Introduction

Understanding the neural basis of individual differences is an important goal of modern neuroscience. A data-driven canonical correlation analysis (CCA) on imaging and behavioral data of a large cohort of subjects in the Human Connectome Project (HCP) has identified a single mode of population variation that links individuals’ functional brain connectivity with their demographic and behavioral measures and also features a behavioral change from negative to positive personal traits¹. However, the structural basis of such a one-dimensional functional connectivity-behavior relationship remains largely unknown. Here, we investigate the potential role of brain structures in this relationship by examining the covariation of the population covariation of the cortical thickness, resting-state connectivity, and behavioral measures.

Methods

We used imaging and behavioral data from 818 HCP subjects². Resting-state functional connectivity among 200 distinct brain regions are provided by the HCP PTN (Parcellation + Timeseries + Netmats) datset³. We also used the cortical thickness map on the brain surface of 32K resolution and included 129 behavioral/demographic measures for analysis. We performed the principal component analysis (PCA) to reduce non-subject dimensions to 100 before applying CCA to explore pairwise relationships among these three modalities. The age, gender, brain size, and head motions were regressed out from the CCA analyses. Permutation test were used to derive statistical significance¹. The first CCA mode was taken as the principal mode for each CCA analysis. The measures of three modalities were correlated with corresponding canonical variates across subjects to derive thickness scores, behavior scores and connectivity scores. For each brain region/parcel, we then count the number of its connections showing significantly increased or decreased strength along the direction of CCA mode to represent regional modulations in functional connectivity. We also used the first principal component to represent an eigen-direction of 6 canonical variates from the three CCA analyses.

Results

The CCA found a single significant mode (r = 0.70, p = 1.0×10-4) linking the cortical thickness with demographic/behavioral measures. Positive and negative personal traits show opposite correlations with this identified CCA mode, similar to what has been previously observed with the connectivity-behavior CCA mode. More interestingly, the cortical thickness across the neocortex displays distinct correlations with this CCA mode: the lower-level sensory regions, including sensorimotor, auditory and visual cortices, show significant positive correlations whereas the higher-order brain regions, mostly the default mode network (DMN) and frontal regions, are characterized by negative values (Fig. 1). The finding suggests that individuals with more positive personal traits tend to have thinner gray matter at the higher-order brain regions but thicker cortex at the sensory areas. We also applied CCA to resting-state connectivity and behavioral data and replicated the previous finding about the connectivity-behavior CCA mode. A close inspection on the modulation of functional connectivity along the connectivity-behavior mode revealed a similar divergent change in these two sets of brain areas. The connectivity strength increase along the mode direction is the highest in the higher-order regions, particularly the DMN, but the lowest at the sensory areas. In contrast, the connectivity strength decrease showed completely opposite pattern of changes (Fig. 2). Finally, we derived the CCA mode between the cortical thickness and functional connectivity data in a similar way and compared it with the previous two modes. The result suggests that the three CCA modes are highly correlated with each other with the same modality of data modulating in a similar way in different CCA analyses (Fig. 3).

Discussion

By examining the population covariation of the cortical thickness, resting-state functional connectivity, and demographic/behavioral measures from a large cohort of subjects, we found a specific direction linking the three modalities of data, along which they all show specific patterns of change. Individuals tend to have more positive personal traits towards one end of this direction, which are also associated with divergent changes at the lower- and higher-order brain regions in both structural and functional imaging data: the thinner cortex and stronger resting-state connectivity at the higher-order areas but opposite modulations at the lower-order sensory regions. The cortical area of these two sets of brain regions has been found to expand disproportionally with the brain size increase, over the developmental course, and across species⁴, suggesting their distinct roles in brain functions. Our findings suggest that divergent changes of the cortical thickness at the brain regions of distinct hierarchy could be an important structural feature that is tightly linked to human behavior.

Conclusion

Divergent changes at the lower- and higher-order brain regions in the cortical thickness and resting-state functional connectivity are the structural and functional imaging features most strongly correlated with inter-subject behavioral variation.

Acknowledgements

This study is supported by the NIH Pathway to Independence Award (K99/R00).