Multiple sclerosis (MS) is a chronic neuroinflammatory and neurodegenerative disease driven by complex pathophysiological mechanisms that contribute to neurologic impairment and disability progression. This thesis explores protein and metabolite biomarkers and employs machine learning models to predict disease trajectory in MS, aiming to improve diagnosis, prognosis, and treatment response.

To address this objective, comprehensive proteomic profiling was conducted on cerebrospinal fluid (CSF) and plasma from individuals with early-stage MS and healthy controls. Differentially expressed CSF proteins were enriched in pathways related to B cell activation, and a linear regression model incorporating 11 of these proteins and age effectively predicted long-term disability progression for up to 13 years. Additionally, logistic regression models based on CSF proteins could distinguish MS from controls and predict short-term disease activity.

Since disease progression in MS is influenced not only by baseline pathology but also by therapeutic interventions, further focus was placed on how dimethyl fumarate (DMF), a common oral treatment in MS, affects plasma and CSF proteomic profiles related to pathological mechanisms in MS. Longitudinal analysis revealed DMF-induced reductions in inflammatory proteins associated with T-helper 1 immunity, underscoring the drug’s ability to modulate this key pathologic pathway. Importantly, baseline levels of specific axonal, glial and myelination-related proteins differentiated responders from non-responders, suggesting a potential role for these biomarkers in guiding treatment selection and optimizing therapeutic strategies.

Expanding the focus beyond proteomics, metabolic dysregulation in MS was examined through the analysis of CSF and normal-appearing white matter (NAWM) metabolites across different disease stages. Metabolites that were most strongly associated with clinical factors in MS were linked to mitochondrial dysfunction, axonal integrity, astrogliosis and demyelination. CSF biomarkers in linear regression models could distinguish MS from unspecific but similar neurological symptoms and differentiate between subtypes of the disease. A random forest model incorporating NAWM metabolites demonstrated high predictive power for long-term disability progression for up to 16 years, offering a promising non-invasive tool for MS prognosis.

Together, these studies provide a comprehensive perspective on MS pathophysiology, presenting protein- and metabolite-based models for enhanced diagnosis, treatment response monitoring, and long-term disease progression assessment. The biomarkers suggested in this thesis lay the groundwork for future translational applications in clinical practice.