Electroencephalography and clinical Neurophysiology, 79 ( 1991 ) 81-93

8l

,*') 1991 Elsevier Scientific Publishers Ireland, Ltd. 0013-4649/91/$03.50 ADONIS 00134649910(}116K

EEG 90682

Review article Neural m e c h a n i s m s underlying brain waves: from neural m e m b r a n e s to networks * Fernando Lopes da Silva Biology Center, Department of Experimental Zoology, University qf Amsterdam, 1098 Amsterdam (The Netherlands) (Accepted for publication: 21 February 1991)

Summary In this review, a n u m b e r of experimental findings and theoretical concepts that have led to new insights into the mechanisms underlying brain waves are presented. At the cellular level, the new evidence that certain types of neuron have intrinsic oscillatory properties that may underlie rhythmic E E G activities is discussed. In particular, the question of whether spindle oscillations are autonomous or input-dependent is addressed. At the neural network level, the main circuits of the thalamus and cortex that are responsible for the occurrence and modulation of spindles and alpha activity are described. In addition, the properties of rhythmic activities outside the alpha band are considered, particularly in relation to the prominent beta activity of the visual cortex. At the theoretical level, the possibility that neural networks may behave as complex dynamic systems with the properties of deterministic chaos is discussed. Finally, the fact that brain rhythms may have functional implications for the working of neural networks is examined in relation to 2 cases: the possibility that oscillations may subserve a gating function, and that oscillations may play a role in the formation of assemblies of neurons that represent given stimulus patterns. Key words: E E G generation; Thalamic cells; Cellular oscillations; Spindles; Alpha activity; Beta activity; Coherence; Network models; Chaos

It may not be surprising that a lecture that carries the prestigious name of Adrian is dedicated to one of the scientific fields where he was a famous pioneer: that of the neural mechanisms underlying brain waves, Appropriately, we must remember that it was the research work of Adrian and Matthews (1934) that initially demonstrated that alpha waves have their origin in the brain, and of Adrian and Yamagiwa (1935) that showed that alpha waves were not in phase over the whole scalp and most probably originated in the occipital region. In the half century that elapsed between those seminal observations, following Berger's discovery, and the present time, a large number of research studies have been dedicated to the phenomenology of brain waves and their behavior in healthy subjects and in disease states. Nevertheless, it has been difficult to unravel the basic mechanisms responsible for the generation of brain waves at the neuronal level. This has led some even to propose extracranial sources for the alpha rhythm, namely the eye muscles, and to suggest

* This review is a slightly rearranged version of the Adrian Lecture delivered at the XI1 International Congress of Electroencephalography and Clinical Neurophysiology, Rio de Janeiro, Brazil, on January

17th, 1990. Correspondence to." Dr. F. Lopes da Silva, Biology Center, Dept.

of Experimental Zoology, University 1098 Amsterdam (The Netherlands).

of Amsterdam, Kruislaan 320,

that this rhythmic activity would reflect cardiac induced electromechanical properties of the brain. These suggestions can be dismissed on account of a number of experimental observations, the most spectacular of which may be the demonstration that alpha rhythms can be recorded from the isolated dog's brain, in the absence of orbital contents, drug effects or pulsatile cerebral blood flow (Hogan and Fitzpatrick 1988). Furthermore, the cortical source of the alpha rhythm in the awake dog was demonstrated using multielectrode intracortical recording (Lopes da Silva and Storm van Leeuwen 1977). In the last decade a number of experimental findings and theoretical concepts have led to new insights into the mechanisms underlying brain waves. In this respect, 3 developments are particularly relevant: at the cellular level, the evidence that certain types cff neuron have intrinsic oscillatory properties that may underlie rhythmic E E G activities; at the neural network level, new findings have clarified the dynamics of the main circuits responsible for the occurrence and modulation of rhythmic behaviour in neural p0pulations; at the theoretical level, the demonstration that neuronal networks may behave as complex dynamic systems with the properties of deterministic chaos has challenged the classic ideas about how E E G signals should be interpreted and analysed. In this review, the evidence supporting these 3 lines of development will be briefly presented and discussed.

82

F. LOPES DA SILVA

EEG rhythmicactivitiesat the cellularlevel: the case of spindles in the 6-14 Hz frequencyrange At the cellular level, an important issue that has been repeatedly discussed in studies of the mechanisms underlying E E G rhythmic activities is the question of whether such rhythms are caused by single cells with pacemaker properties. A significant advance in this discussion was achieved when it was demonstrated by Jahnsen and Llinfis (1984a, b) that some types of thalamic neuron display oscillatory behavior in vitro, even after blockage of synaptic transmission. Most neurons studied in this way tend to generate oscillations in the frequency range of 6-10 Hz. These intrinsic oscillatory properties of some neurons are most probably of importance in shaping the rhythmic behavior of the networks to which they belong. However, these properties may not be sufficient to account for the network rhythmic behavior. One main argument supports this statement: the disconnection of cortically projecting thalamic cells from their inputs, that arise in the reticular nucleus of the thalamus, abolishes spindles in thalamo-cortical relay (TCR) neurons (Steriade et al. 1985). However, single T C R neurons may preserve their intrinsic m e m b r a n e oscillatory properties in vitro (Jahnsen and Llin~s 1984a, b). In order to understand how oscillations may be generated in thalamic nuclei we have to consider the basic structure of these neuronal networks. In the thalamic relay nuclei where this type of rhythmic activity can be recorded, 3 main types of neuron can be distinguished: the thalamo-cortical (TCR) neurons whose axons project to the cortex, the reticular nucleus (RE) neurons that contribute to the inhibitory feedback control of the former and the local, intrinsic, neurons. The RE forms a thin sheet of neurons that surround the anterior and lateral parts of the thalamus and interact synaptically with the T C R neurons. The axons of the RE cells have as main target the T C R ceils. Their dendrites make synaptic contacts with the axons of the T C R ceils. Thus, the T C R and the RE neurons are interconnected by means of a feedback loop as schematically shown in Fig. 1. Both the RE and the local-circuit neurons contain gamma-aminobutyric acid (GABA) as transmitter. This implies that the feedback of the R E on the T C R neurons is GABAergic and thus inhibitory. In a thalamic nucleus, a number of feedback circuits can be distinguished between T C R and RE neurons, In the context of this discussion, it is important to note that the T C R neurons can work essentially according to 2 distinct modes: either (a) as relay cells that simply depolarize and can produce single spikes in response to an adequate input volley, or (b) as oscillatory cells producing bursts of high-frequency spikes that are repeated in a rhythmic fashion. The resting

~. ~

/ ~ i

J-f

f-

I

.........

~ C

i: c. Tri

Synapses

--Q) ~

Gabaergic Excitatory /~ C h o l i n e r g i c

Fig. I. Simplified scheme of a thalamic neural network where the main features are presented. The tbalamo-cortical relay (TCR), the reticularthalamic (RE) and the local-circuit interneurons (IN) are indicated, along with projections from a cholinergic population (ACh). The synaptic triad formed by the dendrites of TCR and IN neuronswith the specific afferent (spec aft) fibers is also indicated. Three types of synapse are shown: excitatory, GABAergic (inhibitory) and cholinergic. The effect of the latter on RE neurons is assumedto be an increase in potassium conductance (gK), whereas on the TCR neurons the effect is the opposite.

m e m b r a n e potential of a T C R neuron determines whether such a neuron will be active in the relay or in the oscillatory mode. How can this take place? To reply to this question, it is necessary to take into account the intrinsic m e m b r a n e properties of these cells, namely their main ionic conductances and the corresponding dynamics. It should be noted that these properties became known only quite recently, mainly due to microphysio[ogical investigations in thalamic slices, in vitro and in isolated brain preparations, by the group of Llin~s and collaborators (Jahnsen and Llinfis 1984a, b). For a comprehensive account of these m e m b r a n e properties, the reader is referred to the original publications and to the excellent reviews by Steriade and Llinfis (1988) and Steriade et al. (1990a, b). Here, we shall consider only the main aspects that are pertinent to understanding how these ceils can exhibit intrinsic m e m b r a n e oscillations even in in vitro conditions. The main ionic conductances of the T C R neurons are summarized in Fig. 2, where it can be seen that the interplay of different ionic conductances (indicated as G or g with the corresponding ion given as subscript)

N E U R A L MECHANISMS U N D E R L Y I N G BRAIN WAVES

oeo~,,,os

s~a

long-lasting potassium-dependent IPSPs mediated by

t GG B

GNa

dendritic sDikes

[nonlnlct~vlting]

somatocand

~o.~p,k,~ GN~ ( ........... ,I

10Hz rhythm

G~(ca~ GK GK(Ca )

/

J

dendritic a , e -

,oboooa,~,,e,

hyoerpolarlzahon

GCa

6Hz,hy~hm

(. . . . . . . . . . . . )

GK(a)

6 Hz 10 HZ

-

gK

~

*g

"

H.T, K(Ca} ""X

~ ~ ~.~i2-'[----~ ~ -55 mV* ~ - - - - - - - - - - - - ' m - - - - " ~ - - - - - - ~ - ~--

[10mY

Ica IPSP * IA ÷ gK{ca)

83

25 ms

Fig. 2. Summary of the main conductances (g) of thalamic neurons and their behavior in the generation of thalamic oscillations. HT, high threshold spike; LT, low threshold spike; IPSP, inhibitory postsynaptic potential; IA, potassium A current; other abbreviations are indicated in text. (Adapted with permission from Steriade and

Llinfis 1988.)

underlies the oscillatory behavior of these cells and determines at which frequency they tend to oscillate, Whether a TCR cell oscillates at 6 or at 10 Hz depends on the initial conditions, i.e., on the level of the membrane potential. The switching from one to the other mode is voltage-dependent. In Fig. 2 it can be seen that the cell tends to oscillate at 10 Hz when the initial state of the membrane potential is relatively more depolarized than in the resting state, and at 6 Hz when it is more hyperpolarized, It is important to consider also the intrinsic properties of the other type of cells, the reticular thalamic neurons (RE) which, jointly with the TCR cells, form the basic neural network of the thalamic nuclei responsible for spindle generation. The R E neurons show also 2 basic modes of operation: tonic and burst firing, The first mode occurs as the cells initially have a rather negative resting potential and it depends on a similar intrinsic mechanism to that of TCR cells, i.e., a calcium-dependent low-threshold action potential located at the proximal dendritic level (Mulle et al. 1986) that is deinactivated by a preceding hyperpolarization. It is interesting to add that, recently, it was proposed that

GABA B receptors of TCR neurons may also play a role in inducing these cells to fire rhythmically (Crunelli and Leresche 1991). These GABA B mediated IPSPs, that are usually preceded by IPSPs mediated by GABA A receptors, would provide the hyperpolarization that is necessary to remove the inactivation of the calcium current responsible for the low threshold action potentials. A basic question is whether the different types of oscillation can be found in vitro in the absence of external inputs, since this would give support to the idea that these thalamic cells may support pacemaker rhythmic activity. Jahnsen and Llinfis (1984a) provided an initial answer to this question, by showing that the oscillatory activity at about 10 Hz may be observed in the absence of external drive in vitro. Having briefly reviewed the intrinsic membrane properties of thalamic neurons that may subserve oscillatory behavior in vitro, we must now consider what is the contribution of the feedback circuits between TCR and RE neurons to the generation of EEG oscillations in vivo. In this context the most direct experimental evidence pertains to the generation of spindle oscillations occurring during the early stages of sleep. Spindles appear in the E E G of most mammals as oscillations with a frequency in the range from 7 to 14 Hz forming bursts lasting for 1.5-2 sec, that recur with an irregular periodicity of about 5-10 sec. This type of spindle can also be seen in animals under barbiturate anesthesia. It was shown by Steriade et al. (1985) that each thalamic nucleus is incapable of generating spindle oscillations after disconnection from the RE nucleus. Therefore the existence of feedback circuits, including the RE nucleus, is necessary for spindling to occur under normal conditions in vivo. A most striking record of how the oscillatory depolarizations in RE neurons jointly occur as oscillatory hyperpolarizations in TCR neurons is shown for the case of a barbiturate-induced spindle in Fig. 3. In this example, the main oscillation is at about 6 Hz, but the rhythmicity is rather irregular. The depth of anesthesia usually affects the frequency of the oscillations, and in unanesthetized drowsy animals the spindle frequency is generally higher than in the anesthetized state (Steriade and Llinfis 1988). The rhythmic hyperpolarizations are due to sequences of long-lasting IPSPs, generated by the inhibitory synapses formed by RE axons on TCR cells. These hyperpolarizations will induce the deinactivation of the low-threshold calcium currents; consequently action potentials are generated in the TCR cells that lead to the feedback activation of the RE neurons; these in turn hyperpolarize the TCR cells, and in this way the oscillatory state will be established. In contrast to the TCR cells, the mean membrane

84

F. LOPES DA SILVA

_

I~ • ass Fig. 3. Cellular oscillations during a barbiturate EEG spindle in a cat. Membrane potential oscillations of a neuron of the reticular nucleus (RE), of a thalamo-cortical neuron (Th-Cx) and of a pyramidal neuron of the cortex (PT). Note that during the oscillatory e v e n t the mean membrane potential of the RE neuron changes in the depolarized direction, whereas that of the Th-Cx neuron tends to be hyperpolarized. (Adapted with permission from Steriade and Desch~nes 1988.) potential of RE neurons is depolarized during the spindle, In conclusion, under the normal conditions of an intact brain, both synaptic interactions and input signals play an essential role in setting the conditions necessary for oscillatory behavior to occur. The frequency of this oscillation depends on the intrinsic m e m b r a n e properties, on the m e m b r a n e potential of the individual neurons, and on the strength of the synaptic interactions. The thalamic nuclei responsible for the generation of rhythmic activity, in addition to the specific sensory inputs, receive 2 main systems of unspecific inputs: (a) the cholinergic brain-stem-thalamic projections and (b) the monoaminergic projections arising from the locus ceruleus and the raphe nuclei, (a) The cholinergic projections have a number of effects: acetylcholine causes hyperpolarization of the RE cells by activating a K + conductance, through muscarinic (M 2) receptors (McCormick and Prince 1986) and, by contrast, it produces a fast depolarization of T C R cells followed by a sequence of hyperpolarization and depolarization, although these effects are species-dependent (McCormick and Prince 1987). What are the resulting effects of the cholinergic systems in the intact animal? We may start by discussing what happens if the cholinergic inputs decrease, and later we shall discuss what the effects are of their stimulation. It has been shown that just before (1 sec) the transition from wakefulness to sleep there is a decrease

in the firing rate of the cholinergic brain-stem inputs (Steriade 1984). This results in the withdrawal of the nicotinic receptor mediated depolarizations on T C R cells. At the same time the inhibitory influence of acetylcholine on RE cells is removed and these two effects together may set the conditions for the oscillatory mode of the whole population of thalamic cells. In contrast, stimulation of the cholinergic inputs leads to a blockade of the spindles, most probably through the inhibition of RE cells (Sillito et al. 1983) causing the opening of the main feedback loop of the thalamus. We should add that the R E nucleus receives also an important input from the basal forebrain cholinergic cell groups (Levey et al. 1987). The stimulation of these forebrain cholinergic inputs provokes a decrease, or even blockade, of spindles whereas a specific lesion of these areas will cause an increase in the tendency for spindles to occur (Buzsaki et al. 1988), and thus affects the E E G accordingly. (b) Monoaminergic inputs consist mainly of noradrenergic fibers arising in the locus ceruleus and serotonergic fibers arising in the raphe nuclei. The local iontophoretic application of noradrenaline or the electrical stimulation of the locus ceruleus, in general, increases the firing rate of T C R cells, through the activation of a-adrenergic receptors. This effect is caused by a blockade of a resting potassium conductance rather than a voltage-dependent one (McCormick and Prince 1988). In general, noradrenaline can enhance the signal-to-noise ratio of the transfer of impulses in several brain structures, including the thalamus (Rogawski and Aghajanian 1980a, b), by enhancing both excitatory and inhibitory responses against the background firing. Recently, it was shown (Pape and McCormick 1989) " that both noradrenaline and serotonin may increase in thalamic cells a cation carrying conductance. In this way a selective dampening of the responsiveness of the cell to large hyperpolarizing inputs would result. Since this hyperpolarization is a necessary condition for triggering the oscillatory behavior, these neurotransmitters would tend to decrease burst firing in these thalamic cells. In conclusion, although monoaminergic fiber systems are not likely to play a primary role in the generation of the spindle mode of E E G activity, they may contribute to the modulation of this activity, and their stimulation may decrease the tendency for spindles to occur.

From spindles to alpha waves In the previous sections I have concentrated on the cellular aspects of the neurophysiology of spindle activity in the frequency range of 6-14 Hz, that commonly

NEURAL MECHANISMS UNDERLYING BRAIN WAVES

appears in the superficial stages of sleep and under barbiturate anesthesia. In comparison, the basic mechanisms of other rhythmic activities are much less well understood. In this section I examine some of these aspects and discuss the possibility of formulating a general model of a neural network that may be used as a paradigm for the generation of rhythmic activities in the brain, The type of rhythmic activity of the brain that is most directly related to the spindles discussed above is, of course, the alpha rhythm. However, the conditions for the occurrence of the alpha rhythm are quite different from those necessary for sleep spindles. The alpha rhythm occurs in the frequency range of 8-13 Hz, during relaxed wakefulness and particularly at eyes closure, mainly over the parieto-occipital cortex, Therefore the frequency range overlaps, although it does not coincide with, that of the barbiturate and sleep spindles. Furthermore, there are a number of essential differences between the two types of rhythmic activity concerning their spectral contents and topographic distributions in the brain, as described previously (Lopes da Silva et al. 1973b). In addition, spindies appear during unconsciousness and are associated with blockade of synaptic transmission through the thalamus (Steriade and Llin~s 1988), whereas alpha waves may increase during some attention tasks (Creutzfeldt et al. 1969; Ray and Cole 1985). Experiments using the clog as experimental model have revealed a number of elementary properties of alpha waves that are relevant to understanding the corresponding mechanisms of generation. Alpha waves with the same characteristics can be recorded from the visual cortex and the visual thalamus, namely the fateral geniculate nucleus and pulvinar (Lopes da Silva et al. 1973a). The alpha rhythms of the visual cortex are generated by cortical neurons forming an equivalent dipole layer at the level of cortical layer I V / V , that corresponds to the activation of the somata and basal dendrites of pyramidal neurons of these two layers (Lopes da Silva and Storm van Leeuwen 1977) as illustrated also in Steriade et al. (1990b). These cortical alpha rhythms appear to be generated in small cortical areas, so-called epicenters, from which the activity spreads in different directions by way of cortico-cortical connections (Lopes da Silva and Storm van Leeuwen 1978). Of all thalamic nuclei studied the pulvinar appears to have the strongest influence in determining the alpha activity of the cortex (Lopes da Silva et al. 1980a, b), as indicated also in Steriade et al. (1990b). At the macroscopic level, the complex dynamics of how alpha waves are generated and propagate in the cortex are becoming understood. However, at the microseopic level the cellular mechanisms of alpha rhythms are still largely undefined because the conditions under which intrinsic membrane properties are

85

best studied, i.e., in vitro, differ from those necessary for alpha rhythms to occur in the awake animal. Nevertheless, the experimental findings mentioned above were incorporated in a model of a simplified chain of cortical neurons (Van Rotterdam et al. 1982) that is an updated extension of our previous model (Lopes da Silva et al. 1974, 1976). Such a neuronal chain is assumed to have the following properties: ( a ) t h e transfer functions of input-output spike densities are, to a first approximation, second-order linear functions with a longer time constant for inhibition than for excitation; (b) the interconnectivity properties are homogeneous in space, i.e., the strength of inhibition is only a function of distance between the interconnected neurons; and (c) the strength of interconnectivity decreases exponentially as a function of distance within the cortex. These model studies address, mainly, the question of how the spatial properties of the distribution of alpha rhythmic activity over the cortex can be explained, taking into account basic neuronal and synaptic features. The model shows that the rhythmic components in the alpha band are less damped along a neuronal chain with similar basic features to those presented above for the thalamic neurons (dynamics of depolarization and hyperpolarization) than the components outside the alpha band. The simulated alpha activity has characteristics that are in qualitative agreement with experimental data on alpha rhythmic activity obtained from the visual cortex of dog (Lopes da Silva and Storm van Leeuwen 1978). The comparison between experimental and theoretical results can be examined in relation first to coherence and second to phase measurements. Coherence analysis of E E G signals recorded from chronically implanted subdural electrodes lying directly on the marginal gyrus revealed a higher correlation between E E G components in the alpha band than between components outside that band (Lopes da Silva et al. 1980a, b). These experimental data support the conclusion from the model study that rhythmic activities within the alpha frequency range are the least damped in the chain. Phase shifts at the alpha frequency (around 12 Hz) measured over the marginal gyrus in the dog reflected phase velocities that were about 30 c m / s e c (Lopes da Silva and Storm van Leeuwen 1978). Using the model, the characteristic length corresponding to this phase velocity and frequency was computed. The order of magnitude of the characteristic lengths was found to be within the range of 160-330 /.tm. The characteristic length represents the distance at which the influence of a neuron on a neighbor is reduced to 37% of the starting value. The figures calculated using the model are in the same range as the dimensions of the basic cortical space module, which is considered to be a

86 vertical cylinder of 200-300 ktm diameter (Chow and Leiman 1970; Braitenberg 1977; Szentfigothai 19781. In addition, several systems of intracortical fibers, which involve much larger distances than the diameter of a column (distances at least one order of magnitude larger), assure the interconnectivity of several columns, These systems of overlapping intracortical connections can lead to the establishment of cooperative action between large populations of cortical neurons and thus to the appearance of dynamic patterns over the cortex (Katchalsky et al. 1974). In conclusion, these spatial properties are important to understand how alpha generators at the microscopic level can cause the formation of clusters of generators, The existence of such clusters is necessary to explain the macroscopic features of alpha rhythms recorded from the cortical surface, as discussed below.

F. LOPES DA SILVA thalamic signal contributed to the alpha rhythms recorded from both cortical sites. This effect was small, although significant, for the signals recorded from the lateral geniculate nucleus, but much larger for those recorded from the pulvinar, indicating that this nucleus exerts an important influence on the alpha activity of different cortical areas. Nevertheless, the cortico-cortical coherence remaining after partialization of the thalamic signals was still appreciable in the majority of cases, indicating that in addition to the existence of thalamo-cortical components underlying cortical alpha activity, there are also cortico-cortical components (or other subcortical sources) that contribute to the generation of a cortical domain of alpha rhythm and to its propagation over the cortex.

Clustered domains of rhythmic activity in the cortex The organization of thalamo-cortical relationships It may be assumed that the intrinsic membrane properties of the thalamic neurons in conjunction with the synaptic interactions between excitatory and inhibitory neuronal populations form the basic condition for the generation of rhythmic activity, such as sleep spindles or alpha waves. Consequently, this thalamic rhythmic activity will impinge on the cortex and lead to the occurrence of cortical rhythmic activity. However, the fact that intracortical coherences of alpha activity measured over the cortical surface are in general larger than any thalamo-cortical coherence measured in the same animals (Lopes da Silva et al. 1973a, 1980a, b) leads to the following question: are these large intracortical coherences exclusively due to an intracortical process independent of thalamic sources, or not? An answer to this question may be obtained experimentally by measuring the effect of eliminating the thalamic influences on the cortex. To avoid making surgical lesions, which most certainly would affect specific and non-specific thalamo-cortical systems, a signal analytical procedure consisting in realizing a sort of 'theoretical thalamic deafferentation,' was proposed (Lopes da Silva et al. 1980a, b). This procedure consists in computing partial coherence functions. In this way the effect of the signal recorded from a thalamic site on the coherence between a pair of cortical signals (e.g., alpha activities recorded from two closely spaced cortical sites) can be estimated. This statistical procedure, called partialization, consists in eliminating from each of a pair of cortical signals the part that can be considered as being determined by, or predictable on the basis of, a third'signal (e.g., in this case the thalamic EEG). For most pairs of cortical signals investigated, there was a significant decrease of the intracortical coherence after partialization, indicating that the

A condition for E E G rhythms to be measurable at the scalp is that the corresponding cortical neurons form a coherent cluster over an appreciable surface of the cortex. This means that for such waves to be recordable at the scalp the underlying cortical surface must involve a relatively large number of cortical roodules occupying a few square millimeters or even centimeters. In a recent study of the spatial variance of E E G and M E G signals in the man (De Munck 1989), it was shown that the alpha activity can be described by a statistical model consisting of clustered dipoles with a fixed position and a randomly fluctuating amplitude, whereas the remaining E E G frequency components appear to have a spatial covariance that only depends on interelectrode distance, as would be expected from activity generated by randomly distributed cortical dipoles. Nunez (1989) has also shown that it is necessary to assume a set of clustered dipoles, i.e., dipoles showing the same orientation and activated in a coordinated way, to account for the potential fields that are commonly recorded from the scalp during the alpha rhythm. The generation of such sets of clustered dipoles with a consistent phase lag can be accounted for, at least for the case of the alpha rhythm, by the model of a cortical chain of neurons interacting by means of cortico-cortical fiber systems as explained above and in more detail in Van Rotterdam et al. (1982). In a comparative physiological study Bullock and McClune (1989) showed that, on average for the cortex of rabbits and rats, the horizontal extent of coherence of the spontaneous E E G when no rhythmic activity is dominant, tends to fall with distance, such that it decreases to 0.5 for most frequency bands within 2.5-5 ram. These authors stress, on the basis of these findings, that the E E G generators are largely independent but synchronized to varying degrees. It appears that in those cases where rhythmic activities dominate the

NEURAL MECHANISMS UNDERLYING BRA1N WAVES EEG, such as during alpha rhythms, the degree of cooperativity increases in a very significant way, such that coherent activity can occur over larger extents of the cortical surface,

Rhythmic activities outside the alpha band In comparison with the extensive studies of rhythmic activities within the alpha frequency range, other rhythmic activities have been much less investigated, with the possible exception of the theta rhythm of the limbic cortex, but the latter has been reviewed elsewhere (Lopes da Silva et al. 1990, 1991). Thus, we consider here only a few of the recent findings that are relevant to understand the mechanisms underlying the generation of rhythmic activities in the low (delta activity) and in the high frequency ranges (beta activity),

(1) Delta acticity and cholinergic modulation One of the recent findings that shed a new light on the generation of delta activity is the demonstration that deafferentiation of the cortex from cholinergic inputs may condition the appearance of cortical delta activity. Detari and Vanderwolf (1987) and Buzsaki et al. (1988) showed that a decrease of the input to the cortex arising in the cholinergic neuronal groups of the basal forebrain could cause slow waves of large amplitude in the cortical EEG.

(2) Beta rhythms, mechanisms and functional releL,ance: the formation of oscillating cell assemblies E E G activity in the high frequency range, usually called beta or even gamma activity (Bressler 1990), appears in many brain areas, and it dominates the local E E G recorded from olfactory areas of the base of the forebrain (Freeman 1975; Bressler and Freeman 1980) and also from the temporal cortex, namely the entorhinal cortex of the cat (Boeijinga and Lopes da Silva 1988). Most interesting is the consistent finding of beta rhythms in the neocortex in association with behavioral conditions when an animal is highly alert and focusing his attention on a target. This was shown to occur in the awake dog while looking attentively at a screen where a visual conditioned stimulus was expected to appear (Lopes da Silva et al. 1970). It should be stressed that different modes of rhythmic activity can be recorded from the same cortical sites depending on the behavioral state of the animal; for example, a rather regular spectral peak in the alpha frequency range when the alert animal has the eyes closed, and a somewhat less regular spectral peak in the beta range as the animal pays attention to a visual stimulus, as shown in Fig. 4. This means that a change in attention does not cause a simple desynchronization of the alpha

87 rhythm, but can lead to the appearance of another mode of rhythmic activity. It is interesting to note that whereas the alpha activity is evident both in the cortex and the thalamus (LGN), the beta activity appears only in the cortex. Also in the alert cat, while focusing his attention on an unsizable mouse, an activity in the frequency range 35-45 Hz was described, in the motor and the parietal association areas (Rougeul-Buser et al. 1983; Bouyer et al. 1987). It is worth noting that cortical rhythmic activity of the same frequency range appears to depend on an intact system of cortically projecting dopaminergic fibers arising from the ventral tegmental areas (Montaron et al. 1982). Recently Freeman and Van Dijk (1987) described, in the visual cortex of the monkey, beta activity with a spectral peak of 30 _+ 3.7 Hz, appearing when the animal performed accurately a conditioned response to a visual conditioned stimulus. Cortical beta activity related to the performance of a motor task has also been reported in man over the parieto-temporo-occipital cortex and the motor cortex (De France and Sheer 1988). An intriguing new and possibly related finding is the discovery by Gray et al. (1988) of oscillations in the firing of visual cortical neurons responding to moving light bars, within the beta frequency range. Using autoand cross-correlation analyses of multi-unit cortical responses of cells that had similar orientation preferences to appropriately oriented moving light bars, these authors found that these responses were oscillatory with a frequency ranging from 40 to 60 Hz and synchronous for neuronal groups, at distances up to about 7 mm. In most cases the phase difference between oscillatory units was zero and usually it did not exceed 3 msec. Similar findings were reported by Eckhorn et al. (1988), who described 'patches of enhanced correlations' where neurons of visual areas 17 and 18, with overlapping receptive fields, had synergistic oscillations in the beta frequency range (20-60 Hz), but also at lower frequencies. In addition, these authors found that the local field potentials showed also oscillations at about 40-60 Hz in response to the same stimulus conditions. However, Eckhorn et al. (1989) noted that the cross-correlation functions between single-unit signals recorded from neighboring cells could have not only a narrow peak around 0 msec, indicating a precise coupling and zero phase, but also a broad peak at 10-20 msec. The establishment of such cortical domains of oscillations may form a mechanism for the representation of the features of a given stimulus. lndeed, the fact that different neuronal elements may display coherent oscillations can be interpreted as a sign that such elements form a cell assembly responsible for the extraction of relevant features of a given stimulus configuration (Von der Malsburg and Singer 1988).

OC lat 150--

100 -

30" 25-

50-

20 15-

~

0o

~o ,27

1020

30

~o

OC

mes

300 -16-

100 ~

"

O-

.

3

6-

-

i

o

lo 127

20

-

1-

30

~o

LGN 6O

8-

/.0

'76-

30

5" 4-

20

~

32-

10

10

' 127 10

' 20

' 30

' /.0

6

2L

p.V /Hz

.--4P.

Hz

lb 117

2~

3'o

~.'o

N E U R A L MECHANISMS U N D E R L Y I N G BRAIN WAVES

EEG s i g n a l s can reflect s o m e functional states of neural networks

E E G signals vary as function of the state and of the area of the brain. These signals reflect spatio-temporal patterns of brain electrical fields that consist of a series of short-lasting quasi-stationary epochs corresponding to what Lehmann et al. (1987) have called brain functional micro-states. To extract information over these spatio-temporal states from records of scalp E E G s (or MEGs) is a problem that is difficult to solve in general terms. The current insight is that the brain processes information in a parallel fashion, where several neuronal assemblies are simultaneously active. These assemblies may occupy different cortical areas that may not be anatomically contiguous. It is therefore necessary for the brain to integrate the operations of multipie anatomically distinct areas. In other words, it may be stated that the brain performs the integration of a distributed set of neuronal activities spread over multipie cortical areas to produce a coherent representation of a stimulus pattern. This essential property of the brain has been analyzed recently by means of connectionist models (Ballard et al. 1983) and interpreted in terms of the theory of neuronal group selection (Edelman 1989). It is likely that the activities of the different neuronal networks are reflected in the local field potentials. However. the fact that these networks are most likely not anatomically contiguous and may occupy cortical areas with different dimensions and positions implies that the resulting E E G activities will form a complex spatio-temporal pattern with a large number of degrees of freedom. Therefore, it may be extremely difficult to extract such a spatio-temporal pattern of activity from the E E G ongoing activity and to relate it to the corresponding cognitive process. Nevertheless, some successful attempts have been made to compute a set of correlations between E E G signals recorded from different sites, occurring during a particular behavior, in order to infer which brain areas are responsible for such a behavior and how they are functionally related, for instance during the preparation for motion (Gevins et al. 1986, 1987; Gevins and Bressler 1988) and cognitive tasks (Klimesch et al. 1988). Even if we have to accept that the ongoing E E G activity can only be used in selected cases to obtain an insight into how the brain processes information, the possibilities offered by the use of event-related potentials or changes

89

in E E G spectral properties (Pfurtscheller and Aranibar 1977; Pfurtscheller et al. 1988) should not be underestimated, although this aspect lies out of the scope of this review. Specially relevant to the present topic is the fact that E E G signals reflect more readily changes in the s t a t e of the underlying networks, particularly when these changes involve a relatively large area, than specific aspects of the information being processed. A most evident example is the change in E E G pattern from wakefulness to sleep. Within the scope of the present discussion, it is of interest to examine whether some E E G features, such as sleep spindles, characteristic of the transition between wakefulness and sleep, may play a functional role in this state transition. A main functional effect of thalamic spindles is that during their presence sensory inputs are prevented from being relayed to the cortex. This was experimentally tested by studying the excitability of thalamic neurons to synaptic inputs (Glenn and Steriade 1982). During the transition from the awake state to superficial sleep as spindles occur in the thalamic nuclei and in the cortex, the thalamo-cortical relay neurons are mainly hyperpolarized. Under these conditions there is a decrease probability of these neurons responding to synaptic inputs arising from the periphery. This means that the E E G activity characterized by spindling activity corresponds to a functional state where the specific thalamic nuclei and the cortex are cut off from sensory inputs. In this way, the oscillatory activity characteristic of the state where spindles occur may subserve a 'gating function,' as regards the flow of specific informarion through the thalamus to the cortex. Also a 'gating function' has been assigned to the rhythmic slow activity (RSA or theta rhythm) of the limbic cortex, as regards the transmission of signals through the entorhino-hippocampal pathway (for review see Lopes da Silva et al. 1990, 1991). It is possible that the most adequate way to accomplish a change of state in a neuronal network of the brain is by means of a transition from a random type of activity to an oscillatory mode. Such an E E G oscillation has 2 important features: (i) it can occur in a rather synchronized form over a large neuronal population, and (ii) it corresponds to a long lasting change in mean m e m b r a n e potential level as encountered in the thalamus (Fig. 3) in the case of spindles and in the hippocampus in the case of theta rhythm (Lopes da

Fig. 4. Power spectra calculated from EEG signals recorded from two sites on the surface of the occipital cortex (oc) in the dog (lateral gyrus, oc-lat, and mesial cortex, oc-mes) and from the lateral geniculate nucleus (LGN) as indicated in Lopes da Silva et al. (1970). The spectra on the (a) left were obtained while the awake animal kept its eyes closed, and those (b) on the right when the animal paid intense attention to a visual object. Note that under both conditions the cortical activity has different dominant frequency components: alpha activity dominates in (a) and beta in (b). At the same time the LGN shows also a change in spectral content but without evidence for beta frequency components. Note also that the dominant beta frequency differs in the two cortical records.

90

Silva 1991). In this way, a generalized and long lasting change of membrane level in an extended population of neurons can be efficiently effectuated. Accordingly, the functional state of the neurons within the population will be biased in a different direction. This is what is meant here by the term 'gating function.' In addition to a gating function, the oscillatory mode of behavior in a neuronal network may be of functional importance in favoring the coupling between different brain areas if they have similar resonance characteristics, as has been suggested for the theta rhythm of the limbic areas (Lopes da Silva 1991). It is not clear whether similar coupling between different areas may also occur for other frequency ranges, such as for sleep spindles, or E E G activity in the beta frequency range. In contrast to E E G spindle activity, the pattern characterized by oscillations in the beta frequency range is typical of a brain state of attentive information processing. Under these conditions the E E G may reflect coherent states of the brain during which cognirive and sensory inputs are being processed (Ba§ar 1990).

Do EEG signals correspond to chaotic oscillations? It is clear that neuronal networks that generate E E G signals are complex systems with non-linear dynamics. Therefore several theoreticians have applied methods of analysis of non-linear dynamic systems to find new ways of characterizing brain functional processes through E E G signals (Babloyantz 1985; see also for review Ba§ar and Bullock 1989). An important observation in this respect is the fact that E E G signals present irregular oscillations, such as can be found in complex systems that have a chaotic attractor. Techniques derived from the theory of non-linear dynamics have been applied to E E G signals, with the expectation that the analysis of these time series may reveal essential properties of the underlying neural networks. These techniques involve the computation of correlation dimension, Lyapunov exponents a n d / o r Kolmogorov entropy. Accordingly, the number of degrees of freedom necessary for the description of the network can be estimated. In general, it has been shown (cf., Babloyantz 1989) that beta rhythmic activity needs more than 10 degrees of freedom for its description, alpha activity appears to need a smaller number of degrees of freedora (correlation dimension around 6 or 8), during sleep stages II, III and IV the number of degrees of freedom decreases, whereas during coma and petit mal epilepsy this number reaches a low value, implying that in this state the dynamics of the networks can be described by only a small number of parameters. Recently Pijn (1990) has shown that in the course of the development of an epileptic seizure of focal onset, the

F. LOPES DA SILVA

E E G of brain areas that become progressively involved in seizure activity switches from a large value of correlation dimension to a much smaller value. In those cases where such a low correlation dimension is found, it may be concluded that the corresponding network can be described as behaving as a chaotic attractor. However, in those cases where a rather high dimension is found, such as in the case of E E G alpha or beta activity, one has to be reserved with respect to such an interpretation. Indeed we have to recognize that it is difficult to estimate with precision whether a chaotic attractor of high dimension may be present when the time series available for analysis is necessarily rather short, in view of the non-stationary character of the E E G signals. In these cases it is difficult to distinguish such an apparent high dimensional attractor from illtered noise. This problem merits further investigation. Nevertheless it is clear that E E G signals may present transitions, or bifurcations, between states such as found in complex systems with non-linear dynamics. A specially interesting property of such systems is that it is possible for a transition between oscillatory and non-oscillatory behavior to occur rather abruptly, depending on the parameters of the system (Glass and Mackay 1988). In this way the 'gating function' described above may be performed in an optimal form.

Conclusions rhythms

on

functional

implications

of brain

Our knowledge of the mechanisms responsible for the generation of E E G rhythmic activity has increased significantly in the last decade. To this development a number of factors have contributed significantly; namely (i) the microphysio[ogical investigations of thalamic neurons, in particular of their intrinsic and synaptic properties, (ii) the application of quantitative methods of signal analysis to estimate the correlation (linear and non-linear) between E E G signals recorded from different brain areas, and (iii) the construction of mathematical and computer models of neuronal networks aimed at simulating how E E G signals are generated by such networks. However, only a few aspects of the E E G have been analyzed employing the approaches discussed here, namely sleep and barbiturate spindles, the alpha rhythm, the rhythmic slow activity (theta rhythm) of the limbic cortex, and the fast rhythms of olfactory and visual areas. Further basic research aimed at clarifying many aspects of these and other E E G rhythmic activities is necessary. Regarding the question of whether oscillations in neuronal networks, as reflected in brain waves, may be simple epiphenomena, or may subserve specific or general functions, 2 issues may be raised: the possibility that oscillations (i) may subserve a 'gating function"

N E U R A L M E C H A N I S M S U N D E R L Y I N G BRAIN W A V E S

and (ii) may mediate the formation of assemblies of neurons that represent a given stimulus pattern. (i) Based on the experimental observations described above, pertaining to the generation of thalamic spindle activity (Fig. 3) and hippocampal rhythmic slow activity, it can be assumed that the change from a random pattern of activity to an oscillatory mode, in a neuronal population, results from the interplay of intrinsic and network properties. Accordingly, I propose that the occurrence of an oscillatory state in a neuronal network provides the conditions for setting the m e a n level of the m e m b r a n e potential within extensive populations of neurons. The fact that the occurrence of rhythmic activity may be rather fast, due to the non-linear characteristics of the neurons, appears to offer a very efficient way of changing the bias on the m e r e brane state within a large population of neurons. In this way the characteristics of the input-output transfer of information through the neuronal network can change dramatically. It appears that oscillatory mechanisms have emerged as the mechanism of choice for a neural mass, in the sense of Freeman (1975), to switch between different behavioral modes. The importance of the cholinergic systems of the forebrain in this switching mechanism was put forward above. (ii) In addition, the question must be considered of whether oscillations may play a functional role in the formation of neural representations of patterns of stimuli. This was first suggested for the olfactory systern where both the local E E G and the single cell activity present an oscillatory modulation in the freq u e n c y r a n g e of 40-90 Hz and are stimulus-dependent (Freeman 1975, 1988). A related phenomenon has been proposed for the visual cortex (Gray et al. 1988). Both single-unit activity and local field E E G records exhibit these oscillations. Furthermore, such oscillatory r e s p o n s e s t o specific visual stimuli appear t o o c c u r s y n chronously across cortical columns, although the strength of interaction decreases with increasing spatial separation (Engel et al. 1990). These experimental findings demonstrate that the local cortical E E G may reflect how underlying neuronal populations may form coherent domains of activity that represent essential features of sensory stimuli. It may be concluded that experimental evidence is accumulating that justifies the statement that E E G signals can reflect the functional states of neuronal networks. 1 acknowledge the excellent assistance of Cristine Knaap-Cabi and lna Ituijsen in composing the manuscript, of Simon Van Mechelen for the artwork, and the critical comments of Jan Pieter Pijn and Wytse W a d m a n .

91

References Adrian, E.D. and Matthews, B.H.C. The Berger rhythm: potential changes from the occipital lobes in man. Brain. 1934, 57: 355-385. Adrian, E.D. and Yamagiwa, D. The origin of the Berger rhythm. Brain, t935,58: 323-351. Babloyantz, A. Evidence of chaotic dynamics of brain activity during the sleep cycle. Phys. Len. (A), 1985, 111: 152-156. Babloyantz, A. Estimation of correlation dimensions from single and multichannel recordings: a critical view. ln: E. Ba~ar and T.H. Bullock (Eds.), Brain Dynamics. Springer, Berlin, 1989: 122-131. Ballard, D.H., Hinton, G.E. and Sejnowski, R.J. Parallel visual computation. Nature, 1983, 306: 21-26. Ba,~ar, E. Chaotic dynamics and resonance p h e n o m e n a in brain function: progress, perspectives and thoughts. In: E. Ba~ar and T.H. Bullock (Eds.), Chaos in Brain Function. Springer, Berlin,

19911: 1-31. Basar, E. and Bullock, T.H. (Eds.). Chaos in Brain Function. Springer, Berlin, 1989. Boeijinga, P.H. and Lopes da Silva. F.tt. Differential distribution of /3 and 0 EEG activity in the entorhinal cortex of the cat. Brain Rcs., 1988,448:272 286. Bower, J.J., Montaron, M.F., Vahn~e, J.M., Albert, M.P. and Rougeul, A. Anatomical localization of cortical beta rhythms in

cat. Neuroscience, 1987, 22: 863-869. Braitenberg, V. On the Texture of Brains. Springer, New York, 1977. Bressler, S. The g a m m a wave: a cortical information carrier'? Trends Neurosci., 1990, 13:161 162. Bressler. S.L. and Freeman, W.J. Frequency analysis of olfactory system E E G in cat, rabbit, and rat. Electroenceph. clin. Neurophysiol., 19811, 50: 19-24. Bullock, T.H. and McClune, M.C. Lateral coherence of the electroencephalogram: a new measure of brain synchrony. Electroenceph. clin. Neurophysiol., 1989, 73: 479-498.

Buzsaki,G., Bickford, R.G., Ponomareff, G., Thai, L.J., Mandel, R. and Gage, F.H. Nucleus basalis and thalamic control of neocortical activity in the freely moving rat. J. Neurosci.. 1988, 8: 40074026. Chow, K.L. and geiman, A.L. Aspects of the structural and functional organization of the neocortex. Ncurosci. Res. Prog. Bull., 1,~71t, s: 157-183. Creutzfeldt, O., Grunewald, G., Simonova, O. and Schmitz, H. Changes of the basic rhythms of the EEG during the performance of mental and visuomotor tasks. In: C R . Evans and T.B. Mulholland (Eds.), Attention in Neurophysiology. Butterworth, London, 1969: 148-168. Crunelli, V. and [,eresche, N., A role t~r G A B A B receptors in excitation and inhibition of thalamocortical cells. Trends Neu-

rosci., 1991, 14: 16-21. De France, J. and Sheer, D.E. Focused arousal, 411-Hz E E G and motor programming. In: D. Giannitrapani and L. Murri (Eds.), The E E G of Mental Activities. Karger. Basel, 1988: 153-168. De Munck, J.C. A Mathematical and Physical Interpretation of the Electromagnetic Field of the Brain. Ph.D. Thesis. University of Amsterdam, 1989. Detari, L. and Vanderwolf, C.H. Activity of cortically projecting and other basal forebrain neurons during large stow waves and cortical activation in anesthetized cat. Brain Res., 1987, 437:1 8, Eckhorn, R., Bauer, R., Jordan, W., Brosch, M., Kruse, W., Munk, M. and Reitboeck, H.J. Coherent oscillations: a mechanism of feature linking in the visual cortex'? Biol. Cybern., 1988, 60: 121-1311.

92 Eckhorn, R., Bauer, R. and Reitboeck, H.J. Discontinuities in visual cortex and possible functional implications: relating cortical structure and function with multielectrode/correlation techniques. In: E. Ba~ar and T.H. Bullock (Eds.), Brain Dynamics. Springer, Berlin, 1989: 267-278. Edelman, G.M. The R e m e m b e r e d Present. A Biological Theory of Consciousness. Basic Books, New York, 1989. Engel, A.K.. K6ning, P., Gray, Ch.M. and Singer, W. Stimulus-dependent neuronal oscillations in cat visual cortex: inter-columnar interaction as determined by cross-correlation analysis. Eur. J. Neurosci., 19911, 2: 588-6116. Freeman, W.J. Mass Action in the Nervous System. Academic Press, New York, 1975. Freeman, W.J. Nonlinear neural dynamics in olfaction as a model for cognition. In: E. Ba~ar (Ed.), Dynamics of Sensory and Cognitive Processing of the Brain. Springer, Berlin, 1988: 19-29. Freeman, W.J. and Van Dijk, B.W. Spatial patterns of visual cortical fast LEG during conditioned reflex in a rhesus monkey. Brain Res., 1987, 422: 267-276. Gevins, A.S. and Bressler, S.L., Functional topography of the h u m a n brain. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Hans Huber, Toronto, 1988: 99-117. Gevins, A.S., Morgan, N.H., Bressler, S.L., Doyle, J . C and Cutillo, B.A. Improved E R P estimation via statistical pattern recognition. Electroenceph. olin. Neurophysiol., 1986, 64: 177-186. Gevins. A.S., Morgan, N.H., Bressler, S.L., Cutillo, B.A., White, R.M,, llles, J., Gree. D.S., Doyle, J.C. and Zeitlin, G.M. t t u m a n neuroelectric patterns predict performance accuracy. Science, 1987. 235: 580-585. Glass, L. and Mackay, M.C. From Clocks to Chaos: the Rhythms of Life. Princeton University Press, Princeton, N J, 1988:248 p. Glenn, L.L. and Steriade, M. Discharge rate and excitability of cortically projecting intralaminar thalamic neurons during waking and sleep states. J. Neurosci., 1982, 2: 1387-1404. Gray, C., K6nig, P., Engel, A. and Singer, W. Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which rcflects global stimulus properties. Nature, 1988, 338: 334-337. Hogan, K. and Fitzpatrick, J. The cerebral origin of the alpha rhythm. Eleclroenceph. clin. Neurophysiol., 1988, 69: 79-81. Jahnscn, H, and LlinAs, R. Electrophysiological properties of guineapig thalamic neurones: an m vitro study. J. Physiol. (Lond.), It,~84a, 349: 205-226. Jahnsen, H. and l,lim~_s, R. Ionic basis for the electroresponsiveness and oscillatory properties of guinea-pig thalamic neurones in vitro. J. Physiol. (Lond.), 1984b, 349: 227-247. Katchalsky, A.R., Rowland, V. and Blumenthal, R. Dynamic patterns o1' brain cell assemblies. Neurosci. Res. Progr. Bull., 1974, 12: 1-187. Klimesch, W., Pfurtscheller, G. and Mohl, W. E R D mapping and long-term memory: The temporal and topographical pattern of cortical activalion. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Hans Huber, Toronto, 1988: 131 143. Lehmann, D., Ozaki, H. and Pol, 1. L E G alpha maps series: brain microstates by space-oriented adaptive segmentation. Electroenceph. olin. Neurophysiol., 1987, 67: 271-288. Levey, A.I., Hallanger, A.E. and Wainer, B.H. Cholinergic nucleus basalis neurons may influence the cortex via the thalamus. Neurosci, Lett., 1987, 74: 7-13. Lopes da Silva, F.H. The rhythmic slow activity (theta) of the limbic cortex: mechanisms of generation, models and functional implications. In: T.H. Bullock and E. Ba§ar (Eds.), Induced Rhythms of the Brain. MIT Press, Cambridge, MA, 1991: in press, Lopes da Silva, F.tl. and Storm van Leeuwen, W. The cortical source of the alpha rhythm. Neurosci. Lett., 1977, 6: 237-241. Lopes da Silva, F.H. and Storm van Leeuwen, W. The cortical alpha rhythm in dog: the depth and surface profile of phase. In: M.A.B.

F. LOPES DA SILVA Brazier and H. Petsche (Eds.), Architectonics of the Cerebral Cortex. Raven Press, New York, 1978: 319-333. Lopes da Silva, F.H., Van Rotterdam, A., Storm van Leeuwen, W. and Tielen, A.M. Dynamic characteristics of visual evoked potentials in the dog. II. Beta frequency selectivity in evoked potentials and background activity. Electroenceph. clin. Neurophysiol., 197{I, 29: 260-268. Lopes da Silva, F.tt., Van Lierop, T.H.M.T., Schrijer, C.F.M. and Storm van Leeuwen, W. Organisation of thalamic and cortical alpha rhythms: spectra and coherences. Electroenceph. clin. Neurophysiol., 1973a, 35: 627-639. Lopes da Silva, F.H., Van Lierop, T.H.M.T., Schrijer, C.F.M. and Storm van Leeuwen, W. Essential differences between alpila rhythms and barbiturate spindles: spectra and thalamo-cortical coherences. Electroenceph. clin. Neurophysiol., 1973b. 35: 640-647. Lopes da Silva, F.tt., Hoeks. A., Smits. H. and Zetterberg, L.H. Model of brain rhythmic activity - the alpha rhythm of thalamus. Kybernetik, 1974, 15: 27-38. Lopes da Silva, F.H., Van Rotterdam, A., Barts, P., Van Heusden, E. and Burr, W. Models of neuronal populations: the basic mechanisms of rhythmicity. In: M.A. Corner and D.F. Swaab (Eds.), Progress in Brain Research, Vol. 45. Elsevier, Amsterdam, 1976: 281-308. Lopes da Silva, F.H., Vos, J.E., Mooibroek, J. and Van Rotterdam, A. Partial coherence analysis of thalamic and cortical alpha rhythms in dog - a contribution towards a general model of the cortical organization of rhythmic activity. In: G. Pfurtscheller el al. (Eds.), Rhythmic L E G Activities and Cortical Functioning. Elsevier, Amsterdam, 1980a: 33 59. Lopes da Silva, F.H., Vos, J.E.. Mooibrock, J. and Van Rotterdam, A. Relative contributions of intracortical and thalamo-cortical processes in the generation of alpha rhythms, revealed by partial coherence analysis. Electroenceph. clin. Neurophysiol., 1980b, 511: 449-456. Lopes da Silva, F.H.. Witter, M.P., Boeijinga, P.H. and Lohman, A.H.M. Anatomic organization and physiology of the limbic cortex. Physiol. Rev., 19911, 71): 453-511. McCormick, D.A. and Prince, D.A. ACh induces burst firing in thalamic reticular neurones by activating a K ~ conductance. Nature, 1986, 319: 4112-405. McCormick, D.A. and Prince, D.A. Actions of acetylcholine in the guinea pig and cat medial and lateral geniculate nuclei, in vitro. J. Physiol. (Lond.), 1987, 392:147 165. McCormick, D.A. and Prince, D.A. Noradrenergic modulation of firing pattern in guinea pig and cat thalamic neurons, in vitro. J. Neurophysiol., 1988, 59: 978-996. Montaron. M.F., Bouyer, J.J., Rougcul, A. and Buser, P. Ventral mesencephalic t e g m e n t u m (VMT) controls electrocortical beta rhythms and associated attentive behaviour in the cat. Behav. Brain Res., 1982, 6:129 145. Mulle, C., Madariaga, A. and Desch~nes, M. Morphology and electrophysiological properties of reticularis thalami neurons in cat. In vivo study of a thalamic pacemaker. J. Neurosci., 1986, 6: 2134-2145. Nunez, P.L. Generation of h u m a n LEG by a combination of tong and short range neocortical interactions. Brain Topogr., 1989, 1: 199-215. Nunez, A., Garc~a-Austt, E. and Bufio. Jr., W. Intracellular theta rhythm generation in identified hippocampal pyramids. Brain Res., 1987, 416: 289-31/1/. Pape, H.-C. and McCormick, D.A. Noradrenaline and serotonin selectively modulate thalamic burst firing by enhancing a hyperpolarization-activated cation current. Nature, 1989, 340: 715-718. Pfurtscheller, G. and Aranibar, A., Event-related cortical desynchronization detected by power measurements of scalp EEG. Elcctroenceph, clin. Neurophysiol., 1977, 42: 817-82(3.

NEURAL MECHANISMS UNDERLY|NG BRAIN WAVES Pfurtscheller, G., Steffan, J. and Maresch, H., ERD mapping and functional topography: temporal and spatial aspects. In: G. Pfurtscheller and F.H. Lopes da Silva (Eds.), Functional Brain Imaging. Hans Huber, Toronto, 1988:117-131. Pijn, J.P. Quantitative Evaluation of EEG Signals in Epilepsy; Nonlinear Associations, Time Delays and Nonlinear Dynamics. Ph.D. Thesis. University of Amsterdam, 1990. Ray, W.J. and Cole, H.W. EEG alpha activity reflects attentional demands and beta activity reflects emotional and cognitive processes. Science, 1985, 228:750 752. Rogawski, M.A. and Aghajanian, G.K. Activation of lateral geniculate neurons by norepinephrine: mediation by an a-adrenergic receptor. Brain Res., 1980a, 182: 345-359. Rogawski. M.A. and Aghajanian, G.K. Modulation of lateral geniculate neurone excitability by noradrenaline microiontophoresis or locus coeruleus stimulation. Nature, 1980b, 187: 731-734. Rougeul-Buser, A., Bouyer. J.J., Montaron, M.F. and Buser, P. Patterns of activities in the ventrobasal thalamus and somatic cortex SI during behavioural immobility in the awake cat: focal waking rhythms. Exp. Brain Res. [Suppl.], 1983, 7: 69-87. Sillito, A.M., Kemp, J.A. and Berardi, N. The cholinergic influence on the function of the cat dorsal lateral geniculate nucleus (dLGN). Brain Res., 1983, 280: 299-307. Steriade, M. The excitatory-inhibitory response sequence in thalamic and neocortical cells: state-related changes and regulatory systerns. In: G.M. Edelman, W.E. Gall and W.M. Cowan (Eds.), Dynamic Aspects of Neocortical Function. Wiley, New York, 1984: 107-157. Steriade, M. and Desch~nes, M. lntrathalamic and brainstemthalamic networks involved in resting and alert states. In: M. Bentivoglio and R. Spreafico (Eds.), Cellular Thalamic Mechanisms. Elsevier, Amsterdam, 1988: 51-76.

93 Steriade, M. and Llinfis, R.R. The functional states of the thalamus and the associated neuronal interplay. Physiol. Rev., 1988, 68: 649-742. Steriade, M., Desch~nes, M., Domich, L. and Mulle, C. Abolition of spindle oscillations in thalamic neurons disconnected from nucleus reticularis thalami. J. Neurophysiol., 1985, 54: 1473-1497. Steriade, M., Domich, L., Oakson, G. and Desch~nes, M. The deafferented reticularis thalami nucleus generates spindle rhythmicity. J. Neurophysiol., 1987, 57: 260-273. Steriade, M., Jones, E.G. and Llin~is, R.R. Thalamic Oscillations and Signaling. Neuroscience Institute Publications. Wiley, New York, 1990a: 431 p. Steriade, M., Gloor, P., Llinfis, R.R., Lopes da Silva, F.H. and Mesulam, M. Report of the IFCN Committee on basic mechanisms. Electroenceph. clin. Neuropl-lysiol., 1990b, 76: 481-509. Szent~gothai, J. The local neuronal apparatus of the cerebral cortex. In: P. Buser and A. Rougeul-Buser (Eds.), Cerebral Correlates of Conscious Experience. INSERM Symp. No, 6. Elsevim/NorthHolland Biomed. Press, Amsterdam, 1978: 131-138. T6mb61, T. Comparative data on the Golgi architecture of interneurons of different cortical areas in cat and rabbit. In: M.A.B. Brazier and H. Petsche (Eds.), Architectonics of the Cerebral Cortex. Raven Press, New York, 1978: 59-76. Van Rotterdam, A., Lopes da Silva, F.H., Van der Ende, J., Viergever, M.A. and Hermans, A.J. A model of the spatial-temporal characteristics of the alpha rhythm. Bull. Math. Biol., 1982, 44: 283-305. Von der Malsburg, C, and Singer, W. Principles of cortical network organization. In: P. Rakic and W. Singer (Eds.), Neurobiology of Cortex. Wiley, New York, 1988: 69-99.