Biomarkers

Alzheimers Dement. 2025 Dec;21 Suppl 2:e105737. doi: 10.1002/alz70856_105737.

ABSTRACT

BACKGROUND: The clinical heterogeneity of Early-Onset Alzheimer's Disease (EOAD) is a key factor behind delayed diagnosis within this young(<65yo) group. However, most research has focused on late-onset amnestic participants and largely underutilized tau-PET, despite its ability to link neuropathology with clinical outcomes. We aimed to characterize tau-based subtypes through a robust data-driven approach in the Longitudinal Early-onset Alzheimer's Disease Study.

METHOD: Baseline [18F]Flortaucipir-PET scans from 365 amyloid-PET-positive participants with sporadic EOAD were quantified in 10 regions: left and right medial temporal, lateral temporal, occipital, parietal, and frontal. Tau-PET values were z-scored against 85 amyloid-PET-negative cognitively normal age-matched controls and fitted into Subtype and Stage Inference (SuStaIn), an unsupervised clustering algorithm that simultaneously models subtypes and progression from cross-sectional data.

RESULT: Three tau-PET-based subtypes were identified (Figure 1): Subtype1 (n = 144, 39.5%) had a typical bilateral temporoparietal pattern, while Subtype2 (n = 111, 30.4%) showed predominant left temporal binding, and Subtype3 (n = 104, 28.5%) showed early occipital tau. Subtypes show no significant demographic differences (Figure 2a) but were associated with clinical presentations. Subtype1 was enriched in amnestic participants, while Subtype2 accounted for 61% participants with primary progressive aphasia, and Subtype3 included 79% participants with posterior cortical atrophy (Figure 2b). At baseline, higher SuStaIn stages were associated with higher CDR-SB (Figure 2c). All subtypes showed longitudinal increase in CDR-SB, but clinical decline was faster in Subtype1 (Figure 2d). When follow-up Flortaucipir-PET scans were fitted to SuStaIn trained on baseline data, 85.6% participants were clustered within the same subtype as their baseline scans; SustaIn stage increased by 0.56/year on average with no difference across subtypes (Figure 3a-b). Modeling voxelwise tau-PET over time revealed striking differences (Figure 3c), as each subtype showed significant accumulation in regions that were relatively spared at baseline: tau-PET increase predominated in the occipital lobe for Subtype1, in bilateral frontal and right temporal areas for Subtype2, and bilateral frontotemporal lobes for Subtype3.

CONCLUSION: SuStaIn was able to identify robust tau-PET-based subtypes that were associated, but not redundant with known clinical phenotypes in AD. Subtypes exhibited differences in prospective clinical decline and patterns of tau-PET changes, highlighting their potential to refine prognosis and improve progression monitoring in clinical practice and trials.

PMID:41505031 | DOI:10.1002/alz70856_105737