REM sleep.
RECORDING AND DATA ANALYSIS
Polysomnography provides an objective way of assessing sleep variables such as total sleep time, the number of awakenings after * sleep onset, sleep latency and sleep efficiency. In addition, sleep stages and sleep architecture can be determined. The widespread acceptance and use of the sleep scoring criteria of Rechtschaffen and Kales (101) have facilitated the comparison of sleep recordings obtained in different laboratories, thereby promoting the exchange of information. Figure 2 depicts a 20-s epoch of stage 4 with a predominance of slow waves in the electroencephalogram (EEG), a low level of electromyographic activity (EMG), regular heart rate and respiration. Figure 3 illustrates a REM sleep epoch with a low-amplitude EEG, rapid eye movements reflected in the electrooculogram (EOG), an almost vanished submental EMG, and irregularities in heart rate and respiration. Finally, Figure 4 (top panel) illustrates the hypnogram of an all-night sleep recording and in the lower panels a synoptic representation of different polygraphic variables which can be obtained on-line by computer-aided signal analysis techniques.
While the general use of the conventional sleep scoring system has greatly promoted human sleep research and sleep medicine, its limitations have become increasingly apparent. One of the main problems derives from the arbitrary specification of some criteria. This is exemplified by the substates of the nonREM sleep stages 2, 3 and 4, for which the major discriminating criterion is the abundance of EEG delta waves within a scoring epoch. The frequency range (0 –2 Hz), the minimum peak-to-peak amplitude of delta waves (75 mV), and the prevalence in a scoring epoch (20 – 50% for stage 3; > 50% for stage 4) are arbitrary. Consequently, variations that are unlikely to be of major physiological relevance (e.g., interindividual or age-related variations in EEG amplitude) may affect the sleep stage distribution. Problems may arise also from drug-induced changes. Drugs affecting those EEG parameters that are critical for scoring (e.g., the amplitude of delta waves) appear to give rise to prominent changes in sleep architecture, whereas equally potent drug effects on other EEG variables (e.g., an augmentation of beta activity) do not affect the sleep stage distribution. For example, after administration of a benzodiazepine hypnotic, the sleep EEG was still massively altered in the drug-free, post-drug night, whereas sleep architecture (as defined by the standard sleep scores) was no longer significantly changed (31). In such applications, the scoring criteria may give rise to misleading results by either exaggerating the drug-induced changes of sleep architecture or by inadequately reflecting alterations of the sleep EEG.
Analysis of the Sleep EEG
Spectral Analysis and Period Amplitude Analysis
Among the EEG variables assessed by computer-aided methods, one of the most important functional variables is "slow-wave activity" ("delta activity") which encompasses components of the EEG signal in the frequency range of approximately 0.5 – 4.5 Hz. A quantitative measure of slow-wave activity can be obtained by spectral analysis (e.g., 28), period amplitude analysis (e.g., 38), or similar methods. An illustration of slow-wave activity is depicted in panel 2 of Figure 4.Controversy exists concerning the respective merits of period-amplitude analysis and power spectral analysis (see 73). Power spectral analysis can be performed by transformation of the digitized signal into the frequency domain by the standardized Fast Fourier Transformation (FFT) procedure (41). Adding the squared real and imaginary parts of the resulting complex values leads to the power spectrum. It is represented by a set of values in adjacent frequency bands, the resolution corresponding to the reciprocal of the duration of the transformed time epoch.
Power spectra of nonREM sleep and REM sleep are illustrated in Figure 5. Note that the values for nonREM sleep exceed those of REM sleep up to 16 Hz. Due to their relationship to functional aspects of sleep regulation, the spectra in the frequency range of slow waves (SWA; red) and sleep spindles (SFA; blue) are of particular interest. A further measure based on power spectral analysis is the coherence spectrum (a correlation in the frequency domain) which may provide information on the functional connectivity between brain sites (5a) and (5b).
Period amplitude analysis is computed entirely in the time domain by determining the zero voltage crossings of the digitized signal and defining the signal portions between two consecutive zero crossings as half waves. For every half wave, a set of parameters can be determined, such as duration (period), amplitude, curve length and area under the curve.
Both methods have been widely used in human and animal research and both have their advocates. Some authors prefer power spectral analysis because it provides independent measures for frequency bands and because its theoretical basis is well established. Others select period amplitude analysis because of its alleged ability to discriminate changes in incidence (number) from changes in amplitude of EEG waves, and its intuitive grasp, since it is performed in the time domain.
Both methods are known to have specific limitations. Thus, period amplitude analysis is unable to detect fast waves riding on top of large slow waves. Power spectral analysis, on the other hand, does not provide separate measures for wave amplitude and wave incidence, and it requires a stationary signal. The two methods do not appear to be equivalent in their practical application, since they do not yield concordant results. Thus, on the basis of period amplitude analysis, wave incidence in the theta range was decreased during recovery sleep after extended waking in humans (66) and rats (95), whereas on the basis of power spectral analysis, theta power was increased in humans (28, 54, 57) and rats (30). The two methods have been compared, with the conclusion that band-pass filtering is necessary prior to period amplitude analysis even for the analysis of the lowest frequency range, and that the indiscriminate use of period amplitude analysis may give rise to spurious results (90, 73).
Pharmacological Effects on EEG Spectra
All-night spectral analysis of the sleep EEG revealed that sleep after intake of a hypnotic differs from drug-free, physiological sleep. The typical EEG changes induced by a benzodiazepine receptor agonist consist of a depression of activity in the low-frequency range (delta and theta bands) and an increase of activity in the spindle frequency range (29, 31). The reduction of low-frequency activity was still present in the post-drug night, at a time when the hypnotic effect had vanished or was strongly attenuated. Two new non-benzodiazepine compounds (zolpidem, an imidazopyridine, and zopiclone, a cyclopyrrolone) induced spectral changes that were very similar to those induced by benzodiazepine hypnotics (120, 34). It has been proposed that a "spectral EEG signature" may reflect the agonistic effect of hypnotics on the GABA-benzodiazepine-receptor complex. The similarity of the changes induced by two benzodiazepine receptor agonists is illustrated in Figure 6.
HOMEOSTATIC ASPECTS OF SLEEP REGULATION
Electrophysiological Markers of nonREM Sleep Homeostasis: Slow-Wave Activity and Spindle Frequency Activity
Researchers recognized as early as 1937 that sleep intensity is reflected by the predominance of slow waves in the sleep EEG (20). Subsequent studies confirmed that the responsiveness to stimuli decreased as EEG slow waves became more predominant (e.g., 126). Under physiological conditions, this EEG variable can be therefore regarded as an indicator of "sleep depth" or "sleep intensity". This statement applies both to humans and to animals. Thus, in the rat EEG, slow-wave activity was recently shown to be inversely related to the rate of spontaneous, brief awakenings from sleep (70). The global declining trend of slow-wave activity across consecutive cycles is illustrated in Figure 7 (average curve) and for an individual in Figure 4.
Sleep spindles constitute transient 12–15 Hz oscillations. The pattern of their occurrence corresponds to a large extent to the pattern of spectral power density in the frequency range of spindles (61). Unlike slow-wave activity, spindle frequency activity (power density in the 12.25 –15.0 Hz band) does not decrease during nocturnal sleep (Fig. 7) and, on average, shows an increasing trend (7). High values are seen at the onset and termination of nonREM sleep episodes.
Neurophysiology of Slow Waves and Spindles
Recently, slow waves and sleep spindles were shown to be closely related to cellular changes at the level of thalamic and cortical neurons (for reviews see: 106, 107). The progressive hyperpolarization of thalamocortical neurons induces membrane potential fluctuations in the frequency range of the sleep EEG. Thus, at moderate stages of hyperpolarization, oscillations are observed in the frequency range of sleep spindles; at more negative membrane potential fluctuations, they occur in the delta range (106, 107). Since the transition from waking to deep nonREM sleep is associated with an increased level of hyperpolarization (75), the progression from "spindle sleep" (stage 2) to slow-wave sleep (stages 3 and 4), and the partly inverse relationship between EEG power density in the spindle and slow-wave ranges (7, 61), seem to have a counterpart at the level of thalamocortical neurons. At the transition from nonREM sleep to either waking or REM sleep, the thalamocortical neurons become depolarized, whereupon the rhythmic oscillations vanish. The recognition of these relationships opens the possibility of investigating sleep homeostasis at the cellular level.
Based on these neurophysiological considerations, the typical changes in the sleep EEG can be interpreted as follows: due to a progressive membrane hyperpolarization in the early part of nonREM sleep, an increasing number of thalamocortical neurons start to discharge in an oscillatory mode. Presumably, some cells become hyperpolarized more readily than others, and, as a consequence, oscillations in the slow-wave range in some neuronal groups would occur concomitantly with spindle oscillations in other groups. This could explain why EEG power density in these two frequency ranges increases concomitantly in the initial part of the nonREM sleep episode (7). As the membrane hyperpolarization progresses further, increasing numbers of thalamocortical cells are expected to switch from the spindle mode to the slow oscillatory mode. On the level of the EEG, this corresponds to the middle part of the nonREM sleep episode, in which spindle frequency activity decreases and slow-wave activity increases (Fig. 7). Fluctuations at the cellular level at frequencies below 1 Hz were recently described (106a),107) and were shown to have a complement in the human sleep EEG (5, 13a).
Experimental Challenges of nonREM Sleep Homeostasis
"Sleep pressure" can be experimentally augmented by extending the waking period. It has been known for a long time that sleep deprivation gives rise to increased slow-wave sleep in the recovery night (see 22 for older references). Webb and Agnew (122) were among the first to show that slow-wave sleep increases as a function of prior waking. The quantitative assessment of slow-wave activity by all-night spectral analysis revealed that a night without sleep (i.e. 40.5 h of wakefulness) resulted in an enhancement of this EEG variable during recovery sleep (28).
Figure 8 (lower panels) depicts the global trend as well as the ultradian dynamics of slow-wave activity over successive nonREM-REM sleep cycles. The prolongation of the waking period causes a prominent rise of slow-wave activity during recovery sleep. A declining trend over 3–4 cycles is evident in both records. Note that the peaks are at a steady low level during the last 4 cycles of recovery sleep.
The enhancement of slow-wave activity by sleep deprivation was confirmed in several studies both in humans and other mammals (for references see 26, 110). The extent of the increase was shown to be a function of the duration of prior waking (58, 111, 113).
It is important to assess whether the effects of sleep deprivation are due to prolonged waking or to some concomitant factor. Sleep deprivation experiments typically involve in addition to the absence of sleep, the maintenance of an upright body posture during the habitual sleep period, sensory stimulation, motor and cognitive activity, social interaction, as well as eating and drinking at night. Nevertheless, there is strong evidence that it is the duration of waking and not other associated factors that determines the EEG response. If sleep deprivation is performed under a constant routine protocol which is similar to the sleep condition (i.e. physical inactivity, semi-recumbent posture, constant dim light), EEG slow-wave activity is increased to a similar level as after sleep deprivation with normal waking activities (51).
"Sleep pressure" can not only be augmented by sleep deprivation but reduced by daytime naps. Naps taken early in the day contain less slow-wave sleep than late naps (92, 87). Direct evidence for a monotonic rise of slow-wave activity in the course of waking was obtained in a study in which naps were scheduled at 2-h intervals throughout the day (16, 54). The rise of slow-wave activity could be approximated by a saturating exponential function and extended beyond the time of habitual sleep when sleep was postponed. Therefore, sleep deprivation does not cause a qualitatively different kind of EEG response, but merely augments a physiological trend.
A reduction of slow-wave sleep and slow-wave activity in nocturnal sleep has been reported after naps in the late afternoon or early evening (82, 66), but not after a nap in the morning (82, 85, 86). The smaller effect of the morning nap was predicted by the two-process model (48). According to this model, the sum of the cumulated slow-wave activity (i.e. slow-wave energy) during the nap and the post-nap sleep episode is equal to the cumulated value of the baseline sleep episode. This prediction was confirmed in two recent studies (67, 123).
REM Sleep Homeostasis
Not only nonREM sleep but also REM sleep is regulated by homeostatic mechanisms. Prolonged sleep deprivation enhanced REM sleep, but its increase in the recovery nights was delayed and persisted longer than the slow-wave sleep rebound (19). Selective REM sleep deprivation augmented REM sleep propensity, which was reflected by the increasing number of interventions required to prevent REM sleep episodes (50, 79, 63). Nevertheless, REM sleep rebound is not a consistent feature, and its extent may be limited (37, 63). There is increasing evidence that it is the inhibitory action of nonREM sleep which may prevent the full manifestation of REM sleep homeostasis. Thus a selective REM sleep deprivation during the first 5 h of sleep induced a REM sleep rebound in the remaining 2.25-h interval of the sleep episode (17). In another study, the induction of a substantial REM sleep deficit by curtailment of sleep duration during either 2 or 4 consecutive nights resulted in a REM sleep rebound in the recovery nights (33, 35). In these different experimental situations, the REM sleep rebound occurred either at a time when slow-wave propensity was low at the end of the sleep episode (17) or, in the partial sleep deprivation paradigm, when it was much less increased than REM sleep propensity (33, 35). Finally, a sleep-dependent disinhibition of REM sleep in the course of the sleep episode was demonstrated in a study using the forced desynchrony protocol (53). Taken together, these results show that slow-wave sleep and slow-wave activity in nonREM sleep, as well as REM sleep are homeostatically regulated. However, if slow-wave propensity is high, the manifestation of REM sleep homeostasis is suppressed.
CIRCADIAN ASPECTS OF SLEEP REGULATION
When subjects live under conditions in which time cues are absent, their sleep-waking rhythm is no longer synchronized to the 24-h cycle and 'free runs'. In early studies, it was observed that under such experimental conditions the cycles of sleep-waking, rectal temperature and other variables exhibited a period close to 25 hours (14, 42, 125). However, when subjects were kept in dim light, the circadian period was much closer to 24 h (24.1-24h 45). In some subjects, the sleep-wake cycle became desynchronized from the rhythm of rectal temperature ("internal desynchronization", 15).
During periods of synchronized or desynchronized free-run, the initiation of sleep tends to coincide with the falling limb of the rectal temperature cycle and sleep termination with the rising limb (108, 128, 129). In addition, long sleep episodes begin frequently close to the maximum of the temperature cycle, whereas short sleep episodes start close to the minimum (44, 129, 130). All of these observations demonstrate that the circadian pacemaker exerts a strong influence on the timing of sleep.
Both sleep propensity and the REM / nonREM sleep ratio are under circadian control. Various reports have stated that REM sleep is modulated by the circadian rhythm. However, barring a modulation of the period of the nonREM—REM sleep cycle, a change in the duration of one sleep state must by necessity be associated with an opposite change in the other. The maximum REM sleep propensity (i.e. the highest REM / nonREM sleep ratio) occurs a short time after the minimum of the rectal temperature cycle (128, 43).
Slow-wave activity in nonREM sleep undergoes little circadian modulation. In experiments in which sleep was initiated at different phases of the 24-h cycle, slow-wave activity and slow-wave sleep decreased during sleep episodes regardless of the circadian phase (10, 59, 53).
Circadian components of sleep regulation interact with homeostatic components, and particular experimental protocols are required to analyze the contribution of each component to sleep propensity and sleep architecture (see section ‘Forced desynchrony protocol' ).
INTERACTIONS OF HOMEOSTATIC AND CIRCADIAN PROCESSES
The relationship between slow-wave sleep and the duration of prior waking has been documented by Webb and Agnew (122) and placed into a theoretical framework by Feinberg (64). However, circadian aspects were not included in these analyses. The two-process model, originally proposed to account for sleep regulation in the rat (21, 23), posits that a homeostatic process (Process S) rises during waking and declines during sleep, and interacts with a circadian process (Process C) that is independent of sleep and waking. The time course of the homeostatic variable S was derived from EEG slow-wave activity. Various aspects of human sleep regulation were addressed in a qualitative version of the two-process model (22). A quantitative version of the model was subsequently elaborated in which Process S varied between an upper and a lower threshold that are both modulated by a single circadian process (46, 47) (Fig. 9). This model was able to account for such diverse phenomena as recovery from sleep deprivation, circadian phase-dependence of sleep duration, sleep during shift work, sleep fragmentation during continuous bed rest, and internal desynchronization in the absence of time cues (47).
Elaboration of the two-process model triggered numerous experimental studies to test its predictions. Thus the time constant of the declining limb of Process S during sleep, which had been derived from the changes of slow-wave activity in an early sleep deprivation study (28), was found to be similar in other studies (16). Moreover, a nap study demonstrated that a saturating exponential process with a time constant of 18.7 h could closely approximate the changes of slow-wave activity during the day, thereby providing direct evidence for the rising limb of Process S (16, 54, 58). A good correspondence between the predictions of the model and the empirical slow-wave activity data was found also in repeated partial sleep deprivation studies (33) and in a sleep deprivation study in habitual short and long sleepers (8).
In an elaborated version of the model (proposed by 16, 55 and formalized by 1) it is the change of S, and not its level, which is proportional to the momentary slow-wave activity. The extension of the model allowed one to address not only the global changes of slow-wave activity, as represented by Process S, but also the changes within nonREM sleep episodes (1). The magnitude of the intra-night rebound after selective slow-wave sleep deprivation in the first 3 h of sleep was in accordance with the prediction.
A series of experiments was designed to test the time course of Process S during extended sleep episodes that were initiated at various phases of the 24-h cycle. Long sleep episodes starting at regular (60, see also 84), phase-advanced (57, 120) and phase-delayed bedtimes (59) were studied. Regardless of the circadian phase, mean slow-wave activity showed a decline over the first 3–4 nonREM-REM sleep cycles and then remained at a constant low level (see Fig. 8, lower right panel).
Another advanced version of the model was subjected to an optimization procedure based on the weighted least-square error method, using the mean time course of empirical slow-wave activity from a large data set (16 subjects, 26 nights) as a template (6). A sensitivity analysis showed the system to be rather robust to moderate variations of the parameter values. To test the performance of the model, the estimated parameter values were used to simulate data from three different experimental protocols. Empirical REM sleep data were used to activate the REM trigger parameter in the model. In general, a close fit was obtained between the simulated and empirical slow-wave activity data and their time course (Fig. 10). In particular, the occurrence of late slow-wave activity peaks during extended sleep could be simulated. Minor discrepancies in later cycles of sleep initiated in the evening or in the morning suggested indirect or direct circadian influences on slow-wave activity. The simulations demonstrated that the model could account quantitatively for empirical data and could predict the changes induced by the prolongation of waking or sleep. In a recent study, the time course of slow-wave activity after a daytime nap could be closely simulated, although some discrepancies were present (123).
Folkard and Åkerstedt (68, 69) simulated the changes of daytime alertness by the combined action of a homeostatic process, a circadian process and a sleep inertia process. It is noteworthy that the time constants derived from sleepiness ratings and from the sleep EEG are in a similar range. This model was recently validated (11) and applied for predicting sleep latency (12) and sleep duration (13).
Attempts were made to integrate various conceptual and modeling approaches. Thus, models proposed by various authors were linked as "modules" to form a combined model (3). Although this model's properties need to be further examined, preliminary simulations demonstrated that it was feasible to incorporate homeostatic, circadian and ultradian factors regulating nighttime sleep and daytime sleep propensity into a single model. Similarly, homeostatic, circadian and ultradian processes were combined in the recent extension of the limit cycle reciprocal interaction model (93).
Recently, modeling was extended to animal sleep by simulating nonREM sleep homeostasis in the rat (71). Slow-wave activity determined for consecutive 8-s epochs over a 72-h recording period (24-h baseline, 24-h sleep deprivation, 48-h recovery; 70) served as the data base for the simulation. Process S was assumed to decrease exponentially in nonREM sleep and increase according to a saturating exponential function in waking and REM sleep. After optimizing the initial value of S as well as its time constants, a close fit was obtained between the hourly mean values of slow-wave activity in nonREM sleep and Process S. In particular, the typical changes of slow-wave activity, such as its biphasic time course during baseline, its initial increase after sleep deprivation and the subsequent prolonged negative rebound could be simulated.
Independence of S and C
There is evidence that homeostatic and circadian facets of sleep regulation can be independently manipulated and therefore may be controlled by separate mechanisms. Thus, during a 72-h sleep deprivation period, the subjective alertness ratings showed a prominent circadian rhythm (9). Conversely, in a study in which the phase of the circadian process (as indexed by body temperature and plasma melatonin) was shifted by bright light in the morning, the time course of slow-wave activity was unaffected (56).
Animal experiments clearly indicate that the homeostatic and circadian facets of sleep regulation are mediated by separate processes. A conflict can be created between their tendencies by ending a 24-h sleep deprivation period in the rat at the onset or in the middle of the dark period, the circadian period of predominant waking (25, 94, 119). In this experimental paradigm, slow-wave activity showed a rebound in two stages: an immediate increase followed by waking and a second, delayed increase at light onset.
Another line of evidence for the independence of homeostatic and circadian processes of sleep regulation stems from experiments in which the circadian facet was disrupted by lesions of the suprachiasmatic nuclei (94, 115, 121). Arrhythmic rats were subjected to 24-h sleep deprivation to test their ability to compensate for sleep loss. Measures of slow-wave activity in nonREM sleep and the amount of REM sleep were increased.
Finally, recent experiments demonstrated that animals exposed to different photoperiods showed dramatic changes in the 24-h distribution of sleep and waking, while nonREM sleep homeostasis (as reflected by slow-wave activity) and its response to sleep deprivation remained unaffected (49, 72).
Forced Desynchrony Protocols
The influence of circadian and homeostatic (i.e. sleep/wake-dependent) factors on sleep can be studied in forced desynchrony protocols. While the usual ratio of sleep and waking is maintained, the period of the sleep-wake cycle is scheduled to be either shorter or longer than the period of the circadian pacemaker.
In a recent study, subjects were scheduled to a 28-h sleep-waking cycle, one third of which was reserved for sleep (52, 53 and 61a). Light intensity during waking was low to minimize light-induced effects on the circadian pacemaker. Rectal temperature, melatonin and cortisol served as circadian phase markers. The results confirmed that sleep propensity is highest at the minimum of the circadian rhythm of rectal temperature. Sleep propensity was lowest 16 h later. Sleep consolidation was optimal when sleep was initiated close to this phase. In addition, spindle frequency activity was influenced by both homeostatic and circadian factors.
Analysis of the forced desynchrony protocol confirmed the following main tenets of the two-process model: in accordance with its basic assumption, slow-wave activity (the marker of Process S) was determined mainly by homeostatic (i.e., sleep-wake dependent) factors, whereas REM sleep (or rather the REM sleep/NREM sleep ratio) depended on both homeostatic and circadian factors. Furthermore, the sleep-related inhibition of REM sleep, postulated in the first version of the model (22), was confirmed. Finally, data showed that not only the timing of sleep but also the changes in daytime vigilance were governed by the interaction of Processes S and C (47); the rising homeostatic sleep pressure during waking was compensated by the declining circadian sleep propensity (4, 32, 62). Conversely, during sleep the rising circadian sleep propensity served to counteract the declining homeostatic sleep pressure, thereby ensuring the maintenance of sleep (52). Data from the forced desynchrony protocol indicated that non-linear interactions also must be considered.
Although the homeostatic and circadian sleep processes can be delineated on the basis of physiological markers, the underlying brain mechanisms remain to be elucidated. Nevertheless, there are promising new developments in this direction.
‘Sleep Substances' and Molecular Genetics
Since the pioneering studies of Ishimori (78) and Piéron (100), a large number of experimental studies have been devoted to the study of putative, endogenous sleep substances (for reviews see 27, 74, 77, 89). Substances that have received particular attention in recent years include the cytokines (interleukin-1, tumor necrosis factor, interferon, etc.), prostaglandin D2 and adenosine (18, 103 ). In assessing the status of these and other chemical entities as 'sleep substances', it is important to specify the criteria upon which such an evaluation is based. In the context of physiological sleep regulation, two criteria were proposed (27). First, a 'sleep substance' should induce a physiological type of sleep, as defined by behavioral, electrographic and physiological criteria. The behavioral signs of sleep include a typical body posture, reduced responsivity to environmental stimuli, a rapid state reversibility and a species-specific alternation between sleep and waking episodes. The electrographic criteria encompass the typical EEG, EOG and EMG changes of nonREM sleep and REM sleep; moreover, these two substates of sleep should exhibit a cyclic alternation. The second, basic requirement for a ‘sleep substance' is its variation in the organism (i.e. change in the level, metabolism, receptor affinity, etc.) in association with sleep and waking. Whereas the first requirement may be met by a number of substances, no unambiguous evidence as yet exists that this is the case for the second requirement. This may be due to the fact that such changes may be difficult to detect because they are restricted to specific brain sites. An additional problem is that several putative 'sleep substances' are known to be involved in functions other than sleep. Therefore, a specific, sleep-related function is difficult to demonstrate.
The search for endogenous sleep-promoting agents has received a new impetus by the advent of molecular genetic techniques (for a review see 118). Various studies have shown that immediate early genes (IEGs) and their products (c-fos, NGFI-A) are modulated by sleep and waking (e.g., 39). The expression of these factors is higher during waking than during sleep and depends on the noradrenergic innervation of the forebrain (40). FOS protein, an IEG product, was recently used to trace the diencephalic circuitry involved in the modulation of arousal (104). However, both IEGs and late genes are influenced by the sleep-wake cycle. Based on the concept of sleep homeostasis, a subtractive hybridization technique was used to identify mRNAs in rat brain whose level was changed after sleep deprivation (102). The products of two of these transcripts were shown to be neurogranin (97) and the novel brain protein dendrin (98). The level of the transcripts as well as the proteins were altered by sleep deprivation in specific brain regions. These are promising first results indicating that the use of contemporary techniques (e.g., differential display) may disclose specific genes and gene products which play a key role in sleep regulation (118a). Another approach based on molecular genetics consists of the use of knockout and transgenic mice as a model of human sleep pathology (e.g., the prion disease fatal familial insomnia; 116, 117).
Local, Use-Dependent Facet of Sleep Homeostasis
Although it has been clearly demonstrated that putative markers of sleep intensity, such as slow-wave activity, are determined by the duration of prior wakefulness, the significance of this effect and the underlying mechanisms remain obscure. Recently, two hypotheses have been advanced which both imply that regional increases in neuronal activity and metabolic demand during wakefulness may result in selective changes in EEG synchronization of these neuronal populations during nonREM sleep. Benington and Heller (18) proposed that adenosine, which is released by neurons and glial cells throughout the central nervous system during periods of high metabolic activity, plays a key role in EEG synchronization. Krueger and Obál (88) hypothesized that factors such as interleukin-1 (IL-1) and the IL-1-receptor antagonist are produced by glial cells in response to neurotransmitter stimulation. As a consequence, these factors are assumed to modulate "sleep propensity" and EEG synchronization locally by strengthening the synaptic connectivity of neuronal groups which have been under-stimulated during wakefulness.
To investigate whether regional activation during wakefulness results in regional changes in EEG power spectra during nonREM sleep, we subjected subjects to unilateral activation of the somatosensory system by vibration of the hand (83). Positron emission tomography (PET) studies have demonstrated that such treatment results in increased regional glucose metabolism and regional cerebral blood flow, primarily in the contralateral somatosensory cortex. Indeed, if the activity of neural elements during wakefulness is a determinant of subsequent slow oscillations in sleep, we predicted that vibration of the hand should result in an enhancement of EEG synchronization in the contralateral somatosensory cortex.
The sleep EEG was recorded from frontal, central, parietal and occipital derivations and subjected to spectral analysis. The interhemispheric asymmetry index was calculated for spectral power in nonREM sleep in the frequency range of 0.25–25.0 Hz. In the first hour of sleep following right-hand stimulation, the interhemispheric asymmetry index of the central derivation was increased relative to baseline, corresponding to a shift of power towards the left hemisphere (Fig. 11). This effect was most prominent in the delta range, was limited to the first hour of sleep and was restricted to the central derivation situated over the somatosensory cortex. No significant changes were observed following left-hand stimulation. Although the effect was small, it was consistent with the hypothesis that the activation of specific neuronal populations during wakefulness may have repercussions on their electrical activity pattern during subsequent sleep.
This hypothesis has been further support by a recent topographical EEG study (124a). One of the major findings was an apparent sleep-wake dependent hyperfrontality of slow-wave activity (Fig. 12). In the first nonREM sleep episode, power in the 2-Hz bin was highest in the anterior derivation (see also 36, 127). However, over subsequent episodes, the anterior preponderance of power declined and vanished entirely in the fourth episode. Such a time-dependent shift in the power ratio was limited to anterior sites (e.g., fronto-central/central-parietal derivation) and was not present at posterior sites (e.g., central-parietal/parietal-occipital derivations) (124a). This result supports the notion of a specific involvement of frontal parts of the cortex, which seem to exhibit the largest activity during wakefulness, in the sleep process (76). In the framework of the two-process model, the results indicate that Process S declines in the anterior region of the brain at a steeper rate than in posterior regions; therefore, the homeostatic nonREM sleep regulating process may exhibit regional differences in the brain.
The functional neuroanatomy of REM sleep has been recently explored with PET and statistical parametric mapping (91). An examination of the regional cerebral blood flow distribution as an index of neuronal activity revealed that the intense and widespread cortical activation during REM sleep was not uniform. In particular, those cortical areas that received input from the amygdala were activated. If the results of the PET study can be related to the EEG spectra of the present analysis, the relative prefrontal deactivation during REM sleep may have a counterpart in the marked antero-posterior shift of EEG power during nonREM-REM sleep transitions (124a).
Taken together, these results indicate that the neuronal processes underlying the sleep EEG differ between brain regions. These data also support the notion that sleep is not only a global brain state but also a local, use-dependent phenomenon (88). It has long been known that unihemispheric sleep prevails in marine mammals (96), and that in dolphins, deprivation of sleep in one hemisphere gives rise to a unihemispheric compensatory response (99). The recent data suggest that this regulatory aspect may be subtly manifested also in terrestrial species.
The fundamental problem behind all investigations of sleep regulation is the function of sleep. The ubiquitous occurrence of sleep suggests that it has been an important factor in evolution. When did sleep arise in the evolutionary process? Considering the two major processes of sleep regulation, it has been suggested that the circadian pacemaker modulating rest and activity is the antecedent of the sleep-wake cycle, and that the homeostatic facet has evolved to provide a more flexible and need-dependent regulation (23, 25). However, due to the scarcity of data, it is still unclear whether such a sequence of events is plausible. A sleep-like resting state that includes a compensatory response to forced activity has been described in invertebrates (80, 81, 109, 112, 114). This is an avenue of research which deserves to be pursued. That ‘recovery' during sleep may involve specific memory functions has been demonstrated by an activity pattern of hippocampal pyramidal cells in the sleeping rat that reflects the changes produced by prior experience (105). The consolidation of memory may go along with a discarding of useless information gathered during waking. Perhaps William Golding had this similar image in mind when he wrote: "Sleep is when all the unsorted stuff comes flying out as from a dustbin upset in a high wind".
The authors' research was supported by the Swiss National Science Foundation and the Human Frontiers Science Program.