A protein panel including pTau217 outperforms pTau217 alone in identifying high tau load in amyloid positive individuals

TL;DR

A blood test combining pTau217 with other proteins better identifies advanced tau pathology in Alzheimer's patients than pTau217 alone.

Contribution

A multi-protein blood panel including pTau217 improves detection of high tau load in amyloid-positive individuals compared to pTau217 alone.

Findings

A multi-analyte panel (AUC=0.93) outperformed pTau217 alone (AUC=0.88) in predicting BraakLate stages.

NPTXR, GDNF, and VGF were identified as complementary proteins to pTau217.

The multi-analyte model had a better fit and lower AIC score than pTau217 alone.

Abstract

Recent anti‐amyloid trial designs for Alzheimer's disease (AD) have aimed to identify amyloid‐β (Aβ)‐positive patients without an advanced tau pathology, as they are most likely to benefit from these therapies. Blood‐based biomarkers might reduce the need to use cerebrospinal fluid (CSF) or positron emission tomography (PET) but it is unclear whether phosphorylated tau‐217 (pTau217) alone would be effective to exclude this high‐tau group at screening. We investigated whether a blood‐based protein panel, including pTau217, could better distinguish early from late‐stage tau pathology in Aβ‐positive patients compared to pTau217 alone.

Aβ‐positive participants from the TRIAD cohort (n = 129; mean [SD] age, 70.4 [8.3] years; females [58.9%]) were classified as BraakLate (Braak V‐VI: n = 51) or BraakEarly (Braak I‐IV: n = 78) by tau PET imaging([18F]MK6240). We employed the NULISAseq CNS Panel to quantify 120 CNS‐related proteins. A bootstrapped (1000x) LASSO regression was used to identify the most recurringly selected proteins for distinguishing BraakLate from BraakEarly. Generalized linear models (GLM) for the multi‐analyte panel and pTau217, adjusted for age and sex, were used and their performance evaluated by ROC analyses and Akaike Information Criterion (AIC) scores. GLMs were also used to estimate probability scores for each patient for belonging to BraakLate.

The bootstrapped LASSO regression retained pTau217, neuropentraxin receptor (NPTXR), vascular growth factor (VGF) and growth‐derived neurotrophic factor (GDNF) in >75% of the iterations. ROC analysis demonstrated that the multi‐analyte panel (AUC=0.93: 95% CI 0.89‐0.98) had a significantly better prediction of BraakLate than pTau217 alone (AUC= 0.88; 95% CI 0.88‐0.94; P
DeLong= 0.004). The fit of the model was also assessed by comparing AIC, where the multi‐analyte panel showed a reduction in the score to detect the Braak category, suggesting a better model fit. This was further supported by an ANOVA comparison between the two models, where the multi‐analyte model was significantly better than pTau217 alone (P
ANOVA < 0.001).

We identified three complementary proteins (NPTXR, GDNF, VGF) to pTau217 that can improve its ability in detect later Braak stages in Aβ‐positive patients. This suggests an immunoassay‐based panel might be a cost‐effective tool to exclude participants with high tau pathology in anti‐amyloid trials designs.

Figures