How do we fall asleep?
- 1 Initiation of sleep
- 2 Circadian cycle
- 3 Borbély model
- 4 Phase response curve (PRC)
- 5 Recursive phase response curve (rPRC)
- 6 Two-component model of sleep in SleepChart
- 7 REM rebound hypothesis
- 8 Sleep-wake flip-flop
- 9 Suprachiasmatic nucleus (SCN)
- 10 Dorsomedial Hypothalamic Nucleus (DMH)
- 11 Ventrolateral Preoptic Nucleus (VLPO)
- 12 Nucleus of the Solitary Tract (NTS)
- 13 Adenosine
- 14 References
Initiation of sleep
We fall asleep when two signals are generated in the brain:
- (H) "too much waking" signal and
- (C) "it is the usual sleep time" signal.
The "too much waking" signal is called the homeostatic signal. While "time to sleep" signal is called the circadian signal. The homeostatic signal is a reflection of network "tiredness". The more you learn, the more you think, the more you process information, the more tired you get mentally. This generates homeostatic sleep propensity. However, homeostatic sleepiness is not enough to fall asleep. You may be dead tired of too much waking or too much learning, but you may still be unable to get a wink. This is where the circadian sleepiness comes in. Circadian sleepiness is maximum during the subjective night period. There is also a mid-day slump in alertness that also has circadian nature. When you are sleepy in both homeostatic and circadian sense, you can finally fall asleep.
The homeostatic signal is generated in the neural networks of the brain. It is associated with slow-wave activity in the EEG. One of its known expressions is an increase in adenosine levels. The effects of adenosine are blocked by caffeine. This is why coffee can temporarily help overcome the homeostatic component of sleepiness. At the same time, caffeine is entirely ineffective against the circadian component. This is why drinking coffee during the subjective night is imprudent and unhealthy. As the waking hours tick on, brain glycogen and ATP reserves are depleted. ATP is degraded to ADP, then AMP, and finally to adenosine. Adenosine then builds up in the brain. This includes a buildup in the basal forebrain (Porkka-Heiskanen 1999), which is the hypothetical source of the neural homeostatic signal. Depletion of the glycogen reserve is also hypothesized to have its own contribution to the homeostatic sleep propensity (Kong et al. 2002). The basal forebrain, which is a cholinergic structure, when active, contributes to the wakefulness and REM sleep. Deactivation of the basal forebrain helps initiate NREM sleep and sleep in general.
The main source of the circadian signal is the suprachiasmatic nucleus (SCN). A set of genes is expressed in a regulatory loop that keeps a 24 hour rhythm of activity. The SCN rhythm can be reset by light, or activity, or other signals (see: Phase response curve (PRC)). The SCN sends most of its fibers to the subparaventricular zone (SPZ) and the dorsomedial hypothalamic nucleus (DMH). One of the hormonal signals produced by the effects of the SCN oscillation is the release of the melatonin from the pineal gland during the subjective night. This led researchers to the idea that melatonin might be a natural help in initiating sleep (given sufficient homeostatic sleepiness).
Integrating homeostatic and circadian signals
The homeostatic signal needs to be integrated with the circadian input. The precise mechanism of the integration is not known, but there are a couple of solid hypotheses on how this might work. The anterior hypothalamus is the presumed site of the integration. The hypothesized integrating nuclei are: the medial preoptic area (MPA), the anterior paraventricular thalamic nucleus (aPVN), and the dorsomedial hypothalamic nucleus (DMH). DMH and MPA send a big bunch of fibers in the direction of the ventrolateral preoptic nucleus (VLPO), which is one of the main brain nuclei responsible for the initiation of sleep.
Sleep control system. See the text for details. See the legend here
Adenosine agonists are also able to activate the VLPO (Scammell et al. 2001). It has been hypothesized that adenosine inhibits anterior hypothalamic and basal forebrain GABAergic neurons that suppress the activity in the VLPO.
The VLPO is thus able to initiate sleep by receiving both the circadian signal from the anterior hypothalamus and the homeostatic signal from endogenous substances (e.g. adenosine) that accumulate in the course of a waking day. The VLPO and its adjacent nuclei are then able to inhibit the histaminergic wake-promoting TMN and other arousal systems in the pons and the midbrain (e.g. LC, DR, LDT, PPT, PeF, vPAG, etc.). Sleep is a direct consequence of the inhibition of the ascending reticular activating system (RAS) which groups those neural structures that keep the cerebral cortex in the waking state. With the depression in the activity of the RAS, we quickly lose interest in demanding intellectual activities. Soon the only thing we can think of is sleep. Once we rest in an undisturbed place, we drift into the dreamland. People who cannot follow their natural body rhythms will often be unable to follow the above scenario.
Neural inhibition of the arousal is also accompanied by a significant drop in ACTH and cortisol, which are chief alertness hormones. Similarly, the levels of serotonin and catecholamines drop, and so does the body temperature. All those processes proceed on parallel tracks and we sleep best when they are all perfectly synchronized. It is awfully easy to put that symphony out of sync by all forms of intervention: excitement before sleep (dopamine), coffee before sleep (homeostat), exercise (adrenaline), etc. Synergistic manipulation also has side effects: sleeping pills, alcohol or marijuana destroy the sleep structure. Even melatonin has its side effects. Sleep is healthiest when all physiological variables change in pre-designed synchrony. This can best be accomplished by following the commandments of one's own body clock.
Not all scientists agree
Dr James M Krueger has championed, for many years, an idea that all advanced neural networks have an inherent ability to enter a sleep state (in particular, cortical columns have this property). A biochemist by education and spirit, Krueger started his investigations from looking for substances that induce sleep. He was inspired by a century old finding that cerebrospinal fluid of sleepy animals contains substances that are able to induce sleep when transferred to otherwise alert animals. Over the last four decades, Krueger has amassed a great body of evidence for the existence of a huge number of sleep regulating substances (SRS) such as adenosine, nitric oxide, TNF, IL-1, GHRH, prostaglanding D2, etc. (Krueger et al. 1999; Krueger et al. 2001). Some of SRSs, like adenosine, build up with mental activity (e.g. as a result of the release of glutamate (Simasko et al. 2005)) and may play a role in sleep homeostasis, while others (e.g. melatonin) are circadian. In his recent publications, Krueger asserts that sleep is a network-emergent phenomenon, and that sleep control nuclei in the brain play only an accessory synchronizing role. Even though the overarching principle may seem to quarrel with the mainstream science of neural sleep control, the body of undisputed facts is overwhelmingly larger than the areas of disagreement. Even though Krueger theories do not seem to explain the computational aspects of sleep, where a neural control of sleep centers seems indispensable, they all align pretty well with the homeostatic aspect of sleep control. In the place where biochemists meet neural network experts and neurophysiologists, we can find the most fruitful field for further exploration of the mysteries of sleep.
The human body clock runs in a cycle of circa 24 hours. That cycle was therefore named a circadian cycle. Understanding the circadian cycle is vital for healthy sleep. My wild guess is that 95% of sleep problems in industrialized nations are caused by the lack of understanding of the circadian cycle, or lack of respect to its power and importance. The cycle is encoded deep in the human genome and cannot be easily changed or overridden. Playing with the circadian cycle may result in long-term health consequences. All cells in the body express various clock genes, however, there is a master clock in the brain that helps synchronize other clocks in the body to run in harmonious synchrony that is vital for health, well-being, longevity, learning, creativity, etc. The master clock is located in the brain and is called the suprachiasmatic nucleus (SCN). Circadian cycles of the SCN result in periodic release of melatonin from the pineal gland. This led to the use of melatonin as a sleep remedy. The popularity of melatonin comes from its natural origins and the possibility of oral administration. However, as melatonin is located downstream of the SCN in the circadian cascade, it does not have the full magic powers of generating complete nighttime circadian states. Even the natural release cycle of melatonin my get misaligned with the sleep-wake cycle in irregular schedules. This limits melatonin applications. It can be used to produce phase shift (e.g. phase advance if taken 1-2 hours before natural bedtime), but it is not the universal sleeping pill as it is often advertised.
An important alertness hormone, cortisol, can be used to map a well-timed circadian cycle. Its levels drop during the first half of sleep, and raise dramatically on waking giving us a sharp waking mind. On the other hand, growth hormone is less dependent on the clock and is released primarily during deep sleep having its hand in the anabolic power of sleep making it important for both the brain and the brawn.
Circadian alertness is partly hormonal and partly neural. The brainstem contains a collection of nuclei know as the reticular activating system. These nuclei, when activated, keep us awake and alert. Those "vigilance nuclei" include the serotonergic raphe nuclei, adrenergic locus ceruleus, parabrachial nuclei, and more. Various lesions to those areas and their connections may result in insomnia or coma.
In 1982, a Hungarian sleep researcher, Alexander A. Borbély published a seminal paper titled "A two process model of sleep regulation". This model has later been described in pretty precise mathematical terms, and is now the mainstay of our understanding how sleep is initiated and how the sleep-wake flip-flop works in healthy sleep in abstraction from the actual neurophysiological interpretation.
In short, Borbely noticed the distinction between the two components of sleepiness: homeostatic sleepiness and circadian sleepiness. Homeostatic sleepiness increases during the day as a result of mental effort. Circadian sleepiness increases at nighttime. Borbely's model argues that for a good night's sleep, you need to go to bed with both components of sleepiness in a high gear. This means that going to sleep early, before your circadian sleepiness kicks in, is a bad idea. You won't be sleepy enough to fall asleep, or your sleep will be shallow and easily interrupted. On the other hand, the model also implies that a premature awakening may clear the homeostatic sleepiness, and we may find it hard to fall back asleep even though the circadian sleepiness ensures we are pretty tired.
An exemplary interpretation of the two process model of sleep for normal sleep and sleep following a sleepless night. Homeostatic sleepiness is denoted as Process S (throughout this article, I use H for mnemonic reasons). Circadian sleepiness is an inverse of Process C. Sleep occurs when C is low and S is high. Additional sleep pressure accumulates after a night without sleep, and the sleep can occur earlier and last longer (it starts at higher homeostatic sleepiness despite slightly lower circadian sleepiness). SWA - slow-wave activity - is a brain wave activity that represents the deepest sleep. TST - total sleep time - is higher after a sleepless night.
During sleep, cortical slow-wave activity (EEG power density range of 0.7 to 4.5 Hz) depends on the duration of prior waking. This is why it is considered a hallmark of homeostatic sleep propensity (Daan et al. 1984). It decreases exponentially after the sleep onset. One of the limitations of the model is that it does not account for NREM-REM exchange, while the homeostatic sleepiness (Process S) might actually increase in the REM phase. Circadian sleepiness correlates with the release of melatonin, but can also be mapped onto core body temperature, or release of other sleep inducing or alertness hormones.
Borbély model in practice
An exemplary interpretation of the two-process model taken from an actual sleep log in SuperMemo. Aqua line represents circadian sleepiness. Green line represents homeostatic alertness (an inverse of the homeostatic sleep propensity). Red line represents overall alertness that is an inverse of overall sleep propensity. Best alertness is achieved when both components of sleepiness are at their lowest.
[find a better graph! esp. one that aligns well with the Wikipedia picture].
In Borbély model, sleep timing is determined by the points in which the curves representing the two processes cross. SuperMemo uses a more intuitive approach, in which both components of sleepiness are integrated heuristically to match the expected course of overall alertness (red line in the graph). Sleep is initiated when the overall alertness drops below a certain level. Sleep may thus be initiated by both components of sleepiness independently, as it may happen in early life, but the timing and duration of sleep will differ for various values of both variables (and the status of the circadian system).
You can "feel" both components of sleep. Homeostatic sleepiness is more likely to be described as feeling "unrefreshed", while circadian sleepiness is more likely to be named "grogginess". In a healthy cycle, you should never see the difference between the two: you wake up fresh, and you get sleepy in the evening when both components of sleepiness kick in making you just "very sleepy". However, if you are jetlagged and groggy, you can feel the unpleasant circadian sleepiness that does not go away and cannot be helped with a nap if your homeostatic sleepiness is too little to fall asleep. On the other hand, after a sleepless night, you may be dead tired and unrefreshed, however, with the morning sunlight you get a new energy to survive yet a couple of hours. As your circadian sleepiness passes by, you may feel homeostatic sleepiness that seems survivable (until the next circadian low hits)(see more: Sleeping against your natural rhythm).
Borbély model and evolution
The homeostatic component of sleep may simply be an unavoidable cost of the evolving neural networks. To prevent catastrophic forgetting, neural networks need to implement an overload protection. That protection is the homeostatic drive to sleep. We do not know how much of that protection is a natural consequence of the network overload, and how much of it is an added effort by the brain to prevent further overload. For example, we can improve cognitive function with the help of caffeine by blocking the adenosine-based component of the homeostatic sleep drive. This proves that the brain provides a degree of network overload protection. The overload will result in progressive decline in recall and memory consolidation in the waking day.
The circadian component evolved long before the neural function of sleep was established. However, it was convenient for the organism to do its neural housekeeping at opportune times. For example, for human hunters and gatherers, night is a time of inaction. This is why hooking up sleep to the night time period made an evolutionary sense. Other animals may have made different choices, however, the circadian cycle is always a good hint on the optimum timing for neural optimization.
Borbély's two-process model has been extended by an additional process W that represents sleep inertia. The basis for that model was self-rated reports of sleepiness (Akerstedt and Folkard 1990). The model used in SuperMemo does not include the sleep inertia factor as it is primarily targeted at studying free running sleep.
Phase response curve (PRC)
Phase response curve (PRC) represents a function that tells us how much a phase of an oscillator shifts in response to selected stimuli depending on the timing of these stimuli. PRCs can be used to study circadian rhythms as well as other biological, physical or electronic system (e.g. the heartbeat).
For example, PRC for light may tell us that applying a green light pulse of a given intensity 1 hour before sleep pushes the circadian cycle forward by 10 minutes (phase delay), while a blue light of a higher density might push the same cycle by 25 minutes. During the subjective night, there is a dead zone when light does not produce shifts in the circadian cycle.
There are many PRCs for different stimuli such as exercise, stress hormones (e.g. cortisol), melatonin, and other stimuli. The crossover time between the delay side and the advance side of the PRC for light is near the core body temperature minimum. The sleep control system seems most sensitive to shorter wavelengths of visible light in suppressing the release of melatonin (Brainard et al. 2001).
Stimuli that cannot shift the cycle or have a negligible impact (e.g. cup of warm milk), can also have a PRC plotted. However, that PRC will be a straight horizontal line running along the phase shift of zero. SleepChart uses an algorithm for plotting the so-called recursive PRC, in which the degree of phase shift is measured in reference to the actual position of sleep episodes in free running sleep without differentiating between the actual causes of the shift. rPRCs differ between people. They also change in response to lifestyle changes.
Changing the length of the circadian period
The existence of the PRC implies that the length of the clock period is under our control. If we apply zeitgebers early or late enough we can affect larger phase shifts that can lengthen or shorten the period of the cycle. Everyone can prove it all to himself or herself with relatively simple measures (e.g. bright lights in the late evening to shift the phase forward, or early morning exercise to shift the phase back). This is why PRCs are very important when treating phase-shift disorders. This is also why lifestyle determines phase shifts and possible sleep problems. This is why the modern lifestyle based on the use of electricity causes an epidemic of DSPS in the young learning generation. Recently, the fact of the adaptability of the body clock period was demonstrated with investigations into a possibility of astronauts adapting to a Martian day (Scheer et al. 2007). This was also demonstrated earlier in rats by various "lifestyle" changes (e.g. wheel restriction increases the circadian clock period).
Phase-shifting neural inputs
Our master clock, the SCN, is affected by 3 major zeitgeber inputs that allow of a phase shifts:
- retinohypothalamic tract (RHT) carries light resetting stimuli and acts via the NMDA receptors. This input bypasses cognitive vision
- intergeniculate leaflet (IGL) projects to the SCN via the geniculo-hypothalamic tract (GHT) and carries motor resetting stimuli
- the raphe nuclei provide serotonergic input that is hypothesized to modulate some aspects of the circadian cycle and might be involved in changes to the circadian cycle in affective disorders. Lesions in this pathway or decrease in serotonin lengthen the active phase in constant darkness without affecting the circadian period
Recursive phase response curve (rPRC)
To study the phase response, scientists need expensive laboratory setups and time-consuming research procedures. However, a simple computational trick makes it also possible to see the effects of phase shifting stimuli in SleepChart without the use of a sleep lab.
SleepChart implements a concept of the Recursive Phase Response Curve (rPRC). The curve is recursive because it is first obtained by computing the impact of phase shifts in sleep episodes in relation to the circadian acrophase computed using statistical methods. Once the first approximation of rPRC is obtained, it can be used to produce a better approximation of the middle of the subjective night line that is then used to generate a better approximation of the rPRC. A few iterations of such a process are sufficient to produce the best fit of the rPRC that corresponds well with the actual sleep data. SuperMemo uses acrophase estimates by using a fixed rPRC that roughly corresponds with rPRCs obtained with SleepChart. Whereas a typical PRC employed in chronobiology maps the response of the sleep system to a single stimulus (e.g. light, exercise, melatonin, or various chemical agents), rPRC is the resultant of all natural sleep delaying factors (incl. light, activity, stress, etc.). It can also be interpreted as a PRC, in which the waking activity forms the input to the free running sleep system. Unlike a PRC which responds to a shifting factor, rPRC responds to the evening phase shift caused by the same factor. As such, rPRC is not a de facto PRC, and all departures from the free running condition invalidate the computation. The main advantage of rPRC is that it can be derived from sleep data without collecting blood samples, saliva samples, or taking core body temperature measurements. This way, SuperMemo can correlate learning with sleep models that use only plain sleep log data as input.
In the presented graphs, Sleep delay (h) stands for the bedtime delay and equals the difference between the actual bedtime and the bedtime as computed by SleepChart from the prior history of sleep. As the measurements refer to free running sleep, little phase advance data is available due to the natural way of waking. The causes of sleep delay may include light, social interaction, stress, a conscious decision to delay sleep, exercise, ingestion of caffeine, medication, etc.
Phase shift (h) stands for a phase shift and equals the difference between two exponentially weighted waking hour averages on two successive days: the day on which the bedtime delay occurred and the following day. Instead of the bedtime hours, waking hours were compared as these are less affected by the homeostatic shift caused by the actual delay thus representing a truer reflection of the actual phase shift.
The flattening of the curve (as compared with a typical PRC) is caused by the recursive reference to actual sleep data, which results from the fact that plotting the circadian acrophase by SleepChart is an approximation based on the same sleep measurements. As a result, polynomial approximation shows a slight increase in phase shifts with increasing delay, which is not the case in typical PRC plots. The deviation of the bedtime hour from the optimum bedtime may result from either environmental delay factors or from the approximation error resulting from heuristic procedures used to plot the circadian function, while sleep onset usually occurs naturally at optimum physiological time. The inherent asymmetry of the graph comes from the fact that earlier bedtime is nearly always natural, while delayed bedtime may be natural or forced. It is the forced bedtime delay that is the main source of phase shifts in free running sleep.
Recursive PRC in DSPS
The graph presented below implies that, in the case considered, delaying sleep by four hours results in a shift of sleep phase equal to 1.4 hours (which seems to be close the maximum shift possible). Phase advance would require a natural onset of sleep that preceded the optimum retirement time by as much as 6 hours. Going to sleep at the optimum hour results in the natural daily delay, in this particular case 1.0 hour, typical of DSPS disorders or conditions of isolation from zeitgebers (e.g. constant lighting).
Recursive PRC and phase advance
Delaying sleep should always be avoided (except for cases where it is used as a form of chronotherapy). The next graph shows how sleep delays can actually advance the sleep phase. This is a reverse situation to the described earlier phase delay caused by an evening melatonin overdose. Where the wakefulness intrudes past the circadian acrophase, which follows the stationary point of the rPRC, phase delays decrease rapidly up to a point where further delay in sleep will push the phase backwards. Naturally, this "method" of phase manipulation is particularly unhealthy as it implies arousal in the middle of the subjective night (see: Health effects of shift-work and jetlag).
Recursive PRC showing phase advanced that can be caused by either (1) bedtime delays of above 5 hours, or (2) bedtime advances of more than 2 hours.
In the presented exemplary graph we can read the following:
- going to sleep at one's natural bedtime causes a 1.3 hour phase shift resulting in a circadian cycle period of 25.3 hours
- phase shifts are eliminated only when going to sleep 2 hours ahead of the natural bedtime
- phase advances of 2 hours are possible, but require an unnatural early bedtime (e.g. as induced with intense exercise such as running a marathon)
- phase delays of more than 2 hours are unlikely
- a delay in bedtime larger than 5 hours can result in an actual phase advance due to the impact of arousal on the morning end of the subjective night. Obviously, this should not be considered a "cure" to phase delays because such serious deviation will seriously affect the quality of sleep and produce major ripples in the control of the circadian cycle
If you run your sleep free and have a sufficiently large set of data (e.g. several months of a sleep log), you can generate your own rPRC data with File : Export : Recursive PRC in SleepChart (you need SuperMemo 15 or later).
Recursive PRC in polyphasic sleep
It is possible to feed SleepChart with data obtained from "Uberman experiments". Obviously, the mere departure from free-running condition makes the outcome hard to interpret. Even the recursive nature of the procedure used to obtain rPRC cannot effectively cope with the lack of the leading circadian crest. With all that in mind, it is still interesting to peek at "Uberman rPRC" as it nicely reflects the chaotic nature of the sleep system subjected to a polyphasic experiment.
A polyphasic sleeper pushes his sleep phase back and forth largely at random. That can only result in a chaos and complete asynchrony of all neural, endocrinal and biochemical processes depending on the circadian component of the sleep cycle. One might expect serious health consequences of such a chaotic input to the system; however, natural defense mechanisms make life quite miserable for those who attempt a struggle against the natural sleep cycle. As a result, those who attempt polyphasic sleep are doomed to drop out sooner or later.
Chaotic phase-shifting input
Chaotic signals sent to the phase-shifting inputs as seen, for example, in polyphasic sleep, may have hard to predict negative consequences for the sleep control system. The risk is not fully known and hard to estimate. It could include in the order of decreasing likelihood:
- desensitization to signals sent by the sleep control system
- long-term instability in the sleep control system
- damage to nerve cells involved in the control of the circadian cycle
The first possibility can actually be observed in shift-workers and people running a constant battle with sleep deprivation. In those individuals, the concept of "refreshed mind" and "refreshing sleep" becomes hazy, and one can observe an increased tolerance to permanent degree of tiredness coming from insufficient sleep or sleep in a wrong circadian phase. In other words, a degree of fatigue becomes a norm.
Instability of the sleep control system is also observed in shift-workers. I am not sure if shift-induced instabilities can become chronic or are fully reversible in a relatively short time. Even in a perfectly tuned sleep control system, minor rhythm perturbations, such as a switch to the DST, can produce regulatory ripples lasting for days. Larger perturbations might, in theory, result in uncoupling of master and slave oscillators with a particularly slow return to a fully stabilized control. Perhaps this kind of uncoupling is the primary factor that underlies a myriad of disorders that plague shift-workers in the long-term.
Two-component model of sleep in SleepChart
SuperMemo uses a two-component sleep model inspired by the publications of Alexander A. Borbély and Peter Achermann. Unlike other models, SuperMemo uses user's sleep data to predict the homeostatic and circadian status of overall sleep propensity. This model is helpful in choosing the optimum time for learning on a given day (given a particular history of sleep). It can also help planning the optimum bedtime in cases where the sleep pattern is highly irregular. The model does not predicate on the timing and duration of NREM and REM sleep episodes.
The model is tuned to fit typical SleepChart data logs. However, there are individual genetic differences that affect the length of the circadian cycle, steepness of the homeostatic decline in alertness, sleep length preferences, sleep architecture, spectral properties of sleep, fragmentation of sleep, etc. This model is limited in accounting for these variables. If you are sleepy against the simulations based on the model, you can probably trust your own instincts better. If you feel alert against the simulations based on the model, you can certainly get down to learning and ignore predictions of the model. Moreover, sleep patterns are a good measure of your sleep control systems only if they are not artificially disturbed (e.g. by forcefully delaying sleep, using alarm clock, using medication, etc.). In other words, if you are not free running your sleep, the presented model may fail to map your circadian rhythms correctly. You can mark blocks as artificially shortened or delayed (Forced awakening and Delayed retirement on the context menu available with a right-click). However, marked blocks will have a limited effect as there is no way of knowing the degree of the cut into the sleeping patterns, and, consequently, knowing the resulting perturbation in the sleep control system produced by artificially modified sleep.
In Borbély model, the timing of sleep is determined by the points in which the curves representing the homeostatic and circadian processes cross. SuperMemo uses a simpler, but more intuitive approach, in which both components of sleepiness are integrated into an overall alertness level (red line in the graph). The advantage of that approach is the possibility of instant feedback from an actual learning process, where the level of memory recall is supposed to correlate directly with the level of alertness determined by the model. The formula for integrating the two components of sleep into overall alertness was chosen heuristically with the help of alertness data gathered in SuperMemo. The purpose of the integration was to achieve the best possible match of the predicted alertness in the model with the average recall level in SuperMemo. As it has been shown earlier, both homeostatic sleepiness and circadian sleepiness affect the grades in SuperMemo, however, only a combined effect of both components provides a good match with the changes of recall for different combinations of homeostatic and circadian sleepiness. In the model used in SuperMemo, sleep is initiated when the overall alertness drops below a certain level. Sleep may thus be initiated by both components of sleepiness independently, but the timing and duration of sleep will differ for various combinations of changes in the homeostatic and circadian sleep propensity. Despite using a different approach to determining the sleep onset, predictions of the model fit the actual sleep log data pretty well in free running condition in cases studied.
To see the predictions of the model for your own sleep data for any given day, make sure you have your sleep log filled out for recent days in SleepChart, and shift-click the day in question in the sleep log.
Two-component sleep model in SuperMemo: The horizontal axis represents time. Blue blocks show the actual sleep episodes. Aqua line shows the 24h circadian sleep drive with a mid-day siesta hump. Green line is an inverse of the homeostatic sleep drive and can be interpreted as homeostatic alertness, which declines exponentially during wakefulness and is quickly restored by slow-wave sleep (for simplicity, as in Borbely model, the entire sleep block is assumed to have a contribution proportional to its length, as the SleepChart model does not account for sleep stages). Yellow vertical lines show the prediction of the circadian acrophase (circadian middle-of-the-night peak). Acrophase computations are done with the help of a phase response curve model (as opposed to a statistical model used in earlier versions of SleepChart). Red line shows the resultant alertness (peaks are best for learning, valleys are best for sleep). For example, Alertness on Oct 1, 2008 at 7:43 was predicted to be at 59% of the maximum but would keep increasing fast in the first 2 hours of wakefulness (a typical symptom of a night sleep that is terminated too early). The picture shows two peaks in alertness on Oct 1, 2008, at 9 am and at 7 pm. Those periods would likely be best suited for learning on that day.
To see a more accurate presentation of your own homeostatic and circadian alertness in SuperMemo, see the Alertness tab in SleepChart.
REM rebound hypothesis
Researchers know that Borbely's two-process model is not complete and does not explain all known properties of sleep, nor even all possible sleep patterns (e.g. various napping habits, newborn sleep, irregular sleep patterns, sleep in psychiatric disorders, etc.). There have been numerous attempts to expand the model by new variables that may show up in specific circumstances (e.g. adding noise to simulate a sleep-wake pattern in autistic children, ultradian dynamics to model NREM-REM occurrence, adding the impact of light intensity, etc.). Borbely and Achermann keep investigating various aspects of sleep that would help make the model more complete. One of their investigative targets is a REM sleep rebound following a REM sleep deprivation. It has already been discovered long ago that REM sleep deprivation reduces alpha activity, waking, and NREM sleep. These are clear signs of REM homeostatic compensation (Borbely et al. 1990, Brunner et al. 1993). It has been proposed that increases in muscle atonia episodes in NREM (MAN) be considered as markers of an increase in REM sleep pressure (Achermann et al. 2002).
For many years now, I have observed an unusual phenomenon in SleepChart logs that I could not explain with the two-process model. In people with irregular sleep, late naps are often exceedingly long and unrefreshing. Those long naps clear up the homeostatic component of sleep propensity, and often result in later bedtimes. In some extreme cases, this can lead to confusion about the optimum timing of sleep. The affected person will nap long enough to lose the sense of the timing of his or her own subjective night. A graph below demonstrates such a classic occurrence.
In the exemplary sleep log above, a middle-aged woman working from home and suffering from a delayed sleep phase syndrome shows a clear and pretty regular progression of the sleep phase from a bedtime at 2 am to sleeping past midday. The lady claims to suffer from irregular sleep, daytime tiredness, and never knowing when to go to sleep to get a "good night's rest". On Sep 24, due to feeling tired, she went for a nap at 6:30 am. This nap unexpectedly lasted 3.5 hours and produced the impression that no more sleep was required on that day. Despite some tiredness, the lady did not go to bed that evening even though the chart clearly says that it was the period of her subjective night and she should retire. After a particularly tiring evening and night, the lady went to sleep 3:30 am on the assumption this was her "night sleep". That sleep was 6 hours long and refreshing enough to "impersonate" the night sleep. This completed the role reversal between the night and siesta periods. The two circadian lows have been swapped in the sleeper's mind. This swap is then reflected in retirement rituals, expectations, and other habits that can perpetuate the reversal for a few days despite a potentially highly unrefreshing sleep. A night-vs-siesta reversal is not stable though. On Sep 25, in the period of the subjective night, a short nap was taken which seemed particularly refreshing. Still the refreshing impression dissipated fast and the third "night at siesta time" followed. On Sep 26, sufficient sleep debt accumulated leading to a "nap" that suddenly turned into 8 hours of deep and refreshing sleep. The sleep pattern flipped back to the norm after 3 prolonged napping episodes. This role reversal cannot be explained with the two-process sleep model. Nor can it be explained with the model employed in SleepChart. Those three outlier naps taken past the siesta time should rather be shorter due to the fact they were not enhanced by the circadian siesta dip. Those naps were also occurring too early to capitalize on the nighttime circadian low. In other words, those "kinky" naps, despite missing the circadian component of sleep, lasted unusually long.
Having seen those kinks in sleep patterns dozens of times, I came to believe that the 2-process model of sleep propensity needs to be extended by a third component. However, it was hard to come up with a sensible hypothesis that would plausibly fit with what we know about the function of sleep and its evolution. A big clue came from interviews with people affected by kinky naps. It appears that those long naps are very often triggered by consumption of alcohol or, in some cases, smoking marijuana. If the timing of alcohol or cannabis administration aligned with the late waking hours, shortly before the subjective night, the kinky nap could follow on the next day. In addition, those naps are preceded by a particularly strong feeling of being unrefreshed in the morning, which is a frequent case in alcohol or cannabis abuse (as much as in the application of sleeping pills or even melatonin). As both substances are known to reduce the proportion of sleep spent in REM, I hypothesized that it is the REM-sleep deficit that might be causing the said sleep perturbations.
I have also documented cases were kinky naps followed a healthy and refreshing night sleep that did not involve alcohol, cannabis nor other substances affecting sleep. Those remaining cases also had another common factor: a substantial one-time delay in optimum bedtime and the resulting sleep phase shift. This would agree with the REM-deficit hypothesis. If sleep is delayed past the circadian REM peak, it is also known to be less REM-rich.
Finally, those kinky naps, unlike the healthy well-timed naps were reported to be dream rich. This could also indicate that they might be involved in REM compensatory function.
If the REM-deficit hypothesis was to be right, we would need to always consider separate homeostatic REM and NREM sleep propensity. In healthy sleep, the REM component might be hard to notice. Some researchers hypothesize that homeostatic REM drive depends on the preceding NREM sleep. If so, homeostatically, healthy night would produce no REM deficit, while waking activity would only produce homeostatic NREM sleepiness.
How could REM deficit produce those prolonged naps? There are some indications that REM sleep can also produce an increase in demand for NREM sleep. Thus those two, functionally vital phases of sleep, could produce a mutually amplifying cycle that would run its course until the demand for both sleep components was fulfilled. Why would REM sleep increase towards the end of normal night sleep? Some of that increase is circadian, some of it might come from the fact that homeostatic NREM sleep demand is satisfied faster. The biological explanation of sleep terminated with REM is difficult, esp. in the light of Buzsaki model of hippocampal "training" in REM. Waking up with a clean slate seems biologically more advantageous. Perhaps that last REM period is responsible for creative breakthroughs of the early morning? Only a detailed mathematical modelling and comparisons with actual sleep cycle measurements could answer the questions about the homeostatic interplay between NREM and REM sleep.
Three components of sleep propensity
SleepChart cannot easily verify the nature of the REM-deficit hypothesis. Not only are sleep stages missing from its logs, detecting REM sleep is not practicable in home conditions amongst users of SuperMemo or SleepChart. However, the third variable needed to explain kinky naps in sleep logs, which I will call the RD variable (for REM deficit), could possibly be included in the two-process model in hope of mathematically explaining the impact of kinks on the estimated sleep phase. As mentioned earlier, those kinky naps do not need alcohol or other REM-suppressing factors, sleep blocks marked as Delayed retirement often cause similar effects due to a wrong phase alignment vs. the circadian REM peak. Once sleep misalignments are explained successfully with the RD variable, it would be up to sleep labs to verify the model using EEG measurements. The interaction between the RD variable and the other two sleep variables (H and C) is not straightforward. For example, high RD would not suffice to initiate sleep, as it is not possible to initiate sleep without an appropriate combination of H and C. High RD and high C might also be insufficient (as it is indicated by sleep logs where sleep is pretty short in the nights that follow kinky naps due to the low H). However, high RD and high H could initiate fully blown sleep and result in kinky naps with possible negative consequences for the subsequent night sleep (low H), and sleep phase. At the moment of writing, I am still now sure how the sleep phase is affected, however, I am pretty sure it is. For example, in the example presented earlier, the sleep phase seems to have been shifted back by a few hours, however, it could as well be caused by the deficiency of the model employed in SleepChart (precisely due to the missing RD variable). The need for both high H and high RD to initiate sleep for low C seems consistent with current research on the mutual interaction between NREM and REM sleep stages where one increases the demand for the other.
The sleep-wake flip-flop is a system of two sets of brain nuclei that produce a rapid switch from sleep to waking, and vice versa. One set of nuclei is responsible for inducing sleep and inhibiting the arousal centers, while the other set acts in the opposite way. Both sets inhibit one another. This means that when there is time to sleep, the sleep centers take an upper hand and turn off the wake centers. Later on, in the morning, the wake centers take control and turn off the sleep centers. This sleep-wake flip-flop is constructed in such a way that the transitions from sleep to wake and back are pretty fast and thorough. In a healthy sleep cycle, we should be half-awake only for a very short time before sleep, and perhaps a little while longer in the morning. Unfortunately this does not mean that we can switch the flip-flop wherever we wish. It also does not imply that we won't feel tired before sleep. Homeostatic increase in sleepiness is a natural process and it proceeds throughout the waking period. It is only the transition from wake to sleep that is fast and the time when homeostatic sleepiness meets a sufficient degree of circadian sleepiness. The sleep-wake flip-flop is stabilized by orexin neurons. As demonstrated by Siegel, the level of orexins (also called hypocretins) is not related to the circadian cycle but to a particular behavior. During a waking activity, e.g. during exercise, the level of orexins may remain high thus preventing the switch in the sleep-wake flip-flop. When the orexin stabilizer is off, narcolepsy enters the picture and the flip-flop becomes unstable causing multiple sleep episodes in a single day in hard to predict circumstances.
The most important components of the sleep-wake flip-flop are:
- the ventrolateral preoptic nucleus (VLPO) on the sleep side, and
- the tuberomammillary nucleus (TMN), locus coeruleus (LC) and dorsal raphe nucleus DR on the arousal side.
Once sleep is initiated, another flip-flop starts operating: the one that is responsible for transitions between NREM and REM sleep.
Suprachiasmatic nucleus (SCN)
Human brain harbors a clock that runs in a cycle that is slightly longer than 24 hours. That clock is called the suprachiasmatic nucleus (SCN) and is located in the anteroventral hypothalamus. The SCN is made of two groups of neurons (10,000 each, 0.25 mm3) situated bilaterally just above the optic chiasm. The SCN is slightly more elongated in women, and there is a marked difference in VIP expressing neurons between sexes (up to twice as many in males)(Swaab et al. 1990). Homosexual men have larger SCNs and twice the number of VP expressing neurons than heterosexual men (Swaab and Hofman 1990). Incidentally, I am pretty sure that this difference is not by choice and it cannot be remedied with self-discipline or by prayer.
In 1972, the SCN has been identified as the body's master clock that can run without environmental cues and receives resetting inputs from the retina. Clock genes in the SCN are responsible for a circadian cycle of gene expression that determines the output from the SCN. The neurons in the SCN express the cycle that finds its reflection in signals that travel from the SCN to other brain nuclei and the rest of the body in various neural and hormonal forms. If we surgically damage the SCN, the circadian cycle wanes or disappears. It can be restored with a transplant of SCN cells.
The SCN signal is most active during the subjective day, esp. in the evening hours. It is the weakest during the subjective night, esp. in the early morning when the body temperature reaches its minimum. If you ever tried to sleep polyphasically, it is the SCN that will bother you and make you crave the core sleep and make you oversleep during the subjective night time. The SCN controls alertness, attention, release of hormones, body temperature, melatonin secretion, feeding, and more. Most of the output from the SCN flows to the subparaventricular zone (SPZ) and the dorsomedial nucleus of the hypothalamus (DMH). Neurons in the dorsal SPZ (dSPZ) affect the circadian rhythm of the body temperature, while those in ventral SPZ (vSPZ) are running the wake and sleep cycle. vSPZ in turn commands the inputs to DMH which is the chief command center for waking behavior, motor activities, cortisol cycles, feeding, etc. DMH affects sleep promoting VLPO and the wake promoting LHA (lateral hypothalamus). Lesions to VLPO and LHA can produce loss of sleep or insurmountable sleepiness respectively.
This central positioning of the SCN and the DMH at the crossroads of the most essential and influential neural pathways controlling behavior is a powerful demonstration of how a tiny group of a few thousand neurons exerts a powerful influence on what we do as active feeding and surviving organisms. This should remind everyone that sleep hygiene is essential for the proper function of this tiny structure in the human brain. Disrupting circadian cycles with alarm clocks, shiftwork and the like can lead to a whole volley of physical and mental disorders. For a thorough review of the interaction between the SCN, the DMH and the rest of the body see Dr Clifford B. Saper "Hypothalamic regulation of sleep and circadian rhythms" (Saper et al. 2005). Interestingly, Dr Saper hypothesizes that it is the DMH that integrates the circadian resetting stimuli such as exercise or social interaction. In rodents, DMH can also be reset by the availability of foods or even the temperature. It is unlikely though that you will be able to combat jetlag or adapt to any shift-work pattern with the help of zeitgebers such as food or temperature.
SCN oscillates with a period slightly longer than 24 hours. To adapt to the 24h world, the oscillation needs to be reset daily to match the daylight cycle of the Earth. The resetting is done with the help of zeitgebers ("time givers") such as light, exercise, feeding, etc. The most important zeitgeber is light. Light signals are received by glutamatergic melanopsin-expressing retinal ganglion cells in the retina (pRGCs). From there, they are transmitted to the SCN via the retinothypothalamic tract (RHT). The impact of light signals and other zeitgebers on the circadian phase is described by the so-called phase response curve (PRC). Most importantly, morning light signal helps reset the cycle. The circadian period gets shortened to match the 24h daylight cycle. With the help of zeitgebers, the oscillator with a slightly longer period is brought back to synchrony with the daylight by a minor SCN-mediated reset. This provides for a stable oscillation. People who cannot effectively cue their oscillators suffer from phase-shift disorders. People suffering from DSPS could experiment with light dimmers, toning down their schedules in the evening, properly timed exercise and bright light in the morning. People with ASPS should use opposite measures (e.g. 3000 lux light in the evening). In addition to light, the SCN is affected by activity. Locomotor activity affects the SCN by activating NPY-containing neurons in the intergeniculate leaflet (IFL) and serotoninergic neurons in the median raphe nucleus (MRN). This is why exercise and social interaction act as powerful zeitgebers.
The neural symphony commanded by the SCN goes awry when we use artificial lighting or do exciting evening activities such as watching TV, surfing the net, playing computer games, reading, etc. It has been hypothesized that light (as well as other stimuli) may affect the SCN in two different ways during the subjective night. Short light pulses simply change the expression of clock genes and result in phase shifts along the PRC. However, constant lighting may result in uncoupling between the SCN neurons and the downstream nuclei affected by the SCN resulting in dangerous arrhythmicity (Ohta et al. 2005). Continuous disruptions to circadian cycles as seen in shiftwork or jetlag may lead to a gradual mental decline as indicated by research in rodents (Ree et al. 1985). Circadian changes associated with aging and Alzheimer's can be correlated with loss of cells in the SCN or changes in its inputs. Vasopressin-expressing cells are particularly prominent in their decline in Alzheimer's. All forms of artificial control of sleep cycles, including the use of alarm clocks, can affect the health of those few precious neurons.
The SCN sends projections to the dorsal PVH (parvicellular paraventricular nucleus) whose neurons project to sympathetic preganglionic neurons in the spinal cord that in turn affect the pineal gland and the release of melatonin. This tells us that melatonin, which is often advertized as a "natural sleeping pill" is produced by the pineal gland downstream from the SCN control. This is why it cannot be considered a central factor controlling circadian sleepiness. Melatonin does produce phase shifts along its unique melatonin PRC(see picture in the PRC section). It is possible that this effect is caused by a direct impact of the melatonin on the SCN. However, early sleep will also result in earlier waking and this will also have a phase shifting effect.
Dorsomedial Hypothalamic Nucleus (DMH)
Dr Saper and colleagues demonstrated that excitotoxic lesions to the dorsomedial nucleus of the hypothalamus (DMH) in rats cause a major impairment to circadian rhythms (Saper et al. 2003). As lesioned animals sleep more, it was suggested that the impact of DMH is predominantly activating even though other explanations of the available findings are also imaginable. It appears that a great deal of output from the SCN travels via the subparaventricular zone (SPZ) to the DMH and only then, via inhibitory tracts, to the VLPO that is responsible for the initiation of sleep. The DMH also projects to the lateral hypothalamic area (LHA) that contains wake-promoting orexin neurons. It has been hypothesized that the DMH might be in the center of control of various variables that change along the circadian cycle such arousal, feeding, locomotor activity, cortisol levels, body temperature, melatonin, etc. Restricted feeding synchronizes circadian rhythms of the DMH so that the highest c-Fos expression and locomotor activity coincide with mealtimes (Saper et al. 2006). As most of the input arrives to the DMH via SPZ, it is important to note that dorsal and ventral portions of the SPZ seem to play different functions. Lesions to the dSPZ reduce circadian rhythms of body temperature, while it is the vSPZ that seems to control sleep-wake cycles and locomotor activity (Saper at al. 2001). The dSPZ controls body temperature via the medial preoptic area (MPO) that includes the median preoptic and ventromedial preoptic nuclei. The DMH is affected by the hormones controlling the appetite, ghrelin and leptin, via the ventromedial nucleus (VMH) and the arcuate nucleus (ARC). VMH enhances lipolysis in adipose tissue and decreases feeding. Dr Saper hypothesized that the DMH may serve as a secondary circadian control center that would enable entrainment of the rhythms to the availability of food. However, from the standpoint of control systems, it would seem biologically more sensible to phase-shift the SCN rather than to employ a second asynchronous or phase-locked oscillator. In humans, it is very hard to influence the circadian cycle in any way other than via a minor phase-shift with the use of various zeitgebers, of which, food is a very weak one. It therefore seems highly unlikely that shiftworkers or jetlagged travellers could tangibly benefit from changes to the timing of their diet. Differences between rats and humans cannot, naturally, be excluded. Nevertheless, the DMH is definitely a very interesting further research target.
Ventrolateral Preoptic Nucleus (VLPO)
The ventral lateral preoptic nucleus (VLPO) is one of the chief brain centers needed to initiate sleep and to maintain slow-wave sleep. Lesions in this area halve the amount of sleep, and result in insomnia combined with persistent tiredness. Both NREM and REM can be affected depending on the type of the lesion. For its role, the VLPO is often called a "sleep switch". In both nocturnal and diurnal animals, the SCN is active during the period of daylight, while the VLPO is primarily active during sleep. Once the VLPO is on, it is believed to maintain inhibition of the monoaminergic and cholinergic excitatory systems that keep the brain cortex "awake". Those VLPO projections go to the tuberomammillary nucleus (TMN) (histamine), lateral hypothalamus-perifornical region (LHA/PF) (orexin), ventral periaqueductal grey (vPAG) (dopamine), locus ceruleus (LC) (noradrenaline), parabrachial nucleus (PBN) and dorsal raphe (DR) (serotonin), lateral tegmentum (LDT) (acetylcholine) and the pedunculopontine tegmental nucleus (PPT) (acetylcholine). The inhibitions enacted by the VLPO are mediated by GABAergic neurons as well as by galaninergic inputs to the histaminergic tuberomammillary nucleus (TMN). Inhibition of the TMN and other alertness nuclei results in a drop in alertness hormones and a drop in cortical activation causing drowsiness. A hypothesis says that separate populations of the VLPO might be responsible for expressing circadian aspects of NREM and REM sleep. A subset of VLPO cells is able to stimulate cholinergic neurons in the LDT and PPT. This contributes to inducing REM bursts that activate the cortex without wakefulness during REM sleep.
The VLPO receives its circadian signal input from the SCN (the main body clock) via the dorsomedial nucleus of the hypothamalus (DMH), which is the other brain clock that is usually synchronized with the SCN. The VLPO neurons do not build up a homeostatic need for sleep, however, some homeostatic mechanisms, such as the intracellular build-up of adenosine, may inhibit aminergic or cholinergic wake centers and thus activate the VLPO. For example, infusion of adenosine agonists into the basal forebrain increases both NREM and REM sleep (Satoh et al. 1999) and increases c-Fos in the VLPO (Scammell et al. 2001). In sleep deprivation, the activity in the VLPO is not much higher than in ordinary waking. This low level of activity persists until the bedtime. Once the sleep begins, VLPO neuron firing rate may double in conditions of sleep deprivation (Saper et al. 2005). This indicates that even though the VLPO does not build up homeostatic sleep propensity, it is impacted by the homeostatic mechanisms in the end. This also indicates that the VLPO is located downstream the circadian and homeostatic signal integrator. Aminergic arousal nuclei such as the TMN, LC and the raphe form a part of the sleep wake switch. That switch is stabilized by orexin/hypocretin cells from the lateral hypothalamus-perifornical region (LHA/PF), esp. during motor activities or feeding. The arousal can thus be maintained uninterrupted despite competing inhibitory influences. The arousal nuclei inhibit the VLPO in waking as much as the VLPO inhibits them back in sleep. Scientists believe that this mutual inhibition forms a classic unbalanced flip-flop with sharp state transitions. This is what helps us fall asleep fast, and wake up fast, spending minimum time in transition, and maximum time in the desired states: alertness or deep sleep.
For more on the place of the VLPO in the sleep control system see Figure 39 in Functional Anatomy of the Hypothalamus and Pituitary.
For a thorough review of the role of various sleep and wake centers, see Saper's "Hypothalamic regulation of sleep and circadian rhythms" (Saper et al. 2005).
Nucleus of the Solitary Tract (NTS)
This section speaks more of fads and fashions in science than of the actual involvement of the nucleus of the solitary tract (NTS) in sleep. I heard of the importance of the NTS for sleep in the early 1980s during my college years when studying biology. Some time later, I added a couple of items on the NTS to SuperMemo to consolidate that knowledge for years to come. When writing my Good sleep, good learning, good life article in 2000, I still mentioned the NTS and how rocking babies to sleep might work even though I knew that destruction of NTS does not lead to insomnia, which should be a big clue. The NTS seems to be more involved in processing signals received from the gut. These signals play only a minor part in sleep control. In the end, I fell victim to the same old affliction that pesters science since its inception. Sometimes it takes the old generation of scientists to die out for a new idea to take hold. Old knowledge makes us more conservative, because not knowing makes us seek answers while knowing makes us passive even if our answers are wrong. Once you believe you know all the answers, there is less pressure to investigate. In the end, many other brain centers play a role comparable to that of the NTS. After all, the brain is a highly connected structure and few things happening in one corner of the central nervous system have no bearing on events in other corners. Consequently, activation of nearly all major nuclei will have an arousing or inhibiting effect within the reticular activating system, which has also been for years a mainstay of our thinking about arousal. At the same time, back in 2000, I hardly mentioned the VLPO, as it was perhaps not fashionable enough. A similar situation, we may or may not face with the SLD, which has emerged as an alternative to the well-established PPN/LDT REM on system. Even the DMH might not be immune to fashions. Like the NTS, it is also involved in feeding behaviors. Perhaps, in a decade, this article will warrant a complete rewrite with a great deal of old fads gone. Equally well, in the era where all new findings in science are available at our fingertips, and we can easily communicate via e-mail and other means, we will all show a lesser tendency to swim with the crowd. More importantly, new investigative technologies are likely to open new areas that might still be subject to fads, while the subject matter discussed in this article will gel out and solidify.
Incidentally, the pain of fashions was once the main factor that pushed me away from peer-review writing to blogging. Writing the presented article was the acme of fun. The article was written using incremental writing, and polished collaboratively as a wiki. Being part of a commercial company, I am not subject to publish-or-perish pressures. This is a precious freedom. Back in 1992, with Dr Gorzelanczyk we studied the literature of the spacing effect and came to conclusion that the mountain of data we collected with SuperMemo, as well as a clear computational formulation of the concept of spaced repetition will sweep the world of education and memory science off its feet. A vast majority of the spacing effect literature of that time was focused on short-term studies (e.g. checking the memory effect after just a week from the trial). Bahrick's study of the retention of Spanish vocabulary was a major and stellar exception. However, Bahrick could only study the retention of vocabulary many years after the original training with no specific data on the timing of exposure to individual words during the period of learning, or during the long period preceding the measurement. In that light, we thought we have all we needed to start a new revolution in learning and in the science of memory. To our monumental disappointment, we could not push our paper through to be published in Memory and Cognition. Our failure came partly due to our inexperience, and the lack of credentials. We both have just come out of the university with MSc degrees. However hard we tried to phrase our paper around the fashionable spacing effect, we were not able to mold it to match the mainstream science of memory. From some old obscure journal, we picked the best-sounding scientific name for our repetition scheduling methodology. We called it repetition spacing. This term mutated later into spaced repetition and remains the only tangible legacy of the original paper, even though it has not yet penetrated the scholarly namespace. Perhaps it never will. More general "distributed spacing" or "distributed presentation" still predominate, while users of computerized flashcard now consistently speak of "spaced repetition". The editors of Memory & Cognition congratulated us on our results and mentioned that the computational aspect of the paper made it suitable to journals devoted to computer algorithms. The world of fashion in memory science was so different from our proposition that no top model took our stance. In the end, we published in a lesser known Acta Neurobiologiae Experimentalis. Sadly, the paper has got only 18 citations in the course of 20 years since publishing, and when it is mentioned, it is quoted with caution. After all, the "optimization algorithm" feels like a black box. It was offered free for anyone, and yet it is hard to study it in action. It is not a neat formula. It is an algorithm, and it can best be run on a computer and studied with computer means. Our line of clothing appeared to be highly unfashionable. It is now commonly used by millions, and new designers hop on board monthly. Scientific community remains largely impervious for now though. It awaits a wave of new talent grown on the feed of spaced repetition.
Fashions in science are part of our collective cognitive prejudices. They slow down progress. They are unavoidable. And still, in the long-run, they regress to the mean of the approximate truth. It is important that for each step back, we can make a dozen steps forward.
Adenosine is one of the endogenous markers of the homeostatic sleep drive. During the waking period, as the cortex and other parts of the brain keeps burning their glycogen reserves, ATP is converted to adenosine which accumulates extracellularly. The role of adenosine was first discovered upon the finding that its systemic administration promotes sleep (Radulovacki et al. 1984).
The increased activity of the cholinergic neurons in the basal forebrain neurons causes a buildup of adenosine that in turn inhibits the activity in that region via its A1 receptor (Strecker et al. 2000). This is one of the hypothetical homeostatic triggers of sleep. The accumulation of adenosine in the basal forebrain is particularly important as it is here that its effect is most pronounced (Strecker et al. 2000). The accumulation in the basal forebrain causes the inhibition of some aminergic waking centers and the disinhibition of the VLPO, which promotes sleep. Infusion of A2A receptor agonists into the rostral basal forebrain increases both NREM and REM (Satoh et al. 1999). Moreover, adenosine A2A receptor agonists stimulate the VLPO. The resulting activation of the VLPO may be measured by the increase of c-Fos activity (Scammell et al. 2001). Some of the presented scenario has recently been questioned upon finding that rats with a 95% loss to cholinergic neurons in the basal forebrain show intact sleep homeostasis despite the lack of the hallmark increase in adenosine.
Adenosine is particularly interesting as its well-known antagonist is caffeine. Caffeine binds to adenosine receptors thus blocking the homeostatic sleep propensity. This proves that network overload is not the cause or at least not the sole cause of the homeostatic sleep drive. The brain has evolved sleep protection mechanisms, in this case involving adenosine, to ensure that before a network overload leads to any significant consequences, homeostatic sleep drive pushes an animal to take a sleep break and do the necessary neural housekeeping.
- Porkka-Heiskanen T., "Adenosine in sleep and wakefulness," Annals of Medicine / Volume 31 / Issue 2 (1999): 125-129
- Kong J., Shepel P.N., Holden C.P., Mackiewicz M., Pack A.I., and Geiger J.D., "Brain Glycogen Decreases with Increased Periods of Wakefulness: Implications for Homeostatic Drive to Sleep," The Journal of Neuroscience / Volume 22 / Issue 13 (July 1, 2002): 5581-5587
- Scammell T.E., Gerashchenko D.Y., Mochizuki T., McCarthy M.T., Estabrooke I.V., Sears C.A., Saper C.B., Urade Y., and Hayaishi O., "An adenosine A2a agonist increases sleep and induces Fos in ventrolateral preoptic neurons," Neuroscience / Volume 107 / Issue 4 (November 28, 2001): 653–663, doi: 10.1016/S0306-4522(01)00383-9
- Krueger J.M., Obál F., Jr, and Fang J., "Humoral regulation of physiological sleep: cytokines and GHRH," Journal of Sleep Research / Volume 8 / Issue S1 (March 1999): 54-59, doi: 10.1046/j.1365-2869.1999.00009.x
- Krueger J.M., Obál F., Jr, Fang J., Kubota T., and Taishi P., "The Role of Cytokines in Physiological Sleep Regulation," Annals of the New York Academy of Sciences / Volume 933 (March 2001): 211-221, doi: 10.1111/j.1749-6632.2001.tb05826.x
- De A., Krueger J.M., and Simasko S.M., "Glutamate induces the expression and release of tumor necrosis factor-alpha in cultured hypothalamic cells," Brain Research / Volume 1053 / Issues 1-2 (August 16, 2005): 54-61, doi: 10.1016/j.brainres.2005.06.044
- Borbély A.A., "A two process model of sleep regulation," Human Neurobiology / Volume 1 / Issue 3 (1982): 195-204
- Daan S., Beersma D.G., and Borbely A.A., "Timing of human sleep: recovery process gated by a circadian pacemaker," American Journal of Physiology - Regulatory, Integrative and Comparative Physiology / Volume 246 / Issue 2 (February 1984): 161-183
- Akerstedt T. and Folkard S., "A model of human sleepiness." In "Sleep ‘90," edited by J. A. Horne (Bochum: Pontenagel Press, 1990), 310-313
- Scheer F.A.J.L., Wright K.P. Jr, Kronauer R.E., and Czeisler C.A., "Plasticity of the Intrinsic Period of the Human Circadian Timing System," PLoS ONE / Volume 2 / Issue 8 (2007): e721, doi:10.1371/journal.pone.0000721
- Brunner D.P., Dijk D.-J., Tobler I., and Borbély A.A.,Effect of partial sleep deprivation on sleep stages and EEG power spectra: evidence for non-REM and REM sleep homeostasis," Electroencephalography and Clinical Neurophysiology / Volume 75 / Issue 6 (June 1990): 492–499, doi: 10.1016/0013-4694(90)90136-8
- Daniel P. Brunner, Derk-Jan Dijk and Alexander A. Borbely, "Repeated Partial Sleep Deprivation Progressively Changes In EEG During Sleep And Wakefulness," Sleep / Volume 16 / Issue 2 (1993): 100-113
- Werth E., Achermann P., and Borbély A.A., "II. Muscle atonia in non-REM sleep," American Journal of Physiology - Regulatory, Integrative and Comparative Physiology / Volume 283 / Issue 2 (March 2002): 527-532, doi: 10.1152/ajpregu.00466.2001
- Swaab D.F., Hofman M.A., and Honnebier M.B.O.M., Development of vasopressin neurons in the human suprachiasmatic nucleus in relation to birth," Brain research. Developmental brain research / Volume 52 / Issues 1-2 (March 1, 1990): 289-293, doi: 10.1016/0165-3806(90)90247-V
- Swaab D.F. and Hofman M.A., "An enlarged suprachiasmatic nucleus in homosexual men," Brain Research / Volume 537 / Issues 1-2 (December 24, 1990): 141-148, doi: 10.1016/0006-8993(90)90350-K
- Saper C.B., Scammell T.E., and Lu J., "Hypothalamic regulation of sleep and circadian rhythms," Nature / Volume 437 (October 27, 2005): 1257-1263, doi: 10.1038/nature04284
- Ohta H., Yamazaki S., and McMahon D.G., "Constant light desynchronizes mammalian clock neurons," Nature Neuroscience / Issue 8 (2005): 267-269, doi: 10.1038/nn1395
- Fekete M., Van Ree J.M., Niesink R.J.M., and De Wied D., "Disrupting circadian rhythms in rats induces retrograde amnesia," Physiology & Behavior / Volume 34 / Issue 6 (June 1985): 883-887, doi: 10.1016/0031-9384(85)90008-3
- Chou T.C., Scammell T.E., Gooley J.J., Gaus S.E., Saper C.B., and Lu J., "Critical Role of Dorsomedial Hypothalamic Nucleus in a Wide Range of Behavioral Circadian Rhythms," The Journal of Neuroscience / Volume 23 / Issue 33 (November 19, 2003): 10691-10702
- Gooley J.J., Schomer A., and Saper C.B., "The dorsomedial hypothalamic nucleus is critical for the expression of food-entrainable circadian rhythms," Nature Neuroscience / Volume 9 (2006): 398-407, doi: 10.1038/nn1651
- Lu J., Zhang Y.-H., Chou T.C., Gaus S.E., Elmquist J.K., Shiromani P., and Saper C.B., "Contrasting Effects of Ibotenate Lesions of the Paraventricular Nucleus and Subparaventricular Zone on Sleep–Wake Cycle and Temperature Regulation," The Journal of Neuroscience / Volume 21 / Issue 13 (July 1, 2001): 4864-4874
- Satoh S., Matsumura H., Koike N., Tokunaga Y., Maeda T., and Hayaishi O., "Region-dependent difference in the sleep-promoting potency of an adenosine A2A receptor agonist," European Journal of Neuroscience / Volume 11 / Issue 5 (May 1999): 1587–1597, doi: 10.1046/j.1460-9568.1999.00569.x
- Radulovacki M., Virus R.M., Djuricic-Nedelson M., and Green R.D., "Adenosine analogs and sleep in rats," The Journal of Pharmacology and Experimental Therapeutics / Volume 228 / Issue 2 (February 1984): 268-274
- Strecker R.E., Morairty S., Thakkar M.M., Porkka-Heiskanen T., Basheer R., Dauphin L.J., Rainnie D.G., Portas C.M., Greene R.W., and McCarley R.W., "Adenosinergic modulation of basal forebrain and preoptic/anterior hypothalamic neuronal activity in the control of behavioral state," Behavioural Brain Research / Volume 115 / Issue 2 (November 2000): 183-204, doi: 10.1016/S0166-4328(00)00258-8