What Is Regulation?

The title of this article is important—I use the word “regulation,” not “control.” Although these two words are frequently used interchangeably, they have different meanings. “Control” applies to mechanisms that cause a process to go faster or slower, higher or lower, and bigger or smaller without reference to any particular value. “Regulation” applies to the use of control mechanisms to achieve or maintain a particular value frequently called a “set point” by engineers. Probably no true “set point” exists in physiology, but rather a range of values that keeps the system and the organism in homeostatic balance. That range may involve a number of modulatory factors that alter the expression of the regulated process. Regulation requires information about the difference between the current state of a system and its optimum level. That information is referred to as negative feedback. Homeostatic systems use negative feedback information to manage their mechanisms of control. We know much about sleep control systems that initiate and terminate sleep and alter sleep stages (e.g., Chen et al., 2018; Pace-Schott & Hobson, 2002; Saper & Fuller, 2017; Scammel et al., 2017), but we know much less about the homeostatic regulation of sleep and sleep stages. One reason for this gap in our knowledge is that we do not know the primary function(s) of sleep. Because negative feedback variables are derivatives of the regulated function, understanding the function would help identify possible negative feedback variables and thereby discover the mechanisms of regulation of the system.

First, we must ask whether sleep is regulated or what aspects of sleep are regulated. For example, is the circadian timing of sleep regulated? The circadian sleep phase is controlled by a fixed period oscillator that resides in the suprachiasmatic nucleus. However, that oscillator, which has an endogenous period different from 24 hours, is entrained by negative feedback information from the environmental light–dark cycle (Duffy & Wright, 2005). That feedback information brings the circadian free-running period into synchrony with the 24-hour day. That feedback information also functions to bring the endogenous rhythm into proper phase relationships following travel across time zones—recovery from jet lag. Thus, we can say that the circadian timing of sleep is regulated. Is the duration of the circadian sleep phase regulated? Prolonging a wake phase generally does not lengthen the subsequent sleep phase of the rhythm. Rather, the subsequent sleep phase is more likely to be shorter so that the circadian period is maintained. Thus, the duration of the circadian sleep phase does not seem to be regulated. However, excessive accumulation of sleep debt, such as from a week of short sleep because of work, may override the circadian control and result in a prolongation of sleep beyond the normal sleep phase.

Putting aside the circadian timing of sleep or wake, there is definitely evidence for homeostatic relationships between sleep and wake. Prolonging wake results in an increased drive for sleep as measured by the Multiple Sleep Latency Test (MSLT) (Carskadon & Dement, 1979) and other measures of sleepiness. The aspect of sleep that most clearly reflects homeostatic regulation is EEG slow-wave activity, or delta power—the spectral power in the 0.75–4.5 Hz range (Borbély, 1982; Dijk & Czeisler, 1995). Another aspect of sleep that appears to be homeostatically regulated is the non-rapid eye movement to rapid eye movement (NREM–REM) cycle (Benington & Heller, 1994).

This article is not an extensive review, so many important papers are not cited—my apologies to colleagues. My purpose in this brief treatment is to hit significant landmark papers that together with much supporting work are leading us to new insights into the regulation of sleep.

Sleep as a Function of Wake

William Dement, the father of sleep medicine in the United States, frequently said, “The function of sleep is to reduce sleepiness.” That is actually a profound statement because it means that there is a homeostatic relationship between wake and sleep, and a measure of that relationship is sleepiness. Several methods have been devised to quantify sleepiness. The Multiple Sleep Latency Test (MSLT) is an objective measure. The instrumented subjects are placed in beds in a quiet, isolated environment, and the measure is how long it takes them to fall asleep. At 10 a.m., for those who had a good prior night of sleep, it might take them between 15 and 20 minutes to fall asleep, but for those who had prior night sleep restricted to 4 hours, they will fall asleep in about 5 minutes. If the MSLT is repeated at different times of day, it reveals the accumulation of additional sleepiness until early to late afternoon, after which the MSLT scores go down as the influence of the circadian wake maintenance effect is expressed throughout the evening and late p.m. A similar objective measure of sleepiness is the Maintenance Wake Test (MWT), which is very similar to the MSLT, except the instrumented subjects are asked to sit quietly in an isolated environment, to see how long they can stay awake. Subjective measures of sleepiness consist of questionnaires; examples are the Stanford Sleepiness Scale, the Epworth Sleepiness Scale, and the Karolinska Sleepiness Scale. The important point is that sleepiness reflects a drive for sleep that accumulates during wake—a classic homeostatic regulatory relationship. What variable of sleep reflects the homeostatic function sleep serves that reduces sleepiness?

Slow Wave Activity as a Homeostatic Sleep Variable

In the 1960s–1980s, studies showed that sleep deprivation in many species of mammals resulted in common changes in the spectral properties of the cortical EEG during NREM sleep (Borbély et al., 1984), and in 1981 the same was shown to be true in humans (Borbély et al., 1981; Borbély & Achermann, 1999). Those changes were an increase in the spectral power in the 0.5–4.5 Hz range—the delta band. That increase was shown to be proportional to the duration of prior wake, and that slow-wave power declined monotonically during subsequent NREM episodes during recovery sleep, as had been shown earlier for the normal progression of NREM episodes across the night (Feinberg et al., 1978). Naps during the wake phase of the daily cycle decreased the slow-wave power recorded during the subsequent sleep phase.

The classic sleep EEG recordings reflect voltage oscillations over large expanses of cortex. The current practice for the clinic is referred to as the 10–20 system, in which 21 electrodes are placed on the scalp at either 10% or 20% intervals of the total distance from front to back and from left to right. EEG traces can be recorded between any two electrodes, and they will differ but show the same basic wave-form patterns. If electrode pairs are close together but located over different regions of the brain, differences between these recordings can reveal differences in local activity. Such recordings led to the description of local sleep (Krueger & Tononi, 2011; Vyazovskiy et al., 2011) activity that reflected prior wakeful local activity. That approach has been extended to high-resolution mapping of local activity across the entire cortex by using large (as many as 256) electrode arrays. Using such a large electrode array makes it possible to compare regional brain activities. One such study showed that the expression of delta activity in a specific brain region is proportional to the level of activity in that brain region during prior wake (Huber et al., 2004). The subjects in this study were trained to use their non-dominant hand or arm to use a computer mouse to direct the cursor to a target on the screen. However, the computer introduced a nonlinearity in the communication between the cursor and the mouse, so the subjects had to learn to make adjustments for that nonlinearity. Comparing the local EEG during subsequent sleep, for the left and right motor cortices over the hand or arm region, showed greater slow-wave activity or delta power over the cortical region that had been engaged in the training. In addition, there was a positive correlation between the delta power measured and the improvement in performance the following day. Thus, the recorded delta power appeared to reflect the negative homeostatic feedback between wake activity and the sleep response. What is that feedback signal, and what does it suggest for a homeostatically regulated function?

Adenosine as the Sleep Homeostatic Feedback Signal

The postulation that adenosine was the feedback signal controlling sleep came originally from Miodrag Radulovacki (2005). Although several studies in a variety of animals between 1954 and 1973 reported that brain administration of adenosine produced behavioral sleep, Radulovacki and colleagues took a quantitative approach, showing that a number of adenosine analogs induced dose-related increases in sleep, both non-rapid eye movement (NREM) and rapid eye movement (REM) (Radulovacki et al., 1984). Based on the effective dose ranges, they also proposed that the action was through the adenosine A1 receptor. Although they used polygraphic electroencephalography (EEG) recordings, their results were quantified only in terms of total durations of wake and sleep states over 6-hour periods following drug administrations. They concluded that their results “indicate a role for adenosine in the regulation of sleep” (p. 273).

The link between the actions of adenosine through the A1 receptor and the EEG measures of sleep homeostasis was made through experiments in which an adenosine analog was administered vis intracerebroventicular (ICV) or intraperitoneal (IP) in rats while recording EEG and subjecting those recordings to Fourier analysis (Benington et al., 1995; Porkka-Heisskanen et al., 1997). The EEG responses to the adenosine analog mirrored the spectral profiles recorded after sleep deprivation. The authors concluded: “The fact that CPA [cyclopentyladenosine] produces changes in the nonREM-sleep EEG that closely match those produced by TSD [total sleep deprivation] suggests that endogenous adenosine mediates the sleep-homeostatic modulation of the nonREM EEG.” They further speculated that the observed effects of adenosine were mediated via direct effects on neurons in the cerebral cortex and thalamus. An alternative hypothesis was that the effects of adenosine on sleep propensity and nonREM-sleep EEG are mediated through inhibition of brainstem cholinergic neurons (Rainnie et al., 1994). Weighing against the hypothesis that the homeostatic action of adenosine is due to its actions in the basal forebrain were the facts that the A1 receptors are widely distributed throughout the cortex and thalamus (Fastborn et al., 1987), NREM slow-wave activity (SWA) is a function of thalamocortical circuits, and local activation of cortical regions during wake alters the SWA in those regions during subsequent sleep (Kattler et al., 1994; for a review, see Krueger et al., 2019).

The role of adenosine acting through the A1 receptor as a negative feedback signal involved in sleep homeostasis is now well supported (Greene et al., 2017), but the story has become more complicated. Convincing evidence for the negative feedback function of adenosine is the finding that knocking out the gene for the A1 receptor eliminates the increase in SWA in response to prior sleep deprivation. However, after the knockout, sleep structure is quite normal, and there are no cognitive deficits in comparison with wild-type (WT) animals as long as both have ad lib sleep. The knockout animals do not respond to sleep deprivation with a homeostatic response in either SWA or sleep extension, and they show cognitive deficits (Bjorness et al., 2009). Thus, there appear to be two levels of sleep regulation: constitutive—the occurrence of the normal sleep–wake cycle, and adaptive—the occurrence of a compensatory response to greater than average activity during preceding wake.

Another line of evidence supports the concept of constitutive and adaptive sleep regulatory processes, and this work also implicates astrocytes as a crucial component of an adaptive adenosinergic signaling mechanism (Halassa et al., 2009). In that study, it was postulated that the adenosine responsible for the sleep homeostatic response is released by astrocytes. The researchers produced mice with conditional expression of a dominant negative form of the SNARE protein necessary for the gliotransmitter release that controls the levels of extracellular adenosine. Expression of the dnSNARE eliminated the in vitro and in vivo responses to A1 antagonists, but not to A1 agonists, indicating that expression of this transgene eliminated the purinergic transmission by the astrocytes. Because the construct was conditional, the effects were reversible. Similar to the A1 receptor knockout mice, the dnSNARE mice had normal structure of ad lib sleep, but they lacked the SWA response to prior sleep deprivation, indicating again that the homeostatic regulation of sleep was different from regulation of the timing and structure of sleep. Unlike the knockout mice, however, these mice did not show the cognitive deficits associated with prior sleep deprivation. Comparing the cognitive results of these two studies is difficult because they used different sleep deprivation methods and different cognitive tasks.

Hubbard et al.’s (2020) study called into question the concept that SWA reflects a homeostatic sleep response. With detailed analysis of the EEG recordings of sleep in mice and in humans, this study confirmed that the EEG delta band consists of two components, one fast (2.5–3.5 Hz) that rises quickly with sleep onset and declines back to baseline in a relatively short time, and one slow (0.75–1.75 Hz) that has a much lower rise at sleep onset and declines back to baseline slowly. A similar distinction was described by Halassa et al. (2009). However, in the Halassa et al. study, the slow component showed the large response to SD and not the fast component. Hubbard et al. (2020) conclude that the typical delta power response that is seen to characterize sleep homeostasis is really an artifact of the combination of the contributions of the two distinct SW components. They hypothesize that the dynamics of the fast delta component reflects the suite of physiological changes that are associated with the transition from wake to sleep. Whereas this argument may be supported by the first NREM episode of the sleep phase, it is difficult to understand how it could apply to the second and subsequent NREM episodes (Borbély et al., 1981; Franken et al., 1991), or for the regional delta power differences associated with local sleep responses resulting from selective activation of specific cortical regions (Krueger et al., 2019).

In consideration of the fact that this article is focused on regulation and not control of sleep, it seems justified to claim that SWA is a feature of sleep that is regulated, and at least one feedback parameter is adenosine. Considering that the feedback parameter in a physiological regulatory system reflects the state of the regulated function, what can we surmise about NREM sleep function from the identification of adenosine as a prominent feedback signal in the regulation of SWA?

Changes in adenosine levels suggest energetic processes. That connection led to the energy reserve hypothesis as a function of NREM sleep (Benington & Heller, 1995a). The only energy reserve in the brain is glial glycogen, and that hypothesis proposed that high levels of regional neural activity during wake could cause the local release of adenosine, inducing sleepiness, and adenosine would continue to be produced during NREM sleep as a byproduct of the resynthesis of glycogen. That wakeful neural activity could regionally deplete glycogen was tested by Kong et al. (2002). They measured glycogen content of different brain regions as a function of sleep or extended wake. Those experiments as well as others on Drosophila (Zimmerman et al., 2004) supported the hypothesis, but other studies did not do so consistently (Franken et al., 2003). However, single time point measurements of global glycogen content did not offer the possibility of revealing the dynamic balance of synthesis and breakdown, nor could large tissue sample glycogen content reveal events at the local level. Rather than looking at tissue glycogen levels, Petit et al. (2002) examined the effects of sleep deprivation on the molecular components of glycogen synthesis. They showed that 6 hours of sleep deprivation induced a threefold increase in one form of glycogen synthase and a twofold increase in expression of protein targeting to glycogen (PTG), which supports glycogen synthesis. These results suggest that during extended wake with high levels of glycogenolysis, changes in gene expression set the stage for glycogen synthesis during subsequent sleep (Allamann et al. 2000; Allaman et al. 2003; Sorg & Magistretti, 1991).

The possibility that the feedback signal regulating delta power reflects brain energetics gains support from Magistretti and Allaman’s (2015) work elucidating the crucial role of astrocytes in providing fuel for neurons. Astrocytes supply fuel to neurons in the form of lactate, and sources of that lactate are glycogenolysis and glycolysis. Injections into the hippocampus of a blocker of astrocytic glycogenolysis impairs hippocampal working memory that can be restored by injections of lactate or glucose into the hippocampus (Newman et al., 2011). Blockage of the transporter for uptake of lactate by neurons also impairs working memory, but that impairment was not recoverable by injections of glucose or lactate. Clearly, lactate supplied to the neurons from astrocytes is essential for the hippocampal working memory functions. How does sleep enter into this relationship? High or extended levels of neural activity during wake creates a demand for astrocytic lactate production. That lactate comes from glycolytic breakdown of glucose coming both from the blood and from glycogenolysis. The glycolytic products either supply lactate to the neurons or enter the TCA cycle to supply ATP for astrocytic functions. Thus, it is possible that extended neural activity of wake puts energetic demands on the astrocytes, lowering their ATP/ADP ratio, resulting in release of adenosine. Adenosine promotes and maintains sleep, during which energy balances can be restored (Scharf et al., 2008).

Another approach by Bellesi et al. (2018), used to investigate the changes in glycogen dynamics associated with wake and sleep, is three-dimensional electronmicroscopic imaging of glycogen granules in the astrocytic perisynaptic processes. Prolonging wake resulted in increasing the number of glycogen granules associated with the synapses, but progressively longer episodes of wake resulted in smaller granule sizes and estimated lower amounts of glucose in the granules. The conclusion was that wake mobilized this energy source, and sleep replenished it. Whether we are considering the early global assessments of the effects of sleep and wake on glycogen content, the data on changes in gene expression associated with glycogen turnover, or Bellesie et al.’s results showing sleep–wake-related changes in the dynamics in glycogen particles in astrocytic perisynaptic processes, these studies point to a homeostatic energetic process as being at least one function of sleep.

The NREM-REM Sleep Cycle as a Regulated Sleep Process

Discussions of the regulatory relationships between wake and sleep commonly ignore the fact that there are two very different sleep states. The assumption is made that non-rapid eye movement (NREM) and rapid eye movement (REM) sleep have different functions, but those functions are both serving needs generated by wake. An alternative view is that only one of these states is in a homeostatic relationship with wake, and its expression generates a need or pressure for the other sleep state. Unless the pressure for the dependent state is satisfied, the state serving a function related to wake cannot be adequately expressed. The possible relationships between NREM and REM sleep have been reviewed extensively by Le Bon (2020), but without considering possible functions of the two states or how they interact functionally. Ideas about functions could lead to hypotheses about feedback variables responsible for the homeostatic relationships. For example, another review of NREM–REM relationships (Vyazovskiy & Delogu, 2014) also supports the view that REM sleep serves a need generated by NREM sleep, as well as a functional hypothesis for the NREM–REM cycling. They propose that NREM sleep provides critical neuronal network maintenance functions for restoration following waking activity, and periodic REM sleep tests the adequacy or completeness of those functions.

Because NREM always precedes REM sleep, it would make sense that NREM is serving a need generated by wake, and REM is serving a need generated by NREM sleep. In other words, NREM sleep is serving a need generated by wake and is therefore in a homeostatic relationship with wake, but REM sleep serves a need generated by NREM sleep, and therefore REM sleep is in a homeostatic relationship with NREM sleep. Features of the NREM–REM cycle are compatible with this possibility. The NREM–REM cycling has a rather regular period that is a species-specific characteristic. In humans, NREM sleep is deeper earlier in the night, and REM sleep episodes are longer later in the night. These features of the sleep cycle are compatible with the hypothesis that NREM and REM sleep are homeostatically linked, as demonstrated by some simple experiments.

Close examination of the spectral profiles of sleep in the rat revealed sudden, coordinated changes in three frequency bands—delta, theta, and sigma—at the transition from NREM to REM sleep (Benington & Heller, 1994). These events were termed NREM to REM Transitions (NRTs). Frequently, NRTs were not followed by sustained REM sleep that would be scored as such by standard sleep scoring protocols. However, when NRTs were used to designate the beginning and end of a sleep cycle, an extremely regular distribution of cycle periods resulted, enabling experiments on variables that influenced sleep cycle periods. Prior studies on the control mechanisms of sleep state transitions led to the hypothesis that sleep cycling was controlled by a fixed period oscillator (Fuller et al., 2007; Hobson et al., 1975, Pace-Schott & Hobson, 2002). Using NRTs to measure sleep cycles led to a means of testing the fixed period oscillator hypothesis. If the period is fixed, changes in the duration of one phase will result in opposite changes in the other phases—an inverse relationship. If, however, the two phases are in a homeostatic relationship, lengthening or shortening of the independent phase will produce a similar change in the dependent phase—a proportional relationship. Results of analysis of about 4,500 rat sleep cycles clearly showed that the relationship between NREM and REM phases of sleep cycles was a proportional relationship, indicating that REM sleep was in a homeostatic relationship with NREM sleep (Benington & Heller, 1995b). Similar results were also obtained by Vivaldi et al. (1994). Similar quantitative relationships between NREM and REM sleep bout lengths were also shown in humans by Barbato & Wehr (1998) and in mice by Park et al. (2021). Further work showed that when an NRT did not successfully result in sustained REM sleep, the next NRT came sooner indicating a rapid build-up of REM need (Benington et al., 1994). An important conclusion from these experiments was that any method of so-called selective REM deprivation rapidly impairs NREM sleep as well and therefore is not a valid approach for identifying REM sleep function.

The studies on relationships between NREM and REM sleep indicate a homeostatic relationship with NREM sleep, creating a need for REM sleep. Recent studies of hippocampal theta power conform to this hypothesis (Bjorness et al., 2018). This REM drive is responsible for the expression of REM sleep to dissipate that drive, thus determining the timing of the sleep cycle. Extending this rationale to include wake, the argument can be made that wake creates a need for NREM sleep, and the subsequent expression of NREM sleep creates a need for REM. What could those “needs” be? According to the energy reserve restoration process described in “Adenosine as the Sleep Homeostatic Feedback Signal,” wake results in a depletion of the glial glycogen reserve that must be replenished during NREM sleep. The quiescence of neural activity during NREM sleep is due to hyperpolarization resulting from increased K⁺ conductance. Thus, one hypothesis for the function of REM sleep is that periodically the cells must return to a relatively depolarized state that facilitates the return of K⁺ to the neurons. To test this hypothesis, rats were dosed with a potassium channel blocker at very low doses (higher doses interfered with the expression of NREM sleep). A low dose was found that did not compromise NREM sleep, but resulted in the elimination of REM sleep for the duration of the drug effect with no subsequent REM rebound (Benington et al., 1995). Thus, it is possible that one function of REM sleep is as simple as ionic homeostasis.

Conclusions

This article focuses on just the regulation of sleep. There are many more studies on the control mechanisms of sleep. The article pointed out the difference between these two types of investigations. Control mechanisms turn a function on or off, make it go faster or slower, but without any reference to a specific value. In contrast, regulatory mechanisms require information about the optimal level of the regulated variable, frequently termed a “set point,” and information about the current state of that variable, termed “feedback information.” Both set points and feedback are in reference to a function, and questions about regulation lead to new insights and hypotheses about function. The function or functions of sleep are one of the greatest unanswered questions in neurobiology, and even in all of biology, so it is very important to pursue understanding of sleep regulation as a path to discovering the function(s) of sleep.

This article discussed three regulated aspects of sleep: circadian timing, slow-wave activity, and the NREM–REM sleep cycle. Circadian timing is the most understood. The function is to synchronize the daily sleep phase with the geophysical cycle, and the feedback is exposure to light. The slow-wave response (EEG delta power) to prior wake is homeostatically regulated, and at least one feedback parameter is adenosine. This information has led to testable hypotheses about a possible function of sleep—the maintenance of brain energy reserves, which remains a productive line of research. NREM–REM cycling is also a regulated component of sleep, with REM sleep being in a homeostatic relationship with NREM sleep rather than with wake. We don’t know what feedback information is used to control this cycling. Identification of such a feedback signal should provide insights into a function of REM sleep.

There are other proposed functions of sleep: for example, synaptic downscaling (Tononi & Cirelli, 2006), glymphatic clearance of wastes (Xie et al., 2013), consolidation of memory (Rasch & Born, 2013), and even non-neural processes such as maintenance of bone health (Everson & Szabo, 2011). Unfortunately, homeostatic regulation, feedback information, and set points have not been identified for any of these putative functions. The discovery of any of these regulatory components would greatly advance our understanding of the functions of sleep.