Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2015 Nov;18(11):1565-7.
doi: 10.1038/nn.4125. Epub 2015 Sep 28.

A Positive-Negative Mode of Population Covariation Links Brain Connectivity, Demographics and Behavior

Free PMC article

A Positive-Negative Mode of Population Covariation Links Brain Connectivity, Demographics and Behavior

Stephen M Smith et al. Nat Neurosci. .
Free PMC article


We investigated the relationship between individual subjects' functional connectomes and 280 behavioral and demographic measures in a single holistic multivariate analysis relating imaging to non-imaging data from 461 subjects in the Human Connectome Project. We identified one strong mode of population co-variation: subjects were predominantly spread along a single 'positive-negative' axis linking lifestyle, demographic and psychometric measures to each other and to a specific pattern of brain connectivity.


Figure 1
Figure 1
(a) The set of SMs (subject measures) most strongly associated with the CCA mode of population variability. SMs included in the CCA are colored blue, while others (grey) were correlated with the CCA mode post-hoc. Vertical position is according to correlation with the CCA mode, while font size indicates SM variance explained by the CCA mode. Here we do not report “secondary” SMs that are highly redundant with those shown here (Supplementary Table 1 shows the complete set of SMs that correlate highly with the CCA mode). See for details of the SMs. (b) The principal CCA mode - a scatter-plot of SM weights versus connectome weights, with one point per subject, and an example subject measure (fluid intelligence) indicated with different colors. The high correlation visualised here indicates significant co-variation between the two datasets. (c) The total variance explained of the original data matrices (shown separately for connectomes and subject measures) is plotted for the first 20 CCA modes. In black/grey is shown the mean and the 5th to 95th percentiles of the null distribution of the same measures, estimated via permutation testing. Using the null distributions to normalize variance explained accounts for the fact that the initial modes are expected to have higher correlations, even in the null scenario, but, as can be seen from the nulls, this is in any case a very small effect.
Figure 2
Figure 2
(a) The 30 brain connections most strongly associated with the CCA mode of population variability. To aid interpretation, the CCA edge modulation weights are multiplied by the sign of the population mean correlation; hence red indicates stronger connections and blue weaker, for high-scoring subjects (and vice versa for low-scoring subjects). (b) Map of CCA connection strength increases (each node’s parcel map is weighted by CCA edge-strength increases, summed across edges involving the node). (c) Group-mean functional clustering: 4 clusters from a hierarchical analysis of all 200 nodes’ population-average full correlation (Supplementary Fig. 3). These fall into two groups: one cluster (blue) contains sensory, motor, insula and dorsal attention regions, and a group of 3 correlated clusters (brown, red, yellow) primarily covering the default mode network and subcortical/cerebellar areas. (d) As b, but showing CCA connection strength decreases. Maps d and b are largely non-overlapping except in insula. b has spatial correlation of +0.40 with the default-mode areas in c (i.e., high overlap), while d shows negative correlation (−0.12). The average connectivity strength increase is approximately double that of the average decrease (as reflected in the predominance of red edges in a; also, a single map averaging across all 200 edges for each node shows a pattern of overall increase highly similar to b; finally, both b and d are thresholded at the 80th percentile of their respective distributions, and if the threshold applied to b were applied to d, none of the strength reductions shown would survive).

Comment in

Similar articles

See all similar articles

Cited by 144 articles

See all "Cited by" articles


    1. Van Essen DC, et al. NeuroImage. 2013;80:62–79. - PMC - PubMed
    1. Beckmann CF, Smith SM. IEEE Trans. Med. Imaging. 2004;23:137–152. - PubMed
    1. Hotelling H. Biometrika. 1936;28:321–377.
    1. Winkler A, et al. NeuroImage. 2015 In Press.
    1. Spearman C. American J. Psychology. 1904;15:201–292.

Online Methods References

    1. Ugurbil K, et al. NeuroImage. 2013;80:80–104. - PMC - PubMed
    1. Jenkinson M, et al. NeuroImage. 2012;62:782–90. - PubMed
    1. Fischl B, et al. NeuroImage. 1999;9:195–207. - PubMed
    1. Marcus DS, et al. NeuroImage. 2013;80:202–219. - PMC - PubMed
    1. Smith SM, et al. NeuroImage. 2013;80:144–168. - PMC - PubMed

Publication types