Validation of a Commercial Chemiluminescence Immunoassay for the Simultaneous Measurement of Three Different Amyloid-β Peptides in Human Cerebrospinal Fluid and Application to a Clinical Cohort

Study approval, study cohort, and sample collection procedures

The collection and archiving of CSF samples in a local biobank and their use for biomarker studies were approved by the ethics committee of the University of Duisburg-Essen. The participants were recruited in the LVR-Clinics Essen, University of Duisburg-Essen, Department of Psychiatry and Psychotherapy and in the Memory Clinic at the Elisabeth Hospital in Essen (Germany). Lumbar punctures were done for diagnostic reasons. All subjects or their legal caregivers gave their informed consent. CSF was obtained by lumbar puncture and processed in a single laboratory according to standardized procedures essentially as described before [13]. In brief, CSF was collected in polypropylene tubes and centrifuged (generally within 90 min after lumbar puncture) for 15 min at 1600×g and approximately 20°C. The supernatant (“CSF”) was stored in aliquots at –80°C until use. Aliquots were thawed only once prior to the analysis.

Clinical and biomarker supported diagnosis

The subjects included in this study underwent lumbar puncture for diagnostic reasons. They were categorized into three diagnostic groups, namely probable AD, possible AD, and non-AD disease controls (DC) according to clinical, neuropsychological, and biomarker criteria. Probable dementia due to Alzheimer’s disease (ADD) and possible AD were diagnosed clinically according to the NINCDS-ADRDA criteria [14]. The non-AD disease control group (DC) comprised patients with dementia of other origins and non-demented patients with heterogeneous other neurological or psychiatric disorders. Clinical diagnoses were assigned to the subjects of the DC subgroup according to the International Classification of Diseases (ICD-10, German modification). The DC group included patients with, e.g., schizophrenia, depressive disorders, normal pressure hydrocephalus, affective and schizoaffective disorders, persistent delusional disorder, addictive disorders and concomitant diseases, e.g. the Wernicke-Korsakoff syndrome, mild cognitive impairment not due to AD, organic mental non-dementia disorders, other non-AD dementias such as vascular dementia, frontotemporal dementia, Lewy body dementia, and dementia due to other systemic diseases. The DC group (n = 90) had a mean age of 68.8±11.8 years (mean±SD) and comprised 47 females (52.2%) and 43 males (47.8%). The mean age in the probable AD group (n = 62) was 73±10.4 years, and the number of female and male patients was 46 (74.2%) and 16 (25.8%), respectively. The possible AD group (n = 27) comprised 13 females (48.1%) and 14 males (51.9%) with a mean age of 73.7±8.4 years. To support the clinical categorization with neurochemical biomarker data, for the majority of the participants, the CSF concentrations of Aβ_1–42, Aβ_1–40, total tau (T-tau), andphospho-tau 181 (P-tau) were measured in technical duplicates by commercially available ELISA kits in the Laboratory of Clinical Neurochemistry and Neurochemical Dementia Diagnostics, Department of Psychiatry and Psychotherapy, University of Erlangen-Nuremberg, Germany. The ELISA kits for the measurements of Aβ_1–42, T-tau, and P-tau were obtained from Innogenetics (Ghent, Belgium) or Fujirebio Europe and the Aβ_1–40 ELISA from IBL Japan or IBL International (Hamburg, Germany). For all subjects, we determined the CSF levels of Aβ₃₈, Aβ₄₀, and Aβ₄₂ by the MSD multiplex assay as described below. The available data regarding the CSF concentrations of Aβ_1–42, T-tau, P-tau, and the Aβ_1–42/Aβ_1–40 ratio as well as our own MSD measurements, were taken into account to support the categorization into the diagnostic groups probable AD (including early disease stages), possible AD and DC.

Electrochemiluminescence multiplex immunoassay for the simultaneous measurements of Aβ₃₈, Aβ₄₀, and Aβ₄₂

The measurements of the Aβ peptides Aβ₃₈, Aβ₄₀, and Aβ₄₂ were performed with the commercially available Mesoscale Discovery (MSD) V-Plex Aβ panel 1 (6E10) multiplex assay kit according to the protocol provided with the kit. Calibrator peptide dilutions were prepared in Diluent-35 (supplied with the assay kit) according to the manufacturer’s instructions. Aliquots of the CSF samples were thawed and then kept on ice until use. The detection antibody solution for one plate was prepared by mixing 60 μL of the 50×SULFO-TAG anti-Aβ 6E10 stock solution with 30 μL of Aβ₄₀ blocker (provided with the assay kit) and 2910 μL of Diluent-100. PBS-T wash buffer was prepared by dissolving 9.55 g of Instamed PBS-Powder (PBS Dulbecco w/o Mg⁺⁺, w/o Ca⁺⁺, Biochrom GmbH, Berlin, Germany) in 1 L of ultrapure water plus 0.5 mL of Tween-20. CSF samples were diluted prior to the measurements with Diluent-35. For diluting calibrator peptides and CSF samples, 1.5 mL Protein LoBind Tubes PCR clean, (Eppendorf AG, Germany) were used.

Assay procedure: The assay plates were blocked for 60 min under shaking (600 rpm) at 21°C after adding 150 μL of Diluent-35 to each well and sealing the plate with an adhesive plate seal. The blocking solution was decanted, and the plates were washed 3 times with 150 μl of PBS-T per well. Then, 25 μl of detection antibody solution was pipetted into each well. After addition of 25 μL per well of diluted sample, calibrator or Diluent-35 (blanks), the plates were covered with an adhesive plate seal and incubated at 21°C for 2 h on a shaker at 600 rpm in the dark. Then, the sample mixtures were decanted, and each well was washed 3 times with 150 μL of PBS-T. Finally, 150 μL of 2×Read Buffer T was added to each well. The electrochemiluminescent signals were recorded on a MSD Sector Imager 6000 or a MSD Quickplex reader and analyzed with the MSD Compass Software (version 3 or 4). The standard curves were calculated automatically with the default weighted four parameter logistic fit (FourPL). The lower limit of detection (LLOD) was calculated for each assay plate as the lowest concentration producing a signal 3 standard deviations above the blank (zero calibrator) with the option “use minimum error estimates” activated. The lower limit of quantification (LLOQ) was defined as the lowest concentration generating a signal 10 standard deviations above the blank.

Partial assay validation

Repeatability and reproducibility

To determine the within-plate (intra-assay) repeatability, 5 different CSF samples were diluted 16 fold with Diluent-35 and analyzed on a single MSD V-Plex assay plate in 9 technical replicates each. For each one of the three Aβ peptides under investigation and for the Aβ₄₂/Aβ₄₀ ratio, the coefficient of variation (% CV, relative standard deviation) was calculated. To assess the inter-plate imprecision (within laboratory reproducibility), five different CSF samples were measured in duplicates after 16 fold dilution on eight different assay plates on eight different days. Additionally, the same 5 CSF samples were tested in 4 technical replicates on a ninth assay plate. The tests were performed with assay kits from a single lot within 12 months. The mean back-calculated CSF concentrations of Aβ₃₈, Aβ₄₀, and Aβ₄₂ for each plate were used to calculate overall means, standard deviations and coefficients of variation for the three Aβ peptides for each individual CSF sample. Finally, the mean inter-assay coefficient of variation for each of the three Aβ peptides was calculated by averaging the % CVs of the five individual CSF samples. To furthermore estimate the reproducibility of the Aβ measurements on a larger scale (between laboratory plus between lot reproducibility), we compared the Aβ concentrations and the Aβ₄₂/Aβ₄₀ ratio in 5 CSF samples determined with assay kits from two different lots and measured in two different laboratories.

Spike recoveries

Spike recoveries were assessed with 5 different individual CSF samples that were measured in duplicates after 16 fold dilution without (neat) and with added calibrator peptides Aβ₄₀, Aβ₃₈, and Aβ₄₂ at three different spike levels. The resulting theoretical concentrations of the spikes in the diluted samples that were measured were 250, 500, and 1500 pg/mL for Aβ₄₀, 175.7, 351.4, and 1054.1 pg/mL for Aβ₃₈, and 22.5, 45 and 135.1 pg/mL for Aβ₄₂. The spike recoveries were calculated according to the following formula, in which c stands for concentration [11]. % recovery = ( c spiked sample - c neat sample ) theoretical c spike * 100

Assessment of Aβ₃₈, Aβ₄₀, Aβ₄₂, and total Aβ concentrations in human CSF samples in a clinical cohort

The CSF concentrations of the three carboxyterminal Aβ peptide variants were measured with the MSD V-Plex / Aβ-panel assay kit in a cohort of n = 203 subjects. All measurements were done in duplicate reads after 16 fold sample dilution in Diluent-35 as described above. The measurements were completed within 12 months with assay kits from a single lot. In total, 9 assay plates were used. For 190 out of these subjects, we additionally measured total CSF Aβ with a related custom MSD singleplex assay employing antibodies mAb 6E10 for capture and mAb 4G8 for detection [17]. For the determination of total Aβ in duplicate reads, the CSF samples were diluted 16 fold with 1% Blocker A in Tris wash buffer (both provided with the assay kit).

Statistics/data evaluation

For statistical evaluation we used Graph Pad Prism 6 and R version 3.1.3 (2015-03-09)–“Smooth Sidewalk”. To test for normally distributed variables we used the Shapiro-Wilk test. Center effects and age effects were tested with ANalysis Of VAriance (ANOVA). The centers were treated as categorical values while the age was considered to be a continuous numerical variable.

For linear modeling of CSF biomarkers we applied a Deming regression (MethComp package Version 1.22.2) since we wanted to account for errors in observations on both the x- and the y-axis. For simultaneous fitting of two normal distributions we used the flexmix package (Version 2.3-13).

Age and center correction

The participants of the study were recruited at two centers, the LVR-Clinics and the Memory Clinic at the Elisabeth Hospital, both located in Essen, Germany. The composition of the study cohort is summarized in Table 1. While there was no statistically significant difference between the centers in terms of the diagnoses (Fisher test p-value: 0.14), significant differences regarding age (t-test p-value: 6.81e-07)) and gender (Fisher test p-value: 0.006) were detected. Age and center effects were superpositioned. The subjects from the Elisabeth hospital were older on average than those from the LVR-Clinics, and the percentage of females was higher. Age and center effects were corrected for simultaneously. To this end we applied a linear model considering both effects concurrently: y = log 2 ( x ) y = a + b * AGE + c * CENTER + ɛ

If age and center effects are corrected (b = 0, c = 0) the slope of the linear model (a) equals the mean of the data. Therefore, the age and center corrected values (indicated by the subscript letter _c as e.g. in Aβ_42c) are calculated as: y c = mean ( y ) + ɛ x c = 2 yc

This method was applied to each variable separately (except for Aβ_42c/Aβ_40c ratios, which were calculated after correction). The correction was performed on a logarithmic scale (log2(n)) in order to use a correction factor rather than an offset (we assume a multiplicative error).

Impact of sample dilution, estimation of minimum required dilution and assessment of parallelism

Five different CSF samples were measured in single reads undiluted and after 2, 4, 8, 16, 32, and 64 fold dilution. The CSF concentrations of the three Aβ peptides and the ratio of Aβ₄₂/Aβ₄₀ were back-calculated for each CSF sample and for each dilution and plotted against the respective dilution factor (Fig. 1). For some, but not all of the tested CSF samples, the back-calculated Aβ concentrations and the Aβ₄₂/Aβ₄₀ ratios varied substantially between different dilutions indicating the presence of interfering substances producing so called “matrix effects”. It appears that in most cases, 8–16 fold dilution of individual CSF samples should be sufficient to essentially neutralize the matrix effects (minimum required dilution, MRD) and at the same time allow for reliable measurements of all three Aβs in human CSF within the detection range of the assay. Accordingly, all subsequent CSF measurements were done with 16 fold prediluted CSF samples.

From the same assay data, parallelism was evaluated with a Logit-log model [15] which served to linearize the relations between relative dilution factors and the readout. For all three Aβ peptides under investigation, the calculated slopes of the serial sample dilutions did not differ statistically significantly from those of the respective standard curves (Supplementary Figure 1) and passed a predefined acceptance criterion of 85–115% in range (Table 2).

Lower limits of detection (LLODs) and lower limits of quantification (LLOQs)

For each one of the three Aβ peptides and for each assay plate, the lower limit of detection (LLOD) was calculated automatically with the Discovery Workbench software from the standard curves as the lowest concentration producing a signal 3 standard deviations above the blank (zero calibrator) with the option “use minimum error estimates” activated. The lower limit of quantification (LLOQ) was calculated as the lowest concentration generating a signal 10 standard deviations above the blank. The mean LLODs (mean±SD calculated from n = 11 assay plates) were 17.4 pg/mL ± 3.8 for Aβ₃₈, 17.7 pg/mL ± 4.8 for Aβ₄₀, and 0.36 pg/mL ± 0.1 for Aβ₄₂. The corresponding mean LLOQs were 35.7 pg/mL ± 7 (Aβ₃₈), 37.1 pg/mL ± 9.3 (Aβ₄₀), and 0.93 pg/mL ± 0.2 pg/mL (Aβ₄₂).

Repeatability and reproducibility

Five CSF samples were analyzed after 16 fold dilution with Diluent-35 in 9 technical replicates on one assay plate. The mean coefficient of variation (intra-plate % CV, relative standard deviation) was 2.9% ± 1.4 (mean±SD) for Aβ₃₈, 3.6% ± 0.9 for Aβ₄₀, and 2.2% ± 0.8 for Aβ₄₂. It should be noted that this particular experiment was performed by a different experimenter and in a different laboratory than all others reported here. The assay kit was from a different lot than all others and the signals were recorded on a different machine (MSD Quickplex versus Sector Imager 6000). The mean intra-plate % CV for the Aβ₄₂/Aβ₄₀ ratio was 4.2% ± 1.1 (mean±SD).

To evaluate the repeatability between different assay plates but within one laboratory (within laboratory inter-plate variation), five different CSF samples were diluted 16 fold and measured in duplicates on eight different assay plates on eight different days. Furthermore, the same 5 CSF samples were tested on a ninth assay plate with 4 technical replicates, each. The tests were performed with assay kits from a single lot within 12 months. The back-calculated CSF concentrations of Aβ₃₈, Aβ₄₀, and Aβ₄₂ for each plate (means calculated from 2 or 4 technical repeats, respectively) were used to calculate overall means, standard deviations and coefficients of variation for the three Aβ peptides for each CSF sample. Finally, the mean inter-assay coefficient of variation for each of the three Aβ peptides was calculated by averaging the % CVs of the five individual CSF samples. One measurement of a single sample was excluded from the analysis, because one of the two technical replicates was below the detection range. The calculated inter-assay coefficients of variation were 9.4% ± 4, 12.9% ± 4.6, and 9.6% ± 2 for Aβ₃₈, Aβ₄₀, and Aβ₄₂, respectively. The overall % CV of the Aβ₄₂/Aβ₄₀ ratios was 7.8% ± 2.8(mean±SD).

To furthermore estimate the reproducibility of the Aβ measurements on a larger scale, we compared the Aβ concentrations in 5 CSF samples determined with assay kits from two different lots in two different labs. The observed between-laboratory plus between-lot % CV values for the three different Aβ peptides in individual CSF samples ranged from 1–34%. The mean % CVs were 13.2±9.9 (mean±SD) for Aβ₃₈, 17±10.5 for Aβ₄₀, and 16.9±14.2 for Aβ₄₂ (Table 3). The respective % CVs of the Aβ₄₂/Aβ₄₀ ratios were substantially smaller.

Spike recoveries

Spike recoveries were assessed with 5 different CSF samples and at three different spike levels. The results are summarized in Table 4. The averaged overall spike recoveries were 117.7% ± 10.5, 114.8% ± 12.3, and 105.2% ± 12.8 (mean±SD) for Aβ₃₈, Aβ₄₀, and Aβ₄₂, respectively. In summary, the overall spike recoveries for Aβ₄₀ and Aβ₄₂ were within the acceptance range of 85–115% while that of Aβ₃₈ was slightly out of range.

MSD measurements of the CSF levels of Aβ₃₈, Aβ₄₀, Aβ₄₂, and total Aβ in a clinical cohort

In the next step, the MSD Aβ triplex assay was applied to the assessment of Aβ₃₈, Aβ₄₀, and Aβ₄₂ in 203 human CSF samples. All multiplex measurements were done within 12 months with assay kits from a single lot. In total, 9 assay plates were used. For a subset of the CSF samples, Aβ_1–42 and Aβ_1–40 concentrations were also available that had been measured by ELISAs under routine conditions in the ISO certified Laboratory for Clinical Neurochemistry and Neurochemical Dementia Diagnostics at the Department of Psychiatry and Psychotherapy, University of Erlangen-Nuremberg. The Aβ₄₂ levels measured with the MSD triplex assay were statistically significantly correlated with the Aβ_1–42 concentrations according to ELISA (n = 179 data points) (Fig. 2A). However, in general the ELISA reported higher concentrations than the MSD assay suggesting a systematic difference between the two assays (Fig. 2B). Similarly, ELISA measurements of Aβ_1–40 (n = 158 data points) correlated with Aβ₄₀ MSD data (Fig. 2C). Again, in most cases the ELISA reported higher concentrations of Aβ₄₀ than the MSD triplex assay suggesting a systematic difference (Fig. 2D).

The baseline statistics for the CSF concentrations of Aβ₃₈, Aβ₄₀, Aβ₄₂ according to the MSD multiplex assay, total CSF Aβ measured by a custom MSD singleplex assay and ELISA data on CSF levels of Aβ_1–40, Aβ_1–42, P-tau, and T-tau are summarized in the upper part of Table 5.

The total Aβ CSF concentrations followed a normal distribution, while neither Aβ₃₈ nor Aβ₄₀ nor Aβ₄₂ passed the normality test. Instead, Aβ₄₂ followed a log-normal distribution (p-value on log scale: 0.955). We thus log-transformed (log2) the values before applying additional statistics. The clinical sample under investigation comprised patients from two different sites, the LVR-Clinics and the Memory Clinic at the Elisabeth Hospital, both located in Essen, Germany. The MSD data showed considerable center effects (upper part of Table 5) and age effects (data not shown). This was also the case for the available Aβ_1–40 and Aβ_1–42 concentrations according to ELISA. The corrected values after simultaneous removal of age and center effects (indicated by the symbol _c as, e.g., in Aβ_42c) are summarized in the lower part of Table 5. The age- and center corrected CSF levels Aβ_38c, Aβ_40c and Aβ_42c determined with the MSD triplex assay did not show significant effects for the used assay plate, neither did the Aβ_42c/Aβ_40c ratio (Supplementary Figure 2).

The Aβ_38c and Aβ_40c concentrations in CSF were strongly and statistically highly significantly correlated with each other and also with total Aβ_c (Fig. 3A-C). The correlations of Aβ_42c with Aβ_38c, Aβ_40c, and total Aβ_c, respectively were substantially weaker but still statistically highly significant (Fig. 3D-F). When the analysis was carried out in the DC subgroup alone, Aβ_42c correlated more strongly with Aβ_38c, Aβ_40c, and total Aβ_c (Supplementary Figure 3). Correlations of similar strength and statistical significance were also observed with the uncorrected CSF Aβ levels (data not shown).

Calculation of assay specific cutpoints

The clinical sample investigated here comprised 203 subjects in total. For 179 out of these, a clinical diagnosis was available at the time of the statistical data analysis. These included 62 patients diagnosed with probable AD and 90 subjects categorized as improbable AD (non-AD disease controls, DC). For defining assay specific cutpoints regarding the age- and center corrected MSD measurements of CSF Aβ_42c and Aβ_42c/Aβ_40c ratios, we applied three different strategies. In the first approach, the correlations between our MSD measurements and the available ELISA data obtained from the Laboratory for Clinical Neurochemistry and Neurochemical Dementia Diagnostics at the University of Erlangen-Nuremberg were used to translate the diagnostic ELISA cutpoints of the reference laboratory into cutpoints for the age- and center corrected MSD data. Figure 4A shows the correlation between Aβ_1–42c ELISA data and Aβ_42c levels determined by MSD assay (in both cases the age- and center corrected data were used). Linear curve fitting was done by Deming regression. Based on the linear regression, the diagnostic Aβ_1–42 ELISA cutpoint of 600 pg/mL of the clinical Laboratory in Erlangen translated into a corresponding cutpoint of 350 pg/mL for Aβ_42c for the MSD assay. By the same approach, the Erlangen cutpoint of 0.05 for the Aβ_1–42/Aβ_1–40 ratio (measured by ELISAs) translated into a cutpoint of 0.067 (Aβ_42c/Aβ_40c) for the MSD triplex assay. The correlation between the Aβ_42c/Aβ_40c ratios (MSD) and the Aβ_1–42c/Aβ_1–40c ratios according to the ELISA measurements is shown in Fig. 4B. In a second approach, ROC curves were calculated for Aβ_42c (MSD) and for the Aβ_42c/Aβ_40c ratio (MSD) for the single value classification of DC (n = 90) versus probable AD (n = 62) (Fig. 4C). The cutpoints obtained with a defined threshold of 90% specificity were 350 pg/mL for Aβ_42c and 0.062 for the Aβ_42c/Aβ_40c ratio. In a third approach, a cutpoint of 0.074 for the MSD Aβ_42c/Aβ_40c ratio was obtained from the intersection of the two normal distributions fitted to the bimodal distribution of the Aβ_42c/Aβ_40c ratio in the study cohort (n = 197) (Fig. 4D).

The ROC analysis summarized in Fig. 4C suggests an improved diagnostic performance of the Aβ_42c/Aβ_40c ratio as compared to the CSF levels of Aβ_42c per se. This would be in agreement with findings from several published studies [12, 18–20]. It should be noted, however, that in the present study, whenever available, CSF biomarker data including the Aβ_1–42/Aβ_1–40 ratio, Aβ_1–42 and, in some cases, also the MSD Aβ₄₂/Aβ₄₀ ratio had been considered for the biomarker supported clinical diagnosis. Thus, unbiased conclusions regarding the diagnostic performance of any of these biomarkers cannot be drawn from this data. To assess the distribution of those cases, in which either only CSF Aβ_42c or only the Aβ_42c/Aβ_40c ratio were below their respective cutpoints, the whole patient cohort was divided into 19 equally sized and partially overlapping groups according to the age and center corrected CSF levels of Aβ₄₀ (Aβ_40c). For each of the 19 quantiles we classified the subjects according to Aβ_42c (threshold: 350 pg/mL) and the Aβ_42c/Aβ_40c ratio (90% specificity threshold: 0.062). The distribution of subjects with (i) a matching classification, (ii) Aβ_42c below cutpoint but normal Aβ_42c/Aβ_40c ratio and (iii) Aβ_42c above but Aβ_42c/Aβ_40c below their respective cutpoints is shown in Fig. 5. In general, those subjects with low Aβ_42c but normal Aβ_42c/Aβ_40c ratios showed relatively low levels of Aβ_40c. In contrast, subjects with normal Aβ_42c but low Aβ_42c/Aβ_40c ratios were more often found among those with relatively high Aβ_40c than those with low Aβ_40c.

Conclusions

Taken together, our observations indicate that the MSD multiplex assay is well suited for the measurement of Aβ₃₈, Aβ₄₀, and Aβ₄₂ in human CSF samples for research purposes. To effectively neutralize matrix effects, 16 fold sample dilution seems to be appropriate. The CSF concentrations of Aβ₄₀ and Aβ₄₂ according to the MSD assay were statistically significantly correlated with ELISA measurements of Aβ_1–40 and Aβ_1–42 in CSF. However, direct comparison of the data obtained with the different assays does not seem appropriate, due to apparent systematic differences between the assay formats. Accordingly, assay specific diagnostic cutpoints have to be defined. Our findings provide further support for the idea that the reported superior diagnostic performance of the Aβ₄₂/Aβ₄₀ ratio as compared to CSF Aβ₄₂ alone is due to normalization to Aβ₄₀, which can be regarded a surrogate marker for total Aβ levels.