Evaluation of the EUROIMMUN automated chemiluminescence immunoassays for measurement of four core biomarkers for Alzheimer’s disease in cerebrospinal fluid

Introduction Robust immunoassays for quantification of Alzheimer's disease (AD)-specific biomarkers are required for routine diagnostics. We report analytical performance characteristics of four new chemiluminescence immunoassays (ChLIA, EUROIMMUN) running on closed, fully automated random-access instruments for quantification of Aβ1-40, Aβ1-42, tTau, and pTau(181) in human cerebrospinal fluid (CSF). Methods ChLIAs were validated according to the guidelines of the Clinical and Laboratory Standards Institute (CLSI). Optimal cut-offs for biomarkers and biomarker ratios were determined using samples from 219 AD patients and 220 patients with AD-related symptoms. For performance comparison, biomarker concentrations were measured in 110 diagnostic leftover samples using the ChLIAs and established Lumipulse G assays (Fujirebio). Results All ChLIAs met CLSI criteria. Overall agreement between assays was 89.0%–97.3 % with highly correlating results (Pearson's correlation coefficients: 0.82–0.99). Passing-Bablok regression analysis revealed systematic differences. Discussion EUROIMMUN ChLIAs showed good analytical performances and represent new valuable tools for diagnostics of AD.


Introduction
Alzheimer's disease (AD) is a neurodegenerative disease with progressing cognitive impairment and the main cause of dementia in persons over 65 years of age.The disease is pathophysiologically based on the formation of plaques and neurofibrillary tangles, as well as the degeneration of neurons and synapses [1][2][3].The AD-related biomarkers beta-amyloid (1-40) (Aβ 1-40 ), beta-amyloid (1-42) (Aβ 1-42 ), total Tau (tTau) and Tau phosphorylated at threonine 181 (pTau(181)) have gained great importance [4][5][6] and are of essential use for laboratory analytics supporting the clinical diagnosis [3,7].The Aβ 1-42 concentration in the cerebrospinal fluid (CSF) of AD patients is inversely proportional to the amount of amyloid plaques and is approximately 50 % lower than in cognitively healthy elderly persons [8].Aβ  shows no or only small changes.Importantly, the ratio Aβ 1-42 /Aβ 1-40 is more reliable than Aβ 1-42 as a single biomarker because individual differences in the production of amyloids can be neglected and negative pre-analytical effects, caused, e. g., by the material and size of the sample tubes and freeze/thaw cycles can be eliminated [4,9].tTau is an unspecific marker of neurodegeneration [10] that is increased in CSF of AD patients.CSF pTau(181) is considered a fairly specific biomarker, with significantly increased levels in AD patients [11].
Whether biomarkers can make a meaningful contribution to diagnosis depends on the quality of the results derived.Collaborative research brought solid advancements regarding established standard operating procedures for lumbar puncture [12,13], international guidelines for handling of CSF samples and standardization of pre-analytical factors [14,15].Furthermore, standardization efforts with certified reference materials (CRM) for CSF Aβ 1-42 have harmonized results between assays of different manufacturers, thus paving the way for globally uniform reference limits and cut-off values [1,16,17].ELISAs for CSF biomarker quantification are still established in clinical laboratories due to their ease of use without need for special equipment [18].However, the utility of results using ELISAs is challenged by the relatively high impact of influencing factors that may lead to great intra-laboratory and inter-laboratory variability [19].In contrast to that, stable and high-precision measurements of CSF biomarker concentrations are possible using closed and fully automated random-access instruments.These instruments for chemilumincescence immunoassays (ChLIAs) improve analytical precision and consequently the majority of laboratories now benefit from accurate, fast and flexible test throughput [20,21].Recent evidence supports the usefulness of CSF biomarkers measured by ChLIA on fully automated random-access platforms for AD diagnosis [22][23][24].
EUROIMMUN has developed four new ChLIAs for quantitation of Aβ 1-40 , Aβ 1-42 , tTau, and pTau(181) in CSF for fully automated processing.This study aimed to evaluate their analytical performance characteristics and to assess agreements between ChLIA results and those of other established assays.
For performance assessment, results of the EUROIMMUN ChLIAs were compared with those obtained using the Lumipulse G β-Amyloid 1-40, β-Amyloid 1-42, Total Tau and pTau 181 (all Fujirebio Inc., Tokyo, Japan) performed fully automated on the LUMIPULSE G 600II [25] according to the manufacturer's instructions by an independent contract laboratory.

Analytical validation
All ChLIAs were evaluated in accordance with the guidelines of the Clinical and Laboratory Standards Institute (CLSI).Intra-lot precision (CLSI EP05-A3, within-run, between-run, within-day and between-day reproducibility) was analyzed using six CSF leftover samples (80 determinations per sample) in triplicate on 20 days with two runs per day.Inter-lot precision was analyzed using six leftover samples (90 determinations per sample) in triplicate on five days with three different lots.
For determination of range of detection capability (limit of blank (LoB), limit of detection (LoD) and limit of quantitation (LoQ), CLSI EP17-A2) sample buffer and four CSF samples with low AD-related biomarker concentration were tested in total 60 times in replicate with three different lots on three different days.
Linearity (CLSI EP06-A) was determined applying three different lots using four replicates of sets of serially diluted leftover samples and sample buffer.Observed and expected results were compared, linearly and non-linearly fitted and the mean coefficient of determination (R 2 ) was calculated.
Interference (CLSI EP07-A3) was studied using four leftover samples spiked with 1 % whole blood and biotin up to a concentration of 10 μg/ml.The recovery rate of the spiked sample compared with the control sample was calculated.
The analysis of stability (CLSI EP25-A) of ChLIA test kits was performed in accordance with the international standard DIN EN ISO 23640:2015 in triplicate using four samples covering the total measurement range of the respective ChLIAs including a negative control.Transport stability was determined using one lot before and after a transport simulation of a four-day period at 2 • C− 33 • C. To investigate accelerated stability, samples were analyzed at various days between zero and 12 (Beta-Amyloid (1-40) and Beta-Amyloid (1-42) ChLIA) or zero and 11 days (tTau and pTau(181) ChLIA) using three different lots stored at 37 • C. The calibration stability was tested at 12 • C− 15 • C at zero, seven, 14, and 28 days with samples, and additionally calibrators and negative control at time point zero using three different lots.On-board-stability of cartridges was investigated using three different lots, of which one has passed the transport simulation, at various days between zero and 60 at 12 • C− 15 • C. On-board-and in-use-stability testing for calibrators was performed between zero and 6 h (Beta-Amyloid (1-40) and (1-42) ChLIA), zero and 10 h (tTau and pTau(181) ChLIA), and at various times between zero and 60 days, respectively, between 18 • C and 25 • C. Stability and recovery rates of positive and negative controls were determined after three freeze/thaw cycles.Real-time stability of cartridges and calibrators was determined at 2 • C− 8 • C at various times between zero and 24 months.

Cut-off determination
For the cut-off determination of the EUROIMMUN ChLIAs, the concentrations of Aβ 1-40 , Aβ 1-42 , tTau, and pTau(181) were measured in CSF samples from 439 patients of a geropsychiatric ward (University Hospital for Psychiatry and Psychotherapy, University Magdeburg).Of those, 219 patients (mean age: 80 years (range:59− 95); 101 female, 61 male, 57 unknown sex) were clinically diagnosed with AD and 220 (mean age: 72 years (range:34− 94); 77 female, 84 male, 59 unknown sex) had other neurological disorders with AD-like symptoms.Patients were diagnosed according to criteria of the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) by psychiatrists, neurologists, and psychologists.The diagnosis was based on results of magnetic resonance imaging of the brain or cerebral computer tomography, CSF analysis, mini-mental state evaluation, Montreal Cognitive Assessment, and CERAD test battery.Measurement of AD-related biomarkers was performed at the University Magdeburg using INNOTEST® immunoassays (Fujirebio).CSF samples were obtained in the morning and collected in 15 ml high clarity polypropylene centrifuge tubes (Falcon®).Demographics and patient diagnosis is summarized in Table 1.The data were evaluated using receiver operating characteristic (ROC) curve analysis.Optimal cut-offs for the determination of Aβ 1-42 , tTau, pTau(181) as well as Aβ 1-42 /Aβ 1-40 , Aβ 1-42 / tTau and Aβ 1-42 /pTau(181) were derived using Youden's index.All calculations were performed using Analyse-it® Validation Edition.

Method comparison: EUROIMMUN ChLIAs versus Fujirebio Lumipulse G
The performance of the EUROIMMUN ChLIAs (applying cut-offs as calculated beforehand) was assessed in comparison with the respective Lumipulse G assays using leftover CSF samples from 110 patients with unknown diagnosis for the laboratory investigators (mean age 57.2, ranging from 11 to 94 years; 49 female, 60 males, 1 unknown age and sex).CSF samples were collected between February and December 2021.After completion of all serodiagnostic analyses, samples were aliquoted in polypropylene tubes and stored at − 20 • C between 2 and 12 months before measurement of AD-related biomarker concentrations using EUROIMMUN ChLIAs.The measurement using the corresponding Lumipulse G assays occurred approximately 3 months later at a contract laboratory.The samples' biomarker concentrations covered the entire measurement range of the EUROIMMUN ChLIAs.Percentage agreement and Cohen's kappa (ĸ) categorized as follows: 0 < κ < 0.2: slight agreement, 0.21 < κ < 0.4: fair agreement, 0.41 < κ < 0.6: moderate agreement, 0.61 < κ < 0.8: substantial agreement, 0.81 < κ < 1: almost perfect agreement [26] were calculated.For quantitative method comparisons, agreement between biomarker concentrations was analyzed by Passing-Bablok regression and Kolmogorov-Smirnov CUSUM test for linearity at a significance level of 5 %.

Ethics statement
The study was performed in agreement with the Declaration of Helsinki and approved by a local ethics committee (University of Magdeburg, registration number: 134/13 and 21/14).Diagnostic left-over samples were collected by a routine diagnostic laboratory (Lübeck, Germany).All samples were processed anonymously.Informed consent was obtained from subjects involved in this study.

Range of detection capability
LoBs of 15.4 pg/ml, 12.7 pg/ml, 3.9 pg/ml, and 1.8 pg/ml were determined for the Beta-Amyloid (1-40), the Beta-Amyloid (1-42), the Total-Tau as well as the pTau(181) ChLIA, respectively (Supplementary Table 4).LoDs were 28.3 pg/ml, 17.5 pg/ml, 5.6 pg/ml, and 3.1 pg/ml and LoQs 41.0 pg/ml, 45.4 pg/ml, 17.4 pg/ml and 9.2 pg/ml, respectively.The specification LoB < LoD/LoQ < lowest concentration of calibrators or 20 % below cut-off were met for all ChLIAs.The upper LoQ (ULoQ) was defined as 20,000 pg/ml, 2,000 pg/ml, 2,000 pg/ml and 400 pg/ml.Lower LoQ (LLoQ, lowest amount of quantifiable measurand) was determined by plotting the mean LoD values against the respective CVs at the point of intersection at a targeted limit of a CV of 8 %.For the tTau ChLIA, LLoQ was determinable for only one of three lots, since no intersection at a CV of 8 % could be found for the two other lots (Supplementary Fig. 1).Results of LLoQ and ULoQ defined the measurement ranges for the respective ChLIAs.

Linearity
A linear relationship between serially diluted samples was found within the measurement ranges for all ChLIAs with relative nonlinearity less than 15 %.R 2 for all ChLIAs was >0.99.(Supplementary Fig. 2).

Interference
There was no whole blood (1 %) and biotin (up to 10 μg/ml) interference detectable in any of the ChLIAs (Supplementary Table 5).

Stability
All ChLIAs were stable in the investigated ranges.Data are shown in Supplementary Table 6.As the area under the curve (AUC) approaches 1, the test system achieves higher combined values of sensitivity and specificity.

Method comparison: EUROIMMUN ChLIAs versus Fujirebio Lumipulse G
The performance of assays was assessed using 110 CSF leftover samples.The Total-Tau ChLIA detected three samples below LLoQ and one above ULoQ.The Lumipulse G Total Tau found 24 samples below LLoQ.Using the Lumipulse G assays for Aβ 1-40, Aβ 1-42 , and tTau determination, the respective concentrations were above ULoQ in one, six and two samples, with one sample, above the ULoQ for Aβ 1-40 and Aβ 1-42 .This sample was excluded from subsequent considerations.All other samples above ULoQ were evaluated as positive and below LLoQ as negative.
The quantitative method comparison revealed constant as well as proportional differences between the assays (Fig. 3).Except for Aβ 1-40 values, values derived by Lumipulse G assays were generally lower than those derived by the EUROIMMUN ChLIAs (Fig. 3).Aβ 1- 42 assays were highly correlated (Pearson's correlation coefficient r = 0.99) and the Passing-Bablok regression showed a slope closest to the identity line (0.87).Lowest correlation (r = 0.82) and low slope (0.63) were observed for the tTau assays (Fig. 3).No significant deviation from linearity (p > 0.1) was found for comparisons between immunoassays.

Discussion
The use of test systems running on fully automated random-access systems for the detection of AD-related biomarkers is highly beneficial for diagnostic laboratories because it improves the analytical precision [20,21].Here, we show that the EUROIMMUN ChLIAs proved to be specific for their respective target analyte.Interferences were not observed.Measurement of CRM yielded recovery rates similar to those observed by others [17] with repeatable results using several lots.Cross-reactivities were excluded except for the Beta-Amyloid (1-40) ChLIA when spiking a sample with Aβ 1-39 to a concentration of 10 ng/ml.Aβ 1-39 is a beta-amyloid isoform that can be found in CSF samples.The in vivo concentration of Aβ 1-39 is approximately 14 % of the concentration of Aβ 1-40 [27,28], which was found to range between 6.7 and 7.8 ng/ml in CSF samples of different disease cohorts (determined by using Beta-Amyloid (1-40) ChLIA, data not shown).As the concentration of Aβ 1-39 used in this study to test cross-reactivities greatly exceeds its average in vivo concentration, the cross-reaction observed here is irrelevant for laboratory diagnostics.Moreover, since the ratio between Aβ 1-39 and Aβ 1-40 seems to be quite stable in different disease conditions [27,28], any misclassification originating from a potential cross-reaction is unlikely as the clinical cut-off applied here for the Beta-Amyloid (1-40) ChLIA was derived using patient samples.
In several studies, cut-offs were derived using samples from AD patients and cognitively healthy individuals.In other studies, control samples for cut-off determination are obtained from patients with neurological diseases without AD-related pathology, such as multiple sclerosis or epilepsy.Presumably, these control patients are of younger age than average AD patients [23,29].Both approaches bear the risk of misrepresenting conditions of the clinical reality [29] and consequently of misdiagnosing patients.As AD shares symptoms with several neuropsychiatric disorders such as frontotemporal or vascular dementia, using samples from these patients could minimize this risk of misdiagnosis and increase diagnostic accuracy of assays [30].One great advantage of the present study is that for cut-off determination, samples from patients with similar symptoms and of similar age compared to AD patients were obtained from a geriatric psychiatry.Moreover, a total number of 439 CSF samples was analyzed, which is a higher number of samples compared to those used in various other studies [23,29].
The comparison of results obtained using the EUROIMMUN ChLIAs and the corresponding Lumipulse G assays revealed high overall agreement (89.0 %-97.3 %) as well as good correlation (r = 0.82− r = 0.99).However, Passing-Bablok regression analysis showed constant as well as proportional differences between test systems.Absolute concentrations differed between tests with results from EUROIMMUN assays showing higher concentrations than Lumipulse G assays, except for Aβ 1-40 concentrations.A reason for this shift could be the lack of CRM for the analytes Aβ 1-40 , tTau and pTau(181).Assessment of trueness for the Beta-Amyloid (1-42) ChLIA using CRM revealed only minor divergence (0.8 %-10.4 %) to the reference concentrations, which has also been observed for the Lumipulse G β-Amyloid 1-42 assay by others [17].Nevertheless, Passing-Bablok analysis indicated systematic differences between EUROIMMUN and Lumipulse G assays for the determination of Aβ 1-42 concentrations.As the Lumipulse G assays were performed by a contract laboratory and EUROIMMUN ChLIAs were run in-house, systematic differences due to inter-laboratory discrepancies cannot be ruled out, although the performance assessment was planned thoroughly to minimize pre-analytical factors.However, these numerical differences have little influence on the diagnostic evaluation since cut-offs of both assays also differ as they were adjusted to samples from different cohorts.Indeed, best agreement was found between Beta-Amyloid (1-42) ChLIA and Lumipulse G β-Amyloid 1-42 (Table 2).
The established Aβ 1-42 /Aβ 1-40 ratio minimizes the influence of pre-analytical factors and thereby variations between laboratories [31].This is advantageous if the guidelines for CSF handling and protocols for standardization of pre-analytical factors cannot be met, for instance when CSF samples are sent to centralized diagnostic laboratories that do not implement pre-analytical protocols regarding sample handling or storage, which might affect the measured biomarker concentration [14].While several initiatives implement algorithms that include most or all of the four individual biomarkers [5,32], some research groups are pushing towards the use of heterologic biomarker ratios [33].The reason being that the ratio including the two hallmark biomarker of Alzheimer's disease (Aβ 1-42 /pTau(181)) should be sufficient to support the differentiation between Alzheimer's and non-Alzheimer's disease.The heterologic ratios Aβ 1-42 /tTau and Aβ 1-42 /pTau(181) are of increasing interest, because their advantage could be the combination of the two different pathological processes (amyloid and tau-pathology) in one parameter, thus simplifying the interpretation of results.Some studies report identical diagnostic performances for the heterologic Aβ 1-42 /pTau(181) and the amyloid Aβ 1-42 /Aβ 1-40 ratio [34,35].In other studies, the heterologic ratios Aβ 1-42 /pTau(181) [23,35,36] and Aβ 1-42 /tTau [23,36,37] slightly outperform the Aβ 1-42 /Aβ 1-40 ratio.In this study, the AUC for pTau(181) determination yielded higher values than for any of the three biomarker  ratios.This has also been previously shown for AUC values for pTau(181) and the Aβ 1-42 /Aβ 1-40 ratio [23,37], while other studies found better values for the Aβ 1-42 /pTau(181) and Aβ 1-42 /tTau ratios than for pTau(181).Although heterologic biomarker ratios presumably show a slightly better performance than the Aβ 1-42 /Aβ 1-40 ratio, their use is more prone to systematic errors caused by pre-analytic effects [38].Accumulation of systematic errors may even lead to a misdiagnosis of patients.Most scientific studies do not represent real-world settings adequately as they are monocentric and use CSF samples from long-term storage, which are processed in batches.Thereby, systematical and individual-related factors and inter-laboratory differences might be neglected and results should be critically evaluated.In contrast to the Aβ 1-42 /Aβ 1-40 ratio, use of heterologic ratios is currently not recommended by diagnostic guidelines.

Limitations
The cut-offs determined in this monocentric study were not additionally validated using an independent cohort, which could be performed in a future multicentric study.The derived cut-offs are however of high quality because they were established using a large panel of 439 patients suffering from different forms of dementias representing an authentic cohort for differential diagnosis of AD.
Although the EUROIMMUN ChLIAs and the Lumipulse G assays were processed on fully automated instruments, analytical variations occurring in the different laboratories might potentially have resulted in lower concordance of measurement results.
Generally, correct sample collection and storage are crucial for the reliability of the test results, and test results should always be interpreted together with those of further diagnostic analyses and based on the clinical picture of the patient.

Conclusion
The newly developed EUROIMMUN ChLIAs for fully automated determination of AD-related biomarkers show good analytical performance characteristics.High agreement rates of results obtained using EUROIMMUN ChLIAs with those obtained using Lumipulse G assays were found.Thus, the novel ChLIAs provide reliable tools for supporting the diagnosis of AD in CSF samples.

Financial support
This research received no specific grant from any funding agency, commercial or non-profit sectors.

Fig. 3 .
Fig. 3. Passing-Bablok regression analysis comparing EUROIMMUN ChLIAs and Lumipulse G immunoassays.(A) Aβ 1-40 , (B) Aβ 1-42 , (C) tTau, (D) pTau(181).The regression line is indicated in solid blue, respective 95 % confidence intervals in dashed blue, and the identity line in dashed red.Slope and intercept denote Passing-Bablok regression parameters for comparisons between the two assays.Values outside the measurement range are plotted as red circles.N, number, LLoQ, lower limit of quantification, ULoQ, upper limit of quantification.(For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.)

Table 1
Demographics and diagnosis of patients.

Table 2
Numbers of samples with normal and abnormal AD-related biomarker concentrations and amyloid and heterologic ratio by means of EUROIMMUN ChLIAs and Lumipulse G assays and results of the qualitative method comparison.N, number of samples; ĸ, Cohen's kappa; CI, 95 % confidence interval.