White matter brain structure predicts language performance and learning success

Abstract

Individual differences in the ability to process language have long been discussed. Much of the neural basis of these, however, is yet unknown. Here we investigated the relationship between long-range white matter connectivity of the brain, as revealed by diffusion tractography, and the ability to process syntactically complex sentences in the participants' native language as well as the improvement thereof by multiday training. We identified specific network motifs by singular value decomposition that indeed related white matter structural connectivity to individual language processing performance. First, for two such motifs, one in the left and one in the right hemisphere, their individual prevalence significantly predicted the individual language performance, suggesting an anatomical predisposition for the individual ability to process syntactically complex sentences. Both motifs comprise a number of cortical regions, but seem to be dominated by areas known for the involvement in working memory rather than the classical language network itself. Second, we identified another left hemispheric network motif, whose change of prevalence over the training period significantly correlated with the individual change in performance, thus reflecting training induced white matter plasticity. This motif comprises diverse cortical areas including regions known for their involvement in language processing, working memory and motor functions. The present findings suggest that individual differences in language processing and learning can be explained, in part, by individual differences in the brain's white matter structure. Brain structure may be a crucial factor to be considered when discussing variations in human cognitive performance, more generally.

Keywords: Cognitive performance; Language performance; Learning process; White matter; Working memory.

FIGURE 1

Behavioral data. For both, single and double embedding, there is a significant difference between days 4 and 1 (p < .005, corrected). For the double embeddings, there are also significant differences between days 1 and 2 as well as between days 1 and 3 (p < .05, corrected). Adapted from Wang et al. (2021)

FIGURE 2

Network motif 1 in the left hemisphere, whose prevalence before the training correlates with the baseline performance on day 1. (a) Sagittal and axial view (created using BrainNet viewer; Xia et al., 2013) of nine brain areas where the network component has highest (absolute) connectivity (labels according to Glasser et al., , in panel b). (b) Chord plot of strongest connections in the network motif, with line thickness indicating the absolute weight of the connection within the motif and the color indicating the sign (red: positive, blue: negative). Note that connections that have negative weightings in the network motif actually correlate negatively with the motif prevalence. Then, 67 of the 185 brain areas are plotted, the main constituents of the network motif (panel a) are highlighted in color. (c) Regression plot of network prevalence before the training (right singular vector V) against the performance of subjects during the first day of training on the double center embedding task (r = .567, p = .046)

FIGURE 3

Network motif 2 in the left hemisphere, whose changes over the training period correlate with the performance change between days 1 and 4. (a) Sagittal and axial view (created using BrainNet viewer; Xia et al., 2013) of 10 brain areas where the network component has highest (absolute) connectivity (labels according to Glasser et al., , in panel b). (b) Chord plot of strongest connections in the network motif, with line thickness indicating the absolute weight of the connection within the motif and the color indicating the sign (red: positive, blue: negative). Note that connections that have negative weightings in the network motif actually correlate negatively with the motif prevalence. Then, 59 of the 185 brain areas are plotted, the main constituents of the network motif (panel a) are highlighted in color. (c) Regression plot of change in network prevalence between scans before and after training (right singular vector V) against change in the performance of subjects between days 1 and 4 of training on the double center embedding task (r = .606, p = .035).

FIGURE 4

Network motif 3 in the right hemisphere, whose prevalence before the training correlates with the baseline performance on day 1. (a) Sagittal and axial view (created using BrainNet viewer; Xia et al., 2013) of nine brain areas where the network component has highest (absolute) connectivity (labels according to Glasser et al., , in panel b). (b) Chord plot of strongest connections in the network motif, with line thickness indicating the absolute weight of the connection within the motif and the color indicating the sign (red: positive, blue: negative). Note that connections that have negative weightings in the network motif actually correlate negatively with the motif prevalence (and therefore positively with the performance, see panel c). Then, 55 of the 185 brain areas are plotted, the main constituents of the network motif (panel a) are highlighted in color. (c) Regression plot of network prevalence before the training (right singular vector V) against the performance of subjects during the first day of training on the double center embedding task (r = −.575, p = .038)