Introduction: Multiple sclerosis (MS) is an autoimmune disease due to chronic inflammation in the central nervous system (CNS) whose prognosis is still uncertain at disease onset.Objectives/Aims: In an effort to anticipate disease severity, we founded the OutcoMS project, in which we aim to unravel mechanisms behind a poor outcome in MS, through the use of transcriptomics at disease onset. For that, we classified patients in aggressive or non-aggressive disease according to the use of high efficacy treatments during the two first years of the disease.Methods: In a first cohort, we collected peripheral blood mononuclear cells (PBMCs) from 22 patients at their first clinical episode during clinically isolated syndrome (CIS). After >2 years, patients were sorted into 11 non-aggressive MS and 11 aggressive MS. Samples were matched with PBMCs from 11 healthy volunteers (HV) and sequenced using scRNAseq technology. Transcriptomic results were then confirmed at the protein level in a second larger cohort of 60 CIS patients composed of 29 evolving toward a non-aggressive and 31 toward an aggressive MS, and 30 HV from the OFSEP cohort with a flow cytometry assay performed on PBMCs.Results: While subsets of CD4+ T cells discriminate MS patients and HV well, we identified a subset of CD8+ T cells that is increased in CIS patients that later convert to aggressive MS compared to non-aggressive MS. This CD8+ T cell population displays a memory phenotype and cytotoxic activity similar to natural killer cells. The increase in this subset in aggressive MS was confirmed using flow cytometry on the OFSEP cohort. Furthermore, a high frequency of these cells showed predictive value for aggressive MS at onset (AUC of 0.673). When combined with sNfL(serum neurofilament light chain) and sGFAP (serum glial fibrillary acidic protein), the AUC increased to 0.843.Conclusion: Our data suggests that severe forms of MS may be driven by a subset of cytotoxic memory CD8+ T cells with NK-like properties. The frequency of this subset of cells in the blood of MS patients, in association with sNfL and sGFAP, can be used as a predictive tool at the first demyelinating event to identify patients who will require high efficacy treatment.< Leer menos