Cholecystokinin-expressing (CCK+) basket cells are key controllers of theta-gamma coupled rhythms in the hippocampus

  1. Alexandra P Chatzikalymniou
  2. Spandan Sengupta
  3. Jeremie Lefebvre
  4. Frances K Skinner

  • Curated by eLife

    eLife logo

    eLife assessment

    This paper reports a useful set of results that uses a reduced network model based on a previously published large-scale network model to explain the generation of theta-gamma rhythms in the hippocampus. Combining the detailed and reduced models and comparing their results is a powerful approach. However, the evidence for the main claim that CCK+ basket cells play a key role in theta-gamma coupling in the hippocampus is currently incomplete.

Reviewed by

Read the full article See related articles

Discuss this preprint

Start a discussion What are Sciety discussions?

Listed in

Log in to save this article

Abstract

Various inhibitory cell types underlie cognitively important brain rhythms and their couplings. Precisely how this is manifest is unclear. Complex network interactions make an understanding of inhibitory cell contributions extremely difficult and using experiments alone is insufficient. Using detailed biophysical models, we obtain hypotheses of how theta and gamma rhythms in the hippocampus are generated and coupled. We find critical contributions by parvalbumin-expressing (PV+) basket cells (BCs), cholecystokinin-expressing (CCK+) BCs and bistratified cells. Based on this, we develop and explore a population rate model and predict that CCK+BCs exert more control relative to PV+BCs for theta-gamma coupling, and that theta frequencies are more strongly affected by PV+BC to CCK+BC coupling relative to CCK+BC to PV+BC. As specific inhibitory cell types can be targeted during behaviour, it is possible to test these predictions. Our work shows that combining models at different scales creates new insights that otherwise would not be revealed.

Article activity feed

  1. eLife

    eLife assessment

    This paper reports a useful set of results that uses a reduced network model based on a previously published large-scale network model to explain the generation of theta-gamma rhythms in the hippocampus. Combining the detailed and reduced models and comparing their results is a powerful approach. However, the evidence for the main claim that CCK+ basket cells play a key role in theta-gamma coupling in the hippocampus is currently incomplete.

  2. eLife

    Reviewer #1 (Public Review):

    This paper investigates potential mechanisms underlying the generation of hippocampal theta and gamma rhythms using a combination of several modeling approaches. The authors perform new simulation experiments on the existing large-scale biophysical network model previously published by Bezaire et al. Guided by their analysis of this detailed model, they also develop a strongly reduced, rate-based network model, which allows them to run a much larger number of simulations and systematically explore the effects of varying several key parameters. The combined results from these two in silico approaches allow them to predict which cell types and connections in the hippocampus might be involved in the generation and coupling of theta and gamma oscillations.

    In my view, several aspects of the general methodology are exemplary. In the current work as well as several earlier papers, the authors are re-using a large-scale network model that was originally developed in a different laboratory (Bezaire et al., 2016) and that still represents the state-of-the-art in detailed hippocampal modeling. Such model reuse is quite rare in computational neuroscience, which is rather unfortunate given the amount of time and effort required to build and share such a complex model. Very often, and also, in this case, the original publication that describes a detailed model provides only limited validation and analysis of model behavior, and the re-use of the same model in later studies represents a great opportunity to further examine and validate the model.

    Combining detailed and simplified models can also be a powerful approach, especially when the correspondence between the two is carefully established. Matching results from the two models, in this case, allow strong arguments about key mechanisms of biological phenomena, where the simplified model allows the identification and characterization of necessary and sufficient components, while the detailed model can firmly anchor the models and their predictions to experimental data.

    On the other hand, I have several major concerns about the implementation of these approaches and the interpretation of the results in the current study. First of all, the detailed model of Bezaire et al. is considered strictly equivalent, in all of its relevant details, to biological reality, and no attempt is made to verify or even discuss the validity of this assumption, even when particular details of the model are apparently critical for the results presented. I see this as a fundamental limitation of the current work - the fact that the Bezaire et al. model is the best one we have at the moment does not automatically make it correct in all its details, and features of the model that are essential for the new results certainly deserve careful scrutiny (preferably via detailed comparison with experimental data).

    An important case in point is the strength of the interactions between specific neuronal populations. This is represented by different quantities in the detailed and simplified model, but the starting point is always the synaptic weight (conductance) values given by Bezaire et al. (2016), also listed in Tables 2 and 3 of the current manuscript. Looking at these parameters, one can identify a handful of connections whose conductance values are much higher than those of the other connections, and also more than an order of magnitude higher (50-100 nS) than commonly estimated values for cortical synapses (normally less than about 5 nS, except for a few very special types of synapse such as the hippocampal mossy fibers). Not surprisingly, several of these connections (such as the pyramidal cell to pyramidal cell connections, and the CCK+BC to PV+BC connections) were found to be critical for the generation and control of theta and gamma oscillations in the model. Given their importance for the conclusions of the paper, it would be essential to double-check the validity of these parameter values. In this context, it is worth noting that, unlike the anatomical parameters (cell numbers and connectivity) that had been carefully calculated and discussed in Bezaire and Soltesz (2013), biophysical parameters (the densities of neuronal membrane conductances and synaptic conductances) in Bezaire et al. (2016) were obtained by relatively simple (partly manual) fitting procedures whose reliability and robustness are mostly unknown. Specifically for synaptic parameters in CA1, a more systematic review and calculation were recently carried out by Ecker et al. (2020); their estimates for the synaptic conductances in question are typically much lower than those of Bezaire et al. (2016) and appear to be more in line with widely accepted values for cortical (hippocampal) synapses.

    Furthermore, some key details concerning the construction of the simplified rate model are unclear in the current manuscript. The process of selecting cell types and connections for inclusion in the rate model is described, and the criteria are mostly clear, although the results are likely to be heavily affected by the problems discussed above, and I do not understand why the strength of external input was included among the selection criteria for cell types (especially if the model is meant to capture the internal dynamics of the isolated CA1 region). However, the main issue is that it remains unclear how the parameters of the rate model (the 24 parameters in Table 4) were obtained. The authors simply state that they "found a set of parameters that give rise to theta-gamma rhythms," and no further explanation is provided. Ideally, the parameters of the rate model should be derived systematically from the detailed biophysical model so that the two models are linked as strongly as possible; but even if this was not the case, the methods used to set these parameters should be described in detail.

    An important inaccuracy in the presentation of the results concerns the suggested coupling of theta and gamma oscillations in the models. Although the authors show that theta and gamma oscillations can be simultaneously present in the network under certain conditions, actual coupling of the two rhythms (e.g., in the form of phase-amplitude coupling) is not systematically characterized, and it is therefore not clear under what conditions real coupling is present in the two models (although a probable example can be seen in Figure 1C(ii)).

    The Discussion of the paper states that gamma oscillations in the model(s) are generated via a pure interneuronal (ING) mechanism. This is an interesting claim; however, I could not find any findings in the Results section that directly support this conclusion.

    Finally, although the authors write that they can "envisage designing experiments to directly test predictions" from their modeling work, no such experimental predictions are explicitly identified in the current manuscript.

  3. eLife

    Reviewer #2 (Public Review):

    The goal of this study is to find a minimal model that produces both theta and gamma rhythms in the hippocampus CA1, based on the full-scale model (FSM) of Bezaire et al, 2016. The FSM here is treated as equivalent to biological data. This seems to be a second part of a study that the same authors published in 2021, and is extensively cited here. The study reduces the FSM to a neural rate model with 4 neurons, which is capable of producing both rhythms. This model is then simulated and its parameter dependencies are explored.

    The authors succeed in producing a rate model, based on 4 neuron types, that captures the essence of the two rhythms. This model is then analyzed at a descriptive level to claim that the synapse from one interneuron type (CCK) to another (PV+) is more effective than its reciprocal counterpart (PV+ to CCK synapse) to control theta rhythm frequency.

    The results fall short on several fronts:
    The conclusions rely exclusively on the assumption that the FSM is in fact able to faithfully reflect the biological circuits involved, not just in its output, but in response to a variety of perturbations. Although the authors mention and discuss this assumption, in the end, the reader is left with a (reduced) model of a (complex) model, but no real analysis based on this reduction. In fact, the reduced model is treated in a manner that could have been done with the full one. Thus the significance of the work is greatly reduced not by what the authors do, but by what they fail to do, which is to properly analyze their own reduced model. Consequently, the impact of this study on the field is minimal.
    Related to the first point, throughout the manuscript, multiple descriptive findings, based on the authors' observations of the model output, are presented as causal relationships. Even the main finding of the study (that one synapse has a larger effect on theta than another) is not quantified, but just simply left as a judgment call by the authors and reader of comparing slopes on graphs.

  4. eLife

    Reviewer #3 (Public Review):

    While full-scale and minimal models are available for CA1 hippocampus and both exhibiting theta and gamma rhythms, it is not fully clear how inhibitory cells contribute to rhythm generation in the hippocampus. This paper aims to address this question by proposing a middle ground - a reduced model of the full-scale model. The reduced model is derived by selecting neural types for which ablations show that these are essential for theta and gamma rhythms. A study of the reduced model proposes particular inhibitory cell types (CCK+BC cells) that play a key role in inhibitory control mechanisms of theta rhythms and theta-gamma coupling rhythms.

    Strengths:
    The paper identifies neural types contributing to theta-gamma rhythms, models them, and provides analysis that derives control diagrams and identifies CCK+BC cells as key inhibitory cells in rhythm generation. The paper is clearly written and approaches are well described. Simulation data is well depicted to support the methodology.

    Weaknesses:
    The derivation methodology of the reduced model is hypotheses based, i.e. it is based on the selection of cell types and showing that these need to be included by ablation simulations. Then the reduced model is fitted. While this approach has merit, it could "miss" cell types or not capture the particular balance between all types. In particular, it is not known what is the "error" by considering the reduced model. As a result, the control plots (Fig. 5 and 6) might be deformed or very different. An additional weakness is that while the study predicts control diagrams and identifies CCK+BC cell types as key controllers, experimental data to validate these predictions is not provided. This weakness is admissible, in my opinion, since these recordings are not easy to obtain and the paper focuses on computational investigation rather than computationally guided experiments.

  5. eLife

    Author Response:

    eLife assessment

    This paper reports a useful set of results that uses a reduced network model based on a previously published large-scale network model to explain the generation of theta-gamma rhythms in the hippocampus. Combining the detailed and reduced models and comparing their results is a powerful approach. However, the evidence for the main claim that CCK+ basket cells play a key role in theta-gamma coupling in the hippocampus is currently incomplete.

    We thank the reviewers for their thorough and thoughtful notes, and we are pleased that there is acknowledgement of the combination of models as a powerful approach. We agree with many of the comments made and we intend to address them in subsequent revisions.

    In particular, we think that our ‘narrative’ as presented was perhaps not as clear as it could have been, based on the somewhat different comments from the reviewers (R#1 and #3). That is, we created a reduced population rate model based on the theta/gamma generation hypotheses from the detailed model and then explored the PRM in more detail to predict cellular contributions. The goal was not to validate the original detailed model per se (R#1) nor to do a fitting of parameters in the PRM directly from the detailed model (R#3). Rather, it was to obtain a set of parameter values in PRM that would be in accordance with the hypotheses of the detailed model that could be fully explored to derive cellular-based predictions that could help design experiments to understand theta/gamma rhythms.

    Responses specific to the Reviewers are given below.

    Reviewer #1 (Public Review):

    This paper investigates potential mechanisms underlying the generation of hippocampal theta and gamma rhythms using a combination of several modeling approaches. The authors perform new simulation experiments on the existing large-scale biophysical network model previously published by Bezaire et al. Guided by their analysis of this detailed model, they also develop a strongly reduced, rate-based network model, which allows them to run a much larger number of simulations and systematically explore the effects of varying several key parameters. The combined results from these two in silico approaches allow them to predict which cell types and connections in the hippocampus might be involved in the generation and coupling of theta and gamma oscillations.

    In my view, several aspects of the general methodology are exemplary. In the current work as well as several earlier papers, the authors are re-using a large-scale network model that was originally developed in a different laboratory (Bezaire et al., 2016) and that still represents the state-of-the-art in detailed hippocampal modeling. Such model reuse is quite rare in computational neuroscience, which is rather unfortunate given the amount of time and effort required to build and share such a complex model. Very often, and also, in this case, the original publication that describes a detailed model provides only limited validation and analysis of model behavior, and the re-use of the same model in later studies represents a great opportunity to further examine and validate the model.

    Combining detailed and simplified models can also be a powerful approach, especially when the correspondence between the two is carefully established. Matching results from the two models, in this case, allow strong arguments about key mechanisms of biological phenomena, where the simplified model allows the identification and characterization of necessary and sufficient components, while the detailed model can firmly anchor the models and their predictions to experimental data.

    On the other hand, I have several major concerns about the implementation of these approaches and the interpretation of the results in the current study. First of all, the detailed model of Bezaire et al. is considered strictly equivalent, in all of its relevant details, to biological reality, and no attempt is made to verify or even discuss the validity of this assumption, even when particular details of the model are apparently critical for the results presented. I see this as a fundamental limitation of the current work - the fact that the Bezaire et al. model is the best one we have at the moment does not automatically make it correct in all its details, and features of the model that are essential for the new results certainly deserve careful scrutiny (preferably via detailed comparison with experimental data).

    An important case in point is the strength of the interactions between specific neuronal populations. This is represented by different quantities in the detailed and simplified model, but the starting point is always the synaptic weight (conductance) values given by Bezaire et al. (2016), also listed in Tables 2 and 3 of the current manuscript. Looking at these parameters, one can identify a handful of connections whose conductance values are much higher than those of the other connections, and also more than an order of magnitude higher (50-100 nS) than commonly estimated values for cortical synapses (normally less than about 5 nS, except for a few very special types of synapse such as the hippocampal mossy fibers). Not surprisingly, several of these connections (such as the pyramidal cell to pyramidal cell connections, and the CCK+BC to PV+BC connections) were found to be critical for the generation and control of theta and gamma oscillations in the model. Given their importance for the conclusions of the paper, it would be essential to double-check the validity of these parameter values. In this context, it is worth noting that, unlike the anatomical parameters (cell numbers and connectivity) that had been carefully calculated and discussed in Bezaire and Soltesz (2013), biophysical parameters (the densities of neuronal membrane conductances and synaptic conductances) in Bezaire et al. (2016) were obtained by relatively simple (partly manual) fitting procedures whose reliability and robustness are mostly unknown. Specifically for synaptic parameters in CA1, a more systematic review and calculation were recently carried out by Ecker et al. (2020); their estimates for the synaptic conductances in question are typically much lower than those of Bezaire et al. (2016) and appear to be more in line with widely accepted values for cortical (hippocampal) synapses.

    Furthermore, some key details concerning the construction of the simplified rate model are unclear in the current manuscript. The process of selecting cell types and connections for inclusion in the rate model is described, and the criteria are mostly clear, although the results are likely to be heavily affected by the problems discussed above, and I do not understand why the strength of external input was included among the selection criteria for cell types (especially if the model is meant to capture the internal dynamics of the isolated CA1 region). However, the main issue is that it remains unclear how the parameters of the rate model (the 24 parameters in Table 4) were obtained. The authors simply state that they "found a set of parameters that give rise to theta-gamma rhythms," and no further explanation is provided. Ideally, the parameters of the rate model should be derived systematically from the detailed biophysical model so that the two models are linked as strongly as possible; but even if this was not the case, the methods used to set these parameters should be described in detail.

    An important inaccuracy in the presentation of the results concerns the suggested coupling of theta and gamma oscillations in the models. Although the authors show that theta and gamma oscillations can be simultaneously present in the network under certain conditions, actual coupling of the two rhythms (e.g., in the form of phase-amplitude coupling) is not systematically characterized, and it is therefore not clear under what conditions real coupling is present in the two models (although a probable example can be seen in Figure 1C(ii)).

    The Discussion of the paper states that gamma oscillations in the model(s) are generated via a pure interneuronal (ING) mechanism. This is an interesting claim; however, I could not find any findings in the Results section that directly support this conclusion.

    Finally, although the authors write that they can "envisage designing experiments to directly test predictions" from their modeling work, no such experimental predictions are explicitly identified in the current manuscript.

    As noted above, our goal was not to validate the original detailed model but to carry out further analysis of the Bezaire model in its re-use, since as noted by this Reviewer, the original publication was limited in validation and analysis. Further validation/extensions of Bezaire et al can be carried out given their acknowledged limitations (some as mentioned by the Reviewer). However, as noted, more detailed models of CA1 microcircuitry now exist (Ecker et al 2020), and it would be interesting to examine whether and how these more detailed models might express theta/gamma rhythms. In essence, we completely agree that all the details of the Bezaire et al model are not automatically correct. We were using it as a biological proxy, albeit imperfect. However, it is able to produce theta/gamma rhythms using parameter values that are experimentally derived in many ways (Bezaire & Soltesz 2013), and with minimal tuning, and thus our assumption is that it captures a potential ‘biological balance’ to generate these rhythms. Hence, we carried out additional simulations and explorations to derive generation hypotheses that are “applied” in the development of the reduced population rate model (PRM). The “ING” aspect is due to CCK+BCs and PV+BCs firing coherent gamma rhythms that are imposed onto the PYR cell population as mentioned in the Results. Without PYR input, they still fire coherent gamma rhythms. Experiments in which theta/gamma rhythms are characterized (CFC, frequencies) with and without the presence of CCK+BCs would allow the main prediction of the modeling work to be explored – i.e., whether CCK+BCs are essential for the existence of these coupled rhythms. We know from Dudok et al that there are alternating sources of perisomatic inhibition, but how they might control theta/gamma rhythms has not been explored to the best of our knowledge.

    We will more fully describe our process for PRM parameters in subsequent revisions as well as formally apply CFC metrics.

    Reviewer #2 (Public Review):

    The goal of this study is to find a minimal model that produces both theta and gamma rhythms in the hippocampus CA1, based on the full-scale model (FSM) of Bezaire et al, 2016. The FSM here is treated as equivalent to biological data. This seems to be a second part of a study that the same authors published in 2021, and is extensively cited here. The study reduces the FSM to a neural rate model with 4 neurons, which is capable of producing both rhythms. This model is then simulated and its parameter dependencies are explored.

    The authors succeed in producing a rate model, based on 4 neuron types, that captures the essence of the two rhythms. This model is then analyzed at a descriptive level to claim that the synapse from one interneuron type (CCK) to another (PV+) is more effective than its reciprocal counterpart (PV+ to CCK synapse) to control theta rhythm frequency.

    The results fall short on several fronts:
    The conclusions rely exclusively on the assumption that the FSM is in fact able to faithfully reflect the biological circuits involved, not just in its output, but in response to a variety of perturbations. Although the authors mention and discuss this assumption, in the end, the reader is left with a (reduced) model of a (complex) model, but no real analysis based on this reduction. In fact, the reduced model is treated in a manner that could have been done with the full one. Thus the significance of the work is greatly reduced not by what the authors do, but by what they fail to do, which is to properly analyze their own reduced model. Consequently, the impact of this study on the field is minimal.
    Related to the first point, throughout the manuscript, multiple descriptive findings, based on the authors' observations of the model output, are presented as causal relationships. Even the main finding of the study (that one synapse has a larger effect on theta than another) is not quantified, but just simply left as a judgment call by the authors and reader of comparing slopes on graphs.

    We agree with this Reviewer that analysis of the PRM is needed and is currently underway. It will hopefully help us understand what ‘balances’ are essential for theta/gamma rhythm expression. However, the overall goal of this work was not to “find” a minimal model per se, but rather to determine how theta/gamma rhythms in the hippocampus are generated (hence building on previous works). However, it was important to use the detailed model (biological proxy – albeit imperfect – see response to Reviewer#1) to obtain hypotheses on which the PRM is based. We do not envisage the minimal model as a `replacement’ for the detailed model in general, but rather, to show that using a combination approach (detailed and/or experimental observations with ‘derived’ reduced models) allows us to gain insight into cellular contributions to rhythm generation. Quantification of observations will be applied in subsequent revisions.

    Reviewer #3 (Public Review):

    While full-scale and minimal models are available for CA1 hippocampus and both exhibiting theta and gamma rhythms, it is not fully clear how inhibitory cells contribute to rhythm generation in the hippocampus. This paper aims to address this question by proposing a middle ground - a reduced model of the full-scale model. The reduced model is derived by selecting neural types for which ablations show that these are essential for theta and gamma rhythms. A study of the reduced model proposes particular inhibitory cell types (CCK+BC cells) that play a key role in inhibitory control mechanisms of theta rhythms and theta-gamma coupling rhythms.

    Strengths:
    The paper identifies neural types contributing to theta-gamma rhythms, models them, and provides analysis that derives control diagrams and identifies CCK+BC cells as key inhibitory cells in rhythm generation. The paper is clearly written and approaches are well described. Simulation data is well depicted to support the methodology.

    Weaknesses:
    The derivation methodology of the reduced model is hypotheses based, i.e. it is based on the selection of cell types and showing that these need to be included by ablation simulations. Then the reduced model is fitted. While this approach has merit, it could "miss" cell types or not capture the particular balance between all types. In particular, it is not known what is the "error" by considering the reduced model. As a result, the control plots (Fig. 5 and 6) might be deformed or very different. An additional weakness is that while the study predicts control diagrams and identifies CCK+BC cell types as key controllers, experimental data to validate these predictions is not provided. This weakness is admissible, in my opinion, since these recordings are not easy to obtain and the paper focuses on computational investigation rather than computationally guided experiments.

    This Reviewer has provided a succinct description of our work which we will leverage in subsequent revisions as we more fully describe our process – thank you. We agree with the Reviewer that we could ‘miss’ cell types and not capture particular balances etc., as we based our PRM on hypotheses from the detailed model. Our PRM and its reference parameter values are ‘designed’ based on hypotheses from our set of explorations of the detailed model, and we were able to determine particular predictions that can be experimentally explored. Subsequent theoretical analyses will help us understand the required ‘balances’ but as noted above (see response to Reviewer#2), we are not aiming for a minimal model (in general), but rather to use such a combined approach (detailed model and/or experimental observations with ‘derived’ reduced models) to come up with (cellular-based) predictions underlying theta/gamma generation. As noted by this Reviewer, specific inhibitory cell recordings are not easy to obtain and we hope our work would help with computationally guided experiments – i.e, even though the reduced model may ‘miss’ other aspects, it would hopefully capture some aspects that are biologically salient for consideration in experimental design and future detailed model explorations.

  6. Version published to 10.1101/2022.06.01.494440 on bioRxiv