APOE4/4 is linked to damaging lipid droplets in Alzheimer's microglia

Abstract

Several genetic risk factors for Alzheimer's Disease (AD) implicate genes involved in lipid metabolism and many of these lipid genes are highly expressed in glial cells. However, the relationship between lipid metabolism in glia and AD pathology remains poorly understood. Through single-nucleus RNA-sequencing of AD brain tissue, we have identified a microglial state defined by the expression of the lipid droplet (LD) associated enzyme ACSL1 with ACSL1-positive microglia most abundant in AD patients with the APOE4/4 genotype. In human iPSC-derived microglia (iMG) fibrillar Aβ (fAβ) induces ACSL1 expression, triglyceride synthesis, and LD accumulation in an APOE-dependent manner. Additionally, conditioned media from LD-containing microglia leads to Tau phosphorylation and neurotoxicity in an APOE-dependent manner. Our findings suggest a link between genetic risk factors for AD with microglial LD accumulation and neurotoxic microglial-derived factors, potentially providing novel therapeutic strategies for AD.

Extended Data Figure 1

a, Doublet score distributions obtained with the Scrublet (0.2.3) Python package. Vertical lines indicate the doublet score thresholds identified per sample based on the distributions shown. b-e, Distribution of total read counts (a), number of genes expressed per cell (b), percent of reads mapped to ribosomal genes (c), and percent of reads mapped to mitochondrial genes (d) within the processed data after quality control. f, Percent of explained variance across the first 48 principal components.

Extended Data Figure 2

a-c, UMAP visualization of the whole snRNA-seq dataset after quality control and batch correction (n=100,317). Cells are colored by cell type annotation (a), subclusters drawn for cell type annotation (b), subject groups (control, AD-APOE3/3, AD-APOE4/4) (c). d, Bar chart indicating the total number of cells per cell type. e-j, Violin plots indicating gene expression levels of marker genes used for cell type annotations across the 6 identified cell types. k-p, UMAP visualization of the whole snRNA-seq dataset colored by cell type marker gene expression levels per cell.

Extended Data Figure 3

a, Volcano plot representing pseudobulk differential gene expression results (see Methods, Microglia pseudobulk differential gene expression) of microglia from control subjects compared to microglia (left) from subjects with AD and the APOE3/3 genotype, (right) from subjects with AD and the APOE4/4 genotype. Select lipid and metabolism-associated genes highlighted in red. b, Volcano plot representing single-cell differential gene expression results of microglia from subjects with AD and the APOE3/3 genotype compared to microglia from subjects with AD and the APOE4/4 genotype. Select lipid and metabolism-associated genes highlighted in red. c, Select KEGG pathway analysis terms and enrichment score for top 200 differential expressed genes in between control and AD-APOE4/4 microglia (top) and control and AD-APOE3/3 microglia (bottom). d-g, UMAP visualization of the microglia colored by identified subclusters (macrophage cluster 10 is not shown) (d), identified microglial states (e), subject IDs (f), and subject groups (control, AD-APOE3/3, AD-APOE4/4). h, Top marker genes identified for the 3 microglial states (HOMEOSTATIC, DAM, LDAM) with ‘one vs. rest’ marker identification. Single-cell differential gene expression was performed with MAST (see Methods, Single-cell differential expression). Heatmap indicates significant (adj. p-value<0.05) log2-fold changes. i, Normalized and z-scored gene expression levels of HOMEOSTATIC, DAM, LDAM marker genes as well as marker genes identified by Olah et al. (2020) across the 11 subclusters identified within the microglia. j, UMAP representation of microglia cells indicating signature scores per cell for HOMEOSTATIC, DAM, LDAM marker genes as well as marker genes identified by Olah et al. (2020). Contour lines indicate kernel density estimates of the signatures across the UMAP space.

Extended Data Figure 4

a, Quantification of Oil Red O counts for each subject. b-c, Additional Oil Red O staining of APOE4/4 AD subjects with IHC staining for Aβ. White arrowheads represent Oil Red O positive cells in or around Aβ plaques. Scale bars (black, bottom right) 50 μm. d, Representative immunofluorescence images of human frontal cortex adjacent to tissue used in snRNA-seq experiments stained for microglia marker IBA1 (green), ACSL1 (red), and DAPI (blue), and amyloid-beta (magenta) in an AD APOE4/4 subject. Scale bars (white, bottom right) 20 μm.

Extended Data Figure 5

a, Expression of microglia marker genes in IMG. b, Toxicity of fAβ at different concentrations in APOE4/4 and APOE3/3 iMG after 24 hour incubation followed by staining with CytoxRed. Each dot represents a replicate well, error bars represent s.e.m, p-value calculated by ANOVA. c, Dose-dependent effect of fAβ on lipid accumulation. Each dot represents a replicate well, error bars represent s.e.m, p-value calculated by ANOVA. d, Representative image of human macrophages (top) untreated (left) or treated (right) with 5 μM Aβ for 24 hours (left) with PLIN2 (green) and Aβ (red) staining. Scale bars (white, bottom right) 50 um. Representative image of mouse BV2 cells (bottom) untreated (left) or treated (right) with 5uM Aβ for 24 hours (left) with LipidSpot (green) and Aβ (red) staining. Scale bars (white, bottom right) 75 μm. e, LysoTracker area per cell after incubation with fAβ in APOE3/3 and APOE4/4 iMG. Each dot represents a replicate well, error bars represent s.e.m, p-value calculated by ANOVA. f, Phrodo zymosan phagocytosis area per cell after incubation with fAβ in APOE3/3 and APOE4/4 iMG. Each dot represents a replicate well, error bars represent s.e.m, p-value calculated by ANOVA. g-h, Transmission electron microscopy of APOE3/3 iMG and APOE 4/4 iMG treated with 5 μM Aβ for 24 hours and untreated iMG. Scale bars (white, bottom right) 2 μm. Each dot represents images of individual cells, error bars represent s.e.m, p-value calculated by ANOVA. i-j, Representative image of APOE4/4 iMG treated with 5 μM Aβ for 24 hours (left) with Plin2 and ACSL1 staining quantification with (i) with representative images (j) (n=4 replicate wells, error bars represent s.e.m, p-value calculated by ANOVA, Scale bars (white, bottom right) 50 μm. k, ACSL1 gene expression measured in human iMG after LPS treatment as described in Hasselmann et al. 2019 (https://rnaseq.mind.uci.edu/blurton-jones/bulkSeq/)

Extended Data Figure 6

a, Volcano plot of differential gene expression analysis of untreated and Aβ treated APOE4/4 iMG. b, Volcano plot of differential gene expression analysis of lipid droplet high and lipid droplet low APOE4/4 iMG. c, Volcano plot of differential gene expression analysis of lipid droplet high and lipid droplet low APOE3/3 iMG. d, Volcano plot of differential gene expression analysis of lipid droplet high APOE3/3 iMG and lipid droplet high APOE4/4 iMG. e, Example gating scheme for separating cells based on lipid droplet content.

Extended data Figure 7

a, Schematic of CRISPR KO screen in APOE4/4 iMG for lipid droplet formation following Aβ treatment. b, Volcano plot representing CRISPR screen results. Effect score represents log₂ fold change in sgRNA counts in lipid droplet negative versus lipid droplet positive cell fraction. Screen hits with P value <0.005 colored blue. c, Live cell imaging of untreated APOE4/4 iMG, 5 μM Aβ treated iMG, and 5 μM Aβ treated iMG with 10 μM GNE-317. The y axis represents average green fluorescence per cell normalized to untreated APOE 4/4 iMG at the first time point and the x-axis represents imaging time points in hours (n=3 replicate wells per condition, error bars represent s.e.m.) (left). Representative images at the final time point (right) with LipidSpot signal represented in green. d, Quantification of PLIN2 staining in untreated, Aβ treated, and Aβ treated with 10 uM GNE-317 conditions iMG for 24 hours. (n=3 replicate wells, error bars represent s.e.m, p-value calculated by ANOVA). e, Quantification of lysotracker staining in untreated, Aβ treated, and Aβ treated with 10 μM GNE-317 conditions iMG for 24 hours. (n=3 replicate wells, error bars represent s.e.m, p-value calculated by ANOVA). f, Measurement of secreted chemokines in cell culture media in untreated, Aβ treated, and Aβ treated with 10 μM GNE-317 conditions iMG for 24 hours. Individual dots represent replicate wells (n=2, error bars represent s.e.m, p-value calculated by ANOVA). g, Select KEEG pathway enrichment terms for the top 200 significant downregulated genes ranked by p-value upon GNE-317 treatment in APOE4/4 iMG when challenged with Aβ. h, Normalized gene expression counts for significantly downregulated genes with GNE-317 treatment in APOE4/4 iMG when challenged with Aβ APOE4/4 iMG (n=3 replicate wells, error bars represent s.e.m., *** P < 0.0001). P-values determined by DEseq2. i, Select KEEG pathway enrichment terms for top 200 significant upregulated genes ranked by p-value upon GNE-317 treatment in APOE4/4 iMG when challenged with Aβ. j, Normalized gene expression counts for significantly upregulated genes with GNE-317 treatment in APOE4/4 iMG when challenged with Aβ APOE4/4 iMG (n= 3 replicate wells, error bars represent s.e.m., *** P < 0.0001). P-values determined by DEseq2. k, Differential gene expression for genes in mTOR and autophagy pathways upon GNE-317 treatment with fAβ challenge in iMG. (n=3 replicate wells, error bars represent s.e.m., *** P < 0.0001). P-values determined by DEseq2. l, Representative images of LC3B staining (yellow) upon fAβ and GNE-317 treatment. The scale bar (white, bottom left) represents 20μm. m, Quantification of LC3B staining by integrated fluorescence intensity per DAPI signal. Individual dots represent replicate wells (n=4, error bars represent s.e.m, p-value calculated by unpaired, two-sided, t-test). n, Toxicity measurements of GNE-317 as determined by cytox-red. (n=4, error bars represent s.e.m, p-value calculated by ANOVA).

Extended data Figure 8.. Detection of TG synthesized in microglia taken up by neurons through labeled ¹³C-glucose tracing.

a, Measurement of ¹³C-labeled triglycerides synthesized in microglia and profiled in neurons by lipidomics after exposure to microglia conditioned media. Microglia grown in uniformly labeled ¹³C-glucose (U-13C6-glucose) were challenged with fAβ or untreated. Each dot represents an individual replicate. n=3, error bars represent s.e.m, p-value calculated by unpaired t-test.

Figure 1.. AD microglia have novel lipid transcriptional state defined by ACSL1

a, Schematic of single-nucleus RNA-seq cohort and workflow (see Methods). b, UMAP representation of all cells (n=100,317) from snRNA-seq, colored by annotated cell type. Data is shown after quality control and batch correction. c,d, Volcano plot representing MAST-based single-cell differential gene expression results (see Methods, Single-cell differential gene expression) of microglia from control subjects compared to microglia from subjects with AD and the APOE3/3 genotype (c) and from subjects with AD and the APOE4/4 genotype (d). Select lipid and metabolism-associated genes highlighted in red. e, Pathway diagram showing placement of differentially expressed gene ACSL1 in pathway starting from free fatty acid to lipid droplet formation. f, Violin plots showing ACSL1 expression across the cell types within the snRNA-seq dataset. Significance results indicate MAST-based adjusted p-values (see Methods, Single-cell differential gene expression). g, Normalized and z-scored gene expression levels of HOMEOSTATIC, DAM, and LDAM marker genes across the 11 subclusters identified within the microglia (top). HOMEOSTATIC, DAM, and LDAM signature scores are shown across the 11 identified subclusters. h, UMAP representation of microglia cells indicating the marker gene-based cell state annotation (top left), and the signature scores per cell for HOMEOSTATIC (top right), DAM (bottom left), and LDAM (bottom right) states. Contour lines indicate kernel density estimates of the signatures across the UMAP space. i, Bar plots indicating the percentage of cells from the three different cellular states (HOMEOSTATIC, DAM, and LDAM) across microglia from control, AD-APOE3/3, and AD-APOE4/4 groups. Chi-square test results indicate the significance of the percentage differences between the groups (*** indicates P<.0001). j, Representative immunofluorescence images of human frontal cortex adjacent to the tissue used in snRNA-seq experiments stained for microglia marker IBA1 (green), ACSL1 (red), and DAPI (blue) in an aged-matched healthy control subject (left), an AD APOE3/3 subject (middle), and an AD APOE4/4 subject. Scale bars (white, bottom right) 20 μm. k, Quantification of percentage of IBA1+ microglia positive for ACSL1. Each dot represents an average quantification for an individual subject. P-value determined by one-way ANOVA, error bar represents s.e.m.

Figure 2.. Intracellular lipid accumulation is linked to AD pathology.

a, Representative Oil Red O staining image for control, AD APOE3/3, and AD APOE4/4 human frontal cortex. Neutral lipids stained with Oil Red O (red) and nuclei stained with hematoxylin (blue). Scale bars (black, bottom right) 50 μm. b, Quantification of Oil Red O staining. Bar plots represent average Oil Red O counts per image for each subject category. Each dot represents average Oil Red O counts for an individual subject averaged over five 20x image fields per individual. P-value determined by one-way ANOVA, error bar represents s.e.m. c, Oil Red O staining of APOE4/4 AD subjects with IHC staining for Aβ. White arrow heads represent Oil Red O positive cells in or around Aβ plaques. Scale bars (black, bottom right) 50 um (left). High magnification of representative Oil Red O stain with IHC staining for Aβ in AD APOE4/4 subject. White arrowheads represent Oil Red O positive cells in or around Aβ plaques. Scale bars (black, bottom right) 20 μm (right). d, Quantification of the frequency of Oil Red O positive cells in various distances from Aβ plaques. Each dot represents an individual subject. P-value determined by one-way ANOVA, error bar represents s.e.m. e, Scatter plot of average Oil Red O counts per subject averaged over five 20x image fields per individual with individual subject’s meta-data. Subject category colored blue for control, orange for AD APOE3/3 subjects, and red for AD APOE4/4 subjects. P-values determined by spearmen correlation. f, Scatter plot of average Oil Red O counts per subject averaged over five 20x image fields per individual with individual subject’s snRNA-seq data. Subject category colored blue for control, orange for AD APOE 3/3 subjects and red for AD APOE4/4 subjects. P-values determined by Spearmen correlation. g, Representative immunofluorescence images of mouse hippocampus tissue stained for microglia marker IBA1 (red), neutral lipids (LipidSpot: green), and DAPI (blue) in control age-matched non-transgenic mice (left), AD mouse model (J20) with human APOE3 knock-in (middle), and AD mouse model (J20) with human APOE4 knock-in (right). Scale bars (white, bottom right) 20 μm. h, Quantification of average percent of IBA1 positive microglia with neutral lipid dye (LipidSpot). Each dot represents individual biological replicate. P-value determined by one-way ANOVA, error bar represents s.e.m.

Figure 3.. iMG upregulate ACSL1 and increase TG lipid synthesis upon fAβ challenge

a, Schematic of isogenic APOE3/3 and APOE4/4 iPSCs and differentiation into iPSC-derived microglia (iMG). b, Live cell imaging of untreated APOE3/3 and APOE4/4 iMG and fAβ treated APOE3/3 and APOE4/4 iMG stained with a green lipid fluorescent dye (Lipidspot). The y axis represents mean green fluorescence per cell normalized to untreated APOE3/3 iMG at the first time point and the x-axis represents imaging time points in hours (n=3 replicate wells per condition, error bars represent s.e.m.). c, Average Lipidspot green fluorescence per cell normalized to untreated of final timepoint in b. Individual dots represent replicate wells (n=3, error bars represent s.e.m, p-value calculated by unpaired, two-sided t-test). d, Primary rat microglia cultured under serum-free conditions untreated (left) or treated with fAβ (red) and stained with a green lipid fluorescent dye (Lipidspot). Scale bar 200 μm. e, Average Lipidspot green fluorescence per cell normalized to untreated images in d. Individual dots represent replicate wells (n=3, error bars represent s.e.m, p-value calculated by unpaired, two-sided, t-test). f, Representative CARS images of APOE4/4 and APOE3/3 iMG treated with fAβ, Scale bars (white, bottom right) 20 μm. g, Quantification of lipid droplet measurements from CARS microscopy of APO4/4 and APOE iMG treated with fAβ. Each dot represents lipid measurements from individual cell measurements. P-value calculated by unpaired t-test. h, Example CARS spectra from fAβ treated iMG (red) and reference spectra for common lipid species (black). i, Schematic of lipidomics measurement of C13-D-Glucose incorporation into TG over time in BV2 cells treated with 5uM fAβ. n=3, error bars represent s.e.m, p-value calculated by on-way ANOVA. j, Incorporation of C13-D-Glucose into triglycerides in microglia after fAβ treatment or untreated cells compared with cells without C13-D-Glucose as control. Each dot represents an individual replicate. n=3, error bars represent s.e.m, p-value calculated by one-way ANOVA. k, Incorporation of C13-D-Glucose into triglycerides in microglia after fAβ treatment over time compared with cells without C13-D-Glucose as control. Each dot represents an individual replicate. n=3, error bars represent s.e.m, p-value calculated by one-way ANOVA. l, Normalized gene expression counts for significant differentially expressed genes between untreated and fAβ treated APOE4/4 iMG (n= 3 replicate wells, error bars represent s.e.m, P-values determined by DEseq2. m, Volcano plot of Genome-wide CRISPR KO screen for BODIPY high and BODIPY low cells using the human monocyte U937 cell line. The confidence score and effect score are determined by CasTLE. Genes passing a 10% FDR cutoff are highlighted in red and blue. n, Average Lipidspot green fluorescence per cell normalized to untreated cells with 5uM fAβ, and 5uM fAβ with 1uM ACSL1 inhibitor (Triacin C). Individual dots represent replicate wells (n=4, error bars represent s.e.m, p-value calculated by unpaired, two-sided t-test). o, Schematic of ATAC-seq and RNA-seq experiments to characterize lipid droplet high and lipid droplet low iMGs. p, Genome-wide comparison of open ATAC-seq enhancer peaks comparing LD low and LD high iMG (left). De novo Motif analysis of differential peaks using HOMER. Motifs enriched in lipid-associated macrophages are highlighted in red. q, Normalized gene expression counts for significant differentially expressed genes between lipid droplet high from lipid droplet low APOE4/4 iMG (n= 3 replicate wells, error bars represent s.e.m, P-values determined by DEseq2, *** P < 0.0001). r, Measurement of secreted chemokines in cell culture media after FACS separation of lipid droplet positive from lipid droplet negative APOE4/4 iMG following 18-hour Aβ treatment. Individual dots represent replicate wells (n=2, error bars represent s.e.m, p-value calculated by unpaired, two-sided t-test). s, Average percent of pHrodo zymosan red positive of total cell population with or without lipid droplets (Lipidspot) after 18-hour Aβ treatment. Individual dots represent replicate wells (n=3, error bars represent s.e.m, p-value calculated by unpaired, two-sided t-test). t, Average of percent of lysotracker red positive of total cell population with or without lipid droplets (Lipidspot) after 18-hour Aβ treatment. Individual dots represent replicate wells (n=3, error bars represent s.e.m, p-value calculated by unpaired, two-sided t-test).

Figure 4.. LD-containing microglia induce Tau phosphorylation and apoptosis in neurons

a, Schematic of LDAM-specific conditioned media exposure to neurons. b, Representative immunofluorescence images of iPSC-derived neurons exposed to no conditioned media (left), lipid droplet high APOE4/4 iMG conditioned media (middle), and lipid droplet low APOE 4/4 iMG conditioned media (left). Cells were stained for DAPI, MAP2 (grey), and phosphorylated Tau (AT8: green). Scale bars (white, bottom right) represent 20 μm. c, Representative immunofluorescence images of iPSC-derived neurons exposed to lipid droplet high APOE4/4 iMG conditioned media (left), lipid droplet high APOE3/3 iMG conditioned media (middle), lipid droplet high APOE KO iMG conditioned media (right). Cells were stained for DAPI, MAP2 (grey), and phosphorylated Tau (AT8: green). Scale bars (white, bottom right) represent 20 μm. d, Quantification of images as presented in c by measuring the AT8 fluorescence area normalized to the MAP2 florescence area. Each dot represents a random filed image (N=18) across 3 replicate wells, error bars represent s.e.m, p-value calculated by ANOVA. e, Representative immunofluorescence images of iPSC-derived neurons exposed to lipid droplet high APOE4/4 iMG conditioned media (left), lipid droplet high APOE3/3 iMG conditioned media (middle), lipid droplet high APOE KO iMG conditioned media (right). Cells were stained for DAPI (blue), MAP2 (grey), and cleaved Caspase-3 (red). f, Quantification of images as presented in e by counting cleaved caspase-3 fluorescence regions normalized to DAPI. Each dot represents a random filed image (N=18) across 3 replicate wells, error bars represent s.e.m, p-value calculated by ANOVA. g, Representative immunofluorescence images of neurons exposed to lipid droplet low APOE4/4 iMG conditioned media (left) and lipid droplet high APOE4/4 iMG conditioned media (right). Cells were stained for neutral lipids (LipidSpot: green) and activated caspase-3 dye (red). h, Quantification of average LipidSpot green fluorescence area cell per well normalized to the lipid droplet low iMG conditioned media treated neurons. Individual dots represent replicate wells (n=3, error bars represent s.e.m, p-value calculated by unpaired, two-sided t-test). i, Schematic of lipidomics experimental design of neurons treated with conditioned media. j, Volcano plot representing lipids detected in neurons after treatment of lipid droplet high iMG conditioned media compared to lipid droplet low iMG conditioned media. TG species are highlighted in red. k, Representation of lipidomic measurements from one lipid species detected in lipidomic analysis. Individual dots represent replicate wells (n=3, error bars represent s.e.m, p-value calculated by one-way ANOVA. l, Schematic of the proposed role of lipid droplet accumulating microglia in neurodegeneration.