Tau-Atrophy Variability Reveals Phenotypic Heterogeneity in Alzheimer's Disease

Abstract

Objective: Tau neurofibrillary tangles (T) are the primary driver of downstream neurodegeneration (N) and subsequent cognitive impairment in Alzheimer's disease (AD). However, there is substantial variability in the T-N relationship - manifested in higher or lower atrophy than expected for level of tau in a given brain region. The goal of this study was to determine if region-based quantitation of this variability allows for identification of underlying modulatory factors, including polypathology.

Methods: Cortical thickness (N) and ¹⁸ F-Flortaucipir SUVR (T) were computed in 104 gray matter regions from a cohort of cognitively-impaired, amyloid-positive (A+) individuals. Region-specific residuals from a robust linear fit between SUVR and cortical thickness were computed as a surrogate for T-N mismatch. A summary T-N mismatch metric defined using residuals were correlated with demographic and imaging-based modulatory factors, and to partition the cohort into data-driven subgroups.

Results: The summary T-N mismatch metric correlated with underlying factors such as age and burden of white matter hyperintensity lesions. Data-driven subgroups based on clustering of residuals appear to represent different biologically relevant phenotypes, with groups showing distinct spatial patterns of higher or lower atrophy than expected.

Interpretation: These data support the notion that a measure of deviation from a normative relationship between tau burden and neurodegeneration across brain regions in individuals on the AD continuum captures variability due to multiple underlying factors, and can reveal phenotypes, which if validated, may help identify possible contributors to neurodegeneration in addition to tau, which may ultimately be useful for cohort selection in clinical trials. ANN NEUROL 2021;90:751-762.

Figure 1.

Scatter plots showing (a) Tau tracer SUVR vs. cortical thickness in the right inferior temporal gyrus ROI along with robust regression fit and ± 1.5 standard deviation lines, (b) T-N mismatch metric vs. age, (c) T-N mismatch metric vs. MMSE scores, and (d) T-N mismatch metric vs. WMH volume.

Figure 2:

Median ROI-wise group residual maps (columns 1–3). More saturated color represents higher residual, indicating higher (red) or lower (blue) atrophy relative to tau burden. Column 4 shows representative examples with arrows indicating features characteristic of the group.

Figure 3.

T-N regression residuals for each ROI and each participant visualized on a heatmap. ROIs on the x-axis are sorted by lobes, participants on the y-axis are sorted by clusters. Shades of red show positive residuals (more atrophy) and shades of blue show negative residuals (more tau). Stacked bar graph in the top panel shows percentage of ROIs showing “mismatch”, as defined by the residual being greater than 1.5 standard deviation away from the regression line.

Figure 4.

Trajectory of cognitive decline for different groups of participants. Slopes shown for each group is estimated by linear mixed-effects model. Groups 2 and 6 are the vulnerable groups showing steeper decline than group 1 (canonical). Groups 3 and 5 are the resilient groups. Group 4 is not shown because of the small group size. Note that the variable maximum time periods for different groups reflect differences in duration of followup data available.

Figure 5.

Tanglegram displays comparing cluster memberships. Top panel: comparison memberships between using left and right hemispheric ROI measures separately (left) vs. using averaged measures (right). Bottom panel: comparison using discretized regression residuals (left) vs. the raw continuous regression residuals (right). In both panels, color for each cluster on the right was chosen to be similar to the cluster on the left that had the most overlap in membership for ease of visualization. Cluster labels (1–6) are arbitrary, as output by the clustering algorithm.