Multiscale examination of cytoarchitectonic similarity and human brain connectivity

Abstract

The human brain comprises an efficient communication network, with its macroscale connectome organization argued to be directly associated with the underlying microscale organization of the cortex. Here, we further examine this link in the human brain cortex by using the ultrahigh-resolution BigBrain dataset; 11,660 BigBrain profiles of laminar cell structure were extracted from the BigBrain data and mapped to the MRI based Desikan-Killiany atlas used for macroscale connectome reconstruction. Macroscale brain connectivity was reconstructed based on the diffusion-weighted imaging dataset from the Human Connectome Project and cross-correlated to the similarity of laminar profiles. We showed that the BigBrain profile similarity between interconnected cortical regions was significantly higher than those between nonconnected regions. The pattern of BigBrain profile similarity across the entire cortex was also found to be strongly correlated with the pattern of cortico-cortical connectivity at the macroscale. Our findings suggest that cortical regions with higher similarity in the laminar cytoarchitectonic patterns have a higher chance of being connected, extending the evidence for the linkage between macroscale connectome organization and microscale cytoarchitecture.

Keywords: BigBrain; Connectivity; Cytoarchitectonic differentiation; Graph theory; Network; Structural type.

Figure 1.

Overview of the data processing steps. (A) An example of a BigBrain image, selected pair of points, and its closest neighbor pair. (B) A manual BigBrain profile was extracted from the BigBrain image according to randomly selected points. (C) BigBrain profile was registered to the Desikan–Killiany atlas and averaged within each cortical region. (D) The average BigBrain profiles were correlated between every two cortical regions to obtain a similarity matrix. In parallel, we used DWI data (E) and performed fiber tracking (F) to reconstruct the brain structural network (G). BigBrain profile similarity was linked to properties of the brain structural network.

Figure 2.

Association of BigBrain profile similarity with structural connectivity at the edge level. (A) BigBrain profile similarity matrix. (B) Group-weighted structural connectivity matrix. (C) BigBrain profile similarity between interconnected regions was significantly higher than between nonconnected regions (t = 9.36, p < 0.0001). (D) BigBrain profile similarity was positively correlated with connection weight (NOS) of the structural network (r = 0.28, p < 0.0001). (E) BigBrain profile similarity was different among short-, mid-, and long-range connections (F_{(df = 389)} = 9.75, p < 0.0001). Short-range connections showed significantly higher profile similarity than mid- (t = 3.42, p = 0.0007) and long-range connections (t = 4.34, p < 0.0001). (F) Regressing out interregional distance and the mean regional volume and surface area, BigBrain profile similarity still correlated with the connectivity strength (r = 0.27, p < 0.0001). (G) Both intra- and interhemispheric BigBrain profile similarity was higher between connected regions compared with nonconnected regions (Left hemisphere [LH]: t = 8.31, p < 0.0001; right hemisphere [RH]: t = 9.74, p < 0.0001 interhemisphere [LH-RH]: t = 7.53, p < 0.0001). (H) Taking LH, RH, and LH-RH connections separately, BigBrain profile similarity consistently showed correlations with connection strength (r = 0.32, 0.23, and 0.47, separately, all p < 0.0001; *significant differences).

Figure 3.

(A) The pattern of regional BigBrain profile similarity (top) and nodal strength (middle: NOS weights; bottom: streamline density weights). (B) Regional BigBrain profile similarity showed significant correlation with nodal strength (top: NOS, r = 0.56, p < 0.0001; bottom: streamline density, r = 0.37, p = 0.0030).

Figure 4.

The association of regional BigBrain profile similarity with (A) betweenness centrality (r = 0.40, p = 0.0075), (B) clustering coefficient (r = −0.38, p = 0.0033), and (C) mean shortest path length (r = −0.50, p = 0.0001) of the group-weighted network (NOS weights).