Association of Amyloid and Tau With Cognition in Preclinical Alzheimer Disease

Key Points Question Is cognitive decline associated with amyloid-β or tau tangles accumulation? Findings In this cohort study that included 60 normal older adults with repeated positron emission tomography measures, the rate of tau accumulation in the inferior temporal neocortex was associated with the rate of cognitive decline. Amyloid accumulation was associated with subsequent tau accumulation, and this sequence of successive amyloid and tau changes in neocortex was found to mediate the association of initial amyloid with final cognition, measured 7 years later. Meaning Amyloid positron emission tomography is useful to detect early Alzheimer pathology; repeated tau positron emission tomography is useful to track disease progression.

eAppendix: PET data processing PET data presented in this work were processed using a specific longitudinal pipeline.
The mean PET images (average Counts Per Second -CPS) from each session were realigned to each other in two steps. A first pass aligned follow-up scan(s) to the baseline scan, then an average image was created and a second pass aligned all scans to the mean image. The mean image was used for co-registration to the MRI closest to the midpoint between FTP session date sfor a subject, allowing a single registration to be done. ROIs from the selected MR were then applied to all time points. The aim of this longitudinal pipeline was to try to maximize the registration between time points, as PET to PET should produce a better fit than a set of PET to MR registrations, and should reduce variability.
The choice of cerebral white matter as a reference region has been well documented for longitudinal PET studies using Aβ tracers. 1-4 The first longitudinal tau-PET studies used different reference regions 3 including cerebellar gray 5 and subcortical white matter 6 . In our dataset (n=60), the Pearson's correlation between baseline FTP data scaled on cerebellar gray or scaled on cerebral white matter was 0.85. The correlation between FTP slope data using either reference was 0.71. The present study also used partial volume correction (PVC). The Pearson's correlation between baseline FTP data with and without PVC was 0.87. The correlation between FTP slope data obtained with and without PVC was 0.80 (cerebral white matter reference region used for both PVC and non-PVC data). In comparison, PVC and the choice of the reference region had smaller impact on PiB cross-sectional and slope data (all R>0.93). We observed associations between PiB slope and FTP slope, and between FTP slope and PACC slope, regardless of PET data processing.
The annual rates of FTP changes were higher in this study than in the recently published longitudinal FTP data from Mayo Clinic. 5 Both studies observed faster rates of FTP change in high-PiB than in low-PiB participants, but the low-PiB participants from the Mayo Clinic did not demonstrate significant FTP changes over a 14-month follow-up.
Longer follow-up time (e.g., 26 months in our study) may more readily reveal early tau changes. FTP reference region may also influence measurement consistency: Mayo used cerebellar crus; we used cerebral white matter. Using cerebellar gray as reference, we observed significantly lower rates of FTP change (cerebellar gray: 0.025SUVr/y, cerebral white matter: 0.041SUVr/y, p<0.0001); however, FTP change t=0-2 was significantly greater than zero using either reference in the forty participants with low-PiB t=0 at baseline (cerebellar gray: 0.021SUVr/y, cerebral white matter: 0.033SUVr/y, one-sample t-tests: both p's<0.0001). Both the Mayo Clinic study and ours used PVC data. The inpress data of Berkeley 6 show rates of FTP accumulation in the inferior temporal neocortex of older adults that are very similar to ours (3% per year in both studies).
They also used subcortical white matter as a reference region and PVC.

eTable1: Associations between PiB change and FTP change in additional ROIs
The temporal meta-ROI is a Freesurfer average between fusiform, inferior temporal, middle temporal, and entorhinal cortex. PiB is measured in a neocortical aggregate (FLR). All models are co-varied for baseline age, sex, education, and PiB slope (t=0 to t=+2).
Covariates are not significant. The temporal meta-ROI is a Freesurfer average between fusiform, inferior temporal, middle temporal, and entorhinal cortex. The model for inferior temporal is also given in the main manuscript ( Top: Spaghetti plots showing the unadjusted PET SUVr and PACC scores versus age at the initial. t=-3, baseline. t=0, and follow-up. t=+2 observations. The cross-sectional statistics provided is baseline PACC or PET data predicted by age at baseline. At older ages, PiB signal is not significantly higher, FTP signal is higher, and PACC is marginally lower. Bottom: PiB, FTP, and PACC slope data are plotted against age. Slope data for each outcome was obtained from a linear mixed-model predicting this outcome over time with a random intercept and time slope per subject. The slope data statistics provided is change in PACC (or change in PET data) from baseline to final follow-up predicted by age at baseline, adjusting for baseline PACC (or baseline PET data). PiB and FTP slopes are not associated with age. PACC decline is significantly faster at older ages, adjusting for baseline PACC.
The blue color denotes participants with low-PiB and the red color participants with high-PiB at the initial PiB observation. Initial and baseline data are dots. Follow-up data are triangles for participants who remained clinically normal (CN) during the study and stars for participants who progressed to MCI. All MCI progressors had high-PiB signal.