Multiple origins of the cortical gamma rhythm



Gamma rhythms (30–80 Hz) are a near-ubiquitous feature of neuronal population activity in mammalian cortices. Their dynamic properties permit the synchronization of neuronal responses to sensory input within spatially distributed networks, transient formation of local neuronal “cell assemblies,” and coherent response patterns essential for intercortical regional communication. Each of these phenomena form part of a working hypothesis for cognitive function in cortex. All forms of physiological gamma rhythm are inhibition based, being characterized by rhythmic trains of inhibitory postsynaptic potentials in populations of principal neurons. It is these repeating periods of relative enhancement and attenuation of the responsivity of major cell groups in cortex that provides a temporal structure shared across many millions of neurons. However, when considering the origins of these repeating trains of inhibitory events considerable divergence is seen depending on cortical region studied and mode of activation of gamma rhythm generating networks. Here, we review the evidence for involvement of multiple subtypes of interneuron and focus on different modes of activation of these cells. We conclude that most massively parallel brain regions have different mechanisms of gamma rhythm generation, that different mechanisms have distinct functional correlates, and that switching between different local modes of gamma generation may be an effective way to direct cortical communication streams. Finally, we suggest that developmental disruption of the endophenotype for certain subsets of gamma-generating interneuron may underlie cognitive deficit in psychiatric illness. © 2010 Wiley Periodicals, Inc. Develop Neurobiol 71: 92–106, 2011


Electrical activity in the EEG gamma band (30–80 Hz) is almost continually present, to some degree, on all recording electrodes in most behavioral and cognitive states including slow-wave sleep. Early observations showed that gamma activity could be significantly enhanced by sensory input (Adrian,1942). This increase in gamma power is sensory modality-specific, with visual stimuli generating largest changes in primary visual cortex, auditory stimuli activating gamma rhythms in temporal cortex, etc. (e.g., Jefferys et al.,1996). Evidence for a functional role for these gamma rhythms in primary sensory processing began to emerge in the late 1980s (Gray and Singer,1989; Eckhorn et al.,1988). In an attempt to link the activity of many millions of neurons in sensory cortex to the nature of the sensory stimulus itself, it was proposed that coherent or precisely synchronous rhythmic activity served to link or “bind” together the outputs of all neurons coding for certain features of an object in the sensorium (for review see Singer and Gray,1995). The idea that the output from a subset of neurons has particular significance if it occurs near-synchronously with the output from other neurons forms the basis of the neuronal assembly hypothesis and, as such has wider implications for cognition in general (Maldonado and Gerstein,1996). Such transient assemblies are seen to occur with a temporal scale which implies involvement of gamma rhythms (Harris et al.,2003).

Gamma rhythms are also particularly implicated in active sensory processes involving attention (Tiitinen et al.,1993; Deco and Thiele,2009) and also in short-term memory processes (Tallon-Baudry et al.,1998). In each case, what is of likely functional importance is the ability of the gamma rhythm to limit, in time, the probability of output generation in any given set of neurons [Fig. 1(A)]. Such a role makes sense, given that the precise timing of inputs from one cortical region to another is critical for the flow of information through cortex (Burchell et al.,1998). This phenomenon forms the basis of the “communication through coherence” hypothesis proposed as a fundamental component of cortical function (Fries,2005). One particular mechanism of gamma rhythm generation appears to be an ideal substrate for such a functional framework. A combination of feedback from active principal cells back onto proximal local circuit interneurons and projections to distal interneurons in other oscillating regions provides a means by which the relative timing of outputs from each region can modify the period duration of each local gamma rhythm and impart synchrony despite long conduction delays (Traub et al.,1996b). However, this is only one of three main modes of local circuit gamma rhythm characterized using in vitro models. Each mode shares the property of trains of gamma frequency, phasic inhibitory events onto principal cells, but the mechanism of recruitment of interneurons into the gamma rhythm, and perhaps the nature of the interneurons themselves, are rather different. Here, we consider the mechanisms and cellular substrates underlying each of these forms of gamma rhythm and discuss their possible physiological significance.

Figure 1.

Gamma rhythms control the spatiotemporal properties of cortical response to input. A: Trains of principal cell somatic IPSPs produce a sparser but more precisely timed population code. Cartoon of rasters representing the spike times for outputs of 10 principal cells driven by afferent input over 0.5 s. In the absence of a common gamma rhythm, spikes are generated with little temporal precision (left panel). If each cell receives a common train of inhibitory synaptic potentials, spike generation is attenuated during each IPSP (Burchell et al.,1998) but facilitated between consecutive IPSPs. The resulting population code is reduced—in terms of number of spikes—but refined in terms of relative spike timing. Such a modification of spike outputs generates a pattern similar to that associated with cell assemblies in vivo (Harris et al.,2003). This phenomenon is achieved by a combination of perisomatic inhibitory current (graph, gray line) and anodal break potentials quantizing optimal output probability over time in each period (black line). B: Each of the three main subtypes of gamma rhythm (ING, PING, persistent) generates similar patterns of principal cell spike timing through trains of IPSPs. However, the different underlying local network mechanisms impart different dynamic ranges in response to changes in network excitation. With tonic depolarisation of interneurons alone (ING: black symbols, mGluR activation of hippocampal area CA1 in the presence of blockers of ionotropic glutamate receptors (Traub et al.,1996a,b)), network frequency changes markedly with increased network drive and shows a clear maximum. With phasic drive to interneurons generated by orthodromic principal cell spikes (PING: gray symbols, tetanic stimulation of Schaffer collateral input to CA1), network frequency is less sensitive to degree of excitation and changes over a broader dynamic range. With phasic drive to interneurons generated by axonal plexus activity (persistent gamma: white symbols, kainate bath application, recording from area CA1) frequency is almost inert to network drive over at least a decade of local gamma power change. Figure adapted from Traub et al.,1996a,b; Hormuzdi et al.,2001.


Three distinct but interrelated forms of gamma rhythm have been characterized in mammalian cortex. With appropriate experimental conditions they can all be seen in relatively simple cortical circuits such as those in hippocampal cornu ammonis. In each case, the rhythm is an emergent property of the local neuronal network, but differs in terms of interneuron recruitment, principal cell involvement and basic dynamic properties [e.g., Fig. 1(B)].

Interneuron Network Gamma

The simplest form of gamma rhythm can be seen in networks of interneurons alone. Pharmacological isolation of interneurons from phasic excitatory neuronal input, coupled with tonic depolarization of these interneurons readily produces a local circuit gamma rhythm characterized by trains of inhibitory postsynaptic events in quiescent principal cells and fast spiking interneurons (Whittington et al.,1995; Traub et al.,1996a). In this mode of gamma generation, the frequency of the rhythm is sharply dependent upon the magnitude of tonic excitation received by interneurons and the kinetics and magnitude of inhibitory postsynaptic events generated by a recurrent interneuronal network. Despite its elegance and simplicity a number of problems occur when attempting to elucidate both underlying mechanism and physiological relevance of this type of gamma rhythm:

  • 1The highly labile nature of frequency under different drives imparts a sensitivity to drive heterogeneity. In computational models of interneuron networks interconnected with phasic synaptic inhibition (but without gap junctions), even a small variability of tonic drives (c.5%) to each interneuron in the local recurrent circuit has detrimental effects on gamma generation (Wang and Buzsaki,1996). Despite this, experimental observations showed that such a rhythm is highly robust. The reason for this dichotomy appeared to be the presence of gap junctions between fast spiking interneurons (Fukuda and Kosaka,2000). Such non-chemical synaptic connections act as low pass filters, effectively sharing slow changes in membrane potential (Hormuzdi et al.,2001). Thus, blockade of gap junction conductance during interneuron network gamma (ING)—in preventing the equilibration of membrane potential and synergistic interactions between electrical and chemical synapses (Tamás et al.,2000) between interneurons—detrimentally affects the generation of this mode of gamma rhythm; network simulation models of ING that include gap junctions are also robust to heterogeneity (Traub et al.,2004).
  • 2The original studies on ING implicated perisomatic targetting, fast spiking interneurons as the substrate for the interneuron network. Of all the interneurons fitting these criteria, parvalbumin immunopositive basket cells appeared the most likely. During experimentally induced gamma rhythms and earlier computational models the decay constant for IPSCs onto interneurons was seen to be in the range of 4–10 ms, with the frequency of the network gamma rhythm exquisitely sensitive to this parameter (e.g., Traub et al.,1996a). In contrast, precise biophysical measurement of decay time constant in basket-basket connections—albeit in nonoscillating tissue—showed considerably faster kinetics, with timeconstants as low as 1–2 ms (Bartos et al.,2002). Such values could still generate ING-like rhythms in the gamma frequency range in computer simulations with alterations in other interneuron properties (Bartos et al.,2007). The discrepancy here may simply be due to inadequate control of membrane voltage in earlier studies. However, the difference is large and it may well be that the basket cell network is not alone responsible for ING in hippocampus. Interneurons form dense heterogeneous local networks involving many different subtypes (Tamás et al.,1998). The involvement of other interneuron subtypes is a recurring theme when attempting to explain experimental observations on gamma rhythms and is dealt with in more detail in later sections.
  • 3With one exception all experimental studies on ING have used highly non-physiological conditions to expose this mode of gamma rhythm generation. Specifically, the presence of ING can only be established using manipulation of tissue bathing media to pharmacologically block fast, ionotropic glutamatergic excitation. This naturally leads one to question whether ING is of any overt physiological relevance at all. However, in cerebellar cortex both principal cells and most of their local circuit interneurons are GABAergic. Here, cholinergic activation via nicotinic receptors generates a gamma frequency local field potential oscillation that is independent of any kainate, AMPA or NMDA receptor activity [Fig 2(A); (Middleton et al.,2008a,b)]. The cerebellar cortex contains vast numbers of glutamatergic neurons (granule cells), but in the absence of any afferent input in the in vitro slice preparation, these neurons appear irrelevant to the continuous gamma rhythm generated by nicotinic neuromodulation.
Figure 2.

Cerebellar gamma rhythms show ING-like properties under physiological conditions. A: Activation of cerebellar cortex with nicotine generates a gamma rhythm independently of ionotropic glutamate receptor activation. Example traces show Purkinje cell layer field potentials (scale bars 100 ms, 20 μV). Power spectra are from 60 s epochs of these data. B: Cerebellar ING transforms a cell-specific Purkinje cell gamma frequency output to a sparse Purkinje cell population gamma rhythm. Extracellular local field potentials (ec) show no locally synchronous gamma frequency activity despite spontaneous gamma frequency spiking in Purkinje cells (PC, left panel, traces show 500 ms of data). In contrast, local field potentials show clear gamma frequency population activity under nicotinic receptor activation despite low (<10 Hz) individual Purkinje cell spike rates. C: Nicotine-induced ING-like rhythm is associated with interneuron gamma frequency spiking and trains of IPSPs in both Purkinje cells and interneurons (stellate cell activity shown in this example). Figure adapted from Middleton et al.,2008b.

Along with the absence of a role for fast glutamatergic excitation, this cerebellar gamma rhythm shows a number of other similarities to experimentally isolated ING in other cortical circuits. Spontaneous principal (Purkinje) cell firing is markedly reduced. Temporally uncorrelated gamma frequency spiking in individual Purkinje cells is transformed into sparse (c. 2 Hz) highly correlated spiking [Fig. 2(B)]. Again, the underlying mechanism, at the local circuit level, appears to be the existence of gamma frequency trains of IPSPs in both principal cells and interneurons [Fig. 2(C)]. However, differences between this gamma rhythm and other cortical ING activity exist. Both hippocampal ING and cerebellar ING require gap junctional coupling between interneurons. In hippocampus, this requirement involves only low frequency coupling of interneuron membrane potential. In cerebellar stellate cells, however, a combination of gap junctional coupling and intrinsic membrane properties results in functional action potential coupling also (Mann-Metzer and Yarom,1999). In addition, gap junctions between principal cells also appear to play a role in cerebellum. The cerebellar ING is always accompanied by—and dependent upon—a very fast oscillation (VFO) which is associated with electrical coupling between Purkinje cells (Traub et al.,2008).

Pyramidal Interneuron Network Gamma

The experimental model of gamma rhythms most likely to correspond to sensory evoked and induced gamma rhythms in cerebral cortex is that involving reciprocal interaction between interneurons and principal cells (Traub et al.,1997). The predominant mechanism underlying this mode of gamma rhythm generation is the phasic excitation (via fast, AMPA receptor-mediated postsynaptic potentials) presented to interneurons following orthodromic spike generation in principal cells (e.g., Whittington et al.,2000). At spike rates in the tens of Hz, the massive convergence of local excitatory inputs onto interneurons (∼1500 pyramids to 1 parvalbumin positive interneuron has been estimated (Buhl and Whittington,2007) is sufficient to overcome frequency-dependent (Thomson,2000) and metabotropic receptor-mediated (Parra etal.,1998; Vignes et al.,1998; Xiao et al.,2001) attenuation of postsynaptic responses, and is therefore able to produce large (2–10 mV at rmp) compound EPSPs. The resulting divergence of outputs from individual interneurons back to the principal cell local population (one interneuron synapsing onto up to 500 postsynaptic targets (Halasy et al.,1996)) leads to temporally modulated principal cell output as with ING rhythms, provided, of course, that different interneurons fire synchronously.

The requirement for high principal cell spike rates is reflected in the situations in which pyramidal interneuron network gamma (PING) is seen. Pressure ejection of glutamate depolarizes both interneurons and principal cells, resulting in spike rates around gamma frequency. Similarly, intense tetanic excitation and transient elevation of extracellular potassium also generate a PING rhythm which may persist for several seconds following stimulus termination (Whittington et al.,1997; LeBeau et al.,2002). In computational models, PING may exist entirely independently from ING, with no requirement for tonic interneuronal depolarization or reciprocal interneuron network activity (Traub et al.,1997). However, experimentally it is much harder to separate these two modes of rhythm generation. The three main models: glutamate pressure ejection, tetanic stimulation and extracellular potassium concentration rises can still generate gamma rhythms when principal cell outputs, or inputs to interneurons are experimentally ablated (Miles and Poncer,1993; Whittington et al.,2001; Traub et al.,2004). Continuous, intense activation of hippocampal circuits with carbachol or kainate may also generate a PING-like rhythm that may also co-exist with some degree of interneuron network activation (Fisahn et al.,2002,2004; Bartos et al.,2007 but see below). In vivo, specific optogenetic drive of fast-spiking interneurons in layers 2–6 enhances cortical gamma rhythms, suggesting a critical role for interneuron drive (Cardin et al.,2009). Activation of principal cells in this study did not enhance gamma. However, the principal cells targeted were in both superficial layers and layer 5, which does not contribute to the predominant gamma generating circuit in neocortex (Roopun et al.,2006). Similarly, gamma rhythms associated with visual sensory responses significantly improve spike timing precision without increasing principal cell spike rates (Rodriguez et al.,2010).

From the above observations, it remains difficult to separate ING from PING at a functional level in physiological conditions. Nevertheless, having reciprocal dynamic interactions between interneurons and pyramidal cells has at least three important consequences for network dynamics. First, when compared to ING, frequency becomes more stable in the face of changes in network excitation [Fig 1(B)] and GABAA receptor kinetics (Whittington et al.,1997). Second, allowing principal projection neurons to modulate interneuron spike timing provides a robust mechanism for establishing long-range synchrony between spatially separate gamma-generating local circuits (Traub et al.,1996a,b). Finally, the nature of synaptic plasticity at excitatory synapses onto interneurons provides a powerful means by which to modulate and stabilize distributed network activity (Lamsa et al.,2007; McBain and Kauer,2009). On the basis of in vitro studies, it seems safest to assume that PING, and indeed ING, exist predominantly as transient phenomena in cerebral cortical networks—occurring during, or immediately following brief, intense periods of principal cell activity. We must therefore look for an alternative mechanism for the more persistent gamma (PG) rhythms associated with sparse principal cell background activity, anticipatory cortical states, and on-going rhythmogenesis during sequential cognitive tasks.

Persistent Gamma

Models of gamma rhythms involving tonic activation of kainate, muscarinic, and metabotropic glutamate receptors generate oscillations that persist for many hours. Although in some situations a small degree of ING can be uncovered (Fisahn et al.,2002), in most cases no tonic excitation of interneurons is apparent. In vivo PG rhythms survive selective genetic ablation of inhibitory synaptic connectivity between parvalbumin immunopositive basket cells (Wulff et al.,2009). Instead, interneurons receive large (up to 20 mV) compound phasic excitatory inputs at gamma frequency suggesting a PING rhythm. However, this is often not the case. Mechanistic differences exist when comparing these PG rhythms with conventional PING—a rhythm in which interneurons are recruited into the local circuit mainly by phasic AMPA-receptor-mediated events generated by orthodromic principal cell spiking. These differences are:

  • 1Very low somatic spike rates in principal cells. Modal somatic spike frequencies from zero to c.4 Hz are seen—a finding that seems surprising, given the large excitatory input to interneurons (Gillies et al.,2002; Cunningham et al.,2003,2004b). Given that each of these pharmacological models is associated with decreased postsynaptic potential amplitude (see above), even with the large degree of convergence of principal cell inputs onto interneurons, it is hard to see how the enormous phasic EPSPs on interneurons can be generated by somatic spiking alone. The somatic action potentials that are seen in these models have predominant inflections on their rising phase, indicative of an initial segment—somatodendritic break. Apparently, most of these somatic spikes are generated antidromically.
  • 2Prominent spikelets are seen in most principal cells in cortical areas (e.g., Cunningham et al.,2004a,b). These spikelets have rapid kinetics and a somatic membrane potential sensitivity that suggests partial back propagation of axonally initiated action potentials (Draguhn et al.,1998; Schmitz et al.,2001).
  • 3PG rhythms are highly gap junction conductance-dependent. Alkalinization enhances the gamma power and carbenoxolone abolished the rhythm completely (Traub et al.,2000,2003a,b; Cunningham et al.,2003,2004b; etc.). The gap junctions involved are not those between interneurons, nor can the effects of carbenoxolone be explained by NMDA receptor blockade (Chepkova et al.,2008) as many of these rhythms do not involve NMDA receptors (Fisahn et al.,1998; Roopun et al.,2006). PG rhythms still occur in vivo and in vitro in the absence of connexin 36 (Hormuzdi et al.,2001; Buhl et al.,2003), and may even be enhanced to the point of instability in some models (Pais et al.,2003). Evidently, the gap junctions required for PG rhythms use a structural protein other than connexin 36; the identity of this putative protein, however, remains to be established.
  • 4There is a richer dynamic in persistent rhythms than just the gamma rhythm alone. In local field potentials, a VFO (c.100–200 Hz) can be seen nested in most periods of the gamma (Traub et al.,2003a,b; Cunningham et al.,2004a,b) that is recorded continuously when connections between principal cells and interneurons are lesioned. This VFO is also seen robustly in IPSP/C recordings of the AMPA receptor-mediated phasic input to interneurons (Whittington and Traub,2003; Gloveli et al.,2005), suggesting an origin somewhere between the principal cell soma and excitatory synapses onto interneurons—that is to say, somewhere along the principal cell axons.

A functional connection between the presence of spikelets in principal cells and a field VFO, coupled with the non-interneuronal gap junction dependence, led to the suggestion of a gap junctionally connected plexus of axons in massively parallel cortical structures (Draguhn et al.,1998). Such connectivity has been seen with dye coupling (Schmitz et al.,2001) and at the ultrastructural level (Hamzei-Sichani et al.,2007). The working hypothesis for PG rhythms therefore differs fundamentally from that of PING. Drive to interneurons is entirely phasic, consisting of compound EPSPs generated by ectopic spike propagation through an electrically coupled principal cell axonal plexus. This idea explains the discrepancy between low (or zero) somatic spike rates in principal cells and the intense phasic excitatory synaptic drive to interneurons. It also explains the presence of VFO nested within the gamma rhythm and the correlation between field potential VFO and the discernible components of the compound interneuronal EPSP (Cunningham et al.,2004b). The lack of necessity for somatodendritic drive may also explain the stable network frequency from such rhythms across large changes in drive/gamma power [Fig. 1(B)] All these properties of PG can be explained by such a mechanism in computational models (Traub et al.,2000).

When the axons in the plexus are strongly coupled by gap junctions, the necessary rates of ectopic spike generation (in computational models) are quite small. Nevertheless, one still needs to ask: what generates ectopic axonal spikes? Overspill of GABA from synapses may be a candidate mechanism (Traub et al.,2003a,b) as may direct axonal excitation via kainate receptors (Semyanov and Kullmann,2001). In neocortex, however, computational modeling predicted a cell-specific source of such activity. Fast rhythmic bursting cells (chattering cells) are active during sensory-induced gamma rhythms in neocortex in vivo (Gray and McCormick,1996) and in vitro (Cunningham et al.,2004a) and their intense bursts of gamma frequency spikes may serve to populate a principal cell axonal plexus with “ectopic” action potentials (Traub et al.,2005). These neurons do not control the frequency of the gamma rhythm (Cardin et al.,2005) but when present in a network, their fast rhythmic bursting behavior is critical for PG rhythm generation (Fig. 3).

Figure 3.

Persistent neocortical gamma rhythms are associated with antidromic spike generation and require fast rhythmic bursting cell involvement. A: Layer 2/3 regular spiking (RS) principal cells show prominent spikelets in somatic recordings during persistent gamma rhythms (left panel). Example traces show 250 ms epochs of activity at resting membrane potential (−63 mV) and at −45 mV. The latter potential was generated by tonic depolarising current injection to inactivate somatic full action potential generation. Note spikelets are distinct from occasionally observed EPSPs (inset) and survive this inactivation of fast perisomatic sodium channels. During layer 2/3 gamma rhythm generation, fast rhythmic bursting cells (FRB) generate intense bursts of spikes at gamma frequency in a manner dependent on persistent sodium channel activation (right panel, 500 ms epochs). Blockade of persistent sodium channels with phenytoin (50 μM) abolished bursting but maintains single spike generation (lower trace). B: Intense burst discharges in FRB cells are required for neocortical persistent gamma rhythm generation. Hippocampal local circuits do not contain FRB cells. Blockade of persistent sodium channels does not significantly attenuate local field potential gamma rhythms in this region (left panel). In contrast, neocortical networks, containing FRB cells, generate gamma rhythms that are critically dependent on persistent sodium channel-dependent bursting in FRB cells (right panel). Scale bars 250 ms, 0.2 mV. Figure adapted from Cunningham et al.,2004a,b.


The commonly accepted origin of inhibition underlying gamma rhythms is the fast spiking, perisomatic targeting, parvalbumin immunopositive basket cell. Original work showed that fast spiking interneurons with cell bodies in hippocampal stratum pyramidale generated gamma frequency outputs (Whittington et al.,1995). Modeling work showed that, for all modes of gamma generation, synaptic inhibition needed to have fast kinetics and target cell compartments electrotonically close to action potential initiation sites (Traub et al.,1997; Kopell et al.,2000; White et al.,2000). However, while some later studies confirmed the dominant role for persiomatic-targetting basket cells (Hájos et al.,2004; Mann et al.,2005), others revealed further underlying complexity.

Somatic vs. Dendritic Principal Cell Targets

Interneurons that target distal dendrites of hippocampal principal cells receive gamma frequency phasic excitation during population gamma rhythms, but most do not generate an accompanying gamma frequency output (Gillies et al.,2002; Hájos et al.,2004; Gloveli et al.,2005). Instead, their action potential generation is resonant around theta frequency (Maccaferri and McBain,1996; Pike et al.,2000). Radiatum and lacunosum/moleculare interneurons are an exception to this rule. They fire robustly at gamma frequency during local circuit gamma rhythms (Hájos et al.,2004). However, their spike timing was not observed to be phase locked to the on-going gamma rhythm. Interestingly, principal cell somatic compartments, despite receiving gamma frequency inhibitory inputs, also demonstrate theta resonance (Cobb et al.,1995; Pike et al.,2000), whereas it is the proximal dendritic compartments that show subthreshold (for sodium spike generation) gamma frequency resonance (Traub et al.,2003a). Matching of input frequency with intrinsic resonance frequency has a marked potentiating effect on resulting neuronal responses (e.g., see Hutcheon and Yarom,2000). Control of orthodromic spike generation in principal cells may therefore involve proximal dendritic gamma frequency inhibition as much as perisomatic inhibition. Gamma frequency IPSPs are seen several hundreds of microns along the apical dendritic shaft of hippocampal principal cells (Gillies et al.,2002) and GABAergic interneurons targeting these cell compartments preferentially (bistratified cells) are as active at gamma frequency as parvalbumin immunopositive, somatic targeting basket cells (Gloveli et al.,2005).

Spatial Scale of Interneuron Networks?

Basket cells that generate the majority of gamma frequency inhibitory input to somata have dense axonal arbours extending ∼0.5 mm in length in CA3 (e.g., Fig. 4). Synchronization over longer distances, in the absence of principal cell involvement, therefore requires the coupling of many interconnected cells. However, bistratified cells (above) and CA1 basket cells can have a breadth of axonal arbour up to three times this distance, suggesting that single gamma frequency output-generating interneurons may synchronize populations of principal cells over 1.5–2 mm. The synchronizing properties of heterogeneous interneuron networks may be even greater. In in vitro experimental gamma rhythm models, the degree of action potential output was much greater in one particular interneuron subtype than in all others: trilaminar interneurons generated pairs of action potentials on almost all periods of a kainate receptor-driven gamma rhythm in area CA3 (Gloveli et al.,2005). These neurons have axon arbours that extend for many millimeters, forming synapses in areas CA3, CA1, and even subiculum (Fig. 3). The postsynaptic targets for these intensely gamma-generating interneurons are thought to be predominantly other interneurons (Sik et al.,1995), suggesting that they may play a role in the patterns of ING seen in hippocampus—particularly, as they generate two spikes per gamma period; such observations may explain the discrepancy between the kinetics of inhibitory postsynaptic events seen in basket cells during gamma rhythms and the kinetics of basket-basket interactions (see above).

Figure 4.

Multiplicity of GABAergic interneurons recruited into gamma rhythms. Cartoon illustrating the many different subtypes of interneuron generating gamma frequency outputs during network gamma rhythms. Axonal and dendritic arbours are both shown. From left to right—Red neuron: CA3 basket cell. Blue neuron: CA3 bistratified cell. Green neuron: CA3 trilaminar neuron (note axon arbour extends across area CA1 to the subiculum). Purple neuron: CA1 basket cell. Red neuron: Layer 2 medial entorhinal cortical basket cell. Orange neuron: Layer 3 medial entorhinal cortical goblet cell. Example reconstructions taken from: Gloveli et al.,2005; Bibbig et al.,2007; Middleton et al.,2008a.

Multiple Interneuron Subtypes Controlling Gamma Rhythms

Previous studies have shown involvement of many diverse subtypes of interneuron in population gamma rhythms (above), so it seems likely that local circuits have more than one mechanism for generating inhibition-based gamma rhythms. In hippocampus, subtle but robust pharmacological differences in PG rhythms generated by carbachol or mGluR activation strongly suggested that different interneuron subtypes were involved in generating each rhythm (Pálhalmi et al.,2004), despite similar network dynamics.

Further evidence from medial entorhinal cortex has shown that two anatomically distinct subclasses of local circuit interneuron in superficial layers compete for control over gamma rhythm generation (Middleton et al.,2008a,b; Fig. 5). Parvalbumin immunopositive layer 2 basket cells and layer 3 goblet cells (Fig. 4) both participate in superficial entorhinal gamma rhythms. However, the degree of involvement of each cell type dictates the frequency of the network rhythm within the gamma band, in a manner dependent on the activity of NMDA receptors in the local circuit. Only entorhinal layer 2 basket cells receive tonic excitation via NMDA receptors and participate in a network gamma rhythm at the higher frequency end of the conventional gamma band (>40 Hz). Removal of this NMDA drive to these cells almost abolishes their output and facilitates the recruitment of goblet cells into a slower (c. 30 Hz) network gamma rhythm (Fig. 5B). Thus, the degree of depolarization of interneurons in entorhinal cortex may serve to dynamically route inputs to the hippocampus, with CA3 networks being tuned to the goblet cell-mediated slower gamma rhythm and CA1 networks being tuned to the basket cell (higher frequency) gamma rhythm (Middleton et al.,2008a,b). Further modulation of input preference during the theta rhythm in area CA1 may form part of the larger scale network dynamic associated with spatiotemporal information processing (Colgin et al.,2009).

Figure 5.

Developmental modification of NMDA receptor mediated excitation of interneurons disrupts gamma rhythm generation in a cell-specific and region-specific manner. A: Lysophosphatidic acid receptor 1 knock-out selectively reduces the number of parvalbumin immunopositive (PV+) cell bodies in adult layer 2 medial entorhinal cortex (mEC). Left panel shows the laminar distribution of PV+ cell bodies in control? mEC. Note these cells are present across layers 2–5. Middle panel shows the relative densities of PV+ cell bodies in each layer in wild type tissue (white bars) and tissue from adult LPA1−/−mice (gray bars). Only layer 2 shows a reduction in PV+ cell count. Right panel: This lamina-specific reduction in PV+ cell bodies is accompanied by the same lamina-specific decrease in persistent gamma rhythm generation. This layer 2-specific deficit is, in part, mimicked by acute blockade of NMDA receptors with ketamine (20 μM). B: NMDA receptor-independent gamma rhythms involve different local circuit interneurons. Spectrograms of local field potential activity show the gamma rhythm remaining after NMDA receptor blockade is of lower amplitude and power in mEC. This lower frequency gamma rhythm is mediated by layer 3 goblet cells and not the layer 2 basket cells underlying control mEC persistent gamma rhythms. Figures adapted from Whittington,2008 (A) and Middleton et al.,2008a (B).


GABAergic Synaptic Responses

In humans, gamma rhythms predominate in most cortical regions during childhood. Significant increases are seen between 3 and 5 years but the rate of change is different in different areas (Takano and Ogawa,1998). As gamma band power increases during childhood, a concomitant increase in cognitive performance is seen (Uhlhaas et al.,2009). This process is disrupted in adolescence but restored with a pronounced increase in gamma band synchronization in early adulthood—a developmental time period coinciding with peak maturation of perisomatic parvalbumin-containing inhibitory interneuronal terminals in primates (Erickson and Lewis,2002).

In rodents, the onset of gamma rhythm generation appears to coincide with the switch to the adult form of GABAergic inhibition (Tyzio et al.,2007). In neonatal tissue, spontaneous neocortical activity is predominantly a GABA-independent beta2 frequency rhythm (Dupont et al.,2006) which survives in certain cortical regions into adulthood (Roopun et al.,2006). Immature cat cortical responses to visual sensory input are at alpha rather than at gamma frequencies. From a network dynamic perspective, this is perhaps not surprising: network rhythms are a result of interaction between intrinsic neuronal properties and the properties of the synaptic and electrical communication between neurons. Neurons “tuned” to synchronize with inhibitory, hyperpolarizing inputs cannot do so when connected by excitatory, depolarizing inputs (e.g., Haas and White,2002). At around P8–12 in rodents, GABAA receptor mediated responses change from depolarising to hyperpolarizing owing to changes in the behavior of chloride transport proteins (Rivera et al.,1999). Prior to establishment of the hyperpolarizing, adult form of GABAergic inhibition, immature circuits in hippocampus may generate occasional spontaneous, brief burst discharges (Leinekugel et al.,2002)—a behavior that can be mimicked in mature tissue with potassium channel blockade (Michelson and Wong,1991). These discharges may drive neocortical circuits and contribute, along with sensorimotor inputs, to the generation of spindle bursts associated with motor system maturation (Khazipov et al.,2004).

Glutamatergic Control of Interneuron Endophenotype

Studies in primates show that persiomatic synapses onto principal cells from parvalbumin-containing neurons (one of the main components of the local gamma generating circuit) develop at a much later time than other features of inhibition (Erickson and Lewis,2002). Both PING and PG rhythms depend predominantly on glutamatergic input to parvalbumin-containing neurons, but evidence suggests that the development and maintenance of interneuron “endophenotype,” vital for gamma rhythm generation, is also sensitive to the degree and type of glutamatergic input received by immature perisomatic targeting interneurons.

In culture, mouse cortical interneurons had an endophenotype that was dependent on activation of NMDA receptors. Blockade of NMDA receptors with ketamine reduced both GAD67 and parvalbumin by c.50% while preventing NMDA receptor-mediated elevations in CREB and ERK (Kinney et al.,2006). The maintenance of parvalbumin and GAD67 levels was specifically associated with activation of NR2A subunit-containing receptors, present in fivefold higher ratios than in principal cells, and could be complemented by activation of the mGluR receptor agonist DHPG—an effective means of inducing gamma rhythms in cortical tissue (Gillies et al.,2002). A relationship between NR2A receptors and PV expression rates is also seen in postmortem cortical samples from patients with schizophrenia (Woo et al.,2004). Interestingly, the opposite relationship between glutamate receptor activation and interneuron endophenotype is seen in cerebellar cultures (Kovács et al.,2003). In this region, where gamma is generated by ING-like mechanisms independently of glutamate receptor activation, elevated glutamate levels decreased the number of neurons displaying GABAergic endophenotype—a phenomenon that could be abolished by blockade of AMPA and NMDA receptors.

Mechanisms underlying this dependence of interneurons on glutamatergic input may be many-fold. Loss of GAD67 and PV in cortical neurons following NMDA receptor blockade has been linked to an increase in superoxide levels secondary to activation of NADPH oxidase (Behrens et al.,2007). In addition, during development the relative ratios of excitation and GABAergic inhibition in local circuits are modified by NMDA receptors and subsequent activation of trkB-mediated signaling cascades (Liu et al.,2007). Genetically reduced trkB expression also reduced GAD67 and PV levels in adult mouse cortex (Hashimoto et al.,2005), suggesting that glutamatergic receptor-induced synaptic plasticity on interneurons during development plays a critical role in the normal maturation of circuits required for generation of inhibition-based gamma rhythms.

Further evidence for a role of interneuron excitation in normal local circuit development comes from studies showing that either pre or postnatal block of NMDA receptors selectively decreases parvalbumin expression levels in adulthood (Abekawa et al.,2007; Wang et al.,2007). Grossly disrupted cell migration and laminar organization of cortex, following methylazoxymethanol administration to pregnant rats, reduced PV expression rates and reduced auditory sensory-induced gamma frequency responses during a latent inhibition paradigm (Lodge et al.,2009). More subtle developmental modifications in NMDA receptor signaling—the lysophosphatidic acid receptor 1 knock-out (LPA1−/−)—also demonstrated a correlation between interneuron NMDA drive and ability of local circuits to generate gamma rhythms in adulthood (Cunningham et al.,2006). The LPA1 receptor is expressed predominantly during development (Allard et al.,1998) where it is associated with neurogenesis and cell migration (Contos et al.,2000; Yan et al.,2003) and potentiates NMDA receptor-mediated postsynaptic responses (Lu et al.,1999). The expression profile of LPA1 after birth parallels the expression profile of other markers associated with interneuron endophenotype. For instance, parvalbumin expression and Kv3.1b potassium channel expression in interneurons start toward the end of the first postnatal week and tend to peak in the second-third week, similarly to LPA1 (Bergmann et al.,1991; Du et al.,1996). Reduced NMDA signaling during early development in this model was associated with a highly spatially selective reduction in parvalbumin-containing cell count that was directly related to the ability of local circuits to generate gamma rhythms (Fig. 5A). This spatially-selective effect was mimicked by acute NMDA receptor blockade which selectively reduced parvalbumin-immunopositive basket cell excitation (Fig. 5B), exposing a different local circuit gamma generator (see above).


Data from in vitro models of gamma rhythms indicate many different possible mechanistic origins for inhibition-based gamma rhythms. The conventional delineation into ING and PING processes is inadequate to explain the multiplicity of patterns of interneuron recruitment into this EEG rhythm. Specifically, many commonly used models of PG rhythms do not fall neatly into either of these categories and are most likely generated by seemingly random processes (ectopic spike generation) in networks of principal cell axons. Having “noise” as a generator of networkprocesses involved in cognition may appear detrimental to current theories of brain function, but stochastic dynamic behavior imparts many computational advantages that may have relevance to sensory signal detectability, dynamic routing of sensory representations, memory formation and attention (Deco et al.,2009). Furthermore, it appears that different brain regions have different intrinsic mechanisms for gamma generation, with the precise origin of the gamma rhythm being modulated by the nature of coexistent neuromodulation and the degree of afferent activity. Selective dysfunction in one type of gamma rhythm may underlie certain psychiatric illnesses such as schizophrenia.