Alterations of brain network hubs in reflex syncope: Evidence from a graph theoretical analysis based on DTI

Abstract Objective We evaluated global topology and organization of regional hubs in the brain networks and microstructural abnormalities in the white matter of patients with reflex syncope. Methods Twenty patients with reflex syncope and thirty healthy subjects were recruited, and they underwent diffusion tensor imaging (DTI) scans. Graph theory was applied to obtain network measures based on extracted DTI data, using DSI Studio. We then investigated differences in the network measures between the patients with reflex syncope and the healthy subjects. We also analyzed microstructural abnormalities of white matter using tract‐based spatial statistics analysis (TBSS). Results Measures of global topology were not different between patients with reflex syncope and healthy subjects. However, in reflex syncope patients, the strength measures of the right angular, left inferior frontal, left middle orbitofrontal, left superior medial frontal, and left middle temporal gyrus were lower than in healthy subjects. The betweenness centrality measures of the left middle orbitofrontal, left fusiform, and left lingual gyrus in patients were lower than those in healthy subjects. The PageRank centrality measures of the right angular, left middle orbitofrontal, and left superior medial frontal gyrus in patients were lower than those in healthy subjects. Regarding the analysis of the white matter microstructure, there were no differences in the fractional anisotropy and mean diffusivity values between the two groups. Conclusions We have identified a reorganization of network hubs in the brain network of patients with reflex syncope. These alterations in brain network may play a role in the pathophysiologic mechanism underlying reflex syncope.

the three-dimensional diffusion of water as a function of spatial location, which can reflect microstructural tissue status and orientation (Alexander, Lee, Lazar, & Field, 2017). This is made possible because water diffusion in tissues is highly sensitive to differences in the microstructural architecture of cellular membranes (Alexander et al., 2017). This sensitivity makes DTI a powerful method for detecting microscopic differences in tissue properties (Alexander et al., 2017). Although DTI has been applied to various diseases, no studies have investigated microstructural architecture in patients with reflex syncope.
We sought to evaluate the global topology and regional hubs reorganization of brain network in patients with reflex syncope using graph theoretical analysis based on DTI. This approach has many advantages. The reproducibility of graph theory metrics from structural connectomes based on DTI is significantly more accurate than studies of functional connectivity based on electroencephalography, magnetoencephalography, or resting state-functional MRI, which have significant variability over time and a high level of withinsubject variability (Deco, Jirsa, & McIntosh, 2011;Gleichgerrcht, Kocher, & Bonilha, 2015;Mehrkanoon, Breakspear, & Boonstra, 2014). In additional, DTI studies can focus on the quantification of intuitive measurements of axonal fibers, revealing direct structural reorganization and connection, whereas studies based on cortical thickness or volumes can only evaluate structural associations indirectly, using statistical dependence or correlation (Deco et al., 2011;Gleichgerrcht et al., 2015;Mehrkanoon et al., 2014). In addition, we investigated microstructural abnormalities of white matter using the diffusion tensor and scalar values, fractional anisotropy (FA) and mean diffusivity (MD), in patients with reflex syncope compared to healthy subjects. Our hypothesis was that there were alterations of the brain network or white matter microstructures in patients with reflex syncope.

| Subjects
This study was conducted with the approval of our institution's institutional review board. This study was conducted prospectively in a single tertiary hospital. Twenty patients diagnosed with reflex syncope were recruited from March 2017 to December 2017. Only patients who satisfied the following criteria were enrolled: 1) two or more episodes of clinically diagnosed reflex syncope, confirmed by a neurologist, based on the patient's clinical history and head-up tilt table test, 2) absence of abnormal findings on electrocardiography, echocardiography, or electroencephalography, and 3) normal brain MRI scan, by visual inspection. These patients were free of any structural heart disease, arrhythmias, diabetes mellitus, or neurologic disease, and none were taking medications that might affect autonomic function. We also enrolled an age-and sex-matched control group of 30 healthy subjects with no significant past medical, neurological, or psychiatric history.

| Image processing and statistical analysis
Graph theoretical analysis was performed using DSI Studio (http:// dsi-studio.labsolver.org). Nodes were defined as anatomical regions, and edges were defined by fiber density. Graph theoretical analysis was performed as follows. First, a tractography was generated from the DTI data, which entails reading and parsing DICOM files, image reconstruction to characterize the major diffusion direction of the fiber, and fiber tracking. Afterwards, the connectivity matrix was generated, which was calculated from the count of the connecting tracts. The Automated Anatomical Labeling (AAL) template was used as the brain parcellation, and every white matter fiber was evaluated to determine its extreme points. This step included acquiring a whole brain fiber track, which placed the seeding at the whole brain level, spatial normalization, and definition of the regions of interest, and building the connectivity matrix. At last, we calculated the global graph theoretical network measures from the connectivity matrix, including the mean clustering coefficient, characteristic path length, small-worldness, global efficiency, and local efficiency, to obtain quantitative information regarding the global network properties.
In addition, we also obtained measures of strength, betweenness centrality, and PageRank centrality to investigate changes in hub organization. We investigated differences in the graph theoretical network measures between patients with reflex syncope and the healthy subjects. A p-value <0.05 was considered significant for all calculations. Comparisons were analyzed with a Student's t test. All of the statistical tests were performed using MedCalc ® (MedCalc Software version 17.8, Ostend, Belgium).
To perform tract-based spatial statistics analysis (TBSS) analysis, all raw DTI data were preprocessed with FSL (http://www.fmrib. ox.au.uk/fsl). First, eddy current distortions and head motion were corrected by spatially normalizing all the diffusion-weighted images.
Afterwards, skull-stripping was applied to exclude nonbrain tissues and regions. At last, we computed the diffusion tensor, as well as the scalar measures, including fractional FA and MD values (FA and MD values were analyzed with protocols provided by TBSS). We normalized individual FA volumes of the two groups to the MNI template space via affine registration. The aligned FA images were averaged to yield a mean FA image and then thinned to create the FA skeleton of the mean FA image. The skeleton represented the common tract pattern of all participants from the two groups. The FA threshold (0.2) was then set on the skeleton to exclude gray matter and cerebral spinal fluid from the final analysis. The FA image for each subject was then projected onto the skeleton. The significance threshold for between-group differences was set at p < 0.05 using threshold-free cluster enhancement in the FSL "randomize" permutation-testing tool (5000 permutations). Regional FA differences were then local-

| Demographic and clinical characteristics of subjects
Of the 20 patients with reflex syncope, 10 (50%) were men and 10 (50%) were women. The mean age was 37.4 ± 14.9 years. Among the 30 healthy control subjects, 15 subjects (50%) were men and 15 subjects (50%) were women. The mean age was 37.4 ± 6.7 years. All healthy subjects had a normal neurological examination and a normal brain MRI on visual inspection. The age and sex of the healthy subjects were similar to those of the reflex syncope patients (p = 0.9957 and p = 1.000, respectively).

| Measures presenting topology and hubs organization
The measures of global topology, such as mean clustering coefficient, characteristic path length, small-worldness, global efficiency, and local efficiency, in patients with reflex syncope were not different from those in healthy subjects (Table 1). However, there were significant differences in the reorganization of hubs organization in the patients with reflex syncope compared to healthy subjects (

| Analysis of DTI scalar values
Analysis of the DTI scans showed no significant differences in FA and MD scalar values in patients with reflex syncope and healthy controls (Figure 1), indicating that brain white matter is similar in the two groups.

| D ISCUSS I ON
To our knowledge, this is the first study to explore the topological organization of brain networks in patients with reflex syncope. We calculated centrality measures for various regions in the brain and discovered that many, predominantly in the frontal lobe, were significantly decreased in patients with reflex syncope compared to healthy subjects, which suggested that there was a reorganization of the brain network hubs in patients with reflex syncope. These alterations in brain network may play a role in the pathophysiologic mechanism underlying reflex syncope.
Although the pathophysiologic mechanism of reflex syncope is not well understood, especially the afferent part of the reflex, the efferent part of the reflex seem to be elucidated with the rapid decrease in heart rate and blood pressure by change of autonomic nervous system control (Alboni & Alboni, 2017 (Shin et al., 2015). Third, chronic complex regional pain syndrome is a debilitating pain condition accompanied by autonomic abnormalities. A previous report using functional MRI demonstrated that the chronic complex regional pain syndrome was associated with a diffuse frontal lobe hyperactivity, which was diminished or decreased with sympathetic nerve blocks (Apkarian, Thomas, Krauss, & Szeverenyi, 2001). Another study with chronic complex regional pain syndrome using voxel-based morphometry also showed gray matter atrophy in the frontal lobe (Geha et al., 2008). Fourth, a previous study using concurrent microelectrode recordings of sympathetic outflow to either muscle or skin and functional MRI concluded that the frontal lobe is involved in the generation of sympathetic nerve activity These previous reports, taken together with the present study, reveal that the frontal lobe is a key brain region for autonomic control, and suggest a link between dysregulated physiologic reactions arising from compromised frontal autonomic control and increased vulnerability to reflex syncope.
However, recently Blanc & Benditt (2016) proposed a new hypothesis to try to explain and evolution of reflex syncope, "the brain self-preservation theory". According to this theory, reflex syncope appears to be a protective mechanism for the brain. Under certain circumstances, the cerebral blood flow can decrease; the faint causes the body to take on a gravitationally neutral position and thereby provides a better chance of restoring brain blood supply and preserving long-term brain function. Thus, it could be possible that our results may be produced by cerebral hypo-perfusion insults.
Repeated hypo-perfusion insults could produce reorganization of network hubs in the brain network of patients with reflex syncope.
Several previous reports demonstrated various structural abnormalities of the brain in patients with reflex syncope. Beacher et al.,  (Kim et al., 2014). In addition, we previously demonstrated that the cortical thickness of orbitofrontal, pericalcarine, postcentral, inferior temporal, and lateral occipital cortex significantly changed in patients with orthostatic hypotension but not in patients with postural tachycardia syndrome (Shin et al., 2015). These discrepancies could be caused by the different subjects and imaging methods used, or produced by the different vulnerable brain structures to hypo-perfusion according to various subjects.
With DTI scans, FA and MD values can be obtained to assess the changes in white matter tracts. In general, FA and MD values are sensitive to microstructural changes in white matter and the microstructural architecture of cellular membranes, respectively (Alexander et al., 2017). Although the direct comparison of FA and MD values at each voxel across different groups can reveal local changes, the F I G U R E 1 Tract-based spatial statistics analysis of reflex syncope patients and healthy controls. Throughout the brain, no differences were detected in fractional anisotropy (FA) or mean diffusivity (MD) between patients with reflex syncope and healthy subjects results of this method can be significantly and adversely affected by unavoidable registration errors and noise (Faria et al., 2010).
Thus, we used the TBSS method, which can reduce these effects by projecting the volumetric data onto a white matter skeleton (Smith et al., 2006). However, in contrast to alterations of brain networks in patients with reflex syncope, we found the FA and MD values of the white matter in the patients with reflex syncope were not different from those in healthy subjects. A plausible explanation for this finding is that the changes of brain network are more sensitive than the structural changes of white matter.
There are several limitations to this study. First, we only investigated 20 patients with reflex syncope. Further studies with larger sample sizes may be needed. Second, we were unable to apply multiple test corrections in the analysis. Because we divided the whole brain into 90 subregions based on the AAL atlas to construct the brain structural network, the appropriate corrected p-value for significance was = 0.00056 (0.05/90, Bonferroni correction) in the analysis. However, the statistical significance with a p-value <0.00056 was too great to apply in the analysis. It could be possible that our results represent false positive due to small sample size without multiple corrections. Thus, we calculated the power of this study using the Power and Sample Size Program (http:// ps-power-and-samplesize calculation. software. informer.com/download/). It revealed that the statistical power of this study was sufficient to exclude type 1 error (all of them had more than 80%). Third, the major limitation of this study was that we could not identify the causal relationship between changes in the organization of the hubs of brain network and reflex syncope. Alterations of brain network may represent a preexisting vulnerability to reflex syncope, whereas we cannot rule out the possibility that these changes are the result of decreased cerebral perfusion or hypoxic episodes.

| CON CLUS IONS
We have described here, for the first time, reorganization of the network hubs in the brain network of patients with reflex syncope.
Such alterations may play a role in the pathophysiologic mechanism underlying reflex syncope.

ACK N OWLED G M ENTS
This work was supported by the Ministry of Science and ICT of the Republic of Korea (NRF-2017R1C1B5015871).

CO N FLI C T O F I NTE R E S T
There are no conflict of interests to declare. Alboni, P., & Alboni, M. (2017). Typical vasovagal syncope as a "defense mechanism" for the heart by contrasting sympathetic overactivity.