CRCNS: Deep Learning to Discover Neurovascular Disruptions in Alzheimer's Disease

The Neurovascular unit (NVU) comprises of cells in the brain vasculature (endothelial cells and pericytes) working in coordination with parenchymal cells (neurons and astrocytes) to maintain brain homeostasis and cognitive function. Intricate functional interactions among NVU cells, referred to as neurovascular coupling, is progressively impaired and NVU composition is severely disrupted in Alzheimer’s disease. However, the underlying pathophysiological mechanisms are poorly understood due to paucity of molecular level information on the less abundant, yet functionally critical, cerebrovascular endothelial cells and pericytes. The fMRI imaging, widely used in the clinic to evaluate neurovascular coupling, may not inform molecular level changes. It is challenging to identify changes in NVU composition using standard histopathological methods, because they lack sensitivity and specificity to locate endothelial cells and pericytes in the brain tissue. Bulk RNA sequencing from postmortem Alzheimer's brain tissue can be used to investigate NVU components, but it measures gene expression averaged across all cells, thus making it difficult to define cell-specific pathways and NVU constituent interactions. Single-cell methods and linear deconvolution techniques are currently employed to analyze bulk RNA sequencing data to determine cell-type-specific gene expression patterns. However, these techniques struggle to capture the molecular signature of low-abundant cells like endothelial and pericytes. The objective of the current study is to develop deep-learning methods to accurately predict the composition and transcriptomic signature of NVU cells, and to map interactions among them. Our central hypothesis is that data-driven deep-learning models, which have the flexibility to capture underlying gene-gene and cell-cell interactions in the brain tissue, will predict the composition and transcriptomic signature of NVU cells more effectively than the conventional methods. In Aim 1, we will design NUGENT, a novel deep-learning framework, to identify cell-type composition and predict cell-type-specific gene expression patterns. In Aim 2, we will validate NUGENT using new scRNA-seq data of NVU constituent cells harvested from Alzheimer’s disease transgenic mice (APPswe/PSEN1dE9) and their non-transgenic littermates. Employing the data generated in Aim 2 and publicly available patient and mouse data on the NUGENT framework, in Aim 3 we will investigate molecular pathways regulating neurovascular coupling in cognitively normal and Alzheimer’s patients. It is highly likely that the proposed studies will help identify molecular determinants of neurovascular dysfunction underlying age-related cognitive decline and Alzheimer’s dementia and facilitate the discovery of novel biomarkers and therapeutic targets.