First amyloid β1‐42 certified reference material for re‐calibrating commercial immunoassays

2.1. Reference material development: raw materials

Human CSF was chosen as the starting material for the CRMs because the first commutability study showed that artificial matrices or CSF spiked with Aβ₄₂ peptide were not commutable. ¹⁴ Freezing such liquid materials was shown to provide adequate stability in the previous studies. ²⁰ The raw material was anonymized human CSF from samples provided by the Sahlgrenska Academy at the University of Gothenburg (Mölndal, Sweden). The CSF material was collected from 19 normal pressure hydrocephalus patients (first presenters) through continuous lumbar drainage and was approximately ranging from 200 to 350 mL per patient. This material was otherwise destined for destruction. Samples were aliquoted and frozen at –80°C at the respective Swedish neurological clinics where they were originally collected. Subsequently, the CSF samples were transported on dry ice to the Sahlgrenska Academy at the University of Gothenburg, where the sample's Aβ₄₂ concentration was determined by ELISA (INNOTEST Amyloid β [1‐42], Fujirebio Europe N.V., Ghent, Belgium).

2.2. Reference material development: preparation and first studies of candidate materials

To process the candidate materials, the 19 CSF donations were thawed once. Three materials were prepared by pooling different Aβ₄₂ level donations leading to ERM‐DA480/IFCC, ERM‐DA481/IFCC, and ERM‐DA482/IFCC with low, middle, and high concentration of CSF Aβ₄₂, respectively. Subsequently, the CSF pools were aliquoted into 0.5 mL polypropylene Maxymum Recovery sterile microvials (Axygen Scientific, Inc., Union City, California, USA) and capped using screw caps with O‐rings. This was performed using an automatic filling/capping machine placed under a laminar flow cabinet equipped with high efficiency particulate air (HEPA) filters. Each pre‐labeled vial was filled with a minimum of 0.5 mL of CSF and stored immediately at –70°C. Detailed explanations on the processing are available in the certification report. ²¹

The assessment of the homogeneity of the processed material was crucial based on the outcome of several feasibility studies and the fact that Aβ₄₂ is very prone to adhesion to surfaces. Homogeneity as well as short‐ and long‐term stability of the materials was assessed using immunoassays from Roche and EUROIMMUN. Furthermore, the candidate CRMs were tested in a commutability study and were commutable for the combination of routine measurement procedures such as EUROIMMUN beta‐amyloid (1‐42), IBL Amyloid‐beta (1‐42) CSF ELISA, INNOTEST β‐AMYLOID (1‐42), Lumipulse G β‐Amyloid (1‐42), V‐PLEX Aβ Peptide Panel 1 (6E10), and Roche Elecsys β‐amyloid (1‐42). ¹⁸

2.3. Reference material development: value assignment using a common calibrant

A stock solution of pure recombinant Aβ₄₂ peptide was prepared as the calibrant for the IDMS‐based RMPs. The peptide was processed based on the procedure described by Broersen et al. which showed the production of monomeric peptide in solution. ²² Recombinant human Aβ₄₂ peptides were purchased deposited on lyophilized hexafluoroisopropanol (HFIP) films (rPeptide, Bogart, Georgia, USA) with a purity of ≥97%. The peptides were dissolved in HFIP at a concentration of 1 mg/mL, vortexed for 30 seconds, and dried with argon. Subsequently, the peptides were dissolved in dimethyl sulfoxide at a concentration of 1 mg/mL and vortexed for 30 seconds. This peptide solution was injected into a HiTrap desalting column (GE Healthcare Europe GmbH, Diegem, Belgium) equilibrated with 20% acetonitrile and 1% ammonium hydroxide in water. Fractions were collected, pooled, and diluted with 20% acetonitrile and 1% ammonium hydroxide. The peptide solution was collected in 0.5 mL polypropylene vials (Sarstedt, Nümbrecht, DE, Germany) with each vial containing 0.1 mL of solution. The vials were immediately stored at –70°C.

The value assignment of the calibrant stock solution was performed by IDMS using a TSQ Vantage mass spectrometer (Thermo Fisher Scientific, Waltham, Massachusetts, USA), with solid phase extraction and liquid chromatography conditions according to the RMP. ¹⁶ , ¹⁷ Six different Aβ₄₂ preparations were used for the calibration of these measurements, producing six datasets for the stock solution. The concentration values of these six aforementioned Aβ₄₂ preparations were determined by the JRC in Geel (Belgium) using amino acid analysis ²¹ based on IDMS ²³ combined with the purity assessment of the peptide solutions. The content for the stock solution (the average of the means of the six datasets) was converted from mass fraction to mass concentration using the density of the stock solution. The resulting mass concentration for the calibrant stock solution was 68.87 ± 3.54 mg/L.

2.4. Reference material development: characterization and value assignment

The vials for the characterization of the material were selected using a random stratified sampling scheme and covered the whole processed batch. Depending on the amount of samples needed, the whole batch is cut into small sub‐batches and one sample is taken from the each of the sub‐batches. Then they are distributed among the studies to the various temperatures and time points.

A key requirement for any reference material aliquoted into units is equivalence between the units. This between‐unit homogeneity was evaluated to ensure that the certified values of the candidate CRMs are valid for all vials of the material, within the stated uncertainties. Therefore, the number of vials selected corresponds to at least the cube root of the total number of vials produced. The selected vials for homogeneity were analyzed both with a fully automated immunoassay (Roche Elecsys β‐amyloid [1‐42], Roche Diagnostics GmbH, Penzberg) ²⁴ and with a manual ELISA (EUROIMMUN beta‐amyloid [1‐42], EUROIMMUN AG, Luebeck), ²⁵ as described in the certification report. ²¹

Stability testing is necessary to establish the conditions for storage (long‐term stability), as well as for dispatch of the materials to the customer sites (short‐term stability). Time and temperature were regarded as the most relevant influences on the stability of the materials. The stability studies were carried out using an isochronous design. In this approach, samples are stored for a particular length of time at different temperature conditions. Subsequently, the samples are moved to conditions where further degradation can be assumed to be negligible or at least reduced (reference conditions). ²⁶ , ²⁷ At the end of the isochronous storage, the samples are analyzed simultaneously under repeatability conditions. For the short‐term stability study, samples were stored at –70°C, –20°C, 4°C, and 18°C, respectively, for 0, 1, 2, and 4 weeks (at each temperature). For the long‐term stability study, samples were stored at –70°C and –20°C for 0, 4, 8, and 12 months (at each temperature). In both case studies, the reference temperature was set to –150°C (stored above liquid nitrogen). Two vials per storage time were selected using a random stratified sampling scheme. From each vial, three subsamples were measured by ELISA (EUROIMMUN beta‐amyloid [1‐42], EUROIMMUN AG, Luebeck). The measurements were performed under repeatability conditions and a randomized sequence was used to differentiate any potential analytical drift from the trends over storage time as described in the certification report. ²¹

The candidate CRMs have been value‐assigned by IDMS using RMPs ¹⁶ , ¹⁷ or variations thereof. The IDMS measurements were calibrated using calibrators prepared from the common calibrant stock solution of Aβ₄₂ as described above. Five laboratories were selected for the value‐assignment measurements based on criteria that comprised both demonstrated technical competence and quality management aspects. Prior to the study commencement, each participant was required to operate a quality system and to deliver documented evidence of their laboratory proficiency in the field of Aβ₄₂ measurements in human CSF by successfully participating in a proficiency test. It was not mandatory to have an accreditation, but it was obligatory to meet the requirements of ISO/IEC 17025. ²⁸

Each laboratory received nine vials each of ERM‐DA480/IFCC, ERM‐DA481/IFCC, and ERM‐DA482/IFCC and was requested to provide 27 independent results per candidate CRM, that is, three per vial. The sample preparations and measurements had to be spread over 3 days to ensure intermediate precision conditions. Each laboratory also received six vials of the Aβ₄₂ stock solution to perform an independent calibration for each day of analysis.

2.5. Handling of the CRM

The following tests were conducted to determine a suitable way of handling the CRM samples. To determine the influence of the plastic surface on the analyte, the CRMs were analyzed using an ELISA (INNOTEST β‐AMYLOID [1‐42], Fujirebio) directly from their original vial as well as after transfer into a separate 1.5 mL polypropylene tube. The measurements were performed in triplicate for each of the three CRMs.

Additionally, this study determined the effect of different thawing methods on the measurement of Aβ₄₂ in CSF pooled samples by ELISA. In total, 16 aliquoted CSF samples in tubes (1.5 mL vials, same reference as the ones used to store the CRMs) stored at –70°C were thawed in four different ways. First, all the vials were placed on the workbench to thaw for 15 minutes at room temperature in a standing position. Of the 16 vials, 4 vials did not receive any further treatment, while 4 other vials were tapped/flicked for five times. Another 4 vials were placed standing in an Eppendorf Thermomixer Comfort for 5 minutes at 300 rpm at room temperature and the remaining four vials were placed in a roller mixer (RotoShakeGenie) for 15 minutes at 30 rpm at room temperature. All the vials were measured by ELISA (INNOTEST β‐AMYLOID [1‐42], Fujirebio) in triplicate.

Another two tests were performed, first to determine the effect of keeping the CRMs at room temperature and then to determine the effect of successive freeze‐thaw cycles. To do so, each CRM level was analyzed at different time points (0, 4, 18, 24, and 48 hours) standing at room temperature for the first test and after 0, 1, 2, 3, and 5 successive freeze‐thaw cycles (from –70°C to room temperature for 25 minutes and re‐frozen at –70°C for 3.5 hours before re‐thawing). For both experiments, two vials per CRM per time point or per number of freeze‐thaw cycles were measured in triplicate. All the vials were measured by ELISA (INNOTEST β‐AMYLOID [1‐42], Fujirebio).

2.6. Recalibration: samples for immunoassay re‐calibration

Remnant de‐identified CSF samples were obtained from Winfried Stöcker's Klinisch‐immunologisches Labor (EUROIMMUN, Germany). The samples were tested for their Aβ₄₂ levels using EUROIMMUN ELISA test. The samples were then pooled based on their Aβ₄₂ concentrations to generate 15 CSF pools encompassing the entire Aβ₄₂ measurement range. CSF pools were mixed and 250 μL were aliquoted per 2 mL screw‐cap micro Sarstedt tube (catalog number: 72.694.600) and stored at –20°C for ∼ 2 to 3 weeks before measurement. The tubes were pre‐rinsed with Aβ1‐40 dilution buffer (EUROIMMUN, Germany) to minimize Aβ₄₂ binding to the tubes. ²⁹

2.7. Recalibration: pre‐analytical handling and sample measurement for recalibration

All samples were sent to the manufacturers of immunoassays and analyzed according to their respective test instructions. The CRMs and the CSF pools were thawed on a roller mixer for 30 minutes at room temperature followed by a short spin down. Both the CRMs and CSF pools were transferred directly to the automation platforms without any transfer to alternative tubes. The automation platforms included the Lumipulse G series (Fujirebio, Belgium), Elecsys (Roche, Germany), and EUROIMMUN Chemiluminescence (ChLIA) RA 10 analyzer (EUROIMMUN, Germany). For each CSF pool, two aliquots were measured in total in duplicate on two separate days. On both days, the samples were loaded along with assay specific calibrators and controls without any change in the batch, operator, or instrument. The two different aliquots of the CRMs were measured at the beginning and at the end of each run in duplicates.

2.8. Recalibration: data analysis and statistics

This study was performed to determine bias between the platforms based on the concentrations calculated using original calibration curve and those re‐calibrated to the CRM values (CRM‐adjusted). Pairwise correlation between the assays was calculated using Pearson's, Spearman's, and Kendall's correlation coefficients. Regression analysis was used to estimate pairwise the bias between the assays: weighted Deming regression ³⁰ was used for original values and Passing‐Bablok regression ³¹ for CRM‐adjusted values. In contrast to the weighted Deming regression, Passing‐Bablok regression is robust against outliers. Therefore, this method is often a preferable option for method comparison analysis. The currently implemented version of this regression assumes that the slope between methods should be close to one, which was expected for comparisons between CRM‐adjusted concentrations. For the comparisons of the original concentrations the expected slope was considerably different from one, therefore weighted Deming regression was used for the analysis. The regression fit Y = a + b X was used estimate expected percentage bias between two methods at median concentration M: (a + b M–M)/M × 100%. The 95% confidence intervals for weighted Deming regression fit were calculated using jackknife method. ³⁰ For Passing‐Bablok regression, percentile bootstrap confidence intervals were used. ³² To reduce uncertainty of bias estimates due to random errors, the method comparisons were done based on the averaged measurements.

Precision of measurements (within‐run and between‐day variability) was investigated for each platform using variance component analysis.

Statistical calculations were conducted in R 3.2.2 using packages mcr 1.2.1 for regression analysis and VCA 1.2.1 for variance component analysis.

3.1. Reference material processing

Three reference materials with high, medium, and low concentrations of Aβ₄₂ were prepared to allow vendors to establish calibration curves by mixing the three reference materials in different ratios. This was done to ensure that the dilution of the lower concentrations with the solvent did not result in non‐linearity of the calibration curve. For each concentration level, a set of CSF samples was mixed to produce a material with the targeted concentration and filled into pre‐labeled Axygen vials within a single day. On the day of the filling, the vials selected for homogeneity, stability, and characterization studies were removed and stored separately. All samples were frozen at –70°C on the day of processing.

3.2. Characterization of processed material

For the homogeneity studies, regression analysis was performed to evaluate potential trends in the analytical sequence as well as trends in the filling sequence. No trend in the filling sequence was observed at a 95% confidence level. Significant (95% confidence level) trends in the analytical sequence were visible and were corrected accordingly. Detailed explanations are available in the certification report. ²¹

For the stability studies, the data were evaluated individually for each temperature and the results were screened for outliers using the single and double Grubbs test at a confidence level of 99%. The short‐term stability study did not show any statistically significant trend for the candidate CRMs at –70°C or –20°C. A significant trend at 4°C was found for ERM‐DA481/IFCC, but not for ERM‐DA480/IFCC and ERM‐DA482/IFCC. Furthermore, a significant trend was observed for all three materials at 18°C. Therefore, it was concluded to ship the material on dry ice to avoid any freeze‐thaw cycles. In the long‐term stability study, none of the trends was statistically significant at a 99% confidence level for any of the temperatures, signifying that the material can be stored at –70°C. The complete stability results can be found in the certification report. ²¹

The value‐assignment measurements were performed by five laboratories using IDMS according to one of the RMPs or variations thereof. ¹⁶ , ¹⁷ The laboratories and their methods were previously validated in a ring trial. ¹⁹ As described in the certification report ²¹ the data were combined by calculating the unweighted average value of the means of five accepted datasets, wherein each set was obtained from a different laboratory (Figure 1). For each laboratory, the variations presented here are a combination of repeatability and day‐to‐day variation of the specific methods. The uncertainty is the expanded uncertainty of the certified value with a coverage factor k = 2 corresponding to a 95% confidence level. The certified Aβ₄₂ mass concentrations in ERM‐DA480/IFCC, ERM‐DA481/IFCC, and ERM‐DA482/IFCC are 0.45, 0.72, and 1.22 μg/L, respectively, with expanded uncertainties (k = 2) of 0.07, 0.11, and 0.18 μg/L, respectively. Subsequently, the first CRMs for Aβ₄₂ in CSF were released in December 2017 and are currently available to users for purchase. ³³

3.3. Handling of the CRM

Aβ₄₂ is very prone to be adhesive to surfaces such as tubes and pipette tips. ²⁹ , ³⁴ Therefore, variability in the handling of the CRMs could lead to different results. ²⁶ , ³⁵ In the present study, an analysis of the three CRMs after their transfer into polypropylene tubes showed a decrease of measured values in comparison to an analysis directly from original CRM vials of ≈ 35%, 33%, and 20% for ERM‐DA480, ERM‐DA481, and ERM‐DA482, respectively (Figure 2A). After five successive tube transfers of ERM‐DA481, there was an ≈ 80% reduction in the original value (data not shown). These results portray the importance of handling the CRMs in the right manner, with an emphasis to avoid as much as possible tube/container transfers prior to analysis.

In addition, the methodology of thawing that includes contact between CSF and vial cap or rotation of vials for homogeneity etc. could have an influence on these materials. In the current study, even though some differences were observed among the four methods of thawing, the variation between the results was not > 8%, irrespective of the methodology used (Figure 2B). Furthermore, even though methods involving rotation such as roller mixing or thermo mixing gave slightly higher values by ELISA measurements, they were not statistically significant.

The time spent at room temperature and freeze‐thaw cycles of the CRMs were also investigated. Statistical evaluations were performed taking into account the conditions in which minimal change is expected as a reference point and then comparing the results applying the principles laid down. ³⁶ In the present study we could observe that keeping the CRMs for several hours at room temperature does not seem to have a direct impact on the measurements (Figure 2C). However, at 48 hours the results for the highest CRM level (ERM‐DA482/IFCC) show a decrease of the Aβ₄₂ concentration. This observed diminution is statistically significant when compared to time point t = 0. Therefore, we recommend using the CRMs shortly after thawing. Regarding the effect of possible freeze‐thaw cycles, we could see a statistically significant decrease in the measured values already after the first cycle (Figure 2D) and therefore we definitely recommend avoiding freeze‐thaw cycles of the materials.

3.4. Test system re‐calibration

EUROIMMUN, Roche, and Fujirebio used their respective assays to measure the Aβ₄₂ concentration in the 15 CSF samples both before and after re‐calibration of their test system with the CRMs. A good overall precision of the measured values was observed across platforms. The variability among duplicate determinations was low for all assays (mean coefficient of variance < 5%) and no considerable run effects were observed (data not shown). The estimation of correlation and bias for the original sample concentrations revealed high correlations across the different platforms (R: 0.97 to 0.99; Figure 3A, B and C). However, a bias was observed at the median concentration across platforms with Roche versus EUROIMMUN at –25% (–28%, –21%), EUROIMMUN versus Fujirebio at 49% (46%, 52%), and Fujirebio versus Roche at –12% (–16%, –8%); Table 1). Following re‐calibration of the test system with the CRMs, a high consistency was observed among the measurement results without any significant bias across platforms at the median concentration of 700 pg/mL (Figure 4A, B and C). The bias observed at a fixed concentration of 700 pg/mL by comparing the platforms to the mean CRM‐adjusted concentrations was 1.16% (0.010%, 2.51%) for EUROIMMUN, 1.00% (–1.04%, 1.41%) for Fujirebio, and –2.06% (–3.15%, 1.97%) for Roche (Table 2). Overall, the deviations from the CRM target values were < 9% across platforms following re‐calibration (Figure 5).

4.1. Intended use

The CRMs ERM‐DA480/IFCC, ERM‐DA481/IFCC, and ERM‐DA482/IFCC are intended for the calibration of In Vitro Diagnostics, quality control, and/or the assessment of the method performance for the measurement of Aβ₄₂ in CSF. Additionally, similar to any reference material, CRMs can be used for establishing control charts and in method validation studies.

The following sample handling recommendations are provided based on our experience studying the CRMs. First, the user should be aware that Aβ₄₂ in CSF can aggregate and bind to surfaces, particularly to glass. The influence of surface on the analyte can be limited by keeping the number of tube transfers to a minimum, prior to measurement. Additionally, despite observing only limited effects due to the different thawing methods, we recommend avoiding vortexing or inverting the CRM vials to prevent contact between the solution and additional surface of the vial. Finally, storing the CRMs at room temperature for extended periods (> 24 hours) will have an impact on their stability. Therefore, it is important to thaw the CRM vials at room temperature and immediately use them for analysis. We recommend storing the materials at –70 ± 10°C and in the dark, avoiding direct sunlight exposure, and avoiding freeze‐thaw cycles.

The commutability of the reference materials was verified for the combination of the following routine measurement procedures which quantifies ether the full‐length of Aβ1‐42 or a mixture of AβN‐isoforms (eg, MSD): EUROIMMUN beta‐amyloid (1‐42) ELISA and ChLIA (EUROIMMUN AG, Luebeck, DE); IBL Amyloid‐beta (1‐42) CSF ELISA (IBL International GmbH, Hamburg, DE); INNOTEST β‐AMYLOID (1‐42) (Fujirebio Europe, N.V., Gent, BE); Lumipulse G β‐Amyloid (1‐42) (Fujirebio Europe N.V., Ghent, Belgium); V‐PLEX Aβ Peptide Panel 1 (6E10) (Meso Scale Discovery, LLC, Rockville, Maryland, USA); Roche Elecsys β‐amyloid (1‐42) (Roche Diagnostics GmbH). Furthermore, it is important that vendors with alternative Aβ₄₂ measurement procedures, except those mentioned above, verify if the three CRMs or their dilutions are commutable for their methods.

The reference material can be obtained from the JRC or one of its authorized distributors. A detailed technical report is also available. ³³

4.2. Immunoassays

With the introduction of various commercial new immunoassays, the specificity and sensitivity of Aβ₄₂ detection in CSF samples has increased. ³⁷ However, the variability observed in the measured Aβ₄₂ concentration values across laboratories has precluded the usefulness of Aβ₄₂ in routine clinical practice. ³⁸ Such a variation has hindered the introduction of a common cut‐off value in the routine diagnostic work‐up for AD. Additionally, studies have reported that the apolipoprotein E (APOE) ε2/ε3/ε4 polymorphism affects the concentration of CSF Aβ₄₂ in AD ³⁹ , ⁴⁰ and in turn impacts uniform cut‐off values. However, APOE ε4 genotype does not modulate CSF Aβ₄₂ concentrations when stratified for cortical uptake of [¹⁸F] flutemetamol, indicating that the Aβ₄₂ levels reveal cortical Aβ deposition independent of the APOE status. In short, all APOE genotypes should have a uniform clinical cut‐off value for CSF Aβ₄₂ and these biomarkers correlate well with cerebral Aβ irrespective of APOE genotype. ⁴¹ Standardization of the pre‐analytical and analytical procedures would further aid in the generation of comparable results. ¹¹ , ³⁷ The Alzheimer's Association Global Biomarkers Standardization Consortium has been working to standardize pre‐analytical parameters such as the CSF sample collection as well as handling procedures. A consensus protocol for collection and analysis of fresh CSF sample for the estimation of Aβ₄₂ was released during the Global Biomarkers Standardization Consortium meeting at the 2018 Alzheimer's Association International Conference held in Chicago. ⁴² , ⁴³ In the current study, the CSF pools were mixed and aliquoted into 2 mL, protein low binding, polypropylene tubes as suggested in the above guidelines. These samples were then frozen at –20°C for subsequent analysis on the different automation platforms without any transfer to alternative tubes. Presently, there are no guidelines for handling frozen CSF samples for biomarker analysis. Therefore, there is an urgent need for "standard operating procedures" for frozen CSF samples including appropriate transport and storage guidelines. Standardization of the analytical procedures has been difficult due to the lack of calibrators with appropriately assigned target values. Therefore, without a reference system, the measurement of Aβ₄₂ values is vendor dependent and results in a systematic bias of the estimated Aβ₄₂ values across different platforms. ³⁷ Consequently, the IFCC Working Group on CSF proteins developed reference methods for the quantification of Aβ₄₂ in CSF. The IFCC and the JRC of the European Commission developed three CRMs from CSF pools with specified endogenous Aβ₄₂ target concentrations. ¹¹ , ²¹ All three CRMs were commutable across different vendor immunoassays for Aβ₄₂.

In conclusion, it is possible to re‐calibrate CSF Aβ₄₂ kits using the CRMs (ERM‐DA480/IFCC, ERM‐DA481/IFCC, and ERM‐DA482/IFCC) and to establish a quality control system for immunoassays. Subsequently, it is important for different vendors to re‐calibrate their test system using the CRMs, as mandated by the EU Directive on In Vitro Diagnostics Medical Devices (Directive 98/79/EC). We are now one step closer to the harmonization of different assays and the establishment of a global cut‐off value for CSF Aβ₄₂ measurements in AD. As a follow‐up, the Global Biomarkers Standardization Consortium intends to organize a follow‐up study in the near future to confirm the findings in a larger cohort. The consortium will work to reach out to other kit manufacturers with an aim to implement the CRMs for test re‐calibration. Further work is currently ongoing to develop additional CRMs for other AD biomarkers including Aβ₄₀, total‐tau, and phosphorylated‐tau.