A factor-based analysis of individual human microglia uncovers regulators of an Alzheimer-related transcriptional signature

Abstract

Human microglial heterogeneity is only beginning to be appreciated at the molecular level. Here, we present a large, single-cell atlas of expression signatures from 441,088 live microglia broadly sampled across a diverse set of brain regions and neurodegenerative and neuroinflammatory diseases obtained from 161 donors sampled at autopsy or during a neurosurgical procedure. Using single-cell hierarchical Poisson factorization (scHPF), we derived a 23-factor model for continuous gene expression signatures across microglia which capture specific biological processes (e.g., metabolism, phagocytosis, antigen presentation, inflammatory signaling, disease-associated states). Using external datasets, we evaluated the aspects of microglial phenotypes that are encapsulated in various in vitro and in vivo microglia models and identified and replicated the role of two factors in human postmortem tissue of Alzheimer's disease (AD). Further, we derived a complex network of transcriptional regulators for all factors, including regulators of an AD-related factor enriched for the mouse disease-associated microglia 2 (DAM2) signature: ARID5B, CEBPA, MITF, and PPARG. We replicated the role of these four regulators in the AD-related factor and then designed a multiplexed MERFISH panel to assess our microglial factors using spatial transcriptomics. We find that, unlike cells with high expression of the interferon-response factor, cells with high expression of the AD DAM2-like factor are widely distributed in neocortical tissue. We thus propose a novel analytic framework that provides a taxonomic approach for microglia that is more biologically interpretable and use it to uncover new therapeutic targets for AD.

Figure 1.. A 23-factor model characterizing microglia function in a broad single-cell dataset.

(A) Summary of collected tissues. Abbreviations: BCN, body of caudate nucleus; CB, cerebellum; DN, dentate nucleus of the cerebellum; DNET, dysembryoplastic neuroepithelial tumor; OC, occipital cortex; SBWM, subcortical white matter; STN, subthalamic nucleus. (B) Density maps showing the UMAP space and the distribution of the 23 microglial factors at the single-cell level. Lighter color indicates a relatively greater density of cells with higher factor scores. Note, the densities are not directly comparable across factors. Showing the top five genes per factor and overarching categories.

Figure 2.. The 23-factor model offers a generalizable reference atlas of microglial functions and cellular processes, allowing for the projection of external data.

(A) Overview of the external datasets projected onto the model for validation (see Table S5). (B, C) Projection of 65,179 myeloid transcriptomes isolated from glioblastoma slice cultures within the reference UMAP space (light gray background). Cells represented as binned, hexagonal ‘meta-cells’, colored by the number of cells within each bin. Showing examples of two treatment conditions (out of 14), dark gray meta-cells show control myeloid cells from the same donors. (C) Barplot comparing proportion of myeloid cells classified as low vs. high-expressing for topotecan across selected factors with significant perturbation compared to untreated cells. (D-G) Projection of 26,044 WT and 5xFAD-MITRG-xenografted human microglia shows upregulation of the GPNMB^High (scHPF_26) factor and downregulation of IFN-I response (scHPF_20) in 5X-xMGs (F, G), capturing signatures observed at the individual gene level. (H-K) Projection of 41,655 iPSC-derived iMGs transcriptomes (Dolan et al., 2023) shows upregulation of two microglial subpopulations highly expressing IFN-I response (scHPF_20) and chemokine (scHPF_5) factors in AN-exposed iPSC-derived iMGs compared to untreated iMGs. (L, M) Projection of 36,473 HMC3 cells treated with various compounds. (M) Expression factor differences between treated and untreated/DMSO-treated HMC3 cells.

Figure 3.. AD-derived microglia upregulate the disease-associated GPNMB^high factor.

(A) UMAP embedding of microglial nuclei (n=84,651) from DLPFC snRNA-seq (ROS-MAP, n=424 donors) compared to the reference model (gray background). Cells represented as binned, hexagonal ‘meta-cells’ colored by the number of nuclei per bin (nbins=100). (B) Linear regression results for donor-level factor scores (mean-aggregated across cells) adjusted for study, age at death, sex, and postmortem interval. Includes a random-effects meta-analysis across the two datasets (see Table S7 and Table S8). Significance levels: *FDR p < 0.05. (C) Forest plots for the meta-analysis of GPNMB^high (scHPF_26) across all three ROS-MAP cohorts in panel B.

Figure 4.. Microglial regulatory network identifies putative GPNMB^High (scHPF_26) factor regulators.

(A) Heatmap showing hypergeometric enrichment of the top-100 loaded genes for each factor across ARACNe regulons (see Table S9). Grey indicates too few overlapping genes for testing. (B) Heatmap showing FGSEA enrichment of DEGs associated with AD-related traits in bulk RNAseq and their top enriched ARACNe regulons, in both the single-cell (sc) and single-nucleus (sn; see Table S11) ARACNE networks (see Table S12). Showing Bonferroni-corrected p-values across target regulons and traits. (C) Reduced GO enrichment results for upregulated regulon targets. Biological processes adjusted FDR p-value < 0.05 and q-value < 0.05. (D) Donor-level regulator associations with factors, adjusted for donor age, donor sex, PMI and sequencing depth. (E) Venn diagram showing predicted target overlap across TFs in the sc-ARACNE network. (F) Variance partitioning showing the unique total contribution of the four TF adjusted (X1∣X2) for covariates (age, sex, PMI, study), as well as individual unique contribution of each TF adjusted (X1∣X2) for covariates and the other three TFs. Asterisks denotes FDR-corrected p-value for the number of regulators from permutation test for redundancy analysis. Significance levels: *** p < 0.001, ** p < 0.01, * p < 0.05.

Figure 5.. In situ, IR^High microglia show spatially-distinct neighborhoods characterized by high IRF7 expression, and co-upregulation of IFN-I response (scHPF_20) genes in astrocytes.

(A) Overview of the experimental design for MERSCOPE (see Table S13 for gene list). (B) UMAP showing the major cell types identified. (C) In situ distribution of IFN-I response (scHPF20) high (i.e., IR^High) vs. low (i.e., IR^Low) cellular niches (see Table S14 and Table S15). (D) In situ association results of GPNMB^High (scHPF_26) enriched TFs (ARID5B, CEBPA, MITF, PPARG) with their ARACNe-predicted targets across tissue layer niches (see Table S14 and Table S15). (E) Regions of interest showing IR^High neighborhoods and the expression of IRF7 ARACNe-predicted targets, MX1 and OAS3. (F) Barplot showing the proportion of IRF7+/MX1+/OAS3+ microglia across IR^High vs. IR^Low microglia. (G) Volcano plot showing DEGs in astrocytes in IR^High versus IR^Low neighborhoods (see Table S16). (H) Regions of interest showing IR^High niches and the expression of IRF7, MX1, OAS3, IFI6 and IFIT3.