# Talk:Neural inhibition

2nd Review/Reviewer #2 - 11/09 The article after the substantial revision is much improved and provides an excellent entry in Scholarpedia in its current form.

The authors may nevertheless consider the following suggestion for potential improvements:

- "GABA-B receptors couple to K+ channels ..." - This statement is true for postsynaptic receptors, but not for presynaptic ones. Presynaptic receptors are mentioned later but this definitive statement may cause some confusion.

Done

- "[GABA-C] receptors are selectively expressed in the retina." - Use "primarily" instead of "selectively"?

Changed as suggested

- The sections "Multiple forms of inhibition" and "Feed-forward, feedback and other network forms of inhibition" could be merged and restructured using subheadings to better distinguish the molecular, ionic, cellular and network mechanisms that underlie the multiple forms of inhibition.

The heading Multiple forms.. is replaced by "Inhibition can be mediated by multiple receptors", which refers better the subject discussed under this heading.

- Consider discussing the cell level effects of interneuron activity ("Interneuron diversity multiplies the computational ability of principal cells") before the network level ("Inhibitory interneurons provide spatio-temporal coordination in cortical networks"). The latter would also serve better as a link to the subsequent section on oscillations.

We would like to keep the order, although we respect the reviewer's point.

- As for the point raised by Reviewer #1, the statement is indeed misleading (even in its new form). The problem in my mind is the temporal aspect: If the two principal cells receive equally strong inputs, independent of the donwstream circuit, they will fire synchronously and activate the interneuron with a delay. Feed-back inhibition can not suppress any of the cells unless the activation is sustained or repeated within the time window of the postsynaptic effect.

Modified to explicitly illustrate that this applies to a spike train.

Second Review of Neural Inhibition - September 7, 2007

This article on inhibition is very much improved from the original version. It more adequately defines inhibition to a broader audience. I have only one concern regarding the scientific content:

The authors again propose a hypothetical model of two principal neurons that "share" a common inhibitory neuron. They then make 3 statements regarding how such a simple network might act...

The first statement is, " Suppose that two principal cells are excited by the same input but the input to principal cell A is stronger that input to principal cell B. If neuron A and B share a common inhibitory interneuron, the activation of the interneuron may prevent the spiking of neuron B."

Aside from the typo ("that" should be "than the" or similar), this statement is correct.

The third statement, "The same asymmetry can be produced if input to neuron A arrives slightly earlier than the input to B," is also correct.

However, the second statement, "the same outcome is expected if the inputs strengths to neurons A and B are equal but the neuron A-interneuron synapse is slightly stronger than neuron B-interneuron connection. The initial minor difference in the inputs results in a large difference in the output of the two neurons." is not correct. If both A and B (principal neurons) are equally driven, and the connection from the interneuron is EQUALLY inhibitory to both A and B, it DOES NOT MATTER WHO DRIVES THE INTERNEURON. THE OUTPUT OF A AND B WILL BE EQUAL. Equal drive + equal feedback/feedforward inhibition = equal output. The inhibitory drive could be intrinsic or driven from a 3rd principal neuron. It wouldn't matter to the output of A and B. I won't bring this issue up again, as I will leave it to the authors and the editors to work out whether it is an appropriate statement. Perhaps this reviewer is wrong about the circuit layout and the mathematics here, and if such is the case, the authors should consider making their argument more clear to avoid readers making a similar mistake in understanding the circuit.

My only other comment is that there are some typos throughout the essay that could be easily corrected. For instance, in the first sentence the word "meaning" should be plural. With these issues addressed, I think this essay will be a useful addition to the Scholarpedia.

## response

We would like to apologize to Reviewer 1 for prematurely returning our “revised” manuscript. We were still waiting for the second review but somehow our working copy became available for a second review through a trick of the web page.

First of all, we would like to thank both thoughtful reviewers for their precious time spent with our manuscript and for the numerous helpful suggestions. In addition to responding to the specific criticism, we have reorganized the text so that in the new version we begin with the definition and forms of inhibition, followed by its network consequences. We agree that our review is somewhat restricted to cortical networks so perhaps the entry Neural inhibition and Computation would be more appropriate, subject to approval by the Editor.

### To Reviewer 1

In the present version, we attempted to address the remaining issues brought up by the reviewer in his/her second review.

1. “Virtually every surface domain…”. Thanks for alerting us to this misleading statement. We meant to say that virtually every segment of the somatodendritc surface of the neuron… and changed the text accordingly.

2. Adding novel components to a system always results in a non-linear increase of combinatorial properties. This is a general principle in complex systems (e.g., Ward, L.M. (2002). Dynamical Cognitive Science. Cambridge, MA: MIT Press). Doubling the brain size with preserved local connectivity (nota bene: neurons crossing to larger territories are considered novel components since their axon features are different from those in the original brain) would be like placing two PCs in parallel; a linear increase in computation. However, we realize that explaining these tangentially important issues in such a short piece is not easy and beyond our abilities, therefore, we eliminated this statement.

3. “Every known excitatory pathway in the cortex has a matching family of interneurons”. We expanded this sentence to clarify the existence of specific interneurons whose axonal targets match the targets of the given excitatory pathway. Whereas this match has not been shown conclusively in every system, no exception has been observed yet.

4. Interneuron-specific interneurons. We changed the word “exclusive” to “preferential” and tempered the claim a bit.

5. “If only excitatory…”

6. “Principal cell system alone cannot carry out any useful computation”. We have tempered this statement by explaining that without inhibition, excitation would recruit more excitation in real brains.

7. “Balance … oscillations”. We have eliminated reference to “balance” since it is indeed a misleading term, which depends on the time-scale in question, and separated the issue from oscillations.

8. Fig. 4. (now Fig. 1). We have carefully considered the suggestion of the referee to incorporate information about the spatial dependence of shunting inhibition. However, we think that this would make the figure even more complicated. If the referee would absolutely insist, we could insert a separate figure to illustrate this point.

9. Previous Fig. 1 was redone as requested.

### To Reviewer 2

We thank the reviewer to point out our oversight in providing a prompt definition of inhibition. In addition to providing a definition, we restructured the manuscript by discussion elementary processes first and discussing various forms of spike suppression, followed by the role of interneurons in cortical network operations. We agree that, ideally, several issues could be discussed in more details but we tried to keep our entry concise with the hope that several keywords, related but not central to our piece, will be discussed by other entries.

1. We agree with the reviewer that homeostasis have various connotations and avoided the term altogether in the revised version.

2. “Local computation demand”. What we meant here is that in “simple” systems, wiring and computation is local (e.g., cerebellum), and these parallel modular systems typically use a limited number of neuron constituents. We have re-written this paragraph to better illustrate the meaning.

3. Please see answer 2. to Reviewer 1. We chose to eliminate this hypothesis rather than spend several sentences for explanation.

4. We have added an example and references to illustrate this claim.

5. This sentence is moderated. In parenthesis, we note though that artificial neuronal networks without inhibition would not work in the “real (physical) world”, as eg. John Hopfield emphasizes this fact.

6. The term “balance” has been eliminated in most places and the sentences have been reorganized.

7. We have carefully considered the suggestion of the referee to incorporate information about the spatial dependence of shunting inhibition. However, we think that this would make the figure even more complicated. If the referee would absolutely insist, we could insert a separate figure to illustrate this point.

Any further comments and improvement will be most appreciated.

## Old Reviews

NEW REVIEW: The article of P. Jonas and G. Buzski is a very exciting read and provides interesting insights into advanced ideas about the function of interneurons in neural circuits of the cortex. However, as a Scholarpedia, an encyclopedia article, it fails the aim to define and explain the term "neural inhibition" to a broader audience.

First of all, the reader expects a definition and an introduction to basic concepts of neuronal inhibition: What is inhibition and what are its mechanisms in the nervous system? What cellular and synaptic elements are involved? What are the molecular substrates, transmitters and receptors mediating inhibition? Although, some of these aspects are briefly touched upon in the last section focusing on shunting inhibition, the authors clearly assume a very high level of background knowledge about neural inhibition. A pregnant example of this problem is the fact that they fail to explicitly state that interneurons' are GABAergic inhibitory neurons, the cellular substrate of inhibition in the cortex, but use the term from the very beginning.

The article in its present form is overwhelmed by aspects of network dynamics, cell assembly formation and the possible computational roles of interneurons. The question emerges whether it wouldn't be more adequate to treat these in a separate article (may be under the title Cell assemblies'?). Similarly, wouldn't the readers be better served by a hyperlink to a separate article on neural network/population oscillations and the clock hypothesis (instead of the current link to the very distantly related Periodic Orbit' article)?

Finally, while much attention is dedicated to shunting inhibition mediated by fast postsynaptic GABAA receptors, other forms are neglected or not mentioned at all (e.g. presynaptic inhibition). The authors should aim for an unbiased account of the mechanisms of neuronal inhibition.

Specific points:

1) The relation of the homeostatic dynamics/balance' of networks and pattern completion and separation is emphasized but only vaguely defined. Are they necessarily related except for the fact that both are underlied by inhibitory mechanisms? Furthermore, the term homeostatic balance' triggers strong associations to classic hypotheses about the disturbed balance of excitation and inhibition in seizure/epileptic disorder. The relationship of neural inhibition and seizure disorders is a theme that many readers would expect to find here.

2) What is local computational demand' and what would be the implicit `no-local computational demand' specific to the cortex? (beginning of section "Interneurons multiply the computational ?")

3) Scaling numbers of cells versus number of types: is there any evidence for the proposed linear and non-linear increase in computational power? The argumentation is very simplistic and misleading. In fact, the subsequent text suggest that different types of interneuron may switch pyramidal cells between different modes of operations - possibly depending on the behavior-related modulatory state of the network. Whereas the total number of cells defines the computational power of the network (plausibly, in a non-linear/supra-linear manner), the number of neuron/interneuron types could define the number of possible states of a complex network (again in non-linear manner), introducing a new dimension in the computational capacity of the network. This hypothesis is not explained properly in the text.

4) The hypothesis that interneurons can "effectively and specifically control the kinetic properties of their target domains" is appealing, but direct experimental evidence supporting this is relatively scarce. It would be helpful if the authors could provide some robust examples with references.

5) "principal cell system alone cannot carry out any useful computation" Why? It is a rather bold statement, without any explanation. Artificial neuronal networks capturing the basic features of excitatory networks without inhibitory elements do perform some useful computation.

6) "Balance of opposing forces, such as excitation and inhibition, often gives rise to rhythmic behavior." In theory balance could be static without oscillations - it is not clear from the text what is the precise mechanism that leads to the emergence of oscillations from the balance of excitation and inhibition. Similarly, for an average reader it remains unclear how interneurons can provide "rhythm-based timing to the principal cells at multiple time scales".

7) While the text gives a good explanation, fig. 4 doesn't illustrate differences in the spatial or temporal integration rules but simply the difference in the initial conditions after the arrival of a hyperpolarizing, shunting or depolarizing inhibitory input. The figure is also too complex. It would be more instructive to show the coincidence arrival of an EPSP and a hyperpolarizing or a shunting/depolarizing IPSP from synapses located on the same dendrite (inhibition in both case) or two different dendrites (inhibition vs. facilitation).

Second Review of "Neuronal Inhibition" by Peter Jonas and Gyorgi Buzsaki This review article on neuronal inhibition by Drs. Jonas and Buzsaki aims to describe the role and importance of inhibitory mechanisms in neuronal network function. This topic is fundamental to neuroscience, and the authors are well chosen, being leading researchers in the field. After the initial modifications, I believe the following issues still need to be addressed:

1: “Virtually every surface domain of cortical principal cells is under the specific control of a unique interneuron class...”

Comment: It is difficult to think of any particular cellular domain, other than the axon, that is "under the specific control of a unique interneuron class". Dendrites, spines, and somata are innervated by multiple interneuron subtypes, even within the same class of postsynaptic neuron.

2: “Adding more interneurons of the same type linearly increases the network’s combinatorial properties (e.g., increasing the network size of soma-targeting basket cells in species with larger brains). Adding novel interneuron types to the old network, even if in very small numbers, offers a non-linear expansion of qualitatively different possibilities.”

Comment: I'm not sure how the authors derive this mathematical relationship. Wouldn't doubling the number of neurons in a network (doubling brain size) more than double (vastly) the "networks combinatorial possibilities"? And even if the authors math were correct, doubling brain size would also double the number of excitatory cells, glia, etc., so a statement that implies a special relationship to interneurons is misleading.

3: "Every known excitatory pathway in the cortex has a matching family of interneurons."

Comment: This statement has the potential to misinform the reader as to what is known about innervation patterns of inhibitory neurons as well as their relationship to specific “excitatory pathways”.

4: "Interneuron-specific interneurons. These neurons have the distinguishing characteristics that their axons completely avoid principal cells and contact exclusively other interneurons. "

Comment: How strong is the evidence for this "class" outside of the hippocampus? Perhaps the authors could say that this and other classes of interneuron "preferentially" innervated their targets, rather than imply that they exclusively innervate those targets. Additionally, it might be good to point out that these classes of neuron are not found in the same proportions in every brain area.

5: “If only excitatory cells were present in the brain, neurons could not create form or order or secure some autonomy for themselves. Interneurons provide autonomy and independence to neighboring principal cells and, at the same time, offer useful temporal coordination…. Separation of inputs in a network with only excitatory connections is not possible.”

Comment: The phrase "not possible" is a bit strong here, and is not necessary to convey the importance of inhibition in networks.

6. Appropriate citations should be provided for the statement, “If GABAergic synapses are activated repetitively, Cl- can accumulate in the postsynaptic cell, and the depolarizing phase of shunting inhibition will become more accentuated.”

7: Figure 1: Given the general audience this review is targeted to, it would be helpful if the spike trains were labeled to identify to which cell [assembly] they belong.

ORIGINAL REVIEW:

This review article on neuronal inhibition by Drs. Jonas and Buzsaki aims to describe the role and importance of inhibitory mechanisms in neuronal network function. This topic is fundamental to neuroscience, and the authors are well chosen, being leading researchers in the field. However, I have several general and specific criticisms of the work as it is currently presented, especially given that it is targeted to a general audience.

First, the article is focused primarily on fast synaptic inhibition mediated by GABA and/or glycine receptors. Given the title, “Neuronal Inhibition”, this article may be overly focused. While the authors briefly discuss GABA-B receptors, they make no mention of inhibition generated by a myriad of other transmitters, including glutamate, working at G-protein coupled receptors. One solution would be to change the title of this work to “Fast Synaptic Inhibition”, and to create a very general, overarching entry for “Neuronal Inhibition” that would link to this article and to others that describe in more detail the various inhibitory mechanisms that exist.

Second, the authors have resorted to unnecessary hyperbole to emphasize the importance of inhibition in neuronal systems. In particular, I felt the following statements from the article were exaggerated, misleading, or incorrect:

Example 1: “Every surface domain of cortical principal cells is under the specific control of a unique interneuron class.” Comment: This statement is incorrect. Further, it is contradicted later in the article with the statement, "Several additional subclasses in the second group seek out two or more overlapping or non-overlapping dendritic regions, and yet other subclasses innervate the somata and nearby dendrites with similar probability."

Example 2: “Adding more interneurons of the same type linearly increases the network’s combinatorial properties.” Comment: This claim is exaggerated. If, for instance, enough fast-spiking basket cells were added to the system, the networks combinatorial properties would actually become reduced as inhibition begins to limit any excitatory spread through the system.

Example 3: “The extensive computational capacity of a single cortical principal cell is seldom utilized at once.” Comment: While I understand the intent of this statement, depending on how one defines computation, it could be argued that a lack of input to specific dendrites will also contribute to the instantaneous “computation” in a neuron (ie, lack of input also influences the output of a neuron).

Example 4: "Interneurons of the second family target specific dendritic domains of principal cells. Every known excitatory pathway in the cortex has a matching family of interneurons." Comment: This second “family” is not defined or identified. Secondly, the statements have the potential to misinform the reader as to what is known about innervation patterns of inhibitory neurons as well as their relationship to specific “excitatory pathways”. The second statement is overreaching to say the least.

Example 5: “The third distinct family of interneurons has the distinguishing characteristics that their axons completely avoid principal cells and contact exclusively other interneurons.” Comment: Again, this third “family” of interneurons is not adequately defined, nor is the author’s classification system universally “recognized”. Perhaps the authors could add a brief discussion about the diversity of inhibitory neurons, including a description of the several methods used to classify them, with special emphasis on the difficulties of classification and why there is not yet consensus on this issue.

Example 6: “Because the different domains of principal cells have different functional dynamics, interneurons innervating those specific domains adapted their kinetic properties to match their targets.” Comment: This is a difficult statement to defend. Presumably inhibitory mechanisms “evolved” in concert with excitatory processes.

Example 7: “If neuron A and B share a common inhibitory interneuron, the gain in neuron A results in a suppression of neuron B’s activity.” Comment: The word "share" implies that both cell A and cell B synapse onto the same inhibitory neuron, which provides equal inhibition to cells A and B. If this is indeed the scheme the authors wish to convey, then the statements that follow regarding the effects of input on cellular output are not correct for every situation. The statement, "the gain in neuron A results in a suppression of neuron B’s activity" is misleading because an increase of the gain of A will also increase feedback suppression of A, just as it also suppresses the out put of B. Further, the statement, "The same outcome occurs if the inputs strengths to neurons A and B are equal but the synapse between neuron A and the interneuron is slightly stronger [than] the neuron B-interneuron connection", is also misleading: what would actually result is identical firing patterns from both neurons.

Example 8: “The high abundance, the extensive axonal arborization, and the fast signaling properties of parvalbumin-expressing basket cells explain the primary role of this interneuron class in the generation of network oscillations.” Comment: It might be preferable if this statement were more descriptive, rather than implying a “cause and effect”. Here would be a good place for a more in-depth discussion that cites the evidence for fast-spiking neuron participation in network oscillations.

Example 9: Regarding long-distance inhibitory connections, the authors state, “They provide the necessary fast conduit for synchronizing distantly operating oscillators and allow for coherent timing of a large number of neurons that are not connected directly with each other.” Comment: Again, this statement may be too strong. How clear is it that there are long-distance inhibitory projections reciprocally connecting all synchronized oscillating networks? Further, which data demonstrate that those projections are “necessary” for coordinated activity? Isn’t it possible that two oscillating networks could be coordinated simply by excitatory projections that innervate local inhibitory neurons in each network?

Example 10: “If only excitatory cells were present in the brain, neurons could not create form or order or secure some autonomy for themselves. Interneurons provide autonomy and independence to neighboring principal cells and, at the same time, offer useful temporal coordination…. Separation of inputs in a network with only excitatory connections is not possible.”

Comment: This reviewer can imagine a network of excitatory neurons that have finely tuned intrinsic non-linear damping and boosting conductances, as well as specific distributions of short term synaptic dynamics, that would allow two separate inputs to generate independent output. I admit that this criticism may seem overly “nit-picky”, and is less significant than the others above, but as this article is targeted to a general audience, I think it would be best to be as accurate as possible. Further, I don’t think the authors need to resort to such a strong statement to convey the very important role that inhibitory units play in network function.

Appropriate citations should be provided for the statement, “If GABAergic synapses are activated repetitively, Cl- can accumulate in the postsynaptic cell, and the depolarizing phase of shunting inhibition will become more accentuated.”

In Figure 1: Given the general audience this review is targeted to, it would be helpful if the spike trains were labeled to identify to which cell they belong.

Finally, it would be nice if the authors provided a concluding paragraph summing up the main points of the article.