Key question:

Cortical thinning is an important hallmark of the maturation of brain morphology during the transition period from childhood to adolescence. A typical maturation sequence is marked by the hierarchical thinning from the primary cortex to the association cortex, which is thought to be mediated by cellular mechanisms, genetic regulation, and biomechanical factors. However, the white matter connectome-based wiring mechanism that underlies such cortical maturations remains unclear. Using connectomics, transcriptomics, and computational modeling analyses on 521 brain scans from 314 participants aged 6-14 years, we present a mechanistic approach to model how the maturational pattern of cortical morphology is shaped by WM connectome architecture from childhood to adolescence.

Methods:

All participants were recruited from the Beijing Cohort in Children Brain Development project and divided into either child or adolescent group using age 10 years as a cutoff. Individual cortex was parcellated into 1000 nodes (219 and 448 nodes as a validation) based on the modified Desikan-Kiliany atlas, and the maturation extent of nodal CT was defined as the T value of the group-wise difference of nodal CT in a mixed linear analysis with sex as covariate. Then, we reconstructed the WM connectome of each child using dMRI-based deterministic tractography and further generate a binary, group-level WM connectome that preserves individual edge length distributions.

First, we identify whether the maturation of nodal CT was related to its direct neighbors in WM network by assessing the correlation between the maturation extent of nodal CT and the mean maturation extent of its directly WM-connected neighbors. We test the significance of such spatial correlations against two baseline null models including a spatial permutation (“spin”) model and a rewiring model (both 1000 times).

Next, we proposed a graph-based diffusion model to simulate the nodal axonal interactions during cortical development by estimating probabilities of a node to other nodes at the nth neighboring scale during random walks (the maximum n: the network diameter) as the nodal diffusive profiles. A support vector regression model was further trained with nodal diffusive profiles at the nth scale as input features to separately predict the nodal CT maturation extent. To identify the dominant regions that lead cortical development, we calculated the cosine similarity between the nodal CT maturation map and the nodal diffusion profiles.

Finally, we used the BrainSpan and Allen Human Brain Atlas datasets to evaluate the differences in the expression levels of genes associated with several neural development events between dominant and non-dominant regions.

Main findings:

Significant nodal cortical thinning is mainly located in the dorsolateral prefrontal, lateral temporal, and parietal regions (t > 4.10, P_bonf < 0.05). A significant correlation was found between the nodal CT maturation extent and the mean of its directly connected neighbors (r_adj = 0.74, P < 0.001, all p_rewired < 0.001, all p_spin < 0.001) at three nodal resolutions. We also employed another two statical models that capturing cortical thinning among different age points and within individuals. Highly reproducible results were found.

In diffusion model analyses, we found the diffusive profiles of a node could significantly predict its CT maturation extent at multiple neighboring scales (r_{1-9 scale}: ranged from 0.65 to 0.75, all p_spin <= 0.001, all p_rewired < 0.001) with higher prediction accuracies at lower scales that highlight community segregation process. The most consistent dominators (p_spin < 0.05) across neighboring scales distributing in bilateral lateral prefrontal parietal, and temporal regions. By exhibiting the step-wise diffusive processes of the two most robust dominators separately in prefrontal and inferior parietal regions, we found that dominator nodes mainly interacted with neighbors within nearby systems at low scales.

Using BrainSpan datasets, we found the transcription level of dominant regions was significantly higher for dendrite (p = 0.014) and synapse development (p = 0.002) but significantly lower for axon development (p < 0.001) and myelination (p < 0.001) than that of nondominant regions. Considering that the BrainSpan dataset only contains 11 sampling neocortex, we validate the analyses using Allen Human Brain Atlas (AHBA) dataset and found a significantly correlated gene list with positive correlations mainly enriched in biological process such as learning or memory and synapse organization as well as cellular component such as glutamatergic synapse, neuron spine, and somatodendritic compartment (all P < 0.05, FDR corrected).

To evaluate the reproducibility of our findings, we replicated all the main analyses using the cross-sectional, multi-site replication dataset from the Lifespan Human Connectome Project in Development (HCP-D) project, which included 301 subjects aged 5 to 14 years. All results were highly consistent.

Conclusions

In conclusion, using neuroimaging, connectomics, transcriptomics, and computational modeling, we found that the maturational pattern of cortical morphology from childhood to adolescence is structurally constrained by the large-scale WM connectome architecture and that such constraints are predominantly located in frontoparietal nodes and are linked with the expression of genes associated with microstructural developmental processes. Thus, our results provide mechanistic insights into the maturation of cortical morphology during development.