Siegel, J. M. Brainstem mechanisms generating REM sleep. In: Principals and Practice of Sleep Medicine, Second Edition. Edited by M. K. Kryger, T. Roth, W. C. Dement. New York: Saunders, 2000.
Brainstem Mechanisms
Generating REM Sleep
J. M. Siegel Ph.D.
Professor of Psychiatry and Biobehavioral Sciences and
Member of the Brain Research Institute, UCLA;
Chief of Neurobiology Research, Sepulveda V.A. Medical Center
In: Principals and Practice of Sleep Medicine, M. H. Kryger et
al., (Eds.), Saunders, 2000.
Supported by the Medical Research Service of the Veterans Administration and PHS grants NS14610, HL41370 and HL/MH/HD/AR/NS59594
REM sleep is the state in which our most vivid dreams occur. For this reason the elucidation of its function has been the goal of some of the greatest ancient and modern thinkers. The modern era of sleep research, resulting from the physiological identification of the REM sleep state in man and animals, has added a new dimension to this search. While modern sleep researchers have offered additional, physiologically based, theories of REM sleep function, the adaptive function of this state remains unclear. However, tremendous progress has been made during the past 40 years in understanding the mechanism producing REM sleep. Since function and mechanism questions ultimately merge, it seems likely that the further exploration of the physiology of REM sleep will, within the next few decades, produce an answer to the age-old question of the function of the "Dream State".
I WHAT IS REM SLEEP?
REM sleep has also been called "paradoxical sleep", "desynchronized sleep", "active sleep" and "dream sleep". Each of these terms reflects a slightly different emphasis on what its essential, defining features are. In most cases this is merely a semantic issue. However, when one studies a "new" species, works with young or brain injured animals, or is limited to surface EEG electrodes, the question of definition becomes critical.
In intact, adult humans, REM sleep is identified by the simultaneous presence of a relatively low voltage (sometimes termed "desynchronized") cortical EEG, an absence of activity in the antigravity muscles (atonia), and periodic bursts of rapid eye movements. The rapid eye movement bursts are often accompanied by changes in respiration and by phasic muscle activity (i.e. twitching of the distal somatic musculature and face). In human infants and in many young animals, "active sleep" the ontogenetic precursor of the REM sleep state, is not accompanied by a low voltage cortical EEG or by muscle atonia. In these young animals, it is identified by the muscle twitches and eye movements that recur periodically during sleep.
Studies with deep electrodes have established that while the EEG of the neocortex is low in voltage during the REM sleep state, the EEG of the hippocampus is increased in size at a 4-10 Hz (theta) frequency (Fig. 1-top). Theta rhythm is produced by the pyramidal cells of CA1, in the dentate gyrus92 and in the medial entorhinal cortex77. High voltage hippocampal theta can also be observed in active waking, particularly during certain classes of movements, at times when neocortical EEG is low voltage142. Another sub-cortically observed phenomenon which is characteristic of REM sleep, is the Ponto-Geniculo-Occipital (PGO) spike. These waves, which are generated in the pons48, propagate rostrally through pathways in the vicinity of the brachium conjunctivum66,100 and project through the lateral geniculate and other thalamic nuclei35 to the cortex. PGO spikes are one of several "phasic" events of REM sleep. Among the most significant of these phasic events are eye movements, changes in respiration87 and irregularities of heart rate10.
One phenomenon that is unique to REM sleep, is the total loss of activity in the antigravity musculature48. In most animals, only the diaphragm and extraocular muscles retain their tone during REM sleep. While muscle tone may also be reduced in non-REM sleep, only in REM sleep is tone consistently absent. The atonia of REM sleep may be interrupted by muscle twitches, which accompany bursts of PGO spikes and eye movements.
All of the above mentioned EEG and EMG phenomena can be readily explained in terms of changes in neuronal activity. Thus, the low voltage neocortical EEG results from the asynchronous activity of thalamo-cortical projection neurons (see Steriade, this volume). The high voltage EEG of nonREM sleep results from synchronized rhythmic activity in adjacent cortical neurons, at approximately the frequency observed in surface electrodes. Unit activity rate does not change in all cortical regions from nonREM sleep to REM sleep, but the pattern changes dramatically24.
Hippocampal unit activity shows the reverse pattern. In REM sleep, hippocampal, dentate and entorhinal units become highly synchronized due to inputs from cholinergic and GABAergic cells in the medial septal nucleus14,88,146. This results in the high voltage 4-8 Hz signal recorded across the pyramidal layers.
The loss of activity in antigravity muscles results from a cessation of discharge in the motoneurons supplying these muscles. Intracellular recordings have demonstrated that this cessation is due to a hyperpolarization of the motoneurons. This hyperpolarization is thought to result from the release of glycine onto the motoneurons (see Chase this volume). In addition to active hyperpolarization, a disfacilitation of motoneurons may also occur in REM sleep. Neurons containing serotonin and norepinephrine innervate motoneurons and depolarize (facilitate) them. A loss of serotonergic facilitation in REM sleep may make a major contribution to the REM sleep-related reduction of muscle tone in the hypoglossal nucleus58. Aminergic facilitation may have a role in the control of tone in other muscle groups64.
II WHERE IS REM SLEEP GENERATED?
A powerful technique for localizing function, is to divide the brain in half and ask "which half has the function of interest?" If this question can be clearly answered, one can repeat the process, continuing the "half split technique" until the critical area is localized. One might suspect that since brain systems are so thoroughly interconnected, any major subdivision of the brain would so disorganize brain function that REM sleep-like activity would not be present in either half. This is clearly not the case. To an astounding degree, the REM sleep generator mechanisms can survive disconnection from well over 95% of the rest of the central nervous system. Conversely, destruction of a small portion of the brainstem can permanently prevent REM sleep. This characteristic of REM sleep has been repeatedly demonstrated in a number of lesion and transection studies, and has allowed us to achieve a relatively precise localization of the neurons critical to this state.
A Transection studies
One can cut through the midbrain in the coronal plane, so as to separate the brainstem from diencephalic and telencephalic structures. Animals with such lesions can survive for weeks or months. These animals manifest a striking dissociation between states in the forebrain and brainstem as defined by polygraph recording; by transecting the neuraxis, two independent generators of behavioral state are created, one in front of and one behind the transection. When transections are placed at levels A or B in Fig. 2, all the brainstem signs of REM sleep can be recorded caudal to the cut. Thus, atonia, rapid eye movements and PGO spike bursts, as well as a REM sleep-like activation of reticular formation units, occurs in a regular ultradian rhythm48,147,89.
A striking phenomenon is that the amount of REM sleep seen after such transections is a function of central temperature. Reducing the body temperature of the pontine cat (i.e., the cat with transections at A or B in Fig. 2) produces a progressive increase in REM sleep. REM sleep constitutes less that 10 percent of the time at a body temperature of 36° and 80 percent at a temperature of 23° 49.
The cerebral cortex of animals with transections at levels A and B shows both high and low voltage states that alternate spontaneously147. No PGO spikes are seen in the forebrain. No eye movements or variation in pupil diameter linked to EEG states occurs, since all of the extraocular motor nuclei and the Edinger Westphal nuclei, controlling pupil diameter, are behind the cut.
REM sleep is present caudal to midbrain transections and absent rostral to such transections. Therefore, one may conclude that structures rostral to the midbrain are not required for REM sleep and that structures caudal to the midbrain are sufficient to generate REM sleep.
Transection at the junction of the spinal cord and medulla (Fig. 2 level C) does not prevent the features of REM sleep from occurring rostral to the cut2,90. Atonia in muscles innervated by spinal motoneurons is, of course, disrupted by the lesion. Transection of the spinal cord at the mid-thoracic level in the otherwise intact animal, produces the Sherrington-Shiff phenomenon, which interferes with atonia generation, although REM sleep is otherwise normal80. Thus, whereas spinal mechanisms interact with rostral structures in the generation of REM sleep atonia, they are not essential for the generation of the REM sleep state itself.
From the above we may conclude that structures caudal to the midbrain and rostral to the spinal cord are necessary and sufficient for REM sleep.
This technique has been carried one step further by transecting between the medulla and the pons and maintaining the animals for extended periods to allow the fullest possible recovery from the transection124,130. As was the case with pontine sections, the brain regions rostral and caudal to the cut produce independent physiologically defined states. The medulla cycles regularly between an activated state and a quiescent state130. The activated state is characterized by high levels of muscle tone, similar to those seen in active waking, and by accelerated respiration and heart rate. The quiescent state is characterized by lower levels of muscle tone, resembling those seen in nonREM sleep, and by slow regular respiration. Periods of muscle atonia are never seen (Fig. 3). Unit activity in the medial medulla during the quiescent state resembles that seen in this region in nonREM sleep, i.e. it is slow and regular. Unit activity during the activated state increases as it does in waking in the intact animal. However, medial medullary neurons do not show the characteristic burst-pause discharge pattern seen in REM sleep. Therefore, the medulla and spinal cord, disconnected from rostral structures, show spontaneous variations in level of arousal, but do not show the medullary signs of REM sleep.
Structures caudal to the pons are not sufficient to generate REM sleep.
A very different picture is seen in rostral structures after transection between the pons and medulla124 (Fig. 2, level E). Three states can be distinguished rostral to the transection (Fig. 1-bottom). The first is a high voltage EEG state without PGO spikes, resembling nonREM sleep. The second is a low voltage state without PGO spikes, resembling waking. The third state is a low voltage state with PGO spikes. The PGO activity occurs in irregular bursts and as isolated spikes in a manner very similar to that seen in REM sleep. Midbrain reticular units show irregular burst-pause patterns of discharge in conjunction with this third state, as they do in REM sleep125 (Fig. 4). There are, however, significant differences between this state and the REM sleep state seen in intact animals. In transected animals, the state of PGO spike bursts with low voltage EEG may last for hours, compared to a maximum of 20-30 minutes for REM sleep in intact cats. Furthermore, rapid eye movements do not consistently accompany the PGO bursts of this state.
Partial transections of the brainstem at approximately this same level, produce somewhat different phenomena47. Transections of the medial portion of the brainstem reticular formation (up to 5-mm lateral to the midline) produce animals that are capable of thermoregulation143, unlike animals with complete transections. In these animals, PGO spikes with EEG desynchrony (as in REM sleep) are seen in a state with motor activation that behaviorally resembles waking. Responsiveness to the environment is maintained, distinguishing this state from normal REM sleep, whereas the presence of PGO spikes and spike bursts distinguish this state from normal waking.
From the above one can see that when the pons and caudal midbrain are connected to mid- and fore-brain structures, some signs of REM sleep are seen in these rostral structures. When the pons and caudal midbrain are connected to the medulla and spinal cord, as in the midbrain decerebrate animal, the defining signs of REM sleep are seen in caudal structures.
One can then transect through the middle of the pons (Fig. 2, level D) and again ask the question "which side has REM sleep?"124,118. After this transection, the caudal pons and medulla cycle between the aroused and quiescent states seen in the medullary animal114. No atonia or other signs of REM sleep are present. The forebrain EEG shows low voltage, waking like states and high voltage nonREM sleep-like states. Eye movements that can be used to help identify sleep-wake states in the forebrain, are limited to those that can be generated by the occulomotor nucleus. This is because the abducens nucleus is caudal to the cut (the trochlear nucleus and nerve are usually damaged by such a transection). The EEG of the forebrain shows an irregular alternation of high voltage and low voltage states. However, the low voltage states are not accompanied by spontaneous rapid eye movements. During low voltage states, vertical pursuit eye movements and changes in pupil diameter can be evoked by external visual stimuli. However, the stereotyped pattern of spontaneous rapid eye movements with myosis (constriction of the pupil) that characterizes REM sleep is not present9. The low voltage state seen in the rostral portion of these preparations resembles waking.
The high voltage state in the forebrain may be accompanied by PGO spikes and PGO spike bursts. However, midbrain unit activity is greatly decreased at these times, directly the opposite of the pattern seen in REM sleep in the intact animal and in animals transected at the ponto-medullary junction. PGO spikes do not occur in the low voltage state. Therefore, with this midpontine transection we have reached the limit of the transection technique. The major defining characteristics of the REM sleep state are absent on both sides of the transection, even in chronically maintained animals, although some aspects of REM sleep are present.
While the foregoing indicates that the caudal pons and adjacent midbrain are necessary for the generation of REM sleep-like states in both rostral and caudal structures, one may ask if this region is in and of itself sufficient to generate the mesopontine aspects of REM sleep?" One can monitor REM sleep signs after transecting both rostral and caudal to the pons, producing an isolated "pons-caudal midbrain" preparation73 (Fig. 2, levels B and E). In this case, the pons generates periodic episodes of rapid eye movements and PGO spikes in a pattern which, in the intact animal, is seen only in REM sleep. Neuronal activity has not been monitored in this preparation. However, the presence of PGO spikes and rapid eye movements in a REM sleep-like pattern is impressive evidence of the pontine control of these basic aspects of REM sleep.
From the above we can conclude that the pons and the caudal midbrain region are both necessary and sufficient to generate some of the basic phenomena of REM sleep.
B Pontine anatomy and nomenclature
Several different systems for naming the nuclei of the brainstem reticular formation have been developed. Different physiologists have employed different naming systems in describing their work, resulting in much unnecessary confusion and controversy. I will briefly review some of the major naming systems applied to the brainstem and how they relate to each other. I will then present the results of studies attempting to further localize REM sleep generating functions within the pons and caudal midbrain reticular formation.
The simplest nomenclature system divides the brainstem reticular formation into the midbrain, pontine and medullary regions. On the ventral surface of the human brainstem there is universal agreement on the meaning of these terms. The cerebral peduncles define the midbrain, the basilar pons defines the pons. The medulla is the region between the pons and the spinal cord, its caudal boundary being defined by the decussation of the pyramidal tracts and the appearance of the C1 ventral root. Dorsally, the caudal limit of the inferior colliculus defines the caudal boundary of the midbrain. The dorsal boundary between the pons and the medulla, a critical point for descriptions of REM sleep generating mechanisms is less clearly defined. In the human brain, the dorsal pons is commonly considered to include the abducens nucleus, the vestibular and cochlear nuclei, the facial nucleus and all adjacent reticular nuclei19. The medulla encompasses the regions caudal to these landmarks. This usage has not been consistently followed in the cat. In some studies in the cat, the abducens nucleus has been considered to mark the caudal most portion of the pons, with the vestibular, cochlear and facial nuclei considered to be in the medulla. Therefore, it is critical to describe where the area of interest is located with respect to the abducens and caudal cranial nerve nuclei, rather than using the terms "pons" or "medulla". Even less clear is the boundary between the "medial" and "lateral" reticular formations. There is no landmark to divide the reticular formation in this way. Common usage in the cat is to reserve the term "medial" for the region from the midline to two millimeters lateral to the midline.
An alternate nomenclature system, going from rostral to caudal, divides the core of the brainstem reticular formation, into the midbrain reticular formation (RF), nucleus reticularis pontis oralis (RPO), nucleus reticularis pontis caudalis (RPC), and nucleus gigantocellularis (NGC) of the medulla. As used by Brodal13, the RPO begins at the rostral limit of the pons and is devoid of giant cells. The RPC is immediately caudal to the RPO and contains giant cells. There is no cranial nerve landmark defining the boundary between these nuclei. However one of the most widely used atlases of the cat brainstem133 places the junction at the level of the dorsal tegmental nucleus. The uncertainty over the definition of the RPO can be seen in the conflicting descriptions given by different authors working with the cat brain. The Snider and Niemer133 atlas has the rostral boundary of RPO at 2.5 mm anterior to stereotaxic zero, while Carli and Zanchetti16 give it as 1.0 mm anterior to zero, and the Reinoso-Suarez91 atlas has the rostral boundary at 2.0 mm posterior to zero. The junction of the RPC and NGC is at the level of the abducens nucleus, although this usage is also not universally followed. The caudal limit of the NGC is at the most rostral extension of the hypoglossal nucleus.
A third system for subdividing the reticular formation was originated by Berman12. This nomenclature replaces the term "nucleus" for "field". Thus, the terms gigantocellular tegmental field (FTG), lateral tegmental field (FTL) and magnocellular tegmental field (FTM) were coined. This nomenclature has only been defined for the cat, and these terms do not correspond with either of the terminologies described above. The FTG comprises the medial portions of the RPC and the rostral tip of the NGC. Most of its area is not coincident with the NGC, with which it is often confused. The FTL runs all the way from the most caudal portions of the medulla to the RPC. The FTM refers to the ventral portion of the NGC.
As research has focused on portions of the reticular formation adjacent to the locus coeruleus nucleus, a more detailed anatomical nomenclature has come into use to describe this region139. The locus coeruleus proper is within the clearly defined central gray region. Immediately ventral and somewhat lateral to the caudal locus coeruleus is a region termed the nucleus subcoeruleus139. Ventral to the locus coeruleus and dorsomedial to the subcoeruleus is the locus coeruleus a . The term "peri-locus coeruleus a " has been coined to describe regions ventral to locus coeruleus a96.
Much of the confusion in the literature about REM sleep generating mechanisms has resulted from the inconsistent use of these various terminologies. Indeed the only reliable way to evaluate and integrate new findings is to carefully inspect the actual histology or refer to atlas coordinates133,12. In the present chapter I have brought together findings from a number of laboratories, and mapped them all on the Berman12 cat atlas plates for comparison (Figs. 2, 5, 7, 9, 11, 12 and 13)
C Lesion studies
In an attempt to further localize the neurons generating REM sleep, a number of investigators have damaged portions of the pons and caudal midbrain within the region defined as critical by transection studies. The first comprehensive study of this problem16 found that electrolytic lesions that destroyed the bulk of the RPO, permanently eliminated REM sleep. Further work has confirmed this finding, while providing a greater localization of the critical neurons. Following the development of techniques that permitted the localization of catecholamine neurons, much research interest was focused on the noradrenergic locus coeruleus, dorsal to the region identified by Carli and Zanchetti. While Carli and Zanchetti had concluded that locus coeruleus lesions did not block REM sleep, a study by Jouvet and Delorme50 found that lesions of this structure did prevent REM sleep. This was disputed in further investigations45,79. Depletion of norepinephrine and relatively selective destruction of norepinephrine neurons with 6 hydroxydopamine also did not prevent REM sleep59. Thus, the consensus of subsequent studies, in agreement with the original findings of Carli and Zanchetti, is that the norepinephrine cells of the locus coeruleus are not critical for REM sleep.
There has been some uncertainty regarding the laterality of the region critical for REM sleep. Carli and Zanchetti's16 lesions that blocked REM sleep included both medial and lateral RPO. One study suggested that the medial RF (i.e. within 2 mm of the midline in the cat) was critical44. However, further studies by the same group and by the Lyon group have concluded that the lateral rather that the medial regions of RPO are critical148,104,22 (Fig. 5). This small region might be critical either because it contains the somas of cells involved in REM sleep generation, or because major axonal pathways traverse this region. These possibilities can be distinguished by using the cytotoxin kainic acid to remove cells in this region. Using these techniques, it was found that extensive lesions of RPO disrupted REM sleep even with minimal axonal damage. The loss of REM sleep produced by the lesions was proportional to the number of cholinergic cells removed, but was not related to the number of noradrenergic cells lost148. Therefore, cholinergic and perhaps other yet to be identified cell types within this region are critical.
To summarize, the lateral region (L2-4 in the cat) of the RPO, ventral to the locus coeruleus, is the brain region most critical for REM sleep.
D Unit recording studies
Guided by the lesion data, researchers have recorded from the pons to observe the activity of neuronal elements that might be involved in REM sleep generation.
1. Medial Pontine Reticular cells. A number of studies have focused on the medial pontine region first implicated in REM sleep control by lesion studies. Cells in this area were found to have very high discharge rates in REM sleep but were silent or had relatively low discharge rates in nonREM sleep41,38. In cats that were well adapted to the head restraint employed in the recording sessions, medial pontine reticular cells have little activity in waking, discharging at rates comparable to those seen in nonREM sleep38. However, in freely moving cats, virtually all medial pontine reticular formation cells discharge in waking at rates comparable to mean REM sleep rates120,121. Discharge rates during REM sleep are positively correlated with active waking rates116. Studies in the awake, freely moving animal have revealed that most medial reticular cells discharge maximally in conjunction with a directionally specific movement of either the head, neck, eyes, limbs or facial musculature124,128. One would expect centrally commanded skeletal movements (whose expression in REM sleep is blocked by motoneuron hyperpolarization) to accompany the rapid eye movement bursts and muscle twitches of REM sleep. Thus, the apparent "selectivity" of discharge in medial reticular cells for REM sleep in cats adapted to head restraint can be seen as a consequence of the reduction in waking motor activity caused by the restraint, rather than indicating any selective role for these cells in REM sleep control115. This conclusion is confirmed by studies showing no disruption of REM sleep when cytotoxins were injected into this region. Even though this manipulation removed almost all cells in the medial pontine "FTG" region, REM sleep was not disturbed, appearing in normal amounts within 24 hours of the lesion104,22. However, these lesions did greatly reduce head movement138.
2. Lateral pontine and medial medullary reticular cells. A different picture has emerged in unit recordings from the lateral pontine and the medial medullary reticular formation. Many cells in these regions have discharge profiles similar to those seen in the medial pontine regions. However, these areas also contain a population of cells that discharges at a high rate throughout REM sleep and that has little or no activity in nonREM sleep82,131,23,110. In waking, these cells are generally silent, even during vigorous movement; however, some become active during head lowering and related postural changes that involve reductions of tone in a number of muscles (Fig. 6)131. The pontine "REM sleep-on" cells are distributed throughout the lateral region implicated by lesion studies in REM sleep control (Fig. 7)95,110. This distribution is significant, since it indicates that the critical lesion removes the somas of cells that are selectively active in REM sleep. Unit recording studies have found that whereas some of the REM sleep-on cells may release acetylcholine, many are not cholinergic110,23,98. However, it is likely that most respond to acetylcholine (i.e., are cholinoceptive). The medullary REM sleep-on cells are located in an area receiving projections from the pontine REM sleep-on region101. The neurochemistry of this medial medullary region will be discussed below. It is likely that many of the medial medullary and pontine REM sleep-on cells use amino acid transmitters, such as glycine, or GABA, as well as peptides.
3. Noradrenergic cells of the locus coeruleus complex and serotonergic cells of the raphe system. Cells in both of these regions have a similar discharge pattern during the sleep-waking cycle. During waking, discharge is very regular (Fig. 8), in contrast to the movement related burst-pause discharge pattern seen in most medial reticular neurons. During the initial stages of nonREM sleep, norepinephrine and serotonin containing cells slow slightly. During the "transition" to REM sleep (defined in the cat as the time at which PGO spikes begin appearing), discharge in both serotonergic and noradrenergic cells slows dramatically. During REM sleep, these cells have their lowest discharge rates and many are completely silent4,37,75,43. The slowing of discharge in these cells in nonREM sleep suggests that reduced release of serotonin or norepinephrine plays a permissive role in the initiation of nonREM sleep107,39. However, as mentioned above, lesion studies have shown that destruction of locus coeruleus neurons does not prevent REM sleep. The minimal discharge rate of these cells in REM sleep suggests some role in the "gating," of aspects of REM sleep. When we discuss PGO spike control below, we will review the very impressive evidence that serotonergic cells regulate PGO spikes. The role of these cells in facilitation of muscle tone will be discussed in the section on muscle atonia, below. The possible relation of the cessation of activity in locus coeruleus and raphe cells to the function of REM sleep will be discussed in the conclusion section.
The tonic discharge of aminergic cells in waking and even in some circumstances in vitro, suggests that active inhibition is involved in their cessation of activity in REM sleep. We have found that the release of GABA, a potent inhibitory amino acid, is increased in both the raphe and the locus coeruleus nuclei during REM sleep83,84. Microinjection of the GABA agonist, muscimol, into the serotonergic raphe nucleus increases REM sleep amounts, indicating that the cessation of discharge in aminergic neurons can facilitate REM sleep83. Electrophysiological studies show that locus coeruleus neurons are tonically inhibited by GABA during sleep31. Studies using the labeling of c-fos, a transcription regulator whose expression is often increased when cells discharge more rapidly, show increased labeling of some cholinergic cells in REM sleep113. There is also substantial labeling of non-cholinergic cells152. Further studies are needed to determine if the labeled cells contain GABA, glutamate, peptides or other transmitters.
Figure 9 presents maps summarizing current knowledge about the localization of REM sleep-off cells. Figure 10 summarizes current findings on the localization of cholinergic and catecholaminergic neurons within the pons.
We can describe the present state of our knowledge about the localization of mechanisms generating REM sleep as follows. Transection studies have determined that the caudal midbrain and the pons are sufficient to generate much of the phenomenology of REM sleep. Lesion studies have identified a region in the caudal midbrain and the pontine tegmentum, corresponding to lateral portions of the RPO and the region immediately ventral to the locus coeruleus, which is required for REM sleep. Unit recording studies have found a population of cells within this region that is selectively active in REM sleep. It does not appear that this small region is sufficient for REM sleep generation, in the same sense that the suprachiasmatic nucleus is sufficient to generate a circadian signal52. Nevertheless, it has become clear that much of the mechanism that drives this very complex behavioral state, is localized to the RPO nucleus of the pons and caudal midbrain.
It should be emphasized that in the intact animal many brain regions distant from the pons actively participate in the control of the REM sleep state. Obviously the phenomenology of REM sleep, such as muscle atonia, cortical desynchrony, rapid eye movements, alteration of sensory thresholds and autonomic changes, requires the recruitment of many brain systems. What is not always appreciated is the subtler role of non-brainstem systems in shaping the structure of REM sleep. While the decerebrate animal has the basic brainstem physiology of REM sleep, closer inspection reveals substantial differences between the "REM sleep" of the isolated brainstem and the REM sleep of the intact animal. PGO spikes and associated eye movements in the decerebrate animal come in regular alternating clusters, distinct from the irregular pattern seen in the intact animal48,40. Removal of the cerebellum, while not blocking REM sleep, alters the amplitude of PGO activity27,81. Lesions of the frontal cortex alter the amplitude and the pattern of spiking27. PGO spikes are tightly coupled to the eye movements of REM sleep, which in humans are correlated with dream imagery. Stimulation of the amygdala increases REM sleep in cats15. The amygdala is the forebrain area most intensely activated in normal, human REM sleep72. Therefore, one must not view the rest of the brain as merely a passive responder to a REM sleep state generated in the pons and caudal midbrain. Instead, present evidence suggests a dynamic interaction between the forebrain and the pons in molding the structure and timing of PGO spikes and the other "phasic" events of REM sleep and in all likelihood the dream imagery of REM sleep.
III DISSOCIATION OF REM SLEEP COMPONENTS
Experimental manipulations and pathological states allow us to further localize and analyze the mechanisms generating REM sleep. Lesion studies have demonstrated that PGO spikes, atonia and EEG desynchrony can be individually dissociated from the REM sleep state. Conversely, stimulation studies have demonstrated that each of these phenomena can be separately evoked.
A. Muscle atonia
1. REM sleep without atonia. Pontine lesions can produce the syndrome of REM sleep without atonia50,33. The critical lesions for producing this effect are much smaller than those required to block the REM sleep state (Fig. 11). Animals with these lesions have relatively normal nonREM sleep. During sleep they have periods of low voltage EEG, rapid eye movements, myosis and PGO spikes, as seen in normal REM sleep. However, muscle tone is present throughout these periods. Depending upon the exact placement of the lesion32, the animal's motor activity during this state will range from a slight raising of the head, to elaborate displays of exploratory and aggressive behaviors. The animal remains generally unresponsive to the environment and can be "awakened" by strong stimulation. With incomplete lesions, one may often see a partial syndrome of REM sleep with muscle tone interrupted by periods of atonia, gradually progressing to REM sleep periods with complete atonia over a period of several weeks. These findings leave little doubt that REM sleep without atonia is a variant of the normal REM sleep state.
The same syndrome can be produced by lesions of the medial medulla105. This finding supports the hypothesis that the output of the atonia systems in the pons is relayed in these regions of the medulla, on its way to the spinal motoneurons. A population of slowly conducting reticulospinal neurons in the nucleus magnocellularis is activated by stimulation that inhibits muscle tone. These neurons may form part of the "final common path" for brainstem atonia57.
2. Atonia without REM sleep. Muscle tone cannot only be suppressed by stimulation of the medial medulla, but also by stimulation of a number of pontine and midbrain areas. These include the ventrally located tegmental reticular fields of the pons, the dorsally located pedunculopontine nuclei93 and most rostrally, the retrorubral nuclei61 (Fig 14). The systems producing muscle tone suppression constitute a widespread brainstem network.
A loss of muscle tone can be evoked by injection of the cholinergic agonist, carbachol, or the cholinesterase inhibitor, physostigmine into the pons. The region of the RPO where injection produces the most immediate and complete loss of tone corresponds very well with the region which, when lesioned, results in REM sleep without atonia (Fig. 12). Depending on the exact injection site, carbachol may evoke only atonia of the skeletal muscles without the other signs of REM sleep. The animal may demonstrate perception of the outside environment by visually tracking stimuli with its extraocular muscles78,51. Larger injections or injections at other pontine sites can produce the full REM sleep pattern11,30,141,3,6,109,144,29. During spontaneous REM sleep, acetylcholine release is increased in the RPO region56,67. Muscarinic receptors are responsible for the cholinergic induction of REM sleep signs in the pons, with M2, M3 and other "non M1" receptors being particularly important145,42,5,99. Cholinergic cells in the pons also synthesize nitric oxide and microinjection of nitric oxide agonists into the pons increases REM sleep amounts, whereas antagonists suppress REM sleep21. The neurons critical for inducing REM sleep also respond to local microinjections of nerve growth factor153. Injection of glutamate at pontine sites where cholinergic agonists are effective also produces atonia60 and REM sleep86.
Thus a coordinated release of glutamate, acetylcholine, nitric oxide and other transmitters into the dorsal pons (nucleus reticularis pontis oralis) can trigger REM sleep or, depending on injection site and agonist dosage, just produce muscle atonia.
A second area in which chemical stimulation can elicit atonia is the medial medulla. Magoun and Rhines69 first reported that electrical stimulation of this area in the decerebrate cat produces an immediate loss of muscle tone. Glutamate stimulation of the nucleus magnocellularis, the rostral portion of the area identified by Magoun and Rhines, will trigger atonia60 (Figure 15). During REM sleep, glutamate release is increased in this region55. While glutamate stimulation was effective in the nucleus magnocellularis, acetylcholine stimulation was not. Acetylcholine release is not elevated in this region during REM sleep54. However, the situation is reversed in caudal portions of the medial medulla, corresponding to the nucleus paramedianus. Here acetylcholine injection produces a suppression of muscle tone, whereas glutamate does not. Acetylcholine release is elevated in the nucleus paramedianus during REM sleep54 but glutamate release is not55. Both the nucleus paramedianus and the cholinoceptive regions of the dorsolateral pons were found to receive a projection from cholinergic neurons in the pedunculopontine nucleus of the pons. This region and medullary cholinergic cells are the main sources of the acetylcholine released in the nucleus paramedianus and the dorsal pons in REM sleep76,111,112,108.
We investigated the nature of the receptor responsible for the suppression of muscle tone after glutamate microinjection. We found that nonNMDA glutamate receptors mediated muscle tone suppression in both the pons and the medial medulla62. When NMDA glutamate agonists were microinjected at sites where nonNMDA agonists produced atonia, motor excitation and locomotion were evoked. These opposite effects of the two types of glutamate receptors provide a mechanism for the "paradoxical" motor phenomena of REM sleep; the combination of rapid eye movements and phasic twitching, with the suppression of muscle tone. We hypothesize that release of glutamate in REM sleep activates both receptor types simultaneously producing the muscle tone suppression of REM sleep and at the same time the motor activation manifested as twitching and rapid eye movements61,62.
Whereas medullary stimulation in the decerebrate cat can produce loss of muscle tone, identical stimulation in the intact animal usually increases muscle tone134. Thus, the forebrain in the normal animal seems to contain mechanisms that produce a net inhibition of the brainstem atonia control mechanism. Furthermore, although animals with transections at the midbrain produce atonia when stimulated in the medulla, acute and chronic transection at the ponto-medullary junction greatly reduces atonia elicited by medullary stimulation124,57 i.e., if the pons is removed, the medulla does not readily produce muscle atonia. This indicates that the atonia system is not a simple "one way" descending pathway. In summary, transection level strongly modulates the effect of medullary activation on muscle atonia. In the intact animal there are mechanisms in the pons and the medulla which contribute to atonia, and mechanisms in the forebrain which block atonia.
In addition to transection level, blood pressure also modulates the tendency towards atonia. Medullary stimulation that produces atonia will produce muscle excitation when blood pressure is lowered as little as 10-20 mm Hg63.
3. Cataplexy. Cataplexy, a symptom of narcolepsy, is the sudden loss of muscle tone during active waking, usually triggered by strong emotions or physical activity. Physiologically it is similar to the "atonia without REM sleep" state mentioned above. Narcoleptics are aware of their environment and have a clear memory of all aspects of the cataplexy episode. It is reasonable to hypothesize that one or more of the mechanisms that have been identified as promoting or inhibiting atonia are malfunctioning in narcoleptics. The result is that a stimulus that elicits a strong emotional response produces arousal in an intact individual, produces cataplexy in narcoleptics. Cholinergic mechanisms seem to play an important role in cataplexy as they do in experimental atonia. Physostigmine, a cholinesterase inhibitor, increases cataplexy in narcoleptic dogs, while atropine blocks spontaneously occurring cataplexy in these animals8. Narcoleptic dogs have increased numbers of cholinergic receptors in the pons and the medial medulla. Receptors are increased at the same sites where stimulation is effective in producing atonia in the decerebrate preparation53. In summary, we can hypothesize that a hypersensitivity of cholinoceptive cells in the pons and/or the medial medulla is a factor contributing to cataplectic attacks.
In the narcoleptic dog, we can study the state of cataplexy at the neuronal level. Surprisingly, most brainstem neurons do not behave in a similar way in cataplexy and REM sleep. Whereas most brainstem neurons in the pontine and the medullary reticular formation increase activity during REM sleep, most decrease activity during cataplexy123. A specialized group of cells within the nucleus magnocellularis of the medial medulla has selective elevations of discharge rate in both REM sleep and cataplexy123. All REM sleep-off (presumably noradrenergic) cells in the locus coeruleus cease discharge immediately prior to and throughout cataplexy episodes149.
Both REM sleep and cataplexy are accompanied by a loss of muscle tone. However, awareness of the environment is preserved in cataplexy. The complete cessation of discharge in locus coeruleus (and medial reticular) cells during cataplexy suggests that these activity of these cells be more tightly linked to muscle tone than to awareness of the environment. Consistent with this behavioral correlation, is work showing that the release of norepinephrine from locus coeruleus cells facilitates motoneurons (reviewed in Wu et al.,149 ).
Individual noradrenergic locus coeruleus cells can have both descending projections to the spinal cord and ascending projections to widespread regions of the cortex. In addition to their relation to muscle activity, the tonic activity of locus coeruleus cells also modulates forebrain activity. Damage to the locus coeruleus increases metabolic rate106 and prevents cortical expression of the c-fos gene17. Stimulation of the locus coeruleus decreases glucose consumption and increases c-fos expression1,137. Therefore the reduction in locus coeruleus discharge in REM sleep is likely to cause an increase in cortical metabolism25 and a decrease in fos expression. One may speculate that locus coeruleus activity serves to integrate descending motor control with forebrain sensori-motor control systems.
Locus coeruleus cells receive a massive projection from the hypocretin (also known as orexin) system whose cell bodies are in the lateral hypothalamus. Recent work has shown that canine narcolepsy is linked to a mutation of the gene coding for the hypocretin 2 receptor, leading to diminished action of hypocretin in several brain areas (reviewed in 117a). The loss of function in the hypocretin system reduces activity in the locus coeruleus and possibly in other monoaminergic and cholinergic cell populations. The reduced tonic activation of locus coeruleus can explain cataplexy, linked to loss of locus coeruleus activity. Reduced tonic activation of cholinergic cell populations, which are involved in maintaining cortical arousal, is a possible explanation for the sleepiness that characterizes narcolepsy.
Thus, two elements of REM sleep, cessation of locus coeruleus discharge and increased activity of nucleus magnocellularis cells contribute to cataplexy. The cessation of discharge in medial reticular cells, seen in cataplexy but not in REM sleep, may also contribute to the reduced motor activity of cataplexy.
Cataplectic attacks are preceded by a marked increase in heart rate129. Systemic blood pressure is not affected during spontaneous attacks, although increases in blood pressure will trigger cataplectic attacks in narcoleptic dogs. In summary, changes in circulatory control appear to play a role in the triggering of cataplectic attacks, just as they do in determining the response to medullary stimulation. The means by which the brainstem systems responsible for cataplexy are triggered is not completely understood. However, it is likely that forebrain degeneration122 releases the brainstem circuits responsible for loss of muscle tone.
Figure 13 summarizes the current state of our knowledge about the anatomical relation of sites where carbachol produces atonia, the location of REM sleep-on and REM sleep-off cells and lesions blocking REM sleep.
B PGO spikes
PGO spikes accompany the eye movements and many of the other phasic motor and sensory events of REM sleep. They occur during a period of phasically enhanced excitability within REM sleep151. Similar potentials can be elicited in waking by intense auditory stimulation that elicit orienting150,102. Transection studies have localized the generator mechanisms responsible for this activity. Transections at the level of the abducens nucleus (Fig 1) allow PGO spikes to occur in a relatively normal pattern and amplitude distribution in the forebrain, rostral to the transection125. Transections and lesions a few millimeters rostral to this level completely block forebrain PGO spikes65. Lesions of the peribrachial region (the region around the superior cerebellar peduncle or "brachium conjunctivum") can produce "REM sleep" without PGO waves or rapid eye movements (Fig. 16). Lesions that spare the peribrachial regions but damage more ventro-medial areas of the pons, produce REM sleep without atonia, but with large numbers of PGO waves and rapid eye movements (Fig.16)114.
PGO waves can be generated in the absence of other REM sleep phenomena by cholinergic stimulation of the pons7. The best sites for this stimulation are in the peribrachial region. Unit activity studies have identified the cellular elements involved in PGO generation. They are localized to the reticular regions around the superior cerebellar peduncle (peribrachial region) and the area below the locus coeruleus20,21,94,74,93,97,135,136. Many of these cells are cholinergic. They have a characteristic short burst of activity before each ipsilateral PGO wave and project to the thalamus.
Serotonin inhibits PGO waves. Serotonin depletion with PCPA produces continuous PGO activity in all behavioral states18. Lesions of the serotonergic dorsal raphe or small cuts lateral to the raphe also produce a release of PGO activity132. In vitro studies have shown that serotonin blocks the burst firing mode of PGO cells by hyperpolarizing them68. In the transition from nonREM to REM sleep the cessation of activity in serotonergic cells allows the PGO cells to begin discharging in bursts, generating PGO waves.
In summary, PGO waves are generated by cholinergic neurons in the peribrachial region which project rostrally. They are inhibited by serotonergic neurons of the raphe system.
C. EEG desynchrony in the neocortex and hippocampal theta.
A single mechanism appears to be responsible for generating the EEG voltage reduction seen in REM sleep and waking (relative to the high voltage EEG of nonREM sleep). Likewise, hippocampal theta is indistinguishable in waking and REM sleep. Both hippocampal theta and EEG desynchrony can exist in the forebrain disconnected from the pons85. This is in contrast to PGO waves, which do not occur in the forebrain after disconnection of the pons.
Hippocampal theta is normally continuous during REM sleep. During periods of PGO spike bursts and associated phasic activity, the theta frequency increases. Rats which are deprived of REM sleep can have REM sleep episodes even when heavily atropinized. Under these conditions the tonic theta of REM sleep is abolished142. However, the theta that accompanies bursts of phasic motor activity is still present. In a similar way, neocortical desynchrony, which is normally present throughout most of REM sleep, is absent in the atropinized rat and cat34. Instead brief periods of EEG desynchrony are present only in conjunction with bursts of phasic activity. Therefore, it has been concluded that both the tonic EEG desynchrony and the tonic hippocampal theta which accompany REM sleep are generated by cholinergic mechanisms. The phasic EEG desynchrony and hippocampal theta accompanying movement and phasic activity in REM sleep has been attributed to a non-cholinergic mechanism142. Theta can be triggered by electrical stimulation of the brainstem reticular formation. The best brainstem site for elicitation of hippocampal theta is in the pontine RF at the level of the RPO146, i.e. at the same level at which cholinergic stimulation is most effective in triggering REM sleep and lesions are most effective in blocking REM sleep. The control of EEG changes in sleep and waking is discussed in greater detail in two other chapters in this section (see chapters by Jones and Steriade).
III A SYNTHESIS OF FINDINGS ON REM SLEEP CONTROL MECHANISMS.
A Are there executive neurons?
We have seen that there are brainstem neurons that discharge during each ipsilateral PGO spike. There are also cells which are tonically active in REM sleep, but which are silent at other times. These neurons may be related to the atonia or other tonic aspects of REM sleep. Still other cell groups are tonically active during both waking and REM sleep and may be related to EEG control. But is there a cell group that serves to coordinate all of the phenomenology of REM sleep, which triggers other cell groups producing each of the defining signs of this state? It is unclear whether such an "executive" cell group exists. While this coordinating property may or may not be manifest within the activity of individual neurons, it is clear that an "executive system" of neurons for the triggering and maintenance of REM sleep resides in the lateral pons and adjacent midbrain. This system may be comprised of an interacting population of cells, each of which is primarily tied to one or more of the physiological aspects of REM sleep, but none of which is, in and of itself, sufficient to trigger the state. The anatomical discreteness of such cell populations is evident in the dissociation of REM sleep signs that can be revealed by stimulation and lesion studies. Conversely, the synaptic linkage of these populations is evident in the fact that such dissociations never occur in the undisturbed animal and only rarely occur in disease, as in narcolepsy.
B What is the function of REM sleep?
Whether the triggering of REM sleep is manifest in a group of "executive neurons" or in an "executive system", these cells must be fulfilling some biological function. While our knowledge of the mechanism has increased enormously, the question of function remains in the realm of speculation, in stark contrast to that of other behavioral activities occupying substantially smaller amounts of time. While this difficult question remains unanswered, I offer the following speculation.
The changes in brain unit activity during REM sleep should provide a clue to its function. We have described two kinds of unit activity unique to REM sleep, REM sleep-on and REM sleep-off cells. REM sleep-off cells in the raphe system appear to have an important role in the gating of PGO spikes. REM sleep-off cells in the locus coeruleus and raphe may have a role in the maintenance of muscle tone in waking. Histaminergic cells of the posterior hypothalamus are also "off" in REM sleep (and in nonREM sleep). Whereas monoaminergic cells are important in shaping the phenomena of REM sleep, the lesions that eliminate REM sleep are centered on the locations of the REM sleep-on cells, not the REM sleep-off cells. The number of REM-off cells appears to be much greater than the number of REM-on cells. The cessation of activity in the aminergic cells may not only be important in the modulation of REM sleep components, but may also be significant for the function of REM sleep. Specifically, I propose that this periodic cessation of discharge prevents desensitization of aminergic receptors, which are continuously activated in waking127. Desensitization would reduce the effectiveness of these transmitters. Experimental support for this concept comes from studies of pontine activity during REM sleep deprivation. Deprivation greatly reduces the amplitude of a noradrenergic mediated inhibition normally seen in the pons after auditory stimulation70. REM sleep deprivation produces a slowing of presumably noradrenergic cells of the dorsolateral pons71, though serotonergic REM sleep-off cells may speed up under similar conditions28. This slowing of noradrenergic cells may be responsible for some of the symptoms of sleep deprivation, since norepinephrine release has been shown to increase the "signal to noise ratio" of information processing in a number of brain regions127. Receptor assays directed at testing this hypothesis have produced both supportive and contradictory findings127,140.
The other major neuronal phenomenon of REM sleep is the burst discharge and elevated discharge rate in a majority of brainstem and forebrain systems124, 41,24,36. This burst discharge, synchronized in adjacent cells126, underlies the rapid eye movements and twitches that characterize this state. It is dramatically different from the slow regular discharge pattern seen in these same cells during nonREM sleep. Does this transmitter release desensitize certain groups of receptors? Does it function to potentiate neuronal circuits? Does it have a role in maintaining intracellular homeostasis? These questions remain unanswered.
FIGURE LEGENDS
1. TOP-Polygraph tracings of states seen in the intact cat. BOTTOM-States seen in the forebrain four days after transection at the pontomedullary junction. EEG, sensorimotor electroencephalogram; EOG, electrooculogram; OLF, olfactory bulb; LGN, lateral geniculate nucleus; HIPP, hippocampus; EMG, dorsal neck electromyogram. (Reprinted from Siegel et al.125 with permission)
2. Outline of sagittal section of the brainstem of the cat drawn from level L=1.6 of the Berman12 atlas indicating the level of key brainstem transection studies. RN, red nucleus; LC, locus coeruleus; 6, abducens nucleus; 7, genu of the facial nerve; IO, inferior olive. H and A-P scales drawn from the atlas. (Reprinted from Siegel118 with permission)
3. States seen in the chronic medullary cat. Note absence of periods of atonia. EKG, electrocardiogram; RESP, thoracic strain gauge. Calibration 50 m V. (Reprinted from Siegel et al.,119 with permission)
4. Midbrain unit, EEG, EOG and LGN activity rostral to chronic transections at the ponto-medullary junction. In upper portion of figure, unit channel displays the output of an integrating digital counter resetting at one-second intervals. In lower portion one pulse is produced for each spike by a window discriminator. (Reprinted from Siegel118 with permission)
5. Lesions producing a complete suppression of REM sleep drawn from Sastre et al.103 and Friedman and Jones26. Note the relatively close agreement on the location of the critical region.
6. Activity of medullary "REM sleep-on" cell. Note tonic activity during REM sleep. In waking, activity is generally absent even during vigorous movement. However, some activity is seen during movements involving head lowering and postural relaxation. (Reprinted from Siegel et al.127 with permission)
7. Location of pontine "REM sleep-on" cells. Redrawn from Sakai95 and Shiromani et al.110.
8. Activity of a "REM sleep-off" cell recorded in the locus coeruleus. (Reprinted from Siegel117, with permission)
9. Location of "REM sleep-off" cells. Redrawn from Sakai95.
10. Closed circles indicate the location of CHAT containing (presumably cholinergic) neurons and open circles the location of tyrosine hydroxylase containing (presumably catecholaminergic) cells in the pons. (From Jones and Beaudet46 with permission)
11. Location of lesions producing REM sleep without atonia. Redrawn from Sakai95.
12. Location of pontine sites producing atonia at the shortest latency after carbachol microinjection. Redrawn from Katayama et al.51.
13. Anatomical relation of REM sleep-on and off cells, carbachol atonia sites, lesions blocking atonia but not preventing REM sleep, and lesions completely blocking REM sleep.
14. Location of pontine and mesencephalic regions whose electrical stimulation produces suppression of muscle tone. Large dots indicate points at which stimulation produces more than 70% inhibition of muscle tone. Widespread pontine and midbrain regions can produce suppression of muscle tone. 4: trochlear nucleus; AQ: aqueduct; IC: inferior colliculus; FTP: paralemniscal tegmental field; MLB: medial longitudinal bundle; NIP: interpeduncular nucleus; P: pyramidal tract; PAG: periaqueductal gray; PG: pontine gray; PPN: pedunculopontine tegmental nucleus; RD: red nucleus; RR: retrorubral nucleus; SC: superior colliculus; SN: substantia nigra; TRC: TRN (tegmental reticular nucleus), central division; TRP: TRN, peri-central division.
15. Schematic map of ponto-medullary inhibitory areas. Electrical stimulation produced atonia at all the points mapped. All electrically defined inhibitory sites were microinjected with glutamate or cholinergic agonists. Filled symbols represent points at which microinjections decreased muscle tone [to less than 30% of baseline values or to complete atonia]. Open circles indicate points at which injections increased or produced no change in baseline values. Glutamate injections are shown in the top, and acetylcholine [Ach] and carbachol [Carb] injections in the bottom. In the bottom, circles and triangles represent Ach and Carb injections, respectively. 4V, fourth ventricle; 5ME, mesencephalic trigeminal tract; 6, abducens nucleus; 7G, genu of the facial nerve; IO, inferior olive nucleus; LC, locus coeruleus nucleus; NGC, nucleus gigantocellularis; NMC, nucleus magnocellularis; NPM, nucleus paramedianus; PG, pontine gray; PT, pyramid tract; SO, superior olive nucleus; T, nucleus of the trapezoid body; TB, trapezoid body.
16. Twenty second polygraph tracings of REM sleep before and after lesions,
together with a coronal section through the center of the pontine lesions.
EEG voltage reduction of REM sleep (recorded from motor cortex) was present
after both lesions. Top: radio frequency lesions of the pedunculopontine
region diminished PGO spikes and eye movement bursts during REM sleep.
Bottom: lesions in the region ventral to the locus coeruleus produced REM
sleep without atonia without any diminution of PGO spike or rapid eye movement
frequency (From Shouse and Siegel114 with permission).
REFERENCES
1.