Article activity feed

  1. Author Response:

    Reviewer #2:

    Weaknesses:

    The principal result isn't hugely surprising: inclusion of the HTR2A map in the model produces ΔGBC changes with a similar spatial topography to that map in the model. The empirical ΔGBC maps are also similar to the HTR2A maps, and so the simulated ΔGBC give a good fit to the empirical ΔGBC data. Yes, the authors demonstrate convincingly that this simulated-empirical ΔGBC fit is stronger than the similarity to the HTR2A map itself, and also to that of various alternative receptor maps and surrogate null models. But the central result does have an element of 'getting out what you put in'.

    The ΔGBC metric is a bit weak as a stand-alone outcome variable. The usual quantity used in this type of model is the goodness-of-fit of simulated to empirical FC. Indeed, the authors have used this calculation in the initial calibration step for their model, where they identified the global coupling strength parameter that yielded the best fit of empirical to simulated FC in the placebo condition, achieving reasonably good fit (Spearman rank correlation r=0.45). However the authors don't report how this FC fit changes with the inclusion of the HTR2A map modulations. It is an open question whether a model with HTR2A-modulations added that improved ΔGBC but not FC fit should be regarded as a better model than one without.

    We believe this issue relates to spatial scales of this class of large-scale models, which operate at the level of brain regions (parcels) and therefore are not well suited to capturing both inter- regional and intra-regional changes in connectivity, because intra-regional connectivity is not explicitly modeled. We consider this bridging across spatial scales (from voxelwise to inter- regional) to be an important direction for future model development to capture pharmacological neuroimaging effects. We now state in the Discussion: “Because our model is defined at the level of cortical parcels, it cannot speak to changes in connectivity that occur over smaller spatial scales, particularly among neurons within the parcels themselves. Our findings indicate that this coarse dynamical description is sufficient to capture regional GBC 447 differences, but future work that goes beyond regional mean-field modeling may be needed to 448 fully resolve the fine- grained effects of pharmacology on within- and between-region FC.”

    The authors do not make clear why it is necessary, and/or why it makes sense to perform GSR on the mathematical model FC anyway. The artifactual contributions to FC that make this necessary for empirical data are by construction not present in modelled data, after all.

    We now state in the Methods: “GSR was also performed in the model, as GSR not only removes artifactual signal components but neuronal signal as well.”

    The model description is very comprehensive but it omits the actual equations used, which are (I believe) the algebraic neural activity covariance equations at ~Eq. 21 in Deco et al. 2014. After 10 equations leading up to this, the methods section simply says "Simulated BOLD covariance matrices were derived by linearizing these equations and then algebraically transforming the linearized synaptic covariance matrix, using a procedure which we previously reported in Demirtaş et al. (2019)." The final algebraic equations should be added, and also emphasize that they are the ones used. Readers less familiar with these models could otherwise be forgiven for thinking that the neural and haemodynamic differential equations listed in Eqs 1-10 were the ones used, which is not the case.

    We now include an expanded description of this in the Methods section.

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

    This paper will be of interest to scientists working on computational modelling of neuroimaging data, and on the neural effects of psychedelic drugs and other pharmacological interventions. The study is well-motivated. The statistical and data analytic methodologies are rigorous and advanced. The with conclusions are well-supported by the presented data. The modelling methodology includes technical innovations that are potentially of broad utility and importance.

    (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):

    In this manuscript, Burt and colleagues use a well-motivated neural mass modelling approach to better understand the neurobiological basis of a recently-demonstrated link between 5HT2A agonists and whole-brain signatures of increased integration - so-called, Global Brain Connectivity (GBC). The authors derive a simple model of both excitatory and inhibitory neurons in the cerebral cortex, whose activity is dependent on the weighted connections between regions, which in turn are modulated by a neural gain parameter. Crucially, this gain parameter is linked to the action of neuromodulatory neurotransmitters (and exogenous ligands), and has a heterogeneous effect that depends on the expression of a range of different classes of g-protein-coupled receptors. The authors include an estimate of this heterogeneity (mRNA maps from the Allen Human Brain Atlas) in their model by having the expression of the 5HT2A receptor mRNA separately effect the gain of both Excitatory and Inhibitory populations. Fitting their model to previously published empirical data, the authors find that the data are best explained by a relative increase in excitatory > inhibitory neural activity, which is consistent with the known mechanism of action of 5HT2A ligands. After providing a number of useful statistical checks, the authors then use a dimensionality reduction approach to relate different aspects of the functional neural signatures to unique aspects of the phenomenology of the psychedelic experience associated with 5HT2A.

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

    Summary:

    The submission by Burt et al. is an interesting and timely contribution to the computational psychiatry and pharmacological fMRI literature. It advances an account of the synaptic mechanism through which LSD influences global brain activity and functional connectivity (FC), through differential modulation of synaptic gain in excitatory and inhibitory neural populations. This is done by extending a previously-introduced computational model of whole-brain neural dynamics to include regional variation in cellular and neurochemical properties (specifically, the spatial profile of serotonin receptors). This regional variation is based on maps from the Allen Human Brain Atlas, which give transcriptomic expression profiles for the HTR2A gene (encoding the 5-HT2A receptor), amongst others. The model's predictions are compared quantitatively to a human subject fMRI dataset with an LSD intervention, and provide a parsimonious and convincing account of observed LSD-induced changes in global brain functional connectivity. The general approach of augmenting a connectome-based network model of whole-brain neural activity with maps of spatially varying neurotransmitter, gene expression, cytoarchitectural, etc. features from various complementary data sources has been pioneered by this group over the past five years.

    Strengths:

    The study represents an extension of an established neuroinformatics-informed modelling methodology to a novel imaging dataset (LSD intervention), and appropriate use of the recently developed (and increasingly widely used) Allen Human Brain Atlas gene expression maps, with clear neurobiological rationale wrt the fMRI dataset.

    The statistical methodologies for comparing spatial maps (spatial autocorrelation-controlled null models) are rigorous and sophisticated.

    The central result, that HTR2A maps improve model performance better than other receptor maps (e.g. dopamine), is an important demonstration of the utility of this approach (although with some caveats; see below).

    The paper demonstrates comprehensive understanding and utilization of a mathematical model for neural population activity in the service of the research question, including well-chosen modifications to represent neuromodulatory influences, a sensible calibration procedure, and mathematical manipulation to test novel hypotheses.

    ...In particular: the authors have developed a modest piece of new mathematical theory that allows them to analytically incorporate global signal regression into the linear(ized) algebraic model for neural activity covariance and functional connectivity. Because the main dependent variable throughout the study is scalar maps across the cortex of global brain functional connectivity (FC), and changes thereof (ΔGBC), global signal regression (GSR) - a standard but not uncontroversial fMRI denoising technique - naturally is of major importance (because the primary effect of GSR is to remove artifactual global correlation patterns). This approach may in the future also be used to study a wide variety of other fMRI data features, for which it is important to know the potential contribution of GSR.

    Weaknesses:

    The principal result isn't hugely surprising: inclusion of the HTR2A map in the model produces ΔGBC changes with a similar spatial topography to that map in the model. The empirical ΔGBC maps are also similar to the HTR2A maps, and so the simulated ΔGBC give a good fit to the empirical ΔGBC data. Yes, the authors demonstrate convincingly that this simulated-empirical ΔGBC fit is stronger than the similarity to the HTR2A map itself, and also to that of various alternative receptor maps and surrogate null models. But the central result does have an element of 'getting out what you put in'.

    The ΔGBC metric is a bit weak as a stand-alone outcome variable. The usual quantity used in this type of model is the goodness-of-fit of simulated to empirical FC. Indeed, the authors have used this calculation in the initial calibration step for their model, where they identified the global coupling strength parameter that yielded the best fit of empirical to simulated FC in the placebo condition, achieving reasonably good fit (Spearman rank correlation r=0.45). However the authors don't report how this FC fit changes with the inclusion of the HTR2A map modulations. It is an open question whether a model with HTR2A-modulations added that improved ΔGBC but not FC fit should be regarded as a better model than one without.

    The authors do not make clear why it is necessary, and/or why it makes sense to perform GSR on the mathematical model FC anyway. The artifactual contributions to FC that make this necessary for empirical data are by construction not present in modelled data, after all.

    The model description is very comprehensive but it omits the actual equations used, which are (I believe) the algebraic neural activity covariance equations at ~Eq. 21 in Deco et al. 2014. After 10 equations leading up to this, the methods section simply says "Simulated BOLD covariance matrices were derived by linearizing these equations and then algebraically transforming the linearized synaptic covariance matrix, using a procedure which we previously reported in Demirtaş et al. (2019)." The final algebraic equations should be added, and also emphasize that they are the ones used. Readers less familiar with these models could otherwise be forgiven for thinking that the neural and haemodynamic differential equations listed in Eqs 1-10 were the ones used, which is not the case.

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

    I would like to thank the editor for the opportunity to review this work, however considering that I do not have direct experience in the biophysical modelling of dynamic systems, I will only comment on the general aspects of the manuscript. In the article entitled "Transcriptomics-informed large-scale cortical model captures topography of pharmacological neuroimaging effects of LSD" the authors integrate brain-wide transcriptomic data into the large-scale circuit modeling in order to simulate neuromodulatory effects of LSD on large-scale spatiotemporal dynamics of cortical BOLD functional connectivity. This analysis builds on their previously published experimental work which identified that LSD impacts global brain connectivity (GBC) [by elevating GBC in sensory cortex and reduced GBC in association cortex] and that these effects are attributable to the agonism of the serotonin-2A receptor (5-HT2A). Using large-scale circuit modeling in combination with high-resolution spatially-defined transcriptomic data the authors now investigate the underlying mechanisms of these LSD-induced changes showing that the model can capture the spatial topography of these changes and demonstrating that the spatial distribution of 5-HT2A [and not other receptors that have an agonistic relationship to LDS] is critical for generating the cortical topography of LSD-induced functional disruptions.

    From the methodological point of view, this study provides incremental extensions to previously published work, however, this can be viewed both as a potential weakness and a considerable strength. In my opinion, the integration of previous findings and a hypothesis-driven approach is a significant advantage. The adaptation of well-known models for large-scale neural dynamics to investigate pharmacologically-induced changes in brain activity extends the modeling approach and provides novel and insightful contributions towards understanding the biophysical mechanisms of LSD-induced changes in functional connectivity by addressing the mechanistic gap which is frequently lacking in imaging transcriptomic studies. Moreover, the model is also capable of capturing the patterns of functional variation across individuals that are linked to their perception of these pharmacologically-induced changes in experience, going beyond group-average estimates that are commonly used in neuroimaging studies. I also appreciate the investigation of the effects of global signal regression which is still widely debated in the neuroimaging community. Overall, the manuscript is methodologically sound, very well-written, and easy to follow, the key claims presented in the article are supported by the data.

    Read the original source
    Was this evaluation helpful?