Phosphatidylethanolamine Binding Protein 1 (PEBP1) in Alzheimer’s Disease: ELISA Development and Clinical Validation

Selection of candidate AD biomarker

The list of dementia biomarker candidates available within the MIRIADE consortium was thoroughly analyzed by means of bioinformatic online tools such as UniProt [14], CSF Proteome Resource [16], Human Protein Atlas [21], STRING [22], and Reactome [23]. A holistic process leading to selection of PEBP1 as candidate AD biomarker is described in detail by (Mavrina et al., unpublished data).

Human CSF samples

All the CSF samples used in this study were collected and biobanked at the Section of Neurology, Department of Medicine and Surgery, University of Perugia (Perugia, Italy).

CSF samples were collected following the same standard operating procedures (SOP) throughout the study and international guidelines [24]. LP was performed between 8:00 and 10:00 a.m. CSF was collected into sterile polypropylene tubes and centrifuged for 10 min at 2000× g (RT). 0.5 mL aliquots (72.730.007, Sarstedt, Germany) were immediately frozen at –80°C pending analysis. Demographic and clinical data of each patient were confidentially stored in an online electronic database.

All patients and neurological controls underwent a comprehensive clinical and neuropsychological evaluation. Diagnosis of AD was made according to the NIA-AA criteria [9] and supported by the analysis of core CSF AD biomarkers (Aβ₄₂, Aβ₄₀, Aβ₄₂/Aβ₄₀ ratio, t-tau, p-tau) measured with the Lumipulse® G600II (Fujirebio, Japan). According to their cognitive performance and to the Clinical Dementia Rate (CDR) scale, patients were stratified in MCI-AD (0.5≤CDR < 1.0) or AD-dem (CDR ≥1.0) groups. Patients qualified as OND (other neurological disorders) were adult patients, diagnosed with minor neurological diseases, other than inflammatory or degenerative disease of the CNS or of the peripheral nervous system. Only properly stored and unthawed CSF aliquots were used for the analysis.

Two CSF pools were generated as unified testing material for the assay development. One pool consisted of selected CSF samples collected from AD patients at the stage of dementia (“CSF AD pool”). The other pool (“CSF Control pool”) consisted of CSF samples collected from patients cognitively intact, but affected by neurological, not neurodegenerative disorders (i.e., headache, peripheral neuropathy, etc.), showing no cognitive impairment after at least a 2-year follow-up.

For the clinical validation, CSF samples collected between 2016 and 2021 were selected from the CSF Biobank (age, gender, and the CSF levels of the classical AD biomarkers of the patients included in the selected cohorts are reported in the Supplementary Table 1).

The study was approved by the local Ethical Committees; written informed consent was obtained from all the subjects included in this study. All the procedures were performed following the Helsinki Declaration.

Antibody testing by immunoblot

The ability of three human anti-PEBP1 antibodies (sc-376925, monoclonal, mouse, Santa Cruz Biotechnology, USA; 372100 monoclonal, mouse, Invitrogen, USA; HPA063904, polyclonal, rabbit, Sigma-Aldrich, USA) to detect the target protein was tested by western blot (WB) on different matrices (i.e., neat and 5x-concentrated CSF pools, human brain homogenate, and human recombinant PEBP1).

Part of each pool described above, underwent 5x concentration with use of the Vivaspin® (Sartorius, Germany) filter (4 mL, 5 kDa). The total protein concentration of neat and concentrated CSF pools was estimated based on the Bradford assay.

A human brain homogenate sample (HBH) of an AD patient (frontal lobe) [25] was resuspended in 1:5 weight/volume ratio with RIPA lysis buffer. The sample underwent a manual homogenization and then, sonication (3×15 s). The sample was incubated on ice for 30 min and then centrifuged for 10 min at 16,000× g. Supernatant was collected and the total protein concentration was measured by the Bradford assay.

Finally, 100 ng human recombinant protein PEBP1 (AR09265PU-N, OriGene, USA) was prepared by diluting it in 20 mM Tris-HCl (pH 8), 150 mM NaCl.

For each anti-PEBP1 antibody (sc-376925/372100/HPA063904) the WB experiment was performed separately. A 12% polyacrylamide gel was prepared and CSF samples, HBH and recombinant protein samples were run in reducing conditions. The samples were transferred from the gel to a nitrocellulose blotting membrane (SF110B, Himedia, India). The quality of the transfer was evaluated by a reversible Ponceau S staining. The membrane was blocked and subsequently probed overnight at 4°C (shaker) with the selected primary antibodies (sc-376925/372100/HPA063904, 1:1000) diluted in 5% bovine serum albumin (BSA), 0.02% NaN₃ in Tris-buffered saline with 0.1% Tween®20 (Sigma-Aldrich, USA) (TBST) with addition of a phenol red as a pH indicator. After the incubation, the membrane was washed with TBST and the secondary antibodies diluted 1:5000 in a blocking buffer were applied for 1 h at RT. Depending on the used primary antibodies, goat anti-mouse (170–6516, Bio-Rad, USA) (for sc-376925 and 372100) or goat anti-rabbit IgG-HRP (170–6515, Bio-Rad, USA) (for HPA063904) were added. Signal development was performed with use of the enhanced chemiluminescence (ECL) solution (SuperSignaltrademark West Pico Plus, ThermoFisher Scientifictrademark, USA) and the radiographic film (Amershamtrademark Hyperfilmtrademark, GEhealthcare, USA).

For each anti-PEBP1 antibody (sc-376925/372100/HPA063904), dot blot analysis was performed to evaluate the potential unspecific binding of human serum albumin (HSA), the most abundant CSF protein. Four types of samples, 50 ng of PEBP1 recombinant protein (AR09265PU-N), 50 ng HSA, 100 ng HSA, and blank (Phosphate-buffered saline, PBS), were applied on the nitrocellulose membrane. The membrane was left for drying and then blocked with 5% milk powder in TBST. The membrane was incubated with the selected primary antibodies (sc-376925/372100/HPA063904, 1:1000) for 1 h at RT. After the incubation with the appropriate secondary antibodies (170–6516/170–6515, Bio-Rad 1:5000 in the blocking buffer), the signal was developed on the radiographic film with use of the ECL solution.

PEBP1 sandwich ELISA development

The ELISA assay was developed and validated following recommendations provided by [26, 27].

In addition to the WB experiments, evaluating performance of the antibodies in reducing conditions, all the antibodies were tested for the ability to recognize PEBP1 in native conditions. To perform this test, PEBP1 recombinant protein (AR09265PU-N) was coated in 96-wells plate (655061, Microlon®, high binding, Greiner Bio-One, Austria) at final concentration of 2μg/mL and 0.2μg/mL.

Six different antibody pairs were tested as the capture and detection antibody (see Supplementary Table 2). The best concentration of both capture and detection antibody was estimated by testing range of different concentrations of each antibody (2μg/mL, 1μg/mL, and 0.5μg/mL) and by comparison of the obtained signal-to-noise ratio.

To improve the performance of the assay, we compared signal obtained with use of PBS-based versus TBS-based buffers as well as different blocking buffers (1% milk powder, 3% milk powder, 1% BSA, 3% BSA). To decrease the background signal, we tested assay diluents and blocking buffers prepared with use of three different BSA reagents (03117332001, Roche, Switzerland, A9647-100G, Sigma-Aldrich, USA, P6154-100GR, Biowest, France). Diverse sample incubation times, various concentrations of the secondary antibody (goat anti-mouse-HRP, Bio-Rad, 170–6516), as well as addition of a detergent (Tween®20) were tested for the ability to increase the CSF signal.

PEBP1 ELISA protocol

The polyclonal rabbit HPA063904 antibody was selected as the most suitable capture antibody. The 96-wells plate was coated with the capture antibody at the final concentration of 1μg/mL (100μL/well) in a 100 mM bicarbonate buffer, pH 9.5, overnight, at 23°C. Empty wells, not predestined for the further use, were coated with 100μL of the 100 mM bicarbonate buffer, pH 9.5. The plate was sealed and incubated overnight at 23°C.

After the overnight incubation, the plate underwent a single wash with 300μL of an in-house prepared PBS buffer, pH 7.4 and subsequently was blocked with 1% BSA-PBS (BSA –Roche, 03117332001) for 1.5 h at 23°C. After blocking, the plate was washed three times with 300μL of PBST buffer (PBS with 0.05% Tween®20). The 7-point standard curve, ranging from 100 ng/mL to 1.56 ng/mL was prepared by diluting the human PEBP1 recombinant protein in 0.1% BSA-PBS-0.05% Tween®20 (see Supplementary Table 3). Blank sample consisted of 0.1% BSA-PBS-0.05% Tween®20. Undiluted CSF samples, standard curve dilutions, and blank were added to the plate (100μL/well) and underwent 2 h incubation at 23°C and 400 rpm. Subsequently, the samples were removed, and all the wells were washed three times with 300μL of PBST buffer. 100μL of the detection antibody sc-376925 at the final concentration of 1μg/mL was added to each well and underwent 1 h incubation at 23°C and 400 rpm. After the incubation, the wells were washed three times with 300μL of PBST buffer. Goat anti-mouse-HRP antibody (Bio-Rad, 170–6516) diluted 1:3000 in 0.1% BSA-PBS-0.05% Tween®20 was added to each well (100μL/well) and the plate was incubated for 1 h at 23°C and 400 rpm. Finally, the plate was washed four times with 300μL of PBST buffer and 100μL of tetramethylbenzidine (TMB; T2885, Sigma-Aldrich, USA) solution was added to each well and incubated for 10 min at 23°C and 400 rpm in the dark.

The reaction was stopped by addition of 100μL of 1 M H₂SO₄ to each well. The plate was briefly mixed on the shaker for approximately 1 min and the optical density (OD) was read at 450 nm (reference wavelength 630 nm) in the Clariostar (BMG Labtech, Germany) plate reader. The 4-parameter logistic model was applied to generate a standard curve. Concentrations of CSF samples were interpolated based on the blank subtracted ODs related to the standard curve. Graphs were generated with use of the GraphPad Prism software (USA).

Technical validation of the developed assay

The ELISA assay underwent thorough technical validation. This consisted in the estimation of the assay range, intermediate precision (inter-assay reproducibility within the laboratory), repeatability (intra-assay), sensitivity (Lower Limit of Quantification, LLOQ; Upper Limit of Quantification, ULOQ), and sample stability (freeze-thaw, prolonged incubation at diverse conditions). Range of the standard curve was established, and the intra- and inter-assay variabilities were assessed based on % coefficient of variation (CV%) of the replicates. In order to monitor performance of the subsequent measurements, an internal control was included in each run.

Evaluation of the preanalytical factors influencing PEBP1 stability

A freeze-thaw experiment was conducted using five consecutively collected CSF samples. Each sample was divided into four 250μl aliquots. One aliquot of each freshly collected sample was immediately stored at –80°C pending analysis, while remaining aliquots of each sample underwent 2, 3, or 4 freeze (30 min at –80°C) and thaw (30 min at RT) cycles. Subsequently, all the samples were measured by the developed ELISA assay.

Sample stability was controlled in a dedicated experiment, employing a CSF pool generated from three individual patients’ samples. CSF pool was divided into ten identical 250μl aliquots. One of the aliquots was stored at –80°C immediately after generation of the pool and served as a reference. Remaining aliquots were subjected to incubation at RT, +4°C or –20°C for 6 h, 24 h, and 48 h (Fig. 1), after which were stored at –80°C pending analyses. Upon completion of the incubation of all the samples, the PEBP1 concentration was measured by the ELISA assay.

OPEN IN VIEWER

Fig. 1

A schematic illustration presenting the workflow of the sample-stability experiment.

Statistical analysis

Continuous variables are represented as the mean±standard deviation (SD). Statistical analysis was performed with use of the GraphPad Prism software (USA). Shapiro-Wilk test was applied to assess data normality. The data did not exhibit normal distribution, hence non-parametric tests were applied. We implemented Mann-Whitney test for two-group comparisons (OND and MCI-AD) and Kruskal-Wallis test for multiple group comparisons (OND, MCI-AD, AD-dem) followed by Dunn’s test for multiple comparisons correction. When non-parametric tests were not applicable, data were log-transformed and analyzed with parametric two-way ANOVA test, followed by Dunnet’s test for multiple comparisons. Correlations between PEBP1 concentration in CSF and core AD biomarkers or PEBP1 concentration and age were assessed by Spearman’s correlation coefficients. A p-value < 0.05 was considered statistically significant.

Assay development

To evaluate the diagnostic performance of PEBP1 we developed an in-house ELISA.

Three different antibodies recognizing different regions of the PEBP1 were selected. All the antibodies exhibited specific recognition of the protein of interest both in reducing conditions (WB) and in native conditions (direct ELISA), while no cross-reactivity with HSA was observed for any of the antibodies (dot blot) (data not shown).

Once the assay was developed (capture antibody HPA063904, detection antibody sc-376925), we conducted several experiments aimed at its optimization. Comparison of PBS-based versus TBS-based assay buffers resulted in a similar reading of the background signal. While the initial premises pointed to possible dependence of CSF signal on Tween®20 addition, the experiments designed to evaluate such effect did not confirm influence of detergent addition on the intensity of CSF signal. Assessment of the assay performance with use of four different blocking buffers (1% milk powder/3% milk powder/1% BSA/3% BSA in PBS buffer) pointed to 1% BSA-PBS as the most effective blocking solution. No difference in blocking effectiveness was observed among buffers prepared with BSA reagent provided by three different vendors (03117332001, Roche, Switzerland, A9647-100G, Sigma-Aldrich, USA, P6154-100GR, Biowest, France) (data not shown).

The range of the standard curve extended from 100 ng/mL to 1.56 ng/mL. LLOQ and ULOQ of PEBP1 detection in CSF were, respectively, 4.7 ng/mL and 100 ng/mL. The CV% for the inter- and intra-assay variability were, respectively, 13.54% and 3.38%.

Analysis of pre-analytical factors for PEBP1 measurement in CSF

To analyze pre-analytical factors which may influence PEBP1 levels in CSF, we performed an experiment evaluating PEPB1 concentration of the consecutively collected CSF samples, subjected to a number of freeze-thaw cycles. We observed a significant increase (p < 0.01 at 4th cycle) of PEBP1 concentration, dependent on the number of freeze-thaw cycles (see Supplementary Figure 1). Such observation led us to perform another experiment, evaluating concentration of PEBP1 in CSF samples being subject to diverse incubation times and conditions. We observed that the samples subjected to prolonged incubation at RT, +4°C or –20°C instead of –80°C, exhibited a visibly elevated concentration of PEBP1 (Fig. 2), while compared to the reference sample (p≤0.0001 for –80°C (7.38 ng/mL) versus RT 6 h (22.03 ng/mL), p≤0.05 for –80°C (7.38 ng/mL) versus +4°C 6 h (11.82 ng/mL), p≤0.05 for –80°C (7.38 ng/mL) versus –20°C 6 h (11.75 ng/mL), p≤0.0001 for –80°C (7.38 ng/mL) versus RT 24 h (50.58 ng/mL), p≤0.0001 for –80°C (7.38 ng/mL) versus +4°C 24 h (18.71 ng/mL), p≤0.01 for –80°C (7.38 ng/mL) versus –20°C 24 h (13.78 ng/mL), p≤0.0001 for –80°C (7.38 ng/mL) versus RT 48 h (134.58 ng/mL), p≤0.0001 for –80°C (7.38 ng/mL) versus +4°C 48 h (25.23 ng/mL), non-significant for –80°C (7.38 ng/mL) versus –20°C 48 h (7.18 ng/mL)).

OPEN IN VIEWER

Fig. 2

PEBP1 concentration in CSF pool sample undergoing incubation at different conditions. Change related to concentration of the CSF pool sample properly stored at –80°C. Two-way ANOVA test, followed by Dunnet’s test for multiple comparisons were applied after log transformation of the data. Actual values in pg/mL are reported in the figure. ^*p≤0.05, ^**p≤0.01, ^***p≤0.001, ^****p≤0.0001 versus reference sample (–80°C).

Diagnostic performance of PEBP1 for AD diagnosis

The first validation included a cohort of 25 patients consisting of 14 MCI-AD subjects (50% female, age (Mean±SD) = 68±6.1) and 11 OND subjects (45.5% female, age (Mean±SD) = 67±9.9). Although a strong trend was visible, no statistically significant difference was observed when comparing both groups (p = 0.137) (Fig. 3A). The MCI-AD group exhibited higher level of PEBP1 concentration in CSF (MCI-AD = 32.47±12.6 ng/mL, OND = 27.5±15.03 ng/mL).

OPEN IN VIEWER

Fig. 3

A) PEBP1 concentration (ng/mL) in CSF of OND and MCI-AD groups included in the first clinical validation (p = 0.137, assessed by Mann-Whitney test). B) PEBP1 concentration (ng/mL) in CSF of OND, MCI-AD, and AD-dem groups included in the second clinical validation (p = 0.644, assessed by Kruskal-Wallis test, followed by Dunn’s test). In boxplots, heights of boxes represent the interquartile range (from 25th to 75th percentile). The line in the middle of the box represents the median. Whiskers and outliers are plotted according to the Tukey’s method.

To confirm these results, a second validation was performed, with a larger cohort encompassing three diagnostic groups. For this analysis, 25 MCI-AD subjects (56% female, age (Mean±SD) = 73±5.5), 29 AD-dem subjects (72% female, age (Mean±SD) = 74±6.7), and 21 OND subjects (48% female, age (Mean±SD) = 56±15.7; 5 samples measured also in the first validation) were included. No statistically significant difference was observed among the groups (p = 0.644) (MCI-AD 22.41±12.17 ng/mL, AD-dem 19.10±7.29 ng/mL, OND 20.38±11.19ng/mL) (Fig. 3B).

Correlation analysis between CSF PEBP1 level and AD biomarkers (Aβ₄₂, Aβ₄₀, Aβ₄₂/Aβ₄₀, p-tau, t-tau) was then performed. CSF PEBP1 concentration had a significant, though weak, positive correlation with t-tau (ρ= 0.25, p < 0.05), p-tau (ρ= 0.23, p < 0.05), and Aβ₄₀ (ρ= 0.26, p < 0.05) when evaluated in the whole cohort (Fig. 4). Analysis of the individual groups showed that correlations were driven by the AD dementia group (PEBP1 versus t-tau ρ= 0.51, p < 0.05, PEBP1 versus p-tau ρ= 0.53, p < 0.05 and PEBP1 versus Aβ₄₀ ρ= 0.60, p < 0.01). Correlation between CSF PEBP1 and age, Aβ₄₂ or Aβ₄₂/Aβ₄₀ was not significant.

OPEN IN VIEWER

Fig. 4

Scatterplot of (A) CSF PEBP1 (ng/mL) versus t-tau (pg/mL) in the whole cohort, (B) CSF PEBP1 (ng/mL) versus p-tau (pg/mL) in the whole cohort, (C) CSF PEBP1 (ng/mL) versus Aβ₄₀ (pg/mL) in the whole cohort, (D) CSF PEBP1 (ng/mL) versus Aβ₄₂ (pg/mL) in the whole cohort. Correlations assessed by Spearman’s correlation coefficients. Lines are added only for the visualization purposes. Statistically significant correlation coefficients are reported for the whole cohort (in black) and subgroups (with the relative color).