APOE4 drives transcriptional heterogeneity and maladaptive immunometabolic responses of astrocytes

Abstract

Apolipoprotein E4 (APOE4) is the strongest risk allele associated with the development of late onset Alzheimer's disease (AD). Across the CNS, astrocytes are the predominant expressor of APOE while also being critical mediators of neuroinflammation and cerebral metabolism. APOE4 has been consistently linked with dysfunctional inflammation and metabolic processes, yet insights into the molecular constituents driving these responses remain unclear. Utilizing complementary approaches across humanized APOE mice and isogenic human iPSC astrocytes, we demonstrate that ApoE4 alters the astrocyte immunometabolic response to pro-inflammatory stimuli. Our findings show that ApoE4-expressing astrocytes acquire distinct transcriptional repertoires at single-cell and spatially-resolved domains, which are driven in-part by preferential utilization of the cRel transcription factor. Further, inhibiting cRel translocation in ApoE4 astrocytes abrogates inflammatory-induced glycolytic shifts and in tandem mitigates production of multiple pro-inflammatory cytokines. Altogether, our findings elucidate novel cellular underpinnings by which ApoE4 drives maladaptive immunometabolic responses of astrocytes.

Figure 1.. ApoE4 drives transcriptional heterogeneity in astrocytes.

a 13,130 astrocytes across four pooled samples are visualized via UMAP in the dimensionally reduced dataset, representing 12 subtypes. b Splitting of the aggregated dataset by genotype and treatment reveals prominent shift in subtype proportions across the factorial design. Visually, there was a noticeable shift away from seven As_subtypes (1, 2, 3, 4, 5, 7, 10) found predominantly in the NaCl-treated mice and accumulation of LPS-associated subtypes (0, 6, 8, 9, 11). Notably, there were clear shifts in subtype proportions with an E3-bias for accumulating As_9, while E4 had predominantly more As_8 and As_11 subtypes. c Calculation of the percentage of each As_subtype by genotype and treatment. d k-means clustered (k = 4) dot plot using the top five differentially expressed biomarkers for each of the LPS-enriched astrocyte subtypes. e Cell trajectory inference analysis calculations of astrocyte pseudotime score plotted against selected gene expression values of representative As8/11 subsets demonstrating E4-associated bias (teal and purple). f H&E brightfield image of the Visium spatial profiling tissue sections. g Expression score of Aldoc, a gene enriched in astrocytes, shows robust tissue-wide expression across all sections. h-k Module score expression visualization utilizing the differentially expressed biomarker genesets for LPS-enriched subtypes As_0, As_6, As_9, and As_8/11, respectively.

Figure 2.. ApoE4 is associated with differential immunometabolic signatures of astrocytes.

a Sunburst plot of the MEGENA co-expression network hierarchy of the scRNAseq pseudo-bulk expression data. This hierarchical network visualization of concentric rings represents root parent modules that are increasingly subdivided into child modules, moving outward. Modules are colored according to significant overlap (FDR <0.05, Fishers exact test) with As_subtypes. b MEGENA module (c1_2) displayed significant overlap with the cell-type enrichment of E4-associated LPS subtypes As_8/11, hub genes (triangles) and their associated networked genes (circles) are indicated wherein there was a significant difference in expression between E4 and E3 astrocytes (FDR <0.05). c Integrating these hub genes as an aggregate module score across our spatial datasets demonstrates striking enrichment across multiple anatomical areas in the E4-LPS condition. d Pathway enrichment analyses of E4-enriched As_8/11 subtypes demonstrate significant correlations with multiple ontologies related to immune signaling and cellular activity, with representative pathways highlighted for each ontology. e Geneset defining the representative As_8/11 enriched ontologies integrated back into spatial datasets for ‘Tnfa-signaling via Nfkb’, showing higher expression across the E4-LPS tissue section, comparatively. f KEGG metabolic pathway enrichment analysis via AUCell examining the LPS-associated As_subtypes shows differential alignment across metabolic pathways, with subtypes As_8/11 having marked divergent profiles compared to the E3-LPS enriched As_9 subtype. Glycolysis/Gluconeogenesis has been highlighted (arrow). g Genes for KEGG:Glycolysis_gluconeogenesis were analyzed across spatial dataset, which demonstrate an enrichment in the E4-LPS tissue. h Log2 fold change from DESeq2 pairwise comparison of E4:LPS vs. E3:LPS through spot segmentation of GFAP⁺ pixels (inset; black and white) with spatial transcripts only from segmented spots (inset; red ‘GFAP^hi’ spots). i NicheNet ligand:receptor interaction Circos plots visualizing putative signaling from GFAP^hi segments to GFAP^lo in the LPS-treated tissue split by genotype (E3 left, E4 right).

Figure 3.. ApoE4 exacerbates pro-inflammatory induced glycolytic shift in human iPSC astrocytes.

a Mitochondrial respiration and cellular glycolytic capacity were assessed via Seahorse assays for iPSC-derived human astrocytes isogenically encoding either E3/E3 or E4/E4 alleles with or without pro-inflammatory stimulation. Consistently, harboring E4 alleles drove a main effect alterations in basal respiration (2way ANOVA, F(1, 87) = 117.5, p<0.0001) and glycolytic capacity (2way ANOVA; F(1, 66) = 69.10, p<0.0001, compared to E3. Challenging E4 iPSC astrocytes with pro-inflammatory stimulus significantly altered patterns in basal glycolysis (Tukey’s; p = 0.0003), maximal respiration (Tukey’s; p<0.0001), and compensatory glycolysis (Tukey’s; p<0.0001), compared to E4:NaCl (n=20–22 technical replicates per group). b Plot demonstrating the aggregate energetic mapping of cell states across basal respiration and glycolysis, with visual shift in E4 astrocytes to move further away from respiration along a primarily glycolytic trajectory. c Bulk RNAsequencing from the above four conditions (n=6 per group) plotting the top 5000 most variable features within PCA (69.8% variance explained) showing clear demarcations across genotype (PC1; 42.4%) and treatment (PC2; 17.4%). d Upset plot comparing four conditions’ DESeq2 FDR-corrected (>2logFC) pairwise comparisons. Comparatively, there were 228 unique DEGs found within the E4:Stim vs. E3:Stim pairwise contrast. e Volcano plot highlighting several up- and downregulated DEGs from E4:Stim vs. E3:Stim contrast. f EnrichR pathway ontology analysis of up-regulated DEGs from E4:Stim vs. E3:Stim. g As_8/11 MEGENA-defined hub gene human orthologues analyzed using 2way ANOVA with Šídák’s posthoc comparisons examining E4-bias in saline or stimulated responses (*p<0.05 for each corrected pairwise contrast). h MT1A orthologue quantified from ROSMAP dataset demonstrates E4-biased (E3/E4 and E4/E4) increased expression (Mann-Whitney, p = 0.0375), when disaggregated by sex only female E4+ carriers had a significant increase in expression (Mann-Whitney, p = 0.0189). i Transcription factor (TF) inference analysis via PyScenic data binarization predicts 17 unique TF regulons in the E4:Stim condition relative to all. j Predicted TF of the REL NFkB canonical signaling expressed as Log2FC from E3:NaCl, with both a main effect found due to harboring E4 (2way ANOVA; F(1,20) = 119, p<0.0001) and pairwise comparison relative to E4:Stim (Tukey’s; p<0.0001).

Figure 4.. Pharmacological targeting of REL abrogates ApoE4-associated immunometabolic responses in human iPSC astrocytes.

a/b Inhibition of REL via IT-603 (20μM) had minimal effect on glycolytic response of isogenic E3 astrocytes, but significantly blunted both basal and pro-inflammatory induced exacerbated glycolytic responses of E4 iPSC astrocytes (Šídák’s multiple comparisons posthoc ***p<0.01; n=20–22 technical replicates per group). c Cellular steady-state metabolomics revealed IT-603 significantly altered the pro-inflammatory induced alterations in isogenic E4 iPSC astrocytes relative to E4:NaCl baseline (Šídák’s multiple comparisons posthoc *p<0.05; n=5–6 technical replicates per group), representative factorial plots for cellular lactate and alanine production are shown. d Quantification of extracellular lactate revealed similar trends as cellular metabolites, with IT-603 driving a significant suppression from vehicle-treated cells (Šídák’s multiple comparisons posthoc *p<0.05; n=5–6 technical replicates per group), but there were no differences in extracellular glucose availability. e Multiplex ELISA analysis of exudate cytokine and chemokines revealed consistent upregulation in the majority of soluble analytes, to which IT-603 was sufficient to significantly suppress many of these responses (n= 6/group; Šídák’s multiple comparisons posthoc ***p<0.05). Representative factorial plots for CCL2, CXCL10, IL6, and IFN are shown. f PCA plot (71.6% variance explained) of ELISA results shows a clear shift of IT-603 treated E4 iPSC astrocytes away from vehicle treated pro-inflammatory responses, with IT-603-alone causing minimal alteration in the exudate production. g Correlation matrix on extracellular lactate vs. ELISA analytes in the E4:Stim vehicle samples, demonstrating clustered enrichment correlation with lactate.