Superiority of serum NFL in predicting Multiple Sclerosis severity

Objective: Serum neurofilament light chain (sNFL) is becoming an important biomarker of neuroaxonal injury. While sNFL correlates with cerebrospinal fluid NFL (cNFL), 40-60% of variance remains unexplained. Assuming that for diseases of the central nervous system (CNS), such as multiple sclerosis (MS), the cNFL better reflects CNS injury, our goal was to develop and validate adjustment of sNFL for relevant confounders, to strengthen cNFL-sNFL correlations. Methods: We used 1,378 matched cNFL-sNFL pairs divided into training and validation cohort with matching data on 11 confounders, neuroexam, and magnetic resonance imaging (MRI). The effect of confounders on cNFL-sNFL relationship was tested using multiple linear regression (MLR) model. Propensity score matching was used to identify effect of spinal cord damage on sNFL levels. Results: In the training cohort (n=898) we correlated 11 confounders with the residuals from cNFL-sNFL linear regression. Four non-overlapping confounders explaining highest proportion of variance (12%: age, 8.7%: blood urea nitrogen, 3%: alkaline phosphatase, and 3.9%: weight) were used in MLR model. The model strengthened the cNFL-sNFL correlation from R2 = 0.52 to 0.64 in the independent validation cohort and strengthened correlation of adjusted sNFL with number of contrast-enhancing lesions (from R2 0.11 to 0.18). However, only sNFL, but not cNFL correlated with MS severity outcomes. Using propensity score matching, we demonstrated that subjects with proportionally higher sNFL to cNFL levels have significantly higher clinical and radiological evidence of spinal cord injury. Interpretation: Superiority of sNFL likely resides in the release of NFL from axons of lower motor or dorsal ganglia neurons directly to blood.