Microglia transcriptional heterogeneity across human and experimental models

Abstract

AbstractBackgroundThe molecular data from multiple brain regions, cohorts, experimental models are seldom integrated comprehensively, although it provides a unique opportunity to study the cellular heterogeneity and molecular changes across AD etiology and progression. We have developed a new reference model to bridge across datasets.MethodWe leveraged single‐nuclei transcriptomic profiles (snRNA‐seq) of human microglia (parietal cortex from 67 donors from the Knight ADRC and DIAN brain banks; N = 15,726; Dataset1) [1], iPSC‐MLC CRISPRi/a‐based genetic screens in human iPSC‐derived microglia (N = 15,350; Dataset2) [2], and WT and R47H TREM2 xenografted microglia (xMGs), isolated from chimeric AD mice (N = 25,407; Dataset3) [3] to build a comprehensive atlas of microglia heterogeneity, that captures transcriptional states fairly well represented in each dataset. We employed Seurat v4 [4] to integrate the data, and scArches [5] to build classifiers.ResultWe identified homeostatic, activated, interferon (IFN), and IL1B clusters across datasets present in all data we had integrated. In addition, we identified MHCII. We observed that some of the clusters of original states were combined into single clusters. For example, CXCL10‐IFN and INF. We also observed that the integrated data shows a more distinctive expression profile of Type I+II IFN. Furthermore, we build a multiclassifier to label novel microglia generated from hundreds of samples, and data from various experimental models. The cross‐validation analyses indicate that the overall accuracy of the classifier (scArchers) is 0.85 (Fig. 1), although it changes for the distinct datasets [5]. For human microglia in Dataset1 the accuracy was 0.79, while for iPSC‐MLC was 0.73 and 0.96 for Dataset 3. Our results indicate that this multi‐model reference can capture subtle differences better that study‐centric analyses of the data.ConclusionWe are repurposing microglia transcriptomic single‐cell data to build a cross‐species and experimental iPSC‐MLC to bridge across cohorts, species, and experiments. Our results, indicate that this new resource allows to simultaneously interrogate gene expression across different experiments and transcriptional states, and providing the foundation to integrate additional single‐cell molecular data from model organism, ATAC‐seq, and spatially resolved transcriptomics.