Mapping Amyloid Networks in Alzheimer’s Disease: A High-Order ICA Approach to Gray and White Matter Pathology

Nigar Khasayeva, Kyle M. Jensen, Cyrus Eierud, Helen Petropoulos, Ruiyan Luo, Enrico Premi, Barbara Borroni, Jiayu Chen, Vince D. Calhoun, Armin IrajibioRxiv 2025 · DOI

TL;DR

This study uses a sophisticated imaging analysis technique (high-order ICA) on PET scans to map the spread of amyloid-beta (Aβ) in both gray and white matter in Alzheimer's disease (AD). They found that Aβ deposition follows specific network patterns and that these patterns change predictably across the disease progression from cognitively normal to mild cognitive impairment (MCI) to full AD dementia, offering new insights into early AD pathology and potential biomarkers.

What Problem Does This Paper Address?

Alzheimer's disease (AD) is characterized by the buildup of amyloid-beta (Aβ) plaques and neurofibrillary tangles, leading to cognitive decline. While changes in gray matter (GM) are well-studied, the involvement of white matter (WM) and the precise spatial patterns of Aβ accumulation across the entire brain, especially in the early stages of the disease (like mild cognitive impairment or MCI), remain less understood. Traditional methods for analyzing brain scans often focus on specific regions of interest or individual voxels, which can miss complex, distributed patterns of pathology and their relationships. This study aims to address this gap by using a data-driven approach to identify and characterize Aβ networks throughout the brain, including WM, and to understand how these networks evolve with disease progression and relate to cognitive function.

Methods

The researchers employed high-order independent component analysis (ICA), a powerful statistical technique for decomposing complex data into underlying independent patterns. Specifically, they applied ICA with a high model order (100 components) to [18F]Florbetapir Positron Emission Tomography (PET) scans from a large cohort. This allowed them to identify distinct, spatially independent networks of Aβ uptake in both gray matter (GM) and white matter (WM). They then analyzed the "loadings" of these components for each individual, which represent how strongly that individual exhibits the pattern of Aβ uptake associated with that network. These network loadings were then compared across different diagnostic groups (cognitively normal (CN), MCI, and AD dementia) and correlated with various cognitive and neuropsychiatric measures.

Datasets

The study utilized data from 716 participants in the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. This dataset included [18F]Florbetapir PET scans, along with demographic, clinical, and cognitive information. For the primary analysis, one time point per subject was used, with the second time point reserved for assessing replicability. The final sample size after quality control and site selection was 624 subjects.

Key Insights and Ideas

  • Amyloid spread isn't random: Aβ deposition doesn't occur haphazardly but rather follows specific, interconnected network patterns within both GM and WM.
  • MCI is a transitional state: The study reinforces the idea that MCI represents a bridge between normal aging and AD, with Aβ network patterns in MCI individuals often falling between those of CN and AD groups.
  • White matter matters early on: The research highlights the significant involvement of WM networks in AD pathology, suggesting that changes in white matter integrity are not just a consequence but potentially an early contributor to cognitive decline.
  • Network patterns reveal disease trajectory: The identified networks show a clear progression of Aβ accumulation and its association with cognitive decline, offering a more nuanced understanding of how the disease unfolds.
  • ICA is a powerful tool for AD research: High-order ICA proves effective in uncovering subtle, distributed patterns of pathology that might be missed by traditional region-based analyses.

Key Results

  • The analysis identified 80 gray matter (GM) networks and 13 white matter (WM) networks showing distinct patterns of Aβ uptake.
  • Network progression: For several GM networks, particularly in the higher cognition-frontal (HC-FR) and triple network-salience (TN-SA) domains, MCI profiles were more closely aligned with AD than with CN profiles, indicating an earlier shift towards AD-like pathology.
  • Cognitive associations: GM networks like the hippocampal-entorhinal complex and precuneus, and WM networks like the retrolenticular internal capsule, showed strong associations with cognitive performance.
  • WM network significance: Notably, the posterior corpus callosum (PCoCa) showed a strong and replicable effect between CN and MCI, suggesting its importance in early AD.
  • Replicability: The findings were largely replicated in an independent subset of the data, underscoring their robustness.

Limitations and Future Work

The authors acknowledge several limitations:

  • Single-tracer emphasis: The study primarily relied on [18F]Florbetapir PET, which measures amyloid-beta. Future work could integrate multimodal imaging (e.g., tau PET, FDG-PET, fMRI) to provide a more comprehensive picture of AD pathology.
  • Cross-sectional design: The study used cross-sectional data, which limits the ability to definitively track individual disease progression over time. Longitudinal studies are needed to confirm these trajectories.
  • Site/tracer heterogeneity: While efforts were made to account for site effects, potential heterogeneity in tracer uptake across different sites could influence results.
  • Cerebellar contributions: The authors suggest that future studies should apply higher model orders, cerebellar-specific parcellation, and partial-volume correction to better understand the specific contributions of cerebellar GM and WM.

Future work will focus on applying multimodal ICA to fMRI and PET data to capture effects at multiple spatial scales and across different stages of AD. They also plan to investigate AD with co-occurring psychosis and depression, which are common comorbidities that can worsen outcomes. The study also highlights the need for harmonized pipelines and longitudinal follow-up to predict conversion from MCI to AD.

Summary updated

Abstract

Abstract INTRODUCTION Alzheimer’s Disease (AD) is a neurodegenerative disorder marked by gray matter (GM) changes driven by amyloid-beta (Aβ) plaques and neurofibrillary tangles. While GM alterations are well documented, spatially distinct patterns of homogeneous Aβ uptake and white matter (WM) involvement remain underexplored. METHODS We applied high-order independent component analysis (ICA) to 716 [18F]Florbetapir PET scans, identifying 80 GM and 13 WM networks. Diagnostic and cognitive associations were evaluated via statistical modeling. RESULTS Identified networks delineated a progression trajectory, with mild cognitive impairment (MCI) profiles in temporoparietal and frontal subdomains more closely aligned with AD than cognitively normal (CN) profiles. GM networks, including the hippocampal-entorhinal complex and precuneus, and WM networks, including the retrolenticular internal capsule, demonstrated robust associations with cognitive performance. DISCUSSION Our findings highlight the utility of high-order ICA in identifying reproducible Aβ networks and the contribution of WM networks, such as the posterior corpus callosum, in the early pathological landscape of AD.
BibTeX
@misc{khasayeva2025mapping,
  author = {Nigar Khasayeva and Kyle M. Jensen and Cyrus Eierud and Helen Petropoulos and Ruiyan Luo and Enrico Premi and Barbara Borroni and Jiayu Chen and Vince D. Calhoun and Armin Iraji and Vince D. Calhoun and Armin Iraji},
  title = {{Mapping Amyloid Networks in Alzheimer’s Disease: A High-Order ICA Approach to Gray and White Matter Pathology}},
  year = {2025},
  doi = {10.1101/2025.03.21.644614},
  howpublished = {bioRxiv},
}
Figure