Network Analysis of the Default Mode Network Using Functional Connectivity MRI in Temporal Lobe Epilepsy

Summary

The Default Mode Network (DMN) in Temporal Lobe Epilepsy (TLE) is analyzed in the resting state of the brain using seed-based functional connectivity MRI (fcMRI).

Abstract

Functional connectivity MRI (fcMRI) is an fMRI method that examines the connectivity of different brain areas based on the correlation of BOLD signal fluctuations over time. Temporal Lobe Epilepsy (TLE) is the most common type of adult epilepsy and involves multiple brain networks. The default mode network (DMN) is involved in conscious, resting state cognition and is thought to be affected in TLE where seizures cause impairment of consciousness. The DMN in epilepsy was examined using seed based fcMRI. The anterior and posterior hubs of the DMN were used as seeds in this analysis. The results show a disconnection between the anterior and posterior hubs of the DMN in TLE during the basal state. In addition, increased DMN connectivity to other brain regions in left TLE along with decreased connectivity in right TLE is revealed. The analysis demonstrates how seed-based fcMRI can be used to probe cerebral networks in brain disorders such as TLE.

Introduction

Functional Connectivity MRI (fcMRI) is a relatively recent analytic approach to fMRI data that quantifies the relationship between different brain regions based on the similarity of their blood oxygenation level dependent (BOLD) signal time series – this is called “functional” connectivity, and is distinguishable from anatomical connectivity that describes the existence of physical connections between regions (e.g., white matter fibers). In a special application of this approach, the time series are collected when the participant is not engaged in a task or is in the so-called “resting state”.

Although first described in 1995¹, there has been immense interest in fcMRI resulting in approximately 1,000 publications related to the technique in 2012. fcMRI has intrinsic benefits over task-based fMRI in (1) that there is no specific task to be performed, (2) subject cooperation is not necessary, (3) datasets can be used to query several different networks, (4) better signal to noise ratio is present likely due to differences in cerebral energetics involved, and (5) circumvention of task-related confounds². As a proof of its concept, fcMRI changes have been shown to correspond with changes in EEG³ and local field potentials⁴ in the brain.

Techniques of fcMRI analysis include ROI/ seed-based techniques, independent component analysis (ICA), graph theory analysis, Granger causality analysis, local methods (amplitude of low frequency fluctuations, regional homogeneity analysis), and others⁵. No single technique has yet demonstrated clear superiority over another, although the most popular methods are seed-based and ICA methods⁶. Seed-based fcMRI correlates temporal fluctuations in BOLD signal from a preselected part of the putative network under study termed the “seed” or “region of interest (ROI)” to all other parts of the brain. Areas of the brain showing BOLD signal correlating to the seed area are thought to demarcate parts of the involved network. In contrast, ICA uses a model-free data-driven analysis to extract spatio-temporally correlated brain areas (Independent Components, ICs) by analyzing the hemodynamic signal characteristics of the whole brain⁵.

In the current manuscript, a description of methods used in a previously published study of resting state seed-based connectivity analysis of the DMN in TLE is presented⁷. TLE is the most common form of adult epilepsy. In addition to seizures, TLE causes dysfunction of multiple brain networks including memory, behavior, thought, and sensory function⁸. The DMN is constituted by cerebral regions subserving conscious, resting-state cognition. The DMN has been reported to be involved in seizures associated with reduced consciousness^9,10. Additionally, the hippocampus is the key structure involved in TLE and has been thought to be component of the DMN. However, the connectivity of the PCC to the hippocampal formation is weaker than with other DMN components, such as medial prefrontal and inferior parietal cortices. This suggests that the hippocampus is either a subnetwork of the DMN or an interacting network^11,12. These commonalities between TLE and DMN raise the possibility that DMN functional connectivity is altered in TLE. This analysis compares the DMN of subjects with TLE to healthy controls to gain insight into the involvement of DMN in TLE. The connectivity of seeds placed in the chief hubs of the DMN - the anterior and posterior hub regions were analyzed¹². Seeds were placed in the posterior hub consisting of the retrosplenium/precuneus (Rsp/PCUN) as well as the anterior hub consisting of the ventromedial prefrontal cortex (vmPFC) in patients having TLE and in healthy controls to identify the posterior and anterior subnetworks of the DMN.

Protocol

1. Subjects

1. The study population of 36 subjects includes 3 groups: right TLE (n=11), left TLE (n=12), and healthy controls (n=13). Obtain written informed consent from all subjects. The study follows the guidelines of the University of California, Los Angeles (UCLA) Institutional Review Board.
2. The epilepsy subject groups should be patients who are candidates for anterior temporal lobe resection as determined by video-EEG monitoring, brain MRI, PET imaging, and neuropsychological testing. Patients should continue their usual medications during the fMRI scan and should not be scanned immediately following a seizure. Make sure that all subjects have normal brain MRIs and are free from neurologic illness (other than epilepsy in the patient groups) or are using neurologic medications.

2. Imaging

1. Use a 3 Tesla MRI system for imaging. Obtain axial slices for functional images using an echo planar imaging (EPI) sequence and for anatomic images using a spoiled gradient recalled (SPGR) sequence.
2. Perform functional imaging using the following parameters: TR = 2,000 msec, TE = 30 msec, FOV = 210 mm, matrix = 64 x 64, slice thickness 4 mm, 34 slices. Use the following parameters for high-resolution structural imaging: TR = 20 msec, TE = 3 msec, FOV = 256 mm, matrix = 256 x 256, slice thickness 1 mm, 160 slices.
3. Each imaging session should last 20 min. Ask participants to relax with eyes closed. No special auditory input is required.

3. Preprocessing of BOLD Data

1. Preprocess the fMRI data using FSL (fMRIB Software Library) software version 4.1.6 (Oxford, UK, www.fmrib.ox.ac.uk/fsl)^13,14. Preprocessing steps should include the following: Use FSL MCFLIRT to remove head movement artifact¹⁵. Use FSL BET to remove nonbrain tissue¹⁶ with BET option -F for BOLD files. This helps one run further analysis steps on the brain tissue alone.
2. In FEAT, run a minimally processed analysis with registration. Select “First-level analysis” and change “Full analysis” to “Pre-stats” from the top two buttons.
   1. Under Pre-stats tab, uncheck “BET brain extraction” and select “None” for “Motion correction” (as these were already done above). Register the functional (BOLD) images to the anatomical (SPGR) images, and then to a standard (MNI) image. This results in the generation of transformation matrices, which are used later during analysis to warp the seeds selected in standard space into the subject’s brain space.
3. Use the generated transformation matrix (named “standard2example_func.mat”) and transform CSF and white matter ROIs into individual BOLD space.
   1. Extract the time series from the CSF and white matter ROIs using the fslmeants command, using the ROI in individual subject space as a mask. Normalize the extracted time series using the software “R”. These time series are used as regressors in the GLM later to remove the corresponding artifactual signals from the analysis.
4. The next step is removal of subject motion related artifacts. For regression of the motion parameters, set the following within FSL FEAT before running it.
   1. Within the data tab, use motion-corrected and brain-extracted file as inputs, set the TR value to correspond to your dataset. Set high-pass filtering using a 100 sec filter. The high pass filtering will remove signals of no interest, which are of very low frequency. A low pass filter to remove high frequency signals will be applied later in step 4.1.
   2. Within the Pre-stats tab, choose "None" under "Motion correction" as it was already done. Uncheck "BET brain extraction" as it was already done. Perform spatial smoothing using a 5 mm full-width half-maximum (FWHM).
   3. Within the stats tab, regress the 6 motion parameters and their temporal derivatives. Select “None” for convolution and check “Apply temporal filtering.” Use the output of FSL MCFLIRT to get text files of movement parameters, which can be then input into the FEAT analysis model to regress these in a General Linear Model (GLM)
   4. Also add the CSF and white matter signals that were extracted and normalized in previous steps to the GLM. Select “None” for convolution, add temporal derivative, and uncheck “Apply temporal filtering”.

4. Statistical Methods

1. The residuals from preprocessing described above should be used for seed-based correlation. These residuals should be first passed through a low-pass filter of 0.1 Hz, then demeaned by subtracting the mean, dividing by the standard deviation, and then scaled by adding 100. Seeds should be defined with a diameter of 6 mm in the standard MNI space using MRICron software.
2. The posterior and the anterior seeds should correspond to the following coordinates: (1) Rsp/ PCUN region (x=2, y=-60, z=36) and (2) ventromedial prefrontal cortex (vmPFC; x=3, y=60, z=-1). These seed locations have been defined within healthy controls and are transformed to the subject space in the next step^17-19.
   1. The seeds should subsequently be transformed to each subject's individual functional brain space from the standard MNI space. For this, use the transformation matrix generated above (named “standard2example_func.mat”) to transform the seed from standard (MNI) space to the individual functional (BOLD) space.
   2. Use the fslmeants command to extract the time series from the previously demeaned and scaled residual, using the seed in the individual subject space as a mask. Normalize the extracted time series using the software “R”.
3. Partial correlations between the seed voxels and all other brain voxels should be calculated separately for each subject for each run. For this, within the FSL FEAT GUI, select “First-level analysis”, and then “Stats + Post-stats”. Within the Data tab, the previously demeaned and scaled residual should be used as input into FEAT.
4. Set the High pass filter cutoff to 10,000, as the residual is already high passed at 100 sec. Within the Stats tab, unselect "Use FILM prewhitening", and use the previously extracted and normalized seed time series in the GLM. Within the Post-stats tab, set the desired Z-stat threshold to a value of 2.0.
5. Prior to running group analysis combining runs within subjects, a Fisher's Z transform should be performed on the COPE (Contrast of Parameter Estimates) file generated from the previously run correlation analysis (step 4.3). Copy registration data from the “reg” directory of the FEAT analysis done in step 3.1 into the correlation run of step 4.3. 
6. Run a higher-level analysis by combining runs within each subject. For this, within the FSL FEAT GUI, select “Higher-level analysis”, and then “Stats + Post-stats”. Within the “Data” tab, choose "Inputs are lower-level FEAT directories" and enter subject's runs from step 4.4. Within the “Stats” tab, choose “Mixed effects: Simple OLS”. Set up model as mean effect; enter value of 1 for each of the subject's run.
7. To combine data over runs between subjects, an Ordinary Least Squares (OLS) simple mixed effects analysis should be used. For this, within the FSL FEAT GUI, choose”Higher-level analysis”, and “Stats + Post-stats”. Within the Data tab, choose "Inputs are lower-level FEAT directories" and enter subjects' combined runs from step 4.5.
8. Within the Stats tab, choose “Mixed effects: Simple OLS”' Set up a model as 3 groups; enter value of 1 for the group each subject belongs, 0 otherwise. Group analysis should be done on each voxel using a one-way ANOVA with three levels which corresponded to the three groups (right TLE, left TLE, and healthy controls).
9. To threshold the Z statistic images use cluster forming threshold of Z > 2.0 and corrected cluster significant threshold of p=0.05 ²⁰. To obtain correct z-values on the correlation map, a reverse Fisher's Z transform should be performed on the results.
10. The following specific contrasts should be compared (1) right TLE > controls; (2) left TLE > controls; (3) right TLE > left TLE; (4) left TLE > right TLE; (5) control > right TLE; (6) control > left TLE; (7) TLE (combined right and left) > control; and (7) control > TLE (combined right and left).

Results

Figure 1 shows the DMN revealed with connectivity from a posterior seed (Rsp/ PCUN, red-yellow colors) and an anterior seed (vmPFC, blue-green colors) and compares the networks found in the different subject groups (Figures 1A-C) and between each other, namely healthy controls compared to all patients with TLE (Figures 1D and 1E), and then healthy controls compared separately to left TLE (Figures 1F and 1G) and righ...

Discussion

Epilepsy is thought to be a network disease, and abnormalities of the involved networks are present during seizures and in the interictal state²¹. Task-based fMRI has been used to analyze abnormalities of the language and memory networks in TLE⁸. FcMRI has inherent advantages in studying the DMN¹² as it is a network mainly active in the resting state. The DMN is a network of brain regions that has been found to be active in awake individuals who are left undisturbed and are engaged in spo...

Materials

Name	Company	Catalog Number	Comments
MRI machine
Linux workstation with image analysis software installed