Predictive potential of serum and cerebrospinal fluid biomarkers for disease activity in treated multiple sclerosis patients

Abstract

Background: Our objective was to explore various biomarkers for predicting suboptimal responses to disease-modifying treatments (DMTs) in patients with MS (pwMS).Methods: We conducted a longitudinal, bicentric study with pwMS stratified based on their DMTs responses. Treatment failure (TF) was defined as the onset of a second relapse, presence of two or more T2 new lesions, or disability progression independent of relapse during the follow-up period. We evaluated intrathecal synthesis (ITS) of IgG and IgM using OCB, linear indices, and Reibergrams. Free kappa light chains ITS was assessed using the linear index (FKLCi). NfL and GFAP in serum and CSF, and CHI3L1 in CSF were quantified. Quantitative variables were dichotomized based on the third quartile. Predictive efficacy was assessed through bivariate and multivariate analyses, adjusting for age, sex, EDSS, acute inflammatory activity (AI) -defined as the onset of a relapse or gadolinium-enhancing lesions within a 90-day window of lumbar puncture-, treatment modality, study center, and time from disease onset to treatment initiation. In case of collinearity, multiple models were generated or confounding variables were excluded if collinearity existed between them and the biomarker. The same methodology was used to investigate the predictive potential of various combinations of two biomarkers, based on whether any of them tested positive or exceeded the third quartile.Results: A total of 137 pwMS were included. FKLCi showed no differences based on AI, no correlation with EDSS and was significantly higher in pwMS with TF (p = 0.008). FKLCi>130 was associated with TF in bivariate analysis (Log-Rank p = 0.004). Due to collinearity between age and EDSS, two different models were generated with each of them and the rest of the confounding variables, in which FKLCi>130 showed a Hazard Ratio (HR) of 2.69 (CI: 1.35–5.4) and 2.67 (CI: 1.32–5.4), respectively. The combination of either FKLC or sNfL exceeding the third quartile was also significant in bivariate (Log-Rank p = 0.04) and multivariate (HR=3.1 (CI: 1.5–6.5)) analyses. However, when analyzed independently, sNfL did not show significance, and FKLCi mirrored the pattern obtained in the previous model (HR: 3.04; CI: 1.51–6.1). Treatment with highefficacy DMTs emerged as a protective factor in all models.Discussion: Our analysis and the fact that FKLCi is independent of EDSS and AI suggest that it might be a valuable parameter for discriminating aggressive phenotypes. We propose implementing high-efficacy drugs in pwMS with elevated FKLCi.