Single‐nucleus RNAseq‐derived pseudo‐temporal modeling of neurodegeneration in astrocytes of older brains

• Published in: Alzheimer's & dementia Vol. 18; no. S4
• Main Authors: Heath, Laura M, Wiley, Jesse C, Cary, Gregory A., Poehlman, William L., Gockley, Jake, Carter, Gregory W, Greenwood, Anna K, Mangravite, Lara M
• Format: Journal Article
• Language: English
• Published: 01.12.2022
• ISSN: 1552-5260; 1552-5279
• DOI: 10.1002/alz.068215

Background We previously developed a temporal model for unobserved molecular changes occurring during late‐onset Alzheimer’s Disease (AD) for bulk RNA‐Seq and proteomic data. Here we apply this method to single‐nucleus RNAseq data from an AD case‐control cohort and identify Alzheimer’s‐related pathway changes at specific stages of trajectories derived from astrocytes to better understand celltype‐specific gene expression across a continuum. Method Manifold learning defines an order across samples based on their similarity of expression. This ordering estimates pseudotime, an inference of molecular disease progression, quantitatively measured as the distance of each sample from the start of the inferred trajectory. We applied this approach to available snRNAseq data from human postmortem brain samples from the ROS/MAP study (N = 48) (Mathys et al., 2019). We identified a cluster of cells expressing astrocyte‐specific markers. Sex‐specific trajectories were calculated after applying a mutual nearest‐neighbors function to correct the expression matrix values for within‐donor variability. We examined associations between pseudotime and AD case/control status. For each tree branch relative to the root (i.e. the branch with the highest proportion of control cells), we performed differential expression analysis, using a method to account for the hierarchical nature of multi‐subject snRNAseq data (NEBULA, He et al., 2021). We then used gene set enrichment analysis to identify state‐specific significant GO terms from a set that have been curated into 16 distinct AD‐relevant biological domains. Result Pseudotime estimates were significantly associated with LOAD status (females, p = 0.0110; males, p = 0.0011), such that “early” (low pseudotime) samples are enriched for controls, and “late” (high pseudotime) samples are enriched for cases. Genes involved in synapse function and structural stabilization were upregulated, while proteostasis and immune response were downregulated, consistently across pseudotime in males, and later in females. Genes involved in the RNA spliceosome were upregulated at earlier stages but not the latest stage in pseudotime in both sexes. We also observed differences between branches in biological domains encompassing mitochondrial metabolism, vasculature, and apoptosis. Conclusion This approach to identifying AD‐related gene expression changes across a continuum provides an opportunity to glean new insights about celltype‐specific genetic drivers of AD in the brain.