Epilepsy enhance global efficiency of language networks in right temporal lobe gliomas

Abstract Aims We analyzed the resting state functional magnetic resonance images to investigate the alterations of neural networks in patients with glioma‐related epilepsy (GRE). Methods Fifty‐six patients with right temporal lower‐grade glioma were divided into GRE (n = 28) and non‐GRE groups. Twenty‐eight healthy subjects were recruited after matching age, sex, and education level. Sensorimotor, visual, language, and left executive control networks were applied to generate functional connectivity matrices, and their topological properties were investigated. Results No significant alterations in functional connectivity were found. The least significant discovery test revealed differences only in the language network. The shortest path length, clustering coefficient, local efficiency, and vulnerability were greater in the non‐GRE group than in the other groups. The nodal efficiencies of two nodes (mirror areas to Broca and Wernicke) were weaker in the non‐GRE group than in the other groups. The node of degree centrality (Broca), nodal local efficiency (Wernicke), and nodal clustering coefficient (temporal polar) were greater in the non‐GRE group than in the healthy group. Conclusion Different tumor locations alter different neural networks. Temporal lobe gliomas in the right hemisphere altered the language network. Glioma itself and GRE altered the network in opposing ways in patients with right temporal glioma.

All patients were scanned within 3 days before surgery.
The pipeline was the same as in the previous study 10 and is shown in the Appendix S1.

| Regions of tumor invasion
Each glioma was segmented in its individual space based on the region with hyper-intensity in T2-FLAIR images. The extent of glioma infiltration was manually and independently determined by two neuroradiologists. If the determined regions varied more than 5%, the final decision was made by a third neuroradiologist who had over 20-year clinical experience. Subsequently, all tumor masks were normalized into the standard space of the Montreal Neurological Institute template by using SPM8 software (http://www.fil.ion.ucl. ac.uk/spm/softw are/spm8).

| Regions of interest
Regions of interest (ROIs) were generated from "brainnetome atlas" (http://www.brain netome.org/). 14 This open-access atlas comprises 246 brain regions, to acquire matrices of FC. Four sub-templates were extracted, which were sensorimotor, language, left executive control, and visual networks. The ROIs in these networks invaded by glioma were excluded. Hence, the inaccurate effect of registration was reduced as much as possible. Detailed information on each ROI in presented in Tables S1-S4.

| Network construction
The mean time series between each two ROIs were compared using Pearson correlation, and subsequently, the FC matrices were constructed. Consequently, we obtained four different FC matrices. Using Student t test via non-parametric equivalent to compare the difference of tumor volume between GRE and non-GRE groups.
Using one-way ANOVA test to compare the difference of age between patients groups and healthy group.
Using one-way ANOVA test via non-parametric equivalent to compare the difference of education level between patients groups and healthy group.
Using to chi-square test to compare the differences of gender, tumor location, and IDH status between GRE and non-GRE groups.

| Graph theoretical measurement
Topological properties of the four sub-networks were analyzed using graph theory measurement, which included global properties (the shortest path length, global efficiency, local efficiency, clustering coefficient, transitivity, and vulnerability), nodal properties (nodal efficiency, nodal local efficiency, nodal clustering coefficient, degree centrality), and small-worldness. 10,15,16 The details of properties were shown in part 2 of the Appendix S1. All matrices were absolutized and binarized to further analyze the topological properties.

| Statistical analyses
Epidemiology characteristics were compared among the GRE, non-GRE, and healthy groups by using Student's t test, Mann-Whitney U test, chi-squared test, Fisher's exact test, and one-way analysis of variance (one-way ANOVA) based on categories of data. All data were tested to ensure whether they were normal/Gaussian distribution. If a group of data did not exhibit a normal distribution, a Student t test or one-way ANOVA test was applied with a non-parametric equivalent.
The differences in FC of the four sub-networks were generated from comparisons between the patient and healthy groups using Student's t test. Moreover, false discovery rate (FDR) was applied to correct the generated results. To found differences in topological properties, we used a series of sparsity thresholds (from 0.17 to 0.33, interval 0.01) consistent with a previously study. 4 For each subject, topological properties were generated according to sparsity. Each property was first analyzed using one-way ANOVA test. Subsequently, post hoc pairwise comparisons were performed on the generated results in global and nodal properties with least significant difference (LSD) test. A significant p-value was lower than 0.05.

| Demographic characteristics
Fifty-six patients met the inclusion criteria, and four patients were excluded, as their periods of anti-epileptic drug use were longer than 30 days. According to the history of preoperative GRE onset, 28 patients were divided into the GRE group (male, n = 11) and the others into the non-GRE group (male, n = 14,

| Functional connectivity differences
Our results revealed no differences in FC of the four sub-networks (sensorimotor, visual, language, and left executive control networks) among the three groups after FDR correction.
Post hoc analysis with the LSD test ( Figure 1

| Differences in small-worldness properties
In the language network, the value of lambda (p < 0.0001) differed among the groups, as determined using one-way ANOVA (Table S5 and

| Differences in nodal topological properties
One-way ANOVA revealed two nodes in the right hemisphere that had differing nodal efficiencies among the three groups in the language network (Table S6 and  Regarding A39rv_R, the non-GRE group had weaker nodal efficiency

| D ISCUSS I ON
In this study, we investigated alterations in functional neural networks induced by right temporal GRE. Our findings indicated that GRE and right temporal DLGGs resulted in altered language networks. Although the altered network differed from the left temporal GRE change (visual network), the trend of right temporal DLGGs and GRE-induced functional network change was the same as that of the left. That is, the GRE-induced functional network change was found to be opposite of that induced by DLGG.
Global efficiency and shortest path length reflect the ability and cost of conveying information, respectively. 17 In our findings, right temporal glioma decreased global efficiency and increased the shortest path length of the language network in the non-GRE group. These changes were related to a neural pathway disruption caused by glioma infiltration. Indeed, the main language network is located in the left hemisphere in right-handed people. 18 The fMRI results suggested that when the left language network was damaged, functional compensation occurred in the cortex of the right hemisphere corresponding to the left language regions. 19 These findings indicate that parts of the language network are located on the right and cooperated with the left network to accomplish language tasks. 20,21 Consequently, if the right-sided language network was damaged by glioma, the global efficiency of the whole language network was decreased in the non-GRE group. Indeed, compared with the GRE and healthy groups, the clustering coefficient and local efficiency in the non-GRE group were increased. These alterations reflect the increase in the number of functional connections, but this does not mean that the pathway between two nodes was shortened or the ability to convey information was increased. Moreover, the right temporal glioma failed to disrupt the main language network (left hemisphere) and was unable to further induce language deficits. Hence, the language network did not drastically reorganize in the non-GRE group. Therefore, we infer that the global decline in the efficiency of the language network was the result of the damage caused by glioma, which increased the burden of the residual language network.
However, in the GRE group, global efficiency did not differ from that in the healthy group. Why was there no decrease in global efficiency in the GRE group? These alterations were associated with GRE-induced network reorganization, but they did not indicate that GRE facilitated recovery of the language network. Unlike primary epilepsy with a long and frequent onset history, the period from onset of GRE to glioma diagnosis and tumor resection is short.
Hence, cortical sclerosis, 22 gray-matter atrophy, 23 and cortical F I G U R E 5 Results of alterations in nodal properties of left nodes in the language network when gliomas grew in the right temporal lobe. The grp GRE = group of patients with glioma-related epilepsy. The grp non-GRE = group of patients without glioma-related epilepsy. The grp healthy = group of healthy participants

| LI M ITATI O N S
The phenomenon that levetiracetam normalized FC was found in patients with primary epilepsy who took levetiracetam over 3 months. 28 In our GRE group, all patients indeed took levetiracetam, but the period of administration was short (not longer than 15 days). To our knowledge, no study has revealed whether taking levetiracetam in a short period would alter topological properties. Hence, we could not determine whether alterations of topological properties in patients with GRE tended to normalize due to levetiracetam administration.
In the future, we will enroll relevant patients to investigate whether taking levetiracetam for a short period can induce alterations of the functional network in patients with GRE and validate the findings in this study.

| CON CLUS IONS
Different tumor locations alter different neural networks.
Temporal lobe gliomas in the right hemisphere altered the lan-

ACK N OWLED G EM ENTS
Thanks to Dr. Meng Lanxi for imaging data acquisition.

CO N FLI C T S O F I NTE R E S T
All authors do not have any conflict of interest.

AUTH O R CO NTR I B UTI O N
SF, YW, and TJ. conceptualized and designed the study. SF and YW.
acquired and analyzed the data, involved in statistics/verified analytical method, and wrote the final draft. YW and TJ. supervised the study. All authors read and approved final version.

DATA AVA I L A B I L I T Y S TAT E M E N T
Anonymized data will be made available on request.