Abstract
Alzheimer's disease (AD) arises from synergistic interactions between amyloid-β, tau, and neurodegeneration, yet it remains unclear how these mechanisms reshape the hierarchical organization of large-scale brain dynamics. Here, we quantified directed causal interactions across 134 participants spanning the AD continuum, comprising 46 amyloid-negative healthy controls (HC-), 36 amyloid-positive healthy controls (HC+), 31 amyloid-positive individuals with mild cognitive impairment (MCI+), and 21 amyloid-positive patients with AD dementia (AD+). Resting-state fMRI was modelled using generative effective connectivity, and hierarchical organization was assessed via trophic levels and directedness. Healthy older adults exhibited a canonical hierarchy in which sensorimotor and frontoparietal regions acted as causal sources, the default mode network (DMN) occupied an intermediate mediating position, and visual-limbic areas functioned as sinks. In contrast, AD+ individuals exhibited elevated trophic levels in the visual network and reduced levels in somatomotor, salience, control, and DMN systems. This shift was accompanied by decreased directedness, indicating a more flattened and less stratified architecture with reduced computational flexibility. MCI+ participants exhibited disruptions in somatomotor and dorsal attention networks. Compared to early-stage HC+, visual and DMN showed similar alterations while the control system returned to baseline before decreasing in AD+. Machine-learning classification distinguished all stages, including subtle differences between HC- and HC+. Hierarchical alterations were shaped by ATN biomarkers and strongly associated with cognitive decline, highlighting trophic metrics as sensitive neuroimaging biomarkers of AD progression. Together, these findings suggest that reduced hierarchical structure represents a core systems-level signature of AD, offering a promising avenue for early detection and therapeutic targeting.