Biases and consistency of different assay methods for neurological biomarkers using Quanterix single-molecule-array technology: A secondary analysis method comparison study

Abstract

Breakthrough technologies such as the Single molecule array (Simoa) technology developed by Quanterix provide higher sensitivity, enabling measurement of central nervous system-abundant proteins in blood. Neurofilament light chain (NfL) and glial fibrillary protein (GFAP) are two proteins that have attracted considerable attention as biomarkers for many neurological conditions. Due to their performance and utility, these biomarkers are available in different Quanterix assays, including single-plex and multi-plex assay setups. Limited research has been conducted to evaluate how modifications in assay formulations may impact overall analytical performance and comparability. We recently established reference intervals (RI) for plasma NfL and GFAP measured in normative samples from the Canadian Health Measures Survey (CHMS) using the Quanterix Neurology 4-Plex E (N4PE) Advantage. The aim of the present study was to perform method comparisons to assess how well CHMS RIs generated on the N4PE assay translate to other assay formulations to facilitate generalizability and uptake. Seven independent comparisons were conducted, each using a total of 80 plasma samples from the CHMS. These samples were divided (n=40 each) based on the two N4PE lots previously used to generate RIs. The following Quanterix assay formulations were evaluated against the N4PE anchor assay: NfL Advantage PLUS (NfL+), GFAP Advantage PLUS (GFAP+), Neurology 2-Plex B Advantage (N2PB), Neurology 2-Plex B Advantage PLUS (N2PB+), Neurology 4-Plex B Advantage (N4PB), Neurology 4-Plex D PLUS (N4PD+), and Neurology 4-Plex E PLUS (N4PE+). Plasma biomarker concentrations were measured using the Quanterix Simoa HD-X analyzer, with assays run as per the manufacturer’s specifications. Assay comparisons were conducted using Spearman correlation and Bland-Altman analysis to assess bias between assays. Both NfL and GFAP concentrations were tightly correlated between assay formulations with rho > 0.9 and a P-value < 0.0001 for all assay crosses. However, the bias between formulations ranged from 0.5% to 42.1%, indicating that data correction may be required to harmonize data generated from different assay formulations. In summary, this study emphasizes the importance of data correction when using different assay methods to ensure data comparability across studies. Additionally, the study results support the applicability of RIs and suggest that when using RIs, corrections should be made based on the assay-specific bias. Future studies should further expand the sample range, particularly in high-concentration conditions such as acute neurological injury, to comprehensively evaluate the performance of different assay formulations.

1. Introduction

1.1. Clinical context

Neurological conditions are managed using clinical assessments that require patients to present with signs and symptoms.¹ Assessment of underlying pathology may involve neuroimaging that is expensive, associated with long wait times, and often requires patient travel, or invasive cerebrospinal fluid (CSF) analysis that requires specialist access. However, the ability to measure brain-derived proteins in blood has transformed the field of neurology. Two particularly prominent neurological blood-based biomarkers are neurofilament light chain (NfL) and glial fibrillary acidic protein (GFAP). NfL is an axonal intermediate filament protein that is released into blood following axonal damage, with elevated NfL levels observed in almost all chronic and acute neurological disorders.^2–5 GFAP is an intermediate filament protein found in mature astrocytes that can be used as an indicator of neuroinflammation when increased in blood.⁶ Both have utility across multiple conditions, and are of particular interest for traumatic brain injury (TBI). Several longitudinal observational studies in North America and Europe^7–9 have confirmed the value of NfL and GFAP across multiple contexts of use (CoUs) relevant to TBI. Assays for GFAP and ubiquitin C-terminal hydrolase L1 (UCH-L1) measured in tandem have been cleared by the US Food and Drug Administration (FDA) to aid in the diagnosis of mild TBI.^10,11 Alzheimer’s Disease (AD) is another important condition where plasma biomarkers are emerging as diagnostic tools. Although biomarkers such as amyloid beta and phosphorylated tau are considered the best diagnostic tools for AD, NfL and GFAP were included in the 2024 National Institute on Aging (NIA) and Alzheimer’s Association (AA) revised criteria for diagnosis and staging of AD,¹² where neurofilament light (NfL) was labelled as a marker of neurodegeneration, and glial fibrillary acidic protein (GFAP) was labelled as an inflammatory marker in the context of AD. Beyond AD, NfL has also shown high utility to differentiate frontotemporal dementia from other forms of dementia^13,14. The FDA has also granted NfL breakthrough device status as a prognostic aid to assess disease activity in multiple sclerosis (MS), representing an additional use case for NfL to assess efficacy of MS treatments.¹⁵

1.2. Quanterix single molecule array technology

The single molecule array (Simoa) technology developed by Quanterix (Quanterix Corporation, Billerica, MA, USA)¹⁶ was the first widely commercialized analytical methodology for blood-based neurology biomarkers that catalyzed a plethora of studies across a wide variety of neurological indications including neurodegeneration and neurotrauma.^{3–5,13,17–19} In the past ten years, there have been almost 1000 publications on blood-based biomarkers using Quanterix technology²⁰ with a large portion focused on neurological indications.

Simoa technology is utilized in automated analyzers including the HD-1 and HD-X from Quanterix.¹⁶ Simoa principles are based on using capture beads to bind analytes in fluid matrices such as plasma, serum, and cerebrospinal fluid, and isolating immunocomplexes in femtoliter-sized array wells.¹⁶ The array well volume accommodates only a single bead-analyte complex and provides a close-quarters environment for the assay enzymes to react with substrate that enables highly sensitive detection.¹⁶ If successfully bound to the bead-analyte complex, the detector antibody generates a unique fluorescent signal depending on the analyte detected.¹⁶ Assay kits are commercially available as single- or multiplex immunoassay kits, each measuring different analytes of interest in a variety of combinations. Importantly, the Quanterix platform has demonstrated robust analytical performance, characterized by high reproducibility, low intra- and inter-assay variability, and excellent sensitivity across repeated runs and operators. These features support its reliability for research applications, especially when precise quantification of low-abundance proteins is critical.

As NfL and GFAP are biomarkers of interest for multiple CoU across a variety of neurological disorders, these analytes are available in multiple Quanterix assay formats, including single-plex assays measuring each marker individually, and in multi-plex formats that include one to three additional analytes (Supplemental table 1). Importantly, Quanterix Simoa assay formulations have changed over time, progressing from the “Discovery” to “Advantage” and now to “Advantage PLUS” format. “Advantage PLUS” assays are designed to improve longitudinal data precision and enhance suitability for large-scale research studies.²¹ According to the manufacturer, newer formulations are also characterized by improved lot-to-lot consistency, thereby supporting greater analytical reliability and reproducibility.²¹

1.3. NfL and GFAP reference intervals (RI)

We recently established RI for plasma NfL and GFAP using n=900 plasma specimens collected in the Canadian Health Measures Survey (CHMS) from age 3–79 years. These measurements were performed using the Quanterix Neurology 4-Plex E Advantage (N4PE) assay.²² CHMS is a national cross-sectional study representative of the Canadian population.²³ RIs provide a reference range based on the 5^th to 95^th percentile of a normative population and can serve as a benchmark comparator for age-adjusted normative levels. RI can facilitate interpretation of test results especially under conditions where individual results are compared to external reference or control groups.

1.4. Purpose and Hypothesis

Building on our prior findings on the variability of NfL and GFAP results across instruments and N4PE lots,²² the purpose of this secondary analysis method comparison study was to evaluate Quanterix assay agreement and potential bias across distinct assay formulations, with a special interest on the applicability of CHMS-derived RI—originally established using the N4PE assay— to other assay formulations. We hypothesized that NfL and GFAP results are translatable across different assay formats.

2. Methods

2.1. Ethical approval

This study was approved by The University of British Columbia Clinical Research Ethics Board, Vancouver, Canada (H19–01445, approved May 23, 2019). CHMS had previously provided de-identified plasma specimens for secondary analysis upon their approval of our study proposal. The parent CHMS study was launched in 2007²⁴ upon approval by Health Canada and the Public Health Agency of Canada Research Ethics Board, Ottawa, Canada.²⁵ Informed consent for CHMS participants was conducted by Statistics Canada.

2.2. Participant and specimen selection

CHMS is a national epidemiological survey run by Statistics Canada focused on understanding the health of Canadians.²³ CHMS is a representative sample of the Canadian population from 3–79 years of age including major ethnic groups, designed to collect data on obesity, hypertension, cardiovascular disease, infectious diseases, and environmental toxins.²⁶ The CHMS began in 2007 with new cycles every 2 years using a cross-sectional design and data collection through an interview and visit to a mobile examination centre where venous blood was collected in EDTA vacutainers and centrifuged at 8° Celsius for 15 minutes at 3,400 revolutions per minute. Plasma was separated and stored directly into a −80° Celsius freezer within a maximum of 4 hours from collection. Extensive data exist on the CHMS cohort including several published RI for hematological, chemistry, immunological, and environmental toxin tests.^26,27 Further data are available through linkage to provincial health records by Statistics Canada. Although CHMS does not collect data on neurological health other than inferences from questions about drug and medication use and chronic health conditions, their consent process excludes participants with dementia. CHMS inclusion criteria are people living in the ten Canadian provinces, between 1–79 years old, and residents of privately occupied dwellings. Exclusions are: persons living in the three Canadian territories, residents of certain remote regions, people living on reserves or other Indigenous settlements, institutionalized individuals, and full-time members of the Canadian Forces. Other exclusion criteria may be present for specific components of the CHMS.²⁴

A total of n=5,200 de-identified plasma specimens from the CHMS Biobank (Cycle 5, collected in 2016–2017) were received at University of British Columbia in 2021/2022 and n=900 of these specimens, selected by CHMS methodologists, were used to generate RIs for NfL and GFAP from age 3–79 years.²² In our published²² and ongoing RI studies,²² a total of n=2,500 CHMS samples were analyzed using two separate lots of the Quanterix (Quanterix Corporation, Billerica, MA, USA) N4PE assay: lot number 503105 (N4PE Lot 1) and lot number 503811 (N4PE Lot 2). From these specimens, n=560 (n=280 from each lot) were selected for this secondary method comparison analysis. All of the selected plasma samples fell within the assay dynamic range for both NfL and GFAP. With many laboratory methods, measurement accuracy may vary across different concentrations.²⁸ In order to measure bias and agreement across as much as the assay dynamic range as possible as recommended in Clinical Laboratory Standards Institute (CLSI) guidelines for method comparison studies,²⁹ samples were selected spanning the full concentration range of the n=2,500 CHMS samples for each of the assay crosses performed. To select plasma samples to be used in each assay cross, previously analyzed specimens were first separated based on the N4PE lot used, then ranked in descending order of GFAP concentration. In the middle of the GFAP concentration range, samples were selected with an even distribution in the middle of the biomarker concentration range. More samples were a priori selected at both the high and low ends of the concentration range where larger variability and loss of linearity occurs,²⁸ GFAP concentration was used as the selection variable as samples with high GFAP concentration were limited in number. All available samples with GFAP concentrations above 195 pg/mL (n=47) were included.

2.3. Biomarker Analysis

The primary outcome measure of this method comparison study is assay agreement and assessment of bias for NfL and GFAP across multiple assay lots and assay formulations, as described below. We performed 7 separate crosses of n=80 samples each to quantify plasma NfL and GFAP levels across a variety of Quanterix assay lots and assay formulations. For each cross, samples were divided (n=40 each) based on the previous two lots used to analyze CHMS samples using the N4PE assay (catalogue number 103760), which was used as the reference, or anchor, assay formulation in this study. A sample size of n=40 per group was chosen in accordance with CLSI guidelines for internal method comparison studies.²⁹ The assay formulations evaluated in this study (Supplemental Table 1) were Quanterix singleplex assays NfL Advantage PLUS (NfL+; catalogue number 104364, lot number 503994) and GFAP Advantage PLUS (GFAP+; catalogue number 104619, lot number 504093), as well as multiplex assays Neurology 2-Plex B Advantage (N2PB; catalogue number 103520, lot number 504071), Neurology 2-Plex B Advantage PLUS (N2PB+; catalogue number 104670, lot number 504330), Neurology 4-Plex B Advantage (N4PB; catalogue number 103345, lot number 504005), Neurology 4-Plex D PLUS (N4PD+; catalogue number 104822, lot number 504300), and Neurology 4-Plex E PLUS (N4PE+; catalogue number 104465, lot number 504182). NfL and GFAP concentrations were measured using Quanterix Simoa HD-X Analyzer (Quanterix Corporation, Billerica, MA, USA) run using the recommended 4x dilution protocol following the manufacturer’s protocol. Each assay included an 8 or 9-point calibrator curve, two internal kit controls, and two plasma controls. All specimens were loaded onto a 96-well plate and assayed in duplicate. The mean value was reported as the result. For both analytes, quality control information was documented across all runs, including absolute average error and recovery for calibrators and controls, as well as the average intra-plate CV for controls and plasma samples (Supplemental table 2). All controls fell within approved ranges stated on the certificates of analysis included with the assay kits, except the kit control 1 NfL for the N4PB kit; all other quality control metrics for this run indicated an isolated error.

2.4. Statistical Analysis

In clinical laboratories, commonly used statistical methods for assessing agreement between measurement procedures involves correlation studies to determine the relationship between variables, and Bland-Altman plots to determine comparability.^30,31 Following CLSI guidelines for method comparison studies,²⁹ we used both correlation plots and Bland-Altman analysis to evaluate the agreement between each assay and the reference N4PE assay.²⁹ For each sample, the mean of two replicate measurements was used to determine biomarker concentration. To assess associations between biomarker concentrations within each assay cross, Spearman correlation was performed to report rho and simple linear regression was used to determine the line of best fit for correlation plots. To assess difference between assays, Bland-Altman analysis was performed to report the average biomarker concentration bias, standard deviation, and the 95% limits of agreement across assays. Extreme outliers were identified by applying the Tukey method, with a fence of 3, to the percent bias of each sample within each cross. Within each cross, bounds were first defined by determining the interquartile range (IQR) of sample biases. The IQRs were multiplied by 3. The resulting number was subtracted from the 1^st quartile to define the lower bound and added to the 3^rd quartile to define the upper bound. Data points outside these bounds for either NfL, GFAP, or both were identified as extreme outliers and were excluded from correlation and Bland-Altman analysis. GraphPad Prism software (Version 10.2.2) (GraphPad Software, Boston, MA, USA) was used to analyze data and generate correlation and difference plots.

3. Results

3.1. Sample concentration range

N=80 CHMS samples were selected for each assay cross, consisting of n=40 samples previously run on each of two N4PE reference lots. All selected samples had biomarker concentrations that fell within the dynamic range of the N4PE assay. For GFAP, the dynamic range of the N4PE assay is 0–40,000 pg/mL, whereas the GFAP concentration range of CHMS samples was 5.8–467.7 pg/mL (Supplemental figure 1). For NfL, the dynamic range of the N4PE assay is 0–2,000 pg/mL, whereas the concentration range of CHMS samples was 1.34–132.0 pg/mL (Supplemental figure 2).

3.2. Correlation across assay and formulation

For both NfL and GFAP, concentrations across assay formulations correlated strongly with the reference N4PE lots (Figures 1, 2), with Spearman’s correlation p-values <0.001 for all crosses. For NfL, the average rho value was 0.97. Simple linear regression analysis of the correlation plots revealed the following beta values averaged from both lots of N4PE: NfL+ β=0.65, N2PB β=0.64, N2PB+ β=0.88, N4PB β=0.74, N4PD+ β=0.82, and N4PE+ β=0.98. Individual correlations and regression equations for each cross are summarized in Table 1. For GFAP, the average rho value was 0.98. Simple linear regression analysis of the correlation plots revealed the following beta values averaged from both lots of N4PE: GFAP+ β=0.88, N2PB β=1.38, N2PB+ β=1.03, N4PB β=1.33, N4PD+ β=1.08, and N4PE+ β=0.94. Individual correlations and regression equations for each cross are summarized in Table 2.

Figure 1. Plasma NfL correlation plots of crossed assay formulations.

Graphs display plasma NfL concentrations derived from N4PE assays (lot 1: panels A, C, E, G, I, K and lot 2: panels B, D, F, H, J, L) from reference CHMS cohort data on the x-axis versus the comparator assay on the y-axis. Comparator assays include: NfL+ (A, B), N4PE+ (C, D), N2PB (E, F), N2PB+ (G, H), N4PB (I, J) and N4PD+ (K, L). Shaded area represents 95% confidence intervals. Coefficients for simple linear regression fitted to the data is shown in Table 1. Abbreviations used: Neurofilament light chain (NfL); Neurology 2-Plex B Advantage assay (N2PB); Neurology 2-Plex B Advantage PLUS assay (N2PB+); Neurology 4-Plex B Advantage assay (N4PB); Neurology 4-Plex D Advantage PLUS assay (N4PD+); Neurology 4-Plex E Advantage assay (N4PE); Neurology 4-Plex E Advantage PLUS assay (N4PE+); NfL Advantage PLUS assay (NfL+)

Figure 2. Plasma GFAP correlation plots of crossed assay formulations.

Graphs display plasma GFAP concentrations derived from N4PE assays (lot 1: panels A, C, E, G, I, K and lot 2: panels B, D, F, H, J, L) from reference CHMS cohort data on the x-axis versus the comparator assay on the y-axis. Comparator assays include: GFAP+ (A, B), N4PE+ (C, D), N2PB (E, F), N2PB+ (G, H), N4PB (I, J) and N4PD+ (K, L). Shaded area represents 95% confidence intervals. Coefficients for simple linear regression fitted to the data is shown in Table 2. Abbreviations used: Glial fibrillary acidic protein (GFAP); GFAP Advantage PLUS assay (GFAP+); Neurology 2-Plex B Advantage assay (N2PB); Neurology 2-Plex B Advantage PLUS assay (N2PB+); Neurology 4-Plex B Advantage assay (N4PB); Neurology 4-Plex D Advantage PLUS assay (N4PD+); Neurology 4-Plex E Advantage assay (N4PE); Neurology 4-Plex E Advantage PLUS assay (N4PE+)

Table 1.

Summary of NfL results for assay crosses^a,b.

Crossed to	N4PE Lot	Extreme outliersc (n)	Spearman’s rho	Simple linear regression		Bland-Altman
				α	β	Bias (%)	SD	95% LoA
NfL+	1	1	0.98	0.21	0.66	−38.6	14.9	−67.7 to −9.4
	2	0	0.99	0.19	0.64	−42.1	11.7	−64.9 to −19.2
N2PB	1	1	0.94	0.91	0.62	−28.6	15.0	−58.1 to 0.8
	2	2	0.99	0.05	0.66	−39.6	7.3	−58.8 to −25.3
N2PB+	1	1	0.95	1.05	0.90	−1.4	19.7	−40.0 to 37.1
	2	3	0.98	0.86	0.86	−7.6	13.5	−34.0 to 18.8
N4PB	1	3	0.96	1.25	0.80	−3.7	14.8	−32.6 to 25.3
	2	1	0.98	1.50	0.70	−16.2	15.5	−46.5 to 14.1
N4PD+	1	3	0.92	0.98	0.76	−17.0	23.7	−63.4 to 29.4
	2	2	0.98	−0.93	0.88	−23.7	19.4	−61.6 to 14.3
N4PE+	1	0	0.98	−0.10	1.00	−0.5	16.7	−33.2 to 32.1
	2	1	0.98	−0.73	0.95	−10.8	11.6	−33.5 to 11.9

^aAbbreviations used: Limits of agreement (LoA); neurofilament light chain (NfL); Neurology 2-Plex B Advantage (N2PB); Neurology 2-Plex B Advantage PLUS (N2PB+); Neurology 4-Plex B Advantage (N4PB); Neurology 4-Plex D Advantage PLUS (N4PD+); Neurology 4-Plex E Advantage (N4PE); Neurology 4-Plex E Advantage PLUS (N4PE+); NfL Advantage PLUS (NfL+); standard deviation (SD)

^bResults for each assay cross generated from a starting sample size of n=40, followed by exclusion of extreme outliers prior to data analysis.

^cExtreme outliers were identified by applying the Tukey method with a k+3 fence.

Table 2.

Summary of GFAP results for assay crosses^a,b.

Crossed to	N4PE Lot	Extreme outliersc (n)	Spearman’s rho	Simple linear regression		Bland-Altman
				A	β	Bias (%)	SD	95% LoA
GFAP+	1	1	0.97	2.16	0.84	−13.7	15.5	−44.0 to 16.6
	2	2	0.97	1.36	0.92	−8.7	10.6	−29.5 to 12.1
N2PB	1	1	0.99	−7.50	1.37	23.2	11.5	0.6 to 45.7
	2	2	0.99	−3.16	1.38	29.0	14.0	1.5 to 56.5
N2PB+	1	1	0.98	0.81	1.06	5.9	15.2	−23.8 to 35.7
	2	3	0.96	4.39	0.99	5.1	14.7	−23.7 to 33.9
N4PB	1	3	0.98	0.12	1.37	30.9	11.9	7.6 to 54.3
	2	1	0.98	11.12	1.30	34.9	14.8	6.0 to 63.9
N4PD+	1	3	0.97	0.18	1.07	6.6	21.0	−34.6 to 47.8
	2	2	0.97	3.57	1.08	11.2	12.6	−13.5 to 25.9
N4PE+	1	0	0.99	0.87	0.86	−13.7	9.0	−31.3 to 3.9
	2	1	0.97	4.08	0.83	−12.8	15.7	−43.7 to 18.0

^aAbbreviations used: Glial fibrillary acidic protein (GFAP); GFAP Advantage PLUS assay (GFAP+); limits of agreement (LoA); Neurology 2-Plex B Advantage assay (N2PB); Neurology 2-Plex B Advantage PLUS assay (N2PB+); Neurology 4-Plex B Advantage assay (N4PB); Neurology 4-Plex D Advantage PLUS assay (N4PD+); Neurology 4-Plex E Advantage assay (N4PE); Neurology 4-Plex E Advantage PLUS assay (N4PE+); standard deviation (SD)

^bResults for each assay cross generated from a starting sample size of n=40, followed by exclusion of extreme outliers prior to data analysis.

^cExtreme outliers were identified by applying the Tukey method with a k+3 fence.

3.3. Assay bias

For NfL, Bland-Altman plots showed consistent bias across the concentration range and absolute biases differed in magnitude between assay crosses. Notably, the N4PE assay consistently produced higher raw NfL concentrations than all of the other assays examined in this study (Figure 3, Table 1). The N4PE+ assay crossed with N4PE lot 1 had the lowest absolute bias of 0.5% (Figure 3 panel C). However, the average absolute bias for both N4PE lot 1 and lot 2 was lowest for the N2PB+ assay at 4.5% (Figure 3 panel G, H). Some assay crosses revealed large amounts of bias, particularly for the NfL+ lot 2 cross with an absolute bias of 42.1% (Figure 3 panel B). The NfL+ assay also had the highest average absolute bias for both lots with a value of 40.7%. All assays generated NfL values lower than the values from the reference N4PE lots, with an average bias of −19.1% across all crosses. The average standard deviation (SD) of the bias was 15.3, ranging from 7.3 to 23.7. N4PE lot 1 consistently exhibited a higher bias than N4PE lot 2 across all assay comparisons, with bias differences ranging from 3.5% to 10.3% and an average of 8.4%. Individual biases with limits of agreement for each cross are presented in Table 1.

Figure 3. Bland-Altman plots of NfL results between crossed assay formulation.

Graphs display average NfL concentration plotted on the x-axis and the average bias of the comparator assay relative to the reference N4PE assay (lot 1: panels A, C, E, G, I, K and lot 2: panels B, D, F, H, J, L) plotted on the y-axis. Comparator assays include: NfL+ (A, B), N4PE+ (C, D), N2PB (E, F), N2PB+ (G, H), N4PB (I, J) and N4PD+ (K, L). The coloured line represents the average bias for the assay cross and the shaded area represents 95% limits of agreement. Results of the Bland-Altman analysis are summarized in Table 1. Abbreviations used: Neurofilament light chain (NfL); Neurology 2-Plex B Advantage assay (N2PB); Neurology 2-Plex B Advantage PLUS assay (N2PB+); Neurology 4-Plex B Advantage assay (N4PB); Neurology 4-Plex D Advantage PLUS assay (N4PD+); Neurology 4-Plex E Advantage assay (N4PE); Neurology 4-Plex E Advantage PLUS assay (N4PE+); NfL Advantage PLUS assay (NfL+)

For GFAP, Bland-Altman plots also revealed consistent bias across the concentration range with differences in absolute bias across different assays (Figure 4). The average absolute bias across all the Bland-Altman plots was 16.3%, and the average SD was 13.9, ranging from 9.0 to 21.0. Only GFAP+ and N4PE+ kits produced GFAP values lower than the reference N4PE lots, with an average bias of −12.2%. N2PB, N2PB+, N4PB, and NB4PD+ kits all produced GFAP values that were higher than the reference N4PE lots, with an average bias of 18.4%. The overall average bias for all assay crosses was 8.2%. N4PE lot 1 exhibited a lower bias than N4PE lot 2 across most assay comparisons. The bias differences between the two lots ranged from −5.8 to 0.9% and had an average of −3.2%. The exception to this was N2PB+, for which N4PE lot 1 had a bias 0.9% higher than N4PE lot 2. Individual biases with limits of agreement for each cross are presented in Table 2. Additionally, these Bland-Altman analyses show that differences do not vary with the magnitude of the data, especially for GFAP. However, for NfL N4PE+ lot 1 and N2PB+ lot 1, and GFAP N4PD+ lot 1, the variation of differences decreases proportionally to the magnitude of the data, suggesting that the difference between the methods varies proportionally with the overall magnitude of the measurement.

Figure 4. Bland-Altman plots of GFAP results between crossed assay formulation.

Graphs display average GFAP concentration plotted on the x-axis and the average bias of the comparator assay relative to the reference N4PE assay (lot 1: panels A, C, E, G, I, K and lot 2: panels B, D, F, H, J, L) plotted on the y-axis. Comparator assays include: GFAP+ (A, B), N4PE+ (C, D), N2PB (E, F), N2PB+ (G, H), N4PB (I, J) and N4PD+ (K, L). The coloured line represents the average bias for the assay cross and the shaded area represents 95% limits of agreement. Results of the Bland-Altman analysis are summarized in Table 2. Abbreviations used: Glial fibrillary acidic protein (GFAP); GFAP Advantage PLUS assay (GFAP+); Neurology 2-Plex B Advantage assay (N2PB); Neurology 2-Plex B Advantage PLUS assay (N2PB+); Neurology 4-Plex B Advantage assay (N4PB); Neurology 4-Plex D Advantage PLUS assay (N4PD+); Neurology 4-Plex E Advantage assay (N4PE); Neurology 4-Plex E Advantage PLUS assay (N4PE+)

4. Discussion

4.1. Contextualizing study results into the literature

The Quanterix Simoa technology has led to extraordinary growth in studies on blood-based biomarkers with NfL and GFAP being amongst the most widely studied analytes. As Quanterix offers singleplex NfL and GFAP assays as well as a variety of multiplexed assays, this study was designed to assess potential bias between different Quanterix assay formulations for NfL and GFAP, including evaluating the newly released “Advantage PLUS” formats and determining how these compare to the discontinued “Advantage” assays This knowledge will be important to determine if data correction or harmonization may be required to facilitate data comparison across studies that used different assay formulations or assay lots.

Comparisons of measurement procedures are commonly done to quantify bias between two or more laboratory tests. Initial comparisons made by the manufacturer are performed to establish and verify claims, which are then published. In a clinical laboratory, further comparisons are often performed to independently quantify bias of a candidate measurement procedure prior to its introduction for usage in the laboratory. For the latter scenario, CLSI guidelines recommend comparing a sample size of at least n=40 samples across methods and comparing methods using difference plots or regression techniques.²⁹

Many studies spanning various states of neurological health have investigated how Simoa assays compare to other immunoassays on different platforms including Siemens Healthineers Atellica,³² Fujirebio Lumipulse,^32–36 Roche Cobas Elecsys,^32,35 SimplePlex Ella,^33,35–37 and Meso Scale Discovery (MSD).^34,38 Many of these studies focused on NfL measured in blood using the Quanterix single-plex Nf-Light assay or the updated Nf-Light V2 assay formulation. These studies all found strong correlations (R > 0.8) between Simoa and other platforms; however, significant difference in concentration values were noted for several platforms.^32–38 The Elecsys and Atellica assays produced the lower raw values, with the Elecsys having the and greatest bias around 85% lower than Simoa values.^32,35 Other assays were found to produce higher raw values compared to Simoa assays, with the MSD assay having the largest increase.^34,38 Across these studies, the Fujirebio Lumipulse had the closest approximation to Simoa NfL values, with a bias ranging from 6–20%.^32–36 One study comparing Simoa GFAP concentrations in plasma using the single-plex Quanterix GFAP Discovery kit had similar findings of strong correlations (R > 0.8) across all platforms along with much larger magnitude of average bias ranging from 107–186%.³⁹ For GFAP, the Abbott Alinity and Ella platforms reported lower concentrations compared to Simoa, while the MSD platform also reported higher values.³⁹ These large differences in values indicate that results from different platforms are not interchangeable, emphasizing the need for harmonization strategies procedures in order to be able to compare datasets produced using different analytical platforms.

Despite Quanterix offering multiple Simoa assay formats for measuring NfL and GFAP, no studies have been conducted to compare the different assay kits available within the Quanterix platform. Moreover, there have been no studies conducted to evaluate the updated “Advantage PLUS” assays which were released to replace the discontinued “Advantage” assays. This comparison will be essential to determine how the newer “Advantage PLUS” format compares to the Simoa assays used in previous studies.

In this secondary analysis method comparison study, we crossed n=560 plasma specimens previously analyzed with two lots of the N4PE assay (n=280 each) onto a total of 7 assay formulations (n=40 each), including single-plex NfL+ and GFAP+ assays as well as N2PB Advantage and Advantage PLUS, N4PB Advantage, N4PD Advantage PLUS, and N4PE Advantage PLUS multiplex assays (Supplemental table 1). For both NfL and GFAP, concentrations tightly correlated between assay formulations with rho>0.9 and a p-value <0.0001 for all assay crosses. However, bias between some formulations ranged from an absolute % bias of 0.5–42.1%, indicating that data correction may be required to harmonize data generated from different assay formulations. These results also underscore the importance of reporting assay lot and formulation across biomarker studies.

4.2. Assay Agreement

In general, results from 7 different assay formulations measuring NfL and GFAP correlate strongly with results from the N4PE assay. Correlation plots show tight regressions with narrow 95% confidence intervals (CI), and beta values close to 1. This demonstrates that, despite bias between assay formulations, the overall ranking and relative differences between analytes remain consistent across assays. This indicates that results generally translate well across different assays. This is important to establish, as if bias is observed between assays, harmonization will be possible using regression models. This would facilitate data comparison across studies using different assay formulations and allow for greater reproducibility, which are critical aspects to allow these biomarkers to move towards clinical use. As the results have high agreement, harmonization can allow for clinical interpretation of biomarker measurements using different assays.

4.3. Assay Bias

The tested assay formulations exhibited an average absolute bias ranging from 0.5% to 42.1% when compared to previous N4PE results. While Quanterix does not provide official guidelines on when harmonization between assay formulations is necessary, higher bias indicates that data harmonization may be required to interpret data across studies that used different assay formulations to quantify NfL and GFAP.

For NfL, N2PB+, followed by N4PE+, gave the closest approximation of N4PE measurements, whereas NfL+ and N2PB results had a higher bias between assays, resulting in lower concentrations. For GFAP, GFAP+, N2PB+, N4PD+, and N4PE+ assays all had absolute bias under 14%, whereas N2PB and N4PB produced higher GFAP concentrations with a greater bias compared to N4PE. Notably, despite being an updated version of the reference N4PE assay, the N4PE+ assay did not yield the closest approximation to biomarker measurements. This suggests that the new Advantage PLUS versions do not necessarily produce results directly comparable to their original counterparts. A similar trend was observed with the N2PB+ assay, which differed from the N2PB assay by an absolute average bias of approximately 20% for both NfL and GFAP.

Comparison across two N4PE lots showed high consistency, with differences rarely exceeding 5%, and are thus well within the manufacturer expected lot-to-lot bias. In this study, N4PE lot 1 consistently provided higher values for NfL and lower values for GFAP than lot 2, consistent with our previous publication of NfL and GFAP RIs.²² The consistency of results across lots importantly speaks to the robustness and translatability of these findings. Therefore, the assays tested here are robust to lot-to-lot differences, suggesting that harmonization may not be required to compare results across studies using the same assay formulation but different lots.

Assay formulations are designed to meet specific research needs, whether by grouping different biomarker combinations or optimizing dynamic ranges. The Quanterix assay formulations can also be adjusted to modify signal strength, which in turn affects the assay’s lower limits of detection. For example, the N4PB assay was specifically developed to generate a lower signal for GFAP, making it more suitable for measuring samples with high GFAP concentrations⁴⁰. Variations in dynamic range can also affect assay sensitivity for specific analytes, which may explain the observed differences in NfL and GFAP measurements. Our results for NfL demonstrate that the reference N4PE assay produced the highest values, with all other tested assay formulations showing a negative average bias in comparison. Conversely, N4PE yielded lower GFAP values compared to certain formulations, particularly N2PB, N2PB+, and N4PB. These findings highlight that differences in assay formulations can influence individual analytes independently rather than uniformly across biomarkers. Consequently, careful evaluation of assay-specific biases is essential to ensure accurate biomarker interpretation and enable meaningful comparisons across studies.

From a research perspective, these findings emphasize the need for transparency in study methods, particularly in clearly identifying the assay used, since using different assays can introduce unwanted bias and limit reproducibility. They also highlight the importance of applying harmonization techniques when pooling biomarker data across different assay formulations. The large bias we found for some assay formulations is of concern when considering the clinical utility of these biomarkers, as standardized harmonization techniques and reference assays will need to be established to allow for meaningful clinical interpretations of results.

4.4. Translatability of Canadian Reference Intervals

Our group recently generated RIs for NfL and GFAP across age 3–79 years²² to enable researchers to appreciate the marked effect of age on NfL and GFAP concentrations in the Canadian population. As RIs enable a test to be interpreted as normal or abnormal, RI generation is an important step in the progression of biomarker assays from research use only toward clinical use. The findings presented in this study highlight the importance of considering the effects of assay formulation on biomarker interpretation, ideally including internal studies to translate results across assays prior to data interpretation. Our results suggest that N2PB+, followed by N4PE+, deliver results closest to the N4PE assay used to generate RIs, as N2PB+ and N4PE+ demonstrated the lowest average absolute bias relative to N4PE. Therefore, these assays are less likely to require the use of a correction factor to compare data back to these RIs. It would be recommended that when using RIs generated from studies to interpret results, investigation of the bias between the specific assays is warranted.

We demonstrate here that bias between assays is consistent across both N4PE lots, allowing researchers to conduct their own transference or harmonization studies using our RIs even without access to the N4PE reference lot utilized. For assays with a known smaller bias, such as N2PB+ and N4PE+, transference studies can conduct following guidelines detailed by the CLSI guideline EP28-A3c.⁴¹ This will show how an independent control group produces expected normative results when analyzed on a different assay. For assays with a larger known bias, harmonization studies can be done to create correction factors following guidelines detailed by the CLSI guideline EP09-A3.²⁹ Here, samples analyzed on both assay formats are used to generate a regression model to extract a correction formula to harmonize data.

4.5. Limitations

This study has several limitations. First, CLSI guidelines recommend using as broad a measurement interval as possible.²⁹ The dynamic range used in this study was limited as the CHMS cohort included normative samples. Thus, NfL and GFAP concentrations of the samples covered only a small portion of the dynamic range of the assay kits. Agreement across assays may differ at higher concentrations, for example, when measured after a neurological injury. Future studies should incorporate samples with elevated concentrations of NFL and GFAP to determine if bias remains consistent across the full dynamic range of the assays. Additionally, this study focused on plasma, which may limit the generalizability of these findings to other study designs that use different biological matrices, such as serum or cerebrospinal fluid, which are commonly used to measure NfL and GFAP.^2,6

Our study is also limited to assay comparisons between N4PE and 6 other assay formulations that include NfL and GFAP, without directly comparing the alternative assays against each other. This means that while we understand how other assays deviated from N4PE, we did not evaluate how they might differ from one another. Additional comparisons will provide a more comprehensive understanding of potential assay biases and determine whether the absolute bias between assays varies depending on which assay is used as the reference or anchor assay.

5. Conclusion

Our findings revealed substantial bias across some Quanterix Simoa assay formulations measuring NfL and GFAP in normative plasma samples, while bias across the two N4PE lots used to generate CHMS RIs was minimal. Tight correlations were observed between assays, suggesting that different formulations generally align with established CHMS RIs. Therefore, correction factors can be employed to achieve data harmonization when substantial inter-assay bias is present.

Data availability

Data will be made available on request.