Cross-Sectional and Longitudinal Validation of Serum Neurofilament Light Chain (NfL) as a Biomarker of Parkinson’s Disease Progression

Background Alterations in neurofilament light chain (NfL), reflecting axonal damage, have been proposed as a biomarker for neurological disorders including Parkinson’s disease (PD). Methods We measured NfL concentrations by immunoassay in (1) a set of longitudinal CSF samples from 82 PD, 14 other neurodegenerative disorders (OND), and 53 healthy controls (HC); (2) A cross-sectional cohort with paired CSF and serum samples from subjects with 151 PD, 344 OND, and 20 HC, and (3a) a large longitudinal validation cohort with serum samples from 375 PD, 178 HC, and 57 prodromal Lewy Body disorder with hyposmia or isolated REM sleep behaviour disorder (iRBD), (3b) 216 symptomatic and 298 asymptomatic LRRK2, GBA, and SNCA mutation carriers in the Parkinson’s Progression Markers Initiative (PPMI). Mean differences in serum NfL levels between diagnostic groups, and correlation with motor signs, cognitive measures and dopamine transporter imaging, were assessed. Linear mixed effects models were used to assess longitudinal changes in log10 (NfL) in relation to demographic, clinical, and imaging variables. Findings In the longitudinal cohort (1) NfL in CSF increased over time and was significantly positively correlated with MDS-UPDRS III motor and total scores in the PD group (p=0·0231 and 0·0081). In the cross-sectional cohort (2) the paired CSF and serum NfL samples were highly correlated (Spearman’s rank ). In the large validation cohort (3a) mean baseline serum NfL was higher in PD (13 ±7·2pg/ml), hyposmics (15±15·1pg/ml), and iRBD (17±8pg/ml) compared with HC (12±6·7pg/ml) but was highest in the seven OND cases (18±7pg/ml). In the genetic cohort (3b), serum NfL levels were lower in asymptomatic than in symptomatic mutation carriers. There was a significant (age-adjusted) longitudinal increase in serum NfL in PD compared with HC. Serum NfL values were significantly positively associated with longitudinal MDS-UPDRS motor and total scores, as well as age. Interpretation We identified NfL as the first blood-based PD progression biomarker. NfL levels in serum samples are increased in PD compared to HC, increase significantly over time in PD, and correlate with a clinical measure of disease severity. Although the specificity of NfL in PD is low and additional, more specific biomarkers are needed, serum NfL is the first blood-based biomarker candidate that could support disease stratification (PD vs. OND), track clinical progression, and might be used to assess responsiveness to neuroprotective treatments.

Interpretation: We identified NfL as the first blood-based PD progression biomarker. NfL levels in serum samples are increased in PD compared to HC, increase significantly over time in PD, and correlate with a clinical measure of disease severity. Although the specificity of NfL in PD is low and additional, more specific biomarkers are needed, serum NfL is the first blood-based biomarker candidate that could support disease stratification (PD vs. OND), track clinical progression, and might be used to assess responsiveness to neuroprotective treatments.

Introduction
Two major obstacles hamper the success of translational Parkinson's disease (PD) research: 1) currently there is no longitudinal fluid biomarker for PD that correlates with clinical disease progression; 2) a definite diagnosis of PD can currently only be made by autopsy and the rate of clinical misdiagnoses, especially in the early stages of the disease, is reported to be high 1, 2 . We and others identified a 10-15% decrease in cerebrospinal fluid (CSF) α -synuclein in several related disorders including PD, multiple system atrophy (MSA), and dementia with Lewy bodies (DLB). 3,4 CSF α -synuclein values show substantial overlap, no significant longitudinal change during 36 months follow-up, and no correlation with the progression of clinical signs and symptoms 5 , limiting its clinical utility as a standalone biomarker. Therefore, additional biomarkers beyond α -synuclein are warranted that could be used for diagnosis and as progression markers in PD.
Neurofilaments are highly phosphorylated neuronal cytoskeleton components composed of three subunits of differing molecular weight maintaining neuronal structure and determining axonal calibre. The 68 kDa neurofilament light chain (NfL) is essential for the assembly of the complex as it forms its backbone and is released into extracellular fluids in response to axonal damage. 6,7 Several studies have shown elevated levels of NfL in the CSF of patients suffering from neurodegenerative conditions. [7][8][9] Further, CSF NfL levels seem to reflect progression of various neurological conditions, including multiple sclerosis and neurodegenerative dementia disorders. 10,11 Ultrasensitive methods have enabled the quantification of NfL in serum. A tight correlation between NfL in CSF and blood has emerged, thereby raising the possibility that NfL could be a blood-based biomarker for neurological disorders. NfL levels have been shown to differentiate sporadic PD from other movement disorders, such as MSA and progressive supranuclear palsy (PSP). 7,12,13 While one small study did not show a longitudinal change in M o l l e n h a u e r 6 NfL levels during progression in PD, 14 longitudinal analyses of NfL in larger multicentre cohorts, including prodromal and monogenetic subjects, have not been carried out.
In this paper, we aim to answer the following questions: 1) Are there differences in baseline NfL levels and changes over time in the different patient groups, taking age and sex into consideration? 2) Does serum NfL correlate with CSF NfL? 3) Are NfL changes over time associated with clinical outcomes?
We hypothesized that serum NfL would: 1) be higher in PD (sporadic and with mutations) than in healthy controls (HC), 2) be even greater in subjects with other neurodegenerative disorders (OND), 3) increase over time with disease progression, 4) be higher in prodromal and asymptomatic subjects than controls, and 5) correlate with clinical outcome measures and/or imaging indices of progression

DeNoPa-and Kassel-training cohorts
The first training cohort from the longitudinal single-centre DeNoPa cohort 15  The second training set included paired CSF and serum samples from the cross-sectional Kassel cohort and comprised 151 PD and 344 OND. We also added 20 randomly selected healthy control (HC) subjects from the DeNoPa cohort who were already included in the first training cohort. Diagnoses were rendered according to published criteria.

Statistical analysis
All analyses were performed with the statistical software R (version 3.6.0; R Core Team 2018), using the R-packages lme4 for the linear mixed effect Models (LME), and geepack for Generalized Estimation Equations (GEE) in the smaller DeNoPa-cohort to test for longitudinal changes over time. The significance level was set to alpha = 5% for all statistical tests. More details on the statistical analysis of the training cohorts can be found in the supplement.

Analysis of the PPMI validation cohort
Spearman rank based correlation coefficients between log10NFL at baseline and the different other predictors were calculated for the entire cohort and in the PD sub-cohort. Estimating the rate of change of log10NFL over time was done via LME modelling allowing for random intercepts only (some models with random slopes had convergence issues and hence we opted for the simpler ones). For all models we adjusted for age, diagnostic groups (PD, HC, OND etc.), sex, and levodopa equivalent daily doses (log10LEDD). Testing 13 variables simultaneously causes an inflation in the occurrence of type I errors that we accounted for by setting the significance level to 0·05/13~0·0038.

Role of the funding source
The funders had no role in the design and conduct of the study, in the collection, management, analysis, and interpretation of the data, in the preparation, review, or approval of the manuscript or in the decision to submit the manuscript for publication.

Following a linear mixed model (Supplementary
Supplementary figure 1) encouraged us to move to the validation step using serum samples from the longitudinal PPMI-cohort.

Serum NfL measures in the PPMI-cohort
The baseline characteristics of the PPMI-cohort are shown in Table 1. Among the 1131 subjects, the prodromal group has the highest median age (67·7 years), followed by the OND group, the genetic PD group, the HC, the PD group, and the unaffected mutation carriers ( Table 1). The baseline serum NfL levels are highest in the prodromal and OND groups with a median of 14·5 pg/ml and an IQR as 7·2 pg/ml, and a median of 14·5 pg/ml and IQR as 10·3 pg/ml, respectively.

Cross-sectional serum NfL values at baseline in the PPMI-cohort
Multiple regression analysis shows that ( Table 2) on average there is a 3% increase in serum NfL for each year of age (p=0·0003). A woman's serum NfL level is on average 1·05 times that of a man's (p<0·0001). Given the same age and sex, patients with OND have a mean serum NfL level 47% higher than that of HC (p=0·008), and those with genetic PD have a mean serum NfL level 21% higher than that of HC (p<0·0001). There are also non-statistically significant trends of increased serum NfL level in PD and prodromal groups compared with HC (6% higher with p=0.08 and 10% higher with p=0·09, respectively).

Longitudinal change of serum NfL over time in PPMI-cohort
Using a linear mixed model ( Table 3) adjusted for sex, baseline age and baseline serum NfL, a healthy subject has an average increase of 3·2% in serum NfL each year (p<0·0001). Compared with HC, the increment rate is on average 1·06 times greater in PD (p<0.0001) and 1·46 times greater in OND (p<0·0001) (Figure 1b). The increment rate in the genetic cohort is not significantly different from that of HC (Table 3; Figure 1c).

Discussion
We investigated NfL levels by sandwich immunoassay in longitudinal and cross-sectional training cohorts from a single centre and validated the findings in a longitudinal multicentre cohort consisting of HC, prodromal and established PD. We also studied symptomatic and M o l l e n h a u e r 1 3 asymptomatic mutation carriers of known PD genetic mutations with follow-ups of up to six years. Results obtained from our training cohorts indicated that serum and CSF NfL levels were higher in PD subjects compared with controls; levels were higher in OND compared with PD.
CSF NfL levels increased longitudinally over four years and correlated with MDS-UPDRS III and total scores.
In the second step, we focused on NfL serum levels in the validation cohort, considering the strong correlation between CSF and serum NfL levels and the advantage of using a peripheral and less invasive source for the specimen. Consistent with the exploratory work in the training cohorts, the main findings in the validation cohort were: (1)  their respective symptomatic group. The levels in all six genetic groups (symptomatic and asymptomatic) remained relatively stable over the two to three years of follow-up and there were no differences in NfL rate of change between the mutation carrier groups. We also observed a higher level of serum NfL in symptomatic mutation carriers relative to PD patients, possibly reflecting a longer duration of disease. Subsequent analyses will explore the effects of disease duration on changes in serum NfL.
Elevated levels of NfL, as seen here in PD and OND compared with HC, have been identified in several other neurological conditions including dementia disorders and multiple sclerosis.
Therefore, this marker for axonal damage is not specific for any disease, 7 but could be useful for exploring specific questions within disease entities. Within PD spectrum disorders, NfL may be particularly of use in discriminating PD from cognate disorders such as MSA, PSP, and DLB, as has been previously described. 12  follow-up. Based on this process, seven subjects in the PD cohort analysed here were found to have diagnoses other than PD after five years follow-up and were thus taken out of the PD cohort. This group of subjects, despite the small size, was separated in the analysis as OND.
Serum NfL levels in this OND group were higher at baseline (before the follow-up and before evolving into another disease) and may provide future diagnostic utility. Additional studies including analysis on samples from post-mortem PD confirmed subjects will further establish the utility of serum NfL measurements in differentiating PD from OND. In numerous previous CSF studies (conducted before ultrasensitive technologies were available) the mean NfL levels in other neurological disorders were found to be markedly increased compared to HC, such as in multiple sclerosis (4·5-fold increase), traumatic brain injury (three-fold increase), PSP, CBD, MSA (3-4·25 fold increase) as well as in other, more slowly progressing neurodegenerative disorders such as Alzheimer's disease (1·5-2-fold increase). 17 Compared to these disorders, the increase is smaller in PD (1·6 in CSF and 1·25 in serum), where α -synuclein aggregation occurs mainly in neurons with high energy turnover in less myelinated axons, 18 while NfL is mainly expressed in larger myelinated axons. 19 Also, in contrast to many more rapidly progressive or acute diseases, the pathological changes in PD and the respective cell loss is only mild. 20 Despite the slight increase shown here, the relatively higher serum NfL levels in the prodromal groups (compared to PD and HC), especially in the iRBD group, may indicate potential for conversion to either PD or Parkinsonian syndromes. Furthermore, longitudinal NfL levels may, in fact, be highest in the early stages of the disease with the greatest disease activity as has been similarly seen in β -amyloid in Alzheimer's disease. 21 M o l l e n h a u e r 1 6 Another variable affecting the increase of NfL levels is age. Across published studies, NfL in CSF and blood increases with age 17 as we also identified in our cohorts (also shown in Supplementary Figure 2). For example, the prodromal groups in PPMI are on average 3-5 years older than the PD group and the iRBD subjects are older than the hyposmic subjects featuring higher NfL values. Age is likely also the main reason why asymptomatic SNCA mutation carriers have the lowest levels in the PPMI cohort-the mean age in this group is 42±4·6 years. The reasons for this positive association of serum NfL and age is explained by structural alterations of the axons with ageing, including vascular disease, metabolic changes, and inflammation. All of these have also been shown to play a role in PD progression 22 , which may also influence the NfL levels in blood.
In contrast to a previous report showing that CSF NfL levels were relatively stable despite disease progression in PD over 12 months 14  follow-up will identify subjects converting to a motor disease or symptomatic disease state. The individual levels may indicate the prognostic direction.
In conclusion, we have identified NfL as the first blood-based PD progression biomarker. We observed slightly higher levels of serum NfL in PD compared to HC even at early stages of the disease, a mild longitudinal increase in serum NfL levels over time and correlations of serum NfL with clinical measures of disease progression in two independent longitudinal cohorts.
Increases in serum NfL were, in general, less than those seen in OND, which are either more rapidly progressive than PD, associated with damage to myelinated tracts (as seen in multiple sclerosis), or associated with significantly greater cell death (as seen in Alzheimer's and extreme in Creutzfeldt-Jakob disease). Our NfL data will be strengthened with continued analyses of these data using additional alternative statistical models as well as follow-up of the cohorts, especially in subjects at risk for disease progression. In addition, monitoring of larger longitudinal cohorts with a focus on prodromal or asymptomatic PD with longer observational time to allow the development of motor disease are needed. This being said, we remain cognizant that increased NfL levels are not specific to PD or any other neurodegenerative disorder. Thus, more specific markers will need to be identified, leading hopefully to a panel of different markers reflecting disease state, rate, and fate. Finally, we note the profound influence of age and gender on serum NfL levels. We, therefore, recommend that age-and sex-based adjustments be applied when interpreting serum NfL levels in clinical research and practice. found that serum NfL is higher in established PD versus HC; There was a significant (ageadjusted) longitudinal increase in serum NfL in PD compared to HC and the longitudinal change in serum NfL correlated significantly with MDS-UPDRS III motor and total scores, suggesting that serum NfL may be a biomarker of clinical progression.

Implications of all the available evidence
This first large multicenter study shows that serum NfL is the first blood-based biomarker candidate that could support disease stratification and track clinical progression in PD.  wrote the manuscript. MD, TYL, WW, FG, DG, TF, HZ, SS, RG, NK, MF, LMC, TS, ABS,   DW, KM, AS, JMC, SH, DG, CMT and CT co-edited the manuscript. BM, SH, DG, and MD had full access to the clinical primary data and take responsibility for the integrity of the data and the accuracy of the data analysis. All authors had access to the data generated in the study including the statistical analysis and decided to submit the paper for publication.  T  h  e  r  a  p  e  u  t  i  c  s  ,  V  o  y  a  g  e  r  T  h  e  r  a  p  e  u  t  i  c  s  a  n  d  I  n  t  e  c   P  h  a  r  m  a  a  n  d  p  e  r  s  o  n  a  l  f  e  e  s  f  o  r  c  o  n  s  u  l  t  i  n  g  f  r  o  m  N  e  u  r  o  c  r  i  n  e  B  i  o  s  c  i  e  n  c  e  s  ,  A  d  a  m  a  s  T  h  e  r  a  p  e  u  t  i  c  s  ,  B  i  o  g  e  n  I  d  e  c  ,  2  3  a  n  d  M  e  ,  A  l  e  x  z  a  ,  G  r  e  y  M  a  t  t  e  r  ,  A  c  o  r  d  a  ,  A  c  The funders had no role in the design and conduct of the study, in the collection, management, analysis, and interpretation of the data, in the preparation, review, or approval of the manuscript or in the decision to submit the manuscript for publication.   Table 1). The dashed line represents a linear fit through the points and the vertical bars give estimates of the errors.  Table 1). The dashed line represents a linear fit through the points and the vertical bars give estimates of the errors.  LEDD mean ± sd 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 0 ± 0 median (min; max) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0) 0 (0; 0)