Changes in white matter functional networks across late adulthood

Abstract

Introduction: The aging brain is characterized by decreases in not only neuronal density but also reductions in myelinated white matter (WM) fibers that provide the essential foundation for communication between cortical regions. Age-related degeneration of WM has been previously characterized by histopathology as well as T2 FLAIR and diffusion MRI. Recent studies have consistently shown that BOLD (blood oxygenation level dependent) effects in WM are robustly detectable, are modulated by neural activities, and thus represent a complementary window into the functional organization of the brain. However, there have been no previous systematic studies of whether or how WM BOLD signals vary with normal aging. We therefore performed a comprehensive quantification of WM BOLD signals across scales to evaluate their potential as indicators of functional changes that arise with aging.

Methods: By using spatial independent component analysis (ICA) of BOLD signals acquired in a resting state, WM voxels were grouped into spatially distinct functional units. The functional connectivities (FCs) within and among those units were measured and their relationships with aging were assessed. On a larger spatial scale, a graph was reconstructed based on the pair-wise connectivities among units, modeling the WM as a complex network and producing a set of graph-theoretical metrics.

Results: The spectral powers that reflect the intensities of BOLD signals were found to be significantly affected by aging across more than half of the WM units. The functional connectivities (FCs) within and among those units were found to decrease significantly with aging. We observed a widespread reduction of graph-theoretical metrics, suggesting a decrease in the ability to exchange information between remote WM regions with aging.

Discussion: Our findings converge to support the notion that WM BOLD signals in specific regions, and their interactions with other regions, have the potential to serve as imaging markers of aging.

Keywords: BOLD; ICA; fMRI; normal aging brain; resting state; white matter (WM).

FIGURE 1

Forty ICs that are derived from group ICA are displayed in three orthogonal planes. The color map overlaid represents the Z score, reflecting the degree of membership of the voxel to the IC. The brighter color indicates a higher Z value. Here only voxels with Z > 2 are displayed.

FIGURE 2

Relationship between within-IC FC and age. Eight ICs that show significant changes across age are shown (p < 0.05, Bonferroni correction). Each panel visualizes the spatial distribution of the IC, as well as scatter plots that represent the age of the subjects (x-axis) versus the within-IC FCs (y-axis). Note that the y-axis does not represent the raw FC values but the residues after gender and head motions are regressed out. The color reflects the density of the scatters. The hotter color indicates a higher density.

FIGURE 3

Pair-wise IC connections whose FC show significant correlations with age. The panel on the left displays the functional links whose FCs significantly decrease over age (p < 0.05, Bonferroni correction). The thicker lines indicate higher r (absolute) values. The ICs involved in the top 8 most significant changes are highlighted in blue, with their distribution maps shown beside the IC labels. Coincidently IC 5 is involved in all those 8 connections of interest so that it is highlighted with a blue rectangle. The panel on the right displays the functional links whose FCs significantly increased over age (p < 0.05, Bonferroni correction). The thicker lines indicate higher r (absolute) values with age. ICs involved in the most significant changes are highlighted in blue, with their distribution maps shown beside the IC labels.

FIGURE 4

Relationship between network metrics and age. Panel (A) displays the relationship between cluster coefficients and age. Each data point on the radar chart indicates r value. Panel (B) displays the relationships between network efficiencies and age. Each data point on the radar chart indicates r value. Panel (C) displays the relationship between network strength and age. Each data point on the radar chart indicates r value. The distributions of IC 5, 6, 10, 20, and 24, whose spatial distributions are visualized in this panel, are considered IC of interest as they exhibit the closest relationships with age in the case of all three measurements. Panels (D,E) display the correlation between global network metrics, including global efficiency and characteristic path length, and age. P-values have been corrected by the Bonferroni method. Note that the y-axis does not represent the raw metrics but the residues after gender and head motions are regressed out.

FIGURE 5

The variation in circuit configuration over age. Panel (A) display the mean FC matrices corresponding to six age groups. The nodes (ICs) are sorted according to baseline circuit configuration calculated based on the youngest age group (40–50 years). The distributions of the three circuits are displayed in panel (B), with labels of ICs that are involved in different circuits shown above. Panel (C) shows how the within- (first row) and inter- (second row) circuits FC vary with age (groups). Each box visualizes the median, 25, and 75 percentile in the FC values of subjects within an age group.

FIGURE 6

Independent components (ICs) that switch their membership to the circuits over age. IC 13 is a member of circuit 1 but propagates to Circuit 2 at older ages. IC 5 is a member of circuit 2 but propagates to circuit 1 at older ages. IC 30 is a member of circuit 3 but propagates to Circuit 2 at older ages.

FIGURE 7

The variation in spectral powers over age. Upper panel: the r values (in descending order) corresponding to the 23 ICs in which the mean low-band powers decreased significantly with aging. Lower panel: four representatives corresponding to the highest r (absolute) values. Note that the y-axis does not represent the raw power values but the residues after gender and head motions are regressed out. P-values have been corrected by the Bonferroni method.