Between January 2016 and July 2018, 70 patients from Tiantan Hospital with a primary diagnosis of prefrontal lobe glioma were retrospectively enrolled in the study. Inclusion criteria were as follows: (a) aged older than 18 years; (b) more than 6 years of school education; (c) no history of antiepileptic drug use; and (d) no history of biopsy, radiotherapy, or chemotherapy. The exclusion criteria were as follows: (a) contraindications for MRI; (b) head motion greater than 3 mm in translation or 3° in rotation; and (c) gliomas involving bilateral prefrontal lobes or leading midline shift.

Finally, 65 patients with prefrontal gliomas were enrolled in the study (29 male and 36 female, left hemisphere n = 35, right hemisphere n= 30). All patients were divided into the GRE and non-GRE groups. Preoperative electroencephalograms (4 hours) were performed in the patients who did not have a clear history of seizures (patients in non-GRE group, n = 36). Moreover, 25 right-handed healthy participants matched to the patients for age, sex, and education level which comprised the control group (12 men and 13 female).

We retrospectively collected patient characteristics from inpatient records including age, sex, education level, Karnofsky performance status, histopathology, isocitrate dehydrogenase mutation status, the extent of tumor resection, information regarding preoperative seizures, type of seizure onset, and history of taking antiepileptic drugs. Follow-up information about epileptic control was obtained via telephone interviews at 1 year postoperatively.

All MR images were acquired using a MAGNETOM Prisma 3T MRI scanner (Siemens, Erlangen, Germany). T1-magnetization prepared rapid acquisition gradient echo was applied to collect anatomical images [flip angle: 8°; repetition time (TR): 2,300 ms; echo time (TE): 2.3 ms; field of view (FOV): 240 × 240 mm²; voxel size: 1.0 × 1.0 × 1.0 mm³; slice number: 192]. T2-weighted sequences were used to acquire glioma images (flip angle: 150°; TR: 5,000 ms; TE: 105 ms; FOV: 240 × 240 mm²; voxel size: 0.5 × 0.5 × 3 mm³; slice number: 33). Finally, an echo-planar imaging sequence was applied for rs-fMRI (flip angle: 75°; TR: 2,000 ms; TE: 30 ms; FOV: 220 × 220 mm²; voxel size: 3.0 × 3.0 × 5.0 mm³; slice number: 30; acquisition duration: 8 min, closed eyes during scanning). We acquired all MRI data within 72 h before tumor resection.

The Graph Theoretical Network Analysis software (https://www.nitrc.org/projects/gretna) (16) was applied for rs-fMRI preprocessing. For each participant, the steps of preprocessing were as follows: a) data transformation (from Digital Imaging and Communications in Medicine to Neuroimaging Informatics Technology Initiative); b) removal of the first five images; c) slice time correction; d) realignment; e) spatial normalization (normalized to echo-planar imaging template) (17); f) spatial smoothing (full width half maximum = 4 mm); g) temporal detrending (linear detrending); h) regressing out covariance (white matter signal: with WMMask_3 mm; cerebrospinal fluid signal: with CSFMask_3 mm; head motion: Friston—24 parameters); i) temporal filtering (0.01–0.1 Hz); and j) scrubbing (using default parameters and the interpolation strategy: linear interpolation, FD threshold = 0.5, previous time point number = 1, subsequent time point number = 2).

The extent of low-grade glioma invasion was manually drawn based on the region of high signals on T2 fluid-attenuated inversion. All tumor masks were then normalized to the Montreal Neurological Institute standard space using the clinical toolbox package in SPM8 (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/), which are shown in Figure 1 . The tumor volume was calculated with the volumetric method based on the individual tumor mask.

To calculate FC within cerebral functional networks, regions of interest (ROIs) were extracted from an open-access brain atlas, “brainnetome atlas” (http://www.brainnetome.org/) (18), which comprises 246 brain regions. To avoid the possibility that tumor invasion may affect the results of FC, the subnetworks that were invaded by gliomas and glioma-related edema were not analyzed. Therefore, in the current study, sub-templates were extracted, including the sensorimotor, visual, auditory, and left/right executive networks (based on tumor location). Details of ROIs are provided in Tables S1 – S5 .

Pearson’s correlation coefficients were used to compare regional mean time series for all possible pairs of nodes that constructed the FC matrix. Consequently, four different FC matrices were generated from the four sub-templates of the sensorimotor, visual, auditory, and left/right executive networks.

Global and nodal topological properties, including cluster coefficient, global efficiency, the shortest path length, local efficiency, nodal efficiency, nodal local efficiency, and small-worldness properties, were calculated for the GRE, non-GRE, and healthy groups using the graph theory analysis (19). All matrices were transformed into absolute value matrices before calculating topological properties. A total of 10,000 random networks were used to calculate the topological properties.

Small-worldness properties, including gamma, lambda, and sigma, represent the efficiency of information delivery: gamma (γ) = C_real/C_random >> 1 (C represented cluster coefficient), lambda (λ) = L_real/L_random ~ 1 (L represented shortest path length), and sigma (σ) = γ/λ > 1 (20). A high value of sigma represents a high efficiency of information delivery.

We compared the GRE, non-GRE, and healthy groups to identify which alterations were caused by each factor using SPSS (19.0 version, IBM) and GraphPad Prism 7 (GraphPad Software Inc., San Diego, USA). According to the type of data, clinical characteristics were compared between the GRE and non-GRE groups using a two-sample t-test, chi-square test, and one-way analysis of variance (ANOVA) which were used to compare among the GRE, non-GRE, and healthy groups.

When comparing the differences of FCs among the three groups, we used one-way ANOVA and post-hoc test with Bonferroni correction. Moreover, an eta-squared correlation was applied to explore the relationship between GRE onset and FC value.

Additionally, a series of sparsity thresholds (from 0.17 to 0.34, interval 0.01) was used to explore group differences in network topological properties, which was consistent with relevant studies (14, 15). For each property, one-way ANOVA was first applied. Subsequently, a post-hoc analysis with the Bonferroni correction was used if the results of ANOVA showed a significant difference (significant p-value was lower than 0.05). Clinical information (age, sex, and education level) was regressed out during statistical comparison in FC and topological properties.

Based on whether patients had preoperative GRE, 29 patients were included in the GRE group (15 and 14 patients with prefrontal gliomas in the left and right hemisphere, respectively). Additionally, 36 patients comprised the non-GRE group (20 and 16 patients with prefrontal gliomas in the left and right hemispheres, respectively) ( Tables 1 , 2 ). No significant differences were observed regarding age, sex, years of education, and Karnofsky performance status between the three groups. Moreover, no statistical difference in tumor volume and isocitrate dehydrogenase mutation status was found between the GRE and non-GRE groups. Furthermore, our postoperative follow-up data showed that no patient with preoperative GRE experienced epilepsy at 1 year after tumor resection. All patients achieved Engel class I.

The FC was compared among the GRE, non-GRE, and healthy groups in the matrices of sensorimotor, visual, auditory, and left/right executive networks (based on tumor location). Except for FC in the sensorimotor network, no significant differences in FC in the other three networks were noticed after Bonferroni correction. In total, 231 functional edges belonged to the sensorimotor network.

When gliomas were located in the left prefrontal lobe, all these six edges were significantly different among the three groups in post-hoc analysis with Bonferroni correction ( Figure 2 and Table S6 ).

Compared with the non-GRE group [0.357 ± 0.055 (mean ± mean standard error)], only one edge had lower FC in the GRE group (0.025 ± 0.057) after post-hoc analysis with Bonferroni correction (p < 0.0001), which connected A6m_L [medial Brodmann area (BA) 6 in the left hemisphere] and Acvl_L (caudal ventrolateral BA 6 in the left hemisphere). Moreover, compared with the healthy group, six edges were identified with significantly lower FC in the GRE group after post-hoc analysis with Bonferroni correction, which connected 1) A6m_L and Acvl_L (Healthy, 0.399 ± 0.037, p < 0.0001); 2) Acvl_R (caudal ventrolateral BA 6 in the right hemisphere) and A6m_L (GRE, 0.108 ± 0.050; Healthy, 0.418 ± 0.047, p = 0.0027); 3) A6m_R (medial BA 6 in the right hemisphere) and Acvl_L (GRE, 0.074 ± 0.064; Healthy, 0.428 ± 0.043, p = 0.0003); 4) A4tl_R (tongue and larynx region in the right hemisphere) and Acvl_L (GRE, 0.131 ± 0.047; Healthy, 0.436 ± 0.040, p = 0.0069); 5) A1/2/3tonla_R (tongue and larynx region of BA 1/2/3 in the right hemisphere) and Acvl_L (GRE, 0.063 ± 0.061; Healthy, 0.329 ± 0.050, p = 0.0091); and 6) A4ul_L (upper limb region of BA 4 in the left hemisphere) and A4ul_R (upper limb region of BA 4 in the right hemisphere) (GRE, 0.431 ± 0.077; Healthy, 0.812 ± 0.057, p = 0.0009). Moreover, no edge of FC was significantly different between the non-GRE and healthy groups.

When gliomas were located in the right prefrontal lobe, three edges of FC were significant different among the three groups in post-hoc analysis with Bonferroni test ( Figure 3 and Table S7 ).

Compared with the non-GRE group, two edges that were from A6m_R to Acvl_R (GRE, 0.118 ± 0.048; non-GRE, 0.437 ± 0.033, p = 0.0002) and A6m_R to A4hf_L (head and face regions of BA 4 in the left hemisphere; GRE, 0.252 ± 0.042; non-GRE, 0.657 ± 0.083, p = 0.0007) were identified with significantly lower FC in the GRE group. Moreover, compared with the healthy group, three edges were identified with significantly lower FC in the GRE group, which connected 1) A6m_R and Acvl_R (Healthy, 0.429 ± 0.048, p < 0.0001); 2) A4hf_L and A6m_R (Healthy, 0.575 ± 0.050, p = 0.0014); and 3) A1/2/3ulhf_R (upper limb, head, and face regions of BA 1/2/3 in the right hemisphere) and A4ul_L (GRE, 0.359 ± 0.089; Healthy, 0.691 ± 0.051, p = 0.0042). Additionally, no edge of FC was significantly different between the non-GRE and healthy groups.

Our results showed a negative correlation between FC of the functional edge connected A6m_L to Acvl_L when the glioma was located in the left hemisphere (r = -0.590, p < 0.0001, eta-squared correlation). A similar result was found when the glioma was located in the right hemisphere (functional edge: connected A6m_R and Acvl_R, r = -0.541, p < 0.0001).

When gliomas were located in the left prefrontal hemisphere, there were some differences in global efficiency (p = 0.0118) and shortest path length (p = 0.0306) among the three groups in the sensorimotor network in one-way ANOVA ( Table S8 and Figure 4 ). After post-hoc analysis with Bonferroni correction, the non-GRE group (0.391 ± 0.009) showed weaker global efficiency than the GRE group (0.430 ± 0.012, p = 0.0113). Moreover, compared with the non-GRE group (2.747 ± 0.081), the shortest path length was significantly shorter in the GRE group (2.473 ± 0.079, p = 0.0292).

When gliomas were located in the right prefrontal lobe, there were some differences in global efficiency (p < 0.0001), shortest path length (p < 0.0001), local efficiency (p < 0.0001), and clustering coefficient (p < 0.0001) among the three groups in the sensorimotor network by one-way ANOVA ( Table S9 and Figure 5 ). After a post-hoc analysis with Bonferroni correction, in the GRE group (0.521 ± 0.003), the global efficiency was significantly greater than those in the non-GRE group (non-GRE, 0.494 ± 0.007, p = 0.0214). Additionally, the shortest path length in the GRE group (1.958 ± 0.029) was shorter than in the non-GRE group (non-GRE, 2.170 ± 0.044, p = 0.0129). Moreover, compared with the healthy group, the GRE and non-GRE groups showed significantly greater global efficiency (hHealthy = 0.416 ± 0.006; GRE vs. healthy, p < 0.0001; and non-GRE vs. healthy, p < 0.0001), local efficiency (GRE = 0.662 ± 0.013, non-GRE = 0.624 ± 0.010, Healthy = 0.554 ± 0.010, GRE vs. healthy, p < 0.0001, and non-GRE vs. healthy, p < 0.0001), and clustering coefficient (GRE = 0.548 ± 0.012, non-GRE = 0.525 ± 0.009, Healthy = 0.475 ± 0.009; GRE vs. healthy, p < 0.0001, and non-GRE vs. healthy, p = 0.0012) after post-hoc analysis with Bonferroni correction. The shortest path length was significantly shorter in the GRE and non-GRE groups than in the healthy group (GRE, 1.959 ± 0.029, non-GRE, 2.170 ± 0.044, Healthy, 2.589 ± 0.047; GRE vs. healthy groups, p < 0.0001, and non-GRE vs. healthy groups, p < 0.0001).

In addition, the gamma (p = 0.0021) and lambda (p < 0.0001) values showed a significant difference among the three groups in one-way ANOVA. Furthermore, after the post-hoc analysis with the Bonferroni correction test, the value of gamma was found significantly higher in the healthy group (1.684 ± 0.088) than that in the non-GRE group (1.288 ± 0.050, p = 0.0017). The value of lambda was found significantly higher in the healthy group (1.143 ± 0.016) than in the GRE (1.029 ± 0.004, p < 0.0001) and non-GRE groups (1.054 ± 0.009, p < 0.0001).

No difference of global topological properties in the other three networks (visual, auditory, and left/right executive) was found among the three groups.

We analyzed the nodal efficiency and nodal local efficiency among the three groups ( Tables S10 , S11 ) whether the gliomas were located in the left or right hemisphere. However, no significant difference was found among these three groups in nodal properties by one-way ANOVA. Moreover, no difference in nodal topological properties was found among these three groups in the other three networks (visual, auditory, and left/right executive).