On 2017 Jun 05, Michael Greicius commented:

Thanks for your interest in the paper. The discrepancy you note arises from the reannotation step described in the manuscript. We reannotated each probe sequence to a gene using the ReAnnotator tool [1]. The used probe-to-gene mapping is listed in supplementary data file S2 [2]. The complete reannotated file is available through sourceforge [3].

[1]http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0139516 [2]http://science.sciencemag.org/highwire/filestream/631209/field_highwire_adjunct_files/2/Richiardi_Data_File_S2.xlsx [3]https://sourceforge.net/projects/reannotator/files/annotations/

On 2017 May 30, Xiang-Zhen Kong commented:

I found that there are 8 of 136 genes (listed below) not available from the Allen Brain website, http://human.brain-map.org/static/download. I used the "Complete normalized microarray datasets". Could you share how you downloaded the expression data with all the 136 genes? Thanks.

8 genes not available from the Allen Brain website: CDK1 PRSS35 SHISA9 SIX3-AS1 TINCR LINC00617 MS4A8 NUPR1L

On date unavailable, commented:

None

On 2017 May 02, Jonas Richiardi commented:

A detailed reply has been posted at https://doi.org/10.1101/132746, showing that our original results stand. We provide a short summary here for convenience.

To a first approximation, all connectivity, like all politics, is local and these two features -- nearness and connectivity -- are challenging to disentangle. Evidence abounds that the majority of connectivity is local, but this critical attribute of functional anatomy is perhaps most efficiently conveyed in a macaque tracer study by Markov et al (2012, see in particular Figure 7). Thus, in undertaking our analysis, we were well aware that spatial nearness is correlated with connectivity, which we addressed using a measure of "fine" tissue-tissue similarity derived from an ontological atlas provided by the Allen Institute. The peer-reviewers at Science were also aware of this confound and insisted on an additional level of tissue-tissue similarity correction which our analysis survives (note that Data File S1 of our paper also included information about "coarse" tissue classes (field coarsetissueclass)).

The commenters suggest that the tissue-tissue similarity correction we applied is inadequate and point out that in the brain samples we used there remains a significant linear correlation between Euclidean distance and correlated gene expression between brain regions. The r-value for this correlation is 0.1. When the coarse tissue-tissue correction is applied, the r-value for the correlation drops to 0.094. In terms of variance explained, this means that slightly less than 1% of the correlated gene expression measure used in our analyses can be explained by Euclidean distance.

We thank the commenters for providing an independent replication of our core results, using the same method and data as in our original paper, and our rebuttal can be summarized in the following 5 points:

• Our analysis survives correction for Euclidean distance (see details in full reply), applied on top of the tissue-tissue similarity correction we used, as well as distance-aware permutation tests. Here, we note that there is an intrinsic contradiction in the commenters' counter-argument that a linear regression of Euclidean distance is inadequate, despite the fact that their critique (see their Figure 1B) is founded on this very same linear correlation.
• The random clusters generated by the commenters, meant to show the non-specificity of our results, consist of nodes that are roughly twice as close to one another as the nodes in the actual functional networks (see Figure 1 in our full reply).
• We replicated results of our connectivity gene in both a mouse connectivity analysis and a resting-state fMRI connectivity analysis. The commenters did not generate gene lists for any of their random cluster analyses to try and replicate in these or other independent datasets.
• Euclidean distance correction will wrongly assign "nearness" to two "neurally distant" regions on the crowns of adjacent gyri (see Figure2 in our full reply). This is essentially the opposite problem of the limitation to tissue-tissue correction that the commenters rightly point out in their figure 1A.
• Correcting the connectivity of two regions for nearness, using any measure, is bound to dilute the measure of connectivity. That our gene list survives two levels of tissue-tissue similarity correction plus a correction for Euclidean distance and is then replicated in a mouse structural connectivity dataset and a human resting-state fMRI connectivity datasets strikes us as strong support for the conclusion that these genes are important to functional connectivity.

Detail of these and several other points are available in the full reply paper at https://doi.org/10.1101/132746.

On 2016 Oct 19, Spiro Pantazatos commented:

Mind the distance: spatial proximity confounds tissue-tissue gene expression correlations reported in this study.

This is a novel and very interesting study. However, the authors do not adequately control for spatial proximity, which, contrary to the authors’ claims in the original article, accounts entirely for high within-network strength fraction according to our recent replication/reanalysis of these same data. Furthermore, “null networks”, (i.e. contiguous clusters with center coordinates randomly placed throughout cortex), also have significantly high strength fractions, indicating that high within-network strength fraction is not related to resting-state networks identified by fMRI.

Here is a link to the full technical commentary and replication/reanalysis write-up with additional supplementary discussion: http://biorxiv.org/content/early/2016/10/04/079202

And here is a link to the replication/reanalysis code on Github: https://github.com/spiropan/ABA_functional_networks

The lead authors are aware of these findings and concerns (I notified them via personal email in March, 2016) and they have let me know they plan to respond. I have submitted the commentary for peer review to Frontiers in Neuroscience. If accepted, they have the option to publish a formal rebuttal/response letter there, and/or respond in the comments section here.

Commentary Abstract

A recent report claims that functional brain networks defined with resting-state functional magnetic resonance imaging (fMRI) can be recapitulated with correlated gene expression (i.e. high within-network tissue-tissue strength fraction, SF) (Richiardi et al., 2015). However, the authors do not adequately control for spatial proximity. We replicated their main analysis, performed a more effective adjustment for spatial proximity, and tested whether 'null networks' (i.e. clusters with center coordinates randomly placed throughout cortex) also exhibit high SF. Removing proximal tissue-tissue correlations by Euclidean distance, as opposed to removing correlations within arbitrary tissue labels as in (Richiardi et al., 2015), reduces within-network SF to no greater than null. Moreover, randomly placed clusters also have significantly high SF, indicating that high within-network SF is entirely attributable to proximity and is unrelated to functional brain networks defined by resting-state fMRI. We discuss why additional validations in the original article are invalid and/or misleading and suggest future directions.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

On 2016 Oct 19, Spiro Pantazatos commented:

Mind the distance: spatial proximity confounds tissue-tissue gene expression correlations reported in this study.

This is a novel and very interesting study. However, the authors do not adequately control for spatial proximity, which, contrary to the authors’ claims in the original article, accounts entirely for high within-network strength fraction according to our recent replication/reanalysis of these same data. Furthermore, “null networks”, (i.e. contiguous clusters with center coordinates randomly placed throughout cortex), also have significantly high strength fractions, indicating that high within-network strength fraction is not related to resting-state networks identified by fMRI.

Here is a link to the full technical commentary and replication/reanalysis write-up with additional supplementary discussion: http://biorxiv.org/content/early/2016/10/04/079202

And here is a link to the replication/reanalysis code on Github: https://github.com/spiropan/ABA_functional_networks

The lead authors are aware of these findings and concerns (I notified them via personal email in March, 2016) and they have let me know they plan to respond. I have submitted the commentary for peer review to Frontiers in Neuroscience. If accepted, they have the option to publish a formal rebuttal/response letter there, and/or respond in the comments section here.

Commentary Abstract

A recent report claims that functional brain networks defined with resting-state functional magnetic resonance imaging (fMRI) can be recapitulated with correlated gene expression (i.e. high within-network tissue-tissue strength fraction, SF) (Richiardi et al., 2015). However, the authors do not adequately control for spatial proximity. We replicated their main analysis, performed a more effective adjustment for spatial proximity, and tested whether 'null networks' (i.e. clusters with center coordinates randomly placed throughout cortex) also exhibit high SF. Removing proximal tissue-tissue correlations by Euclidean distance, as opposed to removing correlations within arbitrary tissue labels as in (Richiardi et al., 2015), reduces within-network SF to no greater than null. Moreover, randomly placed clusters also have significantly high SF, indicating that high within-network SF is entirely attributable to proximity and is unrelated to functional brain networks defined by resting-state fMRI. We discuss why additional validations in the original article are invalid and/or misleading and suggest future directions.

Conflict of Interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

On 2017 May 30, Xiang-Zhen Kong commented:

I found that there are 8 of 136 genes (listed below) not available from the Allen Brain website, http://human.brain-map.org/static/download. I used the "Complete normalized microarray datasets". Could you share how you downloaded the expression data with all the 136 genes? Thanks.

8 genes not available from the Allen Brain website: CDK1 PRSS35 SHISA9 SIX3-AS1 TINCR LINC00617 MS4A8 NUPR1L