Evaluating the diagnostic performance of six plasma biomarkers for Alzheimer’s disease and other neurodegenerative dementias in a large Chinese cohort

This cross-sectional study enrolled 668 participants, including 54 with amnesic MCI (aMCI), 191 with AD, 100 with PSP, 67 with FTD, 72 with DLB, and 184 HCs in the Department of Neurology, Xiangya Hospital, Central South University. Among them, 49 patients with aMCI or AD completed a longitudinal analysis. AD and aMCI were diagnosed based on the National Institute on Aging and Alzheimer’s Association criteria (2011) [26], FTD, DLB, and PSP were diagnosed according to their respective clinical criteria [27–30]. Among them, 237 patients completed PiB-PET or CSF biomarkers (Aβ42, Aβ40, t-tau, and p-tau181) detection and were divided into Aβ positive(+) & Aβ negative(-), tau positive(+) & tau negative(-) groups according to the 2018 NIA-AA criteria [31]. PSP was further subdivided into Richardson’s syndrome (PSP-RS) and PSP variant (PSP-v) according to international criteria [30]. FTD was further subdivided into behavioral variant FTD (bvFTD), non-fluent variant FTD (nfvFTD), and semantic variant FTD (svFTD) [32]. All HCs were recruited from the Liuyang community and showed no cognitive decline after 2 years of follow-up.

All patients underwent a comprehensive battery of neuropsychological tests, including the Mini Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and Clinical Dementia Rating (CDR) assessment. All HCs completed the MMSE assessments at baseline and the 1-year follow-up respectively.

T1-weighted MRI was performed using a 3T MRI scanner (Prisma, Siemens, Germany) with a 64-channel head/neck coil, using a magnetization prepared rapid gradient echo (MP-RAGE) sequence with the following parameters: repetition time, TR = 2,300 ms; echo time (TE) = 2.98 ms, 1 mm isotropic voxel size and 176 slices. The CAT12 toolbox (http://dbm.neuro.uni-jena.de/cat/↗) implemented in SPM12 running in MATLAB was used to estimate cortical thickness and grey matter volume. All images were checked with an internal quality assurance protocol following CAT12-recommended values/indicators, segmented into gray matter (GM), white matter (WM), and CSF, and normalized using the DARTEL algorithm. For smoothing, we applied an isotropic Gaussian kernel of 10 mm full width half maximum (FWHM).

¹¹C-labelled Pittsburgh compound B (PiB) PET imaging was performed and analyzed as previously described [33]. Briefly, images were acquired 50 min after PiB was injected intravenously (12 mCi, 1 Ci/µmol, over 20 s). The examinations were performed by qualified nuclear medicine technologists. Cortical regions exhibiting the most distinct radiotracer accumulation in Aβ-positive subjects typically include lateral temporal, frontal lobes, posterior cingulate cortex/precuneus, and the parietal lobes, whereas the sensorimotor cortex and the visual cortex can be relatively spared [34]. The cerebellum was used as the reference region for visual interpretation. The results were interpreted by three experienced nuclear medicine physicians who completed the appropriate training programs provided by the radiotracer manufacturer. Standardized uptake value ratios (SUVR) were calculated as the ratio of regional PiB retention to that in the cerebellar gray matter. ROIs were drawn on the lateral temporal, frontal lobes, posterior cingulate cortex/precuneus, and the parietal lobes. Individuals with SUVR values exceeding the cutoff point of 1.5 in any of these ROIs was classified as Aβ(+).

CSF was collected via lumbar puncture, before centrifuging at 2000 × g, 4℃for 10 min, and stored at -80℃. All CSF biomarkers were measured by ELISA, including beta-amyloid (1–40) (Aβ40) (Eqs. 6511-9601-L), beta-amyloid (1–42) (Aβ42) (Eqs. 6521-9601-L), total-tau (t-tau) (Eqs. 6531-9601-L), and phosphorylated-tau181 (p-tau181) (Eqs. 6591-9601-L) (EUROIMUN, Germany). According to the manufacturer’s instructions, participants with Aβ42 < 550 pg/ml or Aβ42: 551–650 pg/ml along with Aβ42/40 ≤ 0.1 were classified as Aβ-positive (A+), and participants with p-tau ≥ 61 pg/ml were classified as tau-positive (T+).

Genomic (gDNA) was isolated from peripheral blood using the standard phenol-chloroform extraction method. All gDNA samples were diluted to 50 ng/µl. The following primers were used to amplify the 581 bp fragment: forward: 5’-CCTACAAATCGGAACTGG-3’; and reverse: 5’-CTCGAACCAGCTCTTGAG-3’. Polymerase chain reaction (PCR) was performed as described previously. The PCR products were sequenced using an ABI 3730 DNA Analyzer (ABI, Louis, MO, USA).

Venous blood of participants was collected in ethylenediaminetetraacetic acid (EDTA) tubes and centrifuged at 3000 rpm for 15 min at 4℃ within 2 h of collection. The obtained blood samples for plasma were stored at − 80℃, and none of the samples underwent freeze-thaw cycles prior to their use in the experiments.

P-tau181, p-tau217, p-tau231, NfL, GFAP, and α-synuclein were quantified on a fully automated machine (AST-Sc-Lite; AstraBio, Suzhou, China). A detailed description of SMID in the Astra System, including a schematic illustration, standard curves, limit of detection (LOD), and limit of quantitation (LOQ) for measuring plasma biomarkers, as well as an assay comparison between AST-Sc-Lite and Quanterix/Mesoscale Discovery, is provided in Supplementary Figs. 1–5, Supplementary Tables 1–3). The working process of SMID in Astra System is as follows: (1) 25 µL Reagent 1 solution containing 1.2 μm magnetic beads obtained from Merck (EM1-100/40) bond with capture antibody was first mixed with 25 µL sample, followed by 6 min incubation under 40℃. The capture antibody chemically modified on magnetic beads bond to antigen in sample solution. (2) 10 µL Reagent 2 solution containing detection antibody labelled nano-fluorophore was then added. After a quick mix, another 4 min incubation under 40℃ happened to let detection antibody labelled with fluorophore can bind to antigen capture by magnetic beads. (3) The mixture was then moved to a flow-cell (Supplementary Fig. 6). A permanent magnet was pressed onto bottom of flow-cell to gathering magnetic beads straightly down onto bottom surface in the flow-cell channel, which was sized as 2 mm*2 mm*60 mm (height*width*length). Due to a well control of verticality of magnetic field direction to channel surface and high uniformity of magnetic beads distributed in the mixture solution, magnetic beads generated a homogeneous but random single-layer array on the channel bottom surface. By controlling of magnetic beads concentration, agglomeration of beads to beads was strictly avoided, while a rare probability of beads may still gather with no effecting on detection variation. Contributed by a fully automated control, the variation of beads collected on the surface was no more than 1.2% between each measurement. (4) Afterwards, unbonded fluorophore was washed away and fluorescent image of beads array was taken by an integrated camara with excitation light source (Supplementary Fig. 7). (5) The counting number was further used for antigen quantitation. Variation of magnetic beads number gathered in generated single-layer beads array contribute to variation of measuring result directly. Thus, a stable control of beads number is essential. We estimated the stability of beads array under different measurements. 1.2% CV% of beads number for 10 repeats indicated a well-controlled beads array generation approach for further quantitation application. P-tau181 (R64030↗), p-tau217 (R64100↗), p-tau231 (R64050↗), NfL (R64040↗), GFAP (R64060↗), and α-synuclein (R64070↗) assay kits (AstraBio) were used for detection. The technicians were blinded to the group status of the participants.

Statistical analyses were performed using the R software (version 4.2.3). Differences in plasma biomarkers between groups were determined using a univariate general linear model (GLM). The Bonferroni correction was used for multiple comparisons. FC was used to describe the difference between two groups. The differential analysis focused on the differences in the distribution of plasma biomarkers between the HC and disease groups, between AD and other dementias, between different pathologies, between degrees of severity, and between the HC and different PSP and FTD subtypes. Classification analysis was performed using receiver operating characteristic (ROC) curve analysis to estimate the diagnostic ability of the plasma biomarkers using the R package cutpointr (https://github.com/thie1e/cutpointr↗) calculating a cut-off score using the Youden’s index and the area under the curve (AUC) was reported for each comparison. The R package RISCA (https://github.com/cran/RISCA↗) was used to visualize ROC curves. Classification analysis focused on the ability of plasma markers to distinguish aMCI and AD from HC, DLB, FTD, and PSP, as well as distinguish the HC from different PSP and FTD subtypes. Spearman’s test was used to evaluate the correlation between the plasma and cerebrospinal fluid markers. For voxel-based analysis of gray matter volume and blood biomarker concentrations, segmentation and smoothing of the brain model were performed using CAT12. The MRI data were spatially smoothed using an 8 × 8 × 8 mm FWHM Gaussian kernel. Correlations were assessed by GLM analysis using SPM12. To study associations of longitudinal changes in plasma biomarkers with longitudinal cognition, we used linear mixed effects models with the interaction between time and standardized plasma biomarker slopes (derived from subject-level linear regression models, with time as predictor of biomarker levels) as the independent variable. The R package visreg (https://github.com/cran/visreg↗) was used to visualization. Sex, age, and APOE ε4 genotype were included as covariates in the above analysis. The total intracranial volume (TIV) was additionally included as a covariate in gray matter volume analysis.

In this study, we used cross-sectional and longitudinal analyses to investigate AD biomarkers in the Chinese Han population. The cross-sectional cohort included a total of 668 individuals, while in the longitudinal study, a subset of 49 A + T + N + patients with AD or aMCI (male: 46.9%, age at onset: 68.9 ± 8.6 years, CSF Aβ42: 358.2 ± 140.8 pg/ml, CSF Aβ40: 7907.9 ± 3814.8 pg/ml, CSF Aβ42/40: 0.05 ± 0.01 CSF t-tau: 144.2 ± 29.4 pg/ml, CSF p-tau: 567.3 ± 212.2 pg/ml) underwent cognitive assessments and plasma biomarker detection after an average of 1-year period (1.2 ± 0.6 years). The baseline demographics, cognitive assessment scores, and CSF biomarker levels are shown in Table 1. There was no significant difference in sex (p < 0.17), whereas the AD + aMCI patients were younger than HC (p < 0.001). Additionally, there was a higher proportion of APOE E4 carriers (43%) among the AD + aMCI group than the HC group (17%, p < 0.001).

The levels of plasma p-tau181 (FC = 2.5, β = 10.6, p < 0.001), p-tau217 (FC = 1.9, β = 3.4, p < 0.001), p-tau231 (FC = 3.8, β = 23.9, p < 0.001), NfL (FC = 1.8, β = 63.0, p < 0.001), GFAP (FC = 2.1, β = 29.6, p < 0.001), and α-synuclein (FC = 2.0, β = 4053.3, p < 0.001) were significantly higher in the aMCI/AD group compared to the HC group. When compared to other dementias, the levels of p-tau181 (FC = 2.2–2.6, β = 9.4–11.0, p < 0.001), p-tau217 (FC = 1.5–1.7, β = 2.3–2.9, p < 0.001), p-tau231 (FC = 2.5–2.9, β = 19.4–20.7, p < 0.001), and GFAP (FC = 1.4–1.6, β = 17.8–20.3, p = 0.002–0.03) were significantly higher in the aMCI/AD group than in other dementias. Besides, the levels of p-tau217 (FC = 1.2–1.3, β = 0.5–1.1, p < 0.002), p-tau231 (FC = 1.4–1.5, β = 2.7–4.3, p < 0.001), and NfL (FC = 1.6–2.1, β = 41.7–65.9, p < 0.003) were increased in other dementias compared to those in the HC group (Fig. 1).

The levels of p-tau217 and p-tau231 showed the best performance for detecting aMCI /AD from HCs at cut-off values of 4.2 pg/mL and 10.6 pg/mL, achieving AUCs of 0.95 and 0.93, respectively, followed by the levels of p-tau181 and GFAP at cut-off values of 8.7 pg/mL and 24.4 pg/mL, with an AUC of 0.82 and 0.79, and α-synuclein at cut-off value of 3981.6 pg/mL, with an AUC of 0.77. Additionally, the level of NfL at cut-off value of 104.2 pg/mL showed lower power for aMCI/AD diagnosis, with AUCs of 0.62. Interestingly, we found that p-tau217 was the best for detecting aMCI/AD from PSP at cut-off value of 4.5 pg/mL, with an AUC of 0.84; p-tau231 was the best for detecting aMCI/AD from FTD at cut-off value of 10.6 pg/mL, with an AUC of 0.81; and p-tau181 was the best for detecting aMCI/AD from DLB at cut-off value of 8.7 pg/mL, with an AUC of 0.83 (Fig. 2A). Additionally, we compared the diagnostic performance of the logistic regression (LR) model combining six biomarkers, which showed an AUC of 0.96 for aMCI/AD diagnosis. The AUCs for distinguishing aMCI/AD from PSP, FTD and DLB were 0.88, 0.86, and 0.85, respectively (Fig. 2B).

We compared the plasma biomarkers in the Aβ (+) (n = 153) and Aβ (–) (n = 84) groups, as well as the tau (+) (n = 100) and tau (–) (n = 91) groups, and found that the Aβ (+) group showed higher levels of p-tau181 (p = 0.045), p-tau217 (p < 0.001), p-tau231 (p = 0.002), GFAP (p < 0.001), and α-synuclein (p = 0.035) than the Aβ (–) group. Furthermore, compared to patients with tau (–) dementia, those with tau (+) dementia had higher levels of p-tau217 (p = 0.043), and GFAP (p = 0.004) (Fig. 3). We also found that p-tau217, p-tau231, GFAP, and α-synuclein were negatively correlated with CSF Aβ42/40 (r = − 0.2, p = 0.005; r = − 0.2, p = 0.024; r = − 0.2, p = 0.002; r = -0.2, p = 0.005, respectively), while p-tau217 and GFAP were positively correlated with CSF p-tau181 (r = 0.2, p = 0.028; r = 0.2, p = 0.009, respectively) (Fig. 4). We further compared the distribution of plasma biomarkers between A + T+ (n = 91) and A + T- (n = 40) patients within aMCI/AD group, as well as among the A + T+ (n = 5), A + T− (n = 13), A − T+ (n = 3), and A − T− (n = 20) patients within the non-AD (FTD and DLB) groups. The results showed that plasma biomarker levels showed no significant difference between A + T + and A + T- within aMCI/AD group. We also found that in non-AD individuals, NfL levels in the A + T + group were elevated compared to the A − T− group (β = 178, p = 0.009); p-tau217 and GFAP levels were higher in the A + T − group compared to the A − T− group (β = 1.1, p = 0.026; β = 34.1, p = 0.003, respectively); and p-tau217 levels were increased in the A − T + group compared to the A − T− group (β = 1.9, p = 0.015) (Supplementary Figs. 8 and 9).

Compared to patients with aMCI and mild dementia (CDR = 0.5–1), those with moderate-to-severe dementia (CDR = 2–3) had higher levels of NfL (p = 0.013) and GFAP (p = 0.020). P-tau181, p-tau217, p-tau231 and α-synuclein showed no significant differences among the different groups (Supplementary Fig. 10).

In this study, 100 patients with PSP were classified into the PSP-RS (n = 65) and PSP-variant (n = 35) subtypes, and the 76 patients with FTD was divided into the bvFTD (n = 31), nfvFTD (n = 9), and svFTD (n = 19) subtypes. We further explored the diagnostic performance of AD non-specific biomarkers, including GFAP, NFL, and α-synuclein, for detecting the above subtypes. As a result, we found that the α-synuclein level was the best for detecting PSP-v and bvFTD subtypes at cut-off values of 3957.3 pg/mL and 6705.0 pg/mL, achieving an AUC of 0.81 and 0.74, respectively, while the levels of GFAP and NFL presented good diagnostic performance in svFTD and nfvFTD at cut-off values of 22.32 pg/mL and 85.7 pg/mL, with an AUC of 0.74 and 0.69, respectively (Fig. 5).

We constructed a GLM using MRI measurements as the dependent variable and blood biomarkers as independent variables. The results revealed that elevated levels of p-tau181 p-tau217, and GFAP were associated with reduced gray matter volume in the right, left and bilateral temporal lobe, respectively. And p-tau231 and GFAP were associated with reduced gray matter volume in the left and bilateral frontal lobe, respectively (p < 0.05) (Fig. 6).

We further tested the association between longitudinal changes in plasma biomarker levels and longitudinal changes in cognition (as reflected by decreased MMSE scores). Longitudinal changes in p-tau181 levels over time were significantly associated with decreased MMSE scores (p = 0.046) (Fig. 7).