Testing the structural disconnection hypothesis: Myelin content correlates with memory in healthy aging

Abstract

Introduction: The "structural disconnection" hypothesis of cognitive aging suggests that deterioration of white matter (WM), especially myelin, results in cognitive decline, yet in vivo evidence is inconclusive.

Methods: We examined age differences in WM microstructure using Myelin Water Imaging and Diffusion Tensor Imaging in 141 healthy participants (age 20-79). We used the Virginia Cognitive Aging Project and the NIH Toolbox® to generate composites for memory, processing speed, and executive function.

Results: Voxel-wise analyses showed that lower myelin water fraction (MWF), predominantly in prefrontal WM, genu of the corpus callosum, and posterior limb of the internal capsule was associated with reduced memory performance after controlling for age, sex, and education. In structural equation modeling, MWF in the prefrontal white matter and genu of the corpus callosum significantly mediated the effect of age on memory, whereas fractional anisotropy (FA) did not.

Discussion: Our findings support the disconnection hypothesis, showing that myelin decline contributes to age-related memory loss and opens avenues for interventions targeting myelin health.

Keywords: Aging; Cognition; Disconnection; Executive Functions; Memory; Myelin; Speed; Structural; Water.

Figure 1.

Participant Flow Diagram.

Figure 2.

Regions of Interest. CP = cerebral peduncles, PLIC = posterior limb of the internal capsule, SLF = superior longitudinal fasciculus, CC = corpus callosum.

Figure 3.

Voxel-wise Associations between Myelin and Memory. Note. Voxelwise analysis using FSL’s randomise. This figure displays significant clusters (TFCE > 0.95, corresponding to p < 0.05). The color bar indicates increasing TFCE values, where warmer colors closer to yellow denote areas of higher statistical significance. Each panel represents a different axial slice of the brain, with slice numbers provided at the top of each image.

Figure 4.

Age-Related Trends in Myelin Content in Late and Early Myelinating Regions. Note. Blue lines represent either quadratic trends or a loess-smoothed trend for the genu corpus callosum. The loess method produces a more robust curve that is less sensitive to extreme values. Red markers indicate turning points or local maxima. R² values are shown for plots with quadratic trends. However, R² is not applicable for loess-smoothed plots, as loess is a nonparametric method that does not provide a straightforward measure of goodness-of-fit like R² in linear regression models.

Figure 5.

Relationship between Age and Cognitive Factor Scores. Note. Each point shows a subject’s average cognitive score, measured in standard deviation units. Cognitive scores are standardized, with a score of zero representing the average performance for our sample. Scores above zero reflect better-than-average performance, while scores below zero indicate lower-than-average performance. The dashed blue line shows linear regression fit for the entire sample, while the solid red line represents the quadratic regression fit derived from a polynomial adjustment of the second degree (y ~ poly(x, 2)) for the entire sample. The shaded grey areas indicate 95% confidence intervals.

Figure 6.

Parallel Mediation of Age-Related Memory by Myelin Content. Note. Memory is measured by Logical Memory, Paired Associates, List Sorting, and Auditory Verbal Learning Task (AVLT). Solid arrows indicate direct paths with standardized coefficients (β), rectangles represent observed variables, ovals denote latent constructs, and curved arrows show residual variances. Sex is coded as 1 for men, 2 for women. Model fit indices for each panel are as follows: Prefrontal: CFI = 1.000, TLI = 1.026, RMSEA = 0.000 (95% CI: 0.000, 0.065), SRMR = 0.040; Genu: CFI = 1.000, TLI = 1.026, RMSEA = 0.000 (95% CI: 0.000, 0.065), SRMR = 0.040; PLIC: CFI = 1.000, TLI = 1.012, RMSEA = 0.000 (95% CI: 0.000, 0.073), SRMR = 0.038.