Novel plasma protein biomarkers: A time-dependent predictive model for Alzheimer's disease

Abstract

Background

The accurate prediction of Alzheimer's disease (AD) is crucial for the efficient management of its progression. The objective of this research was to construct a new risk predictive model utilizing novel plasma protein biomarkers for predicting AD incidence in the future and analyze their potential biological correlation with AD incidence.

Methods

A cohort of 440 participants aged 60 years and older from the Alzheimer's Disease Neuroimaging Initiative (ADNI) longitudinal cohort was utilized. The baseline plasma proteomics data was employed to conduct Cox regression, LASSO regression, and cross-validation to identify plasma protein signatures predictive of AD risk. Subsequently, a multivariable Cox proportional hazards model based on these signatures was constructed. The performance of the risk prediction model was evaluated using time-dependent receiver operating characteristic (t-ROC) curves and Kaplan-Meier curves. Additionally, we analyzed the correlations between protein signature expression in plasma and predicted AD risk, the time of AD onset, the expression of protein signatures in cerebrospinal fluid (CSF), the expression of CSF and plasma biomarkers, and APOE ε4 genotypes. Colocalization and Mendelian randomization analyses was conducted to investigate the association between protein features and AD risk. GEO database was utilized to analyze the differential expression of protein features in the blood and brain of AD patients.

Results

We identified seven protein signatures (APOE, CGA, CRP, CCL26, CCL20, NRCAM, and PYY) that independently predicted AD incidence in the future. The risk prediction model demonstrated area under the ROC curve (AUC) values of 0.77, 0.76, and 0.77 for predicting AD incidence at 4, 6, and 8 years, respectively. Furthermore, the model remained stable in the range of the 3rd to the 12th year (ROC ≥ 0.74). The low-risk group, as defined by the model, exhibited a significantly later AD onset compared to the high-risk group (P < 0.0001). Moreover, all protein signatures exhibited significant correlations with AD risk (P < 0.001) and the time of AD onset (P < 0.01). There was no strong correlation between the protein expression levels in plasma and CSF, as well as AD CSF biomarkers. APOE, CGA, and CRP exhibited significantly lower expression levels in APOE ε4 positive individuals (P < 0.05). Additionally, colocalization analysis reveals a significant association between AD and SNP loci in APOE. Mendelian randomization analysis shows a negative correlation between NRCAM and AD risk. Transcriptomic analysis indicates a significant downregulation of NRCAM and PYY in the peripheral blood of AD patients (P < 0.01), while APOE, CGA, and NRCAM are significantly downregulated in the brains of AD patients (P < 0.0001).

Conclusion

Our research has successfully identified protein signatures in plasma as potential risk biomarkers that can independently predict AD onset in the future. Notably, this risk prediction model has demonstrated commendable predictive performance and stability over time. These findings underscore the promising utility of plasma protein signatures in dynamically predicting the risk of AD, thereby facilitating early screening and intervention strategies.