Enhancing the clinical value of serum neurofilament light chain measurement

Abstract

BACKGROUNDSerum neurofilament light chain (sNFL) is becoming an important biomarker of neuro-axonal injury. Though sNFL correlates with CSF NFL (cNFL), 40% to 60% of variance remains unexplained. We aimed to mathematically adjust sNFL to strengthen its clinical value.METHODSWe measured NFL in a blinded fashion in 1138 matched CSF and serum samples from 571 patients. Multiple linear regression (MLR) models constructed in the training cohort were validated in an independent cohort.RESULTSAn MLR model that included age, blood urea nitrogen, alkaline phosphatase, creatinine, and weight improved correlations of cNFL with sNFL (from R2 = 0.57 to 0.67). Covariate adjustment significantly improved the correlation of sNFL with the number of contrast-enhancing lesions (from R2 = 0.18 to 0.28; 36% improvement) in the validation cohort of patients with multiple sclerosis (MS). Unexpectedly, only sNFL, but not cNFL, weakly but significantly correlated with cross-sectional MS severity outcomes. Investigating 2 nonoverlapping hypotheses, we showed that patients with proportionally higher sNFL to cNFL had higher clinical and radiological evidence of spinal cord (SC) injury and probably released NFL from peripheral axons into blood, bypassing the CSF.CONCLUSIONsNFL captures 2 sources of axonal injury, central and peripheral, the latter reflecting SC damage, which primarily drives disability progression in MS.TRIAL REGISTRATIONClinicalTrials.gov NCT00794352.FUNDINGDivision of Intramural Research, National Institute of Allergy and Infectious Diseases, NIH (AI001242 and AI001243).

Keywords: Multiple sclerosis; Neurodegeneration; Neuroscience.

Figures

Figure 1. Variance between sNFL and cNFL concentrations.

(A) Linear regression model between log10-transformed concentration (pg/mL) of sNFL and cNFL in the training cohort of samples where cNFL levels explain 57% of variance of sNFL levels. (B) Remaining 43% of variance shown as NFL residuals generated as differences between measured sNFL concentration and predicted sNFL concentration calculated from measured cNFL using linear regression model. (C) Eleven potential confounders related to distribution volume (BMI = body mass index, Est Blood Vol = estimated blood volume, height, and weight), protein metabolism/clearance (ALT = alanine transaminase, AP = alkaline phosphatase, AST = aspartate transaminase, BUN = blood urea nitrogen, creatinine, and eGFR = estimated glomerular filtration rate), and age were used as explanatory variables in a multiple linear regression model resulting in varied importance represented as a t statistic of each variable in the model (D). Stepwise regression resulted in retention of 5 confounders in the model (E) that showed increased correlation between measured and predicted sNFL levels both in the training (G) and in the validation (I) cohort in comparison with correlations between measured and predicted values using a simple linear regression model in the same training (F) and validation cohort (H). Confounders in color are the ones selected in the multiple linear regression model that underwent stepwise regression. Green line represents linear regression model with gray shading corresponding to 95% confidence interval. ns, number of samples measured; np, number of patients represented by the samples; CCC, concordance correlation coefficient.

Figure 2. Adjustment for 5 confounders improves correlation of sNFL with number of MRI CELs and eliminates noise.

(A) CELs have been used as a surrogate outcome of blood-brain barrier opening and active inflammation in the brains of patients with MS. Logistic regression that predicts probability of CEL presence/absence and linear regression between NFL and total number of CELs have been tested. A binomial regression classifier was generated to predict dichotomous outcome of present/absent CEL. The area under the curve (AUC), sensitivity, and specificity have been calculated for classifiers using measured cNFL (B and E), measured sNFL (C and F), and sNFL-predicted cNFL (D and G) to predict probability of presence of CELs. Dotted line represents the best probability cutoff value determined in the training cohort with corresponding NFL concentration displayed above the line. Horizontal lines represent medians. Two-sided Wilcoxon 2-sample test evaluated the significance of differences between 2 groups of patients. The linear model between number of CELs (y axes, transformed as natural logarithm of [CEL+1]) and NFL (x axes) shows higher predictive power of cNFL in both training (H) and validation (K) cohorts, compared with sNFL in training (I) and validation (L) cohorts. Adjustment of sNFL for 5 confounders (age, weight, AP, BUN, and creatinine) improved the correlation with number of CELs in both training (J) and validation (M) cohorts compared with measured sNFL. Purple line represents linear regression model with gray shading corresponding to 95% confidence interval.

Figure 3. sNFL correlates better with MS disease severity outcomes than cNFL.

(A) Disease severity in MS is a measure of how fast patients accumulate disability. Slow accumulation of disability over time results in low MS severity (green); fast accumulation of disability results in high MS severity (red). Because it is difficult to measure rates of disability progression prospectively and longitudinally, MS severity outcomes are collected cross-sectionally, measuring past rates of disability progression by normalizing disability to the patient’s age (Age-Related Multiple Sclerosis Severity Score [ARMSS] and Multiple Sclerosis Disease Severity Scale [MS-DSS]) or disease duration (MSSS). Correlation analysis of 3 MS severity outcomes, MS-DSS, MSSS, and ARMSS, with 3 NFL values, measured cNFL, measured sNFL, and sNFL-predicted cNFL, in 2 independent cohorts: training cohort (B) and validation cohort (C). Purple line represents linear regression model with gray shading corresponding to 95% confidence interval. Difference in number of patients/samples used for these analyses is because of exclusion of samples due to missing respective MS severity data.

Figure 4. Two hypotheses explaining superiority of sNFL in predicting MS severity.

(A) Hypothesis 1: Dilution of cNFL due to brain atrophy while sNFL concentration remains unaffected. Brain atrophy was evaluated by brain parenchymal fraction (BPFr) and by semiquantitative measure of brain atrophy (none, mild, moderate, and severe). (B) NFL residuals that fall within IQR (gray) were removed, resulting in a subset of samples with proportionally higher (above the third quartile [teal]) and lower cNFL (below the first quartile [salmon]), with comparable sNFL levels. (C) Paired Wilcoxon rank sum test showed marginally significant difference in BPFr (top left) and total brain atrophy (bottom left) between samples with different cNFL levels in the training cohort. These observations were not confirmed in the validation cohort (top and bottom right). (D) Hypothesis 2: Increase of sNFL due to spinal cord (SC) damage. NFL from damaged peripheral nerves and SC roots is released directly into blood, increasing sNFL concentration while cNFL remains unchanged. SC damage was evaluated using a semiquantitative MRI outcome (a sum of lesion load and atrophy at the level of medulla and cervical spine) and by clinical outcome capturing damage of lower motor neurons (sum of muscle atrophy scores) and damage to peripheral/autonomous nervous system (score for bowel, bladder, sexual, and autonomic dysfunctions) generated from neurological exams digitalized using the NeurEx app. (E) NFL residuals that fall within IQR (gray) were removed, resulting in a subset of samples with proportionally higher sNFL (above the third quartile [teal]) and lower sNFL (below the first quartile [salmon]), with comparable cNFL levels. (F) Paired Wilcoxon rank sum test showed a statistically significant difference in MRI (top left) and clinical (bottom left) outcomes between samples with different sNFL levels in the training cohort; the observed differences were confirmed in the validation cohort (top and bottom right). The box plots depict the minimum and maximum values (whiskers), the upper and lower quartiles, and the median. The length of the box represents the IQR.