A Whole-Brain Functional Connectivity Model of Alzheimer’s Disease Pathology

Abstract

AbstractEarly detection of Alzheimer’s disease (AD) is a necessity as prognosis is poor upon symptom onset. Although previous work diagnosing AD from protein-based biomarkers has been encouraging, cerebrospinal (CSF) biomarker measurement of AD proteins requires invasive lumbar puncture, whereas assessment of direct accumulation requires radioactive substance exposure in positron emission tomography (PET) imaging. Functional magnetic resonance imaging (fMRI)-based neuromarkers, offers an alternative, especially those built by capitalizing on variance distributed across the entire human connectome. In this study, we employed connectome-based predictive modeling (CPM) to build a model of functional connections that would predict CSF p-tau/Aβ42 (PATH-fc model) in individuals diagnosed with Mild Cognitive Impairment (MCI) and AD dementia. fMRI, CSF-based biomarker data, and longitudinal data from neuropsychological testing from the Alzheimer’s Disease NeuroImaging Initiative (ADNI) were utilized to build the PATH-fc model. Our results provide support for successful in-sample fit of the PATH-fc model in predicting AD pathology in MCI and AD dementia individuals. The PATH-fc model, distributed across all ten canonical networks, additionally predicted cognitive decline on composite measures of global cognition and executive functioning. Our highly distributed pathology-based model of functional connectivity disruptions had a striking overlap with the spatial affinities of amyloid and tau pathology, and included the default mode network as the hub of such network-based disruptions in AD. Future work validating this model in other external datasets, and to midlife adults and older adults with no known diagnosis, will critically extend this neuromarker development work using fMRI.Significance StatementAlzheimer’s disease (AD) is clinical-pathological syndrome with multi-domain amnestic symptoms considered the hallmark feature of the disease. However, accumulating evidence from autopsy studies evince support for the onset of pathophysiological processes well before the onset of symptoms. Although CSF- and PET-based biomarkers provide indirect and direct estimates of AD pathology, both methodologies are invasive. In here, we implemented a supervised machine learning algorithm – connectome-based predictive modeling – on fMRI data and found support for a whole-brain model of functional connectivity to predict AD pathology and decline in cognitive functioning over a two-year period. Our study provides support for AD pathology dependent functional connectivity disturbances in large-scale functional networks to influence the trajectory of key cognitive domains in MCI and AD patients.