Shaping brain structure: Genetic and phylogenetic axes of macroscale organization of cortical thickness

Functional and evolutionary patterning converges at the level of macroscale cortical structural organization.


Replication of structural covariance gradients in eNKI dataset
To evaluate whether the observed organizational axes of structural covariance could also be detected in a different dataset with a wider age-range, we evaluated the structural covariance gradients in the eNKI dataset (799 individuals, ages 12-85yrs). Here we observed, similar to the main observations in the HCP dataset, a principal anterior posterior gradient explaining 15% of variance and a secondary gradient traversing from inferior to superior regions explaining 11% of variance. Overall patterns were highly comparable (G1: r=0.81, p_spin<0.001, G2: r=0.88, p_spin<0.001) between HCP and eNKI covariance gradients (Fig S1).

Robustness of principal and secondary gradient
To assess robustness we compared the first two gradients from our main analysis relative to the first two gradients when varying various analysis steps in the gradient computation, exploring different algorithm parameters through the BrainSpace toolbox (29). Other kernels (Pearson, Spearman, Gaussian, or Cosine Similarity) showed all a high correlation between the original G1 and G2 and their respective outputs (r>0.90). Also, the use of an alternative non-linear dimension reduction (Laplacian eigenmap) or linear approach (PCA) give highly similar results (r>0.95). Last, varying the cut-off value of the covariance matrix did not change the outcome (r>0.90).

The third -eight gradient of thickness covariance and genetic correlation of thickness.
Additionally, we studied the third-eight gradient of thickness covariance and genetic correlation of thickness, explaining 5-10% of variance (Fig S3). The third gradient traversed from sensorymotor and mid temporal areas to both frontal and occipital cortices, and a comparable gradient was observed in genetic correlation of thickness. The fourth gradient had a bilateral axis in superior dorsolateral frontal cortex on the one hand and frontal polar, parietal and temporal Shaping Brain Structure polar regions on the other hand. The fifth gradient showed strong lateralization between left temporal parietal regions and right lingual gyrus and corresponded to the sixth gradient of genetic correlation of thickness. The sixth gradient was centered in the right supramarginal gyrus extending to sensory-motor areas on the one hand, and less so in the left sensory cortex, and on the other hand precuneus and para-limbic areas, a similar gradient was not observed in genetic correlation of thickness. The seventh gradient related to sensory-motor, fusiform gyrus and posterior-mid cingulate on the one hand, and temporal regions and precuneus on the other and was most pronounced in the right hemisphere, this gradient was similar to the fifth gradient in coheritability of thickness. The eighth gradient showed a dissociation between temporal parietal regions and posterior-mid cingulate on the one hand, and occipital and sensory regions on the other.

Structural gradients are above and beyond geodesic distance.
Previous work has shown a strong relationship between structural thickness covariance, genetic correlation of cortical thickness, and geodesic distance. Thus, we explored the relationship between organization of structural covariance and geodesic distance. Geodesic distance was defined as the average distance between each of the 400 parcels ipsilaterally (Fig S6). In line with previous reports, we observed a strong relation between structural covariance and geodesic distance (left hemisphere: r=-0.52, p<0.00001, right hemisphere: r=-0.51, p<0.00001).
Moreover, we observed that genetic correlation varied as a function of the organization of
Regressing out effects of distance in covariance in macaques, we observed a correspondence between the principal gradients of structural covariance and the distance-corrected gradients Spearman's r=-0.03) gradient with the dual origin model. Last, comparing the human and macaque distance-corrected covariance gradients, we found a correspondence between the principal gradient in both species (r=0.38, p<0.0005) and the second gradient in both species (r=0.29, p<0.005), but not across gradients (G1macaque-G2human: r=-0.07; G2macaque-G1human: r=0.11).
Relationship between large-scale organization of genetic correlation of regional thickness and microstructure profiles In a last step we evaluated the association between the two main axis of regional covariance topology and cortical microstructure (T1w/T2w) and microstructural covariance gradients (27), in order to qualify and quantify the relation of the observed covariance gradients in thickness to previously reported microstructural organization (27). We probed cortical microstructure at 12 equidistant surfaces sampled between the outer and inner cortical layer in a sub-set of our participants (HCP S900 sample). We observed a strong negative relationship between G1scov and cortical T1w/T2w at all layer depts (-0.34 < r >-0.44) (Fig S7; Table S2). G2scov, however, only showed a significant positive association with the two most outer strata (layer 1: r=0.60, layer 2: r=0.40), but not with layers closer to the GM/WM surface (Fig S7; Table S2).
Subsequently, we probed the association between organizational gradients of within-individual microstructural profile covariance and topological organization of structural covariance of cortical thickness. To do so, we computed the mean microstructural profile covariance (MPC) maps across individuals and preformed gradient decomposition. We observed, as previously reported (27) a primary gradient of cortical microstructural profile covariance traversing a sensory-fugal pattern (22% of variance), and secondary gradient (17% of variance) traversing a pattern from sensory-motor to frontal cortices. We found that the first MPC gradient showed a close correlation with the inferior-superior gradient of genetic covariance of thickness (r=0.62, p_spin<0.005), but not with the posterior-anterior gradient of genetic covariance of thickness (r=-0.02). Conversely, the secondary gradient of MPC was associated with the posterior-anterior gradient of genetic covariance of thickness (r=0.30, p_spin<0.025), but not with the inferior-superior gradient of genetic covariance (r=-0.09, p>0.1).

Functional topography along macro-scale organizational patterns of thickness
We conducted a meta-analysis using the Neurosynth (78) database and estimated the center of gravity across a set of diverse cognitive terms (27,28) along the posterior-anterior and inferiorsuperior macro-scale organization patterns of thickness (Fig S8). In the posterior-anterior gradient we observed a divergence between sensory and visual functions posteriorly and 'working-memory', 'reading', as well as 'motor' and 'action' processing anteriorly. Various terms such as 'emotion' and 'reward' related to both posterior and anterior regions. The inferior-superior gradient on the other hand related to 'motor', 'working memory' and 'action' in superior regions, but 'emotion', 'reward', 'affective', 'pain' in inferior regions.  Table S2. Correlation between layer-dependent T1w/T2w and covariance gradients. Correlation between layer-based T1w/T2w and the first two gradient of thickness covariance (G1 and G2, please see Fig 1).       (25); B) Correlation between geodesic distance and structural covariance between parcels; C) Principal and secondary gradient of geodesic distance; D) Genetic correlation as a function of the binned geodesic distance gradients; E) Correlation between geodesic distance and distance-regressed structural covariance; F) Distance corrected gradient versus covariance gradients reported in Fig 1; G) Covariance gradients while controlling for geodesic distance; H) Distance-corrected gradients versus large-scale functional gradient (28); I) Relation of binned-gradients to distance from archi-and paleocortex; J) Distance-corrected gradients versus laminar differentiation; K) Geodesic distance in Markov parcellation in macaques (32); L) Correlation between geodesic distance and structural covariance between parcels in macaques; M) Principal and secondary gradient of geodesic distance in macaques; N) Covariance as a function of the binned geodesic distance gradients in macaques; O) Correlation between geodesic distance and distance-regressed structural covariance in macaques; P) Distance corrected gradient versus macaque covariance gradients reported in Fig 3; Q) Macaque covariance gradients while controlling for geodesic distance; R) Relation of binned-gradients to distance from archi-and paleocortex in macaques; S) Cross-species comparison transformed human gradient to macaque surface; T) Human versus macaque distance corrected gradient; Right: Scatter plot of both macaque (x-axis) and human (y-axis) distance corrected covariance gradients.

Fig S7. Link between organization of macro-scale organization of thickness and microstructure. A)
Relationship between large-scale organization of genetic correlation of thickness and cortical T1w/T2w; i. T1w/T2w values of equidistant layers between the pial and GM/WM surface and the correlation with the principal and secondary gradient (G1SCOV and G2SCOV) of macro-scale organization of thickness. For visualization purposes only the first (blue), fourth(orange), seventh (yellow), tenth (purple) of 12 probed layers are reported; B) Principal and secondary gradient of microstructure profile covariance (MPC) and the relationship between MPC gradients and G1SCOV and G2SCOV. Fig S8. Neurosynth functional meta-analysis. Meta-analysis maps for diverse cognitive terms were obtained from Neurosynth similar to Margulies et al. (28). We calculated parcel-wise z-statistics, capturing node-term associations, and calculated the center of gravity of each term along the poster-anterior and inferior-superior gradients. The plots depict the average z-score within binned (20-bins) gradient layer of meta-analysis maps.