Plasma biomarkers predict Alzheimer’s disease before clinical onset in Chinese cohorts

Alzheimer’s disease (AD) is the leading cause of dementia worldwide, imposing substantial social and economic burdens. Neuropathological changes precede clinical symptoms by up to two decades, underscoring the need for early, symptom-agnostic detection and intervention. The National Institute on Aging–Alzheimer’s Association (NIA-AA) ATN framework emphasizes biomarker-based definitions of AD using amyloid (A), tau (T), and neurodegeneration (N). While cerebrospinal fluid (CSF) and imaging biomarkers are informative, their invasiveness, cost, and limited accessibility constrain broad implementation. Blood-based biomarkers—Aβ42, Aβ40 and their ratio (A component), phosphorylated tau (T), and total tau and NfL (N)—offer minimally invasive, scalable alternatives. The research question is whether plasma Aβ42, p-tau181, and NfL can predict preclinical and future AD in Chinese populations, potentially enabling earlier identification and risk stratification for disease-modifying trials.

Prior studies have shown that plasma biomarkers aligned with the ATN framework can discriminate AD from controls and sometimes predict future cognitive decline and incident dementia. Plasma p-tau181, NfL, and Aβ42/Aβ40 have shown associations with amyloid PET status, CSF measures, neuropathology, and longitudinal progression in predominantly European and American cohorts. However, results for plasma Aβ42/Aβ40 in predementia stages have been inconsistent across platforms and studies, and plasma total tau generally shows limited diagnostic value compared with phosphorylated tau. Ethnic and racial factors may influence biomarker performance, highlighting the need for data from diverse populations, including Chinese cohorts.

Design: Two independent Chinese cohorts were studied. Cohort 1 was a longitudinal cohort from the China Cognition and Aging Study with 8–10 years of follow-up. At follow-up, there were 126 participants with AD/preclinical AD (PreAD) and 123 controls. Participants were cognitively intact 8–10 years before the study; those who later developed AD with abnormal CSF t-tau/p-tau181 and Aβ42 were designated as preclinical AD. Cohort 2 (replication) was cross-sectional from the familial Alzheimer’s Disease Network (FAD; families with APP, PSEN1, or PSEN2 mutations), including mutation carriers and non-carriers with estimated years from onset (EYO) of approximately 5–8 years.

Biomarkers: Plasma Aβ42, Aβ40, Aβ42/Aβ40, total tau, p-tau181, and NfL were measured at baseline and follow-up (cohort 1) and in cohort 2. CSF counterparts (Aβ42, Aβ40, total tau, p-tau181, NfL) were also measured to assess plasma–CSF correlations.

Ethics: The protocol was approved by the Institutional Review Board; all procedures complied with relevant ethical regulations.

Statistical analyses: Group differences in plasma biomarkers between AD/PreAD and controls were assessed using two-sided Student’s t-tests with Bonferroni corrections where noted. Linear regression (two-sided) quantified correlations between plasma and CSF biomarkers and between baseline and follow-up plasma levels; adjusted R² and P values were reported. Lin’s concordance correlation coefficient (CCC) evaluated within-individual agreement between baseline and follow-up. Logistic regression models, adjusted initially for age, sex, education, and APOE ε4 status, assessed discrimination of AD and PreAD from controls using plasma biomarkers; non-significant covariates (age, sex, education) were excluded from final models. Multicollinearity among plasma biomarkers was examined with tolerance, variance inflation factor (VIF), and condition indices. Discriminative performance was evaluated via ROC curves and AUCs with 95% CIs; likelihood ratio tests assessed the incremental value of APOE ε4; DeLong tests compared AUCs of combined versus individual biomarkers and models with versus without APOE ε4. Replication analyses in cohort 2 evaluated the same combined biomarker models and subgroup ROC analyses by mutation type (APP, PSEN1, PSEN2). Code for analyses is available on Zenodo.

• Group differences (cohort 1, follow-up): Plasma Aβ42 and Aβ42/Aβ40 were significantly lower in AD versus controls (Aβ42: P = 3.57 × 10^-6; Aβ42/Aβ40: P = 1.42 × 10^-12), while p-tau181 and NfL were higher (p-tau181: P = 1.04 × 10^-30; NfL: P = 1.10 × 10^-24).
• Plasma–CSF correlations (cohort 1, follow-up): Significant correlations were observed for Aβ42 (adjusted R² = 0.66, P = 3.15 × 10^-10), Aβ40 (adjusted R² = 0.41, P = 2.10 × 10^-8), p-tau181 (adjusted R² = 0.53, P = 2.86 × 10^-14), and NfL (adjusted R² = 0.85, P = 4.61 × 10^-5); plasma total tau showed a weaker association with CSF tau (adjusted R² = 0.28, P = 2.21 × 10^-6).
• Longitudinal consistency (cohort 1): Baseline and follow-up plasma levels were significantly correlated for all biomarkers (Aβ42: adjusted R² = 0.69, P = 2.68 × 10^-9; Aβ40: 0.66, P = 1.17 × 10^-9; total tau: 0.68, P = 9.03 × 10^-10; p-tau181: 0.62, P = 1.05 × 10^-5; NfL: 0.68, P = 2.16 × 10^-5). CCCs ranged from 0.58 to 0.77, and mean levels differed significantly between time points.
• Predictive performance (cohort 1 ROC/AUC): Combination of Aβ42, p-tau181, and NfL achieved AUC = 0.78 (95% CI: 0.71–0.83, P = 1.22 × 10^-16) at baseline; adding APOE ε4 increased AUC to 0.81 (95% CI: 0.75–0.86, P = 5.11 × 10^-17). At follow-up, the combination with or without APOE ε4 yielded AUC = 0.99 (95% CI: 0.98–1.00; P values ≈ 10^-40). DeLong tests showed combined models outperformed individual biomarkers; APOE ε4 significantly improved baseline AUC (DeLong P = 3.37 × 10^-7) but not follow-up.
• Replication (cohort 2, FAD): The combination of Aβ42, p-tau181, and NfL achieved AUC = 0.79 (95% CI: 0.71–0.88, P = 2.77 × 10^-7), with similar AUCs (0.78–0.80) across APP, PSEN1, and PSEN2 subgroups.
• Overall: Plasma Aβ42 decreased and p-tau181 and NfL increased in preclinical AD; combining these markers discriminated preclinical AD from controls and predicted AD years before clinical onset, including in familial AD replication.

The study addressed whether plasma biomarkers can predict AD during preclinical stages in Chinese populations. Plasma Aβ42, p-tau181, and NfL tracked their CSF counterparts, indicating that peripheral concentrations reflect central pathology. These biomarkers were altered in AD and preclinical AD, and their combination provided superior discrimination compared with individual markers. Logistic models demonstrated strong predictive performance, particularly at follow-up, and the replication in a familial AD cohort supported generalizability across sporadic and autosomal dominant forms. The findings align with prior work in predominantly Western cohorts, reinforcing the applicability of plasma biomarker-based risk stratification in Chinese populations. Notably, plasma Aβ42/Aβ40 did not contribute meaningfully to prediction in this study, and plasma total tau was not informative, underscoring the value of p-tau181 and NfL for the T and N components. These results support using a combined plasma biomarker panel for early detection and trial screening, potentially enabling interventions 5–8 years before clinical onset.

This work demonstrates that a combination of plasma Aβ42, p-tau181, and NfL predicts Alzheimer’s disease years before symptom onset in Chinese cohorts, correlates with CSF biomarkers, and generalizes to familial AD. The combined biomarker model outperforms individual markers and can aid early identification, risk stratification, and clinical trial enrichment. Future research should pursue multicenter, multi-ethnic validation; standardize assays and preanalytical protocols; investigate additional markers (e.g., GFAP, other p-tau isoforms); and evaluate clinical utility, cost-effectiveness, and integration into screening workflows.

• Population and generalizability: Data are from Chinese cohorts; broader validation across ethnicities and settings is needed, as acknowledged by calls for international multicenter studies.
• Assay/platform variability: Inconsistent findings for plasma Aβ42/Aβ40 across studies may relate to measurement platforms; this study also noted no added predictive value from Aβ42/Aβ40.
• Biomarker scope: Plasma total tau was not informative for predementia AD; additional biomarkers (e.g., GFAP, p-tau217) were not included in the final models.
• Sample sizes and design: Although cohort 1 is longitudinal, the replication cohort is cross-sectional and relatively limited; estimates in replication may be less precise.
• Multicollinearity assessment: The authors reported tolerance, VIF, and condition indices consistent with high collinearity but concluded no significant multicollinearity; this discrepancy could affect model interpretation.
• Follow-up intervals: Significant differences between baseline and follow-up levels indicate temporal changes; defining optimal thresholds over time requires further study.