Jump to ContentJump to Main Navigation
Cortical Oscillations in Health and Disease$

Roger Traub, MD and Miles Whittington, PhD

Print publication date: 2010

Print ISBN-13: 9780195342796

Published to Oxford Scholarship Online: May 2010

DOI: 10.1093/acprof:oso/9780195342796.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (oxford.universitypressscholarship.com). (c) Copyright Oxford University Press, 2021. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 26 January 2021

Overview of In Vivo Cortical Oscillations

Overview of In Vivo Cortical Oscillations

(p.31) 3 Overview of In Vivo Cortical Oscillations
Cortical Oscillations in Health and Disease

Roger D. Roger

Miles A. Whittington

Oxford University Press

Abstract and Keywords

brain oscillations occur spontaneously, for example during sleep; as part of evoked responses following specific sensory stimulations; and as a correlate of cognitive and motor processes. In the latter cases, there is evidence that the oscillations are an essential part of the brain computations being performed, motivating study of the basic cellular mechanisms. Cognitively relevant oscillations are often quite fast, above 20 Hz, and even up to several hundred Hz.

Keywords:   sleep, sensory stimulation, cognitive task, evoked responses, fast rhythms

The experimental study of pathological brain oscillations is inextricably mingled with the study of normal oscillations; it is impossible to disentangle them. One must, however, start somewhere, and it is probably appropriate to begin with normal, physiologically occurring brain rhythms, as they occur in vivo—but with our discussion informed by (one might say prejudiced by) our understanding of the cellular mechanisms: mechanisms as determined by in vitro and simulation analyses, to be elaborated upon in later chapters. Because it is our wish that the discussion be so informed, we are constrained as to which oscillations we can consider; there are more in vitro data relevant to some oscillations than to others. For this reason, we shall concentrate in this chapter on oscillations of the following frequencies: gamma (30–70 Hz), beta (10–30 Hz), and VFO (very fast oscillations, >70 Hz). Biological functions can be reasonably hypothesized for gamma, beta, and VFO. In contrast, the slow oscillation (<1 Hz; Steriade et al., 1993d) seems to arise via biochemical mechanisms rather than—if one may put it thus—neuronal processing mechanisms; one function of the slow oscillation may be to support brief epochs of the other oscillations we want to consider, namely, gamma, beta, and VFO. The most widely known cortical oscillation is probably the alpha rhythm, but in vitro analysis of its cellular mechanisms is yet preliminary; we refer the reader to standard textbooks for the clinical features of the EEG alpha (and related) rhythms (Ebersole & Pedley, 2003; Niedermeyer & Lopes da Silva, 1999), and to recent papers on the cellular mechanisms of thalamically generated alpha-frequency oscillations (Hughes et al., 2004; Lörincz et al., 2008).

(p.32) Synchronized neuronal network oscillations arise so readily in brain slice systems, in so many different experimental paradigms, that one is tempted to think that oscillations are inevitable, a straightforward consequence of the intrinsic properties of neurons, and the means by which neurons influence one another, through chemical synapses and gap junctions. Such a view can be rejected. Not all slice preparations oscillate. In the living brain, oscillations occur in some places and not other places, in some states of awareness or attention or movement and not others. One needs to consider not only the cellular mechanisms of each particular oscillation—best addressed with in vitro methods—but also the general conditions under which an oscillation can occur in vivo, and how the brain goes about selecting which neurons are to participate—or, alternatively, which neurons are not to participate. We assume that the significance of any brain oscillation is determined by precise (although not necessarily known) physical characteristics: which neurons are involved; the degree to which axons, somata, and dendrites participate suprathreshold or infrathreshold (i.e., whether different portions of the neurons fire or not); the phase relations between individual neuronal activities and the population as a whole; the ways in which the oscillating neurons affect the activities of “downstream” neurons; and the plastic changes induced, or at least facilitated, by the oscillation. Above all, we must avoid being seduced by mystical concepts such as that synchrony or oscillations matter in and of themselves.

The phenomenology of brain oscillations is sufficiently complex that it will help us to have some conceptual framework, which will allow us to formulate specific hypotheses as to what the oscillations are for, and how the different sorts of oscillations are interrelated. As a prerequisite for describing such a framework, however, some general—and counterintuitive—remarks about brain function are required. Specifically, we need to describe two “modes” in which networks of neurons can operate. We call the two modes “standard” and “nonstandard.”

The Standard Mode of Neuronal Network Operation

This is the operating mode with which all students of neurophysiology are familiar (Kandel et al., 2000), and incorporates concepts going back to Charles Sherrington (1857–1952; Sherrington, 1947) and Donald Hebb (1904–1985; Hebb, 1949). There are two main ideas (Fig. 3.1A). First, neurons “integrate” their inputs, “adding up” their synaptic inputs [even if the details are actually nonlinear (Llinás, 1975; Rall, 1962)], and generating a sequence of action potentials that are a physical consequence of the synaptic inputs (Llinás, 1975; Rall, 1962) so that action potentials propagate forward along the axon and backward into the dendrites (Stuart & Sakmann, 1994), evoking a transient calcium concentration change (Spruston et al., 1995). Second, this activity-dependent dendritic calcium signal in turn secondarily initiates biochemical (p.33)

                      Overview of In Vivo Cortical Oscillations

Figure 3.1 Two modes of spike initiation: Orthodromic spikes can backpropagate to dendrites, whereas antidromic spikes may block in the proximal axon and not propagate into the dendrites. A: “Typically,” action potentials are initiated in the axon by somatic depolarization, itself induced by currents flowing from depolarized dendrites. Under these conditions, the somatic spike not only propagates down the axon, but also “backpropagates” into the dendrites (even if with some decrement); in doing so, the backpropagated spike can generate intradendritic signals that can be compared with the synaptic events that helped to trigger the spike: The synapse/neuron system is sensitive to temporal correlations of input and output. B: Under some—perhaps “atypical” conditions (but more commonly than realized)—spikes are initiated in the axon through the activities of electrically coupled axons, rather than through synaptically induced somatic depolarization. Such axonal spikes still propagate forward down the axon. Axonal spikes may invade the soma to generate full antidromic spikes that can backpropagate (not shown), but they may also decrement to give only a small “spikelet” in the soma, that decrements even further in the dendrites (Larkum et al., 2008; Schmitz et al., 2001). In neither case is there a meaningful presynaptic signal to which the backpropagated signal can be compared. Temporal correlations between input and output appear not to matter.

processes in synaptic specializations, the net effect of which is to influence plastic changes in synaptic efficacies—and, because the calcium signals are activity dependent—the changes of synaptic efficacies are consequences of the temporal correlations between synaptic inputs and synaptically induced action potential outputs: a process now called “STDP,” or spike-time-dependent synaptic plasticity (Amit & Brunel, 1997; Caporale & Dan, 2008; Dan & Poo, 2004; Paulsen & Sejnowski, 2000). Of particular interest and importance is the experimental observation, in accord with Hebb’s original intuition, that (p.34) potentiation (i.e., increase) in excitatory synaptic strength on pyramidal neurons occurs when the excitatory postsynaptic potentials (EPSPs) in question precede—i.e., contribute to the causation of—an action potential, although this time cannot be too long, typically on a time scale of tens of milliseconds (Bi & Poo, 1998; Markram et al., 1997b).

The Nonstandard Mode of Neuronal Network Operation

In this more recently recognized, and less well known, mode of operation, action potentials are initiated in axons of pyramidal cells in a manner independent of synaptic inputs, at least excitatory inputs (synaptic inhibition is able, however, to suppress action potential initiation) (Traub et al., 2003b). The action potentials may propagate forward to evoke transmitter release, especially on interneurons, but invade the soma and dendrites only on occasion (Cunningham et al., 2004a; Larkum et al., 2008; Schmitz et al., 2001; Traub et al., 2003b); mostly (Fig. 3.1B), axonal full spikes are associated only with “spikelets” in the soma and dendrites. As a result, most axonal outputs are not expected to correlate with dendritic calcium transients in pyramidal cells; nor are EPSPs on the pyramidal cells causal for the action potentials in those cells. Unexpectedly, a number of cortical oscillations, at beta-2 (20–30 Hz), gamma, and very fast frequencies reflect this nonstandard mode of neuronal network operation, at least in vitro. This, in turn, raises major questions as to what the function of cortical oscillations can possibly be.

Neuronal Networks Operating in Standard Mode

Synchronized oscillations have natural consequences for neuronal networks operating in the standard mode, although these consequences are not always intuitive. Synchronized network oscillations, by definition, introduce temporal correlations in the firing times of the neuronal somata that participate in the oscillations. Such temporal correlations in turn have consequences both for synaptic integration and for spike-time-dependent synaptic plasticity (STDP), provided that axonal conduction delays are not too large. For synaptic integration, synchrony provides for the possibility of having EPSPs—generated by distinct presynaptic neurons—adding up in a given postsynaptic neuron, the addition taking place with precision determined by the tightness of oscillation synchrony and by the dispersion in conduction delays. This type of synaptic addition is called coincidence detection, and can be important in allowing for individually subthreshold EPSPs to contribute to firing a neuron, especially when EPSPs are rapidly truncated by synaptic inhibition (Abeles, 1982; Azouz & Gray, 2000; Egger et al., 1999; König et al., 1996; Llinás et al., 2002; Prescott et al., 2006; Rodgers et al., 2006). For STDP, synchrony provides for temporally defined relationships between synaptic inputs to, (p.35) and outputs from, multiple neurons; this principle can apply in three different ways:

  1. 1. For common synaptic inputs to a pool of synchronously firing neurons, there will be correlations in the type of plasticity occurring for these inputs onto all of the neurons in the pool.

  2. 2. For synaptic interactions between synchronously firing neurons, there will obviously be temporal correlations between the synaptic inputs to, and the action potential output from, each of the neurons in the pool (i.e., SDTP rules should apply) —and so, secondarily, there should be correlations in the plastic changes induced on each of the neurons. [It needs emphasis, however, that the plastic changes may not always be in the direction one expects, if conduction delays within the synchronously oscillating pool are large enough (Fig. 3.2)].

  3. 3. Likewise, the synaptic outputs of a synchronously oscillating pool of neurons will have correlated plastic changes on their respective postsynaptic targets, provided the target neurons are themselves synchronized with one another. For the output synaptic contacts to have correlated plastic changes—that is, the synapses all tending to

                          Overview of In Vivo Cortical Oscillations

    Figure 3.2 Synchrony of somatic firing does not necessarily imply synchrony of axon-terminal and somatic spikes. Schematic of two pyramidal cells that fire synchronously, as determined by somatic measurements. Spike-timing–dependent plasticity (STDP) is expected, however, to depend on the temporal relations between the presynaptic terminal spike, and the dendritically back-propagated somatic spike. In the case illustrated here, where there are 10-ms axon conduction delays, the presynaptic terminal spikes occur 10 ms before the nearby somatic spikes; either no synaptic plasticity, or synaptic depression, might be the result.

    (p.36) change in the same direction, either up or down—it is not required that the presynaptic and postsynaptic pools be synchronized with each other, only among themselves.

The observant reader will have noticed that the aforementioned consequences, for synaptic integration and plasticity, depend—in the standard mode—only on synchrony, and not on oscillation. The concept of synchrony is, logically and physically, distinct from the concept of oscillation. Why handcuff the two concepts together? A truly basic question. It turns out, however, that experimentally—both in vivo and in vitro—synchrony and oscillations do in fact go together; and when one examines the mechanisms that cause many interconnected neurons all to oscillate at about the same frequency (as we shall do later), one sees that these same mechanisms provide for synchrony. The oscillation and the synchronization emerge together.

Relevance of Nonstandard Synchronized Oscillations

If coincidence detection, and correlated synaptic plasticity, operates for synchronized oscillations in the “standard mode,”, but not in the “nonstandard mode,” why does the brain even have synchronized oscillations in the “nonstandard mode”? This is a paradox. Certainly, coincidence detection—of synaptic inputs—is not relevant when spikes are initiated purely by electrophysiological processes in the axons. Likewise, synaptic plasticity is not likely to occur when synaptic inputs are not causing spike outputs, and where most spikes do not backpropagate into the dendrites. It is not apparent, then, how oscillations generated in “nonstandard” fashion, by axonal and nonsynaptic mechanisms, could be relevant to brain function. While a solution to this paradox is not known definitively, we can offer a simple hypothesis:

Hypothesis: “standard mode” and “nonstandard mode” network behaviors may interact in multiple ways. Consider two networks, A and B, with A nonreciprocally connected to B. Standard-mode synaptic activity in downstream network B [particularly activity involving gamma-aminobutyric acid (GABA) release] may generate action potentials in the projection axons from network A. [This can happen because, as we elaborate upon in later chapters, GABA can excite axons.] These axonal action potentials can antidromically propagate back to network A to provide background drive for nonstandard mode patterns of activity. Such “ectopic” action potential generation is seen in CA3-to-CA1 connections in hippocampus (Avoli et al., 1998; Stasheff et al., 1993a,b), in thalamocortical fibers (Pinault & Pumain, 1989), and may be widespread in the peripheral sensory nervous system (Pinault, 1995). Conversely, oscillations in network A, generated in “nonstandard” mode, may generate synaptic effects in network B. Thus, coincidence detection, and correlated synaptic plasticity, can occur in the downstream network B neuronal targets, even if these mechanisms are not taking place in the “primary” (p.37) oscillating population. Similarly, synaptic input from network A may also influence nonstandard network activity in target network B. Feedforward inhibition via somatodendritic GABA receptors can terminate axonally generated spike propagation quite potently (see Whittington & Traub, 2003). Standard-mode action potential generation, following dendritic excitation, also produces a large, transient calcium load. Even single somatic action potentials lead to large, long-lasting calcium concentration rises in primary axons and collaterals (Baudoux et al., 2003). Such increases in intracellular calcium could act to block gap junctional conductances required to maintain “nonstandard mode” oscillations.

Properties of Some Selected Cortical Oscillations

Gamma Oscillations Induced by Sensory, Particularly Visual, Input

The approximately simultaneous discovery of visually induced cortical gamma oscillations, by two groups in Germany (Eckhorn et al., 1988; Gray & Singer, 1989), had an immediate—and continuing—excitatory effect on systems neuroscience. The discoveries were not only exciting in themselves, but they seemed to point the way to a vital (and unexpected, at least to most people) link between cognition and cellular electrophysiology. The path seemed open to the materialists’ dream: a scientific program for learning about the mind through a defined series of practical neurophysiological experiments. Many were caught up in this dream, including the authors. Although some of the psychological concepts had been discussed earlier, by, for example, von der Marlsburg (von der Malsburg & Buhmann, 1992; von der Malsburg & Schneider, 1986), it is our opinion that much of the credit for the excitement must go to Wolf Singer (b. 1943); Singer saw how the electrophysiological observations took on biological relevance and how this relevance might be best further analyzed. The key idea was to attach meaning to the existence of, or lack of existence of, synchronization of spatially separated neuronal networks, each oscillating at gamma frequency; and to devise an intricate series of in vivo experiments supporting the logical consistency of this postulated meaning. (Of course, the first experimental requirement was to demonstrate that the oscillations existed at all, even at a single cortical site.)

Basic Phenomenology

The classical approach to sensory physiology usually (not always) involves documenting changes in firing rates—in sensory axons or neurons—induced by effective mechanical, visual, or other modality-appropriate stimuli (Hartline, 1938; Rosenblith, 1961). As the information transformations within sensory systems became understood (Hartline & Ratliff, 1958), and as complex—but still biologically relevant—stimuli were used, two avenues of progress developed. (p.38) One avenue led to the development of quantitative models of sensory organs, at least simpler ones (Passaglia et al., 1998). The other led to the recognition that cells in sensory regions of the nervous system, distal to the primary receptive organs, responded to special meaningful features of stimuli: for example a moving spot, fly-like, of interest to a frog (Lettvin et al., 1959); or small moving bars, located in particular areas of the visual field, which bars might be viewed as elementary components of larger and more complex objects that a cat would observe (Hubel & Wiesel, 1959, 1962, 1963). And of course, in retrospect, the existence of responses to meaningful stimuli is hardly surprising.

What was shown in the late 1980s and 1990s for the visual system was that populations of cortical neurons responded to small moving bars not only by changes in firing rates, but by generating locally synchronized gamma oscillations: that is to say, one could (a) record a visually evoked oscillating field potential; one could (b) find individual units firing spikes (or brief high-frequency bursts of spikes—“chattering”) which oscillated at the same frequency as the field; and (c) the unit oscillations and the field oscillations were tightly phase-locked (Eckhorn et al., 1988; Gray, 1994; Gray & Singer, 1989; Gray & McCormick, 1996; Gray et al., 1990, 1992). Oscillating units occurred mostly in superficial cortical layers, but could also be found in deeper layers (Gray et al., 1990; Livingstone, 1996); and the spatial scale over which oscillations were evoked by small stimuli was a few hundred microns, that is, on the scale of one or a few cortical columns (Eckhorn, 1994; Engel et al., 1991; Maldonado et al., 2000). There are temporal aspects to this local response besides the oscillation itself: the induced gamma oscillation has a latent period of tens of milliseconds to hundreds of milliseconds it lasts only a few hundreds of ms and the latent period and oscillation itself are not phase-locked to temporal aspects of the stimulus (although slower modulation is possible; see later) (Gray et al., 1992). Induced gamma oscillations are not an artifact of anesthesia or of the use of cats as an experimental preparation: they are found in awake cats (Gray & Viana di Prisco, 1997), nonhuman primates (Frien et al., 2000; Kreiter & Singer, 1992), and humans (Lachaux et al., 2000).

Once the brain has the possibility of generating fine temporal structure in its local responses (meaning the response of a pool of nearby neurons to a small visual stimulus), then it has the opportunities of exhibiting temporal structure in its responses on larger spatial scales: larger, that is, both in terms of the size of the stimulus, and in terms of the physical separation of the neurons that respond to the stimulus. One way that larger spatial scale temporal structure could, in principle, be expressed is via the synchronization of two or more pools of neurons, each responding to a “piece” of a large stimulus with its own gamma oscillations. Indeed, this is what happens, in at least some experimental settings (Fig. 3.3). The experiments that show synchronization between two separate cortical sites typically use a long moving bar, so that ends of the bar have distinctly separated visual fields, while still possessing the same orientation. Such a stimulus is spatially distributed, but nevertheless of a simple geometry. When the long moving bar stimulus is perturbed, by (p.39)

                      Overview of In Vivo Cortical Oscillations

Figure 3.3 Visual induced gamma oscillations depend on “global stimulus properties.” Data are from extracellular recordings at two sites in the visual cortex, of an anesthetized cat, responding to either a long moving bar (A), or two shorter bars moving in opposite directions (C). A and C also show schematically the receptive fields and orientation preferences (both vertical) of the two sites. With the single long bar, coherent oscillations at ~60 Hz occur at the two sites (B), whereas with two oppositely moving bars, there is nonsignificant (n.s.) coherence (D).

(Reproduced from Engel, Roelfsema et al., 1997 with permission.)

omitting the middle third, or by having the two halves move in opposite directions (Fig. 3.3), then between-site synchronization is lost, although localized sites may still oscillate (Gray et al., 1989).

Binding by Synchrony

Synchronized visually evoked gamma oscillations are found not only with pairs of sites both located in primary visual cortex in one hemisphere; this type of synchronization is also found for sites at homologous locations in opposite hemispheres (Engel et al., 1991a); and in cases where one site is in primary and the other in association visual cortex (Engel et al., 1991b) (although it should be noted that some “association” visual areas also receive (p.40) direct afferents from the lateral geniculate nucleus (see, e.g., p. 25ff. of Payne & Peters, 2002)). Be that as it may, synchronization, with approximately millisecond-scale precision, has been shown to engage brain areas interconnected by collaterals longer than merely within-column axonal branches. For this and other reasons (see later), enormous importance has been attached to the phenomenon (e.g., Engel et al., 1997 and Engel & Singer, 2001, amongst many other reviews). Specifically, it is postulated that synchronization “binds” together all those brain areas responding to a single discrete object, thus giving that object an existence defined as distinct from all other objects, as well as distinct from the “background”—to use a Gestalt psychological means of expression. Other data (see later) are consistent with such an idea, yet the idea is still, in our opinion, something of an extrapolation from the actual observations (not always a drawback, however); apparently contradictory data have also been reported (Thiele & Stoner, 2003). The binding-by-synchrony hypothesis also needs to contend with the experimental observation that moving stimuli are much more effective at inducing oscillations than are fixed stimuli (Gray et al., 1990)—not in accord with personal experience in distinguishing the shapes of objects.

Associations of Synchrony with Behavior and Perhaps Awareness

There are two experiments, in particular, that support the existence of such an association, both performed in cats with surgically induced strabismic amblyopia (i.e., an extraocular muscle of one eye is sectioned, making it impossible for the cat to direct gaze for the two eyes onto a single object simultaneously—as a result, visual images induced by one eye eventually become “suppressed,” as behaviorally confirmed, and as occurs similarly in humans with congenital strabismus). In the first experiment (Roelfsema et al., 1994), it was shown that images projected onto the seeing eye, and onto the nonseeing eye, evoke similar increases in cortical neuronal firing rates; however, only for stimuli projected onto the seeing eye do pairs of neurons synchronize with each other. In the second experiment, performed on awake cats, gratings were projected to one eye or the other, or to both—but when presented to both eyes simultaneously, the gratings were orthogonal to each other, a so-called “rivalry” condition (Fries et al., 1997). Responses during the rivalry condition were compared with responses when only a single eye was stimulated. It was found that visual cortex responding to the seeing eye actually increased the degree of synchronized oscillation with rivalry, whereas cortex responding to the nonseeing eye decreased its degree of synchrony. Because, presumably, the responses to the seeing eye alone are “perceived,” this experiment establishes a correlation between presumed perception and the extent of synchronization of neuronal oscillations.

The oscillations studied by Fries et al. (1997) and Roelfsema et al. (1994), which are presumed to be related to perception, are gamma oscillations. Other studies, in humans and in nonhuman primates, have also suggested a relation (p.41) between perception and enhanced synchrony of gamma oscillations, as well as beta oscillations (Doesburg et al., 2005; Melloni et al., 2007; Tallon-Baudry et al., 2005). There is also a study reporting that enhanced interhemispheric coherence at alpha frequencies correlates with object recognition in humans (Mima et al., 2001b).

In Vivo Cellular Mechanisms: The Importance of Synapses and of “Brain Activation”

Extracellular recordings, of local field potentials and of single or multiple units, can take one only so far in understanding the cellular mechanisms of neuronal oscillations, or indeed of any sort of brain activity. Intracellular recordings are also essential (although still not sufficient to tell us all that we wish to know). Because of technical difficulties, however, rather few studies of in vivo visually evoked gamma oscillations have been undertaken. The studies that have been undertaken all demonstrate the critical role played by synaptic interactions in this type of oscillation (although the possible additional role of electrical coupling has not been addressed in vivo). An example of the data that have been obtained is shown in Figure 3.4, taken from Jagadeesh et al. (1992). These data confirm the extracellular observations that action potentials (in cortex oscillating in response to a visual input) are themselves oscillatory; but the data show as well that prominent subthreshold synaptic activity, at gamma frequency, is occurring, and presumably driving the action potentials, suggesting, but not definitively proving, that synaptic interactions actually are producing the oscillation. Even stronger evidence for a critical, and not just epiphenomenal, role of synaptic interactions would be abolition of the gamma oscillations under the influence of appropriate synaptic blockers—a very difficult experiment to do in vivo.

Further elegant intracellular studies have been undertaken to study the role of so-called chattering cells (fast rhythmic bursting cells or FRB cells in another terminology) in generating visually induced gamma. FRB cells exhibit an intrinsic gamma-frequency bursting rhythmicity (Cardin et al., 2005; Gray & McCormick, 1996; Llinás et al., 1991; Nuñez et al., 1992; Steriade et al., 1998) in response to sustained depolarizing currents, with intraburst firing rates at several hundred Hz (Brumberg et al., 2000) (Fig. 3.5). Not only do chattering/FRB cells respond to experimentally induced tonic depolarization with gamma-frequency bursting; at least some FRB pyramidal neurons, in visual cortex, respond to visual inputs in this way, as shown with intracellular recording (Gray & McCormick, 1996). In addition, unit recordings of visually induced gamma oscillations also indicate chattering (fast rhythmic bursting) [see the data of Dr. Charles M. Gray in Fig. 7.4 of Traub et al. (1999a)]. It is natural, then, to ask if FRB neurons are somehow “pacing” the induced gamma oscillation: for example, the visual input might hypothetically induce a slow depolarization in FRB neurons, which synchronize with one another in some undetermined fashion, and then project (so to speak) the gamma (p.42)

                      Overview of In Vivo Cortical Oscillations

Figure 3.4 Induced gamma oscillations, in cat visual cortex in vivo, are associated with gamma-frequency synaptic potentials. A, B: Whole-cell patch clamp recording of a complex cell in visual cortex, as a bar moves across the visual field, at two different holding potentials: this reveals synaptic potentials, together with (truncated) action potentials in B. C: Spike-triggered average of 8 traces similar to B, with the spikes themselves removed; this also reveals gamma-frequency synaptic oscillations. D: Spikes also have a gamma-frequency probability distribution.

(Reproduced from Jagadeesh et al., 1992 with permission.)

oscillation synaptically onto other neurons. As Cardin et al. (2005) have shown, however, this hypothesis appears to be false (Fig. 3.6). Instead, the visual cortex neurons that do exhibit gamma-frequency bursting (and these all seem to be so-called “simple cells,” a characterization dependent on how they respond to temporally modulated moving gratings) are exhibiting such bursting as a result of oscillatory synaptic inputs, not as a result of tonic depolarization. That is to say, FRB neurons are oscillating as part of a large synaptically interconnected network, and are not acting as pacemakers, at least not in any simple fashion. [It is the case, however, that FRB neurons play an important role in another type of gamma oscillation in vitro (Cunningham et al., 2004b)—more on this in another section.]


Figure 3.5 Some cortical neurons (chattering cells, or FRB [fast rhythmic bursting] cells) generate gamma oscillations intrinsically, upon steady depolarization. The traces show responses of an FRB neuron in cat primary visual cortex, recorded with a sharp microelectrode, responding to various current pulses (indicated by the arrows) and holding potentials. When depolarized enough (top trace), the cell fires tonically, but at intermediate tonic depolarizations (middle traces), there are beta and gamma frequency bursts; within each burst, the cell fires at several hundred Hertz.

(Reproduced from Cardin et al., 2005 with permission.)

Even if tonic depolarized FRB neurons are not driving visually induced gamma in any simple direct fashion, it is still likely that tonic depolarization of cortical neurons, perhaps in combination with a block of intrinsic membrane K+ conductances, is nevertheless required for the oscillations to occur. The reason for suspecting that this is the case derives from the potentiating effect on the oscillations, in vivo, of stimulating the mesencephalic reticular formation (Fig. 3.7; see also Herculaneo-Houzel et al., 1999). The mesencephalic reticular formation contains cholinergic neurons and is contiguous with the pedunculopontine tegmentum (PPT; Armstrong et al., 1983). Stimulation of PPT has been shown to induce slow EPSPs in cortical neurons, along with gamma oscillations throughout cortical columns (and reciprocally connected thalamic cells), via a muscarinic receptor-mediated effect (although the muscarinic receptors mediating this effect could lie on cells along a synaptic pathway rather than on the cortical neurons themselves) (Steriade & Amzica, 1996); interestingly, these latter type of gamma oscillations occur with brain stem stimulation alone, without the requirement for sensory input—as do the (p.44)

Figure 3.6 Even in FRB cells, visually evoked gamma oscillations are driven by synaptic inputs. An intracellularly recorded simple FRB cell is shown responding to a sinusoidal visual grating (top traces), at two membrane potentials: at rest (left), and significantly hyperpolarized (right). Filtered traces of the intracellular recordings (with spikes removed) are shown below. The grating induces waxing and waning gamma oscillations; these are present about to the same extent when the cell is hyperpolarized enough to prevent firing (right), indicating that the oscillations result from synaptic inputs.

(Reproduced from Cardin et al., 2005 with permission.)

gamma oscillations during sleep (see later). Behavioral attention (however this may be mediated) in awake primates also has a potentiating effect on visually induced gamma oscillations (Fries et al., 2001).

The synaptic mechanisms of in vivo visually evoked gamma are not known in any detail. Such mechanisms have been studied in an in vitro model of induced gamma, and these have been reviewed in our earlier monograph (Traub et al., 1999a; see also Chapter 1 of this book). Unfortunately, that model used hippocampal rather than neocortical slices, and so far has not been adapted to the neocortex, at least to our knowledge. Nevertheless, the in vitro hippocampal model, using tetanic electrical stimulation in lieu of physiological sensory inputs, is the only candidate now available. It is therefore appropriate to outline the basic features of tetanic gamma (Traub et al., 1996c, 1999c; Whittington et al., 1997a, 2001), which we consider as hypotheses concerning the events in vivo:

  1. 1. The in vitro tetanic oscillations (Chapter 1, Fig. 1) appear after a latency of tens to hundreds of milliseconds (as in vivo), in association with (p.45)

                          Overview of In Vivo Cortical Oscillations

    Figure 3.7 For synchronized visually induced oscillations to occur, the cortex must be properly activated. Data from anesthetized cat, with extracellular recording and stimulation. (A) Visual responses were recorded in primary visual cortex of the left hemisphere (LH) and right hemisphere (RH); in some cases, stimulation was given to the mesencephalic (midbrain) reticular formation, or MRF. The small boxes in (B) and (C) show peristimulus spike histograms, in response to visual input, with and without MRF stimulation: The amount of firing is roughly comparable in the two cases. The larger boxes show that gamma-oscillatory synchronization between the two hemispheres is, however, present only with MRF stimulation. Stimulation of the pedunculopontine tegmental nucleus is known to produce slow EPSPs in cortical neurons (Steriade & Amzica, 1996).

    (Reproduced from Engel & Singer, 2001 with permission.)

    slow intracellular depolarizations in both pyramidal cells (also seen in vivo; see Fig. 3.6) and in interneurons (not, to our knowledge, tested yet in vivo). The in vitro oscillations also have a duration comparable to that seen in vivo, hundreds of milliseconds.

  2. 2. The intracellular depolarizations, in vitro, are largely caused by activation of metabotropic glutamate receptors (mGluRs). It is not known whether this is the case in vivo. The mGluR activation seen in vitro is not strictly comparable to the metabotropic actions that would be induced in vivo (p.46) by stimulation of brainstem cholinergic nuclei: in the in vitro case, but not necessarily in vivo, the metabotropic activation is induced by the stimulus itself.

  3. 3. The synchrony and period of the in vitro gamma oscillations are primarily determined by GABAA receptor-mediated perisomatic inhibitory postsynaptic potentials (IPSPs). This is suspected to be the case in vivo, but not demonstrated directly.

  4. 4. Pyramidal cells and fast-spiking interneurons, in vitro, oscillate in phase.

  5. 5. Two in vitro hippocampal gamma-oscillating sites can synchronize tightly, despite axonal conduction delays probably greater than 5 ms. Modeling and mathematical analysis (Bibbig et al., 2001; Ermentrout & Kopell, 1998; Traub et al., 1996c) indicate that synchronization is realized by excitatory synaptic inputs to interneurons, arising from the opposite oscillation site; such inputs provide a temporal corrective signal, so to speak, and depend for their efficacy on the rapid time course of α-amino-3-hydroxyl-5-methyl-4-isoxazole-propionate (AMPA)–receptor-mediated excitatory postsynaptic conductances (EPSCs) in fast-spiking interneurons (Geiger et al., 1995, 1997; Miles, 1990). Experimental and simulated synchronized tetanic gamma oscillations, when evoked at two separate sites, are associated with spike doublets in fast-spiking interneurons (Chapter 1, Fig. 1), and the doublets appear to be essential in actually bringing the two-site synchronization about. When the time course of fast-spiking interneuronal EPSCs is slowed down by a transgenic manipulation, then—as predicted—two-site synchronization in the tetanic model is disrupted (Fuchs et al., 2001). These critical predictions on long-range oscillatory synchronization mechanisms remain to be examined in vivo, in a sensory-evoked oscillation paradigm.

Frequency Alone Is Not the Defining Characteristic of Induced Gamma

Sensory-induced oscillations in the turtle visual system (Prechtl, 1994) and in insect olfactory systems (Laurent, 1996; Laurent & Davidowitz, 1994) occur at beta frequencies, not gamma: there is nothing mystical about gamma, in the sense of some universal absolute. The sensible question posed by Singer and his colleagues—why gamma, in particular?—must be answered in terms of mechanism. We can offer two hypotheses as to why mammals use gamma frequencies for sensory-induced oscillations, and the presumed perceptual binding functions that these oscillations may subserve; each hypothesis is based on its own underlying assumption. First, suppose that what counts for sensory induced oscillations is the ability of two oscillating sites to synchronize with each other, without involving oscillations in intermediate tissue, and to synchronize despite considerable conduction delays between the two sites. Then, oscillations based on GABAA receptor-mediated IPSPs, generated by fast-spiking interneurons themselves excited by “narrow” EPSCs can do the (p.47) job; at least this is demonstrated to be true in principle, using in vitro studies. Whether such a mechanism is unique cannot be claimed; but, on the other hand, once evolution has come up with a mechanism that works, alternative mechanisms may not be required. Birds, having wings, do not require helicopter rotors as well. Second, gamma oscillations are believed to interface with synaptic plasticity, based on theoretical grounds, and there is experimental evidence (in vitro) that such an interface actually exists (Whittington et al., 1997b). In that case, then, the period of the gamma oscillation, of order tens of milliseconds, may need to interface with the time constants of processes that determine synaptic potentiation and depression; one of these time constants is that for intradendritic calcium concentration changes—and these have the same order of magnitude as does the gamma oscillation itself (Bibbig et al., 2001; Sabatini et al., 2002).

Other (Nonvisual) Means of Inducing Gamma

The reader must not assume that gamma oscillations are specific to the visual system; at the same time, in reviewing the literature, the reader must be careful to distinguish between gamma oscillations phase-locked to a stimulus (“evoked” oscillations, often examined in the auditory system), and “induced” gamma oscillations not phase-locked to a stimulus. The “induced” visual oscillations discussed in the preceding text are not phase-locked to the stimulus. Indeed, there are also auditory induced gamma oscillations that are not phase-locked to the input stimulus (Palva et al., 2002), and some of these auditory induced gamma oscillations have been proposed to be related to learning (Jeschke et al., 2008). Electrical stimulation of nonprimary auditory thalamus has also been shown to induce cortical gamma oscillations (Barth & MacDonald, 1996; Sukov & Barth, 2001). A rather fast (80 Hz) induced gamma oscillation has been reported, following painful median nerve (i.e., somatosensory) stimulation (Chen & Herrmann, 2001). Cortical gamma oscillations have also been reported in a patient with somatic hallucinations (Baldeweg et al., 1998), but the mechanism of these oscillations may be different than for somatosensory induced oscillations; they may, for example, be more similar to the gamma oscillations that normally occur during slow wave sleep (see later). And of course it has long been known that olfactory stimulation evokes gamma oscillations in multiple parts of the olfactory system (Adrian, 1942, 1950; Eeckman & Freeman, 1990; Neville & Haberly, 2003). Induced gamma oscillations are of general biological interest; the deeper principles involved cannot be completely specific to vision.

Critique of the Binding Hypothesis and Gamma Oscillations

Probably the deepest mechanistic and functional analysis of sensory induced oscillations has been undertaken by Gilles Laurent and his colleagues (Bazhenov et al., 2001; Friedrich & Laurent, 2001; Laurent, 1996, 2002; (p.48) Laurent et al., 1996; MacLeod & Laurent, 1996; MacLeod et al., 1998; Stopfer et al., 1997; Stopfer & Laurent, 1999; Wehr & Laurent, 1996), in the olfactory system of insects. One of the major thrusts of this work has concerned how distinct odors are classified, in the distributed firing patterns of neurons in one olfactory structure or another: as processing proceeds away from the sensory periphery, the representation becomes sparser, in that fewer cells respond to a given odor, and they do so with fewer action potentials. In addition, recognition of the distinction between odors is degraded when population oscillations are disrupted in the insect olfactory system. Evidently, oscillations are important in the analysis of sensory inputs, and in encoding the results of the analysis in such a way as to be useful to other parts of the nervous system. Still, it is not clear what analog “binding”—in the Gestalt sense—has in the olfactory system. There are, to be sure, many odors to be recognized, and they may be intermixed with each other, but is there an analog of a large “object” constituted of smaller “pieces,” so that oscillations generated by each piece need to be synchronized with each other? If extrapolations are possible from the insect olfactory system to the mammalian visual system, then we must consider the possibility that induced oscillations allow for, or contribute to, more effective processing of visual inputs, and for rendering sparser the cellular representation of seen objects; but we must find other tools and models to extend our understanding of the perceptual aspects of synchronized oscillations.

Perhaps an experimental approach would involve a hypothesis such as this: the occurrence of tight synchrony in sensory-induced oscillations allows us to talk about, or write about, what we have seen, now and in the future. To test the hypothesis, we need a method to titrate oscillation synchrony without abolishing consciousness—if such a manipulation is possible.

Fast Oscillations Associated with the Waking State: Expectancy and Short-Term Memory

Visually evoked gamma oscillations occur in anesthetized animals, as well as awake ones. Thus, although synchronized gamma oscillations may be necessary for conscious awareness, the reverse cannot be true. There are, however, other types of fast oscillations, particularly at beta-2 (20–30 Hz) frequencies, that occur during, and in preparation for, motor tasks; and they occur also in other types of cognitive tasks where short-term (lasting seconds) “working” memory is involved. That is to say, the oscillations seem, behaviorally, to be associated with the waking state—even though oscillations having similar morphology, and possibly similar cellular mechanisms, can be found during sleep and even in brain slices.

The behavioral setting of beta-2 oscillations is different than for sensory-induced oscillations: an expectant state before an action, as opposed to possibly passive sensory stimulation; and correspondingly, the cortical locations of beta-2 oscillations are different as well: instead of primary sensory cortices (p.49) and adjacent association cortex, where sensory-induced gamma tends to be found, beta-2 oscillations are found in primary motor, premotor, and supplementary motor cortex, as well as in lateral inferior parietal cortex—but, and one must be careful here, beta-2 is also found in a sensory region, the somatosensory cortex. One must note, however, that somatosensory cortex is both (1) strongly interconnected with motor cortex and also (2) one of the cortical regions of origin (i.e., besides “motor” regions) of the corticospinal tract (Miller, 1987; Rapisarda et al., 1985; Rathelot & Strick, 2007; Toyoshima & Sakai, 1982). To put it another way, somatosensory cortex is, in some sense, also motor cortex.

Delayed match-to-sample and related tasks are one example of a cognitive paradigm giving rise to beta oscillations. In such a task, a person or an animal is presented with a brief stimulus, let us say a visual pattern, which is chosen from a finite repertoire of possible stimuli. The individual is then supposed to remember what the stimulus was, for some period of time (typically one or a few seconds), so that when a second “test” stimulus is presented, that stimulus can be judged as the same or different than the original stimulus: the judgment is then expressed in the form of a motor action, again chosen from a finite repertoire. Alternatively, the subject may simply be expected to produce a specific action, determined by the original stimulus, but only after the delay period. Either of these paradigms tests whether the original stimulus is remembered, and both require immobility during the delay period.

A key set of observations consisted in the demonstration by Miyashita and colleagues (Miyahsita, 1988; Miyashita & Chang, 1988) that, during the delay period, there was a sustained increase in firing rates within particular localized brain regions (inferior temporal cortex for the Miyashita studies), and there was evidence that there might be stimulus specificity in which neurons increased their firing rates, and which did not: thus, stimulus 1 might cause sustained increased firing in neurons A, B, and C; while stimulus 2 would do this for neurons D, E, and F. Working memory was presumably identifiable, then, with a so-called population code: the identity of an object was determined by selecting a subset of neurons out of a group, tonically stimulating this subset, and omitting the stimulation (or actively suppressing) all the other neurons in the group. These experimental observations were subsequently extended from inferior temporal cortex to prefrontal cortex; the involvement of prefrontal cortex in working memory is discussed further in Chapter 5 (Parkinson’s disease).

Significantly, the earlier studies of increased firing, during the delay period, concentrated on the activities of single units or relatively small sets of nearby units (so-called multiunit recordings), rather than on local field potential recordings; the latter signals reflect local averaged synaptic currents, and are generally the most sensitive means of demonstrating that a local population is engaged in a collective oscillation. (Of course, in vivo, unlike the case in selected experimental preparations, synaptic conductances are generally operating, and will not have been suppressed by, say, receptor-blockers or by (p.50) low-calcium media. Thus, synaptic currents—reflecting as they do the summated activities of many cells—are a reliable means of detecting the presence of a collective neuronal oscillation, even if the underlying mechanisms that generate the oscillation do not critically depend on the presence of such currents.) In addition, the earlier investigators were not focused on the issue of whether a neuronal population was oscillating or not, but rather instead on increases and decreases in activation of one cell or another. Nevertheless, as Figure 3.8 demonstrates (in data from an awake macaque monkey, performing a task involving the selection of direction in which to make a saccade after a delay or “memory” period), the delay period is associated not just with increased neuronal activity, but with fast population oscillations as well (note especially Fig. 3.8B, left, and Fig. 3.8Cb). It is of special interest that the increased oscillations quickly become reduced at the end of the delay period, once action is actually taken. Similar observations have been reported by other investigators [in monkeys (Lebedev & Wise, 2000; MacKay & Mendonça, 1995); and in the work of Catherine Tallon-Baudry and her colleagues, in humans (Bertrand & Tallon-Baudry, 2000; Tallon-Baudry & Bertand, 1999; Tallon-Baudry et al., 1997, 1998a,b, 1999a,c, 2001)]. Even in rats, somatosensory cortex (specifically, barrel cortex) fast oscillations, at 25 to 45 Hz, precede an episode of exploratory whisking (Hamada et al., 1999).

Although our discussion has been focused on oscillations occurring during a delayed match-to-sample task, it is important to point out that approximately 25 Hz oscillations have been reported to occur in monkey somatosensory cortex, during intentional reaching movements—the retrieval of raisins, that is, motor activity involving active movement, and not expectant immobility (Murthy & Fetz, 1992, 1996a, 1996b); the oscillations were, however, quite brief, with a mean of about 4 cycles per oscillatory epoch. These oscillations appeared to arise in deeper cortical layers (local field potential maxima at 1 to 2 mm cortical depth; and the oscillations could synchronize between left and right motor cortices, presumably by virtue of the callosal connections between these areas (Jenny, 1979).

During the delay period, in a delayed match-to-sample task, the subject is immobile. Are the oscillations recorded during the delay period involved in holding a memory, or rather (or in addition) in maintaining immobility? When human subjects move a manipulandrum, epochs of immobility (when the manipulandrum is held in a fixed position) are associated with beta oscillations in somatosensory and motor cortex (Fig. 3.9). Figure 3.9 is taken from the work of Stuart Baker and his colleagues, who have in addition shown that motor/somatosensory beta oscillations—in this type of paradigm—are coherent with muscle electrical activity (EMG, electromyogram), and so presumably are coherent with oscillations in the spinal cord (oscillations that can generally not be directly recorded in humans) (Baker et al., 1997, 1999; Kilner et al., 2003; see also Donoghue et al., 1998). Other investigators, for example, Mima et al. (2000, 2001a), have demonstrated, in humans, coherence between motor cortex and muscle over the frequency range 14 to 50 Hz, extending into (p.51) gamma frequencies—coherence presumably mediated, at least in one direction, by the corticospinal tract. The fact that corticospinal tract (pyramidal tract) stimulation can reset motor-cortical beta oscillations is consistent with this notion (Jackson et al., 2002). One must not think exclusively, however, of a unidirectional transmission of motor-cortical beta oscillations to the spinal cord: somatosensory cortex also participates in motor-behavioral beta oscillations in vivo, in synchrony with motor cortex (Witham et al., 2007), and there is evidence that proprioceptive afferents also display oscillatory activity (Baker et al., 2006). With in vitro models, both motor cortex (Yamawaki et al., 2008) and somatosensory cortex (Roopun et al., 2006) have been shown capable of generating beta oscillations, in each case within the deep layers, and involving cells expected to give rise to corticospinal tract fibers (see later chapters). Both in vivo (Baker et al., 2003a) and in vitro, the firing rates of individual somata are lower than the frequency of the field beta: unfortunately, the signal processing within the spinal cord, during motor tasks in humans and nonhuman primates, is little understood. [It is also not known whether motor cortical beta oscillations are specifically transmitted to alpha motor neurons, to gamma (fusimotor) motorneurons, or both—independent pathways to the two pools of motorneurons may exist (Burke et al., 1978; Clough et al., 1971; Koeze, 1973; Koeze et al., 1968; Rothwell et al., 1990).]

There are two issues in the preceding data, on beta oscillations associated with motor activity, which require special consideration. First, not only is motor/somatosensory cortex beta associated with immobility (at least in certain behavioral paradigms), but such beta is increased in Parkinson’s disease, a disorder associated with pathological immobility. [Also, interestingly, beta-frequency motor neuron discharge is characteristic of sustained muscle contractions in humans (Grimby et al., 1981).] Could there be a connection? This question is addressed in more detail in Chapter 5. Second, somatosensory association cortex beta-2 oscillations are generated (in vitro) without a requirement for phasic synaptic transmission: in the “nonstandard” mode, to use terminology introduced earlier in this chapter. Assuming that a similar nonstandard mode is operating in vivo (and this remains to be proved), what about the spinal cord, which is interacting with motor/somatosensory cortex: is the spinal cord oscillating in a standard (synaptic) or nonstandard (nonsynaptic) mode? As noted earlier, our hypothesis is that the spinal cord is operating in the standard mode, with synaptic transmission intact, and with the possibilities for synaptic plasticity to take place according to conventional rules; this hypothesis, however, remains to be verified experimentally.

Some Theoretical Notions to Which the Study of Working Memory has Given Rise

A basic question concerning working memory is this: how is it that a defined population of neurons fires at elevated rates? Does the cellular make-up of this population encode just what the memory is? The late Daniel J. Amit (p.52)

                      Overview of In Vivo Cortical Oscillations

Figure 3.8 Beta and gamma (and also very fast) oscillations occur in parietal cortex during the expectancy phase of a memory task. Tetrode-recorded extracellular data from macaque monkey parietal cortex (area LIP). A: Schematic of the task. Baseline activity is recorded; then a cue is briefly shown in one of eight directions (Northeast, in this example), a memory phase of 1 second occurs, and then the monkey is to make a saccade in the cued direction. B: Field potential activity (polarity reversed), when the saccade is to occur in the “preferred” direction, i.e., in the direction that elicits maximal activity for the particular electrode site. Oscillatory activity is evident in the raw data, in the expanded trace below; C: Likewise, but for a saccade in the opposite (anti-preferred) direction. B: Time–frequency plot of oscillatory activity (~25 to ~100 Hz) during the memory phase (between the two vertical white dashed lines), when the saccade is to occur in the preferred direction (left), but not in the opposite direction (right). C: Oscillatory activity at the single-cell level, as determined by spike-triggered averaging, occurs during the memory phase (below), but not during baseline conditions (above). Please see color insert.

(Reproduced from Pesaran et al., 2002 with permission).

(p.53) (1938–2007) and colleagues, inspired originally by the Miyashita papers on monkey inferior temporal cortex, approached this problem by viewing the relevant neocortical neuronal network in terms of a Hopfield model, with phasically acting recurrent excitatory and inhibitory synaptic connections, and the activated (high firing rate) population appearing as an attractor in the system dynamics (Amit & Brunel, 1997; Griniasty et al., 1993; Miyashita, 1988; Miyashita & Chang, 1988). Another type of model (Camperi & Wang, 1998; Compte et al., 2000; Wang, 1999b, 2001; Wang et al., 2004) depends on N-methyl-D-aspartate (NMDA) receptor–mediated depolarizations in the relevant neurons, possibly augmented by slow intrinsic membrane currents that allow neurons to possess bistable membrane potentials, and with recurrent synaptic inhibition present as well. Such an approach, in common with the Hopfield model, also critically depends on recurrent synaptic connections between the activated neurons, but the connections act predominantly on a slow (hundreds of milliseconds) time course. Selection of the activated subpopulation occurs by transient depolarization [perhaps mediated by a metabotropic glutamate receptor activated inward current (Sidiropoulou et al., 2009)], taking place in particular neurons, rather than emerging solely from the properties of the recurrent synaptic connections. Neither of these types of models accounts for fast (20 Hz and above) network oscillations.

Our own in vitro and modeling work, discussed immediately in the text that follows and also in later chapters, suggests that activation of kainate receptors “turns on” a subpopulation of oscillating neurons, with little participation of recurrent synaptic excitation between the principal neurons—either of brief (AMPA receptor– mediated) or slower (NMDA receptor–mediated) time courses. Unfortunately, the in vitro work does not solve the problem of selection—which exact neurons are to participate in the activated state?—as network activation is brought about by bath application of a drug. One of the reasons why in vivo (p.54) (p.55)

                      Overview of In Vivo Cortical Oscillations

Figure 3.9 Beta oscillations, such as occur in motor and somatosensory cortex during the memory phase of a cognitive task, are coherent between motor cortex and muscle (despite >10 ms conduction delays). (A) Principal descending motor (red) and ascending sensorimotor (blue) pathways. (B) Coherence between motor cortex and muscle during a precision grip task (human data). Coherence at beta frequencies is present when the grip position is steady, and not otherwise. (C) The phase of this corticomuscular coherence is independent of frequency (the red regression line has a slope not significantly different from zero). (D) beta and gamma coherence exists between forearm EMG and the discharge of putative muscle spindle afferents (awake, behaving monkey data). The horizontal red line shows the level of significance. (E) Beta power in different cortical areas of the monkey: power is higher in primary somatosensory and posterior parietal areas than in primary motor cortex (M1). Please see color insert.

(Reproduced from Baker, 2008 with permission.)

(p.56) experiments on delayed match-to-sample, and related paradigms, are so critical concerns just this issue: perhaps the in vivo experiments may shed light on how the brain “decides” to provoke an activated state, with or without network oscillations, in a particular subpopulation of neurons.

Fast Oscillations Associated with the Depolarizing Phase of the Slow Oscillation of Sleep

Working memory (in prefrontal, parietal, and inferior temporal cortex) is apparently associated with the activation of selected brain regions; and within these regions, there appears to be an additional selection of some neurons—but not others—that fire at high rates, this subset of neurons presumably encoding what is to be remembered. Similar principles—consisting of localized cortical activation, and selection within the activated region—perhaps apply to cortical control of movement, and possibly even to all cortical operations characteristic of the waking state. Further, as we have seen (Fig. 3.8), localized activation is associated with fast oscillations.

An interesting contrast to localized cortical activation, discussed earlier, consists of the global cortical activated epochs (each lasting hundreds of milliseconds, with hundreds-of-milliseconds to seconds separation between epochs), that occur during the so-called slow (<1 Hz) oscillation of sleep—a striking phenomenon first discovered by Mircea Steriade and collaborators in the early 1990s (Amzica & Steriade, 1995a, b; Contreras & Steriade, 1995; Steriade et al., 1993a–d). The slow oscillation occurs during slow wave sleep; the term “slow wave sleep” derives, however, not from the slow oscillation, but rather from the delta waves (~2–5 Hz) that occur in this sleep state, the delta waves having been recognized long before the slow oscillation was first discovered (Loomis et al., 1935). Frequencies at 1 Hz and below are ordinarily not recognized in scalp EEG (although they can be, with special methods [Achermann & Borbély, 1997, 1998)]: such frequencies are ordinarily removed with high-pass filters, set so as to exclude artifacts caused by slow impedance changes in the EEG electrode-scalp contacts; in contrast, the slow oscillation of sleep was first recognized by Steriade and colleagues with intracortical field potential electrodes, and with intracellular recordings.

Field potential recordings indicate that the whole cortex, and thalamus, participates in the slow oscillation, approximately synchronously, with between-region phase delays in the tens to greater than 100 ms (Amzica & Steriade, 1995a); in humans, there are individual waves of activity that propagate across the cortex at 1.2 to 7.0 m/s (Massimini et al., 2004); and slow waves triggered by transcranial magnetic stimulation (TMS) also propagate across the cortex (Massimini et al., 2007). [Interestingly, however, transient TMS responses evoked during slow-wave sleep propagate far less than the corresponding responses evoked in the waking state, and the sleep responses also lose the fast oscillations (20–35 Hz) characteristic of TMS-evoked responses during wakefulness (Massimini et al., 2005).] The global synchronization of the slow (p.57) oscillation is maintained, at least in part, by subcortical white matter connections (Amzica & Steriade, 1995b).

Figure 3.10 (left) shows the slow oscillation as recorded with an intracellular electrode from a cortical neuron, probably a pyramidal cell; a filtered version of the extracellular field is shown below. (Chapter 4, Fig. 4.7 illustrates the slow oscillation field, in an unfiltered signal, as measured with an intracortical extracellular electrode—also in an anesthetized cat.) The data in Figure 3.10 (left) were recorded from a cat anesthetized with ketamine-xylazine: the original discovery of the slow oscillation and most subsequent in vivo studies used this preparation (and, of course, intracellular recordings cannot be obtained from humans in situ). Eighty-eight percent of principal cortical neurons were found, in one study (Steriade et al., 1993a), to participate in the slow oscillation, in a manner similar to the cell shown in Figure 3.10 (left): that is, with an alternating series of large sustained depolarizations [containing superimposed action potentials, spikelets (as will be shown later), and synaptic potentials], and large relative hyperpolarizations, each lasting hundreds of milliseconds. Interneurons participate in the slow oscillation as well, as shown by the series of IPSPs occurring during the slow depolarization (“upstate” or “activated state”) (Steriade et al., 1993a), as well as by direct intracellular recording of interneurons (e.g., Fig. 4 of Steriade et al., 2001). Figure 3.10 demonstrates, or at least hints at, the following critical features of the slow oscillation:

  1. 1. Absence of firing in the hyperpolarized state (“downstate”). Although both principal neurons and interneurons fire during the upstate, both classes of neuron are silent during the downstate. Hence, the downstate is apparently not maintained by active synaptic inhibition. Steriade and colleagues used the term “disfacilitation” to refer to the state of affairs whereby the upstate seems to collapse by loss of recurrent synaptic excitation; consistent with this notion, the input resistance of cells is highest during the downstates, when both EPSPs and IPSPs are in abeyance (Contreras et al., 1996b). David McCormick and colleagues, and others as well, using arguments based on their in vitro model of a cortical slow oscillation, have explained the upstate as reflecting a “balance” between synaptic excitation and inhibition (Haider et al., 2006; Hasenstaub et al., 2005; Mao et al., 2001; Sanchez-Vives & McCormick, 2000). As we shall see later, other in vitro data indicate that the balance notion can be only partly true, as an unexpected intrinsic membrane K+ current provides a critical contribution toward terminating the upstate.

  2. 2. Fast oscillations occur during the upstate. Figure 3.10 (left) makes clear that fast oscillations (~20 Hz and above) occur during the upstates, but not during the downstates; Steriade and colleagues elaborated extensively on this finding: one reason was that the occurrence of fast oscillations during slow wave sleep—a brain state (usually) without conscious correlates and producing little residual memory consciously accessible upon awakening—proved that fast oscillations (including, in particular, gamma oscillations) could not be specific for the conscious state, or for cognitive awareness. Yet, this issue may not be so simple (see later). In the rat, there appears to be a differential increase in (p.58) beta-2 EEG oscillations (roughly 20–30 Hz) in slow-wave sleep, as compared with waking and REM sleep; and also a differential decrease in gamma (Maloney et al., 1997). [In humans, 40 Hz (gamma) oscillations have been reported to characterize REM sleep (i.e., dreaming sleep), and to be reduced during slow-wave sleep (Llinás & Ribary, 1993).]

  3. 3. Fast oscillations are coherent between thalamus and cortex. During slow wave sleep (or at least the ketamine-xylazine-induced state that approximates it), the slow oscillation occurs in thalamus as well as cortex, and in phase. But in addition, fast oscillations (>~20 Hz) also occur in the thalamus during the upstates, and there is coherence between thalamus and cortex when “reciprocally connected” portions of each are compared (i.e., when one compares pools of thalamocortical relay cells that innervate particular cortical columns, and layers 6 and 5 of these columns contrariwise innervate the relay neurons and nearby nucleus reticularis cells) (Steriade & Amzica, 1996; Steriade, Contreras, et al., 1996). The experimental determination of reciprocal connectivity is achieved by stimulating the cortical region (respectively, thalamic region), and recording evoked synaptic potentials in the thalamic region (respectively, cortical region).

  4. 4. Fast oscillations during slow wave sleep have spatially limited coherence. Fast oscillations (15–75 Hz) were examined with multiple extracellular electrodes separated by 1 mm, in naturally sleeping and awake cats (Destexhe et al., 1999a). These oscillations were not coherent between distant electrodes, and only rarely and briefly (a few hundred milliseconds) between neighboring electrodes (see also Steriade, Contreras, et al., 1996; Steriade et al., 1995—in the latter study, coherence of fast oscillations was estimated to extend only over about one cortical column, a few hundred microns). Destexhe et al. (1999a) did observe, in one instance, correlations of fast oscillations between electrodes 7 mm apart, but this was during REM sleep.

  5. 5. Sleep spindles occur on the initial phase of the upstate. The slow oscillation of sleep serves as a kind of reference frame, around which other sleep-associated oscillations are temporally organized (Steriade, 2001, 2003, 2005). This type of temporal organization has even been shown for beta oscillations in humans (Mölle et al., 2002). One of the best-known of these sleep-associated oscillations consists of sleep spindles, which occur on the leading phase of the intracellular depolarization portion of the slow oscillation, and concerning which there is a vast literature about the EEG correlates and in vivo cellular mechanisms: much of this literature is reviewed in the Steriade monographs (2001, 2003). Suffice it to say here that sleep spindles (at roughly 9–15 Hz, depending on species) are generated in the thalamus, rather than the cortex: but whether in nucleus reticularis alone (Steriade et al., 1987), or through synaptic interactions between nucleus reticulars and principal thalamic nuclei, remains somewhat controversial—opinions on this issue tending to be influenced by opinions on what constitutes the most appropriate in vitro model of spindles. Thalamic spindles are synaptically projected to cortex and produce spindle-frequency oscillations there that are readily detectable in the (p.59) EEG. Further, reciprocal thalamocortical and corticothalamic synaptic interactions have a major influence on the intrathalamic coherence of spindles (Contreras et al., 1996a).

    For in vitro studies of spindle-like network oscillations, the reader is referred to the studies of McCormick and colleagues (Bal et al., 1995a,b; von Krosigk et al., 1993). Modeling issues have been considered, either in nucleus reticularis alone or in reciprocally connected reticularis and principal thalamus, by Bazhenov et al. (2000), Destexhe et al. (1993, 1994, 1996a), Golomb et al. (1994), Traub et al. (2005a), Wang and Rinzel (1993), and Wang et al. (1995), among others.

    It is well known that the intrinsic properties of thalamic neurons, particularly low-theshold Ca2+ spikes and the h-current, make major contributions to shaping the spindle oscillation (Bal & McCormick, 1993; Contreras et al., 1993; Crunelli et al., 1989; Deschênes et al., 1984; Jahnsen & Llinás, 1984a,b; McCormick & Pape, 1990a); it is perhaps less well appreciated that intrinsic properties of thalamic neurons may contribute to the slow oscillation itself, and not merely spindles, even if the primary “mover” is cortex (Blethyn et al., 2006; Curró Dossi et al., 1992; Hughes et al., 2002b; Leresche et al., 1991; Soltesz et al., 1991; Williams et al., 1997a; Zhu et al., 2006). The latter paper demonstrated that metabotropic glutamate receptor (mGluR) activation could elicit an intrinsic slow oscillation in at least some thalamocortical relay neurons; it should be noted that physiological inputs from the cortex can activate mGluRs in reticularis neurons (Blethyn et al., 2006) and in thalamocortical relay cells (McCormick & von Krosigk, 1992). Consistent with these observations, it has been shown in vivo that removal of overlying cortex prevents the emergence of a thalamic slow oscillation (Timofeev & Steriade, 1996).

  6. 6. Current sinks are different for the slow oscillation and the superimposed fast oscillations. When field potentials are recorded at numerous cortical depths, it becomes apparent that the slow oscillation, and the superimposed fast oscillations, have quite different properties. The slow oscillation has striking phase reversal in the middle cortical layers, at 0.25- to 0.5-mm depths; whereas fast oscillations do not (in vivo) have any clear phase reversal, and instead exhibit multiple sinks and sources with depth (Steriade et al., 1995). As we shall see later, many regions of cortex can have independent generators of fast oscillations in superficial (layers 2 and 3) and deep (layers 5 and 6) sites. Possibly this is not the case for the slow oscillation, which might rather depend on cells in both superficial and deep layers.

  7. 7. The activated state is unlikely, in itself, to encode memories in the form of an attractor, at least in any obvious way. As we have discussed in the preceding text, application of Hopfield-type attractor models to working memory is based, in part, on the notion that the identity of the remembered object (or task) is encoded in the identity of those neurons—within a defined brain region—selected to fire at high rates. For this type of memory to work efficiently—meaning, in particular, that many possible memories can be encoded—then any one memory is encoded sparsely, that is, with a set of (p.60) high-firing neurons that is small relative to the number of neurons in the region (Amit et al., 1985; Amit & Brunel, 1997)—so-called sparse coding. On the other hand, as we have seen in the preceding text (Steriade, Nuñez, & Amzica, 1993), almost 90% of principal cortical neurons are firing during the slow oscillation—hardly sparse! Application of physiological observations on the slow oscillation of sleep, to cognitive issues such as working memory, is then fraught with hazard. We can here identify some of the problems that we consider most pertinent, and which may be addressable with in vitro models of the slow oscillation, and their superimposed fast oscillations: most important, we believe, is the understanding of what determines whether a given pyramidal cell fires at a particular time, during an activated state such as an upstate. “Standard” models of neuronal networks are based on the assumption that timing is determined by membrane depolarization, and the pattern of synaptic inputs—and yet action potentials can be generated in axons, influenced in large part by action potentials in electrically coupled axons. Second, there is the critical issue of what factors determine if a principal cell is to be selected to fire at all during an activated state—is tonic depolarization enough? A given pattern of synaptic inputs? Or a given location in a gap junctional network?

  8. 8. Thalamic neurons participate in the slow oscillation, but cortex can generate a slow oscillation on its own, at least if a sufficient mass of tissue is present. The cortical slow oscillation persists after destruction of the underlying ipsilateral thalamus (Steriade, Nuñez, & Amzica, 1993); it must be possible for the slow oscillation to be generated within the cortex itself. Nevertheless, there seems to be a critical mass of cortex that is required, at least if the proper rhythmicity of the oscillation is to be maintained (Timofeev et al., 2000): a cortical area of about 2 cm2 may be required. How these in vivo data can be reconciled with the existence of slow oscillations in neocortical and entorhinal cortical slices in vitro is not altogether clear. It is interesting that in one in vitro model of the slow oscillation, using thalamocortical slices, there was evidence that thalamic input helped to trigger cortical upstates, even though cortex was able to generate at least some upstates without the thalamic part of the preparation (Rigas & Castro-Alamancos, 2007): this may be the situation in vivo as well.

How Is the Slow Oscillation Generated?

In vivo, the slow oscillation of sleep correlates with the occurrence of certain EEG patterns (delta waves); of behavioral sleep (immobility, lack of awareness); and by the absence of signs pointing to REM sleep (i.e., during slow wave sleep, one does not see rapid eye movements, diffuse skeletal muscular paralysis, or continuous EEG fast rhythms); and the slow oscillation can be aborted by stimulation of the cholinergic pedunculopontine tegmental nucleus (Steriade, Amzica, & Nuñez, 1993). These observations, while obviously important, nevertheless do not provide sufficient information to address a number of questions, however; for example, why there are superimposed fast oscillations and whether there is meaning to the patterns of neuronal activity (p.61) during the slow oscillation. [For example, there are a number of studies purporting to show “replay” of waking firing patterns, recapitulated (so to speak) during sleep, both in hippocampus and in cortex (Euston et al., 2007; Ji & Wilson, 2007). How such replay comes about and what its significance is, however, are not apparent.]

In vitro data (Cunningham et al., 2006b) provide some important clues as to cellular mechanisms of the slow oscillations, data that need to be confirmed with in vivo experiments. We shall discuss some of the data in detail in a later chapter, but mention here the following: our in vitro data provide counterintuitive explanations both for the upstate and for the transition to the downstate. Specifically, our data indicate that the upstate arises through glutamate actions on kainate receptors, rather than AMPA or NMDA receptors; and that the transition to the downstate is initiated by a metabolically regulated intrinsic neuronal K+ current, mediated by ATP-gated K+ channels. The mediation of the upstate by kainate provides an experimental underpinning for one of our basic experimental oscillation protocols, as we discuss next.

In Vitro Fast Oscillations Can Occur that are Continuous, Without a Slow Oscillation

In Figure 3.10B, we provide a first look at in vitro data on neocortical fast oscillations. The figure shows recordings of gamma oscillations in a slice of rat secondary auditory cortex, as well as simulation data from a network model (Cunningham et al., 2004b). [The model contained 1,152 pyramidal cells, both regular spiking (RS) and chattering/fast rhythmic bursting (FRB) as well as 192 fast-spiking (FS) interneurons and 96 low-threshold-spiking (LTS) interneurons. Synaptic receptors were of AMPA and GABAA types, and there was electrical coupling between pyramidal cell axons, and between interneuron dendrites.] Although we expand greatly in later chapters on the cellular mechanisms of cortical gamma oscillations in vitro, the point to be made now is this: that the network model, the in vitro experimental data, and the in vivo data (left part of the figure) resemble each other in numerous ways, particularly in the oscillation frequency and in the subthreshold synaptic potentials—but with one notable difference: the in vivo fast oscillation is broken up at the frequency of the slow oscillation (about 1 Hz), whereas the in vitro oscillation runs continually. We ascribe these respective patterns according to the following hypothesis: the in vitro experiment depends on bath-applied kainate, while in vivo, the upstates are sustained by kainate receptors—as if the slice situation corresponded to a persistent upstate; however, in vivo, the upstates are interrupted by intrinsic ATP-gated K+ channels, whereas in vitro (for unknown reasons), the metabolic state of the neurons is stable enough that such ATP-gated K+ channels do not open—hence the absence of the periodic interruptions of the activated state. It is a striking mystery how and why, in vivo, the brain enters into a state where the cortical neurons appear to be incapable of sustaining their activity for more than a few hundred milliseconds. (p.62)

                      Overview of In Vivo Cortical Oscillations

Figure 3.10 Beta and gamma oscillations occur during the intracellular depolarizations of the cortical slow oscillation of sleep; and these oscillations can be mimicked by drug application in vitro. A: Intracellular and filtered field potential recordings of the cortical slow oscillation, with fast oscillations superimposed on the “upstates” (data from primary somatosensory cortex, in the ketamine/xylazine-anesthetized cat). The intracellular (~0.3 mm) and field activities (both surface and depth) are at least approximately in register throughout the depth for the slow oscillation; and the fast oscillations (>~20 Hz) are also in register. The intracellular recording shows what appear to be oscillating synaptic potentials (compare Fig. 3.4). B: Gamma (30–70 Hz) oscillations in auditory/parietal cortex in vitro (bathed in 400 nM kainate), along with results of a detailed network simulation. The upper part shows that peak gamma power is in layers 2/3. In addition (not well seen here), the fast oscillations are continuous and are not superimposed on and modulated by a slow oscillation. The lower part demonstrates firing behavior in FRB (fast rhythmic bursting) cells, which fire on roughly half the gamma waves; and RS (regular spiking) pyramidal cells, which fire more intermittently. Both types of cell exhibit synaptic potentials at gamma frequency. The in vitro cell potentials resemble those of the in vivo cell in A during the “upstates.”

[Composite reproduced from Traub, Cunningham, & Whittington, 2008 with permission; Data in A from Steriade, Amzica, & Contreras (1996), reproduced with permission. Data in B from Cunningham et al. (2004), reproduced with permission.]

[The astute reader will have noticed as well that, in the vitro data shown in Fig. 3.10, gamma power is far greater in superficial layers than in deep layers. The distribution of frequency and power between layers depends, as we shall show, on both the region of cortex, and on the pharmacological methods used to activate the cortex.]

(p.63) To us, the data in Figure 3.10 provide justification for extrapolation of in vitro oscillation data to at least some in vivo contexts, provided the extrapolation is done critically and carefully.

Very Fast Oscillations Superimposed on Sensory Evoked Potentials

Brief sensory stimulation, in any modality, evokes a series of neural (hence electrically recordable) responses in cortical structures, at early (<~150 ms) and at longer (<~500 ms) latencies, typically consisting of waves that last on the order of tens of milliseconds; these responses are produced as the neural “traffic” proceeds along axons, causes cell firing and then synaptic currents, and in turn influences successive pools of neurons. Evoked responses of this type have been of interest not only to sensory physiologists, but also to physiological psychologists, who study the so-called P300 (a positive cortical potential at 300-ms latency), or contingent negative variation (Sutton et al., 1965); and to clinical neurologists who wish to evaluate sensory pathways in patients with suspected (for example) tumors or demyelinating disease (Ebersole & Pedley, 2003; Halliday, 1967). Neural population events having a similar appearance to evoked responses also occur spontaneously: examples are vertex waves in the EEG during sleep (Ebersole & Pedley, 2003), and physiological sharp waves, described first in the hippocampus by György Buzsáki (1986): one may think of such population events as being (so to speak) internally generated “evoked” responses, or “evoked” potentials. Synchronized epileptiform bursts (see Chapter 4) can also be regarded as a type of internally generated “evoked” response, differing from a physiological sharp wave in quantitative parameters, such as the number of cells participating in the response, and the number of action potentials per participating neuron (Buzsáki, 1986).

Neuronal population responses lasting tens of ms often have very fast oscillations superimposed upon them. As far as we are aware, such an association was first made for epileptiform events in the hippocampus in vitro (Schwartzkroin & Prince, 1977; Wong & Traub, 1983), where the superimposed oscillations were typically several hundred Hertz; however, the association is not confined to pathological, or to in vitro, population events. As Buzsáki and colleagues showed (Buzsáki, et al., 1992; Klausberger et al., 2003; Ylinen et al., 1995a), in vivo hippocampal physiological sharp waves also contain superimposed “ripples” at about 200 Hz. It is critical to understand the relation between the slower spontaneous or evoked responses—which are attributed to synchronized synaptic currents—and the very fast oscillations that are superimposed: critical both for cellular mechanisms, and for proposing reasonable hypotheses as to the functional implications of the firing patterns of the constituent neurons.

Figure 3.11, taken from the work of Daniel S. Barth, shows that very fast oscillations can occur superimposed on genuine evoked responses, produced by stimulating a body part: in the case of this figure, via a rapid induced twitch of the whiskers (trimmed and tied together) on one side of an anesthetized (p.64)

                      Overview of In Vivo Cortical Oscillations

Figure 3.11 Very fast oscillations (~390 Hz in this case) occur superimposed on a somatosensory evoked response in rat barrel cortex. A: Arrangement of the vibrissae (large whiskers) on the rat snout. B: Corresponding arrangement of “barrels” in barrel cortex, a part of primary somatosensory cortex, with overlaid extracellular recording array. C: Classical somatosensory evoked potential (negative upwards), produced by brief displacement of a group of trimmed, tied-together, whiskers. One sees the early positive (P1) and negative (N1) waves, a two sites (solid line and dashed line). D: Corresponding ~390 Hz VFO (called FO in the figure), after digital filtering, 200–1000 Hz. E: Example of phase-aligned VFO at two sites; this VFO is particularly long-lasting.

(Reproduced from Barth et al., 2003 with permission.)

(p.65) rat’s snout. [A note on the nomenclature: the waves are called Px (‘positive” x), where x is an integer designating first, second, third, etc.—in this case first; and Nx, for corresponding negative waves; however, x can also stand for a time in milliseconds, as in P300, meaning the latency in ms when the wave occurs. The “positive” and “negative” refer in turn to the polarity of the signal as measured at the surface of the cortex; and to further confuse matters, signals in the evoked potential literature—following an old EEG tradition—are often plotted inverted (as in Fig. 3.11C), that is with negativity upwards.] Careful study of this so-called somatosensory evoked potential (Fig. 3.11C) shows the superimposed very fast oscillation, which is much easier to see in the filtered signals below. With controlled stimulation of individual whiskers, and pairs of whiskers, it becomes apparent that the very fast oscillations are generated within the cortex, with the thalamic inputs serving as a trigger (Staba et al., 2003); and it becomes apparent that the very fast oscillations are spatially organized within the cortex, in a manner determined by intracortical connections (Barth, 2003; Staba et al., 2005), although whether by synaptic or gap junctional connections, or both, remains to be determined. Notably, layer 4 multiunit neuronal responses can follow rapid mechanical whisker vibrations, 1:1, at frequencies up to 320 Hz (Ewert et al., 2008): thus, very fast oscillations in this rather specialized somatosensory cortex may actually be subserving a direct encoding function.

Ylinen et al. (1995a) had shown that the extracellular field associated with hippocampal sharp wave ripples corresponded faithfully to rhythmical IPSPs in pyramidal cells, motivating the hypothesis that it was networks of interneurons that might actually generate the ripples. There are, however, other ways to account for the observation of Ylinen et al., if one assumes that the pyramidal cell axon plexus is the primary generator of the ripple (Traub & Bibbig, 2000)—we shall return to this critical issue in a later chapter, when we review in vitro data on very fast oscillations. In any case, however, very fast oscillations, superimposed on somatosensory evoked potentials in rat barrel cortex, do not appear to be generated by interneurons: flooding the tissue with GABA does not affect the fast oscillations, although it does abolish the N1 wave (Staba et al., 2004a); and conversely, subconvulsive concentrations of the GABAA antagonist bicuculline actually enhance the very fast oscillation (in terms of producing more waves), without changing either the amplitude or the frequency (Jones & Barth, 2002).

The time course of synaptic excitation between pyramidal cells is relatively fast [deactivation τ = 3 ms in CA3 pyramids at room temperature (Geiger et al., 1995), with physiological temperatures speeding this up, but with dendritic electrotonic filtering slowing down the time course of actual EPSPs (Miles & Wong, 1986)]; but, on the other hand, obtaining stable network oscillations solely through recurrent synaptic excitation is probably not possible (van Vreeswijk et al., 1994). So if neither synaptic inhibition nor synaptic excitation produce very fast oscillations, superimposed on sensory evoked (p.66) repsonses, that presumably leaves—by exclusion—gap junctions. Unfortunately, there is as yet little positive evidence, in vivo, to strengthen this idea (although there is a great deal of in vitro evidence in support of the idea). Genetic knockout of the main neuronal gap junction protein, connexin-36, has little effect on hippocampal sharp wave ripples (Buhl et al., 2003) or on barrel cortex very fast oscillations (Daniel S. Barth, personal communication). It is possible, but not proven, that in the connexin-36 knockout mouse, another gap junction protein is upregulated in pyramidal cells. Interestingly, both humans (Curio 2000; Curio et al., 1994) and piglets show approximately 600 Hz oscillatory components in somatosensory evoked responses, and for piglets the signal has been suggested to arise from thalamocortical axons and terminals in layer 4 (Ikeda et al., 2002, 2005), as well as in cortical somata and dendrites (Okada et al., 2005). It is possible that gap junctions exist within and between these terminals and axons (Hamzei-Sichani et al., 2007.). Very fast oscillations have also been recorded from human anterior temporal cortex, during neurosurgical procedures performed with the patient awake, in response to auditory stimuli—especially if the stimulus was unexpected (Edwards et al., 2005); the cellular mechanisms of the human auditory-stimulated oscillations have not been investigated; however, the existence of such responsiveness in human cortex is of vast importance, particularly for the understanding of the initiation of seizures (see Chapters 4, 13).

Temporal Interactions Between Cortical Oscillations at Different Frequencies

We have seen that oscillations can be superimposed on transient neuronal population events (Fig. 3.11), and also on particular phases of another, slower, oscillation (Fig. 3.10). Perhaps the best-known example of one oscillation superimposed on, and amplitude-modulated, by another is the case of gamma oscillations superimposed on the hippocampal theta rhythm—something that occurs in vivo (Bragin et al., 1995; Soltesz & Deschênes, 1993; Ylinen et al., 1995a), and with in vitro models as well (Fisahn et al., 1998; see Chapter 11; Gillies et al., 2002). Understanding how these sorts of interactions take place, between oscillations of different frequencies, may be important for unraveling the functional importance of each frequency, and how the respective distinct functions might be inter-related (Palva et al., 2005). As always, our default hypothesis is that ideas about function are most readily developed when the cellular mechanisms have been spelled out. This is especially important when it comes to considering the cortical beta-1 (~15 Hz) oscillation, which—at least in vitro—appears to be produced by fitting together (rather than phase-resetting or amplitude-modulating) two simpler oscillations (Kramer et al., 2008; Roopun et al., 2008b).

For now, however, suffice it to illustrate, in the auditory cortex of awake behaving monkeys, examples of phase-resetting of multiple oscillation frequencies by a transient somatosensory stimulus (i.e., a brief stimulus in one (p.67) modality resets oscillations in a part of the brain mostly devoted to a different modality) (Fig. 3.12); and of multiple interactions whereby the amplitude of one oscillation is influenced by the phase of another slower oscillation, during spontaneous activity (Fig. 3.13). This type of data, taken from the work of Peter Lakatos, Charles Schroeder, and colleagues, provides examples of phenomena that may, in principle, be analyzed further with in vitro experimental models. Other examples of cortical oscillatory superimpositions have been reported as well (Canolty et al., 2006).

Cerebellar Oscillations

Although this book’s title begins “Cortical Oscillations …,” meaning “Cerebral Cortical Oscillations …,” there is evidence—at least for the motor system—of interactions of oscillations between widely dispersed brain regions, in a manner that could be functionally meaningful, and that almost certainly has disease implications. Further, the cellular mechanisms of cerebellar oscillations provide interesting contrasts with neocortical mechanisms of oscillations at comparable frequencies, because the synaptic architecture of the cerebral and cerebellar cortices is so different. For these reasons, we shall introduce the subject of cerebellar oscillations here. These fall into several types, including the following:

  1. 1. Oscillations at theta and alpha frequencies, generated among the electrically coupled pool of inferior olivary neurons, and transmitted to the deep cerebellar nuclei and cerebellar cortex via climbing fibers (Blenkinsop & Lang, 2006; Leznik & Llinás, 2005; Martin & Handforth, 2006). Oscillations of this

                          Overview of In Vivo Cortical Oscillations

    Figure 3.12 Auditory evoked activity contains fast oscillations. Extracellular data from awake macaque monkey, recorded with a multisite probe with >15 contacts throughout all cortical layers; stimulation consisted of pure tones and broad-band noise. A: CSD (current source density) of the evoked response; S = supragranular (mainly layers 2 and 3), G = granular (layer 4), I = infragranular. B: Frequency components of the evoked signal at one supragranular site (indicated by the gray arrow). The frequency scale is logarithmic. There is power at theta, alpha, beta, gamma, and VFO frequencies. C: Demonstration that the phase of the oscillations is not random (vertical blue line n graph at the right), but rather is reproducible trial-to-trial. Please see color insert.

    (Reproduced from Lakatos et al., 2007 with permission.)

                          Overview of In Vivo Cortical Oscillations

    Figure 3.13 Example of one oscillation (delta) modulating the amplitude of another (theta). Spontaneous extracellular field potential data, primary auditory cortex of awake monkey. A: Simultaneous field potentials at 20 different cortical depths, recorded with multisite probe (left). Right: current source density (CSD) of this data. S = supragranular, G = granular, I = infragranular, as in Fig. 3.12. B: CSD data from on supragranular site, shown as raw signal (green) and in time–frequency plot (below). There is power in delta, theta and gamma ranges. C: The phase of delta (1.4 Hz) modulates the amplitude of theta (7.8 Hz). Not shown is the additional modulation of gamma amplitude by theta phase. Please see color insert.

    (Reproduced from Lakatos et al., 2005 with permission.)

    sort are relevant in motor control and in physiological tremor, but fall outside the scope of this book.

  2. 2. Very fast oscillations, which occur in the cerebellar cortex of certain genetically modified, and ataxic, mice (Cheron et al., 2004). These are discussed in Chapter 7.

  3. (p.69)
  4. 3. Beta and gamma (as well as lower frequency) oscillations that are generated within the cerebellar cortex. Oscillation coherence has been demonstrated between deep cerebellar nuclei and motor thalamus (ventrolateral and ventral intermediate nuclei), as well as motor cortex, and possibly also the basal ganglia. Such oscillations, involving extensive components of somatosensory and motor systems, have been studied in humans and experimental animals with field potential recordings and with magnetic field measurements. The oscillations are modulated by voluntary movement (usually tending to be suppressed by such activity, with the exception of whisking in rats) and by expectancy (as is the case for cortical beta-2). Coherence between cerebellum and thalamocortical/basal ganglia structures has been reported mostly for alpha and beta frequencies, roughly 8 to 30 Hz (Courtemanche & Lamarre, 2005; Courtemanche et al., 2003; O’Connor et al., 2002; Marsden et al., 2000; Pellerin & Lamarre, 1997; Pollok et al., 2005), but also extending into the gamma range (Soteropoulos & Baker, 2006). Coherence of beta oscillations has also been found between deep cerebellar nuclei and tonically active muscle EMG (Aumann & Fetz, 2004; see Chapter 7).