Altered Integration of Structural Covariance Networks in Young Children With Type 1 Diabetes

Abstract

Type 1 diabetes mellitus (T1D), one of the most frequent chronic diseases in children, is associated with glucose dysregulation that contributes to an increased risk for neurocognitive deficits. While there is a bulk of evidence regarding neurocognitive deficits in adults with T1D, little is known about how early-onset T1D affects neural networks in young children. Recent data demonstrated widespread alterations in regional gray matter and white matter associated with T1D in young children. These widespread neuroanatomical changes might impact the organization of large-scale brain networks. In the present study, we applied graph-theoretical analysis to test whether the organization of structural covariance networks in the brain for a cohort of young children with T1D (N = 141) is altered compared to healthy controls (HC; N = 69). While the networks in both groups followed a small world organization-an architecture that is simultaneously highly segregated and integrated-the T1D network showed significantly longer path length compared with HC, suggesting reduced global integration of brain networks in young children with T1D. In addition, network robustness analysis revealed that the T1D network model showed more vulnerability to neural insult compared with HC. These results suggest that early-onset T1D negatively impacts the global organization of structural covariance networks and influences the trajectory of brain development in childhood. This is the first study to examine structural covariance networks in young children with T1D. Improving glycemic control for young children with T1D might help prevent alterations in brain networks in this population. Hum Brain Mapp 37:4034-4046, 2016. © 2016 Wiley Periodicals, Inc.

Keywords: brain networks; diabetes; graph theoretical analysis; gray matter volume; network efficiency; network resilience.

© 2016 Wiley Periodicals, Inc.

Figures

Figure 1

Global network properties. Changes in network clustering (A) and path length (B) as a function of network sparsity. The 95% confidence intervals and between‐group differences in network clustering (C) and path length (D) as a function of network sparsity. Red circles falling outside of the confidence interval indicate network sparsities at which path length is significantly different between groups. AUC analysis revealed that the area under the path length curve was significantly longer in T1D network. [Color figure can be viewed at http://wileyonlinelibrary.com.]

Figure 2

Structural covariance networks in T1D and HC. The size of each node represents the degree of connectivity. While the network path length was significantly altered in the T1D group compared with HC, we did not find any significant difference in regional network measures between groups after correction for multiple comparisons. Desikan–Killiany Atlas included with FreeSurfer was used as parcellation scheme. The names of the 86 included regions are as follow: Bilateral cerebellum, thalamus, caudate nucleus, putamen, pallidum, hippocampus, amygdala, nucleus accumbens, ventral diencephalon, bank of superior temporal sulcus, caudal anterior cingulate, caudal middle frontal cortex, cuneus, entorhinal cortex, fusiform gyrus, inferior parietal lobule, inferior temporal cortex, isthmus cingulate, lateral occipital, lateral orbitofrontal, lingual gyrus, medial orbitofrontal, middle temporal gyrus, parahippocampal gyrus, paracentral lobule, pars opercularis, pars orbitalis, pars triangularis, pericalcarine fissure, postcentral gyrus, posterior cingulate, precentral gyrus, precuneus, rostral anterior cingulate, rostral middle frontal cortex, superior frontal cortex, superior parietal lobule, superior temporal cortex, supramarginal gyrus, frontal pole, temporal pole, transverse temporal, and insula. [Color figure can be viewed at http://wileyonlinelibrary.com.]

Figure 3

Degree distributions. The log–log plot of cumulative degree distributions of (A) T1D and (B) HC networks thresholded at sparsity of 0.09. The solid line indicates the exponentially truncated power‐law curve fitted to the cumulative degree distribution of the networks (dotted line). The estimated exponent was 1.88 for T1D and 1.81 for HC, the cut‐off degree was 3.05 for T1D, and 3.27 for the HC network. These parameters resulted in R‐square values of 0.98 and 0.96 for T1D and HC distributions, respectively (value close to one represents a good fit). [Color figure can be viewed at http://wileyonlinelibrary.com.]

Figure 4

Between‐group differences in network robustness. A: Changes in the size of the largest component of the network after incrementally removing random nodes. Stars indicate where the difference in the size of the largest remaining components between groups is significant at each increment. B: The 95% confidence interval and between‐group differences in the AUC of network resilience curves revealed that the T1D network is significantly less robust to insult than HC network. [Color figure can be viewed at http://wileyonlinelibrary.com.]

Figure 5

Stability analysis. Changes in AUC of (A) normalized path length and (B) normalized clustering coefficient as a function of sample size for subgroups of T1D (filled circles) and HC (open squares). Both path length and clustering coefficient asymptotically decrease for small group‐size (n <25) and become stable for larger subgroups (n >30). [Color figure can be viewed at http://wileyonlinelibrary.com.]