The focal alteration and causal connectivity in children with new-onset benign epilepsy with centrotemporal spikes

The aim of the current study was to find the epileptic focus and examine its causal relationship to other brain regions in children with new-onset benign childhood epilepsy with centrotemporal spikes (BECTS). Resting-state functional magnetic resonance imaging (fMRI) was performed in 66 children with BECTS and 37 matched control children. We compared the amplitude of low frequency fluctuation (ALFF) signals between the two groups to find the potential epileptogenic zone (EZ), then used Granger causality analysis (GCA) to explore the causal effects of EZ on the whole brain. Children with BECTS had significantly increased ALFF in the right Broca’s area, and decreased ALFF in bilateral fusiform gyrus. The patients also showed increased driving effect from the EZ in Broca’s area to the right prefrontal lobe, and decreased effects to the frontal lobe and posterior parts of the language network. The causal effect on left Wernicke’s area negatively correlated with verbal IQ (VIQ) score. Our research on new-onset BECTS patients illustrates a possible compensatory mechanism in the language network at early stages of BECTS, and the negative correlation of GCA and VIQ suggest the disturbance of epileptiform activity on language. These findings shed light on the mechanisms of and language dysfunction in BECTS.

Functional neuroimaging has produced a more detailed assessment of the language networks and their lateralization that is consistent with the results provided by the gold standard methods 28 . Amplitude of low frequency fluctuation (ALFF) signals can detect regional spontaneous activity by measuring the magnitude of spontaneous blood oxygenation level dependent (BOLD) signals, ultimately reflecting brain activity in a time series. In epilepsy patients ALFF is positively correlated with the number of epileptic discharges, and can thus be used for localizing epileptic foci in the mesial temporal lobe, other cortical and subcortical structures associated with cognitive impairment 29,30 , and possibly regional language networks 31 . Functional connectivity analyses have revealed reduced connectivity between the sensorimotor and language networks in children with BECTS 12,32 .
While static network models aim to describe the structures of brain networks 33,34 , dynamic causal modeling in fMRI implies causation and effective connectivity (directed effects from one region to another) 35 . Granger causality analysis (GCA) is used for inferring the direction of information flow in these brain networks 36,37 . One study on BECTS patients found an increased driving effect from the ROI zone to the right medial frontal cortex and posterior cingulate cortex and decreased causal effects to left inferior frontal gyrus 16 . That study was conducted with patients receiving antiepileptic drugs (AED), however, and AED treatment might interfere with local brain activity 38 . To eliminate the potential effects of AED and avoid the possible influence of long term aberrations in spikes on network connectivity 39 , here we conducted an fMRI study on new-onset treatment-naïve children within three months of their diagnosis with BECTS. We hypothesized that a comparison of the global ALFF signal between patients and healthy controls would find focal functional alterations in the key components of language networks and that the driving effect from the affected zone(s) would illustrate information flow alterations (intraand inter-hemisphere) within the language framework.

Results
Neuropsychological testing. Table 1 shows the demographic, clinical and neuropsychological characteristics of the patients with BECTS and healthy controls. There was no significant difference between patients and control group in FSIQ or PIQ. The VIQ scores in the children with BECTS were slightly lower than the control group but this difference was not significant (t = 1.475, p = 0.143).
Voxel-wise GCA. Seed-to-whole-brain analysis. We chose to set the affected right Broca's area as the seed zone for the analysis. In the seed-to-whole-brain analysis we identified a number of cortical areas and subcortical structures that were driven by the seed region in patients with BECTS (Fig. 2b). The pattern in the healthy controls is demonstrated in Fig. 2a. Compared to healthy controls, the patients showed an increased driving effect from the seed area to the caudate and prefrontal cortex (BA11), along with decreased effects on the bilateral lingual gyrus (BA18), right middle temporal gyrus (BA22), left angular gyrus (BA21) and left hippocampus (BA36; p < 0.05, GRF corrected, two tailed; Fig. 2c; details in Table 2).These results concur with the existing language framework.
Whole-brain-to-seed analysis. Whole-brain-to-seed analysis showed that there was no abnormal positive or negative driving effect from whole brain to seed in the patients with BECTS.

Correlation analysis.
We performed correlation analysis between the VIQ score and the GCA map of the patients (Fig. 3). This analysis revealed a negative correlation between VIQ score and the driving flow from right Broca's area to the left Wernicke's area (t = −0.451, p < 0.05, GRF corrected, two tailed).

Discussion
This study on new-onset treatment-naïve BECT patients found increased local brain activity in the right Broca's area. From that seed area, we then found increased causal effects to the caudate, thalamus, and prefrontal lobe, as well as decreased effects to the middle temporal lobe, angular gyrus, lingual gyrus, and hippocampus; all of which are within the known language network. We also found a negative correlation between VIQ scores and the driving effect to right Wernicke's area. In contrast to some previous reports, we didn't find any significant difference in IQ test between two groups. We found that the patients showed higher ALFF signal in the right Broca's area and decreased signal at bilateral fusiform gyrus. The classical neuropsychological model implies that language is a rigid and monolithic function 40 , however, more recently neuroimaging studies have broadened our perspective on language system 41 . Current opinion suggests that language processing is underpinned by a flexible cerebral network including areas beyond the classical language regions in both hemispheres 41,42 , and that the language system is based on both inter-hemispheric and intra-hemispheric connections 43 . In accordance with this view, the functional network in patients with focal epilepsy might shift from the pathological spike-affected areas to adjacent areas and homologues 19 . Converging evidence also suggests that chronic epileptic activity can result in a developmental shift of language processing from the left to the right hemisphere or event re-route language pathways from traditional to non-traditional areas 44 . The increased ALFF activity that we observed in the right Broca's area may imply a compensatory role for the right Broca's area at the beginning stages of the disease. This view is supported by findings of right homologue area recruitment during language tasks in focal epilepsy patients 22 . Research combining fMRI and DTI has also noted that the right Broca's area serves as a hub in the long-distance dorsal pathway along the arcuate and superior longitudinal fascicle (AF/SLF) and as a key node in high-order cognitive processing 23 .
The decreased ALFF signal observed in fusiform gyrus might illustrate a down-regulation of peripheral components within the dynamic language processing network. The fusiform gyrus is part of the "visual word form area"in occipital lobe 45 , and together with other components, it plays a supporting role in speech and language processing 46 . In our study the deficit of fusiform gyrus recruitment might illustrate the impact of pathological spikes on language system and might underlie the language dysfunction noted in BECTS patients.
GCA analysis has previously been used in epilepsy patients to analyze the EEG and fMRI signal and it enables us to understand better how seizure activity initiates, propagates, and terminates, providing an overview of information flow within the epilepsy network 47,48 , though there is some controversy over how to improve its accuracy 49 . In our study the driving effect from the right Broca's area illustrates that the flexible language system may be based on both core areas including key language components and peripheral areas such as the fusiform gyrus and the para-hippocampal gyrus; these are connected via inferior fronto-occipital fasciculus (IFOF), which serves as an important portion of ventral pathway 50 . One recent theoretical framework proposes that functional interactions between brain regions might change over time, and that the patterns of inter-regional communication may determine the degree of functional specialization for a given region 51 -that is, peripheral areas that are transiently engaged during linguistic processing may come to help complete the processing of language tasks 52,53 . This theory explains the involvement of posterior network areas including angular gyrus, fusiform gyrus, occipital lobe and mid-temporal gyrus in many high-order cognitive tasks including language 54 .
The increased ability for Broca's area to drive frontal lobe coupled with its decreased driving effect on posterior areas may imply elevated information stream density in the frontal-parietal circuit (including frontal lobe and Broca's area) and disrupted connections from core components to peripheral components of the language network including the occipital lobe, angular gyrus and others. Similar findings have been observed in new-onset BECT patients using the Regional Homogeneity (ReHo) approach 55 . This also concurs with other observations that the frontal lobe fails to deactivate task negative network components in epilepsy patients 56 . A study on children with focal epilepsy also found that the ventral language network might be more vulnerable under epilepsy conditions while the dorsal language network was relatively unaffected 57 . A study of subcortical electrical stimulation on the anterior part of the arcuate fascicle linking the inferior parietal and frontal cortices also showed that rostral components can regulate the language system 58 . The imbalance of the rostral-caudal components might be due to the disturbance of temporal segregation of auditory and visual perception and thus of speech comprehension and production, which may further underlie the potential language impairment in long-term epilepsy 24 .
Overall IQ performance did not differ significantly between the patients and the healthy volunteer. This is in accordance with previous findings in children with new-onset BECTS 59 , and may reflect compensation by the increased Broca-frontal connection observed. The short duration of epilepsy in BECTS patients may also contribute to the negative finding 60 . The negative correlation between VIQ and the GCA effect on left Wernicke's area, however might illustrate the disturbance of the right Broca's area. Previous studies have revealed the long-path of Wernicke's area and its right-lateralized connection in the language network 24 . Researchers hypothesize that these connections help Wernicke's area modulate multiple-language information [25][26][27] . The abnormal driving effect from right Broca's area might thus interfere with VIQ performance.
One limitation of the current work is that a cross sectional cannot capture the longitudinal development of the delicate balance of the language network, which would likely provide more evidence of the reorganization and rebalance of these components over time. We also found no significant differences between patients and healthy controls on IQ scores. This could be explained by a compensatory mechanism, or could reflect the relatively limited number of patients represented in this sample. In the future we would like to enlarge the study size to address this question.
This study demonstrates for the first time the imbalance of the language system in the early stages of the disease and a possible rebalancing of the language system under these conditions. We found that children with new-onset BECTS illustrate an increased driving effect from the right Broca's area to the frontal lobe and sub-cortical areas and decreased effects to the caudal parts of the flexible language network. The overall IQ scores did not differ significantly between the patients and healthy controls, but the driving effect from right Broca's area to left Wernicke's area does seem to interfere with VIQ performance, and these disruption may underlie future language dysfunction in these patients. A follow-up longitudinal study may better illustrate the natural history of BECTS and its comorbidities, which will deepen our knowledge of brain development under pathological conditions.  Table 2.
Regions showing abnormal causal effect with epileptogenic zone in patients (seed-to-whole-brain). BA = Brodmann's area; L = left side; R = right side; MNI = Montreal Neurological Institute coordinate; Midtem: middle temporal gyrus; Prefrotal: pre-frontal lobe; NA: not available; +: increased causal effect from seed; −: decreased causal effect from seed; The last two columns show the t value of the corresponding peak voxel within the patient and control group, respectively. Values with "※" show that the mean causal effect of the corresponding cluster is significantly different from zero.   62 , which synthesizes procedures in the Resting State fMRI Data Analysis Toolkit (REST; http://www.restfmri.net) 63 and Statistical Parametric Mapping (SPM8; www. fil.ion.ucl.ac.uk/spm). The first 10 images were removed to ensure steady-state longitudinal magnetization, and the remaining images were then corrected for temporal differences and head motion. After subject selection, neither translation nor rotation parameters exceeded ±2 mm or ±1.5°. The functional images were then registered into 3D T1 image of each participant at a resolution of 3 × 3 × 3 mm 3 . We then spatially smoothed the images with a 6 mm full-width half-maximum isotropic Gaussian kernel, to regress out nuisance signals (white matter, cerebrospinal fluid signals, and 6 head-motion parameters). Finally, linear trends were removed from the time courses and temporal band-pass filtering was performed (0.01-0.08 Hz) ALFF analysis. ALFF is the averaged square root of activity in the low-frequency band (0.01-0.08 Hz) [18].
The ALFF value of each voxel was standardized by dividing the value by the whole-brain mean ALFF value. Two-sample T tests were used to compare the differences in ALFF between the patients and control groups. Using the Gaussian Random Field theory correction (GRF) built into the REST software, we then applied a corrected significance level of voxel level p < 0.05 and cluster level p < 0.05 (GRF corrected, two-tailed).

GCA.
A cluster showing increased ALFF was identified in right Broca's area, and the peak voxel (with a 3 mm radius) in this cluster was used as the seed region for the GCA analysis. The voxel-wise coefficient GCA was performed in the whole brain using REST-GCA, a REST software plug-in ref. 63 . We applied bivariate coefficient GCA to investigate the causal relation between the seed zone and each voxel in the entire brain. One-sample T tests were performed for the causal effects within each group, with an uncorrected significance level of p < 0.05. Two-sample T-tests were then performed on the causal effects between groups using the GRF corrected significance level of voxel level p < 0.05 and cluster level p < 0.05 (two-tailed). To explore whether the neuroimaging measures correlated with the disease features in BECTS children, Pearson's correlation was performed between the causal effect map and VIQ scores at global range.