Resting state functional MRI (rs-fMRI) reveals functional correlations between cortical brain regions. One of the key functional networks is the default mode network (DMN), traditionally associated to self-reflection and working memory¹. This network has been reported to include cerebellar nodes too, which is in line with recent hypothesis that the cerebellum also contributes to non-motor functions². The structural substrate of functional networks can be investigated by reconstructing white matter (WM) pathways connecting network nodes through diffusion MRI, which can also provide a quantitative assessment of their structural integrity. In multiple sclerosis (MS) demyelination and axonal transection can disrupt WM pathways, causing motor and cognitive impairment³. The aim of this study was to investigate the cerebellar contribution to DMN-related performance in healthy controls and MS patients. To achieve this, we reconstructed the DMN structural network from diffusion-weighted data, and then used graph theory to assess network integrity and its association with cognitive performance, with and without the inclusion of cerebellar connections.

Figure 1 summarizes participants’ characteristics (Healthy Controls (HC) and different MS subtypes) and Figure 2 the MRI parameters. The Symbol Digit Modalities Test⁴ (SDMT), a screening test increasingly used in MS studies, was chosen as cognitive performance measure. Diffusion images were processed with FSL⁵ and MRtrix⁶ following an existing pipeline⁷. Track Density Imaging (TDI) maps with 1mm resolution were created by combining CSD and probabilistic tractography (maximum harmonic order: 8; seed: whole brain; step-size: 0.1mm; number of streamlines: 2500000).

In order to build the structural network, we considered as nodes published DMN cortical areas⁸ to which we added the cerebellum⁷, all in MNI152 Atlas space. The DMN map consisted in 8 symmetric binary masks: Frontal Medial Cortex (FMC); Angular Gyrus (AG); Precuneus (P); Middle Temporal Gyrus (MTG). Registration matrices from MNI152 space to TDI⁹ were computed using Niftyreg¹⁰ and FSL and applied to the nodes binary masks. Such masks in TDI space were used as seed and target regions for tractography. Superior cerebellar peduncles⁷ and cerebral peduncles¹¹ were seeded to track respectively afferent ad efferent fibers from cerebellar regions. The corpus callosum and the anterior commissure were seeded to track fibers connecting corresponding cortical regions. Reconstructed tracts were aligned to MNI152 space, summed, thresholded at 50% and binarized resulting in mean masks of the pathways. These were then transformed back to diffusion space of healthy and MS subjects. The mean fractional anisotropy (FA) value of each tract was extracted and exploited as link weight in the framework of graph theory¹² to build the connectivity matrix of the DMN network for each subject. The following parameters were computed and used in the statistical comparison with SDMT: Global efficiency of the network constituted by the traditional DMN (no cerebellum); Global efficiency of the extended DMN including the cerebellum; Mean FA value of the cerebellar tracts; Mean FA value of WM (the mask of the whole WM was obtained in T1 space through the parcellation of the brain with SPM¹³ and then aligned to diffusion space). Comparison of these metrics with SDMT scores were performed with Pearson correlation analysis in SPSS¹⁴.

Pearson correlation coefficients between MRI derived parameters and SDMT are reported along with significance levels in Figure 4 for all groups of patients and healthy subjects. Plots representing these correlations are shown in Figure 5.

The main result was that there is a significant correlation between MRI derived metrics and SDMT performance in all groups of MS patients. In particular, a stronger correlation was found considering network global efficiency rather than WM mean FA value. Pearson correlation coefficients further increased in MS when including tracts connecting the cerebellum to the DMN for the computation of the network global efficiency. This was also confirmed by partial correlation analysis when correcting for WM mean FA values and by the significant correlation also found when considering the mean FA value of only cerebellar tracts in isolation.

Our results support a DMN involvement in MS cognitive impairment, and suggest that disruption of cerebellar-cerebral connections also contributes significantly to the effects that reduction in DMN structural integrity has on cognition. Of the MRI-derived measures we assessed, global network efficiency of the extended DMN had the strongest correlation with SDMT. This suggests that the characterization of networks anatomical organization in the graph theory framework could represent a useful approach to improve the clinical relevance of MRI metrics and their capability of detecting cognition-relevant structural pathology in MS.