Global and localized network characteristics of the resting brain predict and adapt to foreign language learning in older adults

Abstract

Resting brain (rs) activity has been shown to be a reliable predictor of the level of foreign language (L2) proficiency younger adults can achieve in a given time-period. Since rs properties change over the lifespan, we investigated whether L2 attainment in older adults (aged 64-74 years) is also predicted by individual differences in rs activity, and to what extent rs activity itself changes as a function of L2 proficiency. To assess how neuronal assemblies communicate at specific frequencies to facilitate L2 development, we examined localized and global measures (Minimum Spanning Trees) of connectivity. Results showed that central organization within the beta band (~ 13-29.5 Hz) predicted measures of L2 complexity, fluency and accuracy, with the latter additionally predicted by a left-lateralized centro-parietal beta network. In contrast, reduced connectivity in a right-lateralized alpha (~ 7.5-12.5 Hz) network predicted development of L2 complexity. As accuracy improved, so did central organization in beta, whereas fluency improvements were reflected in localized changes within an interhemispheric beta network. Our findings highlight the importance of global and localized network efficiency and the role of beta oscillations for L2 learning and suggest plasticity even in the ageing brain. We interpret the findings against the background of networks identified in socio-cognitive processes.

Figure 1

Distribution of z-scores within each measure of L2 proficiency. Violin plots show the probability density of the data, box plots indicate the median and the respective quartiles. The rhombus at the center represents the mean. Blue and red dots are negative and positive outliers, respectively, defined as values above 1.5 IQR from the median.

Figure 2

Results of the analysis of the relationship between rs connectivity (pre-learning) and L2 changes (complexity, accuracy, fluency). Top row: rs connectivity in alpha-band and L2 complexity change. Bottom row: rs connectivity in beta-band and L2 accuracy change. Left: Cross-subject correspondence between weighted connectivity scores and L2 change in complexity and accuracy. Right: The 30 strongest edges contributing to above relationships, visualized as correlations coefficients of all edges with respect to the fitted model for the respective relationship. Blue indicates negative weights, implying a decrease in L2 performance with stronger connectivity, while red indicates positive weights and hence increase in L2 performance with increase in connectivity. * Adjusted R-squared and p-values as estimated in the PC-based stepwise regression models (see “Results” and Supplementary Material Tables ST3-4 for details). All relationships depicted also survived permutation testing as a second criterion to establish statistical significance (p < .05).

Figure 3

Results of the analysis of the relationship between rs connectivity changes and L2 changes (complexity, accuracy, fluency). Beta-band connectivity changes and L2 changes in fluency were significantly related. Left: Cross-subject correspondence between weighted connectivity scores in beta-band and L2 change in fluency. Right: The 30 strongest edges contributing to this relationship, visualized as in Fig. 2. Blue indicates a decrease in respective L2 performance with stronger connectivity, while red indicates an increase in L2 performance.* Adjusted R-squared and p-values as estimated in the PC-based stepwise regression models (see “Results” and Supplementary Material Tables ST3-4 for details). All relationships depicted also survived permutation testing as a second criterion to establish statistical significance (p < .05).

Figure 4

Schematic illustration of Minimum Spanning Trees (along the lines of Stam et al.). Different configurations consisting of N = 9 nodes and N = 8 edges. (A) Tree with shortest possible diameter and maximum number of leaves. (B) Tree with longest possible diameter and lowest number of leaves. (C) Mixed tree forms with a diameter of 4 (top) and 6 (bottom) and a leaf number of 4 in both cases. For our analyses, diameter and leaf fraction were standardized by the total number of edges and maximum possible leaf fraction, respectively.