Comparative analysis of three platforms for serum NfL quantification in healthy controls and MS patients

Introduction

Since the introduction of highly sensitive immunoassays (such as single molecule array-based analysis, SIMOA [1]) serum neurofilament light chain (sNfL) has been established as a blood-based biomarker for neuroaxonal damage in neuroinflammatory and neurodegenerative diseases [2], [3], [4]. Particularly in the field of multiple sclerosis (MS), sNfL has received widespread attention as a blood-based biomarker for assessing disease activity and progression in conjunction with clinical parameters and imaging [5], 6]. Elevated levels of sNfL are associated with disease activity, response to treatment, and treatment efficacy, making it a candidate for personalized patient management in MS [7], [8], [9], [10]. However, the widespread clinical application of sNfL measurements remains challenging due to the limited availability of testing. Until recently, no clinically certified testing platform was available, and the pioneering SIMOA technology-based platform from Quanterix was “Research-use-only.” Nevertheless, the testing landscape is undergoing a transformation, with one new platform (Siemens) already certified for routine diagnostics, and two other widely available platforms (FujiRebio and Roche Diagnostics) actively seeking certification. The resulting availability of diverse parallel platforms for measuring sNfL demands standardization to ensure consistent and reliable data for patient care and comparable data for scientific questions. Standardization requires defining reference limits based on healthy controls, which for NfL must account for its age-dependent increase [6], 11]. A leading approach is the comprehensive NfL reference database published in 2022 [7] with NfL Z scores and percentile values, allowing the quantification of deviations from healthy levels while adjusting for age and BMI. However, the application of this dataset to sNfL measurements acquired using one of these new platforms demands the demonstration of a high correlation between platforms and of a comparable analytic performance. This is also relevant for the adaptation and harmonization of platform-specific cut-offs, such as the recently published threshold for prognosticating disease activity – defined by the development of new T2 lesions over 2 years – established during the clinical validation of the Atellica IM NfL assay [12]. Therefore, the objective of our multi-center study was to compare three different platforms and assays for the measurement of sNfL concentration (NF-light Advantage V2/Plus by Quanterix, Lumipulse G NfL blood by FujiRebio, and the Elecsys NfL assay by Roche Diagnostics) to analyze analytical correlation and comparability of differentiation between relapse and remission stages in MS.

Materials and methods

Samples

Our study encompassed three cohorts of participants (Figure 1): Healthy blood donors (n=303) were sourced from a blood bank affiliated with the Institute of Transfusion Medicine at the University Hospital Schleswig-Holstein, Campus Kiel. All participants were examined before serum sampling by physicians and certified to be without known neurodegenerative, neoplastic, neuroinflammatory, or neuroinfectious conditions. Demographic information, including age and gender, was recorded, and samples were pseudonymized immediately following the collection.

Figure 1:

Study cohorts. RRMS, relapsing remitting multiple sclerosis; * paired samples of patients in relapse and remission; ** unpaired samples balanced 1:1 in relapse and remission.

A second cohort (high to low samples) was introduced to enrich the range of sNfL levels in the dataset beyond the mostly low levels observed in healthy individuals and treated MS patients. We included 35 samples from patients with MS with known sNfL concentrations covering a broad range of sNfL levels (1.00–232.00 ng/L on Simoa technology-based platform, Quanterix). These were provided by the MS Center, University Hospital Basel, Switzerland.

The third cohort of patients with relapsing remitting MS (RRMS) was included and balanced 1:1 for relapse and remission to compare the analytical validity of differentiating between relapse and remission across different platforms. For remission, a minimum of 30 days since the last relapse was required. Samples from RRMS patients were procured from Basel and the Department of Neurology, Medical University of Graz, Graz, Austria. Participants were required to have a confirmed diagnosis of RRMS, alongside documented information regarding their current disease status (relapse vs. remission). The Basel sample set (n=86) consisted of paired samples from 43 patients at two time points during relapse and remission, whereas the Graz sample set (n=60) provided unpaired samples with a balanced 1:1 representation of 30 patients during relapse and 30 during remission. Together, these 146 samples were designated as the RRMS patient cohort, with a subdivision into remission and relapse.

All samples were stored at −80 °C prior to analysis and were measured immediately after thawing.

Results

We analyzed a total of 484 serum samples from healthy donors (n=303), patients with RRMS (n=146), and low to high samples (n=35) (Figure 1). Participant demographic characteristics are summarized in Table 2.

Cross-platform comparison and outlier analysis

We initially assessed the correlation among the three platforms, combining all samples to ensure a representative spectrum of sNfL values. We observed strong correlations across all three platforms (Pearson’s r=0.71–0.91, p<0.001; Spearman’s r=0.86–0.89, p<0.001; Figure 2A–C and Table 1). Based on Passing-Bablok regression analysis, we identified 14/484 (2.90 %) significantly discrepant samples on at least one platform (outliers; Figure 2D, Supplementary Table 1). These outliers were non-systematic, as neither their direction nor the affected platform exhibited any consistent and significant bias and accrued across all study cohorts, including healthy donors, MS patients, and high to low samples. A repeated measurement of these outliers on the respective platform confirmed these findings, with a coefficient of variation below 10 % in all repeated samples. Of note, in 13/14 of these samples, two of three platforms yielded highly comparable results and only 1/14 samples resulted in largely deviating results from all three platforms (Supplementary Table 1). After exclusion of these outliers, an even higher correlation was observed across all three platforms (Figure 2E, Pearson’s r=0.98–0.99, p<0.001; Spearman’s r=0.90–0.92, p<0.001; Table 1).

Figure 2:

Cross-platform comparison of sNfL values with correlation and outlier analysis. (A–C) Correlation of sNfL values (ng/L) of healthy and MS patients combined between platforms. (A) NF-Light Advantage Adv V2 Plus, Quanterix vs. Elecsys NfL, Roche Diagnostics. (B) NF-Light Advantage Adv V2 Plus, Quanterix vs. Lumipulse G NfL blood, FujiRebio; (C) Lumipulse G NfL blood, FujiRebio vs. Elecsys NfL, Roche Diagnostics. (D) Outliers (marked red) on at least one platform (perpendicular distance from the Passing-Bablok regression line >4 standard deviations). (E) Correlation of all samples following application of conversion formulas. Green triangles Lumipulse G NfL blood, FujiRebio (Pearson correlation r=0.98, p<0.001 and Spearman correlation r=0.92, p<0.001) and purple circles Elecsys NfL, Roche Diagnostics (Pearson correlation r=0.99, p<0.001 and Spearman correlation r=0.91, p<0.001) vs. NF-Light Advantage Adv V2 Plus results on the x-axis.

Clinical use case: evaluation of disease activity in multiple sclerosis patients

Finally, we evaluated the clinical performance of all three platforms in discriminating between relapse and remission in a preselected cohort of patients with RRMS, encompassing equal numbers of patients with relapse and remission status. We compared the performance of (a) raw values obtained from all three analyzers (Supplementary Figure 2); (b) values adjusted using our conversion formulas (Figure 3A–C); and (c) age-adjusted Z scores calculated using our converted values (Figure 3D–F). Age distribution in MS patients is additionally shown in Supplementary Figure 3. Across all three platforms, significant differences were observed between relapse and remission MS patients. Of note, Z scores yielded larger effect sizes compared to absolute converted values (Cohen’s d: 0.84 vs. 0.49 for NF-light Advantage V2/Plus, 1.11 vs. 0.53 for Elecsys NfL, and 0.93 vs. 0.48 for Lumipulse G NfL blood, see Figure 3A–F) as well as absolute original values (see Supplementary Figure 2). Similarly, Z scores performed slightly better in ROC analysis, showing higher AUCs (0.72 vs. 0.69, 0.78 vs. 0.75, and 0.73 vs. 0.71, respectively; Figure 3G and H and Supplementary Table 2), although the differences were not statistically significant. Of note, all three platforms showed very comparable performance in discrimination between relapse and remission on this cohort with a significant – albeit small – advantage for the Elecsys platform (Supplementary Table 2).

Figure 3:

Analytical performance of all platforms to discriminate MS patients with relapse and remission. (A–C) Platform-converted (using our conversion formulas) absolute sNfL values in relapse and remission measured with (A) NF-Light Advantage Adv V2 Plus, Quanterix. (B) Elecsys NfL, Roche diagnostics. (C) Lumipulse G NfL blood, FujiRebio. (D–F) sNfL Z scores based on converted absolute values measured with (D) NF-Light Advantage Adv V2 Plus. (E) Elecsys NfL, Roche diagnostics. (F) Lumipulse G NfL blood, FujiRebio. (G–H) ROC analysis for discrimination between relapse and remission using (G) conversion-formula transformed absolute sNfL values and (H) Z scores based on these converted values measured with NF-Light Advantage Adv V2 Plus (red), Elecsys NfL, Roche diagnostics (green), and Lumipulse G NfL blood, FujiRebio (purple). AUC were not statistically different in G and H.

Discussion

Serum neurofilament light chain (sNfL) has rapidly evolved from a research biomarker to a candidate routine laboratory test for monitoring neuroaxonal injury [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27]. In this comparative study, we evaluated three fully automated research-use-only assays that are currently undergoing CE certification. Across 485 samples from healthy volunteers and patients with MS, Pearson (r>0.98) and Spearman (r>0.90) correlation coefficients demonstrated very high analytical concordance between the platforms, satisfying a key prerequisite for clinical interchangeability. Passing Bablok regression-derived conversion equations aligned the numerical scale of the emergent platforms with the legacy Quanterix Simoa assay, thereby permitting direct comparison with historical cohorts and facilitating meta-analyses of treatment effects. When sNfL values were transformed into age-adjusted Z scores after platform-specific normalization, performance in discrimination between relapse and remission remained comparable to raw concentrations. However, the effect size was larger, supporting recent recommendations [7] that laboratories report age-adjusted Z scores rather than absolute values when interpreting individual results. In our cohort, the Elecsys platform demonstrated a slightly better ability to distinguish between relapse and remission compared to the other two evaluated platforms, both when using absolute values and Z scores. However, since this observation is based on a single cohort and the study primarily focuses on the analytical performance of the compared platforms, it should be interpreted with caution and confirmed in independent studies to assess its reliability and generalizability.

In comparison to previous studies [13], 14], 28], we also observed a strong correlation among three different platforms used for measuring blood NfL concentrations. Notably, our cohort represents the largest dataset to date that compares various analytical platforms for blood NfL quantification in both healthy individuals and patients with neurological disorders. Furthermore, this study investigates the utility of age-adjusted Z scores as an additional approach for harmonizing NfL measurements across different platforms. In contrast, the recent French multicenter analytical study [28] primarily utilized artificially generated samples covering a measurement range comparable to our study, however did not include the Elecsys NfL assay. The study conducted by Mayo Clinic, while encompassing all three assays evaluated in our study, was limited by a smaller data set [13]. Despite differences in study design, the Passing-Bablok regression lines observed in our study (NfL=6.31 × [Elecsys NfL] + 2.33 ng/L; NfL=0.94 × [Lumipulse G NfL blood] + 1.32 ng/L) closely align with those reported by the Mayo Clinic (NfL=6.4 × [Elecsys NfL] − 1.30 ng/L; NfL=0.95 × [Lumipulse G NfL blood] − 0.56 ng/L) [13], indicating the robustness of Passing-Bablok regression as a basis for converting blood NfL values across different platforms. The referenced study also included the fourth and currently only CE-IVDR–certified assay, the Atellica IM Neurofilament Light Chain (NfL), for NfL quantification in MS patients. It demonstrated high correlation across all four platforms (r>0.95) with systematic bias, further supporting the generalizability of using Passing–Bablok regression–based conversion formulas. However, applying such cross-study conversions remains exploratory, as differences in cohorts, sample matrices (serum vs. plasma), and pre-analytical conditions may introduce additional shifts. We therefore encourage larger multicenter comparison studies that include the Siemens Atellica IM to enable direct four-way harmonization and the establishment of robust international reference ranges. This could be further expanded in the event that additional platforms seeking CE-IVDR certification are introduced. To date, none of the previous studies have evaluated the additional utility of age adjusted Z scores, which, as demonstrated in our analysis, provide significant added value in enhancing the harmonization of blood NfL measurements across different analytical platforms and increasing the discriminatory power by normalizing for age effects on NfL concentrations. We found no evidence that epitope differences between antibody pairs influenced clinical validity. Although the targeted epitope by antibody has previously been discussed as a potential factor influencing blood NfL concentrations and its biological half-life [29]. Correlation coefficients and relapse–remission effect sizes in patients with RRMS were virtually identical across all three tested platforms. This observation argues against clinically meaningful epitope-dependent bias at least within the concentration range typical of MS and suggests that the same biological signal is captured irrespective of platform.

A clinically relevant challenge emerged from the identification of platform-specific outliers: in our sample set approximately one in 35 samples deviated by more than four standard deviations on a single analyser yet remained concordant on the other two. These discrepancies persisted on repeat runs of the same platform but disappeared when the specimen was re-analysed on a different system, pointing to matrix-dependent interference potentially in the context of differences driven by epitope specificity of used antibodies or heterophilic antibodies. A pre-analytical influence is considered an unlikely explanation for the outliers, as the results remained consistent in repeated runs. The occurrence of outliers was also noted and addressed in a previous cross-platform comparison, in which 8 out of 122 samples (6 %) – those showing more than 150 % difference between the compared platforms – were identified as outliers and excluded from further analysis [13]. In the absence of a universally accepted gold standard for blood NfL quantification results should be interpreted within the appropriate clinical context, as the absence of a universally accepted gold standard for blood NfL quantification may result in inaccurate measurements on both platforms, potentially in either direction. Future studies incorporating dilution linearity, spike-and-recovery experiments and heterophilic-antibody blocking reagents are required to elucidate the underlying mechanism and to establish whether the problem extends to other neurological disorders such as amyotrophic lateral sclerosis or hypoxic encephalopathy, where sNfL concentrations are markedly higher [26], 27].

Several limitations merit acknowledgement. First, we did not include the Siemens ADVIA Centaur/Atellica or ProteinSimple Ella assays and we therefore cannot generalize our findings to those platforms. Second, differentiation between relapse and remission was assessed using retrospectively selected, equally balanced cohorts; optimal diagnostic thresholds for routine practice must therefore be derived from prospectively assembled unselected cohorts. Third, pre-analytical variables – including sample matrix, storage conditions, and freeze–thaw cycles – were tightly controlled in this study. Although sNfL remains stable for up to 7 days at room temperature and across multiple freeze–thaw cycles [30], these factors can still introduce variability in real-world settings. Laboratories implementing sNfL measurements should therefore incorporate these considerations into their quality-management protocols and performe internal validation of the sample transport and storage. Beyond NfL, several additional blood-based biomarkers – such as phosphorylated tau217, glial fibrillary acidic protein, and amyloid-beta peptides – are currently under investigation for use in routine diagnostics [31], [32], [33], [34]. These markers exhibit differing pre-analytical stability and varying concentrations between serum and plasma, which requires careful consideration when comparing cohorts or analytical platforms and when implementing these assays in routine laboratory practice. Finally, our data do not replace the need for large platform-specific reference datasets; they provide a practical bridge for harmonization of blood NfL measurements across different analytic platforms – an essential step toward the broader implementation of this biomarker in routine clinical practice.

In conclusion, sNfL measurements obtained with three emerging automated immunoassays are highly concordant and can be used interchangeably at the individual level after applying simple conversion equations. We recommend consistently reporting age-adjusted reference values and/or age-adjusted Z scores, as this would also help harmonize results across different platforms. Nevertheless, the observed rate of platform-specific outliers underscores the importance of cross-platform quality control, and continued harmonization efforts – including the development of an international calibrator and reference measurement procedure.