Frequency and Laminar Profile of Feature Specific Visual Activity Revealed by Interleaved EEG-fMRI

Reviewer #1 (Public review):

In this manuscript, Clausner and colleagues use simultaneous EEG and fMRI recordings to clarify how visual brain rhythms emerge across layers of early visual cortex. They report that gamma activity correlates positively with feature-specific fMRI signals in superficial and deep layers. By contrast, alpha activity generally correlated negatively with fMRI signals, with two higher frequencies within the alpha reflecting feature-specific fMRI signals. This feature-specific alpha code indicates an active role of alpha oscillations in visual feature coding, providing compelling evidence that the functions of alpha oscillations go beyond cortical idling or feature-unspecific suppression.

The study is very interesting and timely. Methodologically, it is state-of-the-art. The findings on a more active role of alpha activity that goes beyond the classical idling or suppression accounts are in line with recent findings and theories. In sum, this paper makes a very nice contribution. I still have a few comments that I outline below, regarding the data visualization, some methodological aspects, and a couple of theoretical points.

(1) The authors put a lot of effort into the figure design. For instance, I really like Figure 1, which conveys a lot of information in a nice way. Figures 3 and 4, however, seem overengineered, and it takes a lot of time to distill the contents from them. The fact that they have a supplementary figure explaining the composition of these figures already indicates that the authors realized this is not particularly intuitive. First of all, the ordering of the conditions is not really intuitive. Second, the indication of significance through saturation does not really work; I have a hard time discerning the more and less saturated colors. And finally, the white dots do not really help either. I don't fully understand why they are placed where they are placed (e.g., in Figure 3). My suggestion would be to get rid of one of the factors (I think the voxel selection threshold could go: the authors could run with one of the stricter ones, and the rest could go into the supplement?) and then turn this into a few line plots. That would be so much easier to digest.

(2) The division between high- and low-frequency alpha in the feature-specific signal correspondence is very interesting. I am wondering whether there is an opposite effect in the feature-unspecific signal correspondence. Would the high-frequency alpha show less of a feature-unspecific correlation with the BOLD?

(3) In the discussion (line 330 onwards), the authors mention that low-frequency alpha is predominantly related to superficial layers, referencing Figure 4A. I have a hard time appreciating this pattern there. Can the authors provide some more information on where to look?

(4) How did the authors deal with the signal-to-noise ratio (SNR) across layers, where the presence of larger drain veins typically increases BOLD (and thereby SNR) in superficial layers? This may explain the pattern of feature-unspecific effects in the alpha (Figure 3). Can the authors perform some type of SNR estimate (e.g., split-half reliability of voxel activations or similar) across layers to check whether SNR plays a role in this general pattern?

(5) The GLM used for modelling the fMRI data included lots of regressors, and the scanning was intermittent. How much data was available in the end for sensibly estimating the baseline? This was not really clear to me from the methods (or I might have missed it). This seems relevant here, as the sign of the beta estimates plays a major role in interpreting the results here.

(6) Some recent research suggests that gamma activity, much in contrast to the prevailing view of the mechanism for feedforward information propagation, relates to the feedback process (e.g., Vinck et al., 2025, TiCS). This view kind of fits with the localization of gamma to the deep layer here?

(7) Another recent review (Stecher et al., 2025, TiNS) discusses feature-specific codes in visual alpha rhythms quite a bit, and it might be worth discussing how your results align with the results reported there.

Reviewer #2 (Public review):

The authors address a long-standing controversy regarding the functional role of neural oscillations in cortical computations and layer-specific signalling. Several studies have implicated gamma oscillations in bottom-up processing, while lower-frequency oscillations have been associated with top-down signalling. Therefore, the question the authors investigate is both timely and theoretically relevant, contributing to our understanding of feedforward and feedback communication in the brain. This paper presents a novel and complicated data acquisition technique, the application of simultaneous EEG and fMRI, to benefit from both temporal and spatial resolution. A sophisticated data analysis method was executed in order to understand the underlying neural activity during a visual oddball task. Figures are well-designed and appropriately represent the results, which seem to support the overall conclusions. However, some of the claims (particularly those regarding the contribution of gamma oscillations) feel somewhat overstated, as the results offer indeed some significant evidence, but most seem more like a suggestive trend. Nonetheless, the paper is well-written, addresses a relevant and timely research question, introduces a novel and elegant analysis approach, and presents interesting findings. Further investigation will be important to strengthen and expand upon these insights.

One of the main strengths of the paper lies in the use of a well-established and straightforward experimental paradigm (the visual oddball task). As a result, the behavioural effects reported were largely expected and reassuring to see replicated. The acquisition technique used is very novel, and while this may introduce challenges for data analysis, the authors appear to have addressed these appropriately.

Later findings are very interesting, and mainly in line with our current understanding of feedback and feedforward signalling. However, the layer weight calculation is lacking in the manuscript. While it is discussed in the methods, it would help to briefly explain in the results how these weights are calculated, so that the reader can better follow what is being interpreted.

Line 104 states there is one virtual channel per hemisphere for low and high frequencies. It may be helpful to include the number of channels (n=4) in the results section, as specified in the methods. Also, this raises the question of whether a single virtual channel (i.e., voxel) provides sufficient information for reproducibility.

One area that would benefit from further clarification is the interpretation of gamma oscillations. The evidence for gamma involvement in the observed effects appears somewhat limited. For example, no significant gamma-related clusters were found for the feature-unspecific BOLD signal (Figure 2). Significant effects emerged only when the analysis was restricted to positively responding voxels, and even then, only for the contrast between EEG-coherent and EEG-incoherent conditions in the feature-specific BOLD response. It remains unclear how to interpret this selective emergence of gamma-related effects. Given previous literature linking gamma to feedforward processing, one might expect more robust involvement in broader, feature-unspecific contrasts. The current discussion presents the gamma-related findings with some confidence, and the manuscript would benefit from a more nuanced reflection on why these effects may not have appeared more broadly. The explanation provided in line 230, that restricting the analysis to positively responding voxels may have increased the SNR, is reasonable, but it may not fully account for the absence of gamma effects in V1's feature-unspecific response. Including the actual beta values from Figure 4 in the legend or main text would also help readers better assess the strength and specificity of the reported effects.

Relating to behavioural findings for underlying neural activity, could the authors test on a trial-by-trial basis how behavioural performance relates to the BOLD signal / oscillatory activity change? Line 305 states that "Since behavioural performance in the present study was consistently high at 94% on average and participants were instructed to respond quickly to potential oddball stimuli, a higher alpha frequency might reflect a more successful stimulus encoding and hence faster and more accurate behavioural performance." Also, this might help to relate the findings to the lower vs upper alpha functionality difference.

In Figure 4, the EEG alpha specificity plot shows relatively large error bars, and there is visible overlap between the lower and upper alpha in both congruent and incongruent conditions. While upper alpha shows a positive slope across conditions and lower alpha remains flat, the interaction appears to be driven by the change from congruent to incongruent in upper alpha. It is worth clarifying whether the simple effects (e.g., lower vs upper within each condition) were tested, given the visual similarity at the incongruent condition. Overall, the significant interaction (p < 0.001, FDR-corrected) is consistent with diverging trends, but a breakdown of simple effects would help interpret the result more clearly. Was there a significant difference between lower and upper alpha in congruent or incongruent conditions?

Overall, this study provides a valuable contribution to the literature on oscillatory dynamics and laminar fMRI, though some interpretations would benefit from further clarification or qualification.

Reviewer #3 (Public review):

Summary:

Clausner et al. investigate the relationship between cortical oscillations in the alpha and gamma bands and the feature-specific and feature-unspecific BOLD signals across cortical layers. Using a well-designed stimulus and GLM, they show a method by which different BOLD signals can be differentiated and investigated alongside multiple cortical oscillatory frequencies. In addition to the previously reported positive relationship between gamma and BOLD signals in superficial layers, they show a relationship between gamma and feature-specific BOLD in the deeper layers. Alpha-band power is shown to have a negative relationship with the negative BOLD response for both feature-specific and feature-unspecific contrasts. When separated into lower (8-10Hz) and upper (11-13Hz) alpha oscillations, they show that higher frequency alpha showed a significantly stronger negative relationship with congruency, and can therefore be interpreted as more feature-specific than lower frequency alpha.

Strengths:

The use of interleaved EEG-fMRI has provided a rich dataset that can be used to evaluate the relationship of cortical layer BOLD signals with multiple EEG frequencies. The EEG data were of sufficient quality to see the modulation of both alpha-band and gamma-band oscillations in the group mean VE-channel TFS. The good EEG data quality is backed up with a highly technical analysis pipeline that ultimately enables the interpretation of the cortical layer relationship of the BOLD signal with a range of frequencies in the alpha and gamma bands. The stimulus design allowed for the generation of multiple contrasts for the BOLD signal and the alpha/gamma oscillations in the GLM analysis. Feature-specific and unspecific BOLD contrasts are used with congruently or incongruently selected EEG power regressors to delineate between local and global alpha modulations. A transparent approach is used for the selection of voxels contributing to the final layer profiles, for which statistical analysis is comprehensive but uses an alternative statistical test, which I have not seen in previous layer-fMRI literature.

A significant negative relationship between alpha-band power and the BOLD signal was seen in congruently (EEGco) selected voxels (predominantly in superficial layers) and in feature-contrast (EEGco-inco) selected (superficial and deep layers). When separated into lower (8-10Hz) and upper (11-13Hz) alpha oscillations, they show that higher frequency alpha showed a significantly stronger negative relationship with congruency than lower frequency alpha. This is interpreted as a frequency dissociation in the alpha-BOLD relationship, with upper frequency alpha being feature-specific and lower frequency alpha corresponding to general modulation. These results are a valuable addition to the current literature and improve our current understanding of the role of cortical alpha oscillations.

There is not much work in the literature on the relationship between alpha power and the negative BOLD response (NBR), so the data provided here are particularly valuable. The negative relationship between the NBR and alpha power shown here suggests that there is a reduction in alpha power, linked to locally reduced BOLD activity, which is in line with the previously hypothesized inhibitory nature of alpha.

Weaknesses:

It is not entirely clear how the draining vein effect seen in GE-BOLD layer-fMRI data has been accounted for in the analysis. For the contrast of congruent-incongruent, it is assumed that the underlying draining effect will be the same for both conditions, and so should be cancelled out. However, for the other contrasts, it is unclear how the final layer profiles aren't confounded by the bias in BOLD signal towards the superficial layers. Many of the profiles in Figure 3 and Figure 4A show an increased negative correlation between alpha power and the BOLD signal towards the superficial layers.

When investigating if high alpha (8-10 Hz) and low alpha (11-13 Hz) are two different sources of alpha, it would be beneficial to show if this effect is only seen at the group level or can be seen in any single subjects. Inter-subject variability in peak alpha power could result in some subjects having a single low alpha peak and some a single high alpha peak rather than two peaks from different sources.

The figure layout used to present the main findings throughout is an innovative way to present so much information, but it is difficult to decipher the main findings described in the text. The readability would be improved if the example (Appendix 0 - Figure 1) in the supplementary material is included as a second panel inside Figure 3, or, if this is not possible, the example (Appendix 0 - Figure 1) should be clearly referred to in the figure caption.