Gradient organisation of functional connectivity within resting state networks is present from 25 weeks gestation in the human fetal brain

Reviewer #1 (Public Review):

Summary:

The work studies functional connectivity gradients using advanced resting-state analyses in fetuses and sheds light on pre-existing functional topographies and their continued development during the third trimester of gestation.

Strengths:

The work is novel, and applies state of the art connectomic mapping techniques to study fetal brain organization. The work capitalizes on the existence of large, open access datasets, and shows interesting and impactful findings on the presence of functional topographies from 25GW onwards.

Weaknesses:

To better understand underlying factors in cortical functional organization, the authors could add additional exploratory analyses to assess the role of cortical microstructure/myelin and thalamic connectivity.

Reviewer #2 (Public Review):

In this study, Moore et al. utilise resting-state fMRI data from the Developing Human Connectome Project, applying a recently developed technique ("connectopic mapping") to identify gradients of functional connectivity within resting-state networks in the human foetal brain. Whilst such gradients have previously been identified in adults, this is the first study to explore the topographic organisation of functional connectivity in the foetal brain. Furthermore, the authors describe localised changes within these gradients over the course of gestation, particularly in brain regions implicated in multisensory processing. Together, these results imply that topographic gradients of brain function are present within the developing foetal brain, and continue to develop through gestation. However, the study does not consider critical confounds inherent in the connectopic mapping technique, and as such I do not believe that the data as presented are sufficient to support the conclusions.

Recent evidence (Watson & Andrews, 2023, Neuroimage) has indicated that the connectopic mapping technique employed here can be substantially confounded by spatial autocorrelations present within the data (for instance, occurring naturally due to the inherent smoothness of the BOLD response, and/or introduced artificially during standard data pre-processing steps such as spatial smoothing or interpolation between co-ordinate spaces). These confounds allow connectopic gradients to be obtained even from random data, and which appear highly similar to those obtained from real data, suggesting that these gradients are strongly influenced by such confounds. Consequently, the resulting gradients may be an inevitability of the way the connectopic mapping technique works, rather than reflecting underlying brain functions per se.

In the current study, all of the gradients flow smoothly and continuously along a single axis within every network region, typically oriented relative to the long axis of the region. To put it another way - the connectopic mapping gives fundamentally the same answer in every network region. Such an organisation does feel a bit biologically implausible, and could be more consistent with the gradients representing an inevitable solution of the analysis technique, rather than necessarily reflecting brain function. Indeed, in some cases the gradients do not correspond well to known organisational principles of the regions. For instance, the primary gradient in the principal visual network flows smoothly along a superior to inferior axis, which the authors suggest corresponds to retinotopic polar angle maps - however, polar angle maps would be expected to reverse direction between each visual region, yet such reversals are not present in this connectopic map. The authors note that the foetal gradients appear highly similar to those previously obtained within similar regions in adult participants - this could be indicative of a consistent organisation across development, but would also be consistent with the same confound affecting foetal and adult participants. The reported changes in the gradients across gestation could reflect changes in the extent of these spatial autocorrelations or in the shape of the regions of interest (perhaps in turn resulting from changes in the underlying brain geometry) rather than necessarily reflecting development of brain function or specialisation. None of this precludes the possibility that these connectopic gradients may (at least partially) also reflect genuine brain functions, but it does obfuscate the extent to which they do so. It would be useful for the authors to give some consideration to this issue.

On a different note, could the authors comment on their reason for studying these gradients at the network level. The authors argue (and I agree) that brain function is likely to be organised topographically, rather than split into discrete parcellated regions. Nevertheless, the brain networks the authors choose to use are themselves discrete regions of interest (albeit fairly large ones). Other groups (e.g., Margulies et al, 2016, PNAS) have described coarser-scale connectopic gradients spanning the whole brain. Is there a reason that the authors have chosen to extract network-level gradients, rather than say coarser-scale whole-brain gradients? Have the authors considered examining how whole-brain gradients change over gestation?

Lastly, the correlated changes between gradients and gestation week appear to occur within small localised clusters. Does this reflect local perturbations of the gradient, or is there perhaps a wider change in the gradient as a whole and these clusters reflect extreme points within this that have changed the most (for instance corresponding to an expansion/contraction of the gradient)?