Predicting brain amyloid load with digital and blood-based biomarkers

Abstract

Background: With the recent approval of anti-β-amyloid (Aβ) treatment for Alzheimer's disease (AD), a demand has emerged for scalable, convenient and accurate estimations of brain Aβ burden for the detection of AD that would enable timely, accurate and reliable diagnosis in one's primary care physician's (PCPs) office as called for recently by World Health Organization (WHO).

Methods: MemTrax, a 2-minute online memory test, was selected as the digital biomarker of cognitive impairment, and blood-based biomarkers (BBMs) including Aβ42, Aβ40, P-tau181, GFAP and NfL were used to estimate AD-related metrics in different groups of elderly individuals (n = 349) for comparison with Aβ PET scans of brain Aβ burden. The correlations between MemTrax, MoCA, BBMs and brain Aβ burden, expressed in centiloid (CL) values, were analyzed for predicting CL value alone or in combinations using machine-learning (ML).

Results: Both MemTrax and the MoCA were able to differentiate Aβ status similarly. Integration of MemTrax and BBMs using ML, however, significantly improved the AUCs (over the same with MoCA) for differentiating Aβ status. MemTrax and p-Tau181/Aβ42 composite showed the strongest relationship with CL value among other BBMs. Most importantly, regression analyses of MemTrax and p-Tau181/Aβ42 aptly predicted CL values.

Conclusion: The combination of MemTrax and BBMs provides an accurate, convenient, non-invasive, cost-effective and scalable way to estimate Aβ load, which provides an opportunity for mass screening and timely and accurate diagnosis of AD. Our findings could also facilitate more effective AD clinical management in the PCPs office worldwide for more equitable access to current standard of care.

Keywords: Alzheimer’s disease; Aβ; Blood biomarkers; Centiloid; MemTrax; MoCA.

Fig. 1

MemTrax and blood-based biomarkers (BBMs) in differentiating Aβ status. (A) Receiver-operating characteristic (ROC) curves for distinguishing Aβ status by MemTrax (MTx-%C) and MoCA. (B) ROC curves for differentiating Aβ status by p-Tau181, Aβ42/40, NfL and GFAP. (C) ROC curves for differentiating Aβ status by combining MemTrax and blood biomarkers. (D) ROC curves comparison of MemTrax and MoCA when adding blood biomarkers. (E) Ranked BBMs and digital cognitive metrics in predicting Aβ + or Aβ-. BBMs are marked in orange, while online cognitive metrics are marked in blue

Fig. 2

The two cut-off approach for testing combing digital and blood-based biomarker of Aβ pathology. (A) Combining MemTrax (MTx-%C) and p-Tau181/Aβ42 in differentiating Aβ status. (B) The correlation between p-Tau181/Aβ42 and CL value. (C) Usage of a two-cut-off approach for testing combining MemTrax and p-Tau181/Aβ42 leads to three categories of results: positive, intermediate and negative, increasing the accuracy with which people can be classified as Aβ+/- in PET-CT scan. Individuals classified as having intermediate results are no more than 20%

Fig. 3

Pearson correlation between CL and (A) MemTrax (MTx-%C). (B) MoCA, (C) p-Tau181, (D) Aβ42/40, (E) GFAP, (F) NfL. Aβ, amyloid β; GFAP, glial fibrillary acidic protein; NfL, neurofilament light

Fig. 4

Receiver-operating characteristic (ROC) curves for distinguishing between (A) CU Aβ- and MCI Aβ+, (B) CU Aβ- and AD, (C) MCI Aβ- and MCI Aβ+, and (D) MCI Aβ- and AD dementia participants. ROC curves are presented for A, B, C, and D for (i) Aβ42/ 40, p-Tau181, GFAP, NfL and Aβ42/ 40 + p-Tau181 + GFAP + NfL and (ii) model comprising MemTrax (MTx-%C) + Aβ42/40, MemTrax + p-Tau181, MemTrax + GFAP, MemTrax + NfL, MemTrax + Aβ42/40 + p-Tau181 + GFAP, and MemTrax + Aβ42/40 + p-Tau181 + GFAP + NfL

Fig. 5

Digital and BBMs in predicting CL value. (A) Ranked biomarkers and digital cognitive metrics in predicting CL value. Plasma biomarkers are marked in orange, while online cognitive metrics are marked in blue. (B) Heat map showing predicted CL value by combining p-Tau181/Aβ42 and MemTrax (MTx-%C) using linear regression