Alzheimer's Patient Microglia Exhibit Enhanced Aging and Unique Transcriptional Activation

Abstract

Damage-associated microglia (DAM) profiles observed in Alzheimer's disease (AD)-related mouse models reflect an activation state that could modulate AD risk or progression. To learn whether human AD microglia (HAM) display a similar profile, we develop a method for purifying cell types from frozen cerebrocortical tissues for RNA-seq analysis, allowing better transcriptome coverage than typical single-nucleus RNA-seq approaches. The HAM profile we observe bears little resemblance to the DAM profile. Instead, HAM display an enhanced human aging profile, in addition to other disease-related changes such as APOE upregulation. Analyses of whole-tissue RNA-seq and single-cell/nucleus RNA-seq datasets corroborate our findings and suggest that the lack of DAM response in human microglia occurs specifically in AD tissues, not other neurodegenerative settings. These results, which can be browsed at http://research-pub.gene.com/BrainMyeloidLandscape, provide a genome-wide picture of microglial activation in human AD and highlight considerable differences between mouse models and human disease.

Keywords: Alzheimer’s disease; aging; microglia; neurodegenerative diseases; neuroinflammation; transcriptomics.

Figure 1.. Expression Profiling of Human Cell Populations Sorted from Frozen, Post-Mortem SFG

(A) Experimental overview. See Figure S1 for the FACS gating scheme. (B) Expression of known cell-type markers, derived from previously published human cell data from fresh brains (Zhang et al., 2016), in QC-passing expression profiles indicates high cell-type purity. Each gene was Z score normalized across all profiles of all cell types. See Figure S2 for QC analyses. (C) Principal-component analysis using most variable genes reveals separation of four cell types. The juxtaposition of astrocyte and endothelial cell profiles, and the modest detection of astrocyte markers in endothelial cell samples (see D), may have resulted from astrocytic endfeet (which contain mRNAs; Boulay et al., 2017) remaining associated with endothelial cell bodies. (D) Expression levels ± SEM of selected cell-type markers. (E) Distributions of gene counts in various human microglial gene expression datasets. Each boxplot shows the indicated (10th,25th,50th,75th, and 90th) quantile across all genes, of raw gene counts for each sample of bulk-sorted microglia or, for syn18485175, for each sample’s pseudobulk microglia.

Figure 2.. Sorted Cells from Frozen Human SFG Specimens Exhibit Preferential Expression of Many AD Risk Genes in Microglia

(A) Heatmap of Z scores for each AD risk gene’s normalized reads per kilobase gene model per million total reads (nRPKM) expression value in each sample, with a sample’s Z score for a given gene representing its distance in standard deviations from the mean expression value across all samples for that gene. Gene selection was informed by genome-wide association study (GWAS) reports (Hollingworth et al., 2011; Lambert et al., 2013; Naj et al., 2011; Ramanan et al., 2015; Sims et al., 2017) and specific efforts to identify causal genes in GWAS-identified loci ( Huanget al., 2017; Novikova et al., 2019; Rathore et al., 2018). (B) Expression values are plotted for each AD risk gene in each cell type sorted from frozen SFG of controls (Ctl) or AD patients. Bars and lines represent mean expression ± SEM, with asterisks marking DE in AD versus control cells based on unadjusted DESeq2 p values (*p < 0.05, **p < 0.01, ***p < 0.001).

Figure 3.. Human Microglia Exhibit an AD-Associated DE Profile in Both Frontal and Temporal Cortices

(A) Heatmap of AD DE genes (rows; DESeq2 adjusted p ≤ 0.05 and maximum Cook’s p ≥ 0.01) in control and AD SFG-derived microglia expression profiles (columns, sorted by AD-associated DE). “Panel B genes” indicates genes that were subsequently assayed by qPCR in microglia sorted from FuG tissues, with colors from (B). (B) 4-way comparison of AD-associated DE in SFG microglia measured by RNA-seq (x axis) with DE in FuG microglia measured by qPCR (y axis). Each point represents one gene colored by whether the adjusted p value was ≤ 0.05 in one or both DE analyses (red for SFG RNA-seq, green for FuG qPCR, or blue for both). Corresponding numbers of DE genes are shown near the borders of the plot. For example, the red 11 on the right reflects the number of genes that were significantly up in SFG and trended up but did not meet significance in FuG, whereas the blue 3 at the top right indicates the number of genes significantly upin both regions. Genes were selected manually for validation, consisting of about 1/3 of the DE genes from the RNA-seq study and several other cell-type markers and genes of interest. Diagonal line: y = x. (See Figure S3A for subject-wise SFG-FuG microglia DE correlations, Figure S3B for selected qPCR data plots, and Data S2 columns EK–GH for qPCR expression statistics for all 39 genes in the panel.) (C) SFG microglia DE is reproduced in FuG microglia. DE scores (see STAR Methods) are shown for each SFG and FuG microglia sample, using the 22 SFG DE genes that were included in the qPCR panel. For FuG microglia samples, open circles indicate that a QC-passing SFG RNA-seq microglia profile was not available from that subject. p value, t test. (D) Detection of upregulated HAM profile genes is recapitulated in myeloid-balanced whole AD tissues from frontal and temporal cortical regions and is more robust than DAM changes predicted by mouse microglia profiles. Each study was separately myeloid balanced to create a subset of whole-tissue samples with similar myeloid gene set scores, and neuronal genes were removed from each gene set. (See Figure S4C for division by Braak stage with all samples and all genes included.) Each panel shows gene set scores for the indicated gene sets for each of the myeloid-balanced AD or control samples. Δ, mean log2 fold change; p value, t test.

Figure 4.. DE Genes in Mouse Microglia Studies Are Mostly Unchanged in HAM

(A) Distribution of scores for mouse- and human-derived gene sets in SFG microglia profiles indicates that mouse-derived microglia gene modules undergo little or no change in AD microglia. The DAM gene set was called "neurodegeneration-related" in the previous manuscript. p, (unadjusted) t test; Δ, log2 fold changes in score; *p ≤ 0.05. (See Figures S5A–S5C for heatmaps of individual genes from DAM, microglia, and BrainMyeloid modules.) (B) DE gene set scores, similar to (A) but with DE genes from specific mouse datasets instead of from meta-analysis-derived gene modules. In this case, the scores are DE scores, meaning that they used signed means rather than means (with the sign indicating the direction of DE) so that up- and downregulated genes can be considered together. PVMs relative to parenchymal microglia; age, 22 months relative to young (≤12 month) microglia; cerebellum relative to cortical microglia; infiltrating macrophages (induced by irradiation) relative to tissue-resident microglia. p, (unadjusted) t test; Δ, log2 fold changes in score; *p ≤ 0.05. (For the three comparisons that reached significance, see Figure S5D for 4-way plots of individual gene fold changes in the respective mouse study compared to fold changes in AD versus control SFG microglia. See also Figure S6 for analysis of whether DE genes from the HAM profile are altered in mouse microglia in models of neurodegeneration or other activating conditions.)

Figure 5.. AD-Associated HAM Profile Overlaps Substantially with Age-Related DE Patterns in Human Microglia

(A) 4-way DE plot (analogous to Figure 3B) shows age-related DE from Galatro et al. (2017) on the x axis and AD-related DE on the y axis. Color indicates p ≤ 0.05 significance with aging only (red), with AD only (green), or with both (blue). Most red genes, DE with age, trended in a consistent direction with AD versus control microglia (bottom-left and top-right quadrants), indicating that AD microglia exhibit enhanced aging. The green genes, including APOE, indicate an AD-related signature that is distinct from DE of normal aging. (B) Distribution of subject ages in both studies. (C) Previously reported DE pattern in normal, aged human microglia is recapitulated in control subjects of this study. The 4-way plot shows age-related DE from Galatro et al. (2017)’s dataset on the x axis, as in (A), and age-related DE from this study’s control SFG microglia profiles on the y axis. Genes in red met an adjusted p ≤ 0.05 cutoff in Galatro et al. (2017); other genes are shown as a smoothed density in shades of gray. No DE genes from Galatro et al. (2017) met the p ≤ 0.05 cutoff for age-related DE in our dataset, but most trended in a consistent direction (bottom-left and top-right quadrants). The lack of statistical significance and muted fold changes in our study may resultfromfar fewersamples and our samples coming mainly from subjects. (D) Aging DE score was calculated for each SFG microglia sample in our study—a signed average of the age-related DE genes from Galatro et al. (2017). Regression lines show the increasing trend of this score in both diagnosis groups with age, as well as the elevated score in the AD group relative to controls of similar ages. (E) Aging DE score is elevated in AD microglia relative to controls. y coordinates as in (D); p value, t test.

Figure 6.. Monocyte-Enriched Genes May Contribute to Both Late Aging and AD Microglial Signatures

(A) Example gene expression plots. Each point shows the expression of the indicated gene in a single sample in one of the three studies. In the middle column (Galatro et al., 2017), the dashed line indicates the best linear fit. (B) Monocyte DE profiles relative to microglia are similar in human and mouse studies. The 4-way plot is similar to Figure 3B but with DE genes between monocyte and microglia profiles shown with human and mouse studies on the x and y axes, respectively. (C) Many DE changes elevated or depleted in aged human microglia (x axis) are also elevated or depleted, respectively, in blood monocytes relative to microglia (y axis). The 4-way plot shows DE genes with p ≤ 0.05 in the aging study colored red, DE genes with p ≤ 0.05 and fold change ≥ 8 between monocytes and microglia colored green, and DE genes that meet both criteria colored blue. (D) Heatmap of DE genes from the HAM profile in three datasets. Gene ordering was based on the direction of change in this study and then by effect size (fold change per decade) in aging. The subset of HAM-Down genes that show reduced expression in aged microglia generally shows higher expression in microglia than in monocytes. The subset of HAM-Up genes that show increased expression in aged microglia generally shows higher expression in monocytes than in microglia.

Figure 7.. HAM Signature Is Elevated in Multiple Neurodegenerative Settings, whereas DAM Response Is Weaker in AD Microglia

Control-centered scores (log2 scale) for the indicated gene sets were calculated for each sample in the indicated datasets. For snRNA-seq datasets (Mathys et al., 2019, frozen AD tissues, syn18485175; Jä kel et al., 2019, frozen MS tissues, GEO: GSE118257) and scRNA-seq datasets (Masuda et al., 2019, freshly resected MS lesions, GEO: GSE124335; Hasselmann et al., 2019, human induced pluripotent stem cell [iPSC]-derived xMG into 5xFAD mouse brains, GEO: GSE133433), each datapoint represents a pseudobulk microglia profile from pooling individual nuclei/cells from a given subject. (See Data S4, panels 2 and 3, for definitions of microglia clusters used to generate pseudobulk profiles from sn/scRNA-seq datasets.) Other datasets are bulk-sorted brain myeloid cells from frozen AD tissues (this study, GEO: GSE125050) or fresh mouse model tissues (PS2APP b-amyloid and PS19 Tau-P301S models, GEO: GSE89482 and GSE93180). D, log2 fold change of group means; p values from t test. For syn18485175, t test and D were between low- and high-pathology groups. The p value was omitted for GEO: GSE118257, because only one control sample was available (see STAR Methods). See STAR Methods (Gene Set Analysis section) for gene lists and Figure S7 for depictions of individual DE genes across studies.