Neurofilament levels in patients with neurological diseases: A comparison of neurofilament light and heavy chain levels

Abstract Background Neurofilaments are the major cytoskeletal components of neurons, and cell injury leads to their release into the surrounding area. The aim of this study was to compare the cerebrospinal fluid (CSF) and serum (S) concentrations of neurofilament light chains (NFLs) and phosphorylated neurofilament heavy chains (pNFHs). Methods Neurofilament concentrations were measured in CSF and S samples from 172 patients using three enzyme‐linked immunosorbent assays. Excel, Stata version 13, MedCal version 17.9.7., and NCSS 2007 software were used for the statistical analysis. Results There was a statistically significant correlation between the concentrations of CSF NFL and CSF pNFH (r s = 0.748; n = 89; P < 0.001), but Passing‐Bablok regression showed systematic deviation between the values obtained using the two assays. This indicates that the assays were not interchangeable. CSF pNFH and S pNFH concentrations showed low correlation. The kappa statistic showed moderate conformity between CSF pNFH and CSF NFL concentrations (κ = 0.556). Conclusions The CSF NFL and CSF pNFH assays gave clinically consistent results that reflected the degree of axonal damage, independent of any particular neurological diagnosis. The S pNFH assays had a lower predictive value due to the low correlation coefficient and the kappa index of the CSF pNFH method.

mass of 112.5 kDa, consists of 1020 amino acids. NFH is important for protein-protein interactions, which are regulated locally in the axon by phosphorylation. 1,2 The alpha-internexin protein has a molecular mass of 66 kDa and can form homopolymers; however, due to its instability, this subunit is difficult to detect in laboratory practice. Its gene is on chromosome 10q24.33.
Enzyme-linked immunosorbent assays (ELISAs) or more sensitive techniques, such as electrochemiluminescence immunoassays and single molecule arrays (SIMOAs), can be used to determine NF levels. 3 After axonal injury, NFs are released into the extracellular space.
Accordingly, their concentration in CSF and/or S reflects the degree of axonal damage. 4 The levels of both NFL and NFH are increased in multiple sclerosis (MS), reflecting both neuroaxonal damage in active plaques, which is mediated by the inflammation, and neurodegeneration. 5 In patients with clinically isolated syndrome (CIS), the NFL levels correlate with radiological signs of disease activity (gadolinium-enhancing magnetic resonance lesions) and predict conversion to clinically definite MS with a worse prognosis. [6][7][8] During MS progression, NFH levels correlate with physical disability and changes in brain volume but not with lesion number or volume. The NFH concentration may indicate ongoing neurodegeneration. 5,7,9 Natalizumab-treated patients show a 3-fold decrease in NFL, indicating that this treatment not only has an immunomodulatory effect but may also reduce axonal damage. 10 These effects are also observed in patients with MS who were treated with rituximab, mitoxantrone, or fingolimod. 11,12 However, studies have not demonstrated conclusively that the decline in axonal involvement is not secondary, and anti-NF antibody levels do not correlate with the clinical variants of MS. 13,14 The aims of the study were to compare the cerebrospinal fluid (CSF) concentrations of NFL and pNFH and the CSF and S concentrations of pNFH and to evaluate the correlation of these parameters with the following diagnoses: MS; CIS; inflammatory diseases of the peripheral nervous system (IDPNS); and other inflammatory central nervous system diseases (OIND), noninflammatory neurological diseases (NIND), and no evidence of organic nervous system disease (the control group, Control).

| Samples
Neurofilament light, pNFH, and albumin concentrations were determined in CSF samples that were collected into a polypropylene tube (Sarstedt) and in S samples that were collected into a Serum Gel with Clotting Activator tube (Sarstedt). S and CSF samples were drawn on the same day. The CSF samples were centrifuged at 390 × g for 10 minutes at room temperature, and the S samples were centrifuged at 2500 × g for 6 minutes at 4°C. Both the CSF and S samples were aliquoted into at least three vials (0.3 mL per vial) and stored at −70°C until the analysis. ducible measurement of NFL, as the diagnostic kit did not include a quality control sample; for measuring pNFH, the manufacturer of the diagnostic kit supplied two quality control samples. The kit manufacturers stated that the analytical sensitivity for NFL was 32 ng·L −1 , 27 ng·L −1 for pNFH and 6 ng·L −1 for pNFH sensitive (pNFHs). All samples were measured in duplicate, and the mean intra-assay coefficients of variation for CSF NFL, CSF pNFH, and S pNFH were 1.9%, 3.3%, and 4.2%, respectively.

| Statistical methods
Excel, Stata version 13, MedCal version 17.9.7., and NCSS 2007 were used for the statistical analyses. 16,17 Basic descriptive statistics were used to describe the data, including frequency tables, medians, arithmetic means, standard deviations, and percentiles. The normality of the CSF NFL, CSF pNFH, and S pNFH parameters was verified with the Shapiro-Wilk test of normality. The normality hypothesis was rejected; therefore, nonparametric tests were used, including the Kruskal-Wallis rank test and the two-sample Wilcoxon rank-sum (Mann-Whitney) test. The relationship between the parameters was evaluated by Spearman's correlation coefficient. Data values were categorized as positive and negative. Fisher's exact test was used to test categorized data. Conformity between assay results was evaluated by the kappa index with 95% confidence intervals. Statistical tests were evaluated using a 5% significance level.

| Ethics approval
Informed consent was obtained from all patients at the University Hospital Ostrava who were included in the study. The study was approved by the Ethics Committee of the University Hospital Ostrava as a part of the project "CSF biomarkers of multiple sclerosis" (reference number 400/2017).

| RE SULTS
First, we partially verified diagnostic kits for NFL and pNFH determination. When we evaluated whether the measurements were precise and reproducible, both diagnostic kits showed variation coefficients that were comparable to the values supplied by the manufacturer (Table 1). A total of 172 patient samples were included in the analysis that evaluated the correlation between NF levels and clinical diagnoses.
The analytical characteristics of the studied group are presented in Table 2.
There was a statistically significant correlation between CSF NFL and CSF pNFH concentrations (r s = 0.748; n = 89; P < 0.001).
The regression relationship between these parameters was evaluated using Passing-Bablok regression (Figure 1) NF concentrations according to the diagnosis groups are presented in Figure 2. We evaluated the correlation between the NF concentrations and the different diagnoses. There was a statistically significant relationship between CSF NFL and CSF pNFH in the NIND (r s = 0.793; P < 0.001) and control (r s = 0.811; P < 0.001) diagnosis groups, and between CSF pNFH and S pNFH in the IDPNS (r s = 0.900; P = 0.037) and NIND diagnosis groups (r s = 0.459; P = 0.018) and between CSF pNFH and S pNFHs in the NIND diagnosis groups (r s = 0.435; P = 0.030) ( Table 3).
The correlation coefficient between the CSF pNFH and S pNFH values and between the CSF pNFH and S pNFHs was moderate (r s = 0.579 resp. 0.439), probably due to the differences in the biological material that was analyzed (CSF or S; Figure 3).
The kappa statistic was used to compare the assays based on clinical interpretation because the methods had different reference intervals ( and other neurological diseases. 25 We studied the correlation of CSF NFL with clinical data as well.
The CSF NFL concentration in the MS and CIS diagnosis groups was significantly higher in the subgroup of patients who had EDSS scores of 2.5 or higher 6 months after sampling versus the subgroup of patients with EDSS scores of 0 to 2. These data showed the suitability of using CSF NFL to predict disease severity. Similar results were obtained in the study by Disanto et al, 26

| CON CLUS ION
In this study, we tested three diagnostic kits for the determination of NF concentrations in biological fluids. The NFL ELISA assay had lower sensitivity and was suitable only for CSF analysis, while the pNFH ELISA assay had satisfactory sensitivity and was suitable for S and CSF analysis, the pNFHs only for S analysis. The data showed good correlation and moderate conformity between CSF NFL and CSF pNFH concentrations, indicating that the results can be considered to be consistent. However, the low correlation coefficient and the kappa index found between the S pNFH, even if using a highsensitivity ELISA assay and CSF pNFH meant that the S pNFH and S pNFHs assays gave a lower predictive value. When assessing the relationship of NF concentrations and diagnosis, correlations were found between the concentration of CSF NFL and CSF pNFH in the NIND diagnosis group and in the control group of patients, between the CSF and S pNFH in the IDPNS and NIND diagnosis groups, and between the CSF and S pNFHs in the NIND diagnosis groups. The results confirmed that NFs, whether NFLs or pNFHs, represent an etiologically nonspecific indicator of tissue damage and that it is better to determine their levels in CSF than in S.