APOE modulates microglial immunometabolism in response to age, amyloid pathology, and inflammatory challenge

Abstract

The E4 allele of Apolipoprotein E (APOE) is associated with both metabolic dysfunction and a heightened pro-inflammatory response: two findings that may be intrinsically linked through the concept of immunometabolism. Here, we combined bulk, single-cell, and spatial transcriptomics with cell-specific and spatially resolved metabolic analyses in mice expressing human APOE to systematically address the role of APOE across age, neuroinflammation, and AD pathology. RNA sequencing (RNA-seq) highlighted immunometabolic changes across the APOE4 glial transcriptome, specifically in subsets of metabolically distinct microglia enriched in the E4 brain during aging or following an inflammatory challenge. E4 microglia display increased Hif1α expression and a disrupted tricarboxylic acid (TCA) cycle and are inherently pro-glycolytic, while spatial transcriptomics and mass spectrometry imaging highlight an E4-specific response to amyloid that is characterized by widespread alterations in lipid metabolism. Taken together, our findings emphasize a central role for APOE in regulating microglial immunometabolism and provide valuable, interactive resources for discovery and validation research.

Keywords: APOE; Apolipoprotein E; CP: Neuroscience; DAM; LPS; aging; amyloid; immunometabolism; microglia; scRNA-seq; spatial transcriptomics.

Figure 1.. APOE4 drives immunometabolic changes across the glial transcriptome

(A) Experimental design. Brains from APOE3 and APOE4 mice were analyzed across the lifespan (3, 12, and 24 months of age) and in the presence of an inflammatory challenge (LPS) or AD pathology (amyloid overexpression). (B and C) Number (B) and overlap (C) of differentially expressed genes (DEGs) (p < 0.01) between E3 and E4 brains at each age (bulk RNA-seq). Each circle is a comparison in young (light purple), middle-aged (purple), or aged (dark purple) mice; relative size corresponds to total DEGs. (D) Gene expression of Serpina3n in whole brain. APOE p < 0.001, two-way ANOVA. Error bars denote SEM. (E) The top 10 KEGG pathways most significantly altered by APOE4 in whole-brain tissue. Terms in bold fall under KEGG umbrella pathways of “metabolism” or “immune system.” (F) UMAP showing 24 clusters classified based on canonical gene expression markers. (G) Number of cells per cluster. Bars are colored by individual cluster color from the UMAP in (F). (H) DEGs between E3 and E4 brains within each cell type at each age (scRNA-seq). Young, open bars; middle aged, gray dashed bars; aged, black dashed bars. Dashed lines indicate the number of DEGs in this, as well as two previous, bulk-seq analyses.^, (I) The top 10 KEGG pathways most significantly altered by APOE4 across all cells (scRNA-seq). (J) Venn diagram showing overlap of KEGG pathways differentially expressed between E4 and E3 in the four cell types most affected by APOE4. Numbers represent number of significantly altered pathways in each cell type. The top five overlapping KEGG pathways are listed for each intersection. (K) Heatmap of the top 10 KEGG metabolic pathways altered by APOE in each cell type. Pathways in red show increased expression in E4 cells; blue indicates decreased expression. (B–K) Bulk-seq, n = 3–5 per group; scRNA-seq, 3 biological replicates were pooled together for n = 1 per experimental group. Glycerophos., glycerophospholipid metabolism; Gly,Ser, glycine and serine metabolism; OxPhos, oxidative phosphorylation; PPP, pentose phosphate pathway.

Figure 2.. Age and APOE4 are associated with an increase in “DAM-like” microglia

(A) Gene score plot showing DEGs between E4 versus E3 microglia (y axis) and aged versus young microglia (x axis). Genes labeled in black are common to both DAM/MgND phenotypes. Inset: ridge plot showing DAM/MgND score for each individual microglia as calculated by AUCell. (B and C) E4-specific changes in the microglia transcriptome substantially overlap with AD-relevant gene lists from mouse and human studies. (B) Overlap of published gene lists with DEGs (E4 versus E3) in young (left) and aged (right) microglia (*p < 0.05, hypergeometric distribution test). (C) Expression of select “homeostatic” and DAM/MgND genes in young, middle-aged, and aged microglia. n = 1,422–1,951 cells/group. Error bars denote SEM. (D–F) Aged E4 microglia are enriched for a sub-cluster of cells with a DAM-like expression profile (cluster 6; Mi_6). (D) tSNE (t-distributed stochastic neighbor embedding) of microglia sub-clusters. Top biomarkers for the “homeostatic” clusters (0 and 1) and the “DAM-like” cluster 6 are displayed beneath the cluster labels. (E) Donut charts showing the distribution of aged E3 (left) and aged E4 (right) microglia within each sub-cluster. Clusters labeled in white are enriched in the respective group. (F) Top five Gene Ontology (GO) terms associated with the biomarkers that define Mi_6. (G) SCENIC was used to reconstruct active regulons in each individual microglia and meaningfully cluster cells based on shared activity patterns (binarized). Mi_6 is defined by selective high activity of 16 TFs (red box, “Mi_6 Enriched Regulons”) and the relative absence of activity of other TFs. (H) Ridge plots (top) or tSNE (bottom) showing regulon activity scores for HIF1α (left) and Srebf2 (right).

Figure 3.. APOE4 microglia are metabolically distinct in response to an inflammatory challenge

(A) Experimental design. E3 and E4 mice were injected with lipopolysaccharide (LPS; 5 mg/kg) or saline, and brains were dissected 24 h later for scRNA-seq. (B) Heatmap showing expression of KEGG metabolic pathways in microglia from LPS- or saline-treated mice. (C and D) E3 and E4 brains show enrichment of distinct microglia sub-clusters following LPS treatment. (C) tSNE plot of microglia from LPS- or saline-treated E3 and E4 mice. Colors highlight the 12 microglia sub-clusters. (D) Stacked bar plot showing distribution of experimental groups within each microglia sub-cluster. Top five biomarkers for the two “homeostatic” (0 and 1), E3-enriched (5 and 7), and E4-enriched (8 and 11) clusters are listed below. (E and F) E4 LPS microglia are associated with energy production and OxPhos pathways. (E) Top five GO terms associated with the two E3 LPS-enriched (left, 5 and 7) and two E4 LPS-enriched (right, 8 and 11) clusters. (F) tSNE plots showing higher expression of central carbon (i.e., energy production) pathways in sub-clusters enriched in the E4 LPS brain. Three biological replicates were pooled together for n = 1 per experimental group.

Figure 4.. E4 microglia have increased aerobic glycolysis and higher Hif1α expression

(A) Experimental design. Primary microglia were isolated from E3 and E4 mice and stimulated in vitro with a pro-inflammatory (20 ng/mL IFNγ +50 ng/mL TNF-α) cytokine cocktail prior to Seahorse analysis or targeted metabolomics (both steady-state and stable-isotope-resolved metabolomics). (B and C) Targeted metabolomics on E3 and E4 microglia (n = 21–22 per group). (B) Volcano plot showing changes in steady-state metabolites. (C) Schematic of TCA cycle and glycolysis. Pathways and metabolites associated with pro-inflammatory immunometabolism are highlighted in red, with corresponding bar graphs for E3 and E4 steady-state metabolites overlaid on each. (D) Stable-isotope tracing reveals increased fractional enrichment of fully labeled (m+3) lactate in pro-inflammatory-treated E4 microglia (n = 7–8 per group) after 2 h. (E) Proton efflux rate (PER) (pmol/min/1,000 cells), a measure of glycolysis, measured over time in E3 and E4 microglia during the glycolytic rate assay (Agilent). (F) E4 microglia showed higher basal glycolysis (left) and compensatory glycolysis (right) compared with E3 controls (n = 15–16 per group). (G) ATP production rate (pmol/min/1,000 cells) measured during the ATP rate assay (Agilent) in E3 and E4 microglia, with glycolytic ATP production (GlycoATP) to the left of the y axis and mitochondrial ATP production (MitoATP) to the right (n = 5–9 per group). (H) xy plot with MitoATP displayed on y axis and GlycoATP displayed on x axis. E4 microglia respond to stimulus by dramatically increasing GlycoATP and decreasing MitoATP (red dashed arrow), whereas E3 microglia respond with only a slight increase to GlycoATP and instead show a dramatic increase in MitoATP (blue dashed arrow). (I) Quantitative RT-PCR analysis shows increased Hif1α gene expression in E4 primary microglia (n = 6 per group). Error bars denote SEM. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Two-way ANOVA (D and G), two-tailed t test (F and I), or two-tailed t test adjusted for multiple comparisons (indicated as FDR [false discovery rate]) (B and C).

Figure 5.. Spatial transcriptomics (ST) highlights unique cortical and hippocampal signatures of APOE4, age, and amyloid overexpression

(A and B) ST identifies 17 unique clusters that are anatomically conserved, plus one unique cortical cluster primarily restricted to E4FAD mice (cluster 11, dark green). (A) Spatial transcriptomic plots of brain sections from young, aged, and amyloid-overexpressing E3 and E4 mice. (B) UMAP plot of all 16,979 spots analyzed across all six brains. Clusters were assigned labels based on anatomical concurrence to the Allen Brain Atlas. (C and D) Cluster 11 is enriched in the E4FAD brain and consists of genes related to lipid metabolism and microglial activation. (C) E3FAD and E4FAD brains showing spots belonging to cluster 11. Cluster 11 biomarker genes were re-plotted to scRNA-seq data, showing highest expression in microglia, specifically in Mi_6. (D) Top 10 Gene Ontology terms for cluster 11, highlighting pathways of lipid metabolism and immune activation. (E) Number of spots within each cluster for each experimental group. Clusters are organized by respective brain regions. (F–H) E4 drives gene expression changes primarily in the cortex and hippocampus. (F) DEGs between E4 and E3 brains within each brain region. (G and H) DEGs within the cortex (G) and hippocampus (H) of the 5XFAD mice. Genes labeled in black correspond to DAM/MgND genes. (I) ST plots showing DAM/MgND scores for each spot (calculated with AUCell). n = 1 brain per group.

Figure 6.. APOE4 exacerbates plaque-induced microglial activation and alterations in lipid metabolism

(A) E4FAD brain stained with P2ry12 (green; microglia), GFAP (red; astrocytes), and X-34 (blue) to demarcate amyloid plaques (left). X-34 intensity was quantified to generate a “plaque intensity score” for each individual spatial transcriptomic spot (right). (B) Gene correlation with plaque intensity in E3FAD (blue, left) and E4FAD (red, right) brains. y axis values represent correlation coefficients, with genes at the top of the graph positively correlated with plaque intensity, and genes at the bottom negatively correlated. Distance from center on the x axis represents significance of the correlation (−log10(p adjusted)). DAM/MgND genes are noted in gray. (C) Top five Gene Ontology terms for genes that were positively (left) or negatively (right) correlated with plaque intensity. Some GO terms were uniquely correlated with E4 (red), some uniquely correlated with E3 (blue), and some correlated with plaque intensity regardless of APOE genotype (purple). Venn diagrams show overlap between genes correlated with plaque intensity in E4FAD (red circles) or E3FAD (blue circles) brains. (D–G) Gene networks associated with plaque intensity. (D) The correlation between module eigengenes (MEs) and amyloid plaque intensity. Values in the heatmap are Pearson’s correlation coefficients, and asterisks represent significant correlations: *p < 0.05; ***p < 0.001. Modules with positive values (red) indicate positive correlation of MEs with plaque intensity, modules with negative values (green) represent a negative correlation. (E) Network plots of the top 10 genes with the highest intramodular connectivity (hub genes) in the magenta (top) and red (bottom) modules. (F) UMAP plots map expression of module gene lists (sum) back to the scRNA-seq dataset. (G) Top five Gene Ontology terms associated with the magenta or red modules. (H) Venn diagrams showing overlap of red and magenta modules with oligodendrocyte (OLIG) and plaque-induced gene (PIG) lists from Chen et al. Overlapping genes are listed. (I and J) The E4FAD brain has a high PIG score and the lowest OLIG score. (I) Ridge plots showing PIG (left) and OLIG (right) scores for each experimental group. (J) Spatial expression of PIG (left) and OLIG (right) gene lists.

Figure 7.. Matrix-assisted laser desorption ionization (MALDI) mass spectrometry imaging (MSI) reveals APOE- and region-specific changes in multiple lipid species

(A and B) Expression of glycerophospholipid pathway genes increases with age in whole-brain tissue (A, ‘bulk’) and is highest in aged E4 microglia (B, ‘scRNA-seq’). (C) Experimental workflow for MALDI MSI. (D) Volcano plot of targeted lipid species highlights changes in select phosphatidylcholine, sphingomyelin, ceramide, and triacylglycerol. (E) Heatmap of quantified lipid species (average values across all regions) shows clear clustering by age and amyloid expression, with distinct separation of E3 5XFAD and E4 5XFAD brains. Brackets include multiple possible fatty acid chain lengths and/or double-bond positions. (F) Principal-component analysis (PCA) plot of MALDI MSI-detected lipids shows clear separation of E3 5XFAD and E4 5XFAD brains. (G) Regional intensity of an example lipid from (E) (phosphatidylcholine (16:0/18:2)). (Top) Scans show spatial distribution of lipid across coronal brain sections. (Bottom) Average pixel intensity across each brain region for PC(16:0/18:2). n = 3 per group. Error bars denote SEM. *p < 0.05, **p < 0.01, multiple comparisons ANOVA. Regional data for all scanned lipids can be found in Table S2.