Multimodal and multiscale brain networks : understanding aging, Alzheimer’s disease, and other neurodegenerative disorders
The human brain can be modeled as a complex network, often referred to as the connectome, where structural and functional connections govern its organization. Several neuroimaging studies have focused on understanding the architecture of healthy brain networks and have shed light on how these networks evolve with age and in the presence of neurodegenerative disorders. Many studies have explored the brain networks in Alzheimer’s disease (AD), the most common type of dementia, using various neuroimaging modalities independently. However, most of these studies ignored the complex and multifactorial nature of AD.
The aim of this thesis was to investigate and analyze the brain’s multimodal and multiscale network organization in aging and in AD by using different multilayer brain network analyses and different types of data. Additionally, this research extended its scope to incorporate other dementias, such as Lewy body dementias, allowing for a comparison of these disorders with AD and normal aging. These comparisons were made possible through the application of protein co-expression networks.
In Study I, we investigated sex differences in healthy individuals using multimodal brain networks. To do this we used resting-state functional magnetic resonance imaging (rs-fMRI) and diffusion-weighted imaging (DWI) data from the Human Connectome Project (HCP) to perform multilayer and deep learning analyses. These analyses identified differences between men's and women's underlying brain network organization, showing that the deep-learning analysis with multilayer network metrics (area under the curve, AUC, of 0.81) outperforms the classification using single-layer network measures (AUC of 0.72 for functional networks and 0.70 for anatomical networks). Furthermore, we integrated the multilayer brain networks methodology and neural network models into a software package that is easy to use by researchers with different backgrounds and is also easily expandable for researchers with different levels of programming experience.
Then, we used the multilayer brain networks methodology to study the interaction between sex and age on the functional network topology using a large group of people from the UK Biobank (Study II). By incorporating multilayer brain network analyses, we analyzed both positive and negative connections derived from functional correlations, and we obtained important insights into how cognitive abilities, physical health, and even genetic factors differ between men and women as they age. Age and sex were strongly associated with multiplex and multilayer measures such as the multiplex participation coefficient, multilayer clustering, and multilayer global efficiency, accounting for up to 89.1%, 79.9%, and 79.5% of the variance related to age, respectively. These results indicate that incorporating separate layers for positive and negative connections within a complex network framework reveals sensitive insights into age- and sex-related variations that are not detected by traditional metrics. Furthermore, our functional metrics exhibited associations with genes that have previously been linked to processes related to aging.
In Study III, we assessed whether multilayer connectome analyses could offer new perspectives on the relationship between amyloid pathology and gray matter atrophy across the AD continuum. Subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) were divided into four groups based on cerebrospinal fluid (CSF) amyloid-β (Aβ) biomarker levels and clinical diagnosis. We compared the different groups using weighted and binary multilayer measures that assess the strength of the connections, the modularity, as well as the multiplex segregation and integration of the brain connectomes. Across Aβ-positive (Aβ+) groups, we found widespread increases in the overlapping connectivity strength and decreases in the number of identical connections in both layers. Moreover, the brain modules were reorganized in the mild cognitive impairment (MCI) Aβ+ group and an imbalance in the quantity of couplings between the two layers was found in patients with MCI Aβ+ and AD Aβ+.
Using a subsample from the same database, ADNI, we analyzed rs-fMRI data from individuals at preclinical and clinical stages of AD (Study IV). By dividing the time series into different time windows, we built temporal multilayer networks and studied the modular organization across time. We were able to capture the dynamic changes across different AD stages using this temporal multilayer network approach, obtaining outstanding areas under the curve of 0.90, 0.92 and 0.99 in the distinction of controls from preclinical, prodromal, and clinical AD stages, respectively, on top and beyond common risk factors. Our results not only improved the discrimination between various disease stages but, importantly, they also showed that dynamic multilayer functional measures are associated with memory and global cognition in addition to amyloid and tau load derived from positron emission tomography. These results highlight the potential of dynamic multilayer functional connectivity measures as functional biomarkers of AD progression.
In Study V, we used in-depth quantitative proteomics to compare post-mortem brains from three key brain regions (prefrontal cortex, cingulate cortex, and the parietal cortex) directly related to the disease mechanisms of AD, Parkinson’s disease with dementia (PDD), dementia with Lewy bodies (DLB) in prospectively followed patients and older adults without dementia. We used covariance weighted networks to find modules of protein sets to further understand altered pathways in these dementias and their implications for prognostic and diagnostic purposes.
In conclusion, this thesis explored the complex world of brain networks and offered insightful information about how age, sex, and AD influence these networks. We have improved our understanding of how the brain is organized in different imaging modalities and different time scales, as well as developing software tools to make this methodology available to more researchers. Additionally, we assessed the connections among various proteins in different areas of the brain in relation to health, Alzheimer's disease, and Lewy body dementias. This work contributes to the collective effort of unraveling the mysteries of the human brain organization and offers a foundation for future research to understand brain networks in health and disease.
List of scientific papers
I. Gómez-Ruiz E*, Canal-Garcia A*, Chang YW*, Zhao H, Zufiria-Gerbolés B, Sun J, Veréb D, Westman E, Mijalkov M, Pereira JB, Volpe G. Brain Connectivity Analysis with Multilayer Graphs and Deep Learning. * Authors contributed equally to this work as first authors. [Manuscript]
II. Mijalkov M, Veréb D, Jamialahmadi O, Canal-Garcia A, Gómez-Ruiz E, VidalPiñeiro M, Romeo S, Volpe G, Pereira JB. 2023. Sex differences in multilayer functional network topology over the course of aging in 37543 UK Biobank participants. Network Neuroscience. 7(1):351-76.
https://doi.org/10.1162/netn_a_00286
III. Canal-Garcia A, Gómez-Ruiz E, Mijalkov M, Chang YW, Volpe G, Pereira JB, Alzheimer’s Disease Neuroimaging Initiative. 2022. Multiplex Connectome Changes across the Alzheimer’s Disease Spectrum Using Gray Matter and Amyloid Data. Cerebral Cortex. 32(16):3501-3515.
https://doi.org/10.1093/cercor/bhab429
IV. Canal-Garcia A, Veréb D, Mijalkov M, Westman E, Volpe G, Pereira JB, Alzheimer’s Disease Neuroimaging Initiative. 2024. Dynamic multilayer functional connectivity detects preclinical and clinical Alzheimer’s disease. Cerebral Cortex. bhad542.
https://doi.org/10.1093/cercor/bhad542
V. Canal-Garcia A, Branca RM, Francis PT, Winblad B, Lehtiö J, Nilsson P, Aarsland D, Pereira JB, Bereczki E. Comparative brain regional specific proteomic signatures of Alzheimer’s disease and Lew body dementias. [Manuscript]
History
Defence date
2024-03-04Department
- Department of Clinical Neuroscience
Publisher/Institution
Karolinska InstitutetMain supervisor
Pereira, Joana BragaCo-supervisors
Westman, Eric; Bereczki, ErikaPublication year
2024Thesis type
- Doctoral thesis
ISBN
978-91-8017-201-1Number of supporting papers
5Language
- eng