Background
Cerebrospinal fluid (CSF) biomarkers have been extensively studied as tools for an early diagnosis of Alzheimer’s disease (AD) [
1] and have proven to be cheaper, less demanding in terms of infrastructure and patient management, and most probably capable of showing pathologic alterations slightly earlier than other diagnostic modalities such as, for example, positron emission tomography; the relative invasiveness of the two modalities remains disputable [
2]. In particular, the four core CSF biomarkers, amyloid β (Aβ)1–42, Aβ42/40 ratio, Tau, and pTau181, reliably support AD diagnostics reflecting the hallmark AD pathologies, i.e., amyloidosis and neurodegeneration [
3]. Lumbar puncture (LP) is a routine clinical procedure with a low incidence of complications [
4]. Nevertheless, collection of CSF is accompanied by procedural efforts and inconvenience for subjects, ultimately preventing its use as a screening tool in early, asymptomatic cases and it can also be challenging to use for repetitive monitoring of the disease progression. Hence, there is a strong need to develop blood-based biomarkers that, when applied in a proper context of use, could serve as targeted and relative noninvasive screening tests [
5].
Neurofilaments (Nf) consist of three types of protein chains, differing in molecular mass: a light chain (NfL) of 68 kD, an intermediate chain of 150 kD, and a heavy chain of 190 to 210 kD, and are major components of axonal cytoskeleton [
6]. Each subunit is composed of a double-stranded, highly conserved α-helical core domain, bordered by a head N-terminus and a tail C-terminus. Nf are highly phosphorylated proteins, and the degree of this phosphorylation determines the axon diameter [
6]. Axonal damage leads to release of Nf molecules into the extracellular space and, consequently, into body fluids, such as the CSF or plasma. In line with this, increased blood NfL concentrations were reported in neurodegenerative and neuroinflammatory disorders [
7‐
9]. A very recent report by Mattsson et al. convincingly concludes that plasma NfL could be considered a neurodegeneration biomarker in AD [
10].
In this study, we measured NfL concentrations in plasma samples of AD patients at the stages of mild cognitive impairment (MCI) and early dementia (ADD), and nondemented controls. In all our subjects the clinical and neuropsychologic diagnoses were in accordance with the results of their four core CSF biomarkers (Aβ1–42, Aβ42/40, Tau, and pTau181). For the validation of the assay, we further tested the influence of different preanalytical sample handling procedures (repetitive freezing/thawing, and storage at room temperature or in a refrigerator) on the concentrations of plasma NfL.
Discussion
A growing body of literature postulates a potential application of the plasma concentration of the light chain of the neurofilament protein as a screening tool for neurodegeneration. In our study, the plasma NfL concentration was found to be significantly higher in the AD patients, whose diagnoses were in accordance with the pathological results of the CSF biomarkers, compared with the nondemented control subjects, whose normal cognitive status was in accordance with the unaltered results of their CSF biomarkers. Within the AD group, we observed a tendency towards higher NfL concentrations in patients at the stage of early dementia compared with the stage of MCI. Plasma NfL concentrations correlated inversely with the global cognitive status measured by the MMSE results. Finally, we also provide evidence for the influence of the preanalytical sample handling on the plasma NfL concentrations.
We started our study by examining the influence of sample handling procedures on plasma NfL concentrations. We believe this is a critically important aspect; for example, a sample’s storage and transportation to distant laboratories is a nontrivial issue and has practical implications. Compared with a reference sample (i.e., a deep-frozen aliquot stored unthawed until analysis), we observed a slight but significant increase in the concentrations in the aliquots thawed and refrozen twice or three times, kept for 5 days at the room temperature, or stored in a refrigerator. In contrast, one thawing/refreezing cycle did not systematically alter the NfL concentration, but resulted in nonsystematic changes, i.e., the concentrations increased in some samples but decreased in others, resulting in increased variability but an unchanged average. Whereas in most cases a decrease in the concentration of a protein with storage time or under thawing/refreezing is expected, it is known that some proteins, for example serum albumin, tend to increase in concentration after repetitive thawing/refreezing [
20]. Similarly, in our previous studies, an unsystematic increase in CSF Aβ1–42, Aβ1–40, and Tau in some, but not all, aliquots exposed for more than two thawing/refreezing cycles was observed [
21,
22]. As an explanation, a release of NfL monomers from aggregates, known to form in certain neurodegeneration disorders [
23], might be considered. Interestingly, the data on the stability of NfL in the CSF are ambiguous; whereas one study found rapid decline of concentrations at room temperature or 4 °C [
24], no changes were observed in another study until 3 days, followed by a subsequent decrease [
25]. In any case, it seems reasonable to postulate sending the material to distant laboratories deeply frozen and avoiding more than one intermediate thawing/refreezing cycle.
In agreement with recently published studies on sporadic AD [
10,
26,
27] and familial AD (FAD) [
28], we found increased plasma NfL concentrations in AD patients in the dementia stage (ADD) as well as in MCI subjects with a high probability of underlying AD pathology (MCI-AD), compared with nondemented controls. The results of the current study confirm those reported by Mattsson et al. obtained with the same method and in the same laboratory but on different patient cohorts [
10], not only in terms of the average concentrations and their biological variability (coefficients of variation), but also in terms of the NfL performance as a potential plasma diagnostic test. In both studies, almost identical areas under the ROC curves, contrasting AD patients versus nondemented controls, were obtained (0.853 and 0.87, respectively). A slightly higher average NfL concentration in the controls reported by Mattsson et al. compared with the present study can be explained by the fact that one-third of the controls in the previous paper showed Aβ positivity, whereas in the current study positive Aβ CSF results excluded a subject as a control. This is due to the fact that, in contrast to the papers published previously, patients in the current study were included only if their clinical and neuropsychological diagnoses stayed in agreement with the outcome of the four core CSF biomarkers (Aβ1–42, Aβ42/40, Tau, and pTau181) conservatively interpreted according to the Erlangen Score algorithm [
13,
14]. We are aware that such an approach has advantages and disadvantages; it enables more reliable stratification of the cases, but it excludes the possibility of a direct comparison of the diagnostic utility of the plasma NfL with any of the CSF biomarkers.
Our finding of a positive association between plasma NfL concentration and age is in agreement with previously reported studies on plasma [
9,
10] and CSF [
29,
30]. A weak but significant association between serum NfL and age at onset of a disease was also reported in primary progressive aphasia (PPA) [
8]; however, another study did not find a correlation of NfL with age after adjusting for the estimated age of onset of a disease in FAD [
28]. Mechanisms of this age-dependent increase in the NfL concentrations in body fluids, and also in persons without clinical signs of neurodegeneration, are unclear thus far. It was hypothesized that aging leads to a subclinical axonal degeneration and, in consequence, to the release of Nf molecules [
29]. The same group also proposed that subclinical cerebrovascular changes might be considered as an explanation, since cerebrovascular pathology is common and known to increase with age; finally, vascular copathology is also commonly observed in AD [
31]. Irrespective of the underlying mechanisms, the association of the NfL concentrations with age has implications for the diagnosis-oriented interpretation of the results. First, an age-dependent cutoff needs to be established, and calculation of such a cutoff is not a trivial task. Perhaps the best approach is by applying such statistical tools such as LDA; in such a case, however, the slope of the line discriminating the groups (i.e., the age-dependent cutoff) clearly depends on the distribution of the parameters in question (here NfL concentrations and age) in these groups. If they are not age-matched, as in our study and in some other reports [
27,
28], a line best discriminating AD patients from the controls has a negative slope, which might look contradictory to common sense (i.e., in spite of NfL concentration increasing with age, its cutoff decreases). This would be different if the two groups were age-matched, as in the study by Mattsson et al. [
10]. In such a case, the discriminatory line would have a positive slope (i.e., it would increase with age). Secondly, metrics of the performance of the NfL concentrations as a potential diagnostic test also depend on age. In this study, we observed an increase in the sensitivity at the cost of a decrease in the specificity with increasing age, leaving overall accuracy practically unaltered. Furthermore, we observed a slight, borderline insignificant increase of the area under the ROC curve with age. To the best of our knowledge, only Mattsson et al. [
10] evaluated the age-dependent AUC of the ROC curve discriminating AD from healthy controls, observing a slight decrease of the AUC from 0.87 to 0.79 when the model was fitted with age, sex, and educational level, instead of all variables considered in their study. We are not aware of any report analyzing age-dependency of any other metrics. We believe that the characteristics found in this study, with an age-dependent increase in the sensitivity at the cost of the decreasing specificity clearly seen in the age range of 60–80 years (i.e., in the range when neurodegeneration is most commonly considered in the diagnosis), further supports the postulated potential application of the plasma NfL as a screening tool for neurodegeneration, rather than as a test for confirming AD diagnosis.
In line with the recently published results [
10], we found an inverse correlation of plasma NfL concentrations with MMSE results. In contrast, none of the CSF biomarkers measured in this study correlated significantly with the MMSE score after controlling for the diagnostic categories. This finding supports our previous results of a lack of association between MMSE score and the CSF results [
32,
33], and remains in agreement with the generally accepted assumption that the CSF biomarkers do not correlate with disease progression at the stage of MCI and later [
34]. Other studies provide evidence that NfL plasma concentrations reflect the dynamics of neurodegeneration processes measured with different metrics. Steinacker et al. found an association between increased NfL concentration and functional decline and progression of atrophy in the left frontal lobe of PPA patients [
8], and Weston et al. reported an association of serum NfL concentration with the time from symptom onset in FAD [
28]. Similarly, increased CSF NfL was found to correlate with decreased MMSE score and with faster brain atrophy over time, as measured by changes in whole-brain volume, ventricular volume, and hippocampus volume in AD [
35]. In multiple sclerosis, CSF NfL reflects acute axonal damage, and hence it might be considered a prognostic biomarker (reviewed in [
6]).
Similar to the previously published findings [
10], we observed a highly significant overall correlation of plasma NfL with CSF biomarkers for AD pathology when the diagnostic categories were not considered. Confirming the previous report, the significance of this correlation disappeared when the diagnostic groups were evaluated separately. Such a correlation pattern, with overall significant correlation that is not observed within particular diagnostic groups, is not surprising when a lack of association between the CSF biomarkers and the disease dynamics as soon as the first cognitive symptoms occur (i.e., from the MCI stage on) is taken into consideration [
34].
Perhaps the most important limitation of our study is the relatively small, age-unmatched groups, which we tried to counterbalance by controlling for age in all statistical analyses. It must be stressed, however, that such discrepancy between age of AD patients and nondemented controls simply reflects the reality that AD patients are older.