Altered functional and structural connectivities in patients with MS

Patients.

We studied 12 consecutive right-handed⁸ patients with RRMS.⁹ There were 11 women and 1 man; their mean age was 36.0 years (range = 25–46 years), their median disease duration was 8 years (range = 3–10 years), and their median Expanded Disability Status Scale (EDSS) score¹⁰ was 1.5 (range = 1.0–2.5). At the time MRI was performed, all patients had been relapse- and steroid free for at least 6 months. The main inclusion criteria were the absence of clinical involvement of the right upper limbs and no previous relapses involving the right upper limb. All patients were treated with immunomodulatory drugs (two of them were treated with glatiramer acetate, and the remaining 10 patients received one of the three available formulations of interferons beta). None of them complained of treatment-related side effects, such as fatigue, at the time of fMRI acquisition. Fourteen sex- and age-matched right-handed⁸ healthy volunteers with no previous history of neurologic dysfunction and a normal neurologic exam served as controls (11 women and 3 men, mean age = 33.7 years, range = 24–42 years). All subjects were assessed clinically by a single neurologist, who was unaware of the MRI and fMRI results. The mean laterality quotient at the Edinburgh Handedness Inventory⁸ was 0.96 (range = 0.90–100) in healthy volunteers and 0.97 (range = 0.90–100) in MS patients. Local ethical committee approval and written informed consent from all subjects were obtained before study initiation.

Functional assessment.

Upper-right limb motor functional assessment was performed for all individuals on the same day MRI was acquired, using the nine-hole peg test (9-HPT) and the maximum finger-tapping frequency. The maximum finger-tapping rate was observed for two 30-second trial periods outside the magnet, and the mean frequency to the nearest 0.5 Hz entered the analysis. No difference was found in the performance of these tests between MS patients and healthy controls (time to complete the 9-HPT: mean = 19.8 seconds, SD = 2.5 seconds for controls; mean = 21.4 seconds, SD 2.6 seconds for patients; finger-tapping rate: mean = 3.4 Hz, SD = 0.8 Hz for controls; mean = 3.6 Hz, SD = 0.7 Hz for patients).

Experimental design.

Using a block design (ABAB) where six periods of activation were alternated with six periods of rest (each period of rest and activity consisting of five measurements), the subjects were scanned while performing a simple motor task consisting of repetitive flexion-extension of the last four fingers of the dominant right hand moving together. A standardized frame was used to restrict the amplitude of the extension to approximately 3 cm from the base. The movements were paced by a metronome at a 1-Hz frequency. Patients were trained before performing the study. The subjects were instructed to keep their eyes closed during fMRI acquisition and were monitored visually during scanning to ensure accurate task performance and to check for additional movements (e.g., mirror movements).

fMRI acquisition.

Brain MRI scans were obtained using a 3.0-T Philips Intera scanner (Philips Medical Systems, Best, the Netherlands) with a gradient strength of 40 mT/m and an eight-channel head coil. Functional MR images were acquired using a T2*-weighted single-shot echo-planar imaging sequence (echo time [TE] = 35 milliseconds, flip angle = 85 degrees, matrix size = 128 × 128, field of view [FOV] = 240 mm², repetition time [TR] = 3.7 seconds). Thirty interleaved 4-mm-thick axial slices, covering the whole brain, were acquired during each measurement. All slices were positioned to run parallel to a line that joins the most inferoanterior and inferoposterior parts of the corpus callosum.¹¹ Shimming was performed for the entire brain using an autoshim routine, which yielded satisfactory magnetic field homogeneity.

Structural MRI acquisition.

On the same occasion and using the same magnet, the following images of the brain were also obtained from all subjects: 1) dual-echo turbo spin echo sequence (TR/TE = 3000/45 to 120 milliseconds; echo train length = 5; flip angle = 90 degrees; matrix size = 256 × 256; FOV = 230 × 230 mm²; 30 contiguous, 4-mm-thick, axial sections); 2) three-dimensional T1-weighted fast field echo (TR/TE = 25/4.6 milliseconds; flip angle = 30 degrees; matrix size = 256 × 256; FOV = 230 × 230 mm²; 220 contiguous, axial slices with voxel size = 0.89 × 0.89 × 1 mm); 3) pulsed-gradient SE echo planar with SENSE (acceleration factor = 2.5; TE/TR = 80/8283.2 milliseconds; acquisition matrix size = 96 × 96; FOV = 240 × 240 mm²; 55 contiguous, 2.5-mm-thick axial slices; after SENSE reconstruction, the matrix dimension of each slice was 256 × 256, and in-plane pixel size was 0.94 × 0.94 mm) and diffusion gradients applied in 32 noncollinear directions, using a gradient scheme that is standard on this system (gradient overplus) and optimized to reduce echo time as much as possible. Two optimized b factors were used for acquiring diffusion-weighted images¹² (b₁ = 0, b₂ = 1000 s/mm²). Fat saturation was performed to avoid chemical shift artifacts. Slices were positioned as those of the fMRI data set.

fMRI analysis.

All image postprocessing was performed on an independent computer workstation (Sun Sparcstation, Sun Microsystems, Mountain View, CA). FMRI data were analyzed using the statistical parametric mapping (SPM2) software developed by Friston et al.¹³ Before statistical analysis, all images were realigned to the first one to correct for subject motion, spatially normalized into the standard space of SPM, and smoothed with a 10-mm, three-dimensional Gaussian filter. Subjects were included in the subsequent statistical analysis if they had a maximum translation/rotation lower than 3.0 mm in the x, y, and z planes.

Changes in blood oxygenation level–dependent contrast associated with the performance of the motor task were assessed on a pixel-by-pixel basis, using the general linear model and the theory of Gaussian fields.¹³ Specific effects were tested by applying appropriate linear contrasts. Significant hemodynamic changes were assessed using t statistical parametric maps. The intragroup activations and comparisons between groups were investigated using a random-effect analysis,¹⁴ with one-sample or two-sample t tests, as appropriate. We report activations below a threshold of p < 0.05, corrected for multiple comparisons (family-wise error). To achieve a more precise definition of the anatomic locations of activated regions in each individual, three-dimensional fast field echo images from each subject were coregistered with the corresponding fMRI data sets and normalized into the same standard space. Then, fMRI results were superimposed onto these high-resolution images and, using cluster analysis on a patient-by-patient basis, we evaluated the coordinates of the centers of the activations in those areas with significantly different relative activations at group analysis.

Analysis of functional connectivity.

The interactions between different regions involved in the motor task were calculated using a dynamic causal model (DCM) approach.¹⁵ Treating the brain as an input-state-output system, DCM estimates how hemodynamic activity in a given brain region (output) depends on its interconnectivity with other brain regions whose activity correlates with the task. Definition of brain regions included in the DCM relied on activation clusters obtained from the analysis of fMRI data from the entire study group (patients vs healthy controls). Time series, which were adjusted for the effect of interest, were extracted from a spherical volume (5-mm radius) centered at the most significant voxel within each cluster in the t statistical parametric map displaying brain areas activated during the motor task at a p < 0.001, uncorrected, in each subject. Volumes of interest were extracted from the primary sensorimotor cortex (SMC), bilaterally, the right cerebellum, and the left supplementary motor area (SMA). We assumed that the effect of the task entered the network via the activation cluster within the left (L) primary SMC. This assumption was based on conventional wisdom and was confirmed by the results obtained from the within- and between-group analyses of task-related activations seen in this study.

For each subject, the DCM was used to investigate the intrinsic connectivity pattern between all regions of interest previously extracted, which were involved in the motor task. To this end, a DCM was constructed in which all regions were assumed to be connected bidirectionally with each other. The intrinsic connectivity strength coefficients were then estimated using a Bayesian approach.¹⁵ A one-sample t test was used to assess the within-group strength of the connectivity coefficients. A Mann–Whitney two-independent-samples test was used to compare the strength of the connectivity coefficients previously estimated between controls and MS patients.

Structural MRI postprocessing.

All structural MRI analysis was performed by a single, experienced observer, unaware of who the scans belonged to and blinded to the fMRI results. Brain lesions were identified on the proton density-weighted scans, and lesion loads (LL) were measured using a local thresholding segmentation technique.¹⁶ The corresponding T2-weighted images were always used to increase confidence in lesion identification.

For DT MRI images, the DT was estimated using a nonlinear regression (Marquardt–Levenberg method), assuming a monoexponential relationship between signal intensity and the b-matrix components.¹⁷ After diagonalization of the estimated tensor matrix, the two scalar invariants of the tensor, mean diffusivity (MD) and fractional anisotropy (FA), were derived for every pixel. Then, using the VTK CISG Registration Toolkit,¹⁸ the rigid transformation needed to correct for position between the b = 0 images (T2 weighted, but not diffusion weighted) and T2-weighted images was calculated. Normalized mutual information¹⁹ was the similarity measure chosen for the matching. The same transformation parameters were then used to coregister the MD and FA images to the T2-weighted images.

A tractography algorithm was used to reconstruct the anatomy of brain white matter (WM) fiber bundles, by means of DT-derived indexes calculated inside the tracts. Because the direct application of tractography in patients is hampered by the presence of regions with decreased FA (indicating an increased uncertainty of the principal diffusion direction), a tractography algorithm was used to construct the probability maps for each WM fiber bundle analyzed, using data from healthy subjects²⁰; these probability maps were then applied to patients' data to derive diffusivity measures inside the WM fiber bundle of interest. In details, we used the following procedure. First, an FA atlas was created using data from healthy volunteers. To this aim, a registration algorithm¹⁷ was used to register controls' T2-weighted images to the Montreal Neurological Institute atlas²¹ with an affine transformation. Then, the transformation was applied to FA maps, already registered into the T2-weighted space, before their averaging to obtain the atlas. Then, using the same data, the probability maps for the corpus callosum (CC), the corticospinal tract (CST), the optic radiation (OR), the superior fronto-occipital fasciculus (SFOF), the uncinate fasciculus, the dentatorubrothalamic (DRT) tract, the cingulum, and the superior longitudinal fasciculus (SLF) were created. The trajectories of the different tracts were obtained by placing regions of interest (ROIs) believed to contain a section of the desired WM fiber bundles on both FA and color-coded maps for the principal diffusion directions,²² by cross-referencing neuroanatomic^23,24 and previous tractography^20,25–27 studies. A two-ROI approach was used to define the CST, the OR, the SFOF, the uncinate fasciculus, and the DRT. Trajectories were checked by qualitative visual assessment after reconstructing them in three dimensions. Figure 1 shows an example of the WM fiber bundle reconstructions in an individual subject before constructing the probability map. Once trajectories were obtained, these were transformed using the same change calculated at the previous step, before their averaging to produce tract probability maps (a threshold of 0.4 was applied).²⁸ Because of the possible presence of atrophy, a nonlinear deformation algorithm²⁹ was used to transform patients' FA maps into the atlas space. Then, WM fiber bundle probability maps were superimposed onto the transformed MD and FA maps. In this way, the contribution of each pixel to the mean value was weighted by its value in the probability map. For each WM fiber bundle, average MD and FA were obtained. With the exception of the CST, for WM fiber bundles with a bilateral location in the brain, the averages of the MD and FA values measured in the R and the L fiber bundles entered the analysis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Behavior of the main white matter fiber bundles of the study in a single subject, before spatial normalization

(A) corticospinal tract, (B) corpus callosum, (C) optic radiation, (D) superior fronto-occipital fasciculus, (E) uncinate fasciculus, (F) dentatorubrothalamic tract, (G) cingulum, and (H) superior longitudinal fasciculus.

Volumes of T2-visible lesion in the different WM fiber bundles were derived by applying fiber bundles probability maps obtained using DT tractography and calculating the volume of lesions inside the fiber bundles. The method is described in detail elsewhere.²⁰

Statistical analysis.

Differences of structural MR-derived metrics between patients and controls were assessed using a nonparametric Mann–Whitney test for two independent samples. Univariate correlations between functional and structural changes were assessed using the Spearman correlation coefficient. All statistical analysis was performed using SPSS for Windows (version 13.0).

Structural MRI.

All healthy volunteers had normal brain MRI scans. In MS patients, the median T2 LL was 11.4 mL (range = 1.0–24.3 mL). In table 1, the median values of T2-visible lesions burdens of each of the different WM fiber bundles of interest are reported. No significant difference was found for LL along the right (R) and L CST (mean R CST LL = 0.24, SD = 0.64; mean L CST LL = 0.32 mL, SD = 0.8; p = 0.8). In table 1, the average MD and FA values of the different WM fiber bundles analyzed in healthy controls and MS patients are reported. Significant differences were found in average MD and FA values of all the WM fiber bundles analyzed between MS patients and healthy controls. In MS patients, average MD was significantly higher in the L CST compared with the R CST (p = 0.001). In table 2, the correlations found in patients with MS, between MD and FA average values of the different WM fiber bundles analyzed, and regional and total LLs are reported. No correlation was found between structural MRI metrics and clinical findings.

Functional MRI: Between-group analysis.

All subjects performed the tasks correctly, and no additional movements were observed during fMRI acquisition. In figure 2, the brain areas with significant activations detected while performing the task in healthy subjects and MS patients are shown. Compared with healthy volunteers, MS patients had more significant activations of the contralateral (L) SMA (SPM space coordinates: −10, −18, 64), the contralateral (L) primary SMC (SPM space coordinates: −26, −16, 48), and the ipsilateral (R) anterior lobe of the cerebellum (SPM space coordinates: 16, −34, −22) (figure 2). No significant differences were found in coordinates of centers of activations of the above-reported areas between patients and controls.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Cortical activations on a rendered brain from right-handed healthy subjects and patients with relapsing-remitting multiple sclerosis (MS) during performance of a simple motor task

Cortical activations on a rendered brain from right-handed healthy subjects (top row) and patients with relapsing-remitting MS (middle row) during performance of a simple motor task with their clinically unimpaired and fully normal functioning right hands (within-group analysis, one-sample t tests, p < 0.05, corrected for multiple comparisons). The between-groups differences (areas with more significant activations in MS patients than controls) are shown in the bottom row (two-sample t test, p < 0.05, corrected for multiple comparisons). Images are in neurologic convention. See text for further details.

Functional connectivity analysis.

The results of the analysis of within-group individual correlations between activity changes in anatomically connected regions are reported in table 3. Compared with controls, MS patients had increased functional connectivity between the R primary SMC and the R cerebellum (p = 0.01) and between the L SMA and the L primary SMC (p = 0.04) (figure 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3 Differences of functional connectivity in the motor network during performance of a simple motor task with their clinically unimpaired and fully normal functioning right hands in patients with relapsing-remitting multiple sclerosis (RRMS) and healthy subjects

The light blue arrows identify the connections with similar path coefficients in healthy subjects and in RRMS patients. The red arrows identify the connections with higher path coefficients in RRMS patients than in healthy subjects. SMA = supplementary motor area; SMC = sensorimotor cortex.

Correlations between measures of functional and structural connectivities.

In patients with MS, significant correlations were found between a) coefficients of connectivity between the R primary SMC and the R cerebellum and average FA (r = −0.73, p = 0.02) and average MD (r = 0.85, p = 0.004) of the DRT tracts, and b) coefficients of connectivity between the L SMA and the L primary SMC and CST LL (r = 0.64, p = 0.04). This correlation remained significant when considering the LL along the L CST separately (r = 0.66, p = 0.03).