In Sync—Orchestrating Perfect Harmony in Neuronal Networks
1 February 2006. Flocks of birds do it, schools of fish do it, so why can’t neuronal networks do it? Well, they can, and they do—move in perfect synchrony, that is. But why is that, exactly? Some answers come from back-to-back papers in the January 5 Neuron. The two papers report different, though complementary, mechanisms for regulating the rhythmic firing of GABAergic neurons—ones that use the amino acid derivative γ-aminobutyric acid (GABA) as a neurotransmitter. Peter Jonas and colleagues at the University of Freiburg, Germany, describe how a process called “shunting inhibition” keeps these GABAergic neurons firing together, while Alberto Bacci and John Huguenard at Stanford University School of Medicine, California, reveal that a self-regulatory mechanism ensures that the neurons fire with precise timing. Together, the papers help explain how the so-called γ oscillations, a feature of GABAergic interneurons, are maintained. These oscillations, of between 30 and 90 Hertz, are thought to be indispensable for basic cognitive tasks, such as feature recognition and processing of content- and context-sensitive information. GABA has also been implicated in the pathophysiology of schizophrenia in several different anatomical systems (for a review, see Wassef et al., 2003; Lewis et al., 2005).
The main function of GABAergic interneurons is to modulate the action of excitatory neurons, or principal neurons such as the pyramidal cells, which excite other neurons to generate action potentials. GABAergic neurons are perfectly suited to this role because they are, for the most part, inhibitory (though a prominent exception was reported recently in Science; see below) and they sit in wait between the excitatory cells and their targets (hence, the term interneurons). However, scientists have been puzzled by how GABAergic networks generate synchronous oscillations. In theory, this can happen if a large enough stimulus impacts the network, because it could rapidly propagate an inhibitory pulse throughout the neurons. This would synchronize the network because all the neurons would then recover at the same time. But as Jonas and colleagues point out, synchrony tends to break down if the stimuli from the excitatory cells are not themselves in synch, as happens in vivo. And while a single, very large excitatory stimulus might counteract this tendency sufficiently to set the pace of the network, there is no experimental data to suggest that this is what happens in real life.
Enter shunting inhibition. This is very different from the classic “hyperpolarization” that dampens neuronal activity. When neurons are hyperpolarized, the balance of ions across the cell membrane is shifted so that it becomes harder to depolarize the membrane and get the neuron to fire. In contrast, shunting inhibition, caused by the opening of chloride channels, results in a hint of depolarization because the equilibrium potential for the chloride channel is only slightly more positive than the normal resting potential of the neuron. But more importantly, the open chloride channels clamp, or shunt, the membrane potential at this new level and prevent any other stimuli from depolarizing the cell further.
A few years ago, GABA receptors were shown to mediate depolarizing events (see Stein and Nicoll 2003), while shunting inhibition evoked by activation of the GABAA receptor, which is coupled to chloride channels, has been evoked to explain the rhythmic firing of neurons in the thalamus (see Bazhenov et al., 1999). In this context, Jonas's team tested if shunting inhibition may operate in hippocampal networks. First author Imre Vida and colleagues measured membrane potentials in basket cells, GABAergic interneurons that fire rhythmic bursts in the γ frequency range. The authors found that the GABA-induced synaptic reversal potential was slightly more positive than the resting membrane potential, showing that these neurons are indeed slightly depolarized, or shunted, by GABA. To check if this affects the response of basket cell networks to excitatory stimuli, Vida and colleagues measured mini-networks comprising around 200 neurons, keeping them artificially clamped at this slightly depolarized membrane potential. The researchers found that this simulated shunting inhibition kept the coherence of the mini-network at its highest levels. In contrast, when the network was artificially hyperpolarized, then coherency was almost five times lower. Furthermore, the shunting inhibition model remained coherent even if the excitatory stimuli were fairly low, thus obviating the need for a large excitatory drive to keep the network in synchrony.
How does this depolarizing inhibition keep the network synchronized? Vida and colleagues discovered that it homogenizes the firing rates of the neurons. It makes fast neurons fire more slowly, and slow neurons fire more quickly. Furthermore, shunting inhibition helps the network stay synchronized even when the frequency of the excitatory stimuli, such as those from pyramidal neurons, is not necessarily constant. In other words, this type of inhibition makes the GABAergic network extremely robust.
But these findings do not completely rule out a role for hyperpolarizing inhibition in GABAergic networks. In fact, in the second paper, Bacci and Huguenard report that “autaptic” or self-innervation in GABAergic neurons is mediated by hyperpolarization, and that this helps the neurons keep precise timing.
These authors were interested in comparing how fast-spiking (FS) interneurons and excitatory pyramidal neurons react when neocortical slices were stimulated with an electrical current. They found that, though the pyramidal neurons initially fired together, by the time they fired for the fourth time, they were out of phase. The fast-spiking interneurons, on the other hand, were still in phase after four spikes and even after firing 12 times there was little “jitter” as all the neurons spiked about the same time.
Because pyramidal neurons are not self-innervated, Bacci and Huguenard then tested if autaptic transmission in GABAergic cells could explain the dramatic differences in spike-timing precision between the two types of cells. When they blocked all the interneuron GABAA receptors with the antagonist gabazine, they found that the precision of spike-timing was dramatically weakened—after just four spikes, the interneurons had begun to fire out of phase. But because gabazine blocks all GABAA receptors, this result did not prove that loss of autaptic transmission, per se, was what abolished the timing. However, when Bacci used the dynamic clamp—an electrical device that allows researchers to challenge neurons with different voltages and currents—to restore only autaptic transmission to these gabazine-treated cells, then spike-timing precision returned.
Having determined that autaptic transmission was essential for firing precision, Bacci next asked whether the effect is mediated by shunting inhibition or hyperpolarization. He found that electrically inducing hyperpolarization improved the precision, whereas when shunting was induced, precision actually decreased. “These data indicate that the autaptic effect on FS cell spike timing is due to the hyperpolarizing component of GABAergic autaptic transmission rather than to its shunting mechanism,” write the authors.
How hyperpolarization and shunting inhibition are coordinated within the same network is unclear, but it may be related to what part of the cell is affected by each. Shunting may be going on primarily in the dendrites, whereas autaptic transmission was shown to be limited to the soma, or cell body, and the nearby dendritic tree (see Tamas et al., 1997). Whatever the explanation, “the new insights into the interneuronal rhythms provided by the Bacci and Huguenard and Vida et al. studies lend support to the idea that network oscillations are an integral and important part of cortical information processing,” write Edward Mann and Ole Paulsen at the University of Oxford, England, in an accompanying Neuron preview. “Although oscillations are hard to avoid in feedback-coupled networks, such as the cortex, evolution has apparently developed mechanisms that further enhance rather than suppress their oscillatory behavior,” they add.
Mann and Paulsen also predict that GABAergic neurons may have even more secrets to divulge, which seems prescient given that barely a week later, Gábor Támas and colleagues at the University of Szeged, Hungary, demonstrated that GABAergic neurons are not limited to an inhibitory role—they excite, too. In the January 13 Science, joint first authors János Szabadics and Csaba Varga and their coworkers reported that axo-axonic GABAergic cells, which innervate axons in the cerebral cortex, can depolarize those axons and initiate synaptic events. This work lends support to the earlier suggestion that GABA can have an excitatory role in the cortex under certain circumstances (see Gulledge and Stuart, 2003). For further information on this surprising finding, see the detailed SRF commentary by Guillermo Gonzalez-Burgos, from the University of Pittsburgh.—Tom Fagan.
Vida I, Bartos M, Jonas P. Shunting inhibition improves robustness of gamma oscillations in hippocampal interneuron networks by homogenizing firing rates. Neuron. January 5, 2006;49:107-117. Abstract
Bacci A, Huguenard JR. Enhancement of spike-timing precision by autaptic transmission in neocortical inhibitory interneurons. Neuron. January 5, 2006;49:119-130. Abstract
Mann EO, Paulsen O. Keeping inhibition timely. Neuron. January 5, 2006;49:8-9. Abstract
Szabadics J, Varga C, Molnar G, Olah S, Barzo P, Tamas G. Excitatory effect of GABAergic axo-axonic cells in cortical circuits. Science. January 13, 2006;311:233-235. Abstract
Comments on News and Primary Papers
Primary Papers: Excitatory effect of GABAergic axo-axonic cells in cortical microcircuits.Comment by: Guillermo Gonzalez-Burgos
Submitted 18 January 2006
Posted 18 January 2006
The paper from Szabadics and colleagues contests two dogmas of GABA neurotransmission: 1) that in the adult brain, the effects of GABA synapses are inhibitory; and 2) that among GABAergic interneuron subclasses, axo-axonic cells (AACs) are probably the most powerfully inhibitory.
Like most GABAergic interneuron subclasses, AACs utilize GABA as their main neurotransmitter and the effects of GABA at AAC synapses are mediated by GABAA receptors, which after binding GABA open a chloride channel. Usually, the opening of GABAA receptor chloride channels produces hyperpolarization and thus inhibition. AACs are unique because they make synapses onto the initial segment of the pyramidal cell axon, close to the site where action potentials are initiated. Because action potential initiation requires depolarization, it was thought that AAC-mediated hyperpolarization was strategically placed: The closer an inhibitory synapse is to the site where action potentials are initiated, the more powerful the inhibition. The claim that AACs were more powerful than other interneurons to inhibit pyramidal cell firing was never verified experimentally. However, very few scientists would have doubted that AAC synapses have an inhibitory effect (Howard et al., 2005).
The surprising findings of Szabadics and colleagues show that after releasing GABA, AACs produce depolarization, instead of hyperpolarization of the pyramidal cell membrane. Moreover, the results show that AAC-mediated depolarization is so powerful that in more than 50 percent of the cases in which a single AAC is stimulated, it depolarizes the postsynaptic pyramidal neuron to the point of making it fire an action potential. This suggests that single AACs can produce much more excitation than can single pyramidal neurons, because single pyramidal neurons rarely, if ever, can make a postsynaptic neuron fire spikes.
The reason AACs are so powerfully excitatory is the same reason they were thought to be strongly inhibitory: They make synapses very close to the action potential initiation site. However, instead of hyperpolarization, the AAC synapses produce depolarization, because (as Szabadics and colleagues show) the chloride concentration gradient at AAC synapses determines a depolarizing, rather than hyperpolarizing effect of GABA.
Like AACs, synapses made by pyramidal cells (the prototypical excitatory neurons of the cerebral cortex) have a depolarizing effect. Unlike AACs, however, pyramidal cell synapses are electrically "far" from the site of action potential initiation. This happens because when depolarizing synaptic signals travel within neurons, they typically attenuate in size. Thus, the longer the distance they need to travel to reach the action potential initiation site, the weaker their electrical effect is. In the case of the AAC synapses, the distance is very short, and thus, the effect of these synapses (now shown to be excitatory) is strong.
Depolarizing effects of GABA have been previously found in very immature neurons, where they have been associated with maturational plasticity of GABA synapses (Kandler, 2004). However, until now in the mature brain, the chloride concentration gradient was thought to determine membrane hyperpolarization everywhere in the cell. Szabadics and colleagues provide data consistent with a potential mechanism that can explain the differential chloride gradient: the differential localization of the chloride transporter KCC2, which is almost absent close to the AAC synapses. However, whether or not this mechanism completely accounts for the differential chloride gradient is somewhat irrelevant, because the physiological data are strong on their own.
I think the reason these findings were not reported before by other groups is mainly because, from a number of different points of view, the experiments performed by Szabadics and colleagues are technically challenging. Discovery of the excitatory effect of AACs required first of all the study of small circuits of neurons in living brain slices in which at least one of the cells belonged to the AAC population and had intact connections with nearby pyramidal cells. This is already challenging, because AACs are not easy to identify prior to recording and are not the most abundant class of GABAergic interneuron. In addition to the low probability of finding AACs, in order to study cells without perturbation of the normal chloride gradients, the researchers had to use perforated patch clamp recording, which is considerably more difficult than standard recording techniques.
A very important aspect of this study is that, although the main findings were observed in cortical tissue obtained from the brain of young rats, the authors report essentially identical results from studying adult human cortical neurons. I think this is very important, because it shows that at least in this respect, the rat cortex is an excellent model for further studies of the role of AACs in cortical microcircuits.
GABA transmission mediated by AACs is thought to be altered in the brain of subjects with schizophrenia. The present data from Szabadics and colleagues are thus highly relevant for current models on the neurobiology of schizophrenia (Lewis et al., 2005). If the findings of Szabadics and colleagues are replicated by other groups, the models that propose a role for AACs in the biological basis of schizophrenia will have to be substantially modified. Being able to use the rat (or mouse) brain as a valid animal model to study this role is tremendously important.
Clearly, before jumping to quick conclusions about the therapeutic effects of pharmacological manipulations of AAC cell-mediated excitation, we need to learn more about the role of AACs in the normal brain. In particular, although individual AACs may have strong excitatory effects, it is not clear that AACs as a neuronal population could produce more excitation than the network of pyramidal cells. Indeed, in the cortex only about 20-30 percent of the neurons are GABAergic interneurons; the remainder are pyramidal cells. Moreover, AACs are a subpopulation within the subpopulation of parvalbumin-containing interneurons (parvalbumin is a calcium-binding protein whose expression by these neurons seems to be altered in schizophrenia). Thus, whereas each pyramidal cell receives synaptic input from hundreds, if not thousands, of other pyramidal cells, it receives input from only 2-4 AACs. This suggests that the main excitatory drive for pyramidal cells comes from other pyramidal neurons and that AAC-mediated excitation may have more selective and subtle roles. Consistent with this idea, recent experiments in living rats suggest that AACs do not discharge at the same time as pyramidal cells, and therefore that AACs are activated in a very precise and selective manner (Klausberger et al., 2003; Somogyi and Klausberger, 2005).
View all comments by Guillermo Gonzalez-BurgosComment by: Kevin Spencer (Disclosure)
Submitted 9 February 2006
Posted 9 February 2006
I recommend the Primary Papers
Comments on Related News
Related News: Gamma Band Plays a Sour Note in Entorhinal Cortex of Schizophrenia ModelsComment by: Bita Moghaddam, SRF Advisor
Submitted 3 April 2006
Posted 3 April 2006
Cortical dysfunction in schizophrenia has been attributed to both inhibitory GABA and excitatory glutamate neurotransmission. Abnormalities in cortical GABA neurons have been observed primarily in the subset of GABA interneurons that contain the calcium-binding protein parvalbumin (PV). The glutamatergic dysfunction is suspected primarily because reducing glutamate neurotransmission at the NMDA receptors produces behavioral deficits that resemble symptoms of schizophrenia. These two mechanisms have been generally treated as separate conjectures when conceptualizing theories of schizophrenia. The paper by Cunningham et al. demonstrates that, in fact, disruptions in PV positive cortical GABA neurons and blockade of NMDA receptors produce similar disruptions to the function of cortical networks.
The authors used lysophosphatidic acid 1 receptor (LPA-1)-deficient mice which, they argue, are a relevant model of schizophrenia because these animals display sensorimotor gating deficits, a critical feature of schizophrenia. They demonstrate that, similar to schizophrenia, the number of PV positive GABA neurons is significantly reduced in LPA-1-deficient mice. Furthermore, the γ frequency network oscillation disruptions they observe in these animals are similar to those seen in wild-type mice treated with the NMDA antagonist ketamine. (γ oscillations have been associated with sensory processing and deficits in γ rhythm generation have been reported in patients with schizophrenia during performance of sensory processing tasks.) The disruptive effect of ketamine on γ oscillations was mediated by a decrease in the output of fast-spiking GABA interneurons causing a disinhibition (i.e., increased firing) of glutamate neurons. These findings are significant because they suggest that cortical NMDA hypofunction may cause the reported GABA interneuron deficits in schizophrenia.
View all comments by Bita Moghaddam
Related News: Gamma Band Plays a Sour Note in Entorhinal Cortex of Schizophrenia Models
Comment by: Patricio O'Donnell, SRF Advisor
Submitted 7 April 2006
Posted 7 April 2006
Animal models of schizophrenia and other psychiatric disorders are receiving increasing interest, as they provide useful tools to test possible pathophysiological scenarios. Some models have been tested with a wide array of approaches and many others continue to develop. If one focuses on possible cortical alterations, a critical issue emerging from many different lines of research using several different models is the apparent contradiction between the hypo-NMDA concept and the data suggesting a loss of cortical interneurons. Is there a hypo- or a hyperactive cortex?
This conundrum has been present since earlier days in the postmortem and clinical research literature, but with the advent of more refined animal models, it may be time to provide a possible way in which these discrepant sets of data can be reconciled. Whether this was the authors’ intention or not, the article by Cunningham and colleagues is an excellent step in that direction. This study used mice deficient in lysophosphatidic acid 1 receptor, a manipulation that reduced the GABA and parvalbumin-containing interneuron population by about 40 percent and disrupted γ (rapid) oscillations in the entorhinal cortex. A key element in this study was the finding that a similar alteration in rapid cortical oscillations was observed with the noncompeting NMDA antagonist ketamine. There is a large body of evidence indicating that interneurons (in particular, the fast-spiking type that include parvalbumin-positive neurons) are critical for synchronization of fast cortical oscillatory activity. As fast oscillations can be envisioned as phenomena with deep impact on cognitive functions, these findings may have bearing on possible pathophysiological scenarios underlying cognitive deficits in schizophrenia. This article does provide a strong indication that antagonism of NMDA receptors may selectively target cortical interneurons. This is in agreement with the work of Bita Moghaddam, who has shown that noncompeting NMDA antagonists can indeed increase pyramidal cell firing and glutamate levels in the prefrontal cortex. Thus, it is conceivable that psychotomimetic agents such as PCP or ketamine exert their cognitive effects by impairing interneuronal activity, hampering the fine-tuning of pyramidal cell firing that is expressed as fast cortical oscillations.
View all comments by Patricio O'Donnell
Related News: Asynchrony and the Brain—Gamma Deficits Linked to Poor Cognitive Control
Comment by: Richard Deth
Submitted 14 December 2006
Posted 15 December 2006
Schizophrenia is associated with dopaminergic dysfunction, impaired gamma synchronization and impaired methylation. It is therefore of interest that the D4 dopamine receptor is involved in gamma synchronization (Demiralp et al., 2006) and that the D4 dopamine receptor uniquely carries out methylation of membrane phospholipids (Sharma et al., 1999). A reasonable and unifying hypothesis would be that schizophrenia results from a failure of methylation to adequately support dopamine-stimulated phospholipid methylation, leading to impaired gamma synchronization. Synchronization in response to dopamine can provide a molecular mechanism for attention, as information in participating neural networks is able to bind together to create cognitive experience involving multiple brain regions.
View all comments by Richard Deth
Related News: Asynchrony and the Brain—Gamma Deficits Linked to Poor Cognitive Control
Comment by: Fred Sabb
Submitted 12 January 2007
Posted 12 January 2007
I recommend the Primary Papers
Cho and colleagues find patients with schizophrenia showed a reduction in induced gamma band activity in the dorsolateral prefrontal cortex compared to healthy control subjects during a behavioral task that is known to challenge cognitive control processes. Importantly, the induced gamma band activity was correlated with better performance in healthy subjects, and negatively correlated with higher disorganization symptoms in patients with schizophrenia. These findings help explain previous post-mortem evidence of disruptions in thalamofrontocortical circuits in these patients.
These findings tie together several different previously identified phenotypes into a unifying story. The ability to link phenotypes across translational research domains is paramount to understanding complex neuropsychiatric diseases like schizophrenia. Cho and colleagues provide an excellent example for connecting evidence from symptom rating scales with behavioral, neural systems and neurophysiological data. Although not specifically addressed by the authors, these data may have important implications for understanding the neural basis of thought disorder as well. Hopefully, these findings will provide a frame-work for examining more informed and specific phenotypes relevant to schizophrenia.
View all comments by Fred Sabb