FULL-LENGTH ORIGINAL RESEARCH

Disrupted anatomic white matter network in left mesial temporal lobe epilepsy *Min Liu, *Zhang Chen, *Christian Beaulieu, and †Donald W. Gross Epilepsia, 55(5):674–682, 2014 doi: 10.1111/epi.12581

SUMMARY

Min Liu completed her PhD at the University of Alberta studying white matter pathology in epilepsy.

Objective: Brain imaging studies have shown widespread structural abnormalities in patients with temporal lobe epilepsy (TLE) within and beyond the affected temporal lobe, suggesting an altered network. Graph theoretical analysis based on white matter tractography has provided a new perspective to evaluate the connectivity of the brain. The alterations in the topologic properties of a whole brain white matter network in patients with TLE remain unknown. The purpose of this study was to examine the white matter network in a cohort of patients with left TLE and mesial temporal sclerosis (mTLE) compared to healthy controls. Methods: Anatomic brain networks of 16 patients with left mTLE were compared to those of 21 healthy controls. A white matter structural network was constructed from diffusion tensor tractography for each participant, and network parameters were compared between the patient and control groups. Results: Patients with left mTLE exhibited concurrent decreases of global and local efficiencies and widespread reduction of regional efficiency in ipsilateral temporal, bilateral frontal, and bilateral parietal areas. Communication hubs, such as the left precuneus, were also altered in patients with mTLE compared to controls. Significance: Our results demonstrate white matter network disruption in patients with left mTLE, supporting the notion that mTLE is a systemic brain disorder. KEY WORDS: Diffusion tensor imaging, Graph theory, Temporal lobe epilepsy, Network, Tractography, Diffusion tensor imaging.

Temporal lobe epilepsy (TLE) is the most common focal epilepsy syndrome. Most TLE cases are associated with mesial temporal sclerosis (MTS), where hippocampal atrophy and/or T2 hyperintensity is observed on magnetic resonance imaging (MRI). Recently, quantitative MRI studies have provided new structural information in TLE by demonstrating extensive abnormalities within and beyond the temporal lobe, including reduced gray matter concentration/atrophy,1 decreased cortical thickness,2 and abnormal white matter in the temporal, frontal, and parietal lobes.3 Accepted January 27, 2014; Early View publication March 20, 2014. *Department of Biomedical Engineering, Faculty of Medicine and Dentistry, University of Alberta, Edmonton, Alberta, Canada; and †Division of Neurology, Department of Medicine, Faculty of Medicine and Dentistry, University of Alberta, Edmonton, Alberta, Canada Address correspondence to Donald W. Gross, Division of Neurology, Department of Medicine, 2E3.19 Walter C Mackenzie Health Sciences Centre, Edmonton, AB T6G 2B7, Canada. E-mail: [email protected] Wiley Periodicals, Inc. © 2014 International League Against Epilepsy

The language and memory impairments seen in patients with TLE have also been related to altered blood oxygen level–dependent (BOLD) activation during memory and language tasks compared to healthy controls in functional MRI (fMRI) studies.4 The growing knowledge of the brain structural and functional changes related to TLE has led to the notion that TLE is a network disorder that affects large neural networks involved in normal brain function.5 With the implementation of graph theoretical analysis on brain functional and structural measurements, features of the brain as an integrated system can be quantified to infer its capacity of information integration and segregation.6 Graph theoretical analysis enables the abstraction and examination of the brain topologic architecture and properties at the network level, which provides a macroscopic perspective rather than the previously performed investigations at the level of local connections. Past network analysis in epilepsy has measured the brain functional network constructed from the coherence

674

675 Disrupted WM Network in Left mTLE of electroencephalography (EEG) signals between different regions and has demonstrated that the network in TLE has a small-world configuration that transitions from a more random configuration toward a regular configuration at seizure onset.7–10 A recent study based on single photon emission computed tomography (SPECT) demonstrated greater connectivity and clustering within the medial temporal and subcortical regions, with an overall reduction of connectivity in the remaining cortex and the whole network in patients with mesial TLE during the ictal compared to the interictal state.11 Other studies based on interictal functional fMRI showed decreased global efficiency in patients with TLE4,12 and cryptogenic localization-related epilepsy13 in comparison to healthy controls. Although functional network analysis offers insights into the dynamics of the brain, it is also necessary to examine the structural network that provides an anatomic substrate for the functional change. One recent structural brain network study based on regional cortical thickness correlation reported greater path length and clustering in patients with TLE.14 Three other network studies based on white matter fiber connections derived from diffusion tensor imaging (DTI) tractography specifically examined the default network,15 limbic network,16 and individual connections in the whole brain connectivity matrix17 in patients with TLE. They reported significantly decreased connection density among default network cortical regions,15 increased cluster coefficient in the limbic network,16 and reduced connectivity between the ipsilateral thalamus and the ipsilateral precentral gyrus and increased connectivity between the ipsilateral inferior parietal lobule and the ipsilateral supramarginal gyrus.17 These results provide evidence of disrupted local structural circuits in the brains of patients with

TLE. A whole brain DTI network study demonstrated decreased clustering and increased path length in seven patients with cryptogenic localization-related epilepsy with severe cognitive impairment compared to patients with little or no cognitive impairment and controls, suggesting an aberrant brain network in patients with partial epilepsy and severe cognitive impairment.18 To date, the alterations in the topologic properties of a whole brain white matter network in a group of patients with TLE remain largely unknown. The purpose of this study was to examine the white matter network using graph theoretical analysis in a cohort of patients with left TLE and MTS (mTLE) compared to healthy controls.

Methods Approval of the research protocol was obtained from University of Alberta Health Research Ethics Board, and informed consent was obtained from all participants. Participants Sixteen patients with left mTLE were recruited from the epilepsy clinic of the University of Alberta Hospital (Table 1), as well as 21 healthy volunteers from the community (controls: mean age  standard deviation (SD) 37  12 years, range 19–58 years; left hippocampal T2: 112  4 msec, range 102–120 msec). All patients with left mTLE demonstrated left temporal lobe ictal seizure onset on video-EEG telemetry and left MTS defined by left hippocampal T2 > 2 SD above the mean of controls (i.e., > 120 msec; see the method used in our previous study19). Note that 15 of the 16 patients with left mTLE have been reported previously with individual tract

Table 1. Demographic and clinical data of patients with left mTLE No.

Age (years)

Gender

Left hippocampal T2 (msec)a

Disease duration (years)

Age of onset (years)

PSb

AVLT-VIIc

CVMTc

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Mean

19 20 23 26 33 33 33 36 36 37 44 46 53 55 56 58 38  13

M F M M F M M M F F M M F F F F 8M/8F

132 146 141 148 127 128 128 138 149 126 131 126 139 126 132 160 136  10

17 19 11 15 34 6 4 31 35 13 42 27 39 28 39 51 26  14

2 1 12 12 1 29 29 6 1 24 2 20 14 27 18 8 13  11

82 55 81 64 73 66 84 61 80 75 70 77 86 84 80 77 75  9

24 9 37 37 25 14 21 25 32 18 49 35 28 38 35 32 29  10

5 10 2 1 37 4 5 4 35 17 5 40 12 2 44 16 14  16

Mean and SD of right hippocampal T2 is 117  5 msec. PS – processing speed (Wechsler Adult Intelligence Scale–Third Edition [WAIS-III]). The mean typical age-scaled standard score is 100. c AVLT-VII – Auditory Verbal Learning Test (trial VII), CVMT – Continuous Visual Memory Test. The mean typical age-scaled standard score is 50. a

b

Epilepsia, 55(5):674–682, 2014 doi: 10.1111/epi.12581

676 M. Liu et al. differences on DTI.3,19 Three cognitive tests were administered to the patients with left mTLE, including processing speed (PS, Wechsler Adult Intelligence Scale–Third Edition [WAIS-III]), Auditory Verbal Learning Test (trial VII) (AVLT-VII), and Continuous Visual Memory Test (CVMT). Scores standardized to age were used to test their correlations to network parameters. Cognitive assessments were not performed on controls. Image acquisition All patients with left mTLE were scanned in the interictal state (at least 24 h after the last seizure). Whole-brain magnetization prepared rapid acquisition gradient echo (MPRAGE) and DTI were acquired on a 1.5T Siemens Sonata scanner (Erlangen, Germany). MPRAGE provided high-resolution three-dimensional (3D) T1-weighted images with 144 axial slices, 1-mm slice thickness with no interslice gap, TR (repetition time) = 1,890 msec, TE (echo time) = 4.38 msec, TI (inversion time) = 1,100 msec, flip angle = 15 degrees, NEX (number of excitation) = 1, acquisition matrix = 256 9 192 (interpolated to 512 9 384), field of view (FOV) = 256 mm 9 192 mm, voxel dimension 1 9 1 9 1 mm3 (interpolated to 0.5 9 0.5 9 1 mm3), and scan time = 6 min 3 s. DTI used a dual spinecho, single shot echo planar imaging sequence with 52 axial slices, 2 mm slice thickness with no interslice gap, TR = 10 s, TE = 88 msec, six diffusion directions with b = 1,000 s/mm2, NEX = 8, acquisition matrix = 128 9 128 (interpolated to 256 9 256), FOV = 256 mm 9 256 mm, voxel dimension 2 9 2 9 2 mm3 (interpolated to 1 9 1 9 2 mm3), and scan time = 9 min 30 s. Construction of weighted cortical networks Preprocessing Motion and eddy current corrections for all DTI images were performed using FSL (v5.0, Functional MRI of the Brain's Software Library, Oxford, United Kingdom). Using SPM8 (Statistical Parametric Mapping, Wellcome Department of Cognitive Neurology, London, United Kingdom), the structural images (MPRAGE) were coregistered linearly to the b0 nondiffusion weighted image of each subject so that they were aligned in the DTI native space (Fig. 1). Definition of network nodes The automatic anatomic labeling (AAL) template was used to parcellate the cerebral cortex into 78 predefined cortical regions (39 for each hemisphere, Supplementary Table S1),20 each representing a node of the cortical network. This parcellation scheme has been used for several previous brain network studies.21,22 Specifically, the T1-weighted structural image of each subject was nonlinearly transformed to ICBM (International Consortium for Brain Mapping) 152 T1 template in the Montreal Neurological Institute (MNI) space (SPM8). The resulting inverse Epilepsia, 55(5):674–682, 2014 doi: 10.1111/epi.12581

deformation map for each subject was then applied to warp the AAL template to the DTI native space using the nearest neighbor interpolation method to preserve the discrete labeling values (SPM8). Whole brain white matter tractography The diffusion tensors were estimated and three eigenvalues and eigenvectors were derived from the diagonalization of the tensor matrix for each subject. Whole brain white matter tractography was performed in DTI-TK TractTool (v2.3.1) using a brute-force streamline-tracking method23 seeding from every voxel in the brain with a fractional anisotropy (FA) > 0.2. The tract propagation terminated when it reached a voxel with FA < 0.2 or with primary eigenvector turning more than 45 degrees. Definition of network edges The deformed AAL template for each subject marked 78 cortical regions in the DTI native space. The number of white matter fibers whose two terminal voxels fell within the area of any pair of the 78 cortical regions, conjointly, was counted. Two cortical regions were deemed connected if at least three connecting fibers were found between them. The counting threshold was applied to reduce, but cannot eliminate, false-positive connections among regions.22,24 Construction of weighted network To quantify the strength of each connection/edge between two cortical regions (node i and j) we defined the edge weight by the product of the number of connecting fibers (fiber number, FN) and mean FA of the connecting fibers, divided by the average volume of the two cortical regions to counteract the bias where larger cortical regions inherently project/receive more fibers (i.e. wij ¼ average volume ofFNFA the two cortical regions ). This weighting factor has been adopted by two previous brain network studies.22,24 The network was calculated using an in-house MATLAB (R2012b, The MathWorks, Natick MA, U.S.A.) script. Graph theoretical metrics Several topologic properties were measured for the weighted anatomic brain network derived from each participant using Brain Connectivity Toolbox (http://sites. google.com/site/bctnet/),25 including the total number of edges K (the only measure independent of connection weights), weighted strength S, weighted clustering coefficient C, weighted characteristic shortest path length L, normalized weighted clustering coefficient c, normalized weighted characteristic shortest path length k, global efficiency Eglob, local efficiency Eloc, regional global efficiency Ereg and hubs (nodes with Ereg > group mean + group SD in each participant group) for each AAL region. A summary of graph theoretical metrics and their interpretations are listed in Supplementary Table S2. For each white matter

677 Disrupted WM Network in Left mTLE

Figure 1. A flowchart for the construction of DTI white matter structural network. (1) The T1-weighted structural image for each subject (B) was coregistered into DTI native space (A) using rigid transformation. (2) The resultant structural image was nonlinear registered to the ICBM 152 T1 template (C) in the MNI space. (3) The AAL template, including 78 cortical regions in the MNI space (E), was inversely warped back to the individual DTI space (F) using the inverse transformation (T1). (4) Whole brain white matter fibers were reconstructed using a deterministic tractography method (D). (5) The white matter fibers connecting any pair of the 78 regions were found and the FA, fiber number (FN), and average volume of the two cortical regions were counted to construct the white matter matrix for each subject. The mean matrices averaging from all participants from each group are shown (G). Epilepsia ILAE

A

B

C

D

E

F

G

network, the connection weights w were normalized by the mean of all weights in the network to keep each participant’s cost at the same level. Statistical analysis General linear model was used to compare the network topologic properties (S, C, L, Eglob, Eloc, Ereg) between patients and controls, with age and gender included as nuisance variables. The results of each Ereg of the 78 regions are corrected by false discovery rate (FDR) at q = 0.05.26,27 For left mTLE group, significantly different topologic properties were correlated with the left hippocampal T2, age of seizure onset, disease duration, PS, AVLT-VII, and CVMT controlling for age and gender. Multiple comparison correction was also applied on correlation results using false discovery rate at q = 0.05. Reliability test The confidence intervals (CIs) of graph theoretical metrics measured in this study were estimated by the sampling

with replacement bootstrap approach with 1,000 randomizations for each group. During each bootstrap step, a random resample of the graph theoretical metric was drawn from the original sample with replacement, and the mean of the random resample was calculated. This procedure was repeated 1,000 times and 1,000 resample means were recorded. Finally, the mean and 95% CI of the 1,000 random resample means were computed.

Results Global property of the WM network Both controls and patients with left mTLE demonstrated “small-world” properties (ccontrols = 2.52  0.15, cLmTLE = 2.81  0.29, kcontrols = 1.18  0.04, kLmTLE = 1.19  0.06, rcontrols = 2.14  0.11, rLmTLE = 2.35  0.18). The patient-versus-control comparisons showed significantly smaller K, increased L, and decreased Eglob and Eloc in patients with left mTLE relative to controls (Table 2). Epilepsia, 55(5):674–682, 2014 doi: 10.1111/epi.12581

678 M. Liu et al. Table 2. Global network properties and between-group comparison results between patients with left mTLE and controls K S C L Eglob Eloc

left mTLE

Controls

T

p-Value

1,402  124 0.18  0.06 0.38  0.02 1.06  0.10 0.95  0.09 1.32  0.08

1,547  71 0.19  0.03 0.38  0.01 0.97  0.05 1.03  0.06 1.41  0.05

4.27 0.83 1.14 3.70 3.51 3.69

Disrupted anatomic white matter network in left mesial temporal lobe epilepsy.

Brain imaging studies have shown widespread structural abnormalities in patients with temporal lobe epilepsy (TLE) within and beyond the affected temp...
1MB Sizes 6 Downloads 2 Views