Background: Spatial transcriptomics ( ST ) data provide spatially-informed gene expression for studying complex diseases such as Alzheimer's disease ( AD ). Existing studies using ST data to identify genes with spatially-informed differential gene expression ( DGE ) of complex diseases have limited power due to small sample sizes. Conversely, single-nucleus RNA sequencing ( snRNA-seq ) data offer larger sample sizes for studying cell-type specific ( CTS ) DGE but lack spatial information. In this study, we integrated ST and snRNA-seq data to enhance the power of spatially-informed CTS DGE analysis of AD-related phenotypes.
Method: First, we utilized the recently developed deep learning tool CelEry to infer the spatial location of ∼1.5M cells from snRNA-seq data profiled from dorsolateral prefrontal cortex ( DLPFC ) tissue of 436 postmortem brains in the ROS/MAP cohorts. Spatial locations of six cortical layers that have distinct anatomical structures and biological functions were inferred. Second, we conducted cortical-layer specific ( CLS ) and CTS DGE analyses for three quantitative AD-related phenotypes -- β-amyloid, tangle density, and cognitive decline. CLS-CTS DGE analyses were conducted based on linear mixed regression models with pseudo-bulk scRNA-seq data and inferred cortical layer locations.
Results: We identified 450 potential CLS-CTS significant genes with nominal p-values<10 -4 , including 258 for β-amyloid, 122 for tangle density, and 127 for cognitive decline. Majority of these identified genes, including the ones having known associations with AD (e.g., APOE , KCNIP3 , and CTSD ), cannot be detected by traditional CTS DGE analyses without considering spatial information. We also identified 8 genes shared across all three phenotypes, 21 between β-amyloid and tangle density, 10 between cognitive decline and tangle density, and 10 between β-amyloid and cognitive density. Particularly, Gene Set Enrichment Analyses with the CLS-CTS DGE results of microglia in cortical layer-6 of β-amyloid identified 12 significant AD-related pathways.
Conclusion: Incorporating spatial information with snRNA-seq data detected significant genes and pathways for AD-related phenotypes that would not be identified by traditional CTS DGE analyses. These identified CLS-CTS significant genes not only help illustrate the pathogenesis of AD, but also provide potential CLS-CTS targets for developing therapeutics of AD.