Article activity feed

  1. Author Response:

    Reviewer #1 (Public Review):

    [...] The first issue that may be a concern is the idea that MRS measures of Glutamate and GABA are appropriate proxy measures of the excitation-inhibition balance. While this is a not uncommon interpretation of the relative levels of these two neurotransmitters, a recent paper in NeuroImage presents data to refute this assertion (Rideaux et al 2021), especially in the visual cortex. However, The current study by Kooschijn et-al is different from the data reported by Rideaux, in that the current data is looking at the relative change in the Glu/GABA ratio (and the relative difference in each Glu and GABA by themselves) between two conditions, and not the overall "balance" at rest, or during activity. This dynamic nature of the data, and the temporal resolution present, may explain why a relationship is found between the change in ratio (and individual levels) and performance in this study. Either way, the authors could address this recent paper of Rideaux and the challenge it may present for their interpretation of Glu/GABA ratios as a measure of E-I balance.

    Thank you for raising this point and highlighting the findings from Rideaux Neuroimage 2021. As we discuss below, and as noted by the reviewer, the design of our experiment and analysis is different from Rideaux Neuroimage 2021, which may explain the difference in findings.

    The data presented in Rideaux 2021 considers the between-subject variance for measures of Glx (glutamate+glutamine) and GABA, averaged across time. The authors show no evidence for a correlation between GABA and Glx across participants. Using our data, we replicate this result when assessing the relationship between GABA and glutamate, quantified using all spectra (r17=0.19, p=0.433; after regressing out sex and age: r17=0.21 p=0.400). While this result is intriguing, comparing average measures of GABA and glutamate across subjects obscures temporal dynamics of these neurometabolites that may more closely relate to fluctuations in excitation and inhibition reported at the physiological level.

    Here, by using ultra-high field fMRI and fMRS, we designed our study to assess task-dependent temporal dynamics in neurometabolites. Thus, we acquired time-resolved, within-subject measures of glutamate and GABA which we compare across two conditions of interest (‘remembered’ and ‘forgotten’). This condition-dependent approach provides a means to assess subtle, task-specific changes in neurochemistry that cannot be observed when taking a bulk-average. Moreover, compared to taking the bulk-average measures used by Rideaux, our approach inherently controls for: (1) between-subject differences in average GABA and glutamate which are affected by demographic (e.g. age and sex); (2) between-subject differences in spectral quality; (3) between-subject differences in tissue composition; (4) between-subject differences in the effect of other neurochemicals on measures of glutamate and GABA. Overall, our time-resolved, within-subject and condition-dependent approach arguably provides a readout for fluctuations in neurometabolites that more closely approximate physiologically relevant shifts in excitation and inhibition.

    However, as noted in the Introduction and Discussion sections of our manuscript, the relationship between fMRS and physiological definitions of EI balance remains complex. The temporal resolution of fMRS remains several orders of magnitude slower than rapid changes in synaptic glutamate and GABA that accompany neurotransmitter release. Moreover, MRS fails to discriminate between different pools of glutamate and GABA and only a fraction reflects neurotransmitter release. Meaningful interpretation of MRS instead derives from the approximately 1:1 relationship between the rate of glutamine-glutamate cycling, which is necessary for glutamate and GABA synthesis, and neuronal oxidative glucose consumption, which indirectly supports neurotransmitter release among other processes (Rothman et al., 2003; Shen et al., 1999; Sibson et al., 1998). We therefore conclude that through careful experimental and analytical design, fMRS can provide a non-invasive marker for physiologically relevant shifts in excitation and inhibition, if indirect and at a coarse spatiotemporal scale. This interpretation nevertheless requires validation, by carefully combining preclinical MRI with invasive methods in animal models in future work.

    It is also worth noting that the relative nature of measures utilised may enhance error propagation between individual measures, increasing the risk of a false positive result. The authors have tried to address this through the use of Monte-Carlo permutation analysis for false errors, and it does go some way to restoring confidence.

    Our analytical approach involves assessing the ratio between glutamate and GABA (‘glu/GABA’) across two conditions (‘remembered’/‘forgotten’). As the reviewer notes, by considering this relative measure, the measured uncertainties in glutamate and GABA propagate through to the uncertainty in the functional relationship of interest (‘glu/GABA remembered’/‘glu/GABA forgotten’). In the revised manuscript we now show the full sampling-error curve for the effect size of the relative measure, where the 95% confidence interval is notably non-overlapping with zero (see newly added panel in Figure 4F showing mean, 95% confidence interval and the sampling-error distribution derived using bootstrapping). Thus, in addition to the Monte-Carlo permutations presented in Figure 4G, by presenting the sampling-error distribution we provide additional evidence to suggest our findings cannot be explained by a false positive.

    We also note that there are advantages to our approach, which we discuss above and reiterate here. Compared to analyses that consider individual measures, our approach controls for a number of factors that can otherwise introduce random errors. These factors include: (1) between-subject differences in average GABA and glutamate which are affected by demographic (e.g. age and sex); (2) between-subject differences in spectral quality; (3) between-subject differences in tissue composition; (4) between-subject differences in the effect of other neurochemicals on measures of glutamate and GABA. Finally, our approach is analogous to standard analytical methods employed for event-related fMRI.

    There are some other potential methodological caveats a reader inexperienced in fMRS (and indeed fMRI) should be aware of:

    • The first is that the MRS sequence utilised is not typically used to measure GABA (a point the authors also note), with GABA typically measured using so called "editing" techniques like MEGA-PRESS. The authors also utilise a non-standard unconstrained fit for GABA in their analysis of the MRS spectrum. While these two non-standard methodologies may weaken confidence the GABA measures, the authors are to be commended on the use of simulations to demonstrate that even if "absolute" measures of GABA via this methodology may be slightly outside the usual norms, this methodology is able to detect changes in GABA of the size of those detected in this experiment. These simulations, coupled with the Monte-Carlo permutation steps for FWE correction are a strength of the paper, and I encourage readers to fully examine the supplementary material for this paper to get a better appreciation for the quality and validity of the data being presented.

    Thank you for this point and for highlighting that the Monte-Carlo simulations are a strength of the paper. Below we summarise the rationale for using an unedited sequence, and detail the analysis pipeline that we implement to detect dynamic changes in glutamate and GABA.

    Edited and unedited acquisition protocols each have their pros and cons. To test our hypothesis concerning dynamic fluctuations in glutamate and GABA, we considered an unedited sequence to be advantageous for the following two reasons: (1) unedited sequences permit acquisition of metabolite data for both glutamate and GABA within the shortest possible timeframe, thus minimizing motion and drift artefacts; (2) unedited sequences allow acquisition of high-quality spectra from a comparatively small (8 ml) volume of interest (VOI), while an edited methods would have required a larger VOI. For these reasons unedited sequences may therefore be more suitable for event-related fMRS. In the Discussion we note that investigations comparing edited and non-edited sequences at 7T reveal no significant difference in the concentration of GABA measurements (Hong et al., 2019).

    Regarding our analysis pipeline, as noted by the reviewer, we do not implement default assumptions that are typically used to obtain static estimates of GABA, where metabolite values are constrained within a predefined (‘physiologically plausible’) range. Instead, we remove these constraints to optimise the analysis pipeline for detecting dynamic changes in GABA. To demonstrate the sensitivity of this approach, we use Monte Carlo simulations to generate MRS spectra while preserving the observed noise in our data. Our simulations show that the observed difference in GABA between ‘remembered’ and ‘forgotten’ conditions is significant from a null distribution that would be expected by chance. Moreover, these simulations show that relative to default settings, our analysis pipeline is more sensitive to detecting dynamic changes in GABA. For readers that may be interested in the more technical motivation behind our approach, we now include a supplementary note in Appendix 1.

    • The exact time locking (or not) of fMRS data acquisition to phases of the stimulus presentation and the subsequent temporal resolution and timeline of changes is not fully explained - and may be somewhat misleading. Given the MRS data were collected in step sizes of 4 secs, it may be hard to understand how a temporal resolution of 2.5 secs for the fMRS data is achieved. Likewise, given that the total Question and ITI time was allowed to vary from stimulus to stimulus, it may be hard to understand how the timeline in figures 4 and 5 are achieved. This could be ameliorated by a better explanation of data collection process in the methods, and how data was averaged to produce the timelines as presented.

    We have now edited the Results and Methods sections to clearly explain the temporal relationship between the fMRS acquisition and the task. We now also include a new Figure 4–figure supplement 1 to illustrate how we estimate a moving average for fMRS data at a higher resolution than the 4 s TR used during acquisition.

    • Lastly, the fMRI technique being used is not a typical echo-planar sequence, and the data produced are by the authors own admission impacted in quality. Given that the hippocampus, and other parts of the anterior temporal lobe are difficult to measure at the best of times, the BOLD signal changes from this data may be less robust than normal. As such, while interesting, the correlation between the Hippocampal BOLD signal and the Glu/GABA ratio changes might be considered tentative, and can only benefit from replication at a later data.

    As already noted in the manuscript, the quality of the fMRI component in the fMRI-fMRS sequence was compromised relative to EPI obtained using non-combined contemporary state-of-the-art fMRI sequences. However, we consider the BOLD effects reported in visual cortex and hippocampus to be robust as they replicate equivalent analyses conducted on a dataset previously acquired using a non-combined multiband EPI sequence on the same task (Barron et al., Cell 2020). This point is now noted in the revised Results and Discussion sections of the manuscript. Given the multiband EPI sequence in Barron et al., 2020 did not include MRS measurements, the reported correlation between hippocampal BOLD and glu/GABA ratio during successful vs unsuccessful recall can only be verified in future work.

    My expertise as a reviewer is usually in the methodological aspects of fMRS studies, and not really in the "psychology" or "cognitive neuroscience" aspects of memory recall. However, from my perspective the authors have addressed most major concerns here, and the experimental design presented would seem to be one that can indeed test the processes they are attempting to test. As such I find the information from this study interesting, further supporting the notion that information and memory is stored in the neo-cortex in some way, and not directly in the hippocampus, and that hippocampal activity works to re-instate this information through disinhibition of these circuits. What would be interesting would be to watch the formation of these memories during the training phase, and see if a similar change in the E-I balance occurs. That is, does the Hippocampus "awaken" latent stored information about the associated visual through disinhibition, or is it actually re-instating the E-I balance, and hence the processing state, of the circuits when the stimuli were presented. (I realise it may not actually be able to disentangle these two ideas - and that they may in fact be the same thing.)

    Thank you for this comment. We agree with the reviewer that in future work it would be highly interesting to observe the dynamics of glutamate and GABA during memory formation. Moreover, comparing fMRS data acquired during the learning and test phase of the inference task will likely provide new insight into the mechanisms that support memory recall.

    In all, I found this study methodologically novel, rigorous and sound, and the conclusions and results intriguing and of interest.

    Rideaux, Reuben. 'No Balance between Glutamate+glutamine and GABA+ in Visual or Motor Cortices of the Human Brain: A Magnetic Resonance Spectroscopy Study'. NeuroImage 237 (15 August 2021): 118191. https://doi.org/10.1016/j.neuroimage.2021.118191.

    We would like to thank the reviewer for their complimentary remarks and constructive comments. We are grateful to the reviewer for highlighting the novelty and rigor of our approach. As outlined above, we have addressed your comments in the revised manuscript by including new analyses and substantial changes to the text. We hope you agree that these changes have significantly improved the manuscript.

    Reviewer #2 (Public Review):

    [...] The current study had numerous strengths. The effort to better understand the mechanisms underlying cortical reinstatement in humans is important, although typically constrained by the inferential limitations of BOLD data. Here, the authors test an EI-based account of reinstatement through the application of simultaneous fMRI and fMRS. Both methodologically and conceptually, the work establishes a framework for exploring and understanding how the brain might implement reinstatement. The results were generally compelling given the relatively narrow hypothesis and supportive of the claims of the authors. One weakness was that, perhaps due to the novel nature of the imaging approach, the reliability/robustness of certain neural results was hard to ascertain, particularly for the BOLD data which the authors acknowledge might be compromised compared to a non-combined sequence. For example, the presence and location (e.g. hippocampal laterality) of some of the effects seemed to strongly depend on key preprocessing decisions (smoothing) and at least one participant was excluded due to data quality issues although the criteria for the decision was not described. Notably, this concern is offset slightly by the convergence (and correlation) of the results across fMRI and fMRS.

    We would like to thank the reviewer for highlighting the methodological and conceptual strengths of our study. We are grateful for the constructive comments and hope the reviewer agrees that the revised manuscript is much improved.

    As the reviewer notes, and as stated in the Methods section, due to the novel nature of the imaging approach, the fMRI component in the fMRI-fMRS sequence was compromised relative to non-interleaved state-of-the-art multiband EPI sequences. Nevertheless, several factors indicate the reliability and robustness of our fMRI results. First, the BOLD effect in visual cortex and hippocampus reported here replicate equivalent analyses conducted previously using data acquired on the same task with a higher quality multiband EPI sequence (Barron et al., Cell 2020). Moreover, the higher quality multiband data acquired previously permitted a searchlight Representational Similarity Analysis (RSA) which further revealed reinstatement of the associated visual cues in hippocampus and visual cortex during inference. This latter result demonstrates the involvement of these two brain regions in associative recall during the inference task, and provides the basis for investigating the underlying mechanism for recall using the data reported here.

    Second, our fMRI results are consistent with other studies investigating associative recall of visual cues, where an increase in BOLD signal is observed in the hippocampus and visual cortex (e.g. Horner et al., 2015; Wimmer and Shohamy 2012).

    Third, at a reduced threshold, the effects in hippocampus are bilateral regardless of the smoothing parameters employed. While the peak on one side did not survive whole-brain FWE correction, the t-statistic for the bilateral hippocampal effect is now reported in the revised manuscript (see revised legend for Supplementary File 4 and Figure 3–figure supplement 1D). Notably, whole-volume FWE statistical correction is more conservative than small-volume correction typically applied to regions of the medial temporal lobe, such as the hippocampus, where the BOLD signal is susceptible to distortion and signal loss.

    Finally, we explain the criteria for excluding a participant from the fMRI data. Namely, if the quality of the data was insufficient to allow co-registration between the EPI and structural scan, then data were excluded. These criteria are now clearly stated in the revised Methods section. Apologies that this was not made clear in the previous submission of the manuscript.

    In summary, despite the EPI in the combined fMRI-fMRS sequence being compromised relative to state-of-the-art multiband EPI sequences, the validity of our fMRI results is supported by a number of factors described above, including replication of previous results. We also note that the novelty of our study primarily lies with the fMRS data and the relationship between the fMRS and fMRI data. As demonstrated in Figure 4–figure supplement 3, the quality of the fMRS data is comparable with recent studies using non-interleaved MRS sequences. Moreover, the Monte-Carlo simulations and related permutation testing for false errors further illustrates the validity of our approach. We hope the reviewer agrees that together these factors demonstrate the reliability and robustness of the neural data we present.

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

    Koolschijn and colleagues present a novel and timely investigation of the balance between excitation and inhibition to explore the role of glutamate and GABA during memory retrieval. The innovative use of rapidly interleaved fMRI and fMRS provides a compelling link between successful retrieval effects in hippocampus and inhibitory/excitatory dynamics in visual cortex. The study itself is well-motivated and well executed, complementing prior cross-species work, and provides an intriguing set of results to support the major claims. This paper will be noteworthy to those interested in hippocampus-mediated cortical dynamics during memory retrieval. The rigorous methodology also demonstrates the utility of fMRS in investigating complex cognitive processes.

    (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 #2 agreed to share their names with the authors.)

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

    This study utilises the still relatively novel technique of interleaved fMRI and fMRS to investigate the role of glutamate and GABA in memory recall within the cortex - specifically the visual cortex - and how this may be driven by the hippocampus. While the use of fMRS to investigate changes in neurotransmitter levels is not new, as a well-defined methodology it is still somewhat immature, with several degrees of freedom in the exact method of application, which often reduces the confidence in the results. One of the strengths of this study is that through careful experimental and analysis design, these degrees of freedom are somewhat constrained and controlled for using a within subject design, and an inbuilt control condition. While this boosts confidence, there are still a few minor issues readers need to consider when assessing the results and subsequent inferences the authors make.

    The first issue that may be a concern is the idea that MRS measures of Glutamate and GABA are appropriate proxy measures of the excitation-inhibition balance. While this is a not uncommon interpretation of the relative levels of these two neurotransmitters, a recent paper in NeuroImage presents data to refute this assertion (Rideaux et al 2021), especially in the visual cortex. However, The current study by Kooschijn et-al is different from the data reported by Rideaux, in that the current data is looking at the relative change in the Glu/GABA ratio (and the relative difference in each Glu and GABA by themselves) between two conditions, and not the overall "balance" at rest, or during activity. This dynamic nature of the data, and the temporal resolution present, may explain why a relationship is found between the change in ratio (and individual levels) and performance in this study. Either way, the authors could address this recent paper of Rideaux and the challenge it may present for their interpretation of Glu/GABA ratios as a measure of E-I balance.

    It is also worth noting that the relative nature of measures utilised may enhance error propagation between individual measures, increasing the risk of a false positive result. The authors have tried to address this through the use of Monte-Carlo permutation analysis for false errors, and it does go some way to restoring confidence.

    There are some other potential methodological caveats a reader inexperienced in fMRS (and indeed fMRI) should be aware of:

    • The first is that the MRS sequence utilised is not typically used to measure GABA (a point the authors also note), with GABA typically measured using so called "editing" techniques like MEGA-PRESS. The authors also utilise a non-standard unconstrained fit for GABA in their analysis of the MRS spectrum. While these two non-standard methodologies may weaken confidence the GABA measures, the authors are to be commended on the use of simulations to demonstrate that even if "absolute" measures of GABA via this methodology may be slightly outside the usual norms, this methodology is able to detect changes in GABA of the size of those detected in this experiment. These simulations, coupled with the Monte-Carlo permutation steps for FWE correction are a strength of the paper, and I encourage readers to fully examine the supplementary material for this paper to get a better appreciation for the quality and validity of the data being presented.

    • The exact time locking (or not) of fMRS data acquisition to phases of the stimulus presentation and the subsequent temporal resolution and timeline of changes is not fully explained - and may be somewhat misleading. Given the MRS data were collected in step sizes of 4 secs, it may be hard to understand how a temporal resolution of 2.5 secs for the fMRS data is achieved. Likewise, given that the total Question and ITI time was allowed to vary from stimulus to stimulus, it may be hard to understand how the timeline in figures 4 and 5 are achieved. This could be ameliorated by a better explanation of data collection process in the methods, and how data was averaged to produce the timelines as presented.

    • Lastly, the fMRI technique being used is not a typical echo-planar sequence, and the data produced are by the authors own admission impacted in quality. Given that the hippocampus, and other parts of the anterior temporal lobe are difficult to measure at the best of times, the BOLD signal changes from this data may be less robust than normal. As such, while interesting, the correlation between the Hippocampal BOLD signal and the Glu/GABA ratio changes might be considered tentative, and can only benefit from replication at a later data.

    My expertise as a reviewer is usually in the methodological aspects of fMRS studies, and not really in the "psychology" or "cognitive neuroscience" aspects of memory recall. However, from my perspective the authors have addressed most major concerns here, and the experimental design presented would seem to be one that can indeed test the processes they are attempting to test. As such I find the information from this study interesting, further supporting the notion that information and memory is stored in the neo-cortex in some way, and not directly in the hippocampus, and that hippocampal activity works to re-instate this information through disinhibition of these circuits. What would be interesting would be to watch the formation of these memories during the training phase, and see if a similar change in the E-I balance occurs. That is, does the Hippocampus "awaken" latent stored information about the associated visual through disinhibition, or is it actually re-instating the E-I balance, and hence the processing state, of the circuits when the stimuli were presented. (I realise it may not actually be able to disentangle these two ideas - and that they may in fact be the same thing.)

    In all, I found this study methodologically novel, rigorous and sound, and the conclusions and results intriguing and of interest.

    Rideaux, Reuben. 'No Balance between Glutamate+glutamine and GABA+ in Visual or Motor Cortices of the Human Brain: A Magnetic Resonance Spectroscopy Study'. NeuroImage 237 (15 August 2021): 118191. https://doi.org/10.1016/j.neuroimage.2021.118191.

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

    The current study sought to better understand excitatory and inhibitory (EI) dynamics in visual cortex as a function of hippocampally-mediated remembering. Specifically, the authors hypothesized that the successful retrieval of visual information would lead to an increase in hippocampal engagement (quantified by fMRI-based measurements of BOLD) and further lead to an increase in excitatory dynamics in visual cortex (quantified by MRS-based measurements of increased glutamate relative to GABA). Using a paradigm employed in rodent studies, but modified for use in humans using virtual reality, the authors found evidence for both hypothesized effects as well as, critically, an across-participant correlation between them. Although much prior work has suggested that the hippocampus plays a critical role in 'reinstating' activity in neocortex at retrieval that was present during encoding, the current study leveraged a novel approach to suggest that such reinstatement might further tilt cortical dynamics towards excitation relative to inhibition.

    The current study had numerous strengths. The effort to better understand the mechanisms underlying cortical reinstatement in humans is important, although typically constrained by the inferential limitations of BOLD data. Here, the authors test an EI-based account of reinstatement through the application of simultaneous fMRI and fMRS. Both methodologically and conceptually, the work establishes a framework for exploring and understanding how the brain might implement reinstatement. The results were generally compelling given the relatively narrow hypothesis and supportive of the claims of the authors. One weakness was that, perhaps due to the novel nature of the imaging approach, the reliability/robustness of certain neural results was hard to ascertain, particularly for the BOLD data which the authors acknowledge might be compromised compared to a non-combined sequence. For example, the presence and location (e.g. hippocampal laterality) of some of the effects seemed to strongly depend on key preprocessing decisions (smoothing) and at least one participant was excluded due to data quality issues although the criteria for the decision was not described. Notably, this concern is offset slightly by the convergence (and correlation) of the results across fMRI and fMRS.

    Read the original source
    Was this evaluation helpful?