Illuminating early microglial dysfunction in Alzheimer's disease through integration of explainable AI and iPSC models

Alzheimer Disease (AD) pathogenesis is multifactorial involving multiple cell types which offers several points of intervention. Microglia, the innate immune cell of the brain, are implicated in AD risk and pathogenesis. However, microglia are phenotypically diverse, and which microglial activities are most relevant to AD are not yet known. Until recently, studies to find microglia signatures in human brain tissue were limited by microglia numbers precluding a full appreciation of how microglial states may change or contribute to the disease course. With the explosion of human brain single cell omics studies the field is now empowered to more clearly define regulation of microglia in aging and disease. However, resolving microglial states from human brain omic data is particularly complicated as they are less transcriptomically distinct from each other than different brain cell types are from each other. Better approaches are needed to uncover the suspected subtle microglial changes happening early in disease progression. To address this problem and nominate additional AD relevant microglial states we integrate our novel Explainable AI technique with our deep learning method, ContrastiveVI designed to overcome heterogeneity in samples and pull-out subtle cell states. We hypothesize that our novel AI approach will enable better distinction of AD specific pathways from general aging. Two examples of microglial AD altered pathways we and others have identified are neuronal surveillance and microglial motility. However, it is not clear to what degree these nuanced AD microglia states reflect responses to neuronal damage or pathologic proteins or both. As proof of concept to study computationally identified pathways, we will model the concomitant stressors of AD stimuli and neuronal injury due to aging. We will expose microglia-neuron hiPSC cultures to Aβ, tau and UV irradiated neuronal conditioned media and perform functional and single cell RNA single nuclei-ATAC sequencing. We will validate computationally predicted AD microglial genetic drivers and pathways with network analysis of our in vitro perturbation models. We further hypothesize that the presence of AD pathologic proteins alters computationally predicted AD specific microglia pathways in the setting of neuronal injury which we will test using functional assays. These complementary studies innovate approaches to single cell brain datasets to find the subtleties of microglial AD states while leveraging iPSC perturbation in vitro studies in a controlled setting to systematically test specific pathways and determine the impacts of specific perturbation on gene regulation.