Altered Topological Organization of Cortical Network in Adolescent Girls with Idiopathic Scoliosis

Adolescent idiopathic scoliosis (AIS) is a multifactorial disease affecting approximately 1–4% of teenagers especially girls at the age of 10–16, but its etiopathogenesis remains uncertain. Previous study has revealed that the cortical thickness in AIS patients is different from that in normal controls. Cortical thickness measurements are known to be strongly correlated between regions that are axonally connected. Hence, a hypothesis is proposed to study the possibility to demonstrate abnormal structural network revealed by cortical thickness in AIS patients. The aim of the study is to investigate abnormalities in the organization of the brain cortical network in AIS patients. This study included 42 girls with severe idiopathic scoliosis (14.7±1.3 years old) and 41 age-matched normal controls (NC, 14.6±1.4 years old). The brain cortex was partitioned into 154 cortical regions based on gyral and sulcal structure. The interregional connectivity was measured as the statistical correlations between the regional mean thicknesses across the subjects. We employed the graph theoretic analysis to examine the alteration in interregional correlation, small-world efficiency, hub distribution, and regional nodal characteristics in AIS patients. We demonstrated that the cortical network of AIS patients fully preserved the small-world architecture and organization, and further verified the hemispheric asymmetry of AIS brain. Our results indicated increased central role of temporal and occipital cortex and decreased central role of limbic cortex in AIS patients compared with controls. Furthermore, decreased structural connectivity between hemispheres and increased connectivity in several cortical regions were observed. The findings of the study reveal the pattern of structural network alteration in AIS brain, and would help in understanding the mechanism and etiopathogenesis of AIS.


Introduction
Idiopathic scoliosis is a 3-dimensional spinal deformity that occurs most commonly in adolescent girls and affects about 1-4% of school teenagers worldwide. A more distinct indication of AIS is the prominence of back when they fully bend forward. Current studies indicate that the etiopathogenesis of AIS remains unknown and scientists believe that it is a multifactorial disease which involves intrinsic factors such as the growth of vertebral body, abnormal morphometry of the posterior elements, asymmetrical growth of the neurocentral cartilage etc., and extrinsic factors such as genetic components and chromosomes, extra-spinal left-right asymmetry, body growth and development, imbalance of muscle, dysfunction of the nervous system, somatosensory function, vestibular system, length of spinal cord, abnormal platelets and abnormality in melatonin metabolism [1]. Central nervous system is one of the major factors related to the etiopathogenesis of AIS.
Recently, various neuroanatomical studies have been proposed to explore the etiopathogenesis of AIS patients. Structural brain MRI analyses, such as region volume [2], cortical thickness [3], comprehensive morphometry of corpus callosum [4], vestibular morphoanatomy [5], volume-based morphometry [6], have been used to investigate the morphological change in the brains of AIS patients. The preliminary analysis of regional brain volume has indicated the anatomical asymmetry in brain regions functionally related to somatic motorcontrol and coordination [2]. The morphoanatomical changes of vestibular system and vestibular network were found to be associated with subclinical postural, vestibular, and proprioceptive dysfunctions in AIS [3,5]. We have recently reported the thinning pattern of cerebral cortex in AIS patients. The findings of cortical thinning pattern also lead us to make a hypothesis: is it possible that altered organization pattern of large-scale structural network could be revealed in AIS patients compared with age-matched healthy individuals?
The small-world architecture (popularly known as six degrees of separation) first quantified by Watts and Strogatz [7] has been widespread utilized to describe the topological properties of networks in social, biological, and man-made systems. Recently the terms of ''network'' and ''small-worldness'' began to raise more interest in the aspect of brain organization which was demonstrated to share similar properties with real-world networks of different nature (e.g., social networks, the Internet, et al) [8,9]. Specifically, when the human brain is considered to be a complex network with nodes representing brain regions and edges representing inter-regional interactions, brain network properties can be derived from generated graphs [9]. On the other hand, it has been confirmed that the structural features of human cortex convey brain connectivity information, e.g., the cortical thickness are known to be strongly correlated between regions that are axonally connected like Broca's and Wernicke's Areas [10]. Thus, a whole-brain cortical network can be obtained from structural MRI data based on a matrix of correlations in cortical thickness between all pairs of parcellated regions [11,12]. Indeed, previous study has demonstrated that the variations in human regional cortical thickness follow the small-world topology at the macroscale level [12], and the alteration in topological organization of brain cortical network has been demonstrated useful to investigate the brain development [13], age-and gender-related change [14,15], and the characteristic mechanisms of a series of neurodegenerative diseases [16,17]. Our goal in this study is to investigate the abnormalities of the organization of the brain cortical network in AIS patients by applying cortical thickness measurements combined with the graph theoretical analysis.

Subjects
This study included 42 girls with severe idiopathic scoliosis (14.761.3 years old, range 13-18) and 41 age-matched normal controls (NC, 14.661.4 years old, range 12-18), which were recruited from the Prince of Wales Hospital in Hong Kong. Inclusion criteria for AIS patients included: (1) 12# age #18 years; (2) right thoracic curves; (3) Cobb angle $40u; (4) availability of collection the patient's medical history from the caregiver. The AIS patients had an average Cobb angle of 54.5u (range 40-90u). Exclusion criteria for both AIS patients and NC included any history of head injury, headache, back injury, weakness or numbness in any limbs, urinary incontinence, nocturnal enuresis or any space occupying lesion found on screening MR imaging. There were no significant differences in age and education between the two groups (all P.0.05). All the subjects were right-handed. All the subjects were required to be normal in detailed neurological examinations. Ethical approval was obtained from the university and hospital ethics committee. Informed written consent was obtained from each subject and her parents.

MRI acquisition
All subjects were scanned using a clinical 1.5T MRI scanner (Sonata, Siemens, Erlanger, Germany) with a quadrature head coil. For each subject, high resolution T1-weighted images covering the whole brain were acquired in the axial orientation using magnetization prepared rapid acquisition gradient echo (MPRAGE) with the following parameters: repetition time = 2070 ms, echo time

Cortical thickness mapping
The whole schematic flowchart of cortical structural network construction is shown in Figure 1. For each subject, freesurfer toolkit (http://surfer.nmr.mgh.harvard.edu) was used to analyze the anatomical MPRAGE images. The freesurfer procedure involving white matter segmentation [18], subcortical mass labeling and filling [19], tessellation and topology fixing, topological defects correction [20], white and pial cortical surface generation [21], and cortical thickness measurement [21]. The cortical thickness at each vertex was measured as the shortest distance to the white surface from each point on the pial surface averaged with the shortest distance to the pial surface from each point on the white surface.

Regional thickness calculation
Using spherical registration [22], the individual inflated surface was normalized to a spherical template, and the brain cortexes were partitioned into 154 cortical regions base on a spherical atlas [23]. The cortical thickness for each cortical region was measured as the average thickness of all vertices defined as belonging to that region.

Interregional correlation
The structural connectivity between two cortical regions was defined as the statistical correlation between the regional thicknesses across subjects. Before the correlation computing, the linear regression was performed to remove the effects of age and mean cortical thickness on the full set of regional thickness in each region. The residuals of the regression were then substituted for the raw regional thickness values. The interregional correlation matrix (1546154) of both NC and AIS group was generated by calculating the Pearson correlation coefficients across NC and AIS subjects between the residual thicknesses of each pair of cortical regions.

Threshold selection
The cortical network is represented by a binarized matrix. Since currently no accepted strategy to define an optimal threshold, the correlation matrix was usually thresholded repeatedly over a series of thresholds. If the same correlation threshold was applied to the correlation matrices of both AIS and NC groups, the number of nodes and edges in two cortical networks would differ from each other and the group difference would not be purely reflected by the alterations of network properties. To control this effect, a range of sparsites were used to substitute the correlation thresholds to binarize the matrices. Sparsity is defined as the total number of edges divided by the maximum possible number of edges in a network, which would result two networks of both groups with same number of edges or wiring cost [24].

Small-world efficiency
The network efficiency metrics, first proposed by Watts and Strogatz [7], were used to reflect the efficiency in communication and organization over a network system. Specifically, the smallworld efficiency introduced here was a quantitative analysis on how efficiently the network of human brain exchanges information at the global and local levels [25]. The global efficiency of a network graph G was defined as the inverse of the harmonic mean of the shortest path length between every two nodes, where d ij is the length of the shortest path connecting node i and j, and n is the total number of nodes. The local efficiency of a network graph G was defined as the average of the local efficiencies of each node in a network, where G i is the subgraph composed of the nearest neighbors of node i. The cortical network can be considered to be a small-world network if it meets the criteria , where the E glob G random ð Þ and E loc G random ð Þ are the averages of global efficiencies and efficiencies of 100 matched random networks that preserve the same number of nodes, edges, and degree distribution as the real network.

Regional nodal characteristics
The betweenness centrality was essentially a measurement of the influence of a node over the information flow between itself and other nodes. The betweenness centrality B i of each node i was defined as the fraction of all shortest paths in the network that pass through it [26] where r hj i ð Þ is the number of shortest path between nodes h and j that pass through node i, n is the total number of nodes. The mean betweenness centrality B G was calculated as the average of the values of the nodal betweenness over all nodes. The node betweenness was normalized by the mean betweenness centrality b i~Bi =B G .
The hubs of the cortical networks were defined as the regions with high values of normalized betweenness (at least one standard deviation greater than the average normalized betweenness). As was headed up in the introduction, the nodes represented different brain regions. Then the hubs are considered to be the brain regions playing significant roles in information transferring and integrating in whole brain communication [11].
The regional efficiency E i of each node i was defined as the inverse of harmonic mean of the shortest path length between itself and all other nodes [24] The regional efficiency measures the average shortest path length between a given node and all the other nodes in the network, which is used to quantify the importance of each node in the communication within the network.
The vulnerability V i of each node i was defined as the drop in global efficiency when this node and its connections were removed from the network graph, reflecting [27] where E i glob is the global efficiency of the network after removing node i. The global vulnerability V G of the entire network was defined as the maximum vulnerability for all of its nodes. The vulnerability analysis is a way to identify critical components and can quantitatively measure the damage on the network performance caused by the hypothetical failure of its elements.

Statistical Analysis
Interregional correlation. Differences in correlation of regional thickness between the AIS and NC groups were examined using z-test. Prior to the statistic, the Pearson correlation coefficients were converted into normally distributed z values using Fisher's r-to-z transform. The between-group differences in correlations were determined by comparing these z values using z-test [28]. The false discovery rate (FDR) [29] with q = 0.05 was used to correct the multiple comparisons.
Network characteristics. Differences in overall graph characteristics (E glob , E loc ) and regional nodal characteristics (B i , E i ) between the AIS and NC groups were examined using nonparametric permutation test [30]. The permutation test proceeded as follows. First, the network characteristics were calculated separately at a given sparsity for each group and their differences were obtained. Then the subjects in both groups were pooled and randomly assigned to either one of two groups consisting of the same size as the original AIS and NC groups. The correlation matrices for these two new groups were recalculated and binarized using the same threshold as in the real network. The network characteristics were calculated for each random group and their differences were obtained. This procedure was repeated for 1,000 times. The 95 percentile points of each distribution were used as the critical values for a one-tailed testing the overall graph characteristics (E glob , E loc , B G , V G ) of type I error of 0.05. The 99 percentile points of each distribution were used as the critical values for a one-tailed test of the regional nodal characteristics (B i , E i ) of type I error of 0.01.

Change in interregional correlation
The altered interregional correlations in AIS patients were shown in Figure 2

Small-world efficiency
To investigate the network characteristics, the cortical networks were constructed using a series of sparsities thresholds. A specific sparsity 0.36 was determined by mean betweenness centrality B G and global vulnerability V G to ensure that all cortical regions were included in the cortical networks while keeping the network more stable (Figure 3). A range of sparsities thresholds (0.36# sparsity #0.5, step = 0.01) were used to verify the network efficiency of cortical networks. Both AIS and NC groups exhibit efficient small-world topology (E glob G ð Þ=E glob G random ð Þ &1, E loc G ð Þ=E loc G random ð Þ w1) in cortical networks, and no significant difference in global efficiency and local efficiency (0.36# sparsity #0.5, all P.0.05, uncorrected) was found.

Distribution of hub regions
In this study, 22 and 26 hubs were identified in cortical network of NC and AIS groups respectively, which were thresholded by a specific sparsity threshold 0.36 ( Figure 4). Hubs in the NC group were distributed in 12 left hemispherical regions and 10 right hemispherical regions (9 frontal regions, 1 temporal region, 5 parietal regions, 1 occipital region, 4 limbic regions, 1 parietofrontal region, and 1 insular region). Hubs in the AIS group were distributed in 17 left hemispherical regions and 9 right hemispherical regions (10 frontal regions, 5 temporal regions, 4 parietal regions, 3 occipital regions, 1 limbic region, 1 parietofrontal region, 1 occipito-temporal region, and 1 insular region). Among them, 10 hubs were the same for both groups including the bilateral middle frontal gyrus, the opercular part of inferior frontal gyrus, the superior frontal gyrus, the middle occipital gyrus, the supramarginal gyrus, the inferior temporal gyrus, and the inferior frontal sulcus in left hemisphere, the angular gyrus, and the cingulate sulcus in right hemisphere. Twelve cortical regions including the isthmus of cingulate gyrus, the long gyrus of insula, the subcentral sulci, the cingulate sulcus, and the frontomarginal sulcus in left hemisphere, and the cingulate gyrus, the opercular and triangular part of inferior frontal gyrus, the supramarginal gyrus, the superior frontal gyrus, the sulcus intermedius primus, and the parieto-occipital sulcus in right hemisphere, were identified as hubs in the NC group but not in the AIS group. Sixteen cortical regions, the bilateral middle frontal sulcus, the bilateral postcentral sulcus, the bilateral superior temporal sulcus, the orbital and triangular part of inferior frontal gyrus, the middle temporal gyrus, the inferior segment of the circular sulcus of insula, the orbital sulcus, and the paracentral sulcus in left hemisphere, the inferior temporal gyrus, the calcarine sulcus, and the occipito-temporal sulcus in right hemisphere, were identified as hubs in the AIS group but not in the NC group.

Change in nodal characteristics
A range of sparsities thresholds (0.36# sparsity #0.5, step = 0.01) were also used to investigate the nodal characteristics of cortical networks. The between-group differences of nodal characteristics (B i , E i ) were shown in Figure 5 and 6, where the mean p-values of the series of sparsities were overlaid on the mean cortical surface. Significantly decreased betweenness centrality B i in AIS group was found in the orbital part of left inferior frontal gyrus (P,0.01, when 0.38# sparsity #0.48, uncorrected, P = 0.0081, P indicates the mean p-values of the series of sparsities from 0.36 to 0.5) and the left paracentral sulcus (P,0.01, when 0.36# sparsity #0.5, uncorrected, P = 0.0048) ( Figure 5). Significantly increased regional efficiency E i in AIS group was found in the left sulcus intermedius primus (P,0.01, when 0.36# sparsity #0.5, uncorrected, P = 0.0054), the right superior occipital gyrus (P,0.01, when 0.36# sparsity #0.5, uncorrected, P = 0.0023), the right superior parietal lobule (P,0.01, when 0.37# sparsity #0.49, uncorrected, P = 0.0059), and the orbital part of right inferior frontal gyrus (P,0.01, when 0.36# sparsity #0.37 and 0.43# sparsity #0.5, uncorrected, P = 0.0085) ( Figure 6).

Altered interregional correlation
Interregional correlation which is a statistical association reflects the structural connectivity or connection between pairs of cortical regions. Only decreased between-hemisphere correlations were observed, which demonstrated that the AIS only negatively affects the between-hemisphere connectivity. Three orbitofrontal regions with increased correlations were observed in AIS patients. Particularly, since the insular cortex is in a position to convey diverse sensory information to the orbitofrontal cortex [31], the increased connectivity of right orbitofrontal-insular circuit in AIS patients may relate with the right thoracic convexity. The reduced correlation in left anterior ascending and horizontal limb of lateral sulcus may be caused by decreased cortical thickness in left lateral sulcus (sylvian fissure) [3]. Most of brain regions with altered correlation were included in the vestibular cortical network previously revealed by caloric stimulation and functional MRI [32], and these findings were also supported by the previous finding of vestibular dysfunction [33] and abnormal morphoanatomy [5] in AIS patients.    Small-world efficiency Both AIS and NC subjects were found to demonstrate smallworld architectures and exhibit both high global and local efficiency, which were compatible with previous structural brain networks studies [11,[14][15][16]. The comparison of global and local efficiency between patients and controls in the current study revealed unchanged network organization in AIS over a wide range of sparsities. Our finding suggests that the small-world topological characteristics were fully preserved in AIS patients.

Alteration of Hub Distribution
The identified hubs in this study were consistent with those obtained from previous similar studies [15][16][17]. There was no significant difference of the hub distribution in frontal (NC/AIS: 9/10), parietal (NC/AIS: 5/4), and parieto-frontal (NC/AIS: 1/1) cortex between NC and AIS groups. In this study, most identified hubs of both groups were frontal and parietal regions, which are in line with the previous studies on brain graphs [34]. Nevertheless, the number of hubs in temporal cortex of AIS group was increased from 1 to 5 compared with NC. This may be related with the dysfunction and abnormal morphoanatomy of vestibular system in the temporal lobe observed in AIS patients [5,33] and supported by the fact that AIS patients are with thinner cortex in bilateral lateral sulcus (sylvian fissure) [3]. Possibly effected by the decreased cortical thickness in lateral occipital regions [3], the number of hubs in the occipital cortex was also increased (from 1 to 3). Interestingly, the number of hubs in the limbic cortex of AIS group was reduced from 4 to 1 compared with NC, which may be a possible compensatory mechanism for increased hub number of temporal and occipital cortex. In short, increased central role of temporal and occipital cortex and decreased central role of limbic cortex were observed in AIS patients. Furthermore, the altered distribution pattern of hubs in left (NC/AIS: 12/17) and right hemisphere (NC/AIS: 10/9) indicated the hemispheric asymmetry in AIS patients, and this is in line with the previous studies of AIS [3,35,36]. In this study, all AIS patients were with right thoracic curves, supporting the findings of hemispheric asymmetry of hub distribution.

Altered Nodal Characteristics
Reduced betweenness was observed in the left paracentral sulcus, which belongs to the primary motor area being responsible for generating neural impulses that pass down to the spinal cord and control the execution of movement. The change suggests that the left paracentral sulcus of AIS patient has decreased influence in the brain communication compared with that of NC. Enhanced regional efficiency was observed in the right superior parietal gyrus, which is a part of the posterior parietal cortex. The posterior parietal cortex is also considered to be part of motor cortical area. The change suggests the increased connectivity of the right superior parietal to all other nodes in AIS patient compared with that in NC. Although no significant abnormal activation was observed among these currently identified regions in previous functional study [37], our findings may also suggests an altered pattern of structural connectivity of motor-related cortex in AIS. Furthermore, the results, 1) unchanged global and local efficiency, 2) four regions with enhanced regional efficiency, and 3) no region with reduced regional efficiency, demonstrates that the enhanced ability of commutation only occurred in several cortical regions.

Conclusions
We applied cortical thickness measurement, interregional correlation, and graph theory to analyze the topological organization cortical network in AIS patients. This study demonstrates for the first time that the AIS brain was associated with changed between-hemisphere connectivity, and changed ability of communication of brain cortical networks. Our results indicate that the cortical network of AIS patients fully preserves the small-world architecture and organization, and further verify the hemispheric asymmetry of AIS brain. These findings were compatible with and complemented our previous neuroanatomical findings, thus enhancing the understanding of underlying etiology of AIS.