Computational Modeling of Information Propagation during the Sleep–Waking Cycle

Round 1

Reviewer 1 Report

This is a solid modelling study, with potentially interesting results for better understanding the dynamics of the sleep-wake cycle.  I found no major flaw in the presentation or methodology. My report, however, outlines minor questions and comments that hopefully can help improve the clarity of the paper.

Line 51 : I’m not sure what elements are associated with ‘respectively’.

Line 53 : Is there a specific model being referenced here? Or do you mean a conceptual model?

Line 54 : Perhaps rephrase?: “Cortical responses to transcranial magnetic stimulation (TMS) have been shown to be local during …

Line 58: Downscaling of the conductances of excitatory synapses?

Line 65: Same comment: the upscaling of the conductances ...?

Line 63-65: Given that the decrease in sensory responsivness occurs in non-primary cortical areas, is the model supposed to represent the primary auditory cortex (like mentioned in the abstract), or higher cortical areas, such as those listed on line 50? Either way, this should be specified explicitly at this point.

Line 64-65: The sleep-waking cycle includes a transition through REM sleep does it not? Does the model have a REM state? Perhaps a comment could answer this question in the intro.

Line 77: The acronym SHY has already been define at this point, so either use it here, or remove the acronym’s definition in the intro.

Linr 87-88: The classical HH model has a K- and an Na-conductance. So perhaps rephrase to: … is governed by equations within the framework of the classical model ….

Line 89: What is the justification for the I_KNa?

Line 106: Add a reference to justify the fact that inter-column connections are purely excitatory, despite the fact the intra-colum connections are both excitatory and inhibitoy.

Line 105: Which Gaussian process? An O-U process? Perhaps specify the autocorrelation function?

Line 105: Could you comment on the validity of the representation of the extrinsic input by a Gaussian process, both during NREM and wakefulness? I agree that during wakefulness the extrinsic activity could resemble a simple O-U process (i.e., obtained from an order 1 SDE with parabolic potential function), but for NREM, the nature of the dynamics changes so much that I would assume a second order SDE (i.e., with intertial effects) or a bistable order 1 SDE would be more accurate representations.

Line 107: I feel that this sentence should start with: “given a presynaptic population k’ and a postsynaptic population k, the time course …

Line 113: I’m confused by equation 3a. If alpha_k’ is an impulse response function (i.e., a function of time representing the response to a single impulse), then shouldn’t the other component of the convolution be the input spike train? I guess the “impulse” mentioned here is should not be interpreted as a single spike, but rather as an impulse in the firing rate. Perhaps add a sentence to remove any potential ambiguity?

Line 118: Eq. 4 should be derived in the appendix.

Line 182: Can a comment be added as to whether the perturbed cortical column is assumed to be forced by the same Gaussian noise as the unperturbed one? Wouldn’t is also receive stochastic input from other columns? Perhaps this input might even be correlated to that of the perturbed column, given that thalamic feedback might be at play?

Line 242: Classifying binary states out of a stochastic bimodal time series is a non-trivial problem, and intermediate values can often be ambiguous. Given the fact that the bimodal distribution of LFP values during NREM is far from dropping down to zero between the peaks, perhaps it could be useful to consider a sort of Schmidt trigger approach, where two thresholds are defined instead of only one: e.g., a down state is identified when the LFP undergoes downward-crossing of the lower threshold and an upward-crossing of the upper threshold. This might be a way to avoid detecting transitions when the LFP meanders around intermediate values, the goal of course being to reduce the number of “false transition detections”.

Line 256: At the beginning of the paragraph, we mention that LFP signals are non-stationary, which justifies the use of a time-frequency analysis tool such as the STFT, but at the end of the paragraph you mention the use of Welch’s method to obtain the average PSD, which what is shown in Figure 2. What is the validity of measuring this average PSD given that the signal is non-stationary? Perhaps a comment could be added regarding the features of characteristic of the non-stationarity?

Line 278: These significance tests.

Line 441: The use of “wider” is ambiguous here. Are you referring to other sensory regions? Higher brain regions? Etc..

Line 458: I’m somewhat impressed by this result, and would be curious to know if this study is the first to show, with a model, that a single physiological parameter could be responsible for the transition between NREM and wakefullness. If so, then perhaps the novelty of this discovery could be made more apparent in the abstract and/or introduction.

Line 459: Has a dynamical analysis been carried out to support this claim? The slow-wave oscillation are indeed suggestive of a limit cycle, but given the stochasticity of the signals, could it not also be possible that the system undergoes a saddle-node instead? In other words, stochastic bistable dynamics can sometimes be mistaken for stochastic limit cycle dynamics.

Figure 4: I think this figure could be optional, it does not show anything else than what is already shown in Figure 2, which obviously is the whole point of this figure, but still, you might consider moving it to supplementary figures.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper shows an implementation of neural-mass models to account for results (such as Massimini et al. 2005), showing a larger propagation of responses in the awake brain, while the response tends to stay local in the sleeping brain.  There are a number of problems with this study which must be solved before any further step.

1. The circuit diagram with the connectivity is wrong.  The authors only consider excitatory connections between excitatory cells, but in reality, cortico-cortical (as well as thalamo-cortical) fibers always contact pyramidal neurons and inhibitory neurons in parallel.  This feature is not in the model, and it could critically change the results.  So I urge the authors to modify their model structure to include the correct connectivity and see if their findings still hold.  If there is no correct connectivity, there would be no point publishing this paper.

2. The main conclusion, that there must be an increase of the excitatory conductance to account for propagation in the waking state, is kind of deceiving.  This seems ad-hoc.  Again, this conclusion may be an artifact of using a wrong connectivity scheme, so this should be re-evaluated.

3. Another study found that the Massimini et al. results can be reproduced by neural-mass models, but without changing the synaptic weights (Goldman et al. Biorxiv 2020).  It is astonishing that the authors do not cite this work because both seem to be from the HBP (is there some internal competition inside the project?)  In any case, this previous model is quite close, it uses neural mass models, it also simulates wake and sleep (up and down states), and it also simulates the propagation of information based on Massimini results.  They find that the response remains local in sleep and propagates in wake, with no change of synaptic strength, so it seems important to compare in detail the two studies.  Why do they get different results?

4. Massimini et al. used a precise measure to quantify the spread of responses in cortex, the "Perturbational Complexity Index" (PCI).  It seems important that the authors provide estimates of the PCI and see whether their model produces results comparable to experimental data in a more quantitative way.

Given the amount of work needed, these points would call for a rejection, but I am willing to be positive and give a chance to the paper.  My recommendation is to consider the paper for publication but only if all the above points are fixed.  

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The proposed work comprises the evaluation of neural-mass models, even proposing an extension to two bidirectionally connected cortical columns. The work is of high relevance and the problem is very well selected. However, I recommend the authors to please address the following queries:

-In the first paragraph of the introduction I think it is worth to mention the role of the Cyclic Alternating Pattern (CAP) (for more information about this topic I recommend the article: Parrino L, Ferri R, Bruni O, Terzano MG. Cyclic alternating pattern (CAP): the marker of sleep instability. Sleep Med Rev. 2012 Feb;16(1):27-45. doi: 10.1016/j.smrv.2011.02.003).

-Please present the objectives and the key novelties of the work (in comparison with the significant body of knowledge that is already available in the literature) in the introduction. To me, the introduction is the weakest point of all the work, a lot of information is provided but the most significant elements for the researchers that want to read this work is missing (objectives, novelties, and state of the art review). I also recommend to indicate the article’s structure at the end of the introduction to clear the used organization for the reader.

-Please further detail where is “The code is available at github” (please provide the link or at least specify the repository).

-Please further detail what are the advantages, relevance, and limitations of the “model extension to two cortical columns” proposed in this work.

-I recommend to divide the discussion section in two, one to discuss the results and compare with other state of the art works (a proper comparison, in terms of model’s development and implementation, with other state of the art works is missing), and the second part to be a new section to present the conclusion of the work (stressing what are the key conclusions reached in this work, what are the limitations, and what are the future works).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf