Clinical implications of serum neurofilament in newly diagnosed MS patients: A longitudinal multicentre cohort study

Abstract

Background: We aim to evaluate serum neurofilament light chain (sNfL), indicating neuroaxonal damage, as a biomarker at diagnosis in a large cohort of early multiple sclerosis (MS) patients.

Methods: In a multicentre prospective longitudinal observational cohort, patients with newly diagnosed relapsing-remitting MS (RRMS) or clinically isolated syndrome (CIS) were recruited between August 2010 and November 2015 in 22 centers. Clinical parameters, MRI, and sNfL levels (measured by single molecule array) were assessed at baseline and up to four-year follow-up.

Findings: Of 814 patients, 54.7% (445) were diagnosed with RRMS and 45.3% (369) with CIS when applying 2010 McDonald criteria (RRMS[2010] and CIS[2010]). After reclassification of CIS[2010] patients with existing CSF analysis, according to 2017 criteria, sNfL levels were lower in CIS[2017] than RRMS[2017] patients (9.1 pg/ml, IQR 6.2-13.7 pg/ml, n = 45; 10.8 pg/ml, IQR 7.4-20.1 pg/ml, n = 213; p = 0.036). sNfL levels correlated with number of T2 and Gd+ lesions at baseline and future clinical relapses. Patients receiving disease-modifying therapy (DMT) during the first four years had higher baseline sNfL levels than DMT-naïve patients (11.8 pg/ml, IQR 7.5-20.7 pg/ml, n = 726; 9.7 pg/ml, IQR 6.4-15.3 pg/ml, n = 88). Therapy escalation decisions within this period were reflected by longitudinal changes in sNfL levels.

Interpretation: Assessment of sNfL increases diagnostic accuracy, is associated with disease course prognosis and may, particularly when measured longitudinally, facilitate therapeutic decisions.

Funding: Supported the German Federal Ministry for Education and Research, the German Research Council, and Hertie-Stiftung.

Keywords: Biomarker; Multiple sclerosis; Neurofilament light chain; Prediction; sNfL.

Fig. 1

sNfL at time point of diagnosis correlates with baseline MRI parameters in a multicentre cohort and predicts clinical activity within the next two years. A) In a single-centre pilot study, patients with CIS according to 2010 McDonald criteria (n = 61; CIS[2010]) were reclassified based on the 2017 version of the McDonald criteria. sNfL levels were significantly higher in patients switching to RRMS (8.9 pg/ml, IQR 5.5-14.3 pg/ml, n = 30; RRMS[2017]) compared to patients remaining CIS (4.7 pg/ml, IQR 3.6-8.5 pg/ml, n = 31; CIS[2017]; p = 0.001 determined by Mann–Whitney-U test). B) Study design of a multicentre approach to assess reclassification of CIS[2010] patients according to 2017 McDonald criteria. C) 1543 MRIs from 809 patients were assessed for sNfL levels comparing MRIs with > 8 T2 lesions (baseline: n = 567; two-year follow-up: n = 573) and 1-8 T2 lesions (baseline: n = 242; two-year follow-up: n = 161). sNfL was higher in patients with > 8 than 1-8 T2 lesions at baseline (13.0 pg/ml, IQR 8.2-23.5 pg/ml; 8.6 pg/ml, IQR 6.1-12.9 pg/ml, p < 0.0005) and two-year follow-up (8.4 pg/ml, IQR 6.1-12.2 pg/ml; 7.0 pg/ml, IQR 5.7-9.0 pg/ml, p = 0.001). Both lesion groups showed a significant decrease of sNfL concentration after two years. D) Patients with Gd+ lesions at the time of serum sampling (n = 298; two-year follow-up: n = 94) had significantly higher sNfL levels than patients without Gd+ lesions (n = 487; two-year follow-up: n = 606). E) Baseline and two-year follow-up sNfL levels were negatively correlated with corresponding MSFC values (r = -0.170, p < 0.0005). F) sNfL levels at baseline were significantly higher in patients suffering at least one relapse up to FU2 (median: 12.2 pg/ml, IQR: 7.9-22.2 pg/ml, n = 337) than patients without relapse (median: 10.6 pg/ml, IQR: 7.9-22.2 pg/ml, n = 477, p = 0.015). G) Baseline sNfL levels correlated with the number of relapses between baseline and FU2 (r = 0.092, p = 0.008). H) A baseline risk score consisting of either presence of Gd+ lesions and T2 lesions (1-8 lesions or more than 8) alone (EDSS cutoff point = 1.94; MSFC cutoff point = 1.68) or in addition to sNfL (EDSS cutoff point = 15.48; MSFC cutoff point = 17.21) was determined for risk stratification at study initiation. The predictive power of the model for the two outcome parameters EDSS (YI = 1.54) and MSFC (YI = 0.3) and at FU2 was assessed by SVM algorithm. Correlation coefficients were increased (EDSS: 0.32 to 0.40, MSFC: 0.30 to 0.44) by the inclusion of baseline sNfL levels in the risk score computation. 10-fold cross validation was performed for the correlation and accuracy parameters. Detailed information on the model specifications are depicted in the table. Group differences were analyzed by mixed linear model procedure or Mann-Whitney-U test. Correlation analysis was performed by Spearman's rank correlation coefficient after exclusion of normally distributed data by Kolmogorov-Smirnov-Test and Shapiro-Wilk-Test. BL: baseline, CIS: clinically isolated syndrome FU2: two year follow-up, Gd: gadolinium enhancing lesions, OCB: oligoclonal bands, RRMS: relapsing remitting multiple sclerosis, T2: lesions in T2 weighted MRI scans, sNfL: serum neurofilament, MSFC: Multiple Sclerosis Functional Composite, YI: Youden's index, SVM: support vector machine. **p < 0.01, ***p < 0.001.

Fig. 2

Application of 2017 McDonald criteria selects patients with increased neuroaxonal damage. A) Flowchart of the study design. B) RRMS[2010] patients (n= 445) had significantly higher sNfL levels than CIS[2010] patients (12.5 pg/ml, IQR 8.0-22.8 pg/ml, n = 445; 10.1 pg/ml, IQR 6.9-17.1 pg/ml, n = 369, p < 0.0005). C) Application of 2017 McDonald criteria to CIS[2010] patients (n = 369) differentiated CIS[2017] (n = 45) from RRMS[2017] (n = 213) patients and resulted in significantly higher sNfL levels in patients reclassified to RRMS[2017] (10.8 pg/ml, IQR 7.4-20.1 pg/ml, n = 213) compared to patients remaining CIS[2017] (9.1 pg/ml, IQR 6.2-13.7 pg/ml, n = 45, p = 0.036 determined by Mann–Whitney-U test and 93.0% accuracy of discrimination computed by Bayesian analysis with caution of the imbalanced sample size). D) sNfL levels were significantly higher in RRMS[2010] patients with positive OCB (13.1 pg/ml, IQR 8.0-24.3 pg/ml, n = 222) compared to those without (10.1 pg/ml, IQR 6.7-15.8 pg/ml, n = 23, p = 0.035, discrimination accuracy = 96.4%). E) Significantly higher sNfL levels were found in RRMS[2010] patients with Gd+ lesions (16.4 pg/ml, IQR 9.3-31.3 pg/ml, n = 213) than in those without (10.2 pg/ml, IQR 7.2-16.1 pg/ml, n = 218, p < 0.0005). F) ROC curves generated for either positive OCB or Gd+ lesion alone (AUC = 0.76, CI 0.70 - 0.83) or in addition to the 90th percentile of sNfL (AUC = 0.84, CI 0.79 - 0.89). AUCs were compared using a Chi-square test (p = 0.035). G) Predictive analysis using SVM was performed yielding a prediction accuracy of 79% for OCBs and/or Gd+ for discriminating CIS and RRMS according to 2017 McDonald criteria. Accuracy increased to 84% by including the 90th percentile of sNfL in addition to the above two variables. Group differences were analyzed by Mann–Whitney-U test and in cases of imbalanced sample sizes additionally validated by Bayesian analysis (C+D). sNfL levels are reported as median. IQR: interquartile range, CIS: clinically isolated syndrome, DIS: dissemination in space, DIT: dissemination in time, Gd: gadolinium, OCB: oligoclonal bands, sNfL: serum neurofilament, RRMS: relapsing-remitting multiple sclerosis, SVM: support vector machine algorithm, ROC: receiver operating characteristic, AUC: area under the curve. *p < 0.05, **p < 0.01, ***p < 0.001, ns = not significant.

Fig. 3

Relationship between 2017 McDonald criteria, therapeutic decision and sNfL levels at two-year follow-up. A) Percentage of patients receiving DMT in CIS[2010], RRMS[2010], CIS[2010]→CIS[2017] and CIS[2010]→RRMS[2017] at two-year follow-up. B) sNfL levels at baseline were significantly higher in patients receiving DMT in the following two years (11.8 pg/ml, IQR 7.5-20.9 pg/ml, n = 727) than patients without specific MS medication until two-year follow-up (9.5 pg/ml, IQR 6.4-14.1 pg/ml, n = 87, p = 0.002). C) Patients were grouped into four classes based on the type of DMT they were receiving at year two follow-up. Median sNfL levels at baseline and two-year follow-up in four patient groups defined above (no DMT: n = 176, basic: n = 392, moderate: n = 134 and high: n = 134). Baseline sNfL levels were correlated with the patient's treatment group at two-year follow-up (r = 0.223, p < 0.0005). D) There was a statistically significant two-way interaction regarding the between- and within-subject factors (time*treatment groups) on sNfL concentration (p < 0.0005). At baseline the high treatment group showed elevated sNfL concentration compared to all other groups (p < 0.0005). We observed no significant difference between the treatment groups at two-year follow-up (ns) but a statistically significant effect of time on sNfL concentration for all four groups (p < 0.0005). E) Delta sNfL (two-year follow-up minus baseline) for patients without DMT and on basic, moderate and high DMT. Calculation was adjusted for baseline sNfL values. High treatment versus basic treatment (p = 0.001) and no DMT (p = 0.001) were significantly different. F) sNfL levels were compared between baseline (BL) and two-year follow-up (FU2). Percentage of patients with sNfL increase (FU2>BL) and sNfL decrease (FU2<BL) are depicted for four treatment groups. The group without treatment showed significantly higher proportion of patients with increased sNfL levels as compared to the groups with either moderate or high treatment. G) Percentage of patients under basic, moderate and high efficacy DMT in CIS[2010] and RRMS[2010] after two-year follow-up. H) Percentage of patients under basic, moderate and high efficacy DMT in CIS[2010] patients and in CIS[2017] and RRMS[2017] patients after reassessment according to 2017 criteria. Group differences were analyzed by Mann–Whitney-U test (B) or a one-way ANCOVA with baseline sNfL values as a covariate (E). For two-way (treatment group*time) interaction a two-way mixed ANOVA with four additional separate one-way ANOVAs with Tukey post-hoc test for multiple comparison for simple main effects/between-subject factors for the different treatment groups was applied (D). Correlation analysis was performed by Spearman's rank correlation coefficient after exclusion of normally distributed data by Kolmogorov–Smirnov-Test and Shapiro–Wilk-Test. Chi-square test of homogeneity was applied to investigate differences in proportions. Post hoc analysis involved pairwise comparison using multiple z-test of two proportions and, where appropriate, a Bonferroni correction. sNfL levels are reported as median. FU2: two-year follow-up, CIS: clinically isolated syndrome, RRMS: relapsing remitting multiple sclerosis, DMT: disease-modifying therapy, basic: interferons and glatirameracetate; moderate: teriflunomide and dimethylfumarate; high: natalizumab, rituximab, fingolimod, ocrelizumab, daclizumab, alemtuzumab, mitoxantrone. ns: not significant, *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 4

sNfL levels reflect therapeutic decisions within four years after diagnosis. A) Sankey diagram illustrating treatment stratification at two-year follow-up and therapy changes between year two and four. Width of arrows reflects quantitative changes between groups. B) High efficacy DMTs (“high” or “moderate”) were initiated more often in patients with higher sNfL levels both at baseline (13.2 pg/ml, IQR 8.0-24.2 pg/ml, n = 304; 9.8 pg/ml, IQR 6.7-15.4 pg/ml, n = 301; p < 0.0005) and two-year follow-up (8.2 pg/ml, IQR 6.0-11.8 pg/ml, n = 304; 7.6 pg/ml, IQR 5.9-11.0 pg/ml, n = 301; p = 0.007). There was a statistically significant two-way interaction on sNfL concentration (p < 0.0005). C) Top: baseline sNfL levels were correlated with the number of medication switches performed between baseline and two-year follow-up (r = 0.179, p < 0.0005). Below: FU2 sNfL levels were correlated with the number of medication switches performed between two-year and four-year follow-up (r = 0.133, p = 0.001). D) sNfL at two-year follow-up was significantly increased in patients with escalation of DMT treatment group between two-year and four-year follow-up compared to patients without escalation of DMT within this period (9.1 pg/ml, IQR 6.4-13.7 pg/ml, n = 107; 7.7 pg/ml, IQR 5.9-11.2 pg/ml, n = 358, p = 0.001). E) Patients with moderate or high DMT group at four-year follow-up were divided in two different groups depending on whether they have already been in these DMT classes at two-year follow up (“induction”: n = 153) or not (“escalation”: n = 82). A schematic illustration of this process is depicted in the left panel. Compared to the “escalation” group patients in the “induction” group showed higher sNfL levels at baseline (14.4 pg/ml, IQR 8.2-28.9 pg/ml; 10.9 pg/ml, IQR 7.5-19.1 pg/ml, p = 0.035) but lower sNfL concentrations at two-year follow-up (7.8 pg/ml, IQR 6.0-12.3 pg/ml; 9.3 pg/ml, IQR 6.3-14.3 pg/ml, p = 0.025). There was no significant difference at four-year follow-up (“induction”: 6.2 pg/ml, IQR 4.8-8.7 pg/ml; “escalation”: 5.7 pg/ml, IQR 4.4-7.9 pg/ml, p = 0.885). Correlation analysis was performed by Spearman's rank correlation coefficient after exclusion of normally distributed data by Kolmogorov-Smirnov-Test and Shapiro–Wilk-Test. Where appropriate, post hoc analysis was performed using a Bonferroni correction. sNfL levels are reported as median and interquartile range (IQR). FU2: two-year follow-up, FU4: four-year follow-up, DMT: disease-modifying therapy, basic: interferons and glatirameracetate; moderate: teriflunomide and dimethylfumarate; high: natalizumab, rituximab, fingolimod, ocrelizumab, daclizumab, alemtuzumab, mitoxantrone. *p < 0.05, **p < 0.01, ***p < 0.001.