Project Title: Bayesian Inference Identifies Key Drivers of Neuroprotection in Alzheimer’s Disease
Project Mentor: Jeff Kuret – Biological Chemistry and Pharmacology
Abstract: Alzheimer’s Disease (AD) is the leading cause of dementia and a major contributor to mortality, with only mitigating therapy offered. AD is characterized by two molecular motifs in the brain: neurofibrillary tangles (NFTs) and amyloid-beta plaques (Aβ plaques). Both of these motifs arise from protein misfolding and propagate to affect the entire brain. The one exception to that trend is the cerebellum, which has repeatedly been shown to have no NFT formation despite having strong connectivity to affected brain regions. Further analysis of brain-regional differences in AD has revealed transcriptomic and proteomic signatures unique to the cerebellum, with the functionality of these signatures being indicative of a “neuroprotective” phenotype against NFT formation.
Using various computational tools, we proceeded to examine how these neuroprotective signatures interact with each other. We utilized a microarray dataset containing transcriptomic data from various brain regions of AD patients and applied differential expression analysis between the cerebellum (CB) and prefrontal cortex (PFC). We then proceeded to use the differentially expressed gene set within various other software tools to create a multiscale network using a Bayesian algorithm. This process allowed us to delineate directed causal relationships between genes and find causal drivers of neuroprotective gene expression, which were validated in biochemical and cellular models.
We determined that these neuroprotective signatures are not only unique to the cerebellum but also are co-expressed differently in the cerebellum. We identified three groups or modules of uniquely coexpressed genes in the cerebellum that was highly enriched for heat shock genes critical to proper protein folding. These modules were enriched for the protein folding gene ontology grouping which is key considering that proper protein folding is required to prevent the misfolding events of NFT formation. We combined these modules and proceeded to construct a multiscale network and find the causal regulators. Afterward, we used a machine learning-based algorithm to see if the determined causal regulators are responsive to AD pathology by predicting clinical measurements of AD. Based on the level of causality in the network and the ability to predict AD clinical measurements, we ranked causal regulators that can be validated experimentally to prevent NFT formation.
Keywords: Bioinformatics, Neuroscience, Genetics
Acknowledgments: Dr. Jeff Kuret, Austin Allen, Chris Ayoub, Ohio Supercomputing Center