Structure-function coupling in highly sampled individual brains

Abstract

Structural connectivity (SC) between distant regions of the brain support synchronized function known as functional connectivity (FC) and give rise to the large-scale brain networks that enable cognition and behavior. Understanding how SC enables FC is important to understand how injuries to SC may alter brain function and cognition. Previous work evaluating whole-brain SC-FC relationships showed that SC explained FC well in unimodal visual and motor areas, but only weakly in association areas, suggesting a unimodal-heteromodal gradient organization of SC-FC coupling. However, this work was conducted in group-averaged SC/FC data. Thus, it could not account for inter-individual variability in the locations of cortical areas and white matter tracts. We evaluated the correspondence of SC and FC within three highly sampled healthy participants. For each participant, we collected 78 min of diffusion-weighted MRI for SC and 360 min of resting state fMRI for FC. We found that FC was best explained by SC in visual and motor systems, as well as in anterior and posterior cingulate regions. A unimodal-to-heteromodal gradient could not fully explain SC-FC coupling. We conclude that the SC-FC coupling of the anterior-posterior cingulate circuit is more similar to unimodal areas than to heteromodal areas.

Keywords: dense sampling; diffusion imaging; functional imaging; individual; structure–function coupling.

Fig. 1

Combined variance contributions by Euclidean distance, path length, and communicability toward FC represented on box and histogram plots. Panels A-C: Boxplot and histogram plot for total variance contribution by Euclidean distance, path length, and communicability toward FC for participants 01(A), 02(B), and 03(C), respectively. Bins on histogram plot represent number of individual-specific parcels in each bin. Across all panels, R² values were scaled to 0.15% of variance explained by SC toward FC, for maximum comparability across participants.

Fig. 2

Combined variance contributions by Euclidean distance, path length, and communicability toward FC represented on individual brains. Panels A-C: Total variance contribution by Euclidean distance, path length, and communicability toward FC for participants 01(A), 02(B), and 03(C). Variance contributions are mapped onto each participant’s specific brain organization, derived using the Infomap community-driven algorithm. R² values were thresholded to 0.15 for easy comparison across participants. Black color indicates R² values = 0.

Fig. 3

Variance contributions toward parcels affiliated with 14 resting state FC networks for participant 01, and 16 resting state FC networks for participants 02 and 03. Panels A-C: Total variance contribution by structural metrics toward parcels affiliated with resting state FC networks for participants 01(A), 02(B), and 03(C).

Fig. 4

Association between structure–function R  ²  for a given region and its position along the macroscale unimodal-heteromodal gradient. Left: The y-axis represents the strength of S-F coupling as the total R² calculated using the multiple linear regression (Vázquez-Rodríguez et al. 2019) approach. The x-axis represents the first principal gradient values obtained using the diffusion embedding (Margulies et al. 2016) procedure. Right: Residuals from the regression plot. PCC and ACC regions exhibited much larger S-F coupling than would be expected given their position on the principal gradient. Residuals are thresholded to values between −0.13 and 0.13, for easy comparison across participants.

Fig. 5

Variance contributions are mapped onto each participant’s specific brain organization, derived using the Infomap community-driven algorithm. Panels A-C: Variance contribution by Euclidean distance toward FC for participants 01(A), 02(B), and 03(C), respectively. Panels D-F: Variance contribution by (weighted) path length toward FC for participants 01(D), 02(E), and 03(F) respectively. Panels G-I: Variance contribution by (weighted) communicability for participants 01(G), 02(H), and 03(I), respectively. R² values were thresholded to 0.15 for easy comparison across participants. Black color indicates R² values = 0.