Theme : Dreams

Brainstem Origin for a New Very Slow (1mHz) Oscillation in the Human Non-REM Sleep Episode

Helli Merica1 and Ronald D. Fortune2
1Geneva University Hospital, Neurophysiology Unit, Geneva, Switzerland
2CERN European Organization for Nuclear Research, Geneva, Switzerland
Abstract
The time-courses of power in the different frequency bands (1-40 Hz) within the non-rapid-eye-movement (NREM) episode of the human sleep electroencephalogram have provided for many years a fascinating window into the sleep process. Here our analysis of the slow-wave band (1-4 Hz) reveals a hitherto unrecognized very slow oscillation of power with mean period ~15 minutes, an instability that appears to be an integral characteristic of the early NREM episode. The neuronal transition probability (NTP) model has already given a mechanism explaining how power in the spindle band peaks consistently before that of slow wave activity. Here we show that an extension of the model, with the hypothesis of a population of sleep neurons alternating between two steady probability states, can simulate the very slow oscillation. In doing so it gives not only the time course of power in the slow wave band, but also the simultaneous time-courses in the spindle and in the fast frequency bands. Animal data suggest that a brainstem neuronal population, toggled by an external switching source, generates these time-courses and dictates them to the thalamus and thence to the cortex. The discovery of the very slow oscillation and the success of the NTP model in interpreting the overall NREM structure may have important implications for both clinical and fundamental sleep research.

Current Claim: A new insight into the structure of the NREM episode of human sleep follows from the revelation of a very slow power oscillation (~1mHz) and its interpretation in the light of the neuronal transition probability model.

Activate the ShortNotes by clicking on this link. Your notes will be stored in this area and automatically retrieved upon your next visit.
The cellular basis of the electroencephalogram (EEG) has been for many years the subject of intense study. Considerable advances have been made over the past decade in establishing the sites of origin and the basic mechanisms that underlie well-defined EEG patterns in sleep. These range from individual waves (Steriade et al., 1993a, 1993b, 1993c; McCormick and Bal, 1997) to complex wave sequences such as spindles (Contreras et al., 1997) or K-complexes (Amzica and Steriade, 1997), where the functional interaction of neural substrates plays an important role in grouping and shaping the various rhythms or wave-sequences (Steriade and Amzica, 1998). These basic findings shed light on events of the order of milliseconds to several seconds in duration and provide building blocks with which the dynamic changes that operate across a night of sleep can be investigated. In addition to the patterns observed directly on the EEG, however, there are derived patterns such as the time-courses of power in the major frequency bands: beta (15-30 Hz) characterizing activity during wake or rapid eye movement (REM) sleep; sigma (12-15 Hz) characterizing spindle activity in light non-rapid-eye-movement (NREM) sleep and delta (0.5-4 Hz) characterizing slow-wave activity in deep NREM sleep. This slow-wave activity includes three different oscillations (Amzica and Steriade, 1998): a thalamically generated delta 1-4 Hz, arising from the interplay of two intrinsic currents of thalamocortical (TC) neurons; a cortically generated delta 3-4 Hz that survives thalamectomy but "takes place on a limited scale;" and a cortically generated slow oscillation 0.1-1 Hz, involved mainly in "triggering, shaping and synchronizing" spindles and delta waves. It is the thalamically generated delta only that concerns us here.

The power time-courses provide a means to examine the mechanisms that underlie EEG changes occurring on the longer time-scale of minutes rather than seconds. The evolution of the delta and of the sigma power spectra has been widely studied and the former, in particular, has been successfully used as a basis for the modeling of the homeostatic regulation of sleep (Borbély, 1982). Observations have also been made of the relationship between sigma and delta power time courses (Lancel et al., 1992; Aeschbach and Borbély, 1993; Uchida et al., 1994; Merica and Blois, 1997). The neuronal transition probability (NTP) model provided, for the first time, a mechanism by which these observations could be explained (Merica and Fortune, 1997). A description of the NTP mechanism is given in the Appendix. This model was conceived on the basis of findings on the generating mechanisms giving rise to the various sleep EEG rhythms, in particular the existence of different modes of oscillation of TC neurons (Steriade et al., 1993a; McCormick and Bal, 1997). The model, supported by observations at the thalamic level of the simultaneous occurrence of spindle and slow wave power and of the temporal precedence of spindle power over delta (Lancel et al., 1992), dealt with the single delta-process (DP) within the early NREM episode. The DP is defined as the simultaneous evolution of power in the delta (d), sigma (s) and beta (ß) frequency bands during one polarizing-depolarizing cycle (i.e., a move towards followed by a move away from deep sleep) of a fixed-size generating population of sleep-associated neurons. Henceforth we shall refer to this population simply as the generating neuronal population. This model revealed the fundamental relationship between all three of these time-courses and answered the question of how, in NREM, power in the sigma band peaks systematically before that in delta. Fitting averaged data from selected individuals, the single DP study was a necessary first step after preliminary indications of the systematic multi-peaking of delta activity within NREM (Sinha et al., 1972; Feinberg and March, 1988; Merica et al., 1997). Feinberg and March (1988) were, in their "pulsatile" theory, probably the first to see the delta-peak as the basic building block of the NREM episode. Here we take up the question of the detailed structure of the spectral power time-courses over the typical NREM episode. This can be done only on data from individual subjects. The distinction is important: averaged data often show only a single delta-peak and, although they give the general behavior of the time-courses, they effectively conceal the structure of the individual courses constituting the average. This is especially true when–as in the NREM episode–there is a wide variation in their shapes. It is perhaps because of a tendency to focus on averaged data that the complex character of the NREM spectral evolution has received little attention. In fact, the simplified qualitative descriptions of the NREM episode given in the literature are applicable, not to the entire episode, but only to the DP within the episode. We will show that in fitting individual subject data, the shapes of the early NREM episode spectral time-courses can be simulated on a neurophysiological basis by an extended version of the NTP model. In doing so, we reveal the systematic oscillatory character of these time-courses.

Activate the ShortNotes by clicking on this link. Your notes will be stored in this area and automatically retrieved upon your next visit.
The study uses data from 14 healthy young subjects (mean age 26±3 years) who were paid volunteers from whom informed consent was obtained in accordance with established ethical principles. The data were selected randomly from our large data bank of all-night sleep recordings carried out under controlled environmental conditions. For each subject, the EEG signal recorded between electrode pairs F4-Cz was digitized at a 128 Hz sampling rate and power spectra were computed by fast Fourier transform, for consecutive 4-second epochs over a frequency range 0.5-30 Hz. This range was divided into the three bands, delta, sigma and beta, as defined above. The NREM duration was normalized to 100% and average power calculated for each 2% time bin. An initial value for the delta power normalization factor was chosen so as to put the maximum delta power at about 0.9 on the vertical scale. The DPs were then identified in an objective manner: initial values for the times at which there was a switch between moving towards and moving away from deep sleep were determined by peak and trough detection using three-point moving average smoothing of the delta time-course. The NTP model was then fitted to each process in turn, with starting power values for each polarizing or depolarizing phase taken as equal to the end values of the previous phase. The free fitting parameters were thus the beta, sigma and delta power normalization factors common to all DPs in the episode, and for each DP the probabilities for the four transitions (ß–>s, s–>d, d–>s, s–>ß) and the switchover times. Since delta power was more precisely measured than either sigma or beta, delta was fitted first. Fine-tuning of the delta parameters was carried out using the coefficient of determination R2 as a goodness-of-fit criterion. The power normalization factors for the sigma and beta curves were then fitted, while retaining the values of the other parameters as already determined for delta. In order to fit beta, we subtracted from the data a constant power (bias) which varied from subject to subject. The durations of individual DPs were then measured and the mean and standard deviation of the oscillation period calculated.

Activate the ShortNotes by clicking on this link. Your notes will be stored in this area and automatically retrieved upon your next visit.

Figure 1

Figure 2
Initial results showed, rather surprisingly, that the optimally fitted sets of probability values were very similar for all DPs in all of the fourteen subjects studied. Fixing the probabilities at the mean-values obtained enabled a considerable reduction in the number of free parameters with very little loss of goodness-of-fit. The common transition probability parameters used for all fittings are: Pßs=0.13, Psd=0.13, Pds=0.20 and Psß=0.60 per 1% of the NREM duration. The first pair of probabilities defines a polarizing steady-state of the generating neuronal population, and the second pair a depolarizing steady-state.

The fit of the NTP model to the data of six representative subjects, out of the total of 14 analyzed, is shown in Figures 1 and 2, and Table 1. As for these six subjects, the results for all 14 show that the overall quality-of-fit for delta is very good, and for sigma and beta is sufficient for the model to bring out their salient features. The average goodness-of-fit R2 values over all 14 subjects, calculated from the sums of squares of the deviations, are 81.4% for delta, 39.2% for sigma and 43.3% for beta. The average number of DPs per NREM episode is 4.6±0.5 and the average duration of the NREM episode is 67.7±16 minutes. The average oscillation period is 14.6±6 minutes (variations given as standard deviations).

The sigma time-course is the central element in the frequency cascade both towards and away from deep sleep, and as such it has the most complicated shape. The extended NTP model is nevertheless able to account for the complex detail of this time-course. The model predicts that throughout the NREM episode: (a) sigma-peaks will generally precede the delta-peaks; (b) the time-course of delta is roughly the mirror image of beta; (c) delta-peaks have starting-slopes depending on their starting-magnitudes; and (d) sigma will go through its first maximum at that instant where beta falls to about one third (1/e) of its original power. These predictions are all verified in our data.

The central feature of our results is that, as in the case of overnight sleep having an average of about four NREM-REM cycles, the first NREM episode itself has an average of about four delta-processes or polarizing-depolarizing cycles of the generating neuronal population. This corresponds to a hitherto unrecognized very slow oscillation (VSO) with a period of about 15 minutes. This result demonstrates that within the early NREM episode, getting to deep slow-wave sleep and returning to REM-sleep or wake is by no means as straightforward as depicted in the literature. This process is random in time with repeated alternations towards and away from deep sleep as the system oscillates between the two steady probability states. The concept of intrinsic instability and oscillation within the NREM episode is supported by recent work on the incursion of beta frequencies into slow-wave sleep (Destexhe et al., 1999).

Activate the ShortNotes by clicking on this link. Your notes will be stored in this area and automatically retrieved upon your next visit.

Figure 3

Figure 4
Although the VSO can be seen by peak-and-trough analysis of the delta time-course without recourse to any model, the population-statistical considerations inherent in the NTP theory allow us to go much further. They explain not only the delta time-course, but the concomitant beta and sigma time-courses as well. Governed throughout the entire NREM episode by a constant probability set, this triple agreement between our theory and the data implies directly that there exist two steady-state environments between which a fixed-size neuronal population oscillates. Each of these environments supplies a statistical tendency for the population neurons to make frequency transitions and is applied simultaneously to each of these neurons. For one environment, this tendency is in the sense of encouraging polarization (cascade ß–>s–>d), and for the other is in the sense of encouraging depolarization (cascade d–>s–>ß). Each of the two environments has an associated pair of constant transition probabilities, with each probability applicable to transitions from a particular frequency band. Only toggling commands external to the population, with their associated probabilities, are required to switch between the two environments and thereby produce the observed VSO. This summarizes a top-down deductive approach based on data at the cortical (EEG) and at the thalamic level. The strength of the NTP model lies in the solidity of the tenets on which it is based and on the good agreement between theory and data, and is independent of the location of the generating population and switch system or of the origin of the probabilities. These are matters for future experimentation.

Nevertheless, there are clear pointers to the location of the generating population: it is generally believed that the gradual hyperpolarization of the TC neurons leading to deep slow-wave sleep is due to the progressive decrease in the firing-rate of the brainstem cholinergic, glutamatergic and monoaminergic neurons. Conversely, that increased brainstem excitatory input depolarizes the TC cell membranes producing a transition to wake or REM sleep (Steriade and McCarley, 1990). If the changes in the TC neuronal frequencies thus parallel the firing-rate changes of the brainstem-thalamus projecting neurons, then the spectral power time-courses of the NTP model cannot be determined at the thalamus but only upstream of it, i.e., at a brainstem sleep neuronal population. This is a hierarchical step up from our previous way of looking at the model, where we considered the thalamus to contain the generating neuronal population (Merica and Fortune, 1997). As we see it now, the brainstem generated firing-rate mode time-courses are reflected in target thalamic and cortical neurons. On this hypothesis, the transitions inherent in the NTP model are brainstem neuron firing-rate transitions and the firing-rate modes have a one-to-one correspondence with the TC neuron oscillation modes they induce (Figure 3). Thus, at any one time within the NREM episode, there should exist in the brainstem population a mixture of firing-rate modes with the relative number of neurons in each mode given by the NTP theory. The thalamus in collaboration with the cortex then has the crucial role of translating the overall brainstem control of TC neuron frequency and power into the production of the observed sequences of spindles and slow-waves (Steriade et al., 1993b, 1993c; Steriade and Amzica, 1998). In fact, animal data directly show the parallelism between brainstem firing-rate change and the move at the EEG both from fast to spindle rhythm (Steriade, 1984) and from NREM to wake or REM sleep (Steriade et al., 1997, p. 304, Fig. 4.15). This implies that there does exist a brainstem firing-rate mode corresponding exactly to each neuronal oscillation mode: beta, sigma and delta. These data, however, focus essentially on differences in firing-rates at state transitions over a time scale of only a minute or so. To verify our hypothesis that the VSO exists at the brainstem, similar experiments are needed over a longer time interval including several NREM episodes. Animal data also show that characteristic NTP cortical time-courses, with the temporal precedence of the sigma power peak over delta, are mirrored at the thalamus (Lancel et al., 1992). While so far we have stressed the brainstem-thalamic-cortical pathways in the control of cortical EEG, it should be kept in mind that there also exists, in parallel, a basal forebrain control which exerts its action via the brainstem-basalo-cortical pathway driven by cholinergic-glutamatergic neurons (Jones, 1993). However, it is as yet unclear how this pathway enters into our theoretical interpretation of the observed time-courses. From these several arguments, we propose that the VSO time-course characteristics observed at the EEG and described by the NTP model are generated at the brainstem and transmitted via the thalamus to the cortex.

Lastly, we may speculate on the location of the DP switch system that toggles the tendency to increase or decrease brainstem firing-rates, and on the mechanism for the probabilistic control of the firing-rate modes. In other words, on where are triggered and how are created and maintained the two steady probability states between which the VSO occurs–questions that are immediately posed by the success of our model in fitting the data. The first question is of intense topical interest. It has been suggested that–under the influence of the suprachiasmatic nucleus and homeostatic/sensory inputs–the arousal system in interaction with sleep promoting neurons, may provide the switch between sleep deepening and sleep lightening (Szymusiak, 1995; Shiromani, 1998; Halász, 1998; Gallopin et al., 2000). The arousal system is largely contained in the brainstem reticular formation, and the sleep promoting neurons in the ventrolateral preoptic nucleus. Their interaction may well provide the DP switch system leading to the VSO in NREM. In the context of the NTP model, this switch system (M in Figure 3) radiates, simultaneously to all neurons of the brainstem sleep neuronal population, the signals to initiate the decrease or increase of the firing-rates. On the mechanism for the probabilistic control of the brainstem firing-rate modes, we may speculate that the firing-rates are modulated by the neurotransmitter release probabilities (Katz, 1969) at the synapses (S in Figure 3) linking the switch system to the brainstem sleep neuronal population. The probabilities in the NTP model would then equate to the probabilities of releasing sufficient neurotransmitter to effect the appropriate firing-rate mode transitions. The corresponding time constants, to match EEG data, would have to be of the order of several minutes.

Conclusions

A new very slow oscillation of EEG spectral power within the NREM episode has been revealed and, in addition, the NTP theory has provided a coherent physiological interpretation of it. Animal studies suggest that this oscillation occurs, not only in the cortex and in the thalamus, but also in the brainstem. Moreover, the parsimonious characterization of NREM by the NTP model parameters gives the possibility of clinical applications in the diagnostics and comprehension of sleep disorders.

APPENDIX

The NTP model was conceived in order to solve a problem that had defied solution for nearly 10 years: how does spindle power consistently peak before delta power in NREM? The model was inspired by the resemblance between the EEG power time-course curves for beta, sigma and delta in the first part of the DP and those for the activities in a three-element radioactive decay chain. These curves are expressed mathematically by exactly the same formulae. The fundamental idea behind both processes is that the rate of decrease of a quantity N is proportional to the quantity itself. This is expressed by the equation: dN/dT= -pN, where N=quantity at time (T) and p=constant=transition probability per unit time. This integrates to N=N0 exp(-pT) where N0=quantity at start time (T=0). Single unit measurements have shown that TC sleep neurons oscillate in different frequency modes depending on the degree of their cell membrane polarization and that there are corresponding firing-rate modes in the brainstem. The sleep-deepening phase of the DP can thus be described as a cascade of neuronal transitions ß–>s–>d, with starting equations: Nß=N0ß exp(-Pß sT) and dNs/dT=-dNß/dT–NsPsd where N0ß is the fixed population size and Ni is the number of neurons oscillating in ith frequency mode at time (T). Solving for Nsand Nd we obtain the equations: Ns=(PßsN0ß/(Psd-Pßs)) (exp(-PßsT) -exp(-PsdT)) and Nd=N0ß–Nß–Ns. For the reverse phase–moving away from deep sleep–in the second part of the DP with transitions d–>s–>ß, similar formulae apply, with transition probabilities Pds and Psß.

The rule that transitions from a given state take place at a rate proportional to the amount of that state present also applies to water running out of a tank: it runs out quickly at first then more and more slowly, i.e., the water level drops exponentially.

This enables us to conceive a simple way of getting an intuitive feel for the dynamics of the NTP mechanism. Suppose that three water tanks, beta, sigma and delta, are placed so that beta can pour into sigma and sigma can pour into delta. Initially only the top tank is filled with water and all taps are closed. The amount by which a tap is opened determines the rate of transition from one tank to the next and is therefore analogous to P.

In Figure 4 the system (at T=0) has all water molecules in the beta state, i.e., a brainstem volume with all sleep neurons firing at a rate corresponding to the beta frequency band. If the top and middle taps are opened simultaneously by suitable amounts, the level in the top tank will drop exponentially. In the middle tank, the level will rise to a maximum, occurring when the amount coming in equals the amount going out, and then fall. In the bottom tank, the water will rise slowly at first, then faster, then less fast (an s-shape). The water levels correspond to the amount of sleep neurons in the firing-rate modes beta, sigma and delta. At all times, the total amount of water, i.e., the total number of sleep neurons involved, remains constant. This cascade, beta–>sigma–>delta corresponds to the neuronal polarization phase. This phase ends when all taps are simultaneously closed at delta peak time, i.e., at the time of switchover from a move towards to a move away from deep sleep.

At this switchover time, the reverse phase is initiated by interchanging the top and the bottom tanks, i.e., a depolarization command is received by the brainstem sleep neuron population. If the top and middle taps are then opened simultaneously by suitable amounts, the levels will change, but this time with delta as the cascade source instead of beta. These reversals can be repeated–each new pair of phases constituting a new delta process–up to the last depolarization phase. At the end of the NREM episode, most of the water will finish up in the beta tank, i.e., most sleep neurons in the beta firing-rate mode (REM sleep or wake).

1. Aeschbach D, Borbély AA. All-night dynamics of human sleep EEG. J Sleep Res 1993; 2: 70-81.
2. Amzica F, Steriade M. The K-complex: Its slow (<1Hz) rhythmicity and relation to delta waves. Neurology 1997; 49: 952-9.

3. Amzica F, Steriade M. Electrophysiological correlates of sleep delta waves. Electroencephal Clin Neurophysiol 1998; 107: 69-83.

4. Borbély AA. A two-process model of sleep regulation. Hum Neurobiol 1982; 1: 195-204.

5. Contreras D, Destexhe A, Sejnowski TJ, Steriade M. Spatiotemporal patterns of spindle oscillations in cortex and thalamus. J Neurosci 1997; 17: 1179-96.

6. Destexhe A, Contreras D, Steriade M. Spatiotemporal analysis of local field potentials and unit discharges in cat cerebral cortex during natural wake and sleep states. J Neurosci 1999; 19: 4595-608.

7. Elul R. The genesis of the EEG. Int Rev Neurobiol 1971; 15: 227-72.

8. Feinberg I, March JD. Cyclic delta peaks during sleep result of a pulsatile endocrine process? Arch Gen Psychiatry 1988; 45: 1141-2.

9. Gallopin T, Fort P, Eggermann E, Caull B, Luppi P-H, Rossier J, Audinat E, Muhlethaler M, Serafin M. Identification of sleep-promoting neurons in vitro. Nature 2000; 404: 992-5.

10. Halász P. Hierarchy of micro-arousals and the microstructure of sleep. Neurophysiol Clin 1998; 28: 461-75.

11. Jones BE. The organization of central cholinergic systems and their functional importance in sleep-waking states.In: Cuello AC, ed. Progress in Brain Research, Vol 98. Elsevier: Science Publishers BV, 1993, pp. 61-71.

12. Katz B. The release of neuronal transmitter substances. Liverpool, England: Liverpool University, 1969.

13. Lancel M, van Riezen H, Glatt A. The time course of s activity and slow-wave activity during NREMS in cortical and thalamic EEG of the cat during baseline and after 12 hours of wakefulness. Brain Res 1992; 596: 285-95.

14. McCormick DA, Bal T. Sleep and arousal: thalamocortical mechanisms. Annu Rev Neurosci 1997; 20: 185-215.

15. Merica H, Fortune RD. A neuronal transition probability model for the evolution of power in the sigma and delta frequency bands of the sleep EEG. Physiol Behav 1997; 62: 585-9.

16. Merica H, Blois R. Relationship between the time courses of power in the frequency bands of human sleep EEG. Neurophysiol Clin 1997; 27: 116-28.

17. Merica H, Blois R, Fortune RD, Gaillard JM. Evolution of delta activity within the NonREM sleep episode: a biphasic hypothesis. Physiol Behav 1997; 62: 213-9.

18. Pfurtscheller G, Lopes da Silva FH. Event-related EEG/MEG synchronization and desynchronization: basic principles. Clin Neurophysiol 1999; 110: 1842-57.

19. Shiromani PJ. Sleep circuitry, regulation and function: lessons from c-Fos, leptin and timeless. Prog Psychobiol Physiol Psychol 1998; 17: 67-90.

20. Sinha AK, Smythe H, Zarcone VP, Barchas JD, Dement WC. Human sleep electroencephalogram: A damped oscillatory phenomenon. J Theor Biol 1972; 35: 387-93.

21. Steriade M. The excitatory-inhibitory response sequence of thalamic and neocortical cells: state-related changes and regulatory systems. In: Edelman GM, Gall WE, Cowan WM., eds. Dynamic aspects of neocortical function. New York: John Wiley & Sons, 1984: pp. 107-57.

22. Steriade M, McCarley RW. Brainstem control of wakefulness and sleep. New York: Plenum Press, 1990.

23. Steriade M, McCormick DA, Sejnowski TJ. Thalamocortical oscillations in the sleeping and aroused brain. Science 1993a; 262: 679-85.

24. Steriade M, Nunez A, Amzica F. A novel slow (<1Hz) oscillation of neocortical neurons in vivo: Depolarizing and hyperpolarizing components. J Neurosci 1993b; 13: 3252-65.

25. Steriade M, Nunez A, Amzica F. Intracellular analysis of relations between the slow (<1Hz) neocortical oscillation and other sleep rhythms of the electroencephalogram. J Neurosci 1993c; 13: 3252-65.

26. Steriade M, Jones EG, McCormick DA. Thalamus Vol 1, Organization and Function. Elsevier: Science Ltd, 1997.

27. Steriade M, Amzica F. Coalescence of sleep rhythms and their chronology in corticothalamic networks. Sleep Research Online 1998; 1(1): 1-10.

28. Szymusiak R. Magnocellular nuclei of the basal forebrain: substrates of sleep and arousal regulation. Sleep 1995; 18: 478-500.

29. Uchida S, Atsumi Y, Kojima T. Dynamic relationships between sleep spindles and delta waves during a NREM period. Brain Res Bull 1994; 33: 351-5.

We thank Dr. R. Blois for providing the data, and B. Bertram and A. Lalji for their technical assistance. We are grateful to Professor T. Landis for his useful comments on the manuscript. This research is supported by the Swiss National Science Foundation. Grant No: 3100-050765.97/1.

Original address of this text :

http://www.sro.org/bin/article.dll?Paper&1840&0&0

Please copy this address to the address bar of your internet browser and press the "enter" key.
(We prefer not to put actual links because often page locations change and then our log files get cluttered with error messages
- if the address does not work try to find it from the homepage of the website in question).