White matter maturation reshapes structural connectivity in the late developing human brain

Abstract

From toddler to late teenager, the macroscopic pattern of axonal projections in the human brain remains largely unchanged while undergoing dramatic functional modifications that lead to network refinement. These functional modifications are mediated by increasing myelination and changes in axonal diameter and synaptic density, as well as changes in neurochemical mediators. Here we explore the contribution of white matter maturation to the development of connectivity between ages 2 and 18 y using high b-value diffusion MRI tractography and connectivity analysis. We measured changes in connection efficacy as the inverse of the average diffusivity along a fiber tract. We observed significant refinement in specific metrics of network topology, including a significant increase in node strength and efficiency along with a decrease in clustering. Major structural modules and hubs were in place by 2 y of age, and they continued to strengthen their profile during subsequent development. Recording resting-state functional MRI from a subset of subjects, we confirmed a positive correlation between structural and functional connectivity, and in addition observed that this relationship strengthened with age. Continuously increasing integration and decreasing segregation of structural connectivity with age suggests that network refinement mediated by white matter maturation promotes increased global efficiency. In addition, the strengthening of the correlation between structural and functional connectivity with age suggests that white matter connectivity in combination with other factors, such as differential modulation of axonal diameter and myelin thickness, that are partially captured by inverse average diffusivity, play an increasingly important role in creating brain-wide coherence and synchrony.

Fig. 1.

Change of several diffusion MRI parameters with respect to developmental age averaged over all brain white matter connections at low resolution. (A) ADC at b = 3,000 s/mm² (R = −0.9078, P < 10⁻¹¹), (B) ADC at b = 1,000 s/mm² (R = −0.6204, P < 0.0005), and (C) FA at b = 1,000 s/mm² (R = 0.5659, P < 0.005). R is the Pearson correlation coefficient.

Fig. 2.

Relationship of network metrics and developmental age. Results shown are for cerebral cortex at two spatial resolutions: (A) n = 66 and (B) n = 241 nodes. For whole-brain data (cortex and deep gray structures) see Fig. S2. Scatter plots show node strength, global efficiency, clustering coefficient, and modularity (left to right). All measures are computed from the weighted SC matrices of individual subjects. Values for clustering coefficient and small-world index are scaled relative to populations of 100 random networks with preserved degree and weight distributions. For R and P values, see Table 1.

Fig. 3.

Relationship of the rank of node betweenness centrality for several cortical regions and developmental age. Dark dots and dashed regression lines represent the right hemisphere and light dots and stippled lines the left hemisphere. Centrality ranks remain largely unchanged. None of the relationships shown here exhibit significant trends, with the exception of the left IP. PCUN, precuneus; PC, posterior cingulate cortex; SP, superior parietal cortex; IP, inferior parietal cortex; MOF, medial orbitofrontal cortex; SF, superior frontal cortex; ST, superior temporal cortex; LOCC, lateral orbital cortex.

Fig. 4.

Modularity and SC–FC correlation. (A) Cortical (n = 241) SC and FC matrices averaged over the younger (<4 y) and older (>13 y) age group. Structural modules are delineated by the superimposed white grid. Although modules are highly conserved (normalized mutual information = 0.82), there is a notable increase in SC–FC correspondence from younger to older brains. Eleven modules (M1–M6 in the right hemisphere, M7–M11 in the left hemisphere) were identified, and the two sets of SC and FC matrices are displayed such that modules correspondence across age is maximized. Modules are centered on the following anatomical locations: M1, occipital cortex; M2, parietal cortex; M3, parietal cortex; M4, orbitofrontal cortex; M5, frontal cortex; M6, temporal cortex; M7, occipital cortex; M8, parietal cortex; M9, orbitofrontal cortex; M10, frontal cortex; and M11, temporal cortex. (B) Increasing statistical relationship between SC and FC across age (R = 0.74, P < 0.005). Data reported in this plot are for SC that was resampled into a Gaussian distribution as described in ref. , and for FC corresponding to raw Pearson cross-correlations between time series. The relationship persists for unresampled SC (R = 0.66, P < 0.01), whole-brain unresampled SC (R = 0.59, P < 0.05), or for FC that has been Fisher z-transformed and normalized (R = 0.74, P < 0.005).