Relating neural oscillations to laminar fMRI connectivity

  1. René Scheeringa
  2. Mathilde Bonnefond
  3. Tim van Mourik
  4. Ole Jensen
  5. David G. Norris
  6. Peter J. Koopmans

  • Curated by eLife

    eLife logo

    Evaluation Summary

    This study uses non-invasive imaging to look at the coupling within and between layers and regions of the human visual cortex during the modulation of attention. The results presented here are a re-analysis of a previously recorded dataset, but the novelty is the analytic technique used to relate laminar connectivity to rhythms. This in principle promises to advance the field of both oscillations and laminar fMRI and could deliver valuable insights. The work provides a non-invasive window on how feedback and feedforward circuitry in the human brain operates. We deem the work of potential interest to a broad audience as it aims to provide direct links between the animal invasive electrophysiology and human neuroimaging fields. However, in its current form, major reservations with respect to the hypothesis space being explored here as well as important analytic and technical caveats remain.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #3 agreed to share their names with the authors.)

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

Laminar fMRI holds the potential to study connectivity at the laminar level in humans. Here we analyze simultaneously recorded EEG and high resolution fMRI data to investigate how EEG power modulations, induced by a task with an attentional component, relate to changes in fMRI laminar connectivity between and within brain regions. Our results indicate that our task induced decrease in beta power relates to an increase in deep-to-deep layer coupling between regions and to an increase in deep/middle-to-superficial layer connectivity within brain regions. The attention-related alpha power decrease predominantly relates to reduced connectivity between deep and superficial layers within brain regions, since, unlike beta power, alpha power was found to be positively correlated to connectivity. We observed no strong relation between laminar connectivity and gamma band oscillations. These results indicate that especially beta band, and to a lesser extent alpha band oscillations relate to laminar specific fMRI connectivity. These differential effects for the alpha and beta bands suggest a complex picture of possibly co-occurring neural processes that can differentially affect laminar connectivity.

Article activity feed

  1. eLife

    Author Response:

    Reviewer #1:

    Weaknesses:

    Although the BOLD data is highly spatially specific, there is just one electrophysiological timeseries per subject. This is no doubt a bi-product of the extensive noise cancellation that is necessary to record within the scanner. The caveat therefore is that the covarying BOLD and electrophysiological changes may derive from different regions.

    We recognize this is a limitation which is also not easily solved by approaches for source analysis, given the nature of the data (only 64 channels) and the usually larger imprecisions related to EEG source reconstruction. We circumvented this by choosing a task that is known from previous studies in MEG to induce changes in multiple frequency bands originating from regions the early visual cortex (Hoogenboom et al., 2006; Hoogenboom et al., 2010; Koch et al., 2009; Muthukumaraswamy and Singh, 2013). Furthermore, the EEG responses are highly similar to invasive recordings in animals from visual regions in the context tasks investigating selective attention (Fries et al., 2008). We mention this limitation now in the introduction (lines 102-111).

    The analysis methods are slightly non-standard, perhaps for good reason. The main thing that stands out is the use of correlation coefficients, rather than regression coefficients, at the first

    level of analysis. This could potentially conflate changes in signal with changes in noise or unexplained variance.

    We chose here for the correlation, since in our opinion this leads to a more interpretable measure of linear association than a regression slope. A regression slope-based analysis will yield different outcomes for the regression of y on x, than for x on y, doubling the number of analyses needed. The different results for a regression of y on x and x on y are often interpreted as implying directionality, which is not warranted and not what we would like to imply with our analysis. The asymmetry is caused by the implicit assumption that x does not contain noise in a regression of y on x. This is valid when x represents a paradigm condition vector, but not when it is a data vector. We therefore opted to use the difference in (Fisher-z-transformed) correlation as our estimate for linear association/connectivity between laminar fMRI signals.

    In both a correlation as well as in a regression approach differences can be attributed to differences in true underlying coupling andin a difference in noise. This is however not different for correlation-based measures in coupling in fMRI than in for instance coupling measures like coherence and phase-locking-factor in electrophysiology. Coherence can be regarded as the frequency domain version of (squared) correlation. The fact that our measure might indeed be related to differences in noise would therefore not be resolved by opting for a regression based approach, and is not different from often used measures of coupling in electrophysiology.

    Reviewer #2 (Public Review):

    Introduction. The introduction provides an overoptimistic view on the current possibilities with respect to the investigation of layer-specific activation or connectivity in the living human brain. Cortical layers cannot yet be segmented, the fMRI measures only provide an indirect signal that is heavily influenced by partial voluming between cortical depths, and EEG and MEG approaches often only measure two compartments due to low spatial resolution. The introduction, however, gives the impression that layer-specific neuronal connectivity can precisely be measured in the living human brain, which is not the case. The authors should take considerably more care with respect to how they introduce the methodology with clear references to the limitations. Also, statements such as "laminar fMRI allows us to study connectivity.." should be removed. In the same vein, I would suggest to replace laminar fMRI and laminar connectivity with cortical depthdependent fMRI and connectivity to account for the above mentioned aspects.

    In laminar fMRI research it is commonly accepted that what we measure are not true layers, but depth dependent fMRI between the boundaries of white/gray matter and gray matter/CSF. For the general audience we will make this distinction clearer and discuss the limitations of the technique (lines 74-80).

    Concept. Whereas the authors provide a model in the introduction that specifies how different frequency bands could relate to cortical depth-dependent connectivity, they do not develop a working hypotheses based on their experimental design. One conceptual step is therefore missing in the introduction, which has to combine present knowledge on the relationship between different frequency bands and present knowledge on how attention influences frequency-specific activation in the visual system to then make statements about which analyses can be performed to test which aspect of the model.

    The primary focus of our study was to investigate how oscillations across several frequency bands in the EEG relate to laminar specific activity. Recent publications on laminar fMRI have demonstrated the possibility of performing laminar level fMRI connectivity analyses, which led us to revisit our previously recorded data in order to explore whether not only laminar specific BOLD amplitude but also laminar fMRI connectivity relates to frequency specific EEG power. Since laminar fMRI, and especially connectivity derived from those measures is very novel, we started this analysis without a preconceived model or notion on how this relation would be. The results from this project should therefore be interpreted as an exploration of how these laminar fMRI derived connectivity measures relate to neural oscillations rather than directly addressing a specific cognitive process like selective attention, or prediction and/or a model of how neural oscillations play a role in these processes. Our experimental paradigm was also not designed to address such processes. We chose a paradigm that is known from previous studies using MEG and EEG to induce changes in multiple frequency bands in the early visual cortex (Hoogenboom et al., 2006; Hoogenboom et al., 2010; Koch et al., 2009; Muthukumaraswamy and Singh, 2013). Furthermore, the EEG responses are highly similar to invasive recordings in animals from visual regions in the context tasks investigating selective attention (Fries et al., 2008). The crude attention/task modulation added to the paradigm (attention On versus Off) was in the first place introduced to induce meaningful variation over subjects in a task effect across the frequency bands modulated by visual stimulation. It was not intended to investigate specific individual processes such as prediction, attention or arousal. The observed effects can therefore also not be ascribed to such specific processes, since they are co-modulated by the task. We will make this more clear in the introduction now. We make this point now explicitly in the introduction.

    • Concept & Methods: With respect to both the concept and the analyses, what is missing is taking into considerations the brain areas that were investigated. Wheres in the abstract the authors only mention "within brain region connectivity" and "between brain region connectivity" also in the Methods section there is no clear relation to the anatomical areas that were investigated, being V1, V2 and V3. The authors rather classify the areas as "high level" and "low level" where V2 is sometimes classified as high-level and sometimes as low-level. The data are therefore not investigated with reference to the anatomy of the visual system. In my view, it would be beneficial if all analyses could be performed with respect specifically to V1-V2 connecitivity and V2-V3 connectivity as well as V1-V3 connectivity so that the specific anatomical interrelations are taken into account. Also, the authors should develop a conceptual framework of how layer-specific attention-driven connectivity changes should influence the visual cortex, and why.

    In the results for between region connectivity we averaged over several connection pairs (V1- V2,V1-V3,V2-V3) and for within region connectivity across regions (V1-V1,V2-V2,V3-V3) before effects in connectivity were correlated with EEG power. There are several reasons why we opted for this approach: First, we wanted to maximally increase the statistical power to observe patterns of association between laminar connectivity and EEG power. Since the analyses as carried out here have not previously been performed, we had no estimate of effect size. Secondly, by averaging over region combinations we drastically limit the multiple comparisons problem, since the number of comparisons scales with the square of the number of regions connectivity is computed between. Third, by averaging over regions, we target more general effects of connectivity between and within regions that are more likely to correspond to patterns observed within other contexts and other modalities. The effect for individual region combinations would likely be more variable.

    For completeness in the first submission we did include the results for every single region combination in the supplementary material (see Supporting Figures S2-S5). We have now included in the main document the results for region combinations V1-V2,V1-V3 and V2-V3 for between region connectivity, and V1-V1, V2-V2 &V3-V3 for within region connectivity, presented alongside the results for the grand average.

    The results for the individual region-pairs suggest that inter- and intra-region connectivity are generally consistent with the average over individual region combinations, but also have unique features.

    Similarities include: A strong negative correlation between beta power and deep-to-deep layer coupling was observed for average inter-regional connectivity. In line with this, for all three individual region pairs (V1-V2,V1-V3,V2-V3) a negative correlation is observed for deep-to-deep layer coupling. Similar patterns can be observed from alpha and beta for intra regionalconnectivity (averaged over all regions) and connectivity within V1,V2 and V3 in isolation.

    Individual features include: The relation between beta and inter-regional coupling shows variation over the individual region-pairs. In particular for V2-V3 connectivity, but also for V1-V2 the relation seems to differ from the pattern observed on average. For V2-V3, deep layer V3 seems to be coupled to both deep and superficial layers in V2, a pattern that might reflect anatomical feedback projections that go from deep layer V3 to both deep and superficial layers in V2.The stronger correlation between deep V1 and more middle deep V2 is however harder to directly place, since direct anatomical connections here are largely absent here. It might therefore reflect an indirect effect.

    Despite some degree of individual variation we think the overall picture is largely consistent. The strongest features present in the averaged results can clearly be observed in each of the individual region-combinations as opposed to the latter being a collection of vastly different random patterns that happen to add up to the average result (see for example the intraregional alpha results).

    With respect to our classification of regions into higher and lower level cortical regions, we based on standard anatomical hierarchies like that of van Felleman & van Essen (Felleman and Van Essen, 1991). Here, V1, V2 and V3 are ordered from low to higher in the visual cortical hierarchy.

    Methods. Given the missing conceptual overview over how attention-induced changes in EEG frequency bands should influence laminar connectivity in the visual system, also the methods lack a clear analyses strategy. The authors computed one correlation between power level of different frequency bands and connectivity between different brain areas without providing an explanation of which question this analysis addresses. The offered results therefore seem random to me, without a clear relationship to an investigated hypothesis.

    The primary focus of our study was to investigate how oscillations across several frequency bands in the EEG relate to laminar specific activity. Recent publications on laminar fMRI have demonstrated the possibility of performing laminar level fMRI connectivity analyses (Sharoh et al., 2019; Huber et al., 2017; Huber et al., 2020), which led us to revisit our previously recorded data in order to explore whether not only laminar specific BOLD amplitude but also laminar fMRI connectivity relates to frequency specific EEG power. Since laminar fMRI and especially connectivity measures derived from it are very novel, we started this analysis without a preconceived model or notion on how this relation would be. The results from this project should therefore be interpreted as an exploration of how these laminar fMRI derived connectivity measures relate to neural oscillations rather than directly addressing a specific cognitive process like selective attention, or prediction and/or a model of how neural oscillations play a role in these processes. Our experimental paradigm was also not designed to address such processes and test hypotheses derived from these. The primary focus of the work presented here is to provide a first insight in how neural oscillations measured by electrophysiological measures relate to cortical depth resolved fMRI coupling, which is usually correlation based. We believe these results will be relevant for research focused on how neural oscillations relate to inter-and intra regional interactions (e.g. (Bastos et al., 2012)(Fries, 2015)), since depth resolved fMRI allows us to study laminar interactions within and between brain regions non-invasively in humans. For this it is important to know if and how neural oscillations relate to laminar fMRI based connectivity measures, of which our research here provides a first insight. It also provides insight into which neural processes underlie observed changes in laminar fMRI based coupling, and is therefore relevant for research using such methods in general.

    Methods. The authors mention that they only analyzed the strongest two connecting vertices within a layer, which was done to improve SNR. In my view, for a connectivity analyses, this is not valid, as it can bias the effect towards superficial connectivity where the SNR and thus correlation is always higher.

    We did not analyze vertex pairs within a layer. We computed vertex pairs that connect the boundary between gray matter and CSF with the boundary of gray matter and white matter based on a high resolution anatomical MRI scan. Between these vertices we sampled 21 points of functional fMRI data using nearest neighbour interpolation. Since not all parts of V1, V2 and V3 will be involved in the task, we selected the most activated vertex pairs for further analysis. This serves as a localizer to select the parts within a region where task related activation is observed. For the main analysis the top 10% activated vertex pairs were chosen based on data collapsed across all depths and all attention conditions. This selection is therefore independent of depth, task condition, and the relation with any EEG feature. For this procedure we actually excluded the top five depth bins to avoid being too biased to superficial depths since it is known that signal to noise is substantially better near the surface of the cortex in part due to larger pial veins. To investigate whether the observed results are not due to this arbitrary threshold of 10%, we repeated the analyses for top 5% and 25% activated vertex pairs, the results of which are included in the supplementary information.

    Methods. The authors report 21 correlations in cortical depth, where their resolution allows to only sample perhaps 2-3 data points. The correlation analyses are therefore oversampled, which influences the statistical results. I would suggest to first run a component analyses across cortical depth, and to then correlate independent components to one another to investigate independent data points.

    The correlations are not oversampled, since the correlations used for the connectivity analyses are over trials, and not over space. These analyses are not influenced by the number of laminar BOLD data points we sample. Furthermore, spatial supersampling is a very common practice in FMRI research. For instance, the default in SPM is to upsample 3 mm isotropic standard voxel (very common for initial acquisition) size to 2 mm isotropic voxel size. In laminar fMRI laminar signals are often upsampled up to several factors above the the original resolution. This is for a number of reasons, well outlined on the laminar fMRI community website, a resource maintained by L. Huber in collaboration with many layer fMRI labs (see: https://layerfmri.com/2019/02/22/how-many-layers-should-i-reconstruct/) and ~20 layers is thought to be optimal.

    For our statistical test we explicitly chose a non-parametric cluster based technique to correct for multiple comparisons that takes dependencies across space into account. Laminar fMRI data are not well suited to decompose into components using techniques like PCA and ICA, since they violate assumptions of orthogonality/independence of the underlying responses in both the spatial as well as the temporal dimensions. To illustrate: in a recent laminar connectivity methods review an hierarchical, iterative ICA approach resulted in data being split up in columnar maps rather than laminar ones (Huber et al., 2020).

    Methods. The authors refer to their previously published paper with respect to the methods, and do not give any speficiations on the image sequence, image resolution, and image processing in this paper. In my view, all basic methodological steps that are critical to understand the paper should be described here.

    We are willing to include all relevant parts of the methodology described in our previous paper. This would involve copying large parts of the methods section, and might have to be coordinated with the publisher of the previous publication for copyright reasons. We would be pleased if the editor could advise us on this issue.

    Results. The figure captions are too short and do not explain the presented data in an appropriate way. In Figure 1, details on the calculated contrasts, number of participants investigated, sampling and analyses methods should be given that allows interpreting the data. Also, it would be beneficial to explain the attention paradigm in a bit more detail in the figure caption so that panel A can be interpreted. In Figure 3, more details should be given on what data are shown, particularly for panel C where the only information given is "attention effect on laminar connectivity" with no further axes labels.

    We extended the figure captions in the revised article.

    Results. I do not fully understand the results as shown in Figure 3. As those form the major part of the manuscript, this needs revision. As said before, I think that the figure and results section would benefit from region-specific data analyses and presentation, but also clear axes labels are needed to allow interpretation of the data. Also, when I interpret the data correctly, correlations are done for altogether 21 different cortical depth, which would not be valid because of artificially inflating the number of correlations, as pointed out above.

    We have extended our analyses and now split original Figure 3 up into current Figures 3 and 4 where we separately depict the results for intra- and inter-regional connectivity. For both intraand inter regional connectivity we have now also shown the region-specific results that underlie these results. We updated the figures and captions to make clearer what is depicted. We addressed the point raised about the 21 data points above. It is not relevant for the analysis presented here.

    Reviewer #3 (Public Review):

    However, a weakness of the technique as currently presented is that patterns of connectivity are only related to oscillations across subjects. It would be more powerful to examine whether the current network state (estimated by trial-by-trial power estimates) relates to laminar connectivity within subject. This would indeed speak to the nature of neuronal communication, which takes place on a moment-to-moment time scale, and which is not reflected in the current analysis. This may explain why laminar patterns of fMRI connectivity were not found to correlate with gammaband oscillatory activity. In addition, the negative effects of attention on fMRI connectivity itself are somewhat puzzling. This may related to the limitations of the task design which do not perfectly separate attention vs. arousal/expectation, as the authors readily discuss.

    The reviewer suggests that a relationship between fMRI connectivity and EEG power within subjects over trials would be more indicative of a direct link between connectivity and neural communication. We agree that establishing such a link would further strengthen the link between neural oscillations and laminar connectivity. This would not be trivial however, since connectivity in (laminar) fMRI is typically expressed as a measure of linear association (e.g. correlation or regression slope) over trials or time. Even at conventional spatial resolution, single trial/time point estimates of the network state are rarely used. These single data-point measures usually indicate to what extent a single data point contributes to the measure over all data points. We did not opt for such an analysis, since such analyses in normal (e.g. resting state) fMRI studies are uncommon, introducing more complexity to a study that already includes considerable novel analytic approaches. Furthermore, research relating fMRI activation and connectivity across subjects with other variables(e.g. clinical test scores, DTI measures, personality traits) is a well established procedure. Here we followed this more common approach.

  2. eLife

    Evaluation Summary

    This study uses non-invasive imaging to look at the coupling within and between layers and regions of the human visual cortex during the modulation of attention. The results presented here are a re-analysis of a previously recorded dataset, but the novelty is the analytic technique used to relate laminar connectivity to rhythms. This in principle promises to advance the field of both oscillations and laminar fMRI and could deliver valuable insights. The work provides a non-invasive window on how feedback and feedforward circuitry in the human brain operates. We deem the work of potential interest to a broad audience as it aims to provide direct links between the animal invasive electrophysiology and human neuroimaging fields. However, in its current form, major reservations with respect to the hypothesis space being explored here as well as important analytic and technical caveats remain.

    (This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #1 and Reviewer #3 agreed to share their names with the authors.)

  3. eLife

    Reviewer #1 (Public Review):

    In this study Scheeringa and colleagues make use of an attention manipulation to examine changes in BOLD connectivity between and regions and layers of the human visual cortex. The modulation of this connectivity (over subjects) with attention is then correlated with EEG changes in alpha, beta and gamma bands. The main findings are that decreases in beta power were associated with increasing coupling between deep layers in separate regions and from deep to superficial layers within regions. Within-region deep to superficial modulation of connectivity was also linked to changes in alpha band power.

    Strengths:

    This is an ambitious study which pushes the limits of what is possible with non-invasive imaging. It combines laminar- resolution fMRI with concurrent EEG recordings. There is a huge potential here and extensions of this method promise a bright future for high spatial and temporal resolution of brain imaging data using concurrent EEG and fMRI.

    The BOLD laminar connectivity analysis is impressive and compelling and the electrophysiological data is remarkably clear given the challenging recording conditions.

    The article raises interesting questions about the relationship between the electrophysiological and BOLD signals. Surprisingly, whilst the beta and alpha band signals covary negatively with BOLD generally, connectivity metrics in these bands seems to be linked with BOLD coupling changes of opposite sign.

    Weaknesses:

    Although the BOLD data is highly spatially specific, there is just one electrophysiological time-series per subject. This is no doubt a bi-product of the extensive noise cancellation that is necessary to record within the scanner. The caveat therefore is that the covarying BOLD and electrophysiological changes may derive from different regions.

    The analysis methods are slightly non-standard, perhaps for good reason. The main thing that stands out is the use of correlation coefficients, rather than regression coefficients, at the first level of analysis. This could potentially conflate changes in signal with changes in noise or unexplained variance.

  4. eLife

    Reviewer #2 (Public Review):

    Major concerns

    - Introduction. The introduction provides an overoptimistic view on the current possibilities with respect to the investigation of layer-specific activation or connectivity in the living human brain. Cortical layers cannot yet be segmented, the fMRI measures only provide an indirect signal that is heavily influenced by partial voluming between cortical depths, and EEG and MEG approaches often only measure two compartments due to low spatial resolution. The introduction, however, gives the impression that layer-specific neuronal connectivity can precisely be measured in the living human brain, which is not the case. The authors should take considerably more care with respect to how they introduce the methodology with clear references to the limitations. Also, statements such as "laminar fMRI allows us to study connectivity.." should be removed. In the same vein, I would suggest to replace laminar fMRI and laminar connectivity with cortical depth-dependent fMRI and connectivity to account for the above mentioned aspects.

    - Concept. Whereas the authors provide a model in the introduction that specifies how different frequency bands could relate to cortical depth-dependent connectivity, they do not develop a working hypotheses based on their experimental design. One conceptual step is therefore missing in the introduction, which has to combine present knowledge on the relationship between different frequency bands and present knowledge on how attention influences frequency-specific activation in the visual system to then make statements about which analyses can be performed to test which aspect of the model.

    - Concept & Methods: With respect to both the concept and the analyses, what is missing is taking into considerations the brain areas that were investigated. Wheres in the abstract the authors only mention "within brain region connectivity" and "between brain region connectivity" also in the Methods section there is no clear relation to the anatomical areas that were investigated, being V1, V2 and V3. The authors rather classify the areas as "high level" and "low level" where V2 is sometimes classified as high-level and sometimes as low-level. The data are therefore not investigated with reference to the anatomy of the visual system. In my view, it would be beneficial if all analyses could be performed with respect specifically to V1-V2 connecitivity and V2-V3 connectivity as well as V1-V3 connectivity so that the specific anatomical interrelations are taken into account. Also, the authors should develop a conceptual framework of how layer-specific attention-driven connectivity changes should influence the visual cortex, and why.

    - Methods. Given the missing conceptual overview over how attention-induced changes in EEG frequency bands should influence laminar connectivity in the visual system, also the methods lack a clear analyses strategy. The authors computed one correlation between power level of different frequency bands and connectivity between different brain areas without providing an explanation of which question this analysis addresses. The offered results therefore seem random to me, without a clear relationship to an investigated hypothesis.

    - Methods. The authors mention that they only analyzed the strongest two connecting vertices within a layer, which was done to improve SNR. In my view, for a connectivity analyses, this is not valid, as it can bias the effect towards superficial connectivity where the SNR and thus correlation is always higher.

    - Methods. The authors report 21 correlations in cortical depth, where their resolution allows to only sample perhaps 2-3 data points. The correlation analyses are therefore oversampled, which influences the statistical results. I would suggest to first run a component analyses across cortical depth, and to then correlate independent components to one another to investigate independent data points.

    - Methods. The authors refer to their previously published paper with respect to the methods, and do not give any speficiations on the image sequence, image resolution, and image processing in this paper. In my view, all basic methodological steps that are critical to understand the paper should be described here.

    - Results. The figure captions are too short and do not explain the presented data in an appropriate way. In Figure 1, details on the calculated contrasts, number of participants investigated, sampling and analyses methods should be given that allows interpreting the data. Also, it would be beneficial to explain the attention paradigm in a bit more detail in the figure caption so that panel A can be interpreted. In Figure 3, more details should be given on what data are shown, particularly for panel C where the only information given is "attention effect on laminar connectivity" with no further axes labels.

    - Results. I do not fully understand the results as shown in Figure 3. As those form the major part of the manuscript, this needs revision. As said before, I think that the figure and results section would benefit from region-specific data analyses and presentation, but also clear axes labels are needed to allow interpretation of the data. Also, when I interpret the data correctly, correlations are done for altogether 21 different cortical depth, which would not be valid because of artificially inflating the number of correlations, as pointed out above.

  5. eLife

    Reviewer #3 (Public Review):

    Scheeringa et al. examine laminar connectivity using fMRI and its relationship to neuronal oscillations. The results presented here are a re-analysis of a previously recorded dataset. The novelty here is the analytic technique used to relate laminar connectivity to rhythms. The main result, which is convincingly demonstrated, is that task modulations of deep-to-deep layer connectivity between areas is related to beta power. This is interesting because other studies (including studies by the authors) have found a negative relationship between fMRI BOLD and beta power within an area. In addition, studies in animals have found something similar: that while beta can decrease in a given area that is involved in a task, it can simultaneously increase in synchronization between areas. The present observation provides a link between these animal studies and non-invasive studies in humans. In addition, this work provides an important bridge between laminar connectivity and oscillations, previously only accessible with invasive recordings. Therefore, in the present study Scheeringa et al. push the field of neuroimaging into an exciting, more mechanistic direction, which promises insights into how cognition modulates laminar circuits.

    However, a weakness of the technique as currently presented is that patterns of connectivity are only related to oscillations across subjects. It would be more powerful to examine whether the current network state (estimated by trial-by-trial power estimates) relates to laminar connectivity within subject. This would indeed speak to the nature of neuronal communication, which takes place on a moment-to-moment time scale, and which is not reflected in the current analysis. This may explain why laminar patterns of fMRI connectivity were not found to correlate with gamma-band oscillatory activity. In addition, the negative effects of attention on fMRI connectivity itself are somewhat puzzling. This may related to the limitations of the task design which do not perfectly separate attention vs. arousal/expectation, as the authors readily discuss.

  6. Version published to 10.1101/2020.09.18.303263 on bioRxiv