Article activity feed

  1. Author Response

    Reviewer #1 (Public Review):

    In the present study, the authors first analyzed simultaneously recorded human EEG-fMRI data and found the fMRI signatures of burst-suppression. Then, they reported such burst-suppression fMRI signatures in the other three species examined: macaques, marmosets, and rats. Interestingly, their results indicated an inter-species difference: the entire neocortex engaged in burst-suppression in rats, whereas most of the sensory cortices were excluded in primates. The fMRI signatures of burst-suppression were confirmed in several species, suggesting that such signature is a robust phenomenon across animals. These findings warrant further investigation into its neural mechanisms and functional implications.

    Major Issues

    1. One of the major findings is that burst-suppression in primates appeared to largely spare sensory cortices, especially V1. However, as seen in the tSNR map for macaques and marmosets (Figure 3 &4 -figure supplement 4), the tSNR around the primary visual cortex was much weaker than other cortices. Moreover, in marmosets, the EPI slices did not cover the entire brain and actually left most of the V1 uncovered as seen in Figure 4. If so, the authors should draw their conclusions very carefully when talking about the differences in V1 across species. It would be better to analyze and discuss how the tSNR differences affect their findings. For example, the author may consider including the tSNR as covariance in their map analysis.

    The tSNR in the occipital cortex—especially in the macaque V1—is indeed lower than in more anterior parts of the brain. The higher noise in V1 may have obscured the burst-suppression signal and hindered its detection. That said, we think that burst-suppression would still be detectable at such low tSNR values. We base this claim on our analysis of another macaque brain region—area TE of the inferior temporal cortex (see our additions to Figure 3–figure supplement 4). The tSNR in areas TE and V1 is comparably low, and yet TE is significantly correlated with asymmetric PCs while V1 is not. Therefore, if the burst-suppression fluctuation was present in V1 we should have still detected it.

    Regarding the marmoset data, part of V1 was indeed left out of our field of view, as explicitly shown in our figures (Figure 4 and Figure 4–figure supplement 3). Though we cannot exclude the possibility that the omitted posterior V1 engages in burst-suppression, we think that it is unlikely to behave any differently to more anterior visual areas. We sought more support for this view by obtaining full-brain fMRI data in one additional marmoset. We present this analysis in a new paragraph of the relevant Results section and in the new Figure 4–figure supplement 5. The asymmetric PC map in this individual showed widespread correlation across the neocortex, extending slightly further caudally compared with the group map presented in Figure 4. However, nearly all of V1—including the occipital pole—was still uncorrelated. Considering both the new full-brain marmoset data and the results from area TE in macaques, we think that our conclusion about the uncoupling of primate V1 during burst-suppression is still justified. That said, we have now explicitly included the relevant concerns in the manuscript text.

    1. To confirm their findings, it would be great to look into the EEG signals around the sensory cortex (e.g., V1) to see whether the findings in fMRI could be also confirmed with EEG.

    EEG signals around V1 were already examined during the previous analysis of the human dataset (Golkowski et al., 2017). As reported there, the EEG signal of the occipital electrodes did contain bursts, which could not be differentiated from bursts detected by more anterior electrodes in terms of onset timing, duration, or spectral content. This might mean that the BOLD signal in VI is truly uncoupled from electrical activity. However, we should also consider that EEG may lack the spatial resolution to detect a different activity originating from V1. As seen in the human map (Figure 3), the external cortical surface is almost exclusively covered with areas engaging in burst-suppression, whereas the ‘uncoupled’ V1 represents a small patch by comparison. Therefore, EEG cannot safely determine the nature of electrical activity in V1. We have added the above arguments to the last section of Results. We expect a conclusive answer to come from future electrophysiological recordings in nonhuman primates. The larger proportional size of visual areas in macaques and marmosets as well as the possibility of invasive intra-cranial recordings make these animals attractive models for addressing this question.

    1. As seen in Figure 2-figure supplement 2, there was a significant anticorrelation with burst-suppression at the ventricular borders. It is unclear whether the authors have done physiological or white matter/CSF/global nuisance regression as most of the rest-fMRI studies did. Please make it clear. If not, please explain why and discuss whether it would affect their results.

    We chose to analyze the data without CSF or global signal regression. CSF regression typically requires extracting the signal of a few voxels within the ventricles. Accurately placing such voxels is feasible in the human brain but challenging in small animal brains, especially in rodents. Rodent ventricles are very thin, making it difficult to place a CSF voxel that will not overlap with surrounding brain tissue. Since we had prioritized making the analysis as similar as possible across species, we decided to also forgo CSF regression in humans. While this was our original motivation for omitting CSF regression, we later came across an even more important concern. As we show in Figure 2–figure supplement 2, the CSF signal is not ‘noise’; rather, it is directly related to burst-suppression, and most likely caused by it. Regressing it out would remove much of the variance explained by burst suppression. The coherence between neural, hemodynamic, and CSF oscillations that we see in burst-suppression likely also occurs in other states characterized by global synchrony, as has been shown for non-rapid eye movement sleep (Fultz et al., 2019).

    We think that global signal regression makes no sense in our case, given that our goal was to study a nearly global signal fluctuation. Global signal regression relies on the assumption that neuronal activity is variable across brain regions while many non-neuronal sources contribute globally to the brain signal (Murphy and Fox, 2017). This assumption does not hold true in cases where the neuronal activity itself is global.

    1. Three different concentrations of the anesthetic sevoflurane were chosen for human participants. The authors found that the high concentration (3.9-4.6%) induced burst-suppression much better than the other two lower concentrations as expected. However, in rats, almost all asymmetric PCs were found at an intermediate concentration (2%) of isoflurane less at the low (1.5%) or high (2.5%) concentration in Rat 1. At the same time, all fMRI runs from Rat 2 with a 1.3% concentration of isoflurane had a prominent asymmetric PC. That is, it seems that only the high concentration of isoflurane could not induce burst-suppression well in rats, which was opposite to those findings in humans. The authors may explain what reasons may cause such differences and whether such differences may affect the major findings in differences between primates and rodents.

    The three sevoflurane concentrations (‘high’, ‘intermediate’, ‘low’) used in humans do not necessarily correspond to the three isoflurane concentrations used in rats (2.5%, 2.0%, 1.5%). Comparing anesthetic concentrations across our datasets is challenging, since anesthetic potency is expected to vary depending on the drug (sevoflurane or isoflurane), animal species, age, and the co-administration of other drugs. Nevertheless, we may estimate equivalent concentrations across species by expressing them as multiples of the minimum alveolar concentration (MAC), i.e. the concentration that produces immobility in 50% of subjects undergoing a standard surgical stimulus.

    For humans, we can use available age-related MAC charts (Nickalls and Mapleson, 2003) to express the three sevoflurane levels as follows: ~1 MAC (2%), 1.5 MAC (3%), 2.2–2.3 MAC (3.9–4.6%). For rats, we can rely on the previously reported isoflurane MAC value of 1.35% (Criado et al., 2000) to derive the following levels: 1.2 MAC (1.5%), 1.6 MAC (2%), 1.9 MAC (2.5 %), and ~1 MAC (1.3%, Rat 2 dataset). According to these conversions, fMRI-detectable burst-suppression occurred in humans at ~2 MAC (with some cases at 1.5 MAC), in the Rat 1 dataset at 1.2–1.6 MAC, and in the Rat 2 dataset at 1 MAC. There seems to be a difference between rats and humans as well as a discrepancy between the two rat datasets. The latter discrepancy could have arisen from differences in the calibration of isoflurane vaporizers at the two research sites (direct measurements of end-tidal anesthetic concentration were not obtained in rats).

    In order to better interpret the observed human-rat difference we tried to also compute the multiples of MAC values for our nonhuman primate data, but this proved to be hard. For common marmosets, we are not aware of any published isoflurane MAC values. For long-tailed macaques, a value of 1.28% has been reported (Tinker et al., 1977), which gives a range of 0.7 – 1.2 MAC for our macaque dataset. However, that probably underestimates the actual depth of anesthesia in our experiments, since many of our macaques were old and MAC is known to decrease with age (Nickalls and Mapleson, 2003). Moreover, the administration of medetomidine during anesthesia induction may have further reduced the MAC (Ewing et al., 1993). Consequently, we cannot provide good MAC estimates for the nonhuman primate data and thus have no reference for comparison with other species.

    Even if we knew the correct MAC value in all cases, it may be an inappropriate means of standardizing anesthetic concentrations for burst-suppression. The endpoint measured by MAC—immobility—is mainly mediated by anesthetic effects on the spinal cord and my not be a good predictor for effects on the brain (Rampil et al., 1993). In fact, burst-suppression itself has been proposed as a more appropriate endpoint for measuring anesthetic potency. The proposed metric (MACBS) is defined as the concentration that produces suppressions longer than 1 s in 50% of subjects and is not linearly related to MAC (Pilge et al., 2014).

    In conclusion, if we reference anesthetic concentrations against the MAC, humans and rats indeed seem to exhibit burst-suppression at different concentration ranges. We are unable to perform the same referencing for non-human primates, due to lack of accurate MAC values. Moreover, it is unclear whether MAC is the appropriate reference to begin with. Discussing all these nuances would make the manuscript too long. That said, we have now added a new paragraph to the Discussion section, drawing attention to the fact that anesthetic concentrations are not standardized across species.

    Reviewer #2 (Public Review):

    The strong point in their manuscript is the originality of their results. Using the fMRI's spatial resolution, they can successfully reveal that not all brain areas are synchronized during the burst suppression. Furthermore, they can find that the difference is the most obvious when comparing primates with rats, which makes sense considering the distance on the phylogenetic tree. As far as I know, this manuscript first reports these points.

    On the other hand, there is a weak point in their method. As they've already discussed this point, they needed to use arbitrary thresholds to evaluate whether there is burst suppression or not. Furthermore, this study cannot reject the possibility of spatial inhomogeneity and/or anesthesia-specific modulation in hemodynamic response. If there is such a mechanism, one can find different results from those obtained through electrical measurements.

    1. The authors found that some sensory areas in primates are excluded from those highly synchronized during the burst suppression. While it is true, I wonder if each voxel in such areas shows burst suppression-like activity that is not synchronized with others. If this is the case, burst suppression can still be a global phenomenon. Though authors seem to investigate this point, they used in-ROI averaged time-series so that it cannot reject the possibility that each voxel inside the ROI is not synchronized but shows burst suppression in its manner. I recommend the authors look into each voxel if this is the case or not.

    The reviewer raises an interesting point by proposing that it is possible for sub-regions within the excluded areas—e.g. within V1—to exhibit burst-suppression out-of-phase with each other, thus cancelling out in the mean V1 BOLD signal. We do not think this is the case, for several reasons. Firstly, we can exclude the possibility that any part of V1 exhibits bust-suppression in-phase with the rest of the cortex. The original first-level GLM analysis was a voxel-based univariate analysis. If any voxels within V1 were correlated with the global burst-suppression pattern, we would have seen it on the maps. We saw no such effect, except for some subjects in which a subset of V1 voxels was anti-correlated with the asymmetric PC (the effect was not significant in our group analysis). This anticorrelation was mostly located close to the ventral horns of the two lateral ventricles, and thus could have arisen by the same cycle of ventricular shrinkage-expansion that we describe in Figure 2–figure supplement 2. Secondly, no large clusters of V1 voxels exhibited burst-suppression out-of-phase with the dominant asymmetric PC. If this was the case, we would have seen a phase-shifted version of the fluctuation on the carpet plots. This still leaves the theoretical possibility that individual V1 voxels (or a few at a time) exhibit transitions between burst and suppression epochs out-of-phase with each other. In our response to the next point, we will explain why there is no way of detecting this with fMRI and we discuss whether such a possibility would even fit the label of burst-suppression.

    1. The other but similar point is about their way to detect burst suppression. Why did they use the principal component? By definition, burst suppression should be defined by the existence of burst and suppressed periods. I cannot understand why they did not simply use this definition to check whether each voxel shows such an intermittent activity to evaluate whether it is a global phenomenon or not.

    Burst-suppression on EEG is characterized by quasi-periodic suppressions of activity, during which the EEG signal drops close to being isoelectric. We cannot apply the same definition to fMRI, because the BOLD signal only represents relative changes and thus has no natural baseline equivalent to isoelectricity. Hence there is no way of telling whether a BOLD signal decrease corresponds to a complete activity cessation (suppression) or simply a relative decline. Therefore, we instead decided to rely on another defining feature of burst-suppression—synchrony. We knew that burst-suppression appears simultaneously across EEG electrodes, which means that large parts of the cortex (the major contributor to EEG signal) would have to be synchronized. Moreover, we knew that transitions between burst and suppression epochs occur on a very slow timescale and would be resolvable at a TR of 2 s. PCA allowed us to isolate the large slow synchronous component in the cortical BOLD signal, though this is hardly the only approach that would work. We chose PCA because it is a simple, deterministic, and easily interpretable algorithm.

    On a related note, even if we could identify complete cessation of activity in the BOLD signal of a single voxel, it is unclear whether that would qualify as burst-suppression per the EEG definition. EEG electrodes pick up activity from areas much larger than a voxel, and thus the very presence of an EEG fluctuation presupposes synchrony on a larger spatial scale. If individual voxel-sized brain areas engaged in burst-suppression out-of-phase, that would probably not register as burst-suppression on an EEG electrode.

    1. Why is there no synchronization during the slow-wave states under light anesthesia? During the slow-wave sleep, it is shown that the entire cortical network is decomposed into a modular-like network structure. Is there synchronization inside each module while no synchrony between modules?

    We do not claim that there is no synchrony in the slow-wave state. We simply state that this state lacks the nearly global cortex-wide fluctuation that is produced by the abrupt transitions between burst and suppression epochs. In fact, the very presence of slow waves on EEG requires synchrony. However, this slow-wave synchrony occurs at a timescale too fast for fMRI to capture, and thus would not directly translate into a global BOLD fluctuation, as burst-suppression does.

    Though the slow-wave state lacks global synchrony on fMRI, it may well exhibit within-module synchrony, as the reviewer suggests. Modules resembling the resting-state networks of wakefulness and sleep have been detected during isoflurane anesthesia in primates (Hori et al., 2020; Hutchison et al., 2011). These experiments were presumably conducted during the slow-wave state: burst-suppression would generate a global network, while the isoelectric state would erase any modular structure. We suspect that functional networks during the anesthetized slow-wave state resemble those present in slow-wave sleep. However, we have not assessed that in our study, since our primary goal was to map burst-suppression.

    Reviewer #3 (Public Review):

    The authors present a multicenter, multimodal rs-fMRI study of the spatial signature of burst suppression in the brain of humans, non-human primates and rats. They have used EEG to identify burst suppression activity in human data from simultaneous EEG-rs-fMRI measurements of subjects under servoflurane anesthesia. After having identified a (neurovascular) rs-fMRI representation of burst activity, the authors show that bursts can equally be identified from MR data alone. After a principal component analysis, bursts and their spatial signature were identified by an asymmetry of the correlation coefficients. Across species the authors identified similar spatial signatures, which were conserved for all (investigated) primates, but differed for rats. While rats showed a pan-cortical involvement, signatures in primates were more complex, e.g., not including the visual cortex.

    In this study, the authors have presented a novel purely MR-based method to identify burst suppression and its spatial signature. Their method may be used to readily identify burst suppression in fMRI data. However, no general threshold for the median of the cortex-wide correlation could be identified. The authors also establish a conserved signature of burst suppression for primates and reveal subtle but important differences to rodents. Both achievements are novel and represent a major advance in the field of neuroimaging.

    The study was well designed, including important control data to rule out artefacts as source of the observed burst suppression patterns. The particular strengths of this study are: (1) including multicentre data (although only rats were scanned at two different sites); and (2) including four species from humans to rats.

    The manuscript was very carefully and well written (I did not even notice a single typo) and the figures were carefully devised, comprehensively illustrating the large amount of data. The authors further provide a comprehensive account of the relevant literature. Towards the end of their discussion they also clarify the difference in terminology used for burst suppression in some recent rodent studies.

    The only (and in my opinion notable) weakness, is the lack of a general threshold for the asymmetry of the median of the cortex-wide correlation coefficients. With such a threshold, rs-fMRI could be readily used to automatically detect burst suppression across species. However, the authors clearly state this shortcoming and openly discuss its implications. I do not think that an altered experimental design or additional data could provide further remedy.

    To conclude: This very comprehensive study was very well designed, extremely carefully performed, presents a novel tool for identification of burst suppression, and provides insight across species. It has clearly translational potential, which however, is limited by the lack of a general threshold for burst suppression detection.

    I congratulate the authors for this very nice piece of work, and the most typo-free manuscript I have ever read.

    We thank the reviewer for the positive and detailed feedback.

    Read the original source
    Was this evaluation helpful?
  2. Evaluation Summary:

    This study reveals that anesthesia-induced burst suppression's spatial patterns differ across humans, macaques, marmosets, and rats. Given that burst suppression is considered a hallmark of unconscious states, these findings are potentially important for us to understand the evolution of the neural correlates of consciousness. In addition, a novel, purely MR-based method is presented to identify and map burst suppression, which may have relevance in both clinical and experimental studies.

    (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 #2 and Reviewer #3 agreed to share their name with the authors.)

    Read the original source
    Was this evaluation helpful?
  3. Reviewer #1 (Public Review):

    In the present study, the authors first analyzed simultaneously recorded human EEG-fMRI data and found the fMRI signatures of burst-suppression. Then, they reported such burst-suppression fMRI signatures in the other three species examined: macaques, marmosets, and rats. Interestingly, their results indicated an inter-species difference: the entire neocortex engaged in burst-suppression in rats, whereas most of the sensory cortices were excluded in primates. The fMRI signatures of burst-suppression were confirmed in several species, suggesting that such signature is a robust phenomenon across animals. These findings warrant further investigation into its neural mechanisms and functional implications.

    Major Issues:

    1. One of the major findings is that burst-suppression in primates appeared to largely spare sensory cortices, especially V1. However, as seen in the tSNR map for macaques and marmosets (Figure 3 &4 -figure supplement 4), the tSNR around the primary visual cortex was much weaker than other cortices. Moreover, in marmosets, the EPI slices did not cover the entire brain and actually left most of the V1 uncovered as seen in Figure 4. If so, the authors should draw their conclusions very carefully when talking about the differences in V1 across species. It would be better to analyze and discuss how the tSNR differences affect their findings. For example, the author may consider including the tSNR as covariance in their map analysis.

    2. To confirm their findings, it would be great to look into the EEG signals around the sensory cortex (e.g., V1) to see whether the findings in fMRI could be also confirmed with EEG.

    3. As seen in Figure 2-figure supplement 2, there was a significant anticorrelation with burst-suppression at the ventricular borders. It is unclear whether the authors have done physiological or white matter/CSF/global nuisance regression as most of the rest-fMRI studies did. Please make it clear. If not, please explain why and discuss whether it would affect their results.

    4. Three different concentrations of the anesthetic sevoflurane were chosen for human participants. The authors found that the high concentration (3.9-4.6%) induced burst-suppression much better than the other two lower concentrations as expected. However, in rats, almost all asymmetric PCs were found at an intermediate concentration (2%) of isoflurane less at the low (1.5%) or high (2.5%) concentration in Rat 1. At the same time, all fMRI runs from Rat 2 with a 1.3% concentration of isoflurane had a prominent asymmetric PC. That is, it seems that only the high concentration of isoflurane could not induce burst-suppression well in rats, which was opposite to those findings in humans. The authors may explain what reasons may cause such differences and whether such differences may affect the major findings in differences between primates and rodents.

    Read the original source
    Was this evaluation helpful?
  4. Reviewer #2 (Public Review):

    The strong point in their manuscript is the originality of their results. Using the fMRI's spatial resolution, they can successfully reveal that not all brain areas are synchronized during the burst suppression. Furthermore, they can find that the difference is the most obvious when comparing primates with rats, which makes sense considering the distance on the phylogenetic tree. As far as I know, this manuscript first reports these points.

    On the other hand, there is a weak point in their method. As they've already discussed this point, they needed to use arbitrary thresholds to evaluate whether there is burst suppression or not. Furthermore, this study cannot reject the possibility of spatial inhomogeneity and/or anesthesia-specific modulation in hemodynamic response. If there is such a mechanism, one can find different results from those obtained through electrical measurements.

    1. The authors found that some sensory areas in primates are excluded from those highly synchronized during the burst suppression. While it is true, I wonder if each voxel in such areas shows burst suppression-like activity that is not synchronized with others. If this is the case, burst suppression can still be a global phenomenon. Though authors seem to investigate this point, they used in-ROI averaged time-series so that it cannot reject the possibility that each voxel inside the ROI is not synchronized but shows burst suppression in its manner. I recommend the authors look into each voxel if this is the case or not.

    2. The other but similar point is about their way to detect burst suppression. Why did they use the principal component? By definition, burst suppression should be defined by the existence of burst and suppressed periods. I cannot understand why they did not simply use this definition to check whether each voxel shows such an intermittent activity to evaluate whether it is a global phenomenon or not.

    3. Why is there no synchronization during the slow-wave states under light anesthesia? During the slow-wave sleep, it is shown that the entire cortical network is decomposed into a modular-like network structure. Is there synchronization inside each module while no synchrony between modules?

    Read the original source
    Was this evaluation helpful?
  5. Reviewer #3 (Public Review):

    The authors present a multicenter, multimodal rs-fMRI study of the spatial signature of burst suppression in the brain of humans, non-human primates and rats. They have used EEG to identify burst suppression activity in human data from simultaneous EEG-rs-fMRI measurements of subjects under servoflurane anesthesia. After having identified a (neurovascular) rs-fMRI representation of burst activity, the authors show that bursts can equally be identified from MR data alone. After a principal component analysis, bursts and their spatial signature were identified by an asymmetry of the correlation coefficients. Across species the authors identified similar spatial signatures, which were conserved for all (investigated) primates, but differed for rats. While rats showed a pan-cortical involvement, signatures in primates were more complex, e.g., not including the visual cortex.

    In this study, the authors have presented a novel purely MR-based method to identify burst suppression and its spatial signature. Their method may be used to readily identify burst suppression in fMRI data. However, no general threshold for the median of the cortex-wide correlation could be identified. The authors also establish a conserved signature of burst suppression for primates and reveal subtle but important differences to rodents. Both achievements are novel and represent a major advance in the field of neuroimaging.

    The study was well designed, including important control data to rule out artefacts as source of the observed burst suppression patterns. The particular strengths of this study are: (1) including multicentre data (although only rats were scanned at two different sites); and (2) including four species from humans to rats.

    The manuscript was very carefully and well written (I did not even notice a single typo) and the figures were carefully devised, comprehensively illustrating the large amount of data. The authors further provide a comprehensive account of the relevant literature. Towards the end of their discussion they also clarify the difference in terminology used for burst suppression in some recent rodent studies.

    The only (and in my opinion notable) weakness, is the lack of a general threshold for the asymmetry of the median of the cortex-wide correlation coefficients. With such a threshold, rs-fMRI could be readily used to automatically detect burst suppression across species. However, the authors clearly state this shortcoming and openly discuss its implications. I do not think that an altered experimental design or additional data could provide further remedy.

    To conclude: This very comprehensive study was very well designed, extremely carefully performed, presents a novel tool for identification of burst suppression, and provides insight across species. It has clearly translational potential, which however, is limited by the lack of a general threshold for burst suppression detection.

    I congratulate the authors for this very nice piece of work, and the most typo-free manuscript I have ever read.

    Read the original source
    Was this evaluation helpful?