Abstract
White matter hyperintensities (WMH) are a key hallmark of subclinical cerebrovascular disease and are known to impair cognition. Here, we parcellated WMH using a novel system that segments WMH based on both lobar regions and distance from the ventricles, dividing the brain into a coordinate system composed of 36 distinct parcels ('bullseye' parcellation), and then investigated the effect of distribution on cognition using two different analytic approaches. Data from a well characterized sample of healthy older adults (58 to 84 years) who were free of dementia were included. Cognition was evaluated using 12 computerized tasks, factored onto 4 indices representing episodic memory, speed of processing, fluid reasoning and vocabulary. We first assessed the distribution of WMH according to the bullseye parcellation and tested the relationship between WMH parcellations and performance across the four cognitive domains. Then, we used a data-driven approach to derive latent variables within the WMH distribution, and tested the relation between these latent components and cognitive function. We observed that different, well-defined cognitive constructs mapped to specific WMH distributions. Speed of processing was correlated with WMH in the frontal lobe, while in the case of episodic memory, the relationship was more ubiquitous, involving most of the parcellations. A principal components analysis revealed that the 36 bullseye regions factored onto 3 latent components representing the natural aggrupation of WMH: fronto-parietal periventricular (WMH principally in the frontal and parietal lobes and basal ganglia, especially in the periventricular region); occipital; and temporal and juxtacortical WMH (involving WMH in the temporal lobe, and at the juxtacortical region from frontal and parietal lobes). We found that fronto-parietal periventricular and temporal & juxtacortical WMH were independently associated with speed of processing and episodic memory, respectively. These results indicate that different cognitive impairment phenotypes might present with specific WMH distributions. Additionally, our study encourages future research to consider WMH classifications using parcellations systems other than periventricular and deep localizations.