TL;DR
This study investigates how genetic risk for Alzheimer's Disease (AD), quantified by a Polygenic Hazard Score (PHS), relates to multimodal brain changes across the AD continuum (healthy controls, mild cognitive impairment, AD). By fusing functional MRI, structural MRI, and amyloid PET data, they found that PHS is associated with impairments in key brain networks, with functional and amyloid changes appearing earlier than structural atrophy. This work highlights the complex interplay between genetics and neurodegeneration in AD.
What Problem Does This Paper Address?
Late-onset Alzheimer's Disease (LOAD) is a complex, polygenic disorder. While genome-wide association studies (GWAS) have identified numerous genetic risk factors, understanding how these genetic predispositions translate into observable brain changes and ultimately lead to cognitive decline remains a challenge. Previous research often focused on single genetic variants or single neuroimaging modalities. This paper addresses two key gaps:
- Capturing overall genetic risk: Individual genetic variants may not fully represent an individual's total genetic risk for AD. The authors propose using a Polygenic Hazard Score (PHS), which integrates multiple risk variants, as a more comprehensive measure of genetic predisposition.
- Integrating multimodal information: AD pathology affects multiple aspects of the brain (structure, function, and molecular deposition). Analyzing these modalities in isolation might miss crucial inter-relationships. The paper aims to leverage multimodal fusion techniques to uncover how PHS-associated changes manifest across different imaging types and how these changes evolve along the AD continuum.
Methods
The researchers employed a sophisticated data-driven approach called Multisite Canonical Correlation Analysis with Reference + Joint Independent Component Analysis (MCCAR + jICA). This method is designed to identify shared patterns across multiple data modalities (fMRI, structural MRI, PET) while being guided by a specific reference variable – in this case, the PHS.
Here's a breakdown of the key steps:
- Data Preprocessing: Standard preprocessing pipelines were used for resting-state fMRI (rsfMRI) to extract fractional amplitude of low-frequency fluctuations (fALFF), for structural MRI (sMRI) to derive gray matter volume (GMV), and for amyloid PET to obtain standard uptake value ratios (SUVR).
- PHS Calculation: A PHS was calculated for each participant, integrating genetic variants associated with AD risk.
- Multimodal Fusion (MCCAR + jICA):
- The preprocessed neuroimaging features (fALFF, GMV, amyloid SUVR) were concatenated for each participant.
- PHS was used as a reference to guide the joint decomposition of these multimodal features. This process aims to find components (spatial maps and their corresponding subject-wise loadings) that are maximally correlated with PHS and also show inter-modality associations.
- Fifteen components were estimated for each modality.
- Joint ICA was then applied to the concatenated spatial maps to obtain final independent components (ICs) and their mixing matrices.
- Statistical Analysis:
- Analysis of Variance (ANOVA) was used to compare the loadings of these components across the three groups (HC, MCI, AD).
- Post-hoc t-tests were performed to pinpoint group differences.
- Pearson correlations were used to examine the relationship between the loadings of the identified multimodal components and various cognitive scores (memory, executive function, language, visuospatial function).
Datasets
The study utilized data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. This is a well-established, publicly available resource for AD research, providing a rich collection of multimodal neuroimaging, genetic, and clinical data. The authors specifically selected participants from ADNI who had undergone T1-weighted structural scans, [18F]-AV45 PET scans, rsfMRI, PHS calculation, and comprehensive neuropsychological assessments.
Key Insights and Ideas
- PHS as a unifying genetic risk marker: The study demonstrates that a single PHS score can be associated with a spectrum of brain impairments, suggesting it effectively captures an individual's overall genetic susceptibility to AD.
- Early functional and amyloid changes: The findings suggest that functional impairments (decreased fALFF) and amyloid deposition (increased SUVR) occur earlier in the AD continuum, even in the MCI stage, preceding significant gray matter atrophy (decreased GMV). This aligns with the amyloid cascade hypothesis.
- Network-specific vulnerability: The identified PHS-associated multimodal pattern primarily involves regions within the Default Mode Network (DMN), Executive Control Network (ECN), and visuospatial network. This highlights the vulnerability of these critical brain networks to AD pathology.
- Multimodal covariation is key: The fusion approach revealed that PHS is linked to the covariation of changes across fALFF, GMV, and amyloid SUVR, suggesting that these modalities are not independently affected but rather interact in the disease process.
- Cognitive decline is linked to multimodal changes: The study found significant negative correlations between the loadings of the identified multimodal components and various cognitive scores, underscoring how these brain alterations contribute to cognitive dysfunction.
Key Results
- A single joint component, correlated with PHS, was identified, showing significant alterations across the AD continuum (HC, MCI, AD).
- This component revealed:
- Decreased fALFF in the precuneus, inferior parietal lobule (IPL), and middle temporal gyrus (MTG).
- Decreased GMV in the precuneus, IPL, and temporal regions.
- Increased amyloid SUVR in the precuneus, IPL, posterior cingulate cortex (PCC), and temporal regions.
- These regions are primarily part of the Default Mode Network (DMN).
- Other brain regions associated with the Executive Control Network (ECN) and visuospatial network also showed PHS-related changes.
- Functional abnormalities (fALFF) and amyloid deposition (SUVR) occurred earlier (HC to MCI transition) than GM atrophy (MCI to AD transition).
- Amyloid SUVR showed a continuous increase across the entire AD continuum.
- The loadings of the identified multimodal components were negatively correlated with memory, executive function, language, and visuospatial function.
Limitations and Future Work
The authors acknowledge several limitations:
- Unbalanced sample sizes: The AD group was significantly smaller than the HC and MCI groups, which could impact the robustness of conclusions for the AD stage. Future studies with balanced groups are recommended.
- Descriptive causality: While the study suggests amyloid as a preliminary pathogenesis, the intrinsic causality between amyloid, fALFF, and GMV is descriptive and requires further investigation.
- Cross-sectional design: The study is cross-sectional, and while it attempts to depict the AD continuum, a longitudinal study would provide more definitive insights into the progression of these changes.
- Exclusion of Tau PET: The lack of Tau PET data in the ADNI dataset used limited the ability to include this crucial marker of AD pathology. Future studies could incorporate Tau PET to further understand genetic effects.
- Focus on PHS: The study solely focused on PHS. Future work could explore other polygenic risk scores (PRS) or derive custom PRS to capture a broader genetic landscape.
The authors suggest that future research should consider these limitations and explore longitudinal designs, incorporate additional biomarkers like Tau PET, and investigate a wider range of polygenic risk scores to further elucidate the complex interplay between genetics and AD pathogenesis.
Summary updated