Disruptions in brain hierarchical organisation reveal Alzheimer's disease progression

Alzheimer’s disease arises from interacting amyloid, tau, and neurodegenerative processes, and is associated with large-scale network dysfunction and reduced brain dynamics complexity. However, it remains unclear whether disease progression alters the causal hierarchical organisation of large-scale brain dynamics. We examined hierarchical brain organisation across the Alzheimer’s disease continuum using multimodal neuroimaging from 134 participants in the Alzheimer’s Disease Neuroimaging Initiative (46 amyloid-negative cognitively normal, 36 amyloid-positive cognitively normal, 31 amyloid-positive mild cognitive impairment, and 21 amyloid-positive Alzheimer’s disease dementia; mean age 73.9 ± 6.2 years). Participants underwent resting-state functional magnetic resonance imaging and positron emission tomography, with amyloid and tau imaging available in 121 individuals. We estimated the generative effective connectivity using whole-brain modelling and quantified hierarchy through trophic levels and directedness using an ecology-inspired framework. Group differences were assessed using non-parametric permutation testing with covariate adjustment and false discovery rate correction. A supervised machine learning framework using logistic regression with nested cross-validation and Monte Carlo resampling evaluated whether hierarchical features could classify disease stage. Associations with amyloid, tau burden, and grey matter volume, and cognitive performance were examined using mixed-effects ridge regression and multiple linear regression adjusted for age, gender, education and APOE-ε4 carrier status. Amyloid-negative cognitively normal exhibited a well-stratified hierarchy in which somatomotor and frontoparietal regions acted as sources, the default mode network occupied an intermediate integrative position, and memory, limbic, and visual systems acted as sinks. Across the clinical spectrum, we observed a progressive flattening of hierarchical organisation, characterised by reduced trophic levels in the somatomotor, salience, control, and default mode networks, and increased hierarchical positioning of visual and thalamic regions. Hierarchical features classified disease stage, including the preclinical distinction between amyloid-negative and amyloid-positive cognitively normal, indicating sensitivity to early stages. Hierarchical alterations were associated with amyloid, tau, and neurodegeneration. Moreover, they predicted cognitive impairment across all clinical tests. These findings suggest that Alzheimer’s pathology selectively disrupts hierarchical brain organisation. Our ecology-inspired framework offers a sensitive, non-invasive approach linking molecular pathology to large-scale network reconfiguration and cognitive decline, providing additional information beyond conventional correlation-based or graph-theoretical metrics, revealing its potential as a neuroimaging marker for early detection and disease staging across the Alzheimer’s continuum.