The bi-phasic behaviour of grey matter networks after the first demyelinating attack.

Multiple sclerosis can be considered a network disease. Accumulating evidence recognizes the following importance of grey matter networks: they only require high-resolution anatomical scans for their extraction, they capture changes beyond detectable atrophy and their alteration is associated with disability progression and cognitive impairment. Therefore, it is crucial to understand their behaviours over the initial years of the disease. This observational longitudinal study aimed to investigate changes in grey matter networks after the first demyelinating attack, and how they correlate with brain damage, disability, and conversion to multiple sclerosis over 3-5 years. So far, in multiple sclerosis, network construction has only been based on cortical grey matter, neglecting a possible role for deep grey matter. We applied a radiomics-based network methodology incorporating both deep and cortical grey matter. Patients recruited within 3 months of disease onset and healthy controls attended study visits at 6 months, 1 year, 3 years and 5 years. Study visits included physical and cognitive scales and brain MRI scans. Individual grey matter networks were constructed by computing the correlations between T1w-based radiomic features extracted from any pair of regions of the Brainnetome atlas and characterized with measures of network integration (global efficiency and characteristic path length), segregation (clustering coefficient and modularity), resilience (assortativity) and smallworldness. Additionally, eigenvector centrality was computed for all brain regions as a measure of nodal influence. We enrolled 89 patients (median follow-up 7 months, range 0-75) and 31 healthy controls. Patients showed higher global efficiency, lower shortest characteristic path length and higher smallworldness than controls suggesting a reorganization that prioritize more efficient global communication over local processing. Over time, patients' networks converged towards healthy controls' values by increasing the shortest characteristic path length and decreasing the smallworldness. Assortativity, and the eigenvector centrality in the right ventromedial putamen decreased compared with controls. All the observed changes were driven by non-converters to multiple sclerosis. This study shows that grey matter networks adopt a biphasic behaviour. They respond to the demyelinating event with an increase in nodal integration and then converge to healthy control values. In the process, however, their network resilience is compromised. This suggests that a single demyelinating event has longer-lasting effects on grey matter networks, even in non-converters, and that studying these networks may reveal relevant changes that are not captured by conventional MRI in the early years of the disease.