In total, 300 participants were included in this study. They consisted of 149 cognitively normal individuals (CN group), 87 patients with amnestic mild cognitive impairment (MCI group), and 64 patients with clinically diagnosed AD dementia (DEM group). These individuals were recruited as part of the Korean Brain Aging Study for the Early diagnosis and prediction of Alzheimer’s disease (KBASE). All participants were given appropriate clinical and neuropsychological assessments according to the KBASE assessment protocol. The details of participant recruitment, clinical diagnosis criteria, and further information were described in our previous report (5).
All participants and (where applicable) their legal representatives read and confirmed the informed consent documents. This project was approved by the Seoul National University Hospital Institutional Review Board.
All participants underwent PiB-PET scans using a 3.0T PET-MR scanner (Siemens Healthineers, Erlangen, Germany). Briefly, each individual was injected intravenously with 555 MBq of [11C] PiB (450-610MBq) PET tracer, which enabled visualization of cerebral amyloid deposition. The degree of amyloid accumulation was calculated by the standardized uptake value ratio (SUVR), which was determined by the automatic anatomic algorithm. The four regions of interest (ROIs) were the lateral temporal, lateral parietal, posterior cingulate-precuneus, and frontal regions. If the SUVR value was 1.4 or higher for at least one of the four ROIs, the individual was defined as PiB-positive (PiB+). The details of the imaging protocols were described in our previous paper (5).
All fasting blood samples were collected at 9:00 AM. Whole-blood samples were gathered in K2 EDTA tubes (BD Vacutainer Systems, Plymouth, UK) and centrifuged at 700 g for 5 min at room temperature (RT). The supernatants were collected, the centrifugation step was repeated, and the tubes were stored at −80°C.
QPLEXTM Alz plus assay (QPLEXTM)
Our QPLEXTMkit utilized the Quantamatrix’s multiplex diagnostics platform (QMAP; Quantamatrix Inc., Seoul, Republic of Korea) with microdisk technology to perform multiplex analyses in a single well (6). This suspension bead array system uses graphically coded beads to expose antigens to the 3D environment. Briefly, human plasma samples were diluted in the diluent buffer and incubated with the coded beads and biotin-conjugated detection antibodies in the provided black 96-well plate for 90 minutes at RT on a shaking incubator at 1000 rpm. The immunocomplexes, including the coded beads, were washed twice with washing buffer on a Biotek-510 magnetic wash station (Biotek, VT, USA). Fifty microliters of diluted R-phycoerythrin-conjugated streptavidin were added to each well, and the plate was incubated for 15 minutes at RT on the same shaking incubator. After three washes, the immunocomplexes were resuspended in 100 μl of washing buffer by tapping. Collected immunocomplexes were analyzed automatically by the QMAPTM system; approximately 30 beads were used for the calculation of each biomarker concentration.
In-house ELISA (IH-ELISA)
ELISA plates were coated with capture antibodies diluted to a working concentration in Dulbecco’s modified phosphate-buffered saline solution (DPBS; Invitrogen, Carlsbad, CA, USA) and incubated at RT. The remaining binding sites were blocked with 1 mg/ml bovine serum albumin (Sigma Aldrich, St. Louis, MO, USA) in DPBS. Human plasma samples were diluted in the same buffer that was provided with the QPLEXTM kit and incubated with the primary antibodies immobilized in the wells. Streptavidin-conjugated horseradish peroxidase was added to the wells, and the plate was incubated again for 20 minutes. The immunocomplexes were detected with a chromogenic substrate solution and the reaction was terminated by the addition of 0.5 M HCl. Absorption was read at 450 nm using an ELISA plate reader (Biotek, Winooski, VT, USA).
All statistical analyses were performed using the Medcalc 17.2 software (Ostend, Belgium) and GraphPad Prism 8 (San Diego, CA, USA). Comparison analyses between two variables were conducted by independent t-test or analysis of covariance (ANCOVA) with correction for age and sex. Correlation analyses were performed using the Pearson’s correlation analysis method. To calculate the discriminatory power, sensitivity, and specificity for the biomarker panels, logistic regression, followed by receiver operating characteristic (ROC) curve analysis was performed. The basic formula of our algorithm used in our analyses was as follows:
(pi, predicted probabilities; an, coefficient values, e.g. a1 = 0.008 for Aꞵ40, a2 = -0.0066 for ACE, a3 = -0.0007 for LGALS3BP, a4 = 0.1322 for POSTN; C, constant, e.g. C = 1.2478; Each biomarker level of the samples was multiplied by coefficient values and pi was calculated)
The formulas, coefficients, and constants could be optimized since there were appropriate outliers and various logistic regression models. Multicollinearities were checked using the values of variance inflation factors (VIF). Finally, the accuracy of each model was calculated using a randomized sample selection method. Randomizing analysis was used to create random groups with even distributions of age (the variable for case identification was ‘age’). In other words, each group had almost same average age (Group 1, average 71.8, range 56 to 86; Group 2, 72.3, range 61 to 84; Group 3, 70.1, range 55 to 85; … etc.). After the randomized sample selection and regrouping, each group was validated by the logistic regression models to calculate the accuracy of each biomarker panel. The average accuracy was calculated as shown in Fig. 3. All statistical outliers were excluded from the cohort according to the Grubb’s double-side outlier test (p < 0.05).