METHODS

Clinical details

We recruited 11 right handed healthy controls (median age 37 years (range 27 to 43); five male, six female) and 14 right handed outpatients with clinically stable relapsing-remitting multiple sclerosis (median age 37 years (range 18 to 52); five male, nine female; mean disease duration 3.6 years (range 1 to 8)). Patient disability was minimal as assessed by the expanded disability status score (EDSS) (median EDSS, 1.0 (range 1.0 to 2.5)). Their clinical features are given in table 1. All patients had been relapse-free for at least three months before the study. No patients had tremor or ataxia. None had weakness or sensory symptoms, or signs affecting the right hand.

All subjects gave their informed consent for participation in the study, which was approved by the ethics committee of the faculty of medicine in the University of Siena, Italy.

Functional assessment

Patients showed no symptoms or clinical signs of upper limb impairment and were able to perform simple figure writing as well as the controls. Before entering the scanner, patients and controls traced around five predrawn figure 8s as accurately and as quickly as possible. The mean area covered outside of the predrawn figures and the mean time taken to draw around them were measured to provide an index of subject performance (performance product  =  mean time × mean area).

fMRI scanning and paradigm

Imaging was undertaken with a Philips Gyroscan ACS NT 1.5 T scanner. BOLD multishot echo-planar images were acquired (time of repetition (TR) = 3000 ms, time of echo (TE) = 40 ms, flip angle = 90°, 20 contiguous axial slices of 7 mm thickness acquired parallel to a line defined by the anterior and posterior commissures and including brain from the base of the cerebellum to the vertex, field of view (FOV) 250×250 mm², 128×128 matrix). High resolution whole brain structural scans were acquired using a T1 weighted sequence (TR = 20 ms, TE = 3 ms, flip angle = 20°, 25 contiguous slices of 5 mm thickness acquired parallel to the AC–PC plane). Turbo spin echo T2 weighted scans were also obtained for each subject (TR = 2075 ms, TE = 90 ms, FOV = 250×250 mm², 256×256 matrix, signal average = 1, providing 50 contiguous slices of 3 mm thickness acquired parallel to the AC–PC plane).

Subjects lying supine in the scanner were asked to write “8” repeatedly on paper in a cursive manner (without lifting the hand) for 30 seconds with a pencil in their right hand, to complete a figure every second. This was alternated with 30 second periods of rest. During both conditions the subjects were asked to fix their gaze on a red flashing light cue positioned at the end of the magnet. Each subject undertook six paired rest and writing blocks for a single functional scan and had six functional scans during each session. All subjects were monitored visually throughout the scanning for non-task-related movements.

Image analysis

The data were processed and analysed using advanced image analysis tools from the Oxford Centre for Functional Magnetic Resonance Imaging of the Brain (FMRIB) Software Library (available at http://www.fmrib.ox.ac.uk/fsl). Before statistical analysis, motion correction using MCFLIRT,¹⁷ spatial smoothing (Gaussian kernal of 5 mm full width at half maximum), intensity normalisation, and non-linear high pass temporal filtering were applied. Statistical analysis was done using the general linear model with FMRIB’s improved linear model (FILM)¹⁸ to generate activation maps expressing relative signal change in active versus rest blocks. Cluster detection was undertaken on voxels at a threshold of z >2.3. A probability threshold of p<0.01 was then applied to determine significant clusters.

Using FLIRT,¹⁷ the functional images were registered to the high resolution T1 weighted scan and transformed into a standard brain space based on the MNI 152 template.¹⁹ Mixed effects analysis using FLAME²⁰ was used to contrast activation patterns between groups.

Generation of ROI masks

Regions of interest (ROI) were defined bilaterally on each individual’s high resolution T1 weighted structural image using anatomical landmarks identified on the T1 weighted structural scan (fig 1):

• Primary motor cortex (M1), defined as the posterior half of the precentral gyrus, extending on the lateral surface of the cortex to the midline and from the vertex of the brain to the level of the superior aspect of the lateral ventricles caudally.

• Premotor cortex (PM), defined as the anterior half of the precentral gyrus, the precentral sulcus, and the middle frontal gyrus up to the lateral fissure anteriorly, the lateral surface of the cortex to the midline and from the vertex of the brain to the level of the superior aspect of the lateral ventricles caudally.

• Crus I, defined in the lateral cerebellar hemisphere,²¹ extending rostro-caudally between the superior and inferior aspects of the dentate nucleus.

• Dentate (DN), defined as midline and paravermian central to cerebellar white matter.²¹

Calculation of lesion volume

Lesion volumes were semimanually defined on T2 weighted axial scans for each patient using edge thresholding.²² Hyperintense lesion volumes of more than 3 mm in diameter were summed to provide a total brain lesion volume (table 1).

Calculations of brain volume

Measurements of brain size (normalised for head size) were made for each patient on T1 weighted structural scans as an estimate of disease burden related atrophy, using SIENAX, a fully automated tool²³ (table 1).

Data analysis

The mean positive signal intensity change in each ROI was obtained from the parameter estimates image for each subject. This allowed the calculation of a lateralisation index (LI  =  [C−I]/[C+I], where C  =  mean contralateral and I  =  mean ipsilateral signal intensity change) for M1, PM, crus I, and DN regions. The LI has values between 1 (positive signal change only in the side contralateral to the hand moved) and −1 (positive signal change only in the side ipsilateral to the hand moved). To assess whether activation was significantly less lateralised in patients than in control subjects, two tailed Mann–Whitney tests were used to compare signal intensity changes and LIs for each ROI. Relations between cortical and cerebellar signal intensity change were assessed for correlations using a Pearson correlation coefficient. As described in previous reports (see, for example, Koski and Paus²⁴ and Horwitz et al²⁵), such correlations can be used as measures of functional connectivity between brain regions. A correction for multiple comparisons was applied.

RESULTS

We studied patients with predominantly early relapsing-remitting multiple sclerosis, along with age matched, healthy controls. The lesion load was modest (mean 5.7 cm³; range 0.2 to 15.4 cm³) for the patients (n = 14), and their mean normalised brain volume was similar to that of the controls (mean (SD) brain volume of patients, 1.59×10³ (100) cm³; controls, 1.61×10³ (40) cm³). Task performance was assessed by a timed figure writing trial. There was no difference in either time (t = −0.774, df = 23, p = 0.447), area included between the template and traced lines (t = 0.054, df = 23, p = 0.958), or performance product (t = −0.711, df = 23, p = 0.484) between controls and patients. No “mirror” movements of the contralateral hand were observed during the writing task for any of the patients.

Both patients and controls activated a distributed network of regions associated with movement, including the motor cortex and cerebellum (fig 2; table 2). There were no differences in localisation between centres of activation in patients and controls. However, consistent with previous studies,⁷ patients showed a more bihemispheric pattern of motor cortex activation. This hemispheric activation asymmetry was assessed quantitatively for premotor (PM) and primary motor (M1) cortical regions of interest. There was a reduction of over 50% in the PM LI in patients (U = 32.50, p = 0.013) (mean (SD) PM LI for controls, 0.37 (0.18); mean PM LI for patients, 0.18 (0.18); (fig 3)). This difference in LI primarily was caused by increased signal intensity changes in the right premotor cortex in patients relative to controls (U = 29, p = 0.008). M1 activation lateralisation was not significantly different between patients and controls (mean M1 LI for controls, 0.60 (0.10); mean M1 LI for patients, 0.54 (0.13)).

The relations between the relative changes in cerebral cortex and cerebellar signal intensities within individual subjects offer measures of neocortical-cerebellar functional connectivity (see, for example, Koski and Paus²⁴ and Horwitz et al²⁵). These were different between patients and controls (table 3). The control group showed the expected^26,27 strong “crossed” correlation between activation changes in the left M1 and right dentate ROI, which was not found in multiple sclerosis patients. No ipsilateral correlations were found in controls, but patients showed a significant ipsilateral correlation between the left PM cortex and the left cerebellar hemisphere (crus I) (table 3).

No significant relation was found between either LI or regional signal changes and brain volumes, lesion volumes, or EDSS.

DISCUSSION

This study extends the description of changes that occur in brain motor control in response to injury from multiple sclerosis. Along with other recent reports,^8,10,28 it emphasises that functional changes occur early in the clinical course of multiple sclerosis, and that they need not be accompanied by identifiable performance deficits. As found in similar previous studies, the patients with multiple sclerosis showed increased ipsilateral premotor activation relative to controls.^7,8,28 What is novel here is the characterisation of the associated, altered neocortical–cerebellar functional connectivity. Increased connectivity was found in patients with multiple sclerosis between the premotor cortex contralateral to the hand moved and the cerebellar cortex (the input region for the cortico-ponto-cerebellar projection system) on the same side, with decreased connectivity relative to healthy controls for the crossed cerebellar output (dentate) projections to the same premotor cortex.²⁹ Potentially adaptive responses to injury thus affect not only activity in specific regions, but also interactions between regions in the distributed network for motor control. The results suggest a role for the cerebellum in mediating adaptive changes to injury from multiple sclerosis.

The cerebellum may have a special role in forms of brain functional plasticity.^30–32 Cerebellar connections allow rapid, activity dependent alterations in synaptic efficiency.^33,34 Functional imaging has defined increases in cerebellar activation with learning that involve both the cerebellar hemispheres and the dentate nucleus, the origin of efferent dentato-thalamo-cortical and dentato-thalamo-striate tracts.²⁹ A notable aspect of earlier stages of motor learning is increased recruitment of the cerebellum bilaterally, in a similar fashion to the pattern found here for patients.¹⁴ Damage to the cerebellum impairs forms of implicit learning.^33,35 Increased cerebellar activity is associated with better outcome in patients after stroke.³⁶

An increase in functional connectivity to the cerebellar hemisphere contralateral to the hand moved was observed here with the patients. Increases in activity in the contralateral cerebellar hemisphere are also found with motor learning in healthy subjects.¹⁴ This may reflect recruitment of uncrossed ponto-cerebellar projections. In studies of the cat, while the majority of ponto-cerebellar projections are crossed, almost a quarter were found to project to the same side.³⁷ This provides a pathway for input from the motor cortex to the cerebellar hemisphere on the same side.

The motor cortex, cerebellum, thalamus, substantia nigra, and basal ganglia together constitute key elements of a distributed motor control network. Subcortical structures project widely to neocortical regions involved in motor control.³⁸ The increased functional connectivity between the motor cortex and cerebellum found in the patients involved the premotor and not the primary motor cortex. Changes in relative premotor cortex activity in the patients account for the more bihemispheric pattern of motor cortex activation in the patients. Previous studies of patients after stroke or neurorehabilitation have suggested the particular importance of activity changes in the premotor cortex to functional recovery.^11,39

Observation on the changes in functional connectivity and patterns of premotor activity may be related. Cerebellar activation contralateral to the hand moved provides a potential pathway by which premotor cortex ipsilateral to the hand moved might be recruited for motor control, through crossed dentate-cortical projections. A subcortical pathway linking activity in motor cortices of the two hemispheres is suggested by observations such as those showing that motor cortex activation ipsilateral to the hand moved does not require an intact corpus callosum.^40,41 However, the directionality of the functional interactions between cerebellum and premotor cortex cannot be defined uniquely on the basis of the functional connectivity data here. There are alternative models consistent with the data. One possibility, for example, is a direct primary interaction between the motor cortices of the two hemispheres, which in turn drives more bilateral cerebellar activity—although this may provide a less compelling explanation, given that the patients also show decreased functional connectivity between the cerebellar activation ipsilateral to, and neocortical activity contralateral to, the hand moved.

There are a few technical issues that should be considered in the interpretation of these data. First, the use of such a relatively mildly affected and homogeneous group (as well as its small size) limits confounding from differences in performance, but precludes adequate testing of the relations between activation in specific brain areas and clinical variables such as disease burden, duration, and disability. Not unexpectedly, correlations between functional changes and these measures were not found (data not shown). Second, early reports argued that increased ipsilateral motor cortex activity in patients may reflect “mirror” movements, raising this as a concern here.⁴² Our patients (like other patient groups^7,43) showed no clinical evidence of mirror movements. Third, in previous studies, we used the relative number of significantly activated voxels for calculation of LI.⁷ However, as the absolute number of activated voxels is a function of the statistical threshold chosen, absolute values of such LIs are not interpreted easily and may have higher variance because of dependence on noise in the near threshold voxels.⁴⁴ These problems were limited by calculation of LI from relative signal intensity changes.⁴⁵

Involvement of multiple processing regions of a distributed motor network in adaptive compensation for the brain injury of multiple sclerosis implies that optimising recovery demands optimising functions of multiple targets and their interactions. Altering levels of diffusely projecting modulatory neurotransmitters that may affect interactions between brain regions is thus a theoretically attractive approach.⁴⁶ Second, because of the distributed nature of control, the relation between lesion distribution and functional impairment may be expected to be complex. In fact, previous work has shown only minimal gains in the strengths of correlations between measures of functional impairment in the motor system and lesion burden when the volume in the corticospinal tract alone is used rather than global brain lesion load.⁴⁷ Together, these observations suggest that in established disease the focus for repair of injury should be on the global rather than the system specific disease burden. Finally, the recognition that interactions between spatially distant nodes in a coordinated network may promote recovery emphasises that in multiple sclerosis, a disease characterised by substantial diffuse axonal destruction, the potential mechanisms of recovery are impaired with progression of the disease, just as are primary effector pathways.