A brain multilayer approach to predicting cognitive decline in healthy aging
Poster No:
1128
Submission Type:
Abstract Submission
Authors:
Barbara Avelar Pereira1, Lars Bäckman1
Institutions:
1Karolinska Institutet, Stockholm, Stockholm
First Author:
Co-Author:
Introduction:
Aging is characterized by decline in cognition. Still, individuals do not age in the same way (1). There are large inter-individual differences in onset and rate of cognitive decline. Longitudinal research typically focuses on pathological aging, such as in Alzheimer's disease (AD). Little emphasis is given to the cognitive heterogeneity that exists in healthy older individuals or to the fact that most processes linked to normal aging, such as atrophy or inflammation, are also present in pathological aging (2). This suggests that understanding age-related brain pathology and cognitive deficits first requires understanding brain and cognitive decline in healthy aging. Neuroimaging techniques have, individually or in conjunction with others, been shown to predict cognitive decline to different extents and at different stages of development (3). However, most studies combining more than one modality typically report results in parallel or compare normalized findings but do not analyze data simultaneously.
Methods:
We tested a multi-dimensional network framework, which aggregates data across three brain modalities into a single model capable of capturing heterogeneity in cognitive aging. For this, we used the Cognition, Brain, and Aging (COBRA) study, a large-scale cohort of 181 healthy older adults (age 64-68 at baseline; 44.8% females) followed over 10 years. There are 3 waves, separated by 5 years, and participants go through MRI (structural and functional) and PET (dopamine D2) at each wave. There is also a full cognitive battery, lifestyle, health, and genetics information. The multilayer model included three modalities (i.e., layers): (1) functional connectivity, (2) grey-matter estimates, and (3) dopamine (DA). All data underwent mean centering, regressing out covariates (e.g., age, sex, intracranial volume), and normalization. These data were used to create a subject similarity network, with nodes representing individuals and edges representing similarities between them using pairwise correlations. Finally, a multi-dimensional network framework (i.e., a multilayer network) was constructed (Fig 1). Several metrics can be derived, but the main outcome was modularity since it captures strength of network partition, which is derived using an iterative generalized Louvain community detection algorithm optimized for multilayer frameworks based on the work by Mucha et al. 2010.
Results:
We postulated that, by detecting common as well as complementary signals across modalities and minimizing the effect of different scales and noise, the multilayer network would provide insights into the development and progression of cognitive decline. We have previously applied this methodology to 490 individuals at risk for AD and shown that the model can distinguish cognitively normal (CN) from AD participants (Fig 2A). The method is data driven and blind to the label of each participant (i.e., CN or AD), but these labels are still created with information included in the model (5). As such, it remained unclear how well the network performs on a healthy sample with a comprehensive brain imaging and cognitive battery not tailored to detect a specific kind of dementia. Our framework identified groups with distinct cognitive profiles (Fig 2B). The groups differed in working memory (t(150.2)=-2.3, p=0.011) and the group with better scores also had higher white-matter integrity (t(165)=-2.8, p=0.003). There were no differences in episodic memory (p=0.445), but the group with better working memory performance and higher fractional anisotropy also had marginally better processing speed (p=0.09).
Conclusions:
A full understanding of cognitive aging warrants the inclusion of interactions within and between modalities. Multimodal biomarkers seem to be good predictors of cognition in healthy aging and correlate with other estimates of brain integrity. Gray matter, functional connectivity, and DA can provide unique insights when investigating differences in healthy populations.
Learning and Memory:
Working Memory
Lifespan Development:
Aging 1
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural)
Multivariate Approaches 2
PET Modeling and Analysis
Keywords:
Aging
Cognition
Data analysis
FUNCTIONAL MRI
MRI
Multivariate
NORMAL HUMAN
STRUCTURAL MRI
1|2Indicates the priority used for review
·Figure 1
·Figure 2
Provide references using author date format
2. Jagust, W. (2013). 'Vulnerable neural systems and the borderland of brain aging and neurodegeneration'. Neuron, 77(2), 219-234.
3. Nyberg, L. et al. (2020). 'Biological and environmental predictors of heterogeneity in neurocognitive ageing: Evidence from Betula and other longitudinal studies'. Ageing research reviews, 64, p.101184.
4. Mucha, P. J. et al. (2010). 'Community structure in time-dependent, multiscale, and multiplex networks'. Science, 328(5980), 876-878.
5. Avelar-Pereira, B., et al. (2022). Decoding the heterogeneity of Alzheimer’s disease diagnosis and progression using multilayer networks. Molecular Psychiatry, 1-10.