The longitudinal relation between executive functioning and multilayer network topology in glioma patients

Abstract

Many patients with glioma, primary brain tumors, suffer from poorly understood executive functioning deficits before and/or after tumor resection. We aimed to test whether frontoparietal network centrality of multilayer networks, allowing for integration across multiple frequencies, relates to and predicts executive functioning in glioma. Patients with glioma (n = 37) underwent resting-state magnetoencephalography and neuropsychological tests assessing word fluency, inhibition, and set shifting before (T1) and one year after tumor resection (T2). We constructed binary multilayer networks comprising six layers, with each layer representing frequency-specific functional connectivity between source-localized time series of 78 cortical regions. Average frontoparietal network multilayer eigenvector centrality, a measure for network integration, was calculated at both time points. Regression analyses were used to investigate associations with executive functioning. At T1, lower multilayer integration (p = 0.017) and epilepsy (p = 0.006) associated with poorer set shifting (adj. R² = 0.269). Decreasing multilayer integration (p = 0.022) and not undergoing chemotherapy at T2 (p = 0.004) related to deteriorating set shifting over time (adj. R² = 0.283). No significant associations were found for word fluency or inhibition, nor did T1 multilayer integration predict changes in executive functioning. As expected, our results establish multilayer integration of the frontoparietal network as a cross-sectional and longitudinal correlate of executive functioning in glioma patients. However, multilayer integration did not predict postoperative changes in executive functioning, which together with the fact that this correlate is also found in health and other diseases, limits its specific clinical relevance in glioma.

Keywords: Cognition; Eigenvector centrality; Functional connectivity; Graph theory; Network neuroscience.

Fig. 1

Schematic overview of the MEG multilayer analysis pipeline. For every participant, magnetoencephalography (MEG) data was preprocessed and projected to the brain; the brain was parcellated according to the Automated Anatomical Labeling (AAL) atlas; the Phase Lag Index (PLI) was used to compute weighted connectivity matrices; minimum spanning trees (MST) of the weighted matrices were constructed using Kruskal’s algorithm; and finally an LxN by LxN supra-adjacency matrix representing a multilayer network of the six MEG frequency bands as layers (all with N = 78 nodes for each AAL region and M = N – 1 = 77 binary intralayer links) was constructed, where diagonal blocks contain the intralayer connections for each frequency band and the off-diagonal blocks the interlayer connections; like the intralayer connections, we set all interlayer link weights to 1, obtaining binary multilayer networks; now, multilayer eigenvector centrality (EC) of the frontoparietal network (FPN) was calculated and averaged for each patient and timepoint. NPA = neuropsychological assessment, EF = executive functioning

Fig. 2

Executive functioning and multilayer integration at both time points. Each panel shows a paired raincloud plot, in which individual data points of each patient at both time points are displayed through the combination of a scatterplot (the ‘rain’), a spaghetti plot, a box plot, and a probability density plot (the ‘cloud’). Panel A shows patients’ Z-scores of the three executive functioning tests. Scores below the dashed line at -1.5 indicate clinically relevant cognitive deficits. Only word fluency changed significantly at the group-level (p = 0.002). Panel B shows patients’ multilayer eigenvector centrality (EC) of the frontoparietal network, which did not change significantly at the group-level.

Fig. 3

Significant associations between multilayer integration and set shifting. (A) Displays the cross-sectional association between multilayer eigenvector centrality (EC) of the frontoparietal network and set shifting performance (n = 35, model p = 0.003), with color codes indicating the included covariate (presence of epilepsy). (B) Shows the longitudinal association between change in multilayer EC (T2-T1) and change in set shifting (T2-T1; n = 28, model p = 0.006), with color codes indicating the included covariate (active chemotherapy at T2). Shaded area: 95% confidence interval.