Increased Neurofilament Light Chain Blood Levels in Neurodegenerative Neurological Diseases

Objective Neuronal damage is the morphological substrate of persisting neurological disability. Neurofilaments (Nf) are cytoskeletal proteins of neurons and their release into cerebrospinal fluid has shown encouraging results as a biomarker for neurodegeneration. This study aimed to validate the quantification of the Nf light chain (NfL) in blood samples, as a biofluid source easily accessible for longitudinal studies. Methods We developed and applied a highly sensitive electrochemiluminescence (ECL) based immunoassay for quantification of NfL in blood and CSF. Results Patients with Alzheimer’s disease (AD) (30.8 pg/ml, n=20), Guillain-Barré-syndrome (GBS) (79.4 pg/ml, n=19) or amyotrophic lateral sclerosis (ALS) (95.4 pg/ml, n=46) had higher serum NfL values than a control group of neurological patients without evidence of structural CNS damage (control patients, CP) (4.4 pg/ml, n=68, p<0.0001 for each comparison, p=0.002 for AD patients) and healthy controls (HC) (3.3 pg/ml, n=67, p<0.0001). Similar differences were seen in corresponding CSF samples. CSF and serum levels correlated in AD (r=0.48, p=0.033), GBS (r=0.79, p<0.0001) and ALS (r=0.70, p<0.0001), but not in CP (r=0.11, p=0.3739). The sensitivity and specificity of serum NfL for separating ALS from healthy controls was 91.3% and 91.0%. Conclusions We developed and validated a novel ECL based sandwich immunoassay for the NfL protein in serum (NfLUmea47:3); levels in ALS were more than 20-fold higher than in controls. Our data supports further longitudinal studies of serum NfL in neurodegenerative diseases as a potential biomarker of on-going disease progression, and as a potential surrogate to quantify effects of neuroprotective drugs in clinical trials.


Introduction
Neurofilaments (Nf) are highly specific major structural proteins of neurons, consisting predominantly of four subunits: Nf light (NfL), Nf medium (NfM) and Nf heavy (NfH) chain and alpha-internexin [1]. Nf are released in significant quantity following axonal damage or neuronal degeneration. Disruption to the axonal membrane releases Nf into the interstitial fluid and eventually into cerebrospinal fluid (CSF) and blood. Therefore, blood Nf levels could be useful for both predicting and monitoring disease progression and for assessing the efficacy and/or toxicity of future neuroprotective treatment strategies.
Several previous studies have demonstrated the presence of NfH and NfL in CSF, which has been assumed to reflect brain pathology more accurately than the peripheral blood compartment [2][3][4][5][6][7][8][9][10][11][12]. However, obtaining longitudinal CSF samples is considered too invasive outside the clinical trial arena, precluding the broader clinical use of Nf. In contrast to CSF, serial blood samples can readily be collected, hence reliable quantification of NfL in blood would be a major stride towards a biomarker of the course of neurodegeneration. Several reports have suggested peripheral blood levels of NfH as a potential marker of neurodegeneration [13][14][15][16][17][18][19][20][21][22]. In contrast to this, there is only one recent study investigating serum NfL; this paper examined the relationship between serum NfL and neurological outcome following cardiac arrest [23].
A commercially available ELISA (UmanDiagnostics NF-light ® assay) uses two highly specific, non-competing monoclonal antibodies (47:3 and 2:1) to quantify soluble NfL in CSF samples but it cannot in its present form be used for analysis of blood samples [24].
We recently compared our highly sensitive electrochemiluminescence (ECL)-based solid-phase sandwich immunoassay for NfH SMI35 (adhering to a previously proposed nomenclature, the soluble fraction of NfH measured is indicated with the capture antibody in the superscript [8]) with NfL, determined by the NF-light ® assay in CSF samples [6]. Importantly, the conventional ELISA showed higher sensitivity compared with the ECL-NfH SMI35 immunoassay [25].
The aim of this study was to develop and validate (both analytically and clinically) a sensitive ECL-based NfL assay suitable for the quantification of NfL in serum at concentrations relevant to clinical settings.

Patients and Control persons and ethics statement
Written informed consent was obtained from all patients in accordance with the Declaration of Helsinki, and the study was approved by the Common Institutional Review Board of the Cantons of Basel. Paired CSF and serum samples were collected during routine diagnostic investigations as indicated by the treating physicians.
Samples were collected and processed at room temperature within two hours. Serum samples were spun at 2,000 g, CSF samples at 400 g at room temperature for 10 minutes, aliquoted in polypropylene tubes and stored at -80°C.
Serum samples from 67 healthy control subjects (HC) were included in the study. For ethical reasons CSF samples were not available from these subjects. The group of control patients (CP) (n=68) consisted of patients who, based on extensive diagnostic evaluation had no objective clinical, structural (cranial magnetic resonance imaging, MRI), laboratory (CSF analysis) or functional (electroencephalography, EEG) deficit. These patients suffered from tension type headache (n=21), lower back pain (n=7), psychiatric disorders (n=26) or miscellaneous non-specific symptoms for which no neurological explanation could be found (n=14). From two of these patients there was not enough CSF left for further analysis. In addition, 49 patients with probable or definite ALS (for three no serum and for one no CSF sample was available) [26], probable Alzheimer's disease (AD) [27], or a Guillain-Barré syndrome (GBS) (for one no serum sample was available) (n=20 each) were included ( Table 1). The analyst was blinded to all clinical data.

Analytical procedure
The 96-well plates (Multi-Array ® plates, Meso Scale Discovery, Gaithersburg, MD) include integrated screen-printed carbon ink electrodes on the bottom of the wells. Coating was done overnight with 30 µl of capture antibody (mAB 47:3,1.25 µg/ml) diluted in PBS (pH 7.4) at 4°C. All following incubation steps were done on a plate shaker (800 rpm) and were preceded by three wash steps with 200 µl of TBS, containing 0.1% Tween 20 (pH 7.5) per well. Non-specific binding sites were blocked with 100 µl of TBS, containing 3% BSA, per well for 1h. After washing, 25 µl of TBS containing 1% BSA and 0.1% Tween 20 was added as sample diluent to each well. 25 µl of standard, control or serum/CSF sample was then added in duplicate and the plate incubated at room temperature (RT) for 2h. After washing, 25 µl of the secondary antibody (mAB 2:1, 0.5 µg/ml) diluted in TBS containing 1% BSA and 0.1% Tween 20 was added to each well and the plate incubated for 1 h at RT. After washing, MSD SULFO-TAG TM labelled streptavidin (0.25 µg/ml), diluted in TBS containing 1% BSA and 0.1% Tween 20, was added and incubated for 1h at RT. Following a final wash, 150 µl of ECL read buffer (MSD) diluted 1:2 with distilled water was added and the ECL signal, detected by photodetectors, measured using the MSD Sector Imager 2400 plate reader. A four-parameter weighted logistic fit curve was generated, sample concentrations extrapolated and analysed using the Discovery Workbench 3.0 software (MSD). If required, samples were appropriately diluted to fall in the range of the standard curve. Non measurable NfL samples were reported as 0 pg/ml.

Statistical analysis
Continuous variables were described by their median and interquartile range (IQR), and categorical variables by numbers and percentages. Comparison of demographic data was performed using the Kruskal-Wallis test, and pairwise post-hoc comparisons using Dunn's post-test or chi-square test as appropriate. Serum and CSF levels of NfL were logtransformed to achieve a normal distribution for subsequent analysis. To control for age as a potential confounding factor, an analysis of covariance with age as covariate and disease group as fixed factor, was performed [7]. Group-specific levels of NfL were expressed as geometric means with 95%confidence intervals. For log-normal variables, the geometric mean equals the median. Correlations were computed by determining the Spearman rank correlation coefficient (r). The cut-off (upper reference range of normal) providing optimal sensitivity and specificity in distinguishing ALS from HC by serum NfL was defined by receiver operating characteristic (ROC) curve analysis. Proportions above and below this cut-off were compared with the Chi-Square test. A two-sided p-value < 0.05 was considered as significant. P-values of post-hoc comparisons were adjusted using a Bonferroni correction. All statistical analyses and graphs were performed using SPSS (Version 15.0 SPSS, Chicago, IL) and Graph Pad Prism 5.02 for Windows (GraphPad Software, San Diego, CA).
Recovery rates were tested in 6 serum samples from healthy volunteers. Recovery of NfL (serum spiked with 50 pg/ml of HPLC purified bovine NfL) was 72% and 114%. For serum spiked with 100 pg/ml of NfL it was 81% and 96%, and for 1,000 pg/ml of NfL recovery was 82% and 116%.

Analytical sensitivity and stability of the analyte
Sensitivity (lowest standard above blank) was calculated as blank signal plus three standard deviations (SD) from 32 assays. The mean blank signal was 138 counts (SD 20.9 counts). The mean signal of the lowest standard (15.6 pg/ml) was 184.5 counts (SD 23.2): accordingly analytical sensitivity was defined to be 15.6 pg/ml. We tested the stability of NfL at room temperature (RT), 4 °C and compared this to samples stored at -80 °C. Four aliquoted serum samples were frozen at -80 °C. The aliquots were thawed on days 0, 3 hours before measurement, days 1, 4 and 8 and stored at RT or 4 °C until analysis. The measured signals were normalised to the signal of the day 0. There was no significant change in signal in samples stored at RT and at 4 °C (RT: day 8: 1.06 ± 0.08 (mean ± SD), p = 0.4063 and 4 °C: day 8: 1.01 ± 0.09, p = 0.1721). Four serum samples were analysed for stability during freeze-thawing cycles. The samples underwent 1, 2, 3, 4 or 5 freeze-thawing cycles and the signal was normalised to the sample freeze-thawed once, without any relevant effect of freeze-thawing on the measured signals (5 freeze-thawing cycles: 1.03 ± 0.03, p = 0.5076).

Parallelism
We studied parallelism between standards and samples by reciprocal dilutions of three serum samples and three standard curves. The obtained signals were normalised to the highest value within this series (100%). The parallel relationship is demonstrated in Figure 2, suggesting the absence of aggregate formation or endogenous binding between NfL and other blood substrates [28] (Figure 2).

Discussion
A highly sensitive method for the detection of a clinically relevant biomarker of neurodegeneration has been developed. Importantly, our method allows us to make use of readily available longitudinal patient blood samples, instead of being restricted to ethically difficult to obtain CSF samples. One potential clinical application for serum NfL levels is demonstrated by the diagnostic sensitivity of 91.3% for ALS, a rapidly progressive neurodegenerative disease [29,30].
We present the first ECL based solid phase immunoassay for the NfL protein in blood based on two non-competitive, monoclonal antibodies. These antibodies have been widely used and validated in a commercial ELISA for CSF measurements of NfL (NF-light ® assay) [5,24,31]. NfL is considered to represent the most abundant and also most soluble Nf subunit [1].
The optimised ECL-NfL assay protocol proved to be highly accurate (intra-assay CV < 6%, inter-assay CV < 24%), sensitive (sensitivity 15.6 pg/ml) and demonstrated linearity and parallelism (Figures 1 and 2) over a wide analytical range (15.6-10,000 pg/ml). In addition we found NfL Umea47:3 to be stable in serum [25]. This is relevant for a potential value to monitor drug effects by serum NfL in ALS where Nf aggregate formation is a key pathological finding [28]. In contrast to NfH SMI34 and NfH SMI35 , no such aggregates were found for NfL Umea47:3 , essentially overcoming the limitations of the Nf "hook effect" (matrix effect) [28]. In this context a more than 20fold elevation of serum NfL Umea47:3 levels in ALS compared to HC cannot be overestimated. Interestingly, the fold-differences between disease groups and CP for serum NfL Umea47:3 was higher compared to the respective CSF levels (serum/CSF: An important and unresolved question is whether or not there is a relevant correlation between Nf levels and age. If present, such a relationship would require age dependent cut-off values [7]. A major limitation to all studies in this field to date [6,7,11,[32][33][34] is that they have not been powered to investigate this potential correlation in the CSF, due to lack of samples from a sufficiently large healthy control group across all age categories. Again, the availability of the present method to investigate this in readily available serum samples is highly relevant. Importantly, we did not find a correlation between serum NfL Umea47:3 levels and age in either HC or CP. Whether or not a possible relationship with age exists for ALS, GBS or AD is questionable, as older patients are often more severely affected and higher age is the most important prognostic factor  in either condition and therefore not independent of the neurodegeneration related release of NfL Umea47:3 . The absence of the Nf hook-effect is an important analytical advantage for quantification of the ECL based serum NfL Umea47:3 assay compared to the serum NfH SMI34 and NfH SMI35 ELISA, as there is no necessity for a time-consuming pre-incubation step with urea [28]. Given the important prognostic information that NfH levels provide on a number of clinical conditions, we anticipate NfL Umea47:3 to be relevant for future studies. Serum NfL Umea47:3 bears the potential for predicting disease progression in ALS [15,35,36] and MS [17,18], detecting particularly disabling acute episodes of optic neuritis or relapses in MS [16], identifying primary and secondary brain damage in stroke [22,37], SAH [13], TBI [19,38] and in the emerging concept of chronic traumatic encephalopathy (CTE) [20,38]. Like serum NfH SMI35 , serum NfL Umea47:3 may also be exploited as a safety biomarker for recognising neurotoxicity [21]. There is already data that serum NfL levels are of comparable prognostic value to NfH SMI35 levels following cardiac arrest [19,23]. Of note there were no controls and no analytical validation data from the NfL assay used in one study [23].
Similar to our previous findings for NfH SMI35 in CSF, a bimodal distribution of serum NfL levels was seen in patients with GBS [6]. There are no previous studies on Nf in blood from patients with GBS. We have earlier shown that CSF levels of NfH are higher in patients with evidence of axonal damage compared to those with purely demyelinating GBS, with CSF NfH levels predictive of outcome [9,39]. Future prospective studies incorporating detailed longitudinal clinical and electrophysiological assessments, and sampling are clearly warranted. These studies will also shed light on the role of proximal versus more distal axonotmesis and secondary axonal peripheral degeneration and the relationship of increased blood NfL levels [40].
Blood levels of Nf have similarly not been investigated in patients with dementia. In our study the differences in serum and CSF NfL levels in AD compared to HC and CP (p<0.0001 and p=0.002) lost significance after age and Bonferroni correction. This is in line with previous investigations where CSF NfH SMI35 levels were increased, but diagnostic sensitivity, and hence potential for clinical use of NfH SMI35 was not superior to that of the benchmark biomarkers total tau, phospho tau, or amyloid beta 1-42 [41,42]. To explore these questions further we are currently expanding our database in a larger and well characterised cohort of AD and control patients.
In summary, we developed and validated a sensitive and reliable assay for measurements of NfL in human blood samples. For the first time, we were able to demonstrate increased blood NfL levels in patients with ALS and GBS. These differences were more pronounced for the ECL-NfL Umea 47:3 assay than those reported in ALS for NfH in previous reports [15,36]. Our data support further studies of serum NfL in well-defined longitudinal cohorts of neurodegenerative diseases. These studies will show if serum NfL measurements can be used as a biomarker for disease progression and as an outcome measure in clinical trials.