|Tamas Freund and Szabolcs Kali (2008), Scholarpedia, 3(9):4720.||doi:10.4249/scholarpedia.4720||revision #89023 [link to/cite this article]|
Interneurons are types of nerve cells, typically found in integrative areas of the central nervous system, whose axons (and dendrites) are limited to a single brain area. This feature distinguishes them from principal cells, which often have axonal projections outside the brain area where their cell bodies and dendrites are located. Principal neurons and their networks underlie local information processing/storage and represent the major sources of output from any brain region, whereas interneurons, by definition, have local axons that govern ensemble activity. While principal cells are mostly excitatory, using glutamate as a neurotransmitter, interneurons most often use gamma-aminobutyric acid (GABA) to inhibit their targets. Since GABA acts mainly through opening ion channels permeable to chloride/bicarbonate or potassium ions in the postsynaptic neuron, interneurons achieve their functional effects via hyperpolarizing large groups of principal cells (although, under some circumstances, they may also mediate depolarization or shunting - see below). Interneurons in the spinal cord may use glycine, or both GABA and glycine, to inhibit principal cells, whereas interneurons of cortical areas or the basal ganglia may release various neuropeptides (cholecystokinin, somatostatin, vasoactive intestinal polypeptide, enkephalins, neuropeptide Y, galanin, etc.) in addition to GABA. In some regions, such as the basal ganglia and the cerebellum, principal neurons are also GABAergic.
Most types of interneuron innervate predominantly local principal cells. The dominant excitatory drive of interneurons may originate from outside the territory (i.e., area/subfield/layer) of their neuronal targets, in which case they are said to mediate feed-forward inhibition. Conversely, if the dominant source of input to an interneuron is the same population of cells which is targeted by the interneuron, it is considered to provide feed-back inhibition (Figure 1). The major functional roles of interneurons are thought to be the control of activity levels within a brain area, coordination/generation of rhythmic and intermittent discharge patters of neuronal ensembles, as well as the gating and gain control of excitatory inputs to principal cells. The outcome of interneuron activity is often synchronization of firing in large principal cell populations, which facilitates synaptic potentiation (enhancement of synaptic efficacy) and the formation of potentiated cell ensembles or networks as correlates of memory traces.
Diversity of interneurons, both in terms of structure and function, increases with the complexity of local networks within a given brain area, and this likely correlates with the complexity of the functions carried out by that brain area. Accordingly, the 6-layered neocortex, as the center of the highest nervous functions such as conscious perception or cognition, has the largest number of interneuron types. A simpler version of cortical architecture can be found in the hippocampus, where the fundamental principles of interneuronal connectivity are the same as in the neocortex, but due to the simple lamination, these circuits can be dissected and the functional roles extrapolated with greater precision. The present description is therefore going to focus largely on interneurons of the hippocampus and neocortex, but occasionally mention analogies and homologies with other areas of the central nervous system.
Diversity and classification of cortical inhibitory interneurons
Inhibition in cortical networks is supplied by a wide range of interneuron types that subserve different functions, and possess specific connectivity features, morphological, electrophysiological and neurochemical characteristics that enable them to fulfil these roles. Their major transmitter is invariably GABA, but several types may contain and co-release neuropeptides as well. Any meaningful classification scheme can only be based on function; therefore, differences in morphological (e.g., the number or laminar distribution of dendrites, the arborization pattern of axons, co-transmitter content, receptor expression pattern) or physiological features (e.g. firing pattern, short- and long-term synaptic plasticity) should be used to generate new cell types only if these differences are causally linked to a functional dichotomy. Detailed reviews have been published by Freund and Buzsáki (1996) and several others (e.g., McBain and Fisahn, 2001; Somogyi and Klausberger, 2005) on this subject. For a functional classification, the most important features are the input-output properties, i.e. the identity of afferents driving a particular cell type, and the sites where these cells exert their inhibitory effects. Since these functional specializations directly determine the morphology of the dendritic trees and axonal arbours of interneuron types, even the purely descriptive morphological details are important from a functional point of view.
Most interneurons innervate different types of target cells (both principal cells and interneurons) approximately in proportion of their occurrence in the neuropil, and thus synapse predominantly on the most abundant cell type, which are local principal cells. However, some special interneurons preferentially target other interneurons. This represents the first dichotomy in the classification scheme, since there are subsets of interneurons containing the calcium binding protein calretinin (CR), or the neuropeptide vasoactive intestinal polypeptide (VIP), which were shown to form synapses predominantly on other GABAergic cells, and were therefore termed Interneuron-Selective interneurons (or IS cells). These IS cells occur in all subfields and layers of the hippocampus, often form dendrodendritic contacts with each other, and are ideally suited to synchronize inhibition throughout the hippocampus. Interneurons with similar target selectivity and calcium binding protein content have been found also in neocortex. IS cells are relatively scarce; the vast majority of interneurons selectively innervate different compartments of principal cells besides establishing a small number of synaptic contacts with other interneurons.
The next major branching in the classification scheme of interneurons is determined by the principal cell compartment which they target, since inhibition targeting pyramidal cell dendrites is functionally distinct from inhibition that primarily affects the perisomatic region. The dendritic tree is the site of termination of most if not all excitatory (largely glutamatergic) afferents of principal cells, as well as the site of short- or long-term plastic changes in their efficacy resulting from the action of various dendritic active conductances, which generate calcium spikes and backpropagating action potentials. Thus, interneurons innervating principal cell dendrites (the Dendritic Inhibitory cells) are designed to control the efficacy and most notably the plasticity of excitatory inputs to principal cells. On the other hand, the perisomatic domain is responsible for the summation of postsynaptic potentials arriving from all dendritic branches, which then may or may not reach threshold for action potential generation. Thus, the perisomatic region - and particularly the axon initial segment, which has the highest concentration of voltage-gated sodium channels - plays a crucial role in the generation of output. Interneurons innervating somata, proximal dendrites or axon initial segments (the Perisomatic Inhibitory cells) are specialized to control output from large principal cell populations, and one of their main functions is to rhythmically synchronize principal cell activity at theta (4-8 Hz) and gamma (30-100 Hz) frequencies.
There have been numerous attempts to classify cortical interneurons on the basis of their electrophysiological characteristics, and to relate the results of such classifications to morphologically defined cell types. In vitro studies typically revealed a wide variety of response patterns to artificial current stimulation, and showed large variability even within the same morphologically defined cell class. These results have led some to propose that a functional cell class may only be fully determined by a conjunction of anatomical and physiological features, leading to a potentially astronomical number of functionally distinct cell types (see Markram et al., 2004). However, the same question arises here as in case of morphological details, i.e., which characteristics are relevant from a functional point of view, which are those that evolved as an adaptive property to enable the cell type to fulfil its job more efficiently, and which are those that exist simply as a side product, but survive similarly to natural mutations, which are abundant in the genome of any living creature. In addition, at least some of the differences seen in physiological behavior may result from the artificial recording conditions in vitro, and a different classification or even more homogeneous responses might result from the conditions (including the concentrations of metabotropically active neurotransmitters and the rates and patterns of excitatory and inhibitory synaptic input) experienced by cortical interneurons in the behaving animal.
A strong argument for the functional homogeneity of anatomically/neurochemically defined cell classes was recently provided by recordings from anatomically identified interneurons of the hippocampus in anesthetized animals during behaviorally relevant population activity patterns. Neurons of the hippocampus (as well as neocortical areas) are known to discharge in a temporally organized manner during various behavioral states (such as exploration and specific stages of sleep); characteristic patterns include oscillatory activity at frequencies in the theta and gamma range, as well as intermittent increases in activity known as sharp waves and associated high-frequency oscillations called ripples. All of these patterns are detectable in local field potential recordings in the appropriate hippocampal subfields, but also in the coordinated activity of principal cells and interneurons. Importantly, the discharge patterns of neurons in relation to these naturally occurring population events offer a powerful and functionally relevant way of classifying cortical interneurons. The results of these studies indicated that each anatomically/neurochemically defined cell type had characteristic (e.g., phase-locked) firing patterns relative to theta and gamma oscillations and sharp-wave-associated ripples (see below for cell-type-specific details), consistent with a functionally homogeneous role of each anatomical class in the generation of population activity (Somogyi and Klausberger, 2005).
Interneurons also differ in the molecular identity and density of voltage-gated ion channels in their cell membranes, and in the molecular composition of their input and output synapses. These features were also found to correlate with anatomical/neurochemical class as defined above, and endow interneurons with cell-type-specific (intrinsic and synaptic) physiological properties, many of which have been shown to be functionally relevant. For instance, parvalbumin-containing basket cells express voltage-gated potassium channels of the Kv3 family, which contributes essentially to the "fast-spiking" behavior of these neurons, allowing them to respond to incoming synaptic input with minimal delay, and to sustain high-frequency spiking activity in response to strong input. Differences in the expression of voltage-gated ion channels between interneuron types also underlies the differential dependence of the subthreshold voltage response on the frequency of stimulation ("impedance profile"). Some neuronal classes show distinct peaks ("resonances") in their impedance profiles at characteristic frequencies, which may contribute to their differential involvement in the generation of network oscillations in various frequency bands. Finally, both excitatory synapses onto interneurons and the inhibitory synapses formed by interneurons on local principal cells have physiological properties that depend on interneuronal class. Synapses may differ in the amplitude and kinetics of the postsynaptic response. Many glutamatergic synapses onto interneurons activate Ca-permeable AMPA-type receptors lacking the GluR2 subunit, which is an important determinant of the rapid synaptic signaling that characterizes these connections. Input and output synapses of interneurons also differ in the type of short-term plasticity (facilitation and/or depression) they express, leading to a differential involvement of various interneuron types in shaping network activity which depends on the rate and temporal pattern of firing in principal cells and the interneurons themselves.
In the sections below, the two largest classes of cortical interneurons (i.e., perisomatic and dendritic inhibitory cells) will be described in more detail, based mostly on data from the hippocampus, but with an emphasis on features that can be generalized to the entire cerebral cortex.
Perisomatic inhibitory cells
The precise site of termination as well as specific input characteristics allows to dissect this cell group into three major types. Axo-axonic, or chandelier cells innervate exclusively the axon initial segments of principal cells, and occur both in the hippocampus and in the neocortex. Each axo-axonic cell forms radially running rows of boutons that climb along the axon initial segments of about 1000-1500 pyramidal cells. Due to the strategic location of their axon terminals, axo-axonic cells have been claimed to simultaneously inhibit output from large principal cell populations. However, recent evidence suggests that their postsynaptic GABAA -receptor-mediated effect may be depolarizing, and, as a consequence, they may be able to discharge the entire pyramidal cell population they innervate, with the aim of synchronizing their output, or to reset conductances in their dendritic trees. In the hippocampus, axo-axonic cell bodies are located within or near the pyramidal cell layer, and have radially running dendritic trees that span all layers, although a few exceptions may have dendrites confined to stratum oriens. Intracellularly labelled axo-axonic cells in the hippocampus had axonal arbours measuring 800 to 1,700 µm. Axo-axonic cells can be recruited by both feed-forward and feed-back input. Their firing activity is suppressed during and after hippocampal sharp wave-associated ripple oscillations (which are associated with synchronous discharges of CA3 pyramidal cells). However, axo-axonic cells often increase their firing just before ripple episodes. Their activity is phase-coupled to the peak of the theta waves recorded extracellularly in stratum pyramidale.
The remaining types of Perisomatic Inhibitory cells are basket cells that form multiple synaptic contacts on the somata and proximal dendrites of principal cells, each basket cell targeting about 1500-2000 of them. The transverse extent of basket cell axonal arbours in the hippocampus was found to be about 1 mm. Basket cells are present in many different brain areas including the cerebral and the cerebellar cortices (in the cerebellum, they inhibit Purkinje cells). In the neocortex and the hippocampus, several subtypes of basket cell have been distinguished. The two major subtypes of hippocampal basket cell can be distinguished easiest on the basis of their neuropeptide/calcium binding protein content; however, this neurochemical difference is coupled to functionally more meaningful unique features related to synaptic inputs and receptor expression patterns.
The parvalbumin (PV)-containing basket cell bodies are located within or near stratum pyramidale, and have radially running dendrites that span all layers. They are driven very efficiently by feed-forward excitatory input, as evidenced, for example, by the tight temporal coupling of the firing of PV-containing basket cells in CA1 to characteristic population activity patterns in CA3 in vivo, although many of them also receive input from local principal cells in a feed-back manner. They form synapses on pyramidal cells via alpha1 subunit-containing GABAA receptors. The cholecystokinin (CCK)-containing basket cell bodies can be located in all layers, but typically in stratum radiatum, and have radially running dendritic trees spanning all layers. CCK basket cells generally receive far less local glutamatergic inputs than PV cells, but are selectively innervated by serotonergic afferents from the median raphe via 5HT3 receptors. They also express nicotinic alpha7 and alpha4, as well as presynaptic cannabinoid (CB1), GABAB and estrogen1 receptors, and form synapses on pyramidal cell bodies and dendrites largely via GABAA receptors enriched in alpha2 subunits.
In contrast to PV basket cells, their predominant local drive is feed-back, and due to their large membrane time constants they are uniquely capable of summating feed-forward and feed-back drives over long time windows. During theta activity, both basket cell types fire at the peak of extracellularly recorded field potential waves (out of phase with pyramidal cells, which fire at the trough), with PV cell firing shifted slightly towards the descending, and CCK cell firing towards the ascending phase. While PV cells strongly increase their firing during sharp waves, CCK cell discharges remain unchanged. This is consistent with the fact that CCK cells can be activated only by a combination of feed-back and feed-forward drives, while ripples in CA1 are driven by CA3 in a feed-forward manner.
Although both basket cell types are primarily responsible for rhythmic synchronization of the activity of large principal cell populations at theta and gamma frequencies, there is a considerable division of labour between them associated with their differential behaviour during sharp waves, and with different excitatory drives. A synthesis of the above features led to the proposal (Freund, 2003) that the electrically and synaptically coupled, and mostly locally driven, ensembles of PV-containing basket cells are indispensable components of the oscillating cortical hardware; they represent a precision clockwork without which no cortical operations are possible. The activity of a similar syncytium of CCK-containing basket cells is superimposed on the PV basket cell-entrained network, conveying emotional and motivational effects carried by serotonergic and cholinergic pathways. In addition, actions of the CCK cell ensemble are highly modifiable by local neuromodulators and retrograde signal molecules, which may allow further fine tuning of principal cell cooperation. Impairment of this tuning system likely results in mood disorders such as anxiety. Interestingly, most if not all of the inputs and receptor expression patterns that distinguish CCK cells from PV cells are strongly implicated in anxiety (for reviews see Freund, 2003; Freund and Katona, 2007).
Dendritic inhibitory cells
This group of interneurons is the most diverse both morphologically and functionally. Dendritic Inhibitory cells are present in many different parts of the nervous system, including the cerebellum (where stellate cells target Purkinje cell dendrites and Golgi cells innervate granule cell dendrites), the olfactory bulb (where inhibitory granule cells and periglomerular cells establish dendro-dendritic synapses onto excitatory mitral cells), and all areas of the cerebral cortex. In the neocortex, a large variety of Dendritic Inhibitory interneurons have been described (for review see Markram et al., 2004). These include Martinotti cells, which target mainly the apical tuft region of pyramidal cells, contain the neuropeptide somatostatin (and are thus to some extent similar to O-LM cells of the hippocampus, as described below), double bouquet cells and bipolar cells, which target mostly basal dendrites, as well as bitufted cells and neurogliaform cells. However, the precise functions of these neocortical cell types have been difficult to identify. The functionally relevant classification based on afferent and efferent connectivity is easier to determine in the hippocampus, where afferent pathways and cellular compartments are confined to different laminae. Distinct types of dendritic interneurons have evolved to control glutamatergic inputs of principal cells from different sources. The so-called Oriens – Lacunosum-Moleculare (O-LM) interneurons selectively terminate in stratum lacunosum-moleculare in conjunction with excitatory input coming from the entorhinal cortex. They have a cell body with horizontally oriented dendritic tree which is confined to the stratum oriens in the CA1 subfield but reaches all layers except stratum lacunosum-moleculare in the CA3 subfield. In the dentate gyrus, the analogous neurons have been named HIPP (Hilar Perforant Path associated) cells, since their somata and dendrites are confined to the hilus, whereas the axon is extensively arborizing in the outer two-thirds of the molecular layer, in perfect register with the entorhinal input to granule cells. Both of these cell types are driven exclusively in a feed-back manner by recurrent collaterals of local principal cells. They are characterized by the presence of the neurochemical marker somatostatin (SOM) in conjunction with the absence of calbindin (CB). The axonal arbour of O-LM cells is rather focused in the transverse plane, and elongated in the longitudinal direction, covering an area of about 400 by 800 µm. During theta activity, they fire together with the pyramidal cells at the trough of extracellularly recorded field potential waves, and are silenced during sharp wave-associated ripples. Their function may involve the blockade of synaptic plasticity of entorhinal inputs in time windows when the largest number of local pyramidal cells discharge. This occurs during the trough of theta, and provides efficient feed-back activation for the O-LM cells. However, when pyramidal cells fire in phase precession mode, ahead of the population firing – this happens when they are conveying specific signals e.g. by place field-dependent firing – they can escape feed-back inhibition from the O-LM cells, and their entorhinal input synapses can be potentiated.
The other major excitatory pathway innervating hippocampal principal cells derives from pyramidal cells of the CA3 subfield, which give rise to Schaffer collaterals that terminate in strata radiatum and oriens in area CA1. The interneurons that specialize in controlling this input are called Bistratified cells. This cell type has an axon that arborizes in perfect register with Schaffer collaterals, i.e. in strata radiatum and oriens, skipping stratum pyramidale, has a spread of approximately 2 mm in both transverse and septotemporal directions, and forms synaptic contacts with second or third order thin dendrites of pyramidal cells, in the vicinity of spines that receive the excitatory input. The axon of a subtype is confined to stratum radiatum only, and there seems to be a continuum between this subtype and the classical bistratified arbours with a varying size of a stratum oriens component. The bistratified cell bodies can be found in all layers, and possess radially oriented dendritic trees. The molecular characteristics of this cell type are controversial, but may include the expression of CB and/or PV.
There is a CCK-containing subset that has considerable amounts of axon entering str. lacunosum-moleculare, and is therefore not considered as a classical bistratified cell. It is probably closer to the type of CCK-containing cells that selectively innervates str. lacunosum-moleculare, and is called LM-Perforant Path associated (LM-PP) cell. The dendritic tree of both bistratified and LM-PP interneurons arborizes most extensively in str. radiatum or lacunosum-moleculare, which explains their predominantly feed-forward drive.
Bistratified cells fire in phase with O-LM and pyramidal cells during theta activity (i.e. at the trough of the extracellularly recorded field potential waves), whereas they are strongly excited during sharp wave-associated ripples consistent with a strong feed-forward input from Schaffer collaterals. Their likely role is a feed-forward gain control of the efficacy of Schaffer collateral synapses, as well as the regulation of synaptic plasticity via inhibiting active conductances, calcium spikes and backpropagating action potentials in the dendrites.
Ivy cells, which were discovered only recently, also belong to the group of dendritic inhibitory interneurons. Their name reflects the dense, fine axonal meshwork which targets predominantly the basal and apical tufts of the dendritic tree of pyramidal cells. Cells of this type are suggested to be more numerous than parvalbumin-expressing neurons (previously thought to be the most abundant type of hippocampal interneuron), and are characterized by the expression of the enzyme nitric oxide synthase together with neuropeptide Y. They fire at a rather low rate, but phase-locked to network patterns such as theta and gamma oscillations. The slow rise and decay of the IPSCs they evoke in their target pyramidal cells and their overall similarity to neurogliaform cells in neocortex suggest that their postsynaptic effects may also be mediated, at least in part, by GABAB receptors.
Individual dendritic inhibitory cells of any type provide 2 to 20 synapses onto a single target pyramidal cell, which are scattered all over the dendritic tree. Thus, the contribution of a single dendritic inhibitory cell to the control of dendritic functions of principal cells is likely to be negligible. Thus, they have to fire in synchrony. Interneuron-selective interneurons (IS cells, see above) may synchronize dendritic inhibitory cells, thereby enabling them to exert a powerful effect on dendritic electrogenesis and associated synaptic plasticity.
Besides the classic local inhibitory interneurons described so far, several GABAergic cell types were recently discovered in the cortex which, in addition to innervating local principal cells or interneurons, send axon collaterals outside their host area. Although these neurons, by virtue of the existence of their long-range projections, do not meet the definition of interneuron given at the beginning of this article, they can probably fulfil similar roles within the local network through their short-range axon collaterals, and are therefore commonly referred to as "long-range interneurons". Long-range interneurons include the hippocampo-septal (HS) cells, which express SOM and CB, and have a cell body in stratum oriens of CA1-3, in str. lucidum and radiatum of CA3, and in the hilus of the dentate gyrus. They have horizontal dendrites that remain in str. oriens in CA1, and are restricted to the dentate hilus, which is consistent with a predominantly feed-back drive. The dendrites are extremely long, suggesting that the neuron is sampling input from a large population of local principal cells, apparently monitoring the level of global population synchrony. This information is then relayed to the medial septum, where their targets are the GABAergic pacemaker cells that project back to the hippocampus to selectively innervate all interneuron types. The local axon collaterals of HS cells span very large distances, they can cross the entire hippocampus from the rostral to the temporal pole, and one subset appears to innervate predominantly other interneurons (including other HS cells). At least some HS cells also project via a myelinated axon to the subiculum, the main target of CA1 pyramidal cell output. Thus, these HS cells are ideally suited to synchronize inhibition, and thereby the entire principal cell population, along the hippocampo-septo-hippocampal loop that is crucial for the generation of theta activity. There are HS cells that innervate mostly principal cell dendrites, and interneurons only in proportion of occurrence in the neuropil, and are thus functionally different from the interneuron-selective HS cells. There are reports of so-called backprojection neurons, which innervate all hippocampal subfields, form synapses primarily with principal cells, and are very similar to the HS cells. They may in fact also be HS cells whose septally projecting axon was accidentally not labelled.
Functions of cortical interneurons
Compared to the abundance of information on the morphological, physiological, and molecular properties of interneurons and their synaptic connections, there is still a relative paucity of direct information on the functional roles of interneurons in cortical computations. In this section, we attempt to briefly summarize some of the more compelling ideas about the possible functions of interneurons.
Perhaps the oldest and simplest idea is that interneurons maintain physiological activity levels in the brain, preventing runaway excitation in recurrent cortical networks. A similar role in the stabilization of network dynamics has been ascribed to the feedback inhibition mediated by Renshaw cells in motor regions of the spinal cord. This notion also offers a simplistic explanation for the critical involvement of interneurons in epilepsy. There is evidence that enduring changes in the level of excitation are accompanied by a corresponding change in the overall level of inhibition; however, transient imbalances between excitation and inhibition can also be induced. In the hippocampus as well as in the neocortex, changes in the level of interneuronal firing have been observed to accompany behaviorally relevant novel experiences, and probably contribute to enabling the plastic changes that are induced by such learning events.
As mentioned earlier, interneurons make a critical contribution to the generation of network oscillations, and synchronize the activity of principal cells during oscillatory and transient brain states (see Mann and Paulsen, 2007, for a recent review). Perisomatic interneurons in particular are thought to be indispensable for the generation of fast (gamma frequency) population rhythms, although the exact nature of their contribution could vary between different regions. Some forms of gamma oscillation may be generated through an interaction between excitatory and inhibitory neuronal populations; however, pure interneuronal networks, connected via GABAergic synapses (and often through electrical synapses formed by gap junctions as well), are also known to be capable of generating rhythmic, synchronous activity in the gamma frequency range. The rhythmic inhibitory input generated by such interneuron populations can in turn effectively synchronize the activity of local principal cell ensembles. This may enhance the efficiency of information transmission by increasing the likelihood of coincident spiking, which is an effective trigger of action potential firing in downstream neurons. Oscillatory activity in neuronal populations can also serve as a temporal reference signal for phase encoding, a form of temporal coding where the timing of action potentials with respect to the phase of an ongoing oscillation carries information. Such an encoding scheme may be utilized by hippocampal principal cells (also known as "place cells"), whose phase of firing in relation to the local theta oscillation (as well as their firing rate) changes systematically as the animal moves around in its environment (a phenomenon known as "phase precession").
In addition to maintaining homeostasis and providing a temporal framework for principal cell activity, interneurons likely play some more direct roles in cortical computations. Interneurons targeting specific dendritic regions can selectively gate excitatory input from different sources, thereby changing their relative contributions to the output of the cell. Dendritic inhibition may also control various forms of synaptic and cellular-level plasticity through its interaction with active dendritic processes. At the network level, both feed-forward and feed-back inhibition act to reduce the number of simultaneously active principal cells, working towards the creation of sparse representations (Acsády and Káli, 2007), which are thought to be advantageous for both sensory processing (Olshausen and Field, 2004) and long-term memory (Treves and Rolls, 1994). Feedback inhibition also introduces direct competition among members of a local principal cell population, whereby an increase in the activity of one cell tends to decrease the activity of its fellows. Such competition can be a simple but effective means of noise suppression, and - especially if complemented by local recurrent excitation - mediates selection between competing inputs, and may even implement complex computations such as working memory and decision-making in the neocortex (see. e.g., Machens et al., 2005).
- Acsády L. and Káli S. (2007) Models, structure, function: the transformation of cortical signals in the dentate gyrus. Prog. Brain Res. 163:577-599.
- Freund T.F. (2003) Interneuron Diversity series: Rhythm and mood in perisomatic inhibition. Trends Neurosci. 28:334-340.
- Freund T.F. and Buzsáki G. (1996) Interneurons of the hippocampus. Hippocampus 6:347-470.
- Freund T.F. and Katona I. (2007) Perisomatic inhibition. Neuron 56:33-42.
- Machens C.K., Romo R. and Brody C.D. (2005) Flexible control of mutual inhibition: a neural model of two-interval discrimination. Science 307:1121-1124.
- Mann E.O. and Paulsen O. (2007) Role of GABAergic inhibition in hippocampal network oscillations. Trends Neurosci. 30:343-349.
- Markram H., Toledo-Rodriguez M., Wang Y., Gupta A., Silberberg G. and Wu C. (2004) Interneurons of the neocortical inhibitory system. Nat. Rev. Neurosci. 5:793-807.
- McBain C.J. and Fisahn A. (2001) Interneurons unbound. Nat. Rev. Neurosci. 2:11-23.
- Olshausen B.A. and Field D.J. (2004) Sparse coding of sensory inputs. Curr. Opin. Neurobiol. 14:481-487.
- Somogyi P. and Klausberger T. (2005) Defined types of cortical interneurone structure space and spike timing in the hippocampus. J. Physiol. 562:9-26.
- Treves A. and Rolls E.T. (1994) Computational analysis of the role of the hippocampus in memory. Hippocampus 4:374-391.
- Valentino Braitenberg (2007) Brain. Scholarpedia, 2(11):2918.
- Olaf Sporns (2007) Complexity. Scholarpedia, 2(10):1623.
- James Meiss (2007) Dynamical systems. Scholarpedia, 2(2):1629.
- Roger D. Traub (2006) Fast oscillations. Scholarpedia, 1(12):1764.
- Eugene Roberts (2007) Gamma-aminobutyric acid. Scholarpedia, 2(10):3356.
- Howard Eichenbaum (2008) Memory. Scholarpedia, 3(3):1747.
- Peter Jonas and Gyorgy Buzsaki (2007) Neural inhibition. Scholarpedia, 2(9):3286.
- Rodolfo Llinas (2008) Neuron. Scholarpedia, 3(8):1490.
- Jeff Moehlis, Kresimir Josic, Eric T. Shea-Brown (2006) Periodic orbit. Scholarpedia, 1(7):1358.
- Robert E. Burke (2008) Spinal cord. Scholarpedia, 3(4):1925.
- Arkady Pikovsky and Michael Rosenblum (2007) Synchronization. Scholarpedia, 2(12):1459.
- Petilla Interneuron Nomenclature Group (2008) Petilla terminology: nomenclature of features of GABAergic interneurons of the cerebral cortex. Nat. Rev. Neurosci. 9:557-568.
The authors are grateful to Dr. Gabor Nyiri for assistance with the preparation of the figures and for useful comments on the manuscript.