Could tau-PET imaging contribute to a better understanding of the different patterns of clinical progression in Alzheimer's disease? A 2-year longitudinal study

Alzheimer's disease shows heterogeneous clinical presentations and variable rates of cognitive decline. Tau-PET enables in vivo measurement of tau pathology, which correlates closely with symptoms. Prior studies demonstrate prognostic value of baseline tau-PET for atrophy and cognition, but longitudinal tau-PET findings have been inconsistent across regions and patient subgroups, with some reports of decreasing tracer uptake over time. This study asked how [18F]-flortaucipir uptake changes over ~2 years in mild cognitive impairment/mild dementia due to AD, how these changes relate to cortical atrophy progression, and whether regional tau-PET progression relates to decline in specific cognitive domains. The authors hypothesized that tau SUVr would increase variably by region, that tau-PET progression would correlate with atrophy progression, and that tau-PET progression would associate with domain-specific cognitive decline.

The authors summarize that longitudinal tau-PET studies are relatively few and show mixed results: increases reported in temporal, frontal, parietal, and occipital regions, or broadly except medial temporal lobes; some studies observe decreases over time, sometimes attributed to measurement error. Relationships between baseline tau load and subsequent change are inconsistent (both positive and negative associations reported), with faster accumulation in women and younger patients in some studies. Domain-specific links between tau change and cognition remain underexplored. This context motivates a careful, methodologically rigorous longitudinal analysis linking regional tau change, atrophy, and cognitive domains.

Prospective longitudinal cohort (Shatau7-Imatau; NCT02576821; ethics-approved). Participants: 27 AD patients at MCI/mild dementia stages with AD phenotype (amnestic n=18; biparietal n=9), positive CSF AD profile (t-tau/Aβ42 > 0.52) and amyloid-positive [11C]-PiB PET (GCI > 1.45), CDR ≤ 1; and 12 cognitively normal amyloid-negative controls (MMSE ≥ 27, normal testing, CDR=0, no complaints, GCI < 1.4). Baseline assessments: clinical and neuropsychological battery (MMSE, CDR, composite memory, instrumental, executive scores), 3T MRI (Siemens Prisma) with 3D T1 MP-RAGE, [11C]-PiB PET, and [18F]-flortaucipir PET (Tau1). Follow-up: annual clinical/neuropsychological assessments for 2 years and repeat MRI and [18F]-flortaucipir PET at ~2.3±0.4 years (Tau2). PET acquisition on HRRT; PiB 330.5±63.6 MBq (40–60 min), flortaucipir ~382.0±14.5 MBq at Tau1 and 368.2±35.6 MBq at Tau2 (80–100 min). Reconstructions included corrections and PSF modeling. Quantification: SUVr images using eroded (2 mm) supratentorial white matter as primary reference (also analyzed eroded cerebellar gray; similar results; WM chosen for lower variability and repeatability advantages). VOIs: entorhinal; temporal meta-VOI (entorhinal, parahippocampus, fusiform, inferior/middle temporal); lateral temporal; lateral parietal; medial parietal; frontal; occipital (Ossenkoppele et al. definitions). Longitudinal pipelines: FreeSurfer 6.0 longitudinal stream for cortical thickness; SPM12 pairwise longitudinal registration and CAT12 for voxelwise atrophy maps (Jacobian modulation; 8-mm FWHM smoothing). PET longitudinal processing used ANTs to create within-subject linear PET templates and nonlinear MRI templates for optimal Tau2–Tau1 alignment; annualized subtraction images computed ((Tau2–Tau1)/Δt) and smoothed (8-mm). VOIs were aligned to native PET via ANTs transformations; mean SUVr extracted; annualized change computed. Statistics: R 3.6.1; baseline group differences via ANCOVA; longitudinal trajectories via linear mixed-effects models with group×time interaction and random intercepts; covariates: age, sex (and ApoE genotype, CDR-SOB where specified). VOI analyses with Bonferroni correction. Voxelwise analyses via VoxelStats 1.1 with age, sex, ApoE, CDR-SOB covariates; RFT-based multiple comparisons correction (clusterwise p<0.001; cluster-forming p<0.001). Relationships between imaging changes (Tau2–Tau1 or MRI2–MRI1) and cognition modeled with mixed-effects at the voxel level including interactions with time; separate models for tau-PET and MRI atrophy.

- In AD, average tau SUVr increased over ~2 years in most VOIs (e.g., +6% in inferomedial temporal meta-VOI; +6.7% in frontal VOI), but decreased in the lateral temporoparietal cortex (notably left). Controls showed minimal change (≈+1.3% temporal meta-VOI; +1.6% frontal). Group differences in frontal Tau2–Tau1 were nominally significant (p≈0.008–0.018) but did not survive multiple comparison correction. Voxelwise, AD showed positive Tau2–Tau1 in bilateral frontal and right inferomedial temporal regions and negative changes in left-dominant temporoparietal cortex; only a small left orbitofrontal cluster differed from controls. - Baseline tau vs change: Strong negative correlations between Tau1 and Tau2–Tau1 in temporal and lateral parietal cortices (temporal meta-VOI r^2=0.79, p=0.0015; lateral temporal r^2=0.83, p=0.017; lateral parietal r^2=0.70, p=0.022). ROC using temporoparietal meta-VOI Tau1 discriminated increase vs decrease in Tau2–Tau1 with AUC=0.852; optimal cut-off Tau1=1.6, defining low-Tau1 (n=17) and high-Tau1 (n=10) subgroups. - Subgroup trajectories: Low-Tau1 patients showed increases in SUVr across temporoparietal and frontal cortices (significant vs controls in temporoparietal regions). High-Tau1 patients showed frontal SUVr increases but temporoparietal decreases (significant vs controls on the left). High-Tau1 patients were younger and declined clinically faster. - Atrophy progression: AD exhibited progressive cortical atrophy in temporoparietal and frontal regions; compared to controls, significant temporal lobe differences (VOI and voxelwise), extending to posterior-inferior parietal cortex voxelwise. High-Tau1 showed greater atrophy progression than low-Tau1 in temporal and parietal lobes; hippocampal volume loss did not differ between subgroups. - Tau change vs atrophy change: No significant correlations between Tau2–Tau1 and MRI2–MRI1 regionally or voxelwise (VOI r^2 values were low: lateral parietal 0.05; lateral temporal 0.11; frontal 0.009). - Imaging change vs cognition: Cognitive decline associated with increased frontal tau SUVr and with decreased temporoparietal tau SUVr (left predominant) across domains. In low-Tau1 subgroup, only modest associations were seen between increased SUVr (scattered, predominantly left frontal) and memory decline. By contrast, declines in memory, instrumental, and executive composite scores showed strong, anatomically congruent associations with regional atrophy progression (temporal for memory; parietal for instrumental; fronto-parietal for executive).

Findings indicate heterogeneous longitudinal tau-PET trajectories in early symptomatic AD, with region- and baseline-load–dependent patterns. High baseline temporoparietal tau predicted paradoxical decreases in SUVr in the same regions, alongside faster clinical decline and greater temporoparietal atrophy, while frontal SUVr continued to increase. The decrease in temporoparietal SUVr is unlikely to be an artifact given robust longitudinal registration, alternative reference regions, and PVE checks; it may reflect biological changes that reduce tracer affinity in advanced pathology (e.g., transition to ghost tangles, isoform composition shifts, or post-translational modifications). Clinically, cognitive decline related strongly to regional atrophy progression and only weakly to tau SUVr increases; notably, declining cognition also associated with decreasing temporoparietal SUVr, supporting the interpretation that reduced tracer binding can mark advanced, destructive stages. Thus, while baseline tau burden predicts future decline, longitudinal tau-PET change is a more complex signal than a simple increase and may not directly index neurodegenerative progression, which is better captured by atrophy.

Tau-PET can help stratify AD patients by progression patterns: higher initial temporoparietal tau load and younger age identify a subgroup with faster clinical decline, frontal tau increases, and temporoparietal SUVr decreases over time. These dynamics likely reflect biological maturation of tau pathology with reduced tracer affinity in advanced stages. For therapeutic trials, inclusion criteria and outcome measures should account for these patterns: consider focusing tau-PET endpoints on regions with consistent increases (e.g., frontal) and incorporate MRI-based atrophy progression due to its strong linkage with cognitive decline. Future research should validate these findings in larger cohorts, over longer durations, and investigate tracer sensitivity to tau maturation states.

Relatively small AD sample (n=27) reduces power and generalizability, potentially affecting voxelwise analyses. The follow-up duration (~2 years) may be insufficient to capture longer-term tau biology transitions. Although multiple methodological safeguards were used, residual measurement variance and potential confounds cannot be fully excluded.