1. Introduction
Neurofilament light chain (NfL) is a cytoskeletal protein exclusively expressed in neurons. After axonal damage, NfL is released consecutively into the extracellular fluid, the cerebrospinal fluid (CSF) and the blood [
1]. Using ultrasensitive single molecule array assays (Simoa) it is possible to measure low concentrated NfL, not only in CSF but also in blood samples, with very high sensitivity. Blood levels correlate strongly to those measured in CSF [
2,
3,
4].
Elevated NfL levels in CSF and peripheral blood have been described in a number of neurological conditions including Alzheimer’s disease (AD). According to Bridel et al., the highest NfL levels are seen in HIV-associated dementia, frontotemporal dementia, vascular dementia, amyotrophic lateral sclerosis and atypical parkinsonian syndromes [
5]. However, because of its low specificity for AD, NfL is not suitable for the differentiation between AD and other dementias [
6,
7]. Recent studies have shown that plasma tau (pTau) phosphorelated at different sites (pTau181, pTau217, pTau231) might facilitate the differentiation between AD and other dementias [
8]. The potential application for NfL is in the early detection of AD and the assessment of the course of the disease [
6].
Therefore, blood based measurements of NfL (in serum or plasma) are promising, easily accessible biomarkers of neurodegeneration, and may thus contribute to define disease severity, quantify time to disease milestones and identify disease progression in AD [
9].
It has been shown that AD patients have higher CSF and blood levels of NfL compared to cognitively unimpaired persons and persons with mild cognitive impairment [
9,
10,
11,
12]. High CSF and blood NfL levels in AD patients were found to correlate with worse performance in cognitive tests [
9,
10,
12,
13,
14] and may predict cognitive decline [
15,
16,
17]. Elevated NfL blood levels have been found even in the presymptomatic stage in familial [
13,
18,
19] and sporadic AD [
16,
20], and have been suggested as a marker for tracking neurodegeneration in the different stages of AD [
9,
17,
20,
21].
We here contribute to the validation of sNfL as a neurodegeneration marker in AD by assessing the association between sNfL, cognitive functions, neuropsychiatric symptoms, and brain atrophy in the Prospective Dementia Registry-Austria (PRODEM-Austria). We hypothesized that baseline sNfL levels (1) relate to performance in cognitive tests and predict cognitive decline over time, (2) relate to neuropsychiatric symptoms and, (3) relate to cortical volume loss, particularly in AD signature regions. Our findings rely on a cohort of 237 patients with AD that were prospectively recruited in four university clinics in Austria and were followed over a mean period of 18 months using identical investigational protocols.
2. Materials and Methods
Participants: Data are from the prospective registry on dementia in Austria (PRODEM-Austria), a longitudinal, multi-centre cohort study supported by the Austrian Alzheimer Society. Inclusion criteria of PRODEM-Austria were diagnosis of dementia, no need for 24 h-care and availability of a caregiver who was able to provide information about the patients’ condition. Patients that could not sign the informed consent or had severe co-morbidities were excluded from the study.
For the present study, we used data collected from 23 June 2009 through 31 May 2017. All PRODEM participants had possible or probable AD according to the NINCDS-ADRDA criteria [
22], the state-of-the art criteria at the time of patient recruitment, and serum NfL (sNfL) measurements. Exclusion criteria were other dementias than AD and cerebrovascular disease within six months before the sNfL measurement. Moreover, seven patients who were defined as outliers because of sNfL concentrations > 3 standard deviations of the study cohort mean were also excluded. The final study cohort consisted of 237 participants.
All participants underwent a detailed clinical assessment, cognitive testing, and laboratory evaluation at baseline, after six months, after one year and after two years. Of the included participants 221 (93.3%) had come to at least one follow-up visit. The mean follow-up time was 22 months. If participants had come to more than one follow-up visit, we used the one year follow-up visit.
A subgroup of 125 participants underwent MRI at baseline, and of those, 111 had at least one follow-up examination. The mean follow-up time in the MRI subgroup was 18 months. There was no significant difference in demographics and outcome measures between the MRI subgroup and the whole cohort.
Serum NfL: Serum samples were obtained by peripheral venipuncture and were frozen at −80 °C according to standard procedures [
23]. The sNfL was quantified by an ultra-sensitive single molecule array (Simoa) on a Quanterix SR-X Analyzer (Quanterix, Billerica, MA, USA) using the commercially available NF-light
® assay. All measurements were done it duplicate to assure inter-assay precision by the calculating the coefficient of variation (CV). Results with CVs below 20% were accepted for statistical analysis.
Neuropsychological Tests: Global cognition was assessed using the mini mental state examination (MMSE) and the test battery “Consortium to Establish a Registry for Alzheimer’s Disease (CERAD)-Plus” [
24]. The CERAD-Plus test included subtests for verbal memory and recognition, constructional praxis, figural memory, confrontational object naming, verbal fluency, and cognitive flexibility. The Neuropsychiatric Inventory (NPI) was used to assess frequency and severity of neuropsychiatric and behavioural symptoms. Besides the NPI total score, we defined three subgroups of neuropsychiatric symptoms based on Garre-Olmo et al. as follows: a “psychotic cluster” (including “delusions” and “hallucinations”), an “emotional cluster” (including “agitation”, “irritability”, “depression” and “anxiety”) and a “behavioural cluster” (including “euphoria”, “disinhibition”, “aberrant motor behaviour” and “apathy”). For each symptom we calculated a subscore by multiplying the frequency of the symptom by its severity. The NPI total score and the psychotic, emotional and behavioural cluster were calculated as sum of the corresponding subscores [
25].
Magnetic Resonance Imaging: This was performed either on a 1.5 Tesla (T) whole body MR scanner (Avanto or SymphonyTim; Siemens Healthcare, Erlangen, Germany) or on a 3 T whole body MR scanner (TrioTim; Siemens Healthcare, Erlangen, Germany), with 1 mm isotropic resolution. Brain volumetry was assessed using the FreeSurfer image analysis suite (version 6.0, freely available online
https://surfer.nmr.mgh.harvard.edu/, accessed: 11 March 2022). Technical details are described elsewhere [
26,
27]. We defined six regions of interest (ROI) of the volumetric data that have been shown to be sensitive for early AD. The ROI included the hippocampus, the parahippocampal cortex, the cuneus, the precuneus, the entorhinal cortex, and the inferior parietal cortex [
28]. All regions were normalized for the estimated total intracranial volume and averaged between right and left sides.
Statistical analysis: Statistical analysis was performed using the IBM SPSS Statistics (Version 26.0, SPSS Inc., Chicago, IL, USA). Normally distributed continuous variables are presented using the mean and standard deviation (SD), nominal data is presented with absolute numbers and percentages. To assess normal distribution, we used the Shapiro–Wilk test and visual inspection of histograms. In case of non-normality, we used log-transformation to achieve normal distribution.
We used Pearson’s correlation and multiple linear regression to assess the relationship between (1) sNfL and cognitive function and (2) sNfL and cerebral atrophy. As it has been shown that sNfL increases with age, especially after around 60 years of age [
29], we corrected for age in the linear regression analysis. In the analysis of the cognitive function, we further corrected for sex and education and in the analysis of the cerebral atrophy we corrected for sex. Because of statistical extreme values and non-linearity of the NPI results, we used Spearman’s correlation to assess the relationship between sNfL and neuropsychiatric and behavioural symptoms.
For the longitudinal analysis, we calculated the annualized absolute change of the cognitive and the MRI measures for each individual. All tests were 2-sided and a p-value (P) below 0.05 was considered as statistically significant. We used Bonferroni correction for multiple testing.
4. Discussion
In this prospective study in patients with AD, higher sNfL levels were associated with worse performance in cognitive tests, behavioural symptoms and smaller cerebral volumes of regions typically affected by AD pathology [
28]. The sNfL levels at baseline predicted cognitive decline but not deterioration of neuropsychiatric behavioural symptoms and brain volume loss.
The observed association with cognitive performance is in line with previous results [
9,
11,
12,
14,
15,
30,
31,
32]. We found the strongest association between sNfL and cognitive functioning on subtests of language function. A somewhat weaker association was found between sNfL and constructional praxis. Like other studies we also found that baseline sNfL levels predicted more rapid future cognitive decline [
15,
16,
21,
33].
The sNfL was associated with the presence of neuropsychiatric symptoms measured by the NPI total score. Among the three symptomatic clusters of neuropsychiatric symptoms sNfL related only to the behavioural cluster. The correlation was modest, yet independent of age and sex. So far, only one other study has examined the relationship between CSF, NfL, and neuropsychiatric symptoms in AD patients and failed to observe a significant correlation [
34]. It is unclear whether sNfL is directly related to neuropsychiatric symptomatology or the association is indirect because behavioural abnormalities indicate faster decline of cognitive functions [
17,
35]. Previously published data suggested an association between behavioural abnormalities and tau-associated pathology, as well as axonal degeneration in AD [
36].
The brain regions that were strongest related to sNfL were the temporal lobes, hippocampal and parahippocampal cortex, as well as the entorhinal cortex. These findings replicate previous findings that reported associations of sNfL with lower volumes of hippocampus [
9,
31] and nonhippocampal AD signature regions [
9,
31,
37]. We failed to find an association between baseline sNfL and progression of AD related atrophy. Presumably, the mean observational time of our study of 18 months was too short to detect change that exceeds measurement variability of automated longitudinal volumetric assessment [
38,
39].
The strengths of our study are the extensive diagnostic work-up including a detailed cognitive and neuropsychiatric behavioural assessment plus structural MRI scanning. One limitation of the study is that we used the NINCDS-ADRDA criteria for the diagnosis of AD [
22]. These criteria were standard at the time of patient recruitment into the current investigation. Recently, the criteria have been revised, yet it was shown that the NINCDS-ADRDA criteria provide a diagnostic sensitivity and specificity of 81% and 70%, respectively, in clinicopathologic studies [
40]. Further limitations were the relatively short follow-up time of 22 months in the whole cohort and 18 months in the MRI subgroup and the absence of a healthy control group.