Tau profiling across Alzheimer's disease staging reveals vulnerability to disease pathophysiology

Abstract

Background: Inter-individual variability in tau topography challenges the propagation hypothesis of tau aggregates.

Methods: To address this gap, we propose the Manifold Component Analysis (MCA), for identifying pseudo-continuous profiles informed by the spatial continuity of stage regions.

Results: Longitudinal and cross-sectional MCA in large aging cohort identified individual profiles (N = 753) expressing tau load in the entorhinal, limbic and neocortical regions. Using these profiles, we found neuropsychological and blood-based milestones of early and late disease stages. Finally, we also found evidence of rapid tau load increases and cognitive decline centered at the early-to-mid neocortical stages of Alzheimer's disease.

Conclusions: Stage system based on tau load and spreading profiles across cortical areas provide a compelling framework for inferring pathophysiological prognosis in Alzheimer's disease.

Keywords: 18F-MK6240 PET; Alzheimer’s disease; Component analysis; Dimensionality reduction; Neurofibrillary tangles; PET braak staging; Propagation; Tau.

Fig. 1

Surface PET Braak masks used as input and output for MCA. Left: MCA discrete mask input, in lateral (top) and medial (bottom) views. Right: a pseudo-continuous representation of the discrete mask shown in the left, following MCA. Centre: colormaps corresponding to both discrete and continuous representations

Fig. 2

MCA reveals hierarchical tau accumulation. MCA-derived PET Braak region profiles are shown segregated by the corresponding PET Braak stages, following cohort-derived thresholds based on established criteria (2.5 SD above young subjects’ respective PET Braak region values). Insets in the top are showing cortical distributions of tau that are below the threshold (cyan), above the threshold (magenta), and the current area of sampling (yellow)

Fig. 3

MCA Profiles of TRIAD tau-PET data, stratified by diagnosis and other criteria. A: all subjects’ ¹⁸F-MK-6240 data profiles stratified by APOE considering the presence of at least one 2 (blue), two 3 (green) and at least one 4 (red) alleles; B: profiles stratified by the Clinical Dementia Rating (CDR) scores; C: by amyloid positivity defined by established threshold; D: by diagnosis represented by cognitively normal (CN), mild cognitive impairment (MCI) and Alzheimer’s Disease (AD) dementia subjects. Shaded lines indicate standard error across subjects within a given group

Fig. 4

MCA reveals rapid progression of tau during Braak stages 4 and 5. A: group-wise profiles per Braak stage for baseline (blue) and latest follow-up (red). B: percent change between baseline and latest follow-up across the entire MCA-derived PET Braak region continuum, for each Braak stage subjects’ group displayed on the left. Polynomial fits of 4th degree are displayed as dotted black lines for each Braak stage group’s follow-up– baseline difference

Fig. 5

Cognitive decline rapidly accelerates during stages 4 and 5. Absolute values of partial correlations between the MCA-derived PET Braak regions’ continua and neuropsychological assessment measurements are shown. The correlational fingerprints are sorted by the highest values of fitted 4th degree polynomial curves which are shown as dotted black lines, wherein the second column continues the first

Fig. 6

Neuroinflammation and hyperphosphorylation accelerate during stages 3, 4 and 5. Partial correlations between the MCA-derived PET Braak regions’ continua and several plasma biomarkers are shown. A: correlations between the MCA-derived PET Braak regions’ continua and GFAP, p-tau181 and p-tau231, respectively, using the corresponding ranges indicated by squiggly brackets; B: scatterplots and regression lines showing the respective correlations, segregated into the main diagnostic groups– CN, MCI and AD dementia

Fig. 7

Disease landmarks estimated using MCA. The curves represent 4th degree polynomials akin to those used in the previous figures