Abstract
Serum light-chain neurofilaments (sNfL) have been investigated as a potential minimally invasive biomarker that could help in the diagnosis of patients with cognitive symptoms. We assessed the correlation between sNfL and cerebrospinal fluid (CSF) biomarkers (sNfL versus CSF NfL, ρ= 0.70, p < 0.001), the performance of sNfL in distinguishing controls from patients (controls versus frontotemporal dementia, area under curve 0.86), and sNfL differences in mild cognitive impairment according to amyloid-β (Aβ) deposition (Aβ versus non-Aβ, p = 0.017). Our results support the role of this biomarker in the screening and risk stratification of patients followed in a neurological consultation of a tertiary center.
Keywords
INTRODUCTION
A great effort has recently been dedicated to the search for low-cost, minimally invasive blood-based biomarkers that could help in the screening, diagnosis, and prognosis of patients with cognitive symptoms [1]. The use of techniques like single molecule array (Simoa) that are sensitive to very low concentrations in serum is paving the way to achieve this goal [2].
Alzheimer’s disease (AD) biomarkers in cerebrospinal fluid (CSF), like amyloid-β42 (Aβ42) and phosphorylated tau (p-tau), have shown great correlation with brain amyloid burden and tangle pathology, respectively, and are well established in updated research diagnostic criteria [3] and biological definitions of this disease [4]. The validation of these biomarkers in serum is being extensively investigated with variable performance [5].
Additional blood biomarkers are needed with possible application in other neurodegenerative diseases [1]. Light-chain neurofilaments (NfL) are a marker of nonspecific axonal degeneration [1] and were proven to be useful 1) in the diagnosis of frontotemporal dementia (FTD), distinguishing it from cognitively unimpaired (CU) patients, mild cognitive impairment (MCI), and AD [6–9]; 2) in the prediction of cognitive progression [8–11]; and 3) as a surrogate of cortical atrophy [7, 8, 10].
The diagnostic accuracy of NfL in biologic fluids is highly variable in the literature. As an example, the area under curve (AUC) reported to distinguish AD from FTD varies between 0.51–0.75 [7, 8, 12]. We assessed the performance of NfL in serum (sNfL) in the differential diagnosis of cognitive symptoms in a large cohort of patients followed in the neurological consultation of a tertiary center. We additionally evaluated the correlation between CSF biomarkers and sNfL and how sNfL varied according to Aβ deposition in MCI.
METHODS
Patient selection and investigation
We retrospectively collected demographics and clinical data from patients followed in our dementia consultation, which had sNfL analysis performed at the age of 40 years or above, and for whom CSF biomarkers and/or genetic diagnosis were requested by assistant neurologist for diagnostic clarification. We included patients fulfilling current diagnostic criteria for FTD [13, 14], AD [15], or MCI (regardless of presumed neuropathology) [16] (n = 285). In this patient-group, the diagnosis was supported by CSF biomarkers (Aβ42 or Aβ42/Aβ40 ratio, total tau (t-tau), and p-tau) in 279 patients. A definite diagnosis of frontotemporal degeneration was attempted by genetic testing, finding 21 patients with progranulin and 13 with C9orf72 mutations.
We additionally considered a group of CU controls (without memory complaints and normal cognitive performance) (CTRs) and age at collection≥40 years old (n = 261). Considering that the age distribution of this group was significantly younger than some patients groups (CTRs 63.2; MCI 67.2; AD 64.0; FTD 62.9 years; p = 0.001), we performed a stratified sampling of CTRs randomly selecting 20 individuals for each age quartile of patients (n = 80).
The study was conducted according to the Declaration of Helsinki.
Measurement of NfL in serum and CSF
Blood samples were collected into serum separation tubes, centrifuged at 1,800 g for 10 min at 4°C, and stored at -80°C. NfL were later measured by Simoa technology using the NF-light Advantage kit (SR-X platform, Quanterix). This technique uses an array of femtoliter-sized wells where a conventional ELISA assay is performed on microscopic beads, and reaction fluorescence digitally captured by an optic microscope. Simoa was chosen to evaluate sNfL as it allows the measurement of subfemtomolar concentrations [17].
NfL in CSF (CSF NfL) were measured in 271 patients with cognitive symptoms and 25 CTRs. CSF was diluted 1 : 1 and NfL were measured using a validated ELISA assay (NF-light; Uman Diagnostics). Measurement of CSF AD biomarkers was described elsewhere [9].
Statistical analysis
All statistical analyses were performed using R statistical software (version 4.0.1) and RStudio (Version 1.3.959). Data analysis used tidyverse, graphics plotting used ggplot2 and receiver operating characteristic (ROC) curves used plotROC software packages.
Age at the time of blood sampling and gender were compared across diagnostic groups with non-parametric Kruskal-Wallis test (with Dunn’s test as posthoc analysis).
To analyze the correlation between sNfL and CSF biomarkers we used Pearson (r) or Spearman (ρ) correlation, depending on the fulfilment of linear model assumptions (analysis of residuals and global validation of linear models assumptions with gvlma function were applied).
Non-normality of residuals and sNfL (even after log transformation) and heteroscedasticity were concluded using Shapiro-Wilk normality test and Bartlett test of homogeneity of variances, respectively, and gvlma function. Therefore, non-parametric Kruskal-Wallis test was used to analyze differences in diagnostic groups (with Dunn’s test as posthoc analysis). A robustness analysis applying parametric models and adjusting for some of the above limitations supported the results (see supplementary material).
We assessed the performance of sNfL in distinguishing CTRs, MCI, AD, and FTD patients plotting ROC curves. Sensitivity and specificity were obtained for the optimal cut-off point estimated maximizing the Youden index.
The sNfL comparison between MCI patients with Aβ deposition (A+) versus Aβ negative (A-) in the Aβ tau, neurodegeneration (ATN) classification system was performed using Wilcoxon rank-sum test (not reported for AD as the number of patients in the A- group was residual).
RESULTS
Patient demographic and clinical characterization
The 285 patients were classified in 3 groups: MCI (n = 119; 41.8%), AD (n = 81; 28.4%), and FTD (n = 85; 29.8%). We obtained statistically different ages when we compared all 3 diagnostic groups and controls (CTRs 65.1±10.8; MCI 67.1±9.1; AD 64.0±6.4; FTD 62.9±6.8 years; p = 0.001), with a posthoc analysis showing AD and FTD patients significantly younger than MCI. No gender differences were found across groups (p = 0.15). Demographics, clinical staging, and biomarkers characterization are provided in Supplementary Table 1.
Description of NfL in CSF and serum and their correlation
CSF NfL were distinct among groups, and patients with FTD showed the highest concentration (CTRs 723±402; MCI 1057±1322; AD 1411±731; FTD 4936±4077 pg/mL; p < 0.001). Posthoc analysis did not show statistical differences in the pair CTRs/MCI.
sNfL reproduced the results observed in CSF, with CTRs showing the lowest and FTD the highest concentrations of the biomarker, and differences across groups being statistically significant (CTRs 15±14; MCI 19±15; AD 23±13; FTD 50±45 pg/mL; p < 0.001). A posthoc analysis showed differences across all groups (Fig. 1A).
A) sNfL were statistically different across diagnostic groups (p < 0.001), with FTD showing the greatest concentrations of the biomarker. Posthoc analysis showed differences between all diagnostic group pairs. Vertical axis presented in logarithmic scale. B) ROC curves between diagnostic groups and CTRs showed the best performance for FTD with an AUC of 0.86. C) ROC curves between FTD and AD groups showed and AUC of 0.74 with a specificity of 90.1%, and the curve between AD and MCI showed and AUC of 0.63. Diamond shows the OCP, calculated by the Youden index. AD, Alzheimer’s disease; AUC, area under curve; CTRs, cognitively unimpaired controls; FTD, frontotemporal dementia; MCI, mild cognitive impairment; OCP, optimal cut-point; ROC, receiver operating characteristic; sNfL, Light-chain neurofilaments in serum.
CSF NfL correlated well with sNfL with ρ= 0.70 (p < 0.001).
Diagnostic accuracy of NFL in serum
ROC curves were plotted to assess the performance of sNfL in the distinction of CTRs from MCI, AD, and FTD (AUC = 0.62; 0.75; 0.86, respectively). The best performance was achieved for the CTRs/FTD curve, and its optimal cut-point was 20.9 pg/mL, with a sensitivity of 76.5% and a specificity of 85% (Fig. 1B). ROC curves for AD/MCI and FTD/AD showed an AUC of 0.63 and 0.74, respectively (Fig. 1 C). The last one showed a high specificity (90.1%) and a low sensitivity (57.6%).
Correlation between sNfL and AD biomarkers in CSF
When we considered patients with MCI and AD, CSF AD biomarkers correlated well with sNfL (Aβ42 ρ= – 0.32, p < 0.001; p-tau ρ= 0.16, p = 0.025; t-tau ρ= 0.26, p < 0.001; Aβ42/Aβ40 ratio ρ= – 0.17, p = 0.02; Fig. 2A). However, when only patients in the AD continuum were considered (i.e., MCI and AD with evidence of Aβ deposition), no significant correlation was obtained. FTD patients showed a positive correlation between t-tau and sNfL (r = 0.31, p = 0.004; Fig. 2B).
Correlation between CSF AD biomarkers and sNfL. A) MCI and AD patients showed significant correlations between all CSF AD biomarkers and sNfL. B) For patients with FTD, only CSF t-tau showed a positive correlation with sNfL. Biomarkers named with “Log” represent logarithmic values. Abeta42CSFLog, logarithmic value of Aβ42 in CSF; AD, Alzheimer’s disease; BM, biomarker; CSF, cerebrospinal fluid; FTD, frontotemporal dementia; MCI, mild cognitive impairment; n.s., nonsignificant; PTauCSFLog, logarithmic value of p-tau in CSF; RatioAbeta42_40CSF, Ratio Aβ42/Aβ40 in CSF; sNfLLog, logarithmic value of sNfL; TauCSFLog, logarithmic value of t-tau in CSF.
Association between sNfL and ATN status in MCI patients
Patients with MCI were divided in two groups according to their ATN status: evidence of Aβ deposition (i.e., A+, with any T or N) versus no evidence of Aβ deposition (i.e., A-, with any T or N). No age (A + 67.6 (±8.8), A- 67.0 (±9.2) years; p = 0.66), or gender differences (A + 41.0%, A- 38.6% males; p = 0.81) were found. We concluded that patients with evidence of Aβ deposition were associated with higher sNfL (A + 22±14, n = 39; A- 18±15 pg/mL, n = 80; p = 0.017) (Fig. 3).
Patients with MCI and A + in the ATN classification system (i.e., with evidence of Aβ deposition) showed higher values of sNfL than those with A-. Vertical axis presented in logarithmic scale. A+, Aβ deposition; A-, no evidence of Aβ deposition ATN, Aβ, tau, neurodegeneration system; MCI, mild cognitive impairment; n.s., nonsignificant; sNfL, light-chain neurofilaments in serum.
DISCUSSION
Studies validating the accuracy of peripheral biomarkers in well characterized cohorts of patients with cognitive decline are utterly important. In this short report, we present the performance of sNfL as a biomarker in the diagnosis of patients in a large cohort of a tertiary center.
We confirmed that sNfL can be used as a proxy of CSF NfL obtaining a good correlation between CSF NfL and sNfL (ρ= 0.70). This result was recently reported using Simoa in plasma samples in AD (ρ= 0.59) [18], corroborated in a meta-analysis (r = 0.59) [19], and reproduced for other neurodegenerative diseases [7, 8, 10].
Regarding the correlation between sNfL and CSF AD biomarkers, we found that although CSF AD biomarkers correlate well with sNfL in MCI and AD patients, this correlation was absent when we considered only A + patients. Although a smaller sample effect could account for this finding, this was unexpected as Aβ and tau pathology are precursors of neurodegeneration, and NfL are considered a non-specificity neuronal injury marker. Results reported in the literature on biomarker correlation are unclear [20]. Mattsson et al. found this correlation for all CSF AD biomarkers when he considered a cohort of CU controls and MCI and AD patients, but not when he analyzed AD patients individually [18]. On the contrary, another study reported good correlations for AD patients [21]. In the FTD patients, CSF t-tau was the only CSF AD biomarker that correlated with sNfL, as both biomarkers translate neurodegeneration not necessarily caused by AD pathology [1].
When analyzing MCI, we concluded that A + patients showed higher sNfL concentration. This result can be interpreted considering the higher probability of MCI patients in the AD continuum converting to dementia [4], the growing evidence that sNfL also associate with cognitive deterioration [10, 22], and the variation of sNfL in the different ATN status [23].
The differences in sNfL among behavioral variant FTD, AD, MCI, and controls were initially reported for a small cohort of patients [7], and later extended to include other diagnoses [12, 24]. Our results replicated these findings, showing that sNfL may be useful in distinguishing diagnostic groups, and their mean concentration increases from CTRs to MCI, AD and FTD.
Finally, in our sample sNfL showed the highest accuracy in distinguishing CTRs from FTD patients. A previous work reported an almost perfect biomarker for the pair CTRs/FTD with AUC = 0.97, using Simoa [8]. Another recent study publishing the accuracy of sNfL in two multicenter cohorts, the largest published to date, reported an AUC for the pair CTRs/FTD in the interval 0.92–0.62, having our result landed in that interval (AUC = 0.86) [12]. Even another study, which analyzed multiple serum biomarkers in a smaller cohort, reported an AUC of 0.85 for the pair MCI+AD/FTD (sensitivity 89%, specificity 75%) [24].
Our study has two important strengths: it presents a cohort obtained from a neurological consultation and not primarily for investigational purposes with a relevant number of patients and controls, and the clinical diagnosis was supported by CSF biomarkers. As limitations, we point out: 1) It refers to a single neurological center with a limited number of neurodegenerative diseases and no pathological confirmation (some FTD patients had genetic confirmation); 2) There are age differences between diagnostic groups. Although age associates with higher sNfL concentrations [20], this fact does not affect the validity of results, as MCI patients were older than AD and FTD groups, and this could even accentuate the difference in sNfL between these groups (see Supplementary Material); 3) stratified sampling on CTRs was considered an appropriate strategy to adjust for age in this group, but it considerably reduced the available sample; 4) AD group is younger than expected from epidemiology, a fact that might be explained by a selection bias of patients with atypical clinical AD presentations for which biomarkers were requested for diagnostic clarification.
With higher life expectancy resulting in an increase in the prevalence of dementia, health care systems need easily accessible and affordable biomarkers that could help in the diagnosis and prognosis of patients with cognitive symptoms. As a potential strategy to stratify risk in primary care, some authors suggest the measurement of serum p-tau, and subsequently of NfL in low probability cases of AD [25]. Our study supports this strategy, as sNfL performed well in distinguishing FTD from CU patients.
Further studies like ours are needed to translate sNfL to real clinical contexts.
Footnotes
ACKNOWLEDGMENTS
The authors have no acknowledgments to report.
FUNDING
This work was financed by the European Regional Development Fund (ERDF), through the Centro 2020 Regional Operational Programme under project CENTRO-01-0145-FEDER- 000008:BrainHealth 2020, and through the COMPETE 2020– Operational Programme for Competitiveness and Internationalization and Portuguese national funds via FCT– Fundação para a Ciência e a Tecnologia, I.P., under project UIDB/04539/2020: CIBB. A.S.-S., M.L., and M.J.L. were supported by the Portuguese Foundation for Science and Technology (DFA/BD/6393/2020, SFRH/BD/144001/2019 and PD/ BD/135108/2017, respectively). The funding agency had no role in the study design, sample collection, data analysis, or the writing of the article.
CONFLICT OF INTEREST
M.L. and I.B. are Editorial Board Members of this journal but were not involved in the peer-review process nor had access to any information regarding its peer-review.
DATA AVAILABILITY
Anonymized data are available upon reasonable request to senior author for investigational purposes.
