A whole-brain functional connectivity model of Alzheimer's disease pathology

Abstract

Introduction: Alzheimer's disease (AD) is characterized by the presence of two proteinopathies, amyloid and tau, which have a cascading effect on the functional and structural organization of the brain.

Methods: In this study, we used a supervised machine learning technique to build a model of functional connections that predicts cerebrospinal fluid (CSF) p-tau/Aβ₄₂ (the PATH-fc model). Resting-state functional magnetic resonance imaging (fMRI) data from 289 older adults in the Alzheimer's Disease Neuroimaging Initiative (ADNI) were utilized for this model.

Results: We successfully derived the PATH-fc model to predict the ratio of p-tau/Aβ₄₂ as well as cognitive functioning in older adults across the spectrum of healthy and pathological aging. However, the in-sample fit magnitude was low, indicating a need for further model development.

Discussion: Our pathology-based model of functional connectivity included representation from multiple canonical networks of the brain with intra-network connectivity associated with low pathology and inter-network connectivity associated with higher levels of pathology.

Highlights: Whole-brain functional connectivity model (PATH-fc) is linked to AD pathophysiology. The PATH-fc model predicts performance in multiple domains of cognitive functioning. The PATH-fc model is a distributed model including representation from all canonical networks.

Keywords: cognition; connectome‐based predictive modeling; distributed networks; functional connectivity; resting‐state.

FIGURE 1

Schematic presentation of the CPM model. This figure helps illustrate the process of constructing a CPM model. CPM, connectome‐based predictive modeling.

FIGURE 2

Visual depiction of the high (in red) and low (in blue) PATH‐fc model and the in‐sample fit for the two models and the combined model (in green). (A) Presents the anatomical distribution of the high and low PATH‐fc model. The 677 edges included nodes in all the macroscale regions. (B) Internal model fit of the high, low, and combined PATH‐fc model to the ADNI pathology data. (C) Internal model fit of the combined PATH‐fc model to predict cognitive functioning in PACC, ADNI‐MEM, and ADNI‐EF, respectively. The scatterplots represent Spearman correlation between observed scores (AD pathology, PACC, ADNI‐MEM, and ADNI‐EF) and predicted scores for the same metric derived using cross‐validation. Histogram represents the randomly shuffled connectivity matrices and respective metric pairings over 1000 iterations to compute the final p‐value for the significance. AD, Alzheimer's disease; ADNI, Alzheimer's Disease Neuroimaging Initiative; ADNI‐EF, Alzheimer's disease neuroimaging initiative executive functioning; ADNI‐MEM, Alzheimer's disease neuroimaging initiative executive memory; CU, cognitively unimpaired; MCI, mild cognitive impairment; PACC, Preclinical Alzheimer's Cognitive Composite score.

FIGURE 3

Presentation of the involvement of the 10 canonical networks in the high (A) and low (B) PATH‐fc model with nodes parcellated based on Noble et al. (2017). Ribbons in the ring plot visualization are proportional to the representation of each network in the PATH‐fc model. The matrix presented in the top and bottom panels shows the relative contribution of each network to the high and low PATH‐fc model, respectively. These network contributions have been adjusted using the formula described in Greene et al., 2018.