The visual cortex produces gamma band echo in response to broadband visual flicker

The aim of this study is to uncover the network dynamics of the human visual cortex by driving it with a broadband random visual flicker. We here applied a broadband flicker (1–720 Hz) while measuring the MEG and then estimated the temporal response function (TRF) between the visual input and the MEG response. This TRF revealed an early response in the 40–60 Hz gamma range as well as in the 8–12 Hz alpha band. While the gamma band response is novel, the latter has been termed the alpha band perceptual echo. The gamma echo preceded the alpha perceptual echo. The dominant frequency of the gamma echo was subject-specific thereby reflecting the individual dynamical properties of the early visual cortex. To understand the neuronal mechanisms generating the gamma echo, we implemented a pyramidal-interneuron gamma (PING) model that produces gamma oscillations in the presence of constant input currents. Applying a broadband input current mimicking the visual stimulation allowed us to estimate TRF between the input current and the population response (akin to the local field potentials). The TRF revealed a gamma echo that was similar to the one we observed in the MEG data. Our results suggest that the visual gamma echo can be explained by the dynamics of the PING model even in the absence of sustained gamma oscillations.

• Now we conducted additional analysis to describe topographies of the gamma echo and grating induced gamma oscillations at the sensor and source levels (see, lines 161-174).
The results showed that both the gamma echo and induced gamma oscillations originated in the primary visual cortex and did not appear to propagate outside of this area.
-It is not clear to me the difference between endogenous gamma and gamma echoes (and flicker responses). Related to the previous point, would it be possible to localize precisely to V1 the gamma echoes, to support the authors' claim that these are originated in V1, whereas endogenous gamma is originated in V2? Also, what would be the functional role -if any-of gamma echoes and endogenous gamma oscillations? In this light, why would the endogenous gamma be higher in the visual hierarchy (as suggested by the authors)?
• Indeed, the link between induced gamma activity and the gamma echo needed to be better characterized. To address this point, we computed time-frequency representation of MEG power at the occipital sensor with the strongest response to flicker and described the results (see, lines 148-154). In short, the visual grating stimuli induced gamma oscillations at around 57 Hz (grand average), while the frequency of the gamma echo was around 48 Hz. This result together with the source level analysis suggests that the gamma echo and induced gamma oscillations are produced by different but neighboring sources. This is also consistent with our recent work by Duecker et al. (2020;bioRxiv).
-Your simulation shows that an oscillatory input at the right resonance frequency (48Hz) is amplified by the network, producing an amplified peak in the power spectral density. Isn't this result in contradiction with previous findings showing that visual stimulation may not entrain gamma oscillations? (reference [46] in the paper, from the same senior author as the current study).
• This is an excellent point which prompted us to quantify the induced gamma oscillations (see Fig. 3). In this revised version, we reconsidered and refined parameters of our model, so that the results slightly changed (lines 287-302). In contrast to our previous observations, the amplification was not narrowband but rather peaked at around 48 Hz and decayed towards higher frequencies (see, Fig. 10E). This result closely resembled findings by Gulbinaite et al., 2019 at higher frequencies > 40 Hz (see, Fig. 3C in Gulbinaite et al., 2019). For convenience, we combined Fig. 10E and Fig. 3C from Gulbinaite et al., 2019 into a single figure R1 (see below).
• Indeed, study by Duecker et al., 2020 (old reference [46]) did not report a resonance frequency, which may seem like a contraction to our current findings. However, considering the source modelling results, we conclude that the gamma echo and the induced gamma oscillations are produced by different generators; while the gamma echo is produced by area V1, the endogenous gamma oscillations are produced slightly downstream (possibly area V2). This is now fully consistent with the work of Duecker et al, 2020 demonstrating the visual flicker does not entrain endogenous gamma oscillations. We have updated the Discussion section accordingly (lines 373-390). -The model's coefficients and parameters have not been explored at all in this study. However, it could be interesting to investigate how the TFR changes properties as a function of some relevant parameters. Ideally, this could shed some light on the differences between participants (peak frequency, amplitude, etc.). Additionally, it would be best to motivate the choice of the parameters' value (are they all the same as in the cited Izhikevich model ?) and their biological plausibility. Would the coefficients' value be different in other brain regions (e.g., higher in the visual system), thus predicting differences in the TFR?
• The exploration of parameters has helped us to ensure that the dynamics of the model is indeed robust. As expected from the literature the GABAergic conductance is a key parameter to set frequency of the network.
-Why equalizing the number of trials per condition in each participant? Isn't it reasonable to assume that the two conditions (coherent and incoherent motion) are orthogonal to the TRF?
• We equalized the number of trials per condition because the resulting TRF is an average of the TRF computed for individual trials. This procedure serves to avoid a potential bias in TRF estimation related to unbalanced number of trials. We added this note in the Method section (lines 497-498). were assessed for the same participant (S1), but at slightly different occipital sensors (~4 cm apart). The reason is that the sensor with strongest response to flicker did not capture the alpha echo, so we arbitrarily selected an occipital gradiometer that captured both alpha-band and gamma-band echoes. We updated the text accordingly, "Note that the TRF was computed for an occipital gradiometer that captures both the alpha-band and gamma-band TRF" (line 159-161).
• We have made the data and code available on OSF website (https://osf.io/fe8x5/). The MATLAB script "data/ge_TRF.m" can be used to reproduce the gamma echoes.
-It would be interesting, in the next study, to replicate these results on a larger population. In alpha-band echoes, some participants don't manifest any reverberation, possibly due to the cortex's anatomical configurations. If this is not the case in the gamma echoes (from figure 3, it seems that 5 participants out of 5 show them quite clearly), this could provide some indications that different mechanisms are involved in the two frequency bands.
• There are several differences between alpha-band and gamma-band echoes. First, while alpha echo occurs at around 0.2 s, gamma echo has much earlier onset at 40 ms. Second, gamma echo is largely localized in the primary visual cortex, but alpha echo propagates over the cortex (e.g., Alamia and VanRullen, 2019). Third, the alpha-echo spans for up to 10 cycles, while gamma echo vanishes after 2-3 cycles. Thus, all these different characteristics suggest that the echoes may not be strongly related and perhaps, generated by different mechanisms. We now added this to the Discussion (lines 345-351).
-Related to the previous point, it could be interesting to discuss/speculate about the differences between the echoes in the two frequency bands: do they both reflect two resonance properties of different neuronal networks, the gamma one more locally and the alpha one more globally? Or do they underlying different mechanisms and functions?
• Now we incorporated these points to the Discussion section under the title "Relationship between alpha and gamma echoes" (lines 343-372).
Minor details : -in the 'power spectral density' the epochs are defined as 5 seconds long; in the previous paragraph, it was 4. Which one is it?
• Indeed, there was a typo in the description and now we corrected it accordingly (lines 501-502). To avoid confusion, in the revised version we made MEG and simulated data trials of equal length (4 s).
-It seems that some figures are referenced in the wrong place. The 'Model' paragraph refers to figure 2A, but the model is in figure 4. Similarly, later the connectivity matrix refers to figure 3A, but I see it in 4B.
• We apologies for this confusion. Now we corrected the figure references as follows: "We modelled the neuronal populations of cortical areas as a network of interconnected excitatory and inhibitory neurons (Fig. 5A)".

Reviewer #2:
In this manuscript the authors applied a broadband flicker (1-720 Hz) to drive human visual cortex while measuring the MEG response. Moreover they estimated the temporal response function (TRF) between the visual input and the MEG response, which revealed an early response in the 40-60 Hz gamma range as well as in the 8-12 Hz alpha band.
Simulations of a network model for gamma oscillations with a broadband signal complete the results, thus allowing the authors to estimate numerically the TRF and to compare it with the one from MEG study.
The writing of the article is often sloppy and many paragraphs need to be re-written in order to improve clarity. Some examples are reported in the detailed list of comments/questions in the following.

Moreover I find the Introduction and the Discussion too focused on the experimental results, while a sufficient literature regarding the state-of-art of the models is neglected.
• We now described relevant models in the Introduction (lines 85-93) and Discussion (lines 405-422) sections.
All in all the results of this manuscript are more oriented towards numerical experiments than the experimental ones, thus creating an imbalance in the discussion.
• Now we added more experimental results including source modelling of the gamma echo and induced gamma oscillations, and time-frequency representation of the induced gamma activity at sensor level. These results are crucial for understanding neurophysiological basis of the gamma echo. We hope that the discussion now is more balanced.

INTRODUCTION
1. Citing the text: Furthermore, the GABAergic feedback also serves to synchronize the population activity [14,15] "The pyramidal neurons also play an important role in the mechanism." In each cycle [...] The sentence reported in "..." seems to have been forgotten there.
• We removed the redundant sentence.

Why do you use such a limited number of neurons (200 in total)?
• In this revised version, we increased the number of neurons to 500, following methodology in Quax et al., 2017, PLOS Comp Biol. Now we also explored the impact of the network size on the model firing rate (frequency) and power (see, lines 242-247).

How is chosen the connectivity matrix?
• In this revised version, we refined the model parameters, so the connectivity has been changed. Detailed description of the connectivity matrix is now provided in the text (lines 587-602).

"The temporal response function (TRF) of a system..."
The acronym TRF has been already introduced before; there is no necessity to reintroduce it now.

Merge the paragraphs "TRF show individual resonance frequency of gamma echo" and "
The gamma echo is close to 50 Hz but not due to line noise" for a better comprehension of the results shown in Fig. 3. At the moment it is necessary to read the latter paragraph in order to understand what is written in the previous one, where Fig. 3 is cited.

Could you better explain the motivations underlying the choise of connection strength? In the text is written: "we used a relatively low connection strength between Ecells (cee = 0.001), high connection strength between I-cells (cii = 0.200), while keeping connectivity between E-to-I-cells (cei = 0.050) and I-to-E-cells (cie = 0.010) at a moderate level"
A biophysical justification would help.
• We now provided a detailed explanation in the text (lines 587-602).
8. Fig. 4: Merge panels C and D. The information now contained in panel C is negligible. Add in panel A the symbols for 2 external curves impinging both E and I-cells.
9. Fig. 4: If you simulate for longer times, do you still see oscillations? Locking at panel E, it seems that the amplitude of the oscillations is shrinking, thus suggesting a possible transient phenomenon.
• We now simulated signal of 4 seconds long, to make it consistent with MEG data. As described in the text (lines 230-234), the transient effect lasted less than 0.5 s, and we discarded the first 1 s of the signal in all the analyses.
10. Fig. 4: A dependence of the results on the amplitude of the external current need to be shown in order to justify the choise of 25 pA.
• Indeed, the amplitude of input currents strongly determine the output characteristics of the model. Now we explored the parameter space of the input currents while preserving the connectivity strength and the network size (lines 207-221).

Fig. 5: How much do the results depend on the amplitude of the noise on the E-cells?
What happens for smaller and bigger amplitude values with respect to 16pA?
• We now explored the impact of the amplitude of broadband signal on the characteristics of gamma echo (lines 274-281). The results clearly showed that a specific input current (in this case ~4 µA) is required to generate the gamma echo that is comparable with that in MEG. Lower or higher input currents produced responses with incomparably long or short (respectively) decay times.
12. The definition of the external current as "a sinusoidal function with a given frequency (0 -100 Hz)" is misleading. It should be clearly mentioned the employed value (or values) of the used frequency.
• We improved the description of the oscillatory input (lines 297-299).
13. Fig. 6: In panel C it is calculated the average membrane potential of the E-cells. Is it the same also for the previous figures? If yes, please write it explicitly.
• Yes, all this is now corrected and clarified in the caption.
14. Fig 6: Which is the range of observability of the shown phenomenon? If you vary the frequency of the external current, in which frequency range do you still observe it?
• Now we explored the model response to different oscillatory inputs (lines 312-321) and described the range in which the amplification effect is observed.
Are you able to observe both alpha and gamma peaks at the same time?
• We did not observe both alpha and gamma peaks because the model parameters were tuned to produce gamma band echo. In the Discussion we now write: "In future work it would interesting to extend the model framework such that the network can produce alpha oscillations as well." (lines 416-417).
From what I see observing panel D, the emergent frequency is simply the frequency of the external stimulation.
• Yes, this observation is correct. Our new results showed that there was a minimum input current of 5 µA, above which, frequency of the network oscillations matched the frequency of the input current. This suggests that the frequency of external stimulation does not directly translate to the firing rate. Now we described this in the text (lines 312-321).
15. Fig. 6: is the process shown in panel C stationary? What does it happen if you simulate for longer times?
• As we discussed earlier (point 10), the transient effect was relatively short (~0.5 s), and we discarded the first 1 s of the signal from the analysis to make sure that the remaining signal is stationary (see, lines 230-234).

DISCUSSION
16. Please replace "Using broadband visual input stimuli were here provide evidence for [..}" with "Using broadband visual input stimuli we here provide evidence for [..]" • Corrected.
To conclude, the manuscript cannot be published in this form: the text is sloppy and a detailed investigation of the phenomenon is not presented. In order to understand the origin of the phenomenon, a deeper investigation has to be provided. At the moment I just see (probably) transient phenomena, that are shown just for few exemplary cases, without giving a clear view on the dependence on the amplitude and the frequency of the external stimulation.