Neural control of REM sleep and motor atonia: current perspectives

Abstract

Purpose of Review

Since the formal discovery of rapid eye movement (REM) sleep in 1953, we have gained a vast amount of knowledge regarding the specific populations of neurons, their connections, and synaptic mechanisms regulating this stage of sleep and its accompanying features. This article discusses REM sleep circuits and their dysfunction, specifically emphasizing recent studies using conditional genetic tools.

Recent Findings

Sublaterodorsal nucleus (SLD) in the dorsolateral pons, especially the glutamatergic subpopulation in this region (SLD^Glut), are shown to be indispensable for REM sleep. These neurons appear to be single REM generators in the rodent brain and may initiate and orchestrate all REM sleep events, including cortical and hippocampal activation and muscle atonia through distinct pathways. Several cell groups in the brainstem and hypothalamus may influence SLD^Glut neuron activity, thereby modulating REM sleep timing, amounts, and architecture. Damage to SLD^Glut neurons or their projections involved in muscle atonia leads to REM behavior disorder, whereas the abnormal activation of this pathway during wakefulness may underlie cataplexy in narcolepsy.

Summary

Despite some opposing views, it has become evident that SLD^Glut neurons are the sole generators of REM sleep and its associated characteristics. Further research should prioritize a deeper understanding of their cellular, synaptic, and molecular properties, as well as the mechanisms that trigger their activation during cataplexy or make them susceptible in RBD.

1. Introduction

Sleep in mammals comprises at least two distinct states: slow wave sleep or non-rapid eye movement sleep (NREM sleep) and rapid eye movement sleep (REM sleep). These sleep stages are accompanied by specific alterations in brain activity and, consequently, in physiology and behavior.^1–7 Electroencephalogram (EEG) and electromyogram (EMG), which measure brain and muscle activity, are considered excellent biomarkers to distinguish wake and sleep stages. During wakefulness, EEG mainly comprises low-amplitude high-frequency waves, often called “desynchronized” EEG, reflecting higher cortical activity.^1–6,8,9 EMG activity is high and variable, reflecting higher muscle tone and motor movements in wakefulness. EEG activity slows down upon entry to NREM sleep, and a “synchronized” EEG consisting of high-voltage slow waves with predominant delta activity (0.5 – 4 Hz; also referred to as slow wave activity) ensues.^1–9 The magnitude of delta activity is considered a measure of sleep depth and positively correlates with the homeostatic need to NREM sleep. EMG activity decreases significantly during this stage, but muscle tone is preserved. In addition to EEG slowing and muscle relaxation, NREM sleep is accompanied by decreased brain and body temperatures and reduced sympathetic activity.

On the other hand, REM sleep is characterized by desynchronized EEG similar to wake but comprises regularly occurring theta waves with a dominant 5–9 Hz theta frequency.^3,6,9–12 These theta rhythms reflect the higher activity of the hippocampus, suggesting a role for REM sleep in hippocampal-dependent learning and memory processes. In addition to activated EEG, REM sleep is associated with characteristic rapid, jerky, and conjugate eye movements^{3,6,7,10–13}. These eye movements correlate with the ability to recall dreams, which preferentially occur in REM sleep. Moreover, the pattern of eye movements appears to correlate with the emotional content of the dream. While isolated slow movements are associated with neutral or positive emotions, phasic burst-type movements are associated with negative emotions in the dream.^14,15 Another key feature of REM sleep is the complete loss of muscle tone in almost all skeletal muscles, except for the diaphragm, respiratory muscles, extraocular muscles, and inner ear muscles.^{3,6,10–12,16,17} Muscle atonia is a protective mechanism that prevents the dream enactment (acting out the dreams). Against the backdrop of atonia, intermittent phasic twitches can be observed in many muscles, especially cranial muscles. Based on muscle twitches and eye movements (or other signs), REM sleep is often divided into two microstates - tonic REM sleep comprising the periods of complete atonia with no or fewer eye movements, and phasic REM sleep comprising the periods with muscle twitches and dense eye movements.^18,19 Finally, REM sleep is accompanied by autonomic instability with higher heart rate and respiratory rates, loss of thermoregulation, and penile or clitoral erection, but these features have been understudied.^3,16,17

This review will focus on the neural circuits regulating REM sleep and its two primary features, cortical/hippocampal activation and muscle atonia, with particular emphasis on recent research employing conditional genetic approaches to manipulate specific chemically-defined neural populations.

2. Brainstem structures controlling REM sleep

2.1. A single cell group in the dorsolateral pons generates REM sleep

Investigation of neural structures responsible for REM sleep began in Jouvet’s lab in France. They first showed that cats with transections in front of the pons could still exhibit periods of rapid eye movements with concurrent muscle atonia, indicating that the brainstem is sufficient to generate REM sleep.²⁰ Transections at various levels of the brainstem then revealed that selective signs of REM sleep could be observed consistently on the side connected to the pons, indicating that the pons is indispensable for REM sleep to occur.^21–27 Electrophysiological and pharmacological investigations helped to pinpoint the specific pontine region generating REM sleep. For example, Sakai and colleagues injected small doses of carbachol into roughly 120 sites in the brainstem of cats and demonstrated that the subcoeruleus (SC) region, located ventral to the locus coeruleus in the dorsal pontine tegmentum, was the only area capable of inducing REM sleep with latencies of less than five minutes. Consistently, this region was found to contain a high density of “REM-on” neurons that fired selectively before and during REM sleep.^28,29 Finally, electrolytic or neurotoxic lesions of the SC in cats resulted in profound reductions in REM sleep, convincingly demonstrating that SC is the critical site for REM sleep generation.^30–34 In addition to REM sleep reduction, the SC-lesioned cats displayed incomplete muscle atonia, leading to overt motor activities ranging from simple myoclonic jerks to complex behaviors such as exploration, grooming, chasing, attacking, and flight while in REM sleep.^35–38 As vivid dreams are more common in REM sleep, these involuntary motor activities are viewed as dream-enactment behavior. These findings assumed greater significance several years later when similar dream enactment behavior was observed in humans, leading to the discovery of REM-behavior disorder (RBD).³⁹

In contrast to cats, SC lesions in rats produced only minor effects on REM sleep or muscle atonia.⁴⁰ However, it was later discovered that the area responsible for generating REM sleep in rats and mice might be situated rostral to the subcoeruleus region, ventral to the laterodorsal tegmental nucleus (LDT), and was named the sublaterodorsal nucleus (SLD).^41,42 Lesions in the SLD induced robust reductions in REM sleep (up to 50% in rats) and deficits in muscle atonia characterized by high EMG tone, phasic muscle twitches, and dream-enactment behavior similar to SC-lesioned cats.⁴² These data confirmed that the SLD is the SC-homologue in rats and mice, necessary for generating REM sleep and motor atonia.

REM sleep was still present, albeit in reduced amounts, after SLD or SC lesions, leading to the question of whether other structures can independently generate this state. The forebrain structures have been ruled out even by early transection studies in generating REM sleep,^20,22 but a role for medullary structures has been constantly debated. Several medullary sites, including supraolivary medulla⁴³, ventral medulla,⁴⁴ and dorsomedial medulla,^45,46 were found to contain REM-active neurons, and inhibition or ablation of those neurons reduced REM sleep amounts. However, two recent studies convincingly demonstrated that REM sleep cannot occur in the absence of SLD.^47,48 These studies employed a diphtheria-toxin-based conditional genetic method, producing more complete neuronal lesions than those produced by neurotoxins in the previous studies. These diphtheria toxin-mediated SLD lesions led to a complete absence of REM sleep (vs. 40–50% reduction in REM sleep after ibotenic acid lesions), which persisted for at least nine months of the follow-up period.⁴⁸ While these lesioned rats unsuccessfully attempted to enter REM sleep, full-fledged REM sleep never occurred, and consequently, the dream-enactment behavior was undetectable. Thus, the SLD appears to be the sole generator of REM sleep, but the above medullary sites may regulate this state by directly or indirectly modulating the SLD activity.

2.2. Glutamatergic neurons in the SLD are necessary for REM generation

The SLD contains a mixed population of neurons where glutamatergic (Glut) and GABAergic (GABA) neurons are the majority, and both these subsets appear to be active during REM sleep. For example, Lu and colleagues acutely exposed albino rats to dark during their light period, which substantially increased REM sleep along with an increase in cFos in the SLD GABAergic (SLD^GABA) neurons.⁴² In contrast, Luppi and colleagues induced selective REM sleep deprivation in rats and found that both SLD^GABA and SLD^Glut neurons exhibited higher cFos after REM rebound.^49,50 Recent studies that measured the neural activity using genetically encoded calcium indicators confirmed that a large proportion of Glut and a smaller proportion of GABA neurons in a broader region that includes the SLD are activated during REM sleep.⁵¹ Both SLD^Glut and SLD^GABA neurons project to the REM-suppressing region in the midbrain (see below), but SLD^GABA neurons innervate more densely.⁴² Similarly, the REM-active SLD^Glut neurons (based on cFos) project to the ventromedial medulla (VMM) and the spinal cord (which mediate muscle atonia; see below) but not to the intralaminar thalamus, originally thought to be necessary for cortical activation during REM sleep.⁵² Thus, the SLD^GABA and SLD^Glut neurons were thought to be involved, respectively, in REM state generation and muscle atonia.^42,52

Recent studies using conditional genetic approaches manipulated the neural activity or ablated the specific subpopulations of the SLD to determine their precise role in REM sleep regulation. Selective activation of SLD^Glut neurons using optogenetic or chemogenetic methods reliably increased REM sleep transitions and total amounts, whereas inhibitions reduced REM sleep.^48,53 Specific elimination of Glut neurotransmission from SLD neurons by blocking the synthesis of functional vesicular glutamate transporters (Vglut2) either via a Cre recombinase (Cre)-mediated conditional knockout in mice or by using short hairpin RNA (shRNA) in rats strongly reduced REM sleep levels by 40–50%. These animals also displayed high-amplitude muscle twitches and overt motor behaviors during REM sleep, similar to that observed after incomplete SLD lesions.^52,54 However, the most striking finding is that selective ablation of SLD^Glut neurons completely abolished REM sleep in mice (Fig.1), similar to DT-mediated SLD lesions in rats.⁴⁸ In contrast, neither the loss of GABA neurotransmission by blocking the synthesis of functional vesicular GABA transporter (Vgat) from the SLD neurons nor the ablation of SLD^GABA neurons altered REM sleep amounts or motor atonia (Fig. 1).^48,54 These data demonstrated that SLD^Glut neurons may be the primary or presumably the only generators of REM sleep state and its accompanying features. Thus, any alterations in REM sleep amounts, architecture, or timing in health or disease conditions are likely caused by changes in SLD^Glut neuronal morphology or activity dynamics.

Figure 1. Loss of REM sleep after SLD lesions.

Wen et al⁴⁸., injected a Cre-dependent adeno-associated viral vector expressing diphtheria-toxin subunit A (DTA) into the SLD of transgenic mice expressing Cre recombinase and Green florescent protein (GFP) in glutamatergic (Vglut2-GFP mice; A) or GABA neurons (Vgat-GFP mice; B). The DTA gets expressed in Cre/GFP-positive (Green) neurons and deletes them, but expresses a red florescent protein, mcherry, in Cre-negative neurons in the injection site (red cells in A and B). Note that there are no green cells within the injection site, indicating the complete loss of Glut (A) or GABA neurons (B). REM sleep was absent in mice lacking SLD^Glut neurons (C), but their levels remained unaltered in mice lacking SLD^GABA neurons (D). Modified with permission from Wen et al.⁴⁸

3. Potential mechanisms of REM sleep control

3.1. SLD neurons are inhibited by a REM-off region in the midbrain

In the early 1970s, Petitjean and colleagues discovered that bilateral destruction of a small region in the pontomesencephalic junction overlapping the periaqueductal gray (PAG) and the adjacent reticular formation in the isthmus induced a robust increase in REM sleep⁵⁵, suggesting a REM-inhibitory region may exist at this level of the neuraxis. These electrolytic lesions could have also damaged the dorsal noradrenergic bundle and other tracts passing through this region. However, the application of a GABAergic agonist, muscimol, induced a robust increase in REM sleep, confirming the presence of REM-suppressing neurons at the pontomesencephalic junction.⁵⁶ In this study, muscimol induced REM sleep within short latency and produced a maximal increase in REM amounts (in some cases, REM sleep occurred uninterrupted for up to 4h) when the injections were placed in the ventrolateral part of PAG (vlPAG) and the adjoining lateral pontine tegmentum (LPT).⁵⁶ This same region was also identified as a ‘REM-off site’ by tracing the convergence of descending pathways from the preoptic area (POA) and lateral hypothalamus (LH), previously known to promote and inhibit REM sleep, respectively.⁴² Finally, focal neurotoxic lesions of the vlPAG/LPT in rats and mice substantially increased spontaneous REM sleep amounts, establishing the REM-suppressing role of this region.^42,57

The increase in REM sleep following muscimol into the vlPAG/LPT persisted even in cats with mesencephalic transections that ablate the ascending projections,⁵⁶ suggesting that descending inhibitory inputs from this region to the SLD are responsible for the REM-suppressing effects. Consistent with this idea, the vlPAG/LPT projects heavily to the SLD; indeed, the vlPAG/LPT contributed the highest percentage of inputs, especially to the SLD^Glut neurons, in the entire brain.^42,58,59 Glut and GABA neurons in the vlPAG/LPT monosynaptically innervate the SLD^Glut neurons.^42,58–60 However, GABA neurons exhibited a ‘REM-off’ activity profile. These neurons expressed cFos during selective REM sleep deprivation^42,49. In addition, a subpopulation of GABA neurons displayed a ‘REM-off’ activity pattern, i.e., their activity starts decreasing a few seconds before the REM sleep onset, remains low throughout the REM sleep episode, and abruptly increases when the mice wake up at the end of REM sleep.^59,61 Consistent with this activity pattern, brief optogenetic stimulations of the vlPAG/LPT^GABA neurons strongly suppressed the NREM→REM sleep transitions and reduced REM bout duration and chemoactivation suppressed REM sleep for several hours.^44,59–62 Conversely, photoinhibition of vlPAG/LPT^GABA neurons, specifically during NREM sleep, rapidly initiated REM sleep. Chemoinhibition or ablation of these neurons increased REM sleep mainly by increasing the number of REM episodes, suggesting that reduced activity of vPAG/LPT^GABA neurons primarily contributes to REM sleep initiation.⁶¹ Stimulation of vlPAG/LPT^GABA terminals inhibited the SLD^glut neurons (in vitro), greatly reduced NREM→REM transitions, and abruptly terminated the ongoing REM sleep episodes.⁵⁹ In contrast, optogenetic inhibition of vlPAG/LPT^GABA terminals in the SLD increased REM sleep transitions, but such inhibitions were ineffective in mice lacking glutamatergic neurotransmission in the SLD⁵⁹. These data strongly support the long-standing hypothesis that the vlPAG/LPT^GABA neurons may generally prevent the REM occurrence by inhibiting the SLD^Glut neurons, and the withdrawal of this inhibition allows resumption of SLD^Glut activity and thereby, REM sleep (Figure. 2).

Figure 2: Circuit mechanisms orchestrating REM sleep.

In mammals, REM sleep is generated by single cell group in the dorsal pons, the SLD^Glut neurons. These REM-on neurons are inhibited during wake and NREM sleep by a subset of GABA neurons in the vlPAG/LPT that are ‘REM-off’ (Red). REM sleep is triggered when the REM-off vlPAG/LPT neuron activity reduces and disinhibits the SLD^Glut neurons. These REM-on and REM-off neurons are under the control of several forebrain and brainstem cell groups contributing to the homeostatic and allostatic control of REM sleep. Most important ones are i) another subpopulation of GABA neurons within the vlPAG that are REM-on (pink), ii) Orx neurons (dark blue), iii) MCH neurons (Green) and iv) brainstem cholinergic and monoaminergic neurons (not shown).

SLD - Sublaterodorsal nucleus; vlPAG - Ventrolateral periaquductal gray; LPT- lateral pontine tegmentum; LH- lateral hypothalamus

Dark blue – Orexin neurons; Green - MCH neurons; Brown: GABA neurons

However, the vlPAG/LPT^GABA neurons also project to several other regions, including the LDT cholinergic neurons and locus coeruleus (LC) and lateral hypothalamus (LH), which are known to participate in REM control.^59,61 Interestingly, REM suppression was also observed after activating the vlPAG/LPT^GABA terminals in the LH.⁵⁹ It is unclear if the same neurons projected to both the SLD and the LH and these effects were caused by backpropagation of action potentials activating the soma. At this time, it is safe to conclude that the REM sleep-suppressing effects of vlPAG/LPT^GABA neurons may primarily be due to inhibition SLD^Glut neurons, but additional pathways may reinforce these effects.

In addition to these long-range pathways, local microcircuits within the vlPAG/LPT may also contribute to REM control. Electrophysiological and calcium imaging studies found that a subpopulation of vlPAG/LPT^GABA neurons increases their activity during REM sleep.^{49,59,61,63,64} Indeed, this REM-on population may be larger than the REM-off GABA population in this region⁶⁴. Inactivation or lesions that included both these populations (and other cell types in case of neurotoxic lesions) increased REM sleep, but the REM bouts were fragmented with more frequent and short REM bouts.^42,54,61,65 Grace and colleagues demonstrated that several REM episodes after vlPAG/LPT inhibition may belong to an intermediate state consisting of both NREM and REM features.⁶⁵ Using a computer simulation model, they suggested that such a state could have been caused by combined inhibition of REM-on and REM-off populations rather than the inhibition of REM-off alone.⁶⁵ Consistent with this, vlPAG/LPT^GABA neurons were found to receive local inhibitory inputs and are inhibited by their activation.⁶¹ Thus, REM-on and REM-off GABA neurons within vlPAG/LPT may inhibit each other, and this mutual inhibition may play a vital role in REM transitions and in stabilizing the REM sleep bouts. Nevertheless, REM-off GABA neurons may be the primary output pathway, and their actions on SLD^Glut neurons underlies the gating control of REM sleep (Fig.2).

3.2. Flip-flop model of REM sleep control with an amendment:

Based on cFos and lesion data from rats, Lu and colleagues proposed a ‘flip-flop’ switch model to explain REM sleep generation and transitions.⁴² Per this model, reciprocal inhibition between the vlPAG/LPT^GABA and the SLD^GABA neurons determine the onset and termination of REM sleep. The SLD^GABA neurons inhibit the vlPAG/LPT^GABA neurons and disinhibit their own firing in the due process, leading to stable REM sleep and vice versa.⁴² GABA Importantly, reduced activity of either side leads to less stable REM episodes and more frequent transitions into and out of REM sleep, which is consistent with the REM sleep phenotype observed after lesion or inactivation of the vlPAG/LPT or SLD (incomplete) lesions in rats or mice.^{42,48,54,57,65} However, as described above, recent studies have ruled out a major role for SLD^GABA neurons but established SLD^Glut neurons as REM generators.^48,54 Thus, the flip-flop model of REM sleep needs to be revised to incorporate these recent data. As indicated in the previous section, REM-off vlPAG/LPT^GABA neurons during wake and NREM sleep tonically inhibit REM-on SLD^Glut neurons preventing the occurrence of REM sleep, whereas the suppressed activity of vlPAG/LPT neurons disinhibits the SLD^Glut neurons, initiating REM sleep.^16,66,67 Moreover, NREM-to-REM transitions are not abrupt as NREM→Wake transitions; they are slow and take several seconds to minutes. Also, not all but only about one-third of NREM sleep episodes transition into REM sleep in rodents, and REM sleep may follow an ultradian pattern. Thus, another mechanism must operate during periods between two consecutive REM sleep (inter-REM intervals), which tracks REM pressure and dictate these transitions. Interestingly, REM-off vlPAG/LPT^GABA neurons also decrease slowly over minutes during inter-rem intervals and reset during REM sleep, and these changes correlate with REM sleep pressure.⁶¹ Thus, rapid vs. slow dynamics of REM-off vlPAG^GABA neuron activity may underlie rapid REM transitions vs. slower homeostatic and ultradian control of REM sleep.^42,61 In contrast to the flip-flop model, this new model indicates a unidirectional control by the vlPAG/LPT^GABA on SLD neurons but not vice versa. In other words, inhibition of vlPAG/LPT^GABA neurons by SLD^GABA neurons may not be a key determinant of REM sleep genesis. While this model indicates that REM sleep episodes begin with the suppression of activity in the REM-off vlPAG/LPT^GABA neurons and the consequent disinhibition of the SLD^Glut neurons, periods of ‘REM sleep-like activity was observed in the isolated pons (i.e., after transections at both sides of the pons)⁶⁸ suggesting an intrinsic pacemaker-like activity of SLD^Glut neurons may generate REM sleep. Taken together with SLD^Glut lesions completely abolishing REM sleep, it appears that the SLD^Glut neurons are sufficient for REM sleep generation, but the vlPAG/LPT and the various other cell groups may merely ‘fine-tune’ the timing and architecture of this state.

Termination of REM sleep episodes, on the other hand, could be due to the resumption of activity in the wake-promoting cell groups, at least in laboratory rodents, as most, if not all, of their REM sleep bouts culminate in wakefulness. The orexin (Orx)-producing neurons in the LH and the noradrenergic (NE) neurons in the locus coeruleus appear most important for this control. For example, activity in these neurons is lowest during REM sleep but increases a few seconds before wake-onset.^69,70 Optogenetic activation of LC^NE or Orx neurons abruptly terminate ongoing REM sleep episodes and elicits arousal.^71,72 In contrast, most other wake-promoting neurons in the lateral hypothalamus (LH), pedunculopontine tegmentum (PPT), LDT, dorsal raphe (DR), parabrachial nucleus (PB) and basal forebrain (BF) and ventral tegmental area (VTA) are active during both wake and REM sleep, and they induce arousals from NREM sleep but not from REM sleep.^73–84 Interestingly, dorsal raphe serotonergic neurons are shown to promote sleep or have no effect on wakefulness.^83,85 Finally, the activity of histaminergic neurons in the tuberomammillary nucleus (TMN) does not anticipate wake transitions but starts increasing ~1 sec after wake-onset⁷⁰ and, therefore, an unlikely candidate for terminating REM episodes. Thus, Orx and LC^NE neurons, rather than other wake-promoting regions, may participate in REM termination, which is likely accomplished by activating the vlPAG/LPT^GABA neurons and/or directly inhibiting the SLD neurons. However, Orx neurons may also prevent REM transitions and reduce REM sleep via vlPAG/LPT or paradoxically increase REM sleep via SLD neurons (see below). Additional work is necessary to determine the precise contribution of LC^NE and Orx neurons in REM termination and to delineate their interactions with the SLD^Glut and vlPAG/LPT^GABA neurons.

3.3. Influence of other cell groups on the REM circuit

While the REM-on SLD^Glut and REM-off vlPAG/LPT^GABA neurons constitute the primary REM sleep circuit, several other cell groups may influence their activity and modulate REM sleep. Among them, the most studied are the melanin-concentrating hormone-(MCH) and orexin (Orx)-producing neurons located in the LH and the cholinergic and monoaminergic cell groups in the brainstem.

MCH neurons:

MCH neurons are maximally active during REM sleep, but a subpopulation may also be active during wake (Wake-REM sleep active).^86–89 Nevertheless, activation of MCH neurons by optogenetic or chemogenetic methods consistently promoted REM sleep either by facilitating NREM→REM transitions or increasing the duration of REM bouts^90–96. MCH neurons project to both the primary REM control sites, the vlPAG/LPT, and the SLD, as well as to several wake-promoting cell groups, including the tuberomammillary nucleus, locus coeruleus, and other monoaminergic populations.^93,97–100 MCH neurons projecting to the vlPAG/LPT, but not those directly projecting to the SLD, are activated during REM sleep hypersomnia that follows selective REM deprivation. Consistently, focal inhibition of MCH terminals in the vlPAG/LPT reduced spontaneous REM transitions and almost completely abolished the higher rate of REM transitions induced by MCH neuron (soma) activation.⁹³ These data suggest that MCH neurons primarily inhibit the vlPAG/LPT^GABA neurons, thereby disinhibiting the SLD^Glut neurons to facilitate REM sleep (Fig 2). The vlPAG/LPT^GABA neurons also project back to LH, and selective activation of their terminals in the LH terminated ongoing REM episodes⁵⁹, suggesting the MCH neurons are likely inhibited. Thus, in addition to disinhibiting the SLD^Glut neurons, MCH neurons may disinhibit themselves and reinforce their firing to facilitate REM transitions and sustain REM bouts. Finally, photoactivation of MCH terminals in the TMN prolonged the REM sleep episodes⁹⁶, indicating that MCH neurons could inhibit the TMN and presumably other wake-promoting cell groups and delay the wake-onset to sustain the REM episodes.

Orexin neurons:

Orx neurons are maximally active during active wakefulness, but their activity is also higher during REM sleep, although juxtacellular recordings found they are almost silent during REM sleep^89,101–103. Moreover, Orx neuron firing positively correlates with muscle tone across sleep-wake stages, and even during REM sleep, phasic muscle twitches are accompanied by the burst firing of Orx neurons^89,101,102. Loss of orexin neurons, peptides, or receptors in humans and animals causes abnormal intrusions of REM sleep and muscle atonia (or muscle weakness) episodes during active wakefulness (sleep-onset REM and cataplexy; discussed below)^104–109, indicating that the Orx system may inhibit REM transitions, an action antagonistic to MCH neurons. While extensive evidence supports this REM-inhibitory role of Orx neurons, a recent study found that a subset of orexin neurons directly innervating the SLD may promote or stabilize REM sleep.¹⁰³ Activating Orx neuron terminals in the SLD prolonged REM sleep episodes, whereas silencing them reduced REM sleep amounts and disrupted muscle atonia.¹⁰³ Nevertheless, these REM-promoting or stabilizing actions can be attributed to a minor proportion (about 8%) of Orx neurons, while the majority may exert an inhibitory effect on REM sleep, which may be via activation of vlPAG/LPT neurons (Fig 2).

Cholinergic and monoaminergic neurons:

Pharmacological and electrophysiological studies originally suggested that the cholinergic promote and monoaminergic neurons suppress REM sleep^110,111. However, neurotoxic lesions restricted to the PPT or the LDT containing cholinergic neurons did not reduce REM sleep^112,113. Similarly, lesions including the monoaminergic neurons in the LC or DR did not impact REM sleep levels^{31,38,113,114}. While these lesion studies did not differentiate cell types, genetic manipulations specifically targeting cholinergic neurons also do not indicate a necessary role for these neurons in REM sleep regulation^114,115. However, monoaminergic neurons, especially LC neurons, like the vlPAG neurons, must be silenced for REM sleep to occur^69,116–118. Several classical and recent studies confirmed that monoaminergic neurons cease firing, and their tone is either very low or absent during REM sleep.^69,117,118 Several antidepressants that increase the monoaminergic tone suppress REM sleep for hours along with an increase in cFos in the vlPAG/LPT region.^119,120 Thus, monoaminergic neurons may tonically activate the vlPAG/LPT neurons to prevent REM sleep, and both these structures may be turned off by an inhibitory mechanism, permitting REM entry.

4. SLD^Glut neuronal pathways generating REM sleep signs

4.1. SLD^Glut projections to the medulla and spinal cord orchestrate atonia in the spinal musculature

As described above, loss of SLD^Glut neurons or Glut neurotransmission from the SLD neurons caused REM sleep without atonia and dream-enactment behavior in rats and mice, demonstrating that SLD^Glut neurons are necessary for normal expression of REM sleep atonia in skeletal musculature.^52–54 On the other hand, classical work, especially from the laboratories of Michael Chase and Jerry Siegel, has demonstrated that glycine-mediated post-synaptic inhibition of spinal motor neurons by reticulospinal premotor neurons in the ventromedial medulla (VMM) underlies REM sleep atonia, and this site might be the crucial relay via the pontine REM-on region (i.e., SLD) inhibit spinal motor neurons.^10,121–128 Tracing studies confirmed that SLD^Glut neurons monosynaptically project to GABA/glycinergic neurons in the VMM.^42,43,52,129 Also, specific loss or inactivation of VMM^GABA/Gly neurons resulted in REM sleep without atonia characterized by exaggerated phasic twitches and involuntary motor activities similar to those observed in mice with loss of SLD^Glut neurons.^129–131 The VMM comprises many subregions, including paramedian reticular nucleus, ventral gigantocellular nucleus (GiV; also referred as pre-supraolivary medulla (pSOM) in mice and rats or magnocellular tegmental field in cats), ventral and rostral paragigantocellular nucleus and gigantocellular nucleus-pars alpha (known as the magnocellular nucleus in cats)^132–134. However, GABA/Gly knockouts or neuron inactivation centered at the GiV in front of the inferior olive produced more potent effects in terms of the motor behavior responses and the magnitude of phasic twitches, whereas those missing this region produced only minor effects.^{43,44,131,135} Consistent with this, selective deletion of a subset of SLD neurons projecting to the GiV region attenuated REM sleep atonia and induced dream-enactment behavior in mice.¹²⁹

In addition to VMM, SLD^Glut neurons directly project to lamina VII of the spinal cord containing GABA/Gly interneurons.^42,54 Specific elimination of GABA/Gly neurotransmission from C3-C4 level of spinal cord induced twitching and jerking movements in the upper extremities, although occasional twitching also was observed in the lower extremities and tail during REM sleep.⁵⁴ In contrast, whole-body movements or any form of locomotion were never observed in these mice. These effects were milder compared to those produced by SLD^Glut or VMM^GABA eliminations, but it is important to note that these spinal cord knockouts were confined to c3-c4 levels. Collectively, the available data indicate that SLD^Glut neurons activate VMM^GABA/Gly and spinal interneurons to induce atonia in spinal musculature (Fig.3).

Figure 3: Circuit mechanism orchestrating REM sleep atonia.

When the SLD^Glut neurons that promote REM sleep become active, they stimulate the GABA/Glycinergic neurons in the VMM and spinal cord. This results in glycine-mediated post-synaptic inhibition of ventral horn motor neurons (blue), which induces atonia in the spinal muscles. It is probable that the mechanism responsible for atonia in cranial muscles is similar, but there may also be a need for the withdrawal of monoaminergic signaling.

SLD - Sublaterodorsal nucleus; vlPAG - Ventrolateral periaqueductal gray; VMM – Ventromedial medulla.

Purple – glutamatergic; Red: GABA/Glycinergic; Blue: cholinergic neurons

4.2. Cranial muscle atonia may have similar mechanisms

While the spinal muscle atonia circuits have been extensively investigated, cranial muscle atonia circuits are still far from clear. Nevertheless, atonia in masseter and genioglossus muscles, innervated by the trigeminal (Mo5) and hypoglossal (Mo12) motor nuclei, are relatively better understood. Similar to spinal motor neurons, Mo5 and Mo12 are also subjected to glycinergic postsynaptic inhibition during REM sleep.^122–124 Activation of VMM^GABA/Gly neurons decreased the amplitude and duration of inspiratory-related EMG bursts in tongue muscles, confirming the inhibitory actions of VMM^GABA/Gly neurons on Mo12 neurons.¹³⁶ VMM^GABA/Gly neurons also project to Mo5, facial nerve nucleus, and accessory nerve nucleus¹²⁹ suggesting that they may also be inhibited similarly, but direct evidence is currently lacking. Moreover, these projections to cranial nerve nuclei appear collaterals from the same VMM^GABA/Gly neurons projecting to the spinal cord ventral horn.¹²⁹ Taken together with their role in spinal muscle atonia, these tracing data strongly suggest that VMM^GABA/Gly neuronal inhibition may underlie cranial muscle atonia as well. Interestingly, VMM^GABA/Gly neurons do not project to oculomotor, trochlear, and abducens nerve nuclei,¹²⁹ which is consistent with the intact muscle activity (eye movements) in the ocular muscles during REM sleep. While these data point out a common circuit for spinal and cranial muscle atonia, pharmacological blockade of GABA and/or glycine neurotransmission in Mo12 or Mo5 had minimal effects on atonia in genioglossi and masseter muscles.^137–140 In contrast, combined blockade of excitatory adrenergic and serotonergic receptors abolished the REM sleep-like depression of Mo12 motoneuronal activity in rats.^141–143 These data suggest that the disfacilitation of monoaminergic inputs to Mo5 and Mo12 and possibly other cranial motor neurons contribute to cranial muscle atonia. Thus, glycinergic inhibition and monoaminergic disfacilitation may underlie cranial muscle atonia.

The phasic twitches during REM sleep are more frequent in cranial muscles than spinal muscles. They appear to be governed by glutamatergic drive onto the cranial motor nuclei. For example, Mo5 receives significant glutamatergic projections from the parvocellular reticular nucleus (PCRt) and, to a certain extent, from the paramedian reticular area (PMnR)^144,145. Lesions of PCRt in rats or elimination of glutamate neurotransmission eliminated REM twitches in masseter muscles¹⁴⁶. PCRt glutamatergic neurons also project to the facial (Mo7) and hypoglossal (Mo12) nuclei¹⁴⁵, suggesting their role in facial and tongue muscle twitches, but direct evidence is currently lacking.

4.3. Cortical and hippocampal activation during REM sleep

Desynchronized EEG with high theta activity (5–12 Hz range with a prominent 7 Hz rhythm) during REM sleep reflect the higher cortical and hippocampal activities in this stage^{3,6,11,147–149}, and independent pathways may control them. Cortical activation during REM sleep is hypothesized to utilize the same circuit activating the cortex during wakefulness. Thalamocortical neurons by the brainstem ascending arousal system were initially considered to underlie cortical activation during wakefulness.^150–153 However, recent findings from Lu, Saper, and colleagues convincingly demonstrated that BF rather than the thalamus arouses the cortex during wakefulness.¹⁵⁴ BF receives robust input from the parabrachial nucleus in the pons¹⁵⁴. Neurotoxic lesions in the BF and PB led to a coma-like state in rats with persistent slow EEG with ~1Hz oscillations.¹⁵⁴ In contrast, activation of PB glutamatergic neurons or their terminals in the BF resulted in EEG activation and behavioral arousal, whereas activation of PB terminals in the thalamus had no impact.⁷⁹ These data confirm that the PB^Glut→BF pathway is necessary for cortical activation during wake. It is likely that SLD^Glut neurons activate this circuit during REM sleep (Fig. 4). While the SLD sends moderate projections to the PB, their importance in EEG activation during REM has not been directly demonstrated.

Figure 4: Circuits orchestrating EEG features of REM sleep.

SLD^Glut neurons activate the cortex through relays in the parabrachial nucleus (PB) and the basal forebrain (BF), causing desynchronized EEG. On the other hand, they activate medial septal neurons projecting to the hippocampus to generate theta rhythms.

BF- basal forebrain; Ctx: Cortex; Hip - hippocampus; MS – medial septum; PB - Parabrachial nucleus; SLD – Sublaterodorsal nucleus.

Purple – glutamatergic; Red: GABAergic neurons;

On the other hand, Hippocampal theta rhythms are driven by neurons in the medial septum (MS) projecting to the hippocampus (septohippocampal pathway)^155–157. Both GABA and cholinergic neurons in the medial septum project to the hippocampus, while lesions including both these populations, but not the cholinergic population alone, abolished REM theta rhythms in rats, suggesting the crucial role for MS^GABA neurons in theta generation.^158–160 Consistently, optogenetic inhibition of MS^GABA neurons attenuated theta rhythms during REM sleep in mice without affecting REM sleep amounts.¹⁶¹ MS receives direct projections from the SLD, and selectively activating these projections increases REM sleep theta power⁴⁸. Thus, the SLD may activate the septo-hippocampal pathway to orchestrate theta rhythms during REM sleep (Fig.4).

5. Dysfunction of REM sleep circuits

5.1. REM behavior disorder (RBD)

RBD is a parasomnia characterized by loss of inherent muscle atonia during REM sleep; hence, individuals with RBD tend to enact their dream and display overt motor behaviors, including jerking, jumping, punching, kicking, vocalization, screaming, and shouting while in REM sleep.^162–164 Most of these behaviors are violent or aggressive, leading to self-injuries and injuries to partners.^162–164 The RBD is relatively common, with an estimated prevalence of 0.5 – 1.25% in the general population affecting about 40–100 million individuals worldwide¹⁶⁵; these numbers could go up as many cases go unrecognized. Another critical observation by Schenck and colleagues is that most RBD patients eventually develop Parkinson’s disease (PD).¹⁶⁶ Subsequent studies confirmed these observations and found that about 50% of idiopathic RBD (iRBD) patients develop PD or other alpha-synuclein diseases, such as multiple system atrophy and dementia with Lewy bodies within 12 years after RBD diagnosis.^167–172 RBD is now considered a prodromal marker for PD, and it is important to note that other markers do not have such a high predictive value.^167,173,174 Thus, understanding the neural basis of RBD assumes even greater significance in this context.

As one can speculate from the animal literature on REM atonia circuits, RBD in humans is associated with damage to brainstem structures. Several studies have found focal ischemic, inflammatory, degenerative, or demyelinating lesions involving the dorsal pontine tegmentum or confined to the subcoeruleus region (the human equivalent of SLD) in patients with RBD.^175–183 Even unilateral lesions of the subceoruleus could cause RBD.^176,184,185 Similarly, acute hemorrhage damaging the medial medulla and, presumably, the descending pontine projections was found in one RBD patient.¹⁸⁴ However, reports on RBD patients with medullary lesions are rare, presumably because these lesions often lead to more severe motor and cardiovascular complications. Lesions involving other regions connected to the subcoeruleus – amygdala or anterior thalamus also observed in RBD, but interestingly with no apparent damage to the brainstem.^186,187 It is likely that these lesions reduced the SLD neuronal activity. In contrast, no RBD was observed in some cases with lesions in the pontine tegmentum, and these individuals were reported to exhibit even lower muscle tone during REM sleep compared to healthy controls.¹⁸⁸ However, these pontine lesions were more medial, likely missing the subcoeruleus region. The reduced muscle tone during REM sleep in these patients could be due to fewer muscle twitches, suggesting that phasic REM twitches may originate from this region.¹⁸⁸ Collectively, neuroimaging studies in humans agree with animal data and confirm that lesions in the subcoeruleus and its connections may be the fundamental cause of RBD.

While a significant proportion of RBD patients ultimately develop PD and a subset of PD patients display RBD symptoms, dopaminergic deficiency alone does not lead to RBD in humans or animals. For example, in several human cases of RBD, lesions do not include substantia nigra (SNc) but are restricted to brainstem regions.^175–177 Similarly, experimental SNc lesions neither induce RBD-like motor behavior nor affect REM atonia. However, several iRBD patients display impairment of SNc and loss of dopamine activity in the striatum, which receives extensive SNc dopaminergic inputs.^189–195 Thus, the degenerative changes could begin simultaneously in the SLD and SNc, but PD motor symptoms do not appear until the loss of 70% or more of SNc dopamine neurons. Another possibility is that neurodegenerative changes begin in the brainstem, causing RBD, and progress rostrally to the midbrain, causing PD. Supporting this idea, rats chronically exposed to rotenone, a pesticide known to cause PD, exhibited RBD features, which could be temporally correlated with the progressive alpha-synuclein aggregation and neuronal apoptosis in the SLD and VMM regions¹⁹⁶. Importantly, injections of preformed α-synuclein fibrils into the SLD induced degeneration of SLD neurons followed by widespread propagation of α-synuclein pathology, respectively, accompanied by RBD-like behaviors and PD phenotypes in mice.¹⁹⁷ Irrespective of the mechanisms, i.e., simultaneous vs. progressive neurodegeneration in SLD and SNc, diagnosis of RBD provides a ~ ten-year window for therapeutic interventions to protect the remaining SNc dopamine neurons.¹⁹⁸ Such early interventions could significantly aid in curbing PD in its early stages or preventing the phenoconversion altogether.

5.2. Narcolepsy with Cataplexy

Narcolepsy (type 1; NT1) is another parasomnia associated with REM circuit dysfunction, and this order is characterized by two prominent features - sudden loss of muscle tone triggered by strong positive emotions during wake (cataplexy) and excessive daytime sleepiness. Narcolepsy may also occur without cataplexy (type 2). It is well established that NT1 is caused by dysfunction of Orx-producing neurons^199–201. Narcoleptic humans exhibit loss of Orx neurons in the LH and low levels of orexin-A in their cerebrospinal fluid. Similarly, Orx system dysfunction (neurons, neuropeptides, or receptors) in animals leads to cataplexy and other narcolepsy symptoms. As atonia during cataplexy is similar to that in REM sleep, abnormal activation or disinhibition of REM atonia circuits during wake is considered responsible for cataplexy. Consistent with this idea, activation of SLD^Glut neurons or VMM^GABA/Gly neurons increased spontaneous cataplexy in mice.^53,129 In contrast, inhibition of SLD^Glut neurons or their terminals in the VMM or activation of vlPAG/LPT^GABA neurons that may, in turn, inhibit the SLD^Glut neurons attenuated both spontaneous cataplexy and emotional cataplexy in narcoleptic mice.^53,59,129 In contrast to these data, activating the SLD during active wakefulness did not induce muscle atonia in rats with intact orexin system.⁴⁸ Further, SLD lesions in the Orx-KO mice reduced, but did not abolish, cataplexy in Orx-KO mice.⁴⁷ It is important to note that REM sleep was completely absent in these mice, indicating all REM-generating SLD neurons were presumably lost. These data suggest that cataplexy can occur even without REM-generating SLD neurons. Thus, activation of the REM atonia circuit (SLD-VMM pathway) is unlikely to be the only or the final common pathway for cataplexy. Supporting this idea, lesions of the midbrain locomotor region, especially the LPT, produced cataplexy.^42,202 Glut neurons in this region express Orx receptors and project to VMM or spinal cord motoneurons.²⁰² Thus, abrupt deactivation of LPT^Glut neurons and their descending projections may also contribute to cataplexy.

As strong positive emotions generally trigger cataplexy, one can presume that the cataplexy circuits must be under the control of the limbic forebrain controlling emotional responses. In this regard, the medial prefrontal cortex (mPFC) and central nucleus of the amygdala (CeA) appear crucial. Neurotoxic lesions in these regions reduce both spontaneous and emotional cataplexy^203,204. Moreover, activation of CeA^GABA neurons increases both types of cataplexy while their inhibition selectively attenuates emotional cataplexy.^205,206 These GABA neurons heavily innervate the mPFC and vlPAG/LPT, and mPFC, in turn, projects to the vlPAG/LPT.^207–209 Thus, the deactivation of vlPAG/LPT neurons by mPFC and CeA, in the absence of orexins, may trigger emotional cataplexy (Fig 5).

Figure 5: Circuit dysfunction underlying cataplexy.

Loss of Orx neurons is the fundamental cause for cataplexy. Positive emotional stimuli may activate MCH, CeA^GABA and mPFC neurons. Their activity on to the vlPAG/LPT and SLD neurons, when unbalanced in the absence of Orx neurons, triggers cataplexy.

CeA – central amygdala; mPFC – medial prefrontal cortex; LPT- lateral pontine tegmentum; MCH-melanin-concentrating hormone; Orx -Orexin; SLD - Sublaterodorsal nucleus; vlPAG - Ventrolateral periaqueductal gray.

Another cell group implicated in cataplexy is MCH neurons, primarily based on their close association with the Orx neurons. Despite the drastic loss of Orx neurons, MCH neurons remain intact in narcoleptics.²¹⁰ MCH neurons expressed cFos after periods of cataplexy in Orx-KO mice, and the number of cFos+ MCH neurons positively correlated with cataplexy amounts.²⁰³ In contrast, calcium imaging studies suggest that the activity of these neurons does not correlate with cataplexy or precataplexy behaviors.⁸⁷ Nevertheless, activation of MCH neurons in Orx-KO mice increased cataplexy, and systemic administration of MCH receptor antagonist completely suppressed it, suggesting abnormal activity of MCH neurons may contribute to cataplexy.⁹⁴ Interestingly, loss of MCH neurons aggravated the cataplexy levels in narcoleptic mice lacking orexin neurons.²¹¹ Considering the mutual inhibition between MCH and orexin neurons, these data suggest that the loss of either side may leave the animals more prone to cataplexy while activation of each side produces opposite effects on cataplexy. Manipulating MCH neuron activity, specifically during periods of cataplexy, may be necessary to ascertain their role in the pathophysiology of NT1.

6. Conclusions

One of the most important discoveries in recent years is that REM sleep may be eliminated permanently by the selective ablation of SLD neurons.^47,48 These findings suggest that REM sleep is generated exclusively by the SLD^Glut neurons. Thus, all other brain regions that impact REM sleep levels, including the caudal brainstem, must do so by directly or indirectly influencing SLD^Glut neuron activity. SLD^Glut neurons orchestrate specific REM sleep features via multiple pathways. They induce spinal muscle atonia primarily by activating the VMM^GABA/Gly neurons but decreased monoaminergic neuronal activity also appears to contribute to cranial muscle atonia. Similarly, SLD^Glut neurons activate the cortex and hippocampus via separate pathways involving PB^Glut and MS^GABA neurons, respectively, leading to desynchronized EEG with high theta during REM sleep. Understanding these pathways helped immensely in understanding the neural basis of REM parasomnias, including cataplexy and RBD. Cataplexy may be caused by abnormal activation of the SLD^Glut neurons and their downstream atonia pathway during the wake period, although concurrent inhibition of LPT neurons may also be necessary. Similarly, REM behavior disorder (RBD) is caused by damage to the SLD^Glut neurons and their connections. However, further research is needed to determine the precise mechanisms that activate SLD neurons during cataplexy, underlying causes of SLD neuronal loss in RBD, and the molecular mechanisms driving progression of RBD to PD and alpha synucleiopathies.