Mouse to Human Modeling of Microglia Single-Nuclei Transcriptomics Identifies Immune Signaling Pathways Associated with the Orexin Network

Introduction: Alzheimer’s disease (AD) is a progressive neurodegenerative disease characterized by memory loss and behavior change. Studies have found that dysregulation of microglial cells is pivotal to the development of AD. These mechanisms have been studied in mouse models to uncover potential therapeutic biomarkers. Despite these findings, there are limitations to the translatable biological information from mice to humans due to differences in physiology, timeline of disease, and the heterogeneity of humans. To address the inter-species discrepancies, we developed a novel implementation of the framework Translatable Components Regression (TransComp-R) which integrates microglia single-nuclei mouse and human transcriptomics data to identify biological pathways in mice predictive of human AD. Here, we compared adjustments to our model with sparse (sPCA) and principal component analysis (PCA). To determine therapeutic translatability, we tailored our model in the context of the orexin network, a series of signaling networks in the sleep and awake cycle reported to be associated with AD.

Materials and

Methods: We accessed publicly available data through Gene Expression Omnibus, obtaining single-nuclei RNA sequenced (snRNAseq) mouse 5xFAD/WT (GSE20683) and human AD/Control (GSE175814) data. One-to-one pairwise orthologs were kept between species for the analysis. Following the isolation of microglia with the Seurat R package, we performed differential expression analysis on mice and human data between control and condition with the MAST package to benchmark comparable biology across species. We then applied TransComp-R to the mouse and human data. TransComp-R projects human samples into mouse PC space to determine which mouse principal components (PCs) significantly reflect human biology. This is done by using the PC projections in a generalized linear model (GLM) and determining significant PCs. Significant PCs were selected by a stepwise Akaike Information Criterion (AIC) algorithm. The top and bottom 10% of the mouse PC loading genes on significant PCs were further correlated with differentially expressed genes from dual and selective orexin receptor antagonist signatures from the Library of Integrated Network-based Cellular Signatures (LINCS) database through a Fisher’s exact Test (p-value < 0.05). Finally, we ran gene set enrichment analysis with the fgsea package to identify the biological pathways that best described the separation between AD and control as well as relationships between dual and selective orexin receptor antagonists.

Results, Conclusions, and Discussions:

Results: Differential expression between control and AD in mouse and human data showed only 24 shared genes. We identified several mouse PCs from regular and sparse PCA which were reflective of human biology. From 50 regular principal components, our stepwise AIC model resulted in 22 PCs with an AIC of 1474.5. From 50 sparse principal components, our stepwise AIC model resulted in 21 PCs with an AIC of 1378.7. Both regular and sparse principal components had pathways relevant to AD biology. DEGs of selective orexin receptor antagonists SB-408124 and SB-334867 were significantly associated with PCs. One PC was enriched in several pathways, such as oxidative phosphorylation, ribosome, and TGF-beta which were upregulated in AD. DEGs of dual orexin receptor antagonists suvorexant and almorexant were also significantly associated with PCs. One PC was enriched in several pathways, such as ribosome, oxidative, antigen processing and presenting, and oxidative phosphorylation which were upregulated in AD.


Conclusions: We conclude that the tool TransComp-R is effective for use with single-cell RNA sequencing data. From our two different implementations, both regular and sparse principal component analysis perform similarly in identifying biological elements that adequately separate human data by phenotype. Also, we were able to identify pathways in a mouse AD model that accurately reflected human AD biology. Some of these upregulated pathways were further significantly associated with AD drug scores. We found similarities and differences in the pathways upregulated in AD in dual and selective orexin receptor antagonists, revealing differences in their functionality in treating AD.

Discussion: In this study, we developed a new implementation of the cross-species modelling framework TransComp-R for use with snRNAseq microglial data and therapeutic screening in AD. This workflow expands the use of TransComp-R and emphasizes its effectiveness in extracting biological insights from a variety of data types. It can be utilized for different disease states and animal models as a framework for identifying important elements of mouse data which are relevant to therapeutic development. The addition of the use of single-cell data opens doors for the use of other sparse and large-scale genomic datasets.

Acknowledgements (Optional): The authors acknowledge content from BioRender.com used in Figure 1A, Figure 2A-B, and Figure 3.