Validation of the LUMIPULSE automated immunoassay for the measurement of core AD biomarkers in cerebrospinal fluid

*Corresponding author: Johan Gobom, Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The SahlgrenskaAcademy,University of Gothenburg, Gothenburg, Sweden; and Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden, E-mail: johan.gobom@neuro.gu.se. https://orcid.org/0000-0001-6193-6193 Lucilla Parnetti, Samuela Cataldi, Giovanni Bellomo and Roberta Rinaldi, Laboratory of Clinical Neurochemistry, Section of Neurology, University of Perugia, Perugia, Italy Pedro Rosa-Neto and Serge Gauthier, Department of Neurology and Neurosurgery, McGill University Research Centre for Studies in Aging, Douglas Research Institute, Le Centre intégré universitaire de santé et de services sociaux (CIUSSS) de l’Ouest-de-l’Île-de-Montréal, Psychiatry and Pharmacology and Therapeutics, McGill University, Montreal, QC, Canada; andMontreal Neurological Institute, Montreal, QC, Canada Martin Vyhnalek, Ondrej Lerch, Jan Laczo and Katerina Cechova, Department of Neurology, SecondMedical Faculty, Charles University, Prague, Czech Republic; Motol University Hospital, Prague, Czech Republic; and International Clinical Research Center, St. Anne’s University Hospital, Brno, Czech Republic. https://orcid.org/0000-0002-5845-1854 (K. Cechova) Marcus Clarin, Ulf Andreasson and Kaj Blennow, Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden; and Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden Andrea I. Benet, Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden; and Translational Neuroimaging Laboratory, McGill Centre for Studies in Aging, McGill University, Montreal, QC, Canada Tharick A. Pascoal, Neserine Rahmouni, Jenna Stevenson and Mira Chamoun, Translational Neuroimaging Laboratory, McGill Centre for Studies in Aging, McGill University, Montreal, QC, Canada Manu Vandijck, Else Huyck and Nathalie Le Bastard, Fujirebio Europe N.V., Ghent, Belgium Daniel Alcolea and Alberto Lleó, Department of Neurology, Memory Unit, Hospital de la Santa Creu i Sant PauBiomedical Research Institute Sant Pau-Universitat Autònoma de Barcelona, Barcelona, Spain; and Centro de Investigación Biomédica en Red Enfermedades Neurodegenerativas (CIBERNED), Madrid, Spain Marcel M. Verbeek, Department of Laboratory Medicine, Radboud University Medical Center, Nijmegen, The Netherlands; and Department of Neurology, Radboud Alzheimer Centre, Radboud University Medical Center, Donders Institute for Brain, Cognition and Behaviour, Nijmegen, The Netherlands Nicholas Ashton, Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden; Wallenberg Centre for Molecular and Translational Medicine, University of Gothenburg, Gothenburg, Sweden; King’s College London, Institute of Psychiatry, Psychology & Neuroscience, Maurice Wohl Clinical Neuroscience Institute, London, UK; and NIHR Biomedical Research Centre for Mental Health & Biomedical Research Unit for Dementia at South London & Maudsley NHS Foundation, London, UK Henrik Zetterberg, Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden; Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital, Mölndal, Sweden; Department of Neurodegenerative Disease, UCL Institute of Neurology, London, UK; and UK Dementia Research Institute at UCL, London, UK Katerina Sheardova, Department of Neurology, Second Medical Faculty, Charles University, Prague, Czech Republic; and First Department of Neurology, Faculty ofMedicine,MasarykUniversity and St. Anne’s University Hospital, Brno, Czech Republic Jakub Hort, Department of Neurology, Second Medical Faculty, Charles University, Prague, Czech Republic;Motol University Hospital, Prague, Czech Republic; and First Department of Neurology, Faculty of Medicine, Masaryk University and St. Anne’s University Hospital, Brno, Czech Republic Clin Chem Lab Med 2021; aop


Introduction
Alzheimer's disease (AD) is the most common cause of dementia [1]. The neuropathological hallmarks of the disease are amyloid plaques, composed of amyloid beta (Aβ) peptides [2], and intraneuronal neurofibrillary tangles, consisting of hyperphosphorylated tau protein (pTau) [3]. The most abundant form of Aβ in the extracellular plaques is a 42-amino acid peptide denoted Aβ 1-42 [4]. Aβ peptides are formed through the cleavage of the transmembrane amyloid precursor protein [5]. Tau in its native form, i.e. without abnormal phosphorylation, is a protein found intracellularly and is involved in the stabilization of the microtubules [6].
In AD, the typical biomarker pattern in cerebrospinal fluid (CSF) is a combination of decreased levels of Aβ 1-42 and increased levels of pTau and total tau (tTau) [4]. CSF tTau is measured using assays based on antibodies that detect all tau isoforms, independently of phoysphorylation status [4]. These biomarkers are used extensively in research but have not yet become included in the clinical diagnostic criteria for AD, although clinicians use them already now as support for the diagnosis [7]. The Aβ 1-42 concentration may be normalized to the concentration of the 40-amino acid-long form of beta-amyloid (Aβ 1-40) to obtain the Aβ 1-42/Aβ 1-40 ratio, which has proved to yield improved diagnostic performance compared to Aβ 1-42 alone [8]. While the ratio decreases the impact of preanalytical sources of errors, e.g., adsorption of peptides to CSF collection tubes and pipetting inaccuracy, it has also been suggested that it may serve to compensate for physiological variation in the expression and processing of the amyloid precursor protein [8,9].
Enzyme-linked immunosorbent assays (ELISA), such as the INNOTEST β-amyloid 1-42, hTau Ag, and PHOSPHO-TAU(181P) assays, have for a long time been used to measure the CSF biomarkers. A drawback of manually performed ELISA assays is that they are prone to variability; minor variations in the quality of the reagents, laboratory environment, and execution of the assay protocol may significantly affect the result, thereby making it difficult to obtain consistent results over time and between labs [10]. Such variation has been greatly decreased by the implementation of automated immunoassay platforms with pre-supplied ready-to-use reagents that have become available in recent years from several assay manufacturers.
In this study, we evaluated the performance of LUMI-PULSE G in relation to established immunoassays for measurement of the core AD biomarkers, and for LUMI-PULSE G amyloid β 1-42, performed a comparison with an LC-MS based certified reference method [21]. We also evaluated intra-and inter-laboratory and longitudinal variability, and by analysis of three cohorts of clinically diagnosed patients, determined clinical cutpoints for Aβ42, Aβ42/Aβ40, pTau and tTau.

Method comparisons
Method comparisons were performed at the neurochemistry laboratory, Sahlgrenska University Hospital/Mölndal, using de-identified CSF-samples from the hosptal routine analysis. Samples were selected based on initial INNOTEST results and re-analyzed on the same occasion by LUMIPULSE G and the respective assay. The use of surplus CSF from the routine analysis was approved by the Ethics Committee at the University of Gothenburg (EPN -Gothenburg, Aug 11, 2014).
Reproducibility CSF samples having low, medium, and high concentrations of the four analytes were analyzed by LUMIPULSE G in three laboratories. The experiments were performed in two separate rounds: in the first round, Total Tau was measured at the University of Gothenburg, Sweden (Lab "a" in Figure 2), at Radboud University, Netherlands (Lab "b"), and at FujiRebio, Belgium (Lab "c"). In the second round, pTau 181, β-amyloid 1-42, and β-amyloid 1-40 were measured at Radboud University, FujiRebio, and Sant Pau-Biomedical Research Institute, Spain (Lab "d"). The homogeneity of the sample aliquots was verified by repeated measurments by at the FujiRebio laboratory, Belgium. In each laboratory, the samples were analyzed in triplicate, twice per day over five days. For some measurements, there was insufficient sample volume avaliable; these measurements were then excluded from the comparison. The standard deviations and coefficients of variation (CV) for intra-laboratory repeatability and inter-laboratory reproducibility were calculated according to ISO 5725-2.
Inter-and intra-lab variation were also assessed by analyzing samples within the Alzheimer's Association Quality Control (AA-QC) program, administered at the Clinical Neurochemistry Laboratory at the University of Gothenburg, Sweden [10]. In the AA-QC program, CSF samples with known concentrations of the AD biomarkers are periodically dispatched to laboratories around the world for method validation to monitor inter-lab variability of AD biomarkers. For LUMIPULSE G, only results of β-amyloid 1-42 and total tau were available. Individual CSF samples were analyzed on seven different occasion. For β-amyloid 1-42 a longitudinal sample consisting of pooled CSF was analysed at seven rounds with 4-19 participating laboratories. For tTau we used data from a pooled CSF sample analysed at four time points with 4-19 participating laboratories in each round.
Long-term consistency of measurments for all four assays, performed on a single LUMIPULSE G instrument was evaluated by analyzing aliquots of two CSF pools; one composed of patient samples with AD-like core biomarker profile, and one with normal biomarker levels. Samples were analyzed 71 times approximately once per week during 18 months in Gothenburg.

Clinical cutpoint determination
For determination of clinical cutpoints for tTau, pTau, and Aβ42, CSF samples from three cohorts were used; from McGill University (Translational Biomarkers for Aging and dementia (TRIAD), Canada), University of Perugia (Italy), and Brno/Praha (Czech Republic). Demographics and biomarker data of the samples are listed in Table 5.
For the TRIAD cohort (n=101), lumbar puncture (LP) was performed under local anesthesia, using an 18 ga. "introducer" to penetrate the interspinous ligaments, followed by dural puncture using the 24 ga. Sprotte atraumatic needle. Twenty nine milli-liter of fluid was collected with polypropylene syringes into 10 mL polypropylene tubes (Sarstedt, part no. 62.9924.294). The first four ml was sent to the clinical laboratory for determination of albumin, total protein, glucose and cells. The remaining 25 mL was transferred to polypropylene tubes and centrifuged at 20°C for 10 min at 2,200×g. The CSF was then rapidly frozen for permanent storage at −80°C until analysed on the LUMIPULSE instrument.
The Perugia cohort was sampled at the Center of Memory Disturbances of the University of Perugia. CSF samples were obtained from subjects that were consecutively recruited between January 2012 and June 2016 and followed up for at least two years. The cohort consisted of 58 patients with probable AD diagnosed according to the NIA-AA criteria [24], regardless of CSF biomarker profile, and 37 nondemented controls. All patients underwent a baseline clinical examination by experienced neurologists, detailed neuropsychological assessment including Mini-Mental State Examination (MMSE), blood chemistry, MRI and lumbar puncture (LP). Neurological controls included cognitively normal subjects, with other neurological diseases such as headache, epilepsy and polyneuropathies, who showed no evidence of progression to dementia after at least two years of follow-up. Patients with subjective memory complaints were not included in the control group. CSF samples were collected via LP from

Biomarker
Reference system n Slope Intercept r Slope and Intercept denote Passing-Bablok regression parameters for comparisons between LUMIPULSE G and the respective reference system. R, Pearson's regression coefficient.
8:00 to 10:00 a.m. after overnight fasting, following a standardized procedure [12] and according to international guidelines [25,26]. CSF (10-12 mL) was taken from the L3-L4 or L4-L5 interspace, immediately collected in sterile polypropylene tubes (Sarstedt cat. nr. 62.610.201), and gently mixed to avoid possible gradient effects. The samples were centrifuged at 2000×g for 10 min at room temperature, aliquoted (0.5 mL) in polypropylene tubes (Sarstedt cat. nr. 72.730.007) and stored at −80°C pending analysis. The Czech cohort was collected within the Czech Brain Aging Study (CBAS) [27]. CBAS is a prospective longitudinal memory clinic- Analytical variation is expressed as standard deviation (SD) and coeficcient of variation (CV) of intra-laboratory repeatability and interlaboratory variation, respectively.
based multicentre study recruiting non-demented adults 55+ years of age. Both CBAS centres in Prague and Brno work as a low-threshold facility; hence, the participants are mostly volunteers who enter by self-referral, with memory complaints expressed by themselves or the family or who were referred by general practitioners, local specialists or the Czech Alzheimer Society to one of the memory clinics. All study participants underwent a standard set of procedures, including neurological and comprehensive neuropsychology examinations, as well as laboratory and vital function assessments. Sociodemographic, personal, pharmacological and family history data were collected. Participants and their informants completed multiple questionnaires about cognitive complaints and lifestyle factors. MRI scans of 1.5 or 3 T were performed every 24 months or earlier when a participant converted to dementia or progressed towards cognitive impairment at an unusual rate. Genotyping was carried out at baseline. In a subset, CSF and/or amyloid PET was performed.

Method comparisons
The LUMIPULSE G assays β-amyloid 1-42, β-amyloid 1-40, total tau, and pTau 181 were compared to the corresponding  "n" denotes the number of laboratories participating in each round of measurments.
Aliquots of a single pooled CSF sample ("longitudinal sample") were measured on seven (Aβ) and four (tTau) occasions, respectively to assess longitudinal stability.

Reproducibility
The intra-laboratory repeatability CVs ( Figure 2, Table 2). were below 6% for all four analytes, with the largest variation for a single analyte was observed for β-amyloid 1-40 (6.2%), while the β-amyloid 1-42/1-40 ratio had a CV of 6.2%. Inter-laboratory reproducibility CVs were similar in magnitude, with the largest CV observed for β-amyloid 1-42 (6.5%). Reproducibility was also studied within the AA-QC program. β-amyloid 1-42 and total tau were measured in (different) CSF samples, analysed on seven occasions in different laboratories ( Figure 3A, B, Table 3). The interlaboratory variation varied significantly between measurement rounds, from 7.6-21% for β-amyloid 1-42, and from 3.5-10.6% for Total Tau. Longitudinal measurement of a single sample indicated no significant drift over time for β-amyloid 1-42 ( Figure 3C, Table 3), but again high CV variation between rounds (7.1-18.9%). For total tau ( Figure 3D, Table 3), there was a slight downward trend over time, with the last measurement 20% lower than the first, while the CVs were more uniform between the measurement rounds (4.3-8.9%) compared to β-amyloid 1-42.
Long-term assay stability was also assessed by measuring aliquots of two CSF sample pools, analyzed on a single LUMIPULSE G instrument in Gothenburg twice per week over an 18-month period ( Table 4). For the AD pool (CSF Pool 2), the CV:s for all analytes were below 4%, while for the normal pool (CSF Pool 1) they were slightly higher, but for all analytes within 8%.

Diagnostic performance and cut-off values
To determine the diagnostic performance of the LUMI-PULSE G β-amyloid 1-42, β-amyloid 1-42/1-40, total tau, and pTau 181 assays and establish cut-off values for use in a clinical setting, we analyzed CSF from three cohorts of AD patients and non-demented controls (Table 5). Cut-off values were determined by ROC curve analysis of data from all cohorts combined. Histograms depicting the distributions of the biomarker concentrations in the groups and ROC curves are shown in Figure 4, and the established cut-off values, sensitivities and specificities are listed in Table 6, including also values for each cohort separately. The optimal cut-off values were <409 ng/L för total tau, <50.2 ng/L for pTau 181, and >526 ng/L for β-amyloid 1-42. Because of the high sensitivity and specificity of the Aβ42/Aβ40 ratio, this biomarker displays a clear bimodal distribution in patient populations. Therefore, it was possible to use an alternative approach to determining the cut-off value, by using mixture model analysis of data from a large number of patient CSF samples (n=2,782) analyzed in a clinical routine laboratory setting ( Figure 5, Table 6), resulting in a cut-off of >0.72. Aliquots of two CSF pools; one with normal (Pool ) and one with AD-type (Pool ) core biomarker profile, were measured once a week over  months (n=) at the Neurochmeistry Laboratory in Gothenburg.

]
Demographic information of the patient cohorts used for cut-off determination of tTau, pTau and Aβ. AD, Alzheimer's disease patients; CN, controls.

Discussion
In the present study, we show good diagnostic performance for LUMIPULSE G β-amyloid 1-42, β-amyloid 1-42/40, total tau and pTau 181. All LUMIPULSE G assays show strong linear correlation with INNOTEST amyloid β 1-42, hTAU Ag and pTau 181, and with MSD for Aβ42/Aβ40, as well as good correlation with the LC-MS reference method for Aβ 1-42. A slight deviation from linearity can be seen for pTau 181, for which data points in the lower range, below 50 ng/L, appear to follow a regression line with a more shallow slope. Notably, this feature is also visible in the control group data in a recent study by Leitao et al. [14].
Two previous studies have reported large bias between the INNOTEST and LUMIPULSE G β-amyloid 1-42 assays [15,16] but since December 2018, the LUMIPULSE G assay has been adjusted to harmonize the measurements to the certified reference materials (CRMs) for Aβ 1-42 from the International Federation of Clinical Chemistry and Laboratory Medicine (IFCC). The previous values, as measured before the update, are divided by 1.46 in order to be harmonized with the CRM. This translates to a decrease of about 31.5%, which to a great extent explains the bias observed in previous studies.
The CSF Aβ 1-42/1-40 ratio correlates well between LUMIPULSE G and MSD. However, if the individual concentrations of Aβ 1-40 and Aβ 1-42 are compared for correlation, there is an approximate two-fold higher concentration of Aβ 1-40 in LUMIPULSE. This discrepancy highlights the need for a CRM for Aβ40 to harmonize results between assay platforms. An important aspect of immunoassay platforms for diagnostic use is the ability to generate reproducible results over time and between labs. In the past, the limited ability of many immunoassay methods for the AD biomarkers has made it difficult to compare results reported in different studies, with reproducibility CVs of over 30% described [32]. Our results show remarkably similar variation for repeatability and reproducibility comparisons, with reproducibility CVs ranging between 5.4 and 6.5% for Aβ42 and between 2.1 and 2.8% for tTau. These results indicate that the use of an automated system with presupplied reagents in a closed cartridge format can greatly improve reproducibility of results between labs.
Data from the AA-QC program show a wide spread of reproducibility CVs when comparisons were performed, ranging between 7.6 and 21% for Aβ42 and 3.5-10.6% for tTau. A possible reason for this difference is that the reproducibility study was performed under optimal conditions, using the same calibrator and kit lots in all labs and analyzing the samples as part of a study. In the AA-QC program, in contrast, the samples may be measured with different reagent lots and are handled as in clinical routine, where analytical errors are more likely to occur. The longitudinal sample of pooled CSF showed no deviation over time for Aβ42 or tTau. At this point, longitudinal data is not available for LUMIPULSE G pTau 181 or β-Amyloid 1-40, as they were only recently included in the program.
Cut-off values were determined in a previous study by Leitao et  The histogram in Figure 4D shows that there are several subjects in the control group that have low Aβ42/Aβ40 ratio. These are possibly individuals with incipient amyloid pathology who do not yet manifest AD symptoms. In the ROC curve analysis, they may lead to underestimation of the optimal cutpoint. For Aβ42/Aβ40, the a clear bimodal distribution made it possible to use mixture model analysis to calculate the cutpoint, thus resulting in a slightly higher value compared to that obtained by ROC curve analysis.
In conclusion, the results presented here suggest that the fully automated LUMIPULSE assays for the CSF AD biomarkers are fit for purpose in clinical laboratory practice. Further, they corroborate earlier presented reference limits for the biomarkers.
Research funding: Brno team is supported by the project no. LQ1605 from the National Program of Sustainability II (MEYS CR) and Praha team is supported by Ministry of Health of the Czech Republic, grant no. 19-04-00560. MMV is supported by the BIONIC project (nr. 733050822), which has been made possible by ZonMW (part of the Dutch national 'Deltaplan for Dementia'; zonmw.nl/dementiaresearch), and by grants from the Selfridges Group Foundation, and National Institutes of Health, USA [grant number 5R01NS104147-02]. JG is supported by Alzheimerfonden (AF-930934),  ROC curve analysis and cut-point determination for tTau, pTau, Aβ, and Aβ/Aβ based on data recorded from three cohorts of AD patients and controls, for each conhort separately as well as combined. For Aβ/Aβ, an alternative method for cut-point determination was used, based on mixture model analysis of clinical patient data from Gothenburg.