White matter tract correlations with spoken language in cerebrovascular disease

Abstract

Assessment of spoken language is a promising marker for cognitive impairment in individuals with cerebrovascular disease. However, the underlying neurological basis for spoken language beyond single words and sentences remains poorly defined in this cohort, particularly with respect to white matter. This study aimed to examine and compare white matter hyperintensity volumes and diffusion tensor metrics in normal-appearing white matter (NAWM) as potential correlates of spoken language performance. Baseline imaging and spoken language data were obtained from the cerebrovascular disease cohort of the Ontario Neurodegenerative Disease Research Initiative (n = 127; age: 55-85 years). Most participants had subclinical or very mild strokes, with very little to no aphasia symptoms. Spoken language samples were analysed to compute 10 different measures related to syntax, productivity, lexical diversity, fluency, and information content. Structural and diffusion MRI data were analysed to segment white matter hyperintensities and tracts. Normalized white matter hyperintensity volume, as well as average fractional anisotropy and mean diffusivity in the normal-appearing portion of eight white matter tracts, were correlated with the 10 spoken language measures using canonical correlation analyses. White matter and spoken language variate scores for individual participants then were correlated separately in male (n = 86) and female (n = 41) participants to probe potential sex differences. Spoken language performance was significantly associated with the fractional anisotropy (r_c = 0.51, P = 0.041) and mean diffusivity (r_c = 0.56, P = 0.011) of NAWM, particularly in the left superior longitudinal fasciculus, but not with white matter hyperintensity volumes (r_c = 0.41, P = 0.80) in the same tracts. Measures related to syntax, fluency, and information content loaded most strongly in the spoken language variate. No significant sex differences were found in NAWM microstructure, and female and male participants exhibited similarly strong associations between spoken language and NAWM microstructure (fractional anisotropy: z = 1.44, P = 0.15; mean diffusivity: z = 1.03, P = 0.30). These results suggest that diffusion MRI in NAWM may be superior to white matter hyperintensity volumetrics when evaluating the role of white matter tract integrity on cognitive outcomes in people with relatively mild cerebrovascular pathology. These results also demonstrate that multi-domain spoken language analysis is sensitive to underlying white matter microstructure in participants with cerebrovascular disease without significant aphasia, supporting its value as a tool for assessing cognitive status.

Keywords: cerebrovascular disease; diffusion tensor imaging; language; magnetic resonance imaging; white matter.

Graphical Abstract

Figure 1

Sample MR images. Examples of the various types of MR images acquired (T₁, T₂, FLAIR, PD) or generated (SABRE-LE, FA, MD) in the current study are shown for a single participant with cerebrovascular disease. The SABRE-LE image shows the segmentation of normal-appearing grey matter (beige), normal-appearing white matter (green), white matter hyperintensities (orange), periventricular lacune (lime green), ventricular CSF (blue), and sulcal CSF (brown). FLAIR = fluid-attenuated inversion recovery; PD = proton density; SABRE-LE = Semi-Automatic Brain Region Extraction-Lesion Explorer; FA = fractional anisotropy; MD = mean diffusivity.

Figure 2

Extraction of normal-appearing white matter (NAWM) in selected white matter pathways implicated in spoken language. Shown are segmentations of four white matter tracts (ILF, SLFp, SLFt, and UNC) obtained using TRACULA before (yellow) and after (magenta) removal of white matter lesions and non-NAWM, along with corresponding T₁-weighted anatomical images. The tract segmentations are superimposed on the Semi-Automatic Brain Region Extraction-Lesion Explorer (SABRE-LE) images (beige = normal-appearing grey matter; green = NAWM; orange = white matter hyperintensities; lime green = periventricular lacune; blue = ventricular CSF; brown = sulcal CSF). For each tract, mean diffusion tensor imaging (DTI) metrics were calculated in NAWM only. Note that white matter pathway streamlines are visualized in Supplementary Fig. 1. ILF = inferior longitudinal fasciculus; SLFp = superior longitudinal fasciculus—parietal bundle; SLFt = superior longitudinal fasciculus-temporal bundle; UNC = uncinate fasciculus.

Figure 3

Canonical correlation loadings of diffusion tensor imaging and spoken language measures in individuals with cerebrovascular disease. Canonical correlations examined the association between diffusion tensor imaging (DTI) metrics in both hemispheres of the brain and spoken language performance. The canonical loadings of each DTI and spoken language variable onto their respective variates are shown for (A) fractional anisotropy (n = 127, r_c = 0.509, P  = 0.041) and (B) mean diffusivity (n = 127, r_c = 0.557, P  = 0.011). Spoken language variables can be categorized into several performance domains, as shown in the legends. Tracts with larger magnitude canonical loadings are interpreted as having stronger associations with spoken language measures that also had larger magnitude canonical loadings. Note that, depending on the variable, a negative loading does not necessarily reflect worse performance. ILF = inferior longitudinal fasciculus; SLFp = superior longitudinal fasciculus (parietal bundles); SLFt = superior longitudinal fasciculus (temporal bundles); UNC = uncinate fasciculus; MLU = mean length of utterance (in words); SI = subordination index; Clauses/Ut = mean number of clauses per utterance; WPM = words per minute; MATTR = moving-average type-token ratio; % Maze words = percentage of maze words; Wd dys/Ut = word-level dysfluencies per utterance; % Main events = proportion of main events; CIUs/min = correct information units per minute.

Figure 4

Canonical variate scores of females and males. Canonical variate scores from the CCAs of diffusion tensor imaging (DTI; Variate 1) and spoken language performance (Variate 2) were calculated for each participant (represented as individual dots). The scores of females (n = 41) and males (n = 86) were then correlated separately to examine potential sex differences in the association between DTI and spoken language. These sex-specific variate score correlations (both Pearson r and Fisher z-scores), along with 95% confidence ellipses, are shown for the CCAs of (A) fractional anisotropy and (B) mean diffusivity. Males and females had similarly strong correlations for both FA (z = 1.442, P = 0.149) and MD (z = 1.032, P = 0.302).