Uncovering plaque-glia niches in human Alzheimer's disease brains using spatial transcriptomics

Abstract

Background: Amyloid-beta (Aβ) plaques and their associated glial responses are hallmark features of Alzheimer's disease (AD), yet their interactions within the human brain remain poorly defined.

Methods: We applied spatial transcriptomics (ST) and immunohistochemistry (IHC) to 78 postmortem brain sections from 21 individuals in the Religious Orders Study and Memory and Aging Project (ROSMAP). We paired ST with histological data and stratified spots into major categories of plaque-glia niches based on Aβ, GFAP, and IBA1 intensity. Leveraging published ROSMAP single-nucleus RNA-seq data, we examined differences in gene expression, cellular composition, and intercellular communication across these niches. Neuronal and glial changes were validated by IHC and quantitative analyses. We further characterized glial responses using gene set enrichment analysis (GSEA) with known mouse glial signatures and human AD-associated microglial states. Finally, we used iPSC-derived multicellular cultures and single-cell RNA sequencing (scRNA-seq) to identify cell types that, upon short-term Aβ exposure, recapitulate the glial responses observed in the human spatial data.

Results: Low-Aβ regions, enriched for diffuse plaques, exhibited transcriptomic profiles consistent with greater neuronal loss than high-Aβ regions. High-glia regions showed increased expression of inflammatory and neurodegenerative pathways. Spatial glial responses aligned with established gene modules, including plaque-induced genes (PIGs), oligodendrocyte (OLIG) responses, disease-associated microglia (DAM), disease-associated astrocytes (DAA), and human AD-associated microglial states, indicating that diverse glial phenotypes emerge around plaques and shape the local immune environment. IHC confirmed elevated neuronal apoptosis near low-Aβ plaques and greater CD68 abundance and synaptic loss near glia-high plaques. In vitro, iPSC-derived microglia-but not astrocytes-exposed to Aβ displayed transcriptomic changes that closely mirrored the glial states identified in our ST dataset.

Conclusions: Our study provides a comprehensive spatial transcriptomic dataset from human AD brain tissue and bridges spatial gene expression with traditional neuropathology. By integrating ST, snRNA-seq, and human multicellular models, we map cellular states and molecular events within plaque-glia niches. This work offers a spatially resolved framework for dissecting plaque-glia interactions and reveals new insights into the cellular and molecular heterogeneity underlying neurodegenerative pathology.

Supplementary information: The online version contains supplementary material available at 10.1186/s44477-025-00002-z.

Keywords: Alzheimer’s Disease; Amyloid-beta; Glia; Human Brain; Spatial Transcriptomics.

Fig. 1

Spatial Transcriptomics Profiles of Postmortem DLPFC from 21 individuals. A Diagrammatic summary of experimental workflow. Per individual, up to four 10-μm frozen sections of DLPFC were subjected to spatial transcriptomic sequencing using Visium. For each ST section, two adjacent sections were stained for DAPI, Aβ, GFAP, and IBA1 and imaged by fluorescence microscopy. IHC images were then aligned to the middle ST section, enabling spot-level quantification of IHC intensity for each stain and subsequent identification of genes associated with amyloid plaques and/or glial reactivity. Cartoon graphics created with BioRender.com (B) UMAP visualization of all ST spots from all individuals. Clusters were annotated manually based on the enriched expression of known layer- and/or cell-type-specific genes. C Spot clusters overlaid on a representative section (Sect. 8_C). D The expression level of SNAP25 (pan-neuronal), MBP (oligodendrocytes; white matter), and PCP4 (Cortical layer 5) overlaid on a representative tissue section (Sect. 8_C). E Heatmap showing z-score of average expression of a subset of cluster-enriched genes. F Heatmap of GSEA normalized enrichment score (NES) for cell2location predicted neural sub-cell-types abundance across all spots comprising each spot cluster (BH-FDR: *p < 0.01, **p < 0.001, ***p < 1e-4)

Fig. 2

Stratification of tissue microdomains by adjacent-section IHC intensities. A Representative images of different plaque types from two ST-adjacent sections. Six spots (1–6; including all major plaque types) are shown with a 55-µm spot area superimposed to scale. B We stratified all ST spots by Aβ intensity into no (white; log₂(avg Aβ intensity + 1) between 2.5 and 4), low (yellow; log₂(avg Aβ intensity + 1) between 4 and 6.5), or high (blue; log₂(avg Aβ intensity + 1) > 6.5) groups. The distribution of Aβ IF intensity is shown for each group. C A subset of 781 spots from ST-adjacent sections were selected for manual plaque-type annotation (Supplementary Table 5). The proportion of plaque types is shown, stratified by low or high intensity of Aβ [odds ratio (OR) and p-value based on Fisher’s Exact test indicated]. D Spots were further stratified by the abundance of glia (GFAP: x-axis, IBA1: y-axis). Scatterplots show the average GFAP and IBA1 intensities for each ST spot among Low Aβ (left) and High Aβ (right) groups (red: glia-low, blue: glia-high). Gray-colored spots had intermediate GFAP/IBA1 intensity and were not sorted into a group

Fig. 3

Differential effects of low vs. high Aβ on the local transcriptome. A Schematic of the three contrasts tested. Aβ and glia graphics created with BioRender.com (B) Volcano plot of low vs. high Aβ. Genes with FDR-adjusted p < 0.05 in purple. C Scatterplot of Aβ effects stratified by glia-low or glia-high. Genes with FDR-adjusted p < 0.05 in red and blue for glia-low and glia-high conditions, respectively. D Venn diagram showing the number of DEGs found across Aβ contrasts. E GSEA enrichment of Aβ contrasts showing GO terms and canonical pathways (*p < 5e-4, **p < 5e-5, ***p < 5e-6) (F) GSEA enrichment of Aβ contrasts for relevant genesets (*p < 5e-2, **p < 5e-3, ***p < 5e-4). PIG³¹: plaque-induced genes, OLIG³¹: oligodendrocyte gene module, DAM³²: disease-associated microglia, DAA³³: disease-associated astrocytes; MG0-12: microglia states in human AD brains⁸. G Boxplots of representative genes for each Aβ contrast (LMM, FDR-corrected; *p < 0.05, ***p < 5e-4). H Images of representative plaques from FFPE tissue sections stained with DAPI (blue), Aβ (4G8; yellow), NEUN (cyan), and cleaved Caspase 3 (magenta) (scale bar = 25 um). I Average number of Caspase 3 puncta within NEUN + nuclei (left) or total nuclei (right) in the area within and surrounding plaques. Points represent average values from > 100 ROIs for each of 10 AD individuals (paired t-test; *p = 0.041, **p = 2.61e-4)

Fig. 4

Aβ plaques influence intercellular communication. A Differential ligand-receptor (LR) interaction genes between low Aꞵ vs. high Aꞵ (y-axis) and low Aꞵ vs. no Aꞵ spots (x-axis). Each dot corresponds to an LR pair tested; the axis represents the t-statistics from linear mixed model tests accounting for repeated donors. Significant LR pairs (FDR < 0.05) are highlighted in red (low Aꞵ vs. high Aꞵ only), blue (low Aꞵ vs. no Aꞵ only), or purple (significant in both comparisons). B Selected Gene Ontology terms enrichment from LR differentially expressed in low Aꞵ vs. no Aꞵ and low Aꞵ vs. high Aꞵ. Both ligand and receptor genes were considered for the enrichment analysis. P-values were adjusted using the g:SCS method from gprofiler. Colors in the heatmap represent the signed -log10(adj. P). *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001. C Representative Sect. (17_D) image colored by normalized expression of selected differentially expressed LR gene pairs. All the statistical analyses were performed on gray matter spots only

Fig. 5

Differential effects of glia abundance on the local transcriptome. A Schematic of the three contrasts tested. Aβ and glia graphics created with BioRender.com (B) Volcano plot of glia-high vs. glia-low. Genes with FDR-adjusted p < 0.05 in green. C Scatterplot of glia effects stratified by low or high Aβ. Genes with FDR-adjusted p < 0.05 in yellow and blue for low and high Aβ conditions, respectively. D Venn diagram on the number of DEGs found across glia contrasts. E GSEA enrichment of glia contrasts for GO terms and canonical pathways (*p < 5e-4, **p < 5e-5, ***p < 5e-6). F GSEA enrichment of glia contrasts for Aβ/glia-relevant genesets (*p < 5e-2, **p < 5e-3, ***p < 5e-4). PIG³¹: plaque-induced genes, OLIG³¹: oligodendrocyte gene module, DAM³²: disease-associated microglia, DAA³³: disease-associated astrocytes; MG0-12: microglia states in human AD brains⁸. G Boxplot of representative genes found for each glia contrast (LMM, FDR-corrected; *p < 0.05, ***p < 5e-4, n.s. not significant). H Images of representative high Aβ plaques from FFPE tissue sections stained with DAPI (blue), Aβ (4G8; yellow), IBA1 (cyan), GFAP (magenta), and synaptophysin (SYP; green, top-left panel) or NEUN (green, bottom-left panel). SYP and NEUN display lower-intensity staining surrounding glia-high plaques compared to glia-low plaques (quantification on right). Points represent average values from each of 10 AD individuals quantified in the area surrounding plaques (scale bar = 25 µm; paired t-test; **p < 0.005, ***p < 1e-10, n.s. not significant)

Fig. 6

Plaque-surrounding reactive glia modifies intercellular communication. A Differential ligand-receptor (LR) genes between glia-high vs. glia-low under High Aꞵ spots (y-axis) and glia-high vs. glia-low under low Aꞵ spots (x-axis). Each dot corresponds to an LR pair tested; the axis represents the t-statistics from linear model tests accounting for repeated donors. Significant LR pairs (FDR < 0.05) are highlighted in blue (glia-high vs. glia-low under low Aꞵ only), yellow (glia-high vs. glia-low under high Aꞵ only), or green (significant in both comparisons). B Example Sect. (15_D) colored by normalized expression of selected differentially expressed LR gene pairs. All the statistical analyses were performed on gray matter spots only. C Normalized expression levels by stratified spot groups for selected differentially expressed LR pairs. (*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001). D Selected Gene Ontology terms enrichment from LR differentially expressed between glia-high vs. glia-low under low Aꞵ, high Aꞵ, and combined were considered for the enrichment of both ligand and receptor genes. P-values were adjusted using the g:SCS method from gprofiler. Colors in the heatmap represent the signed -log10(adj. P) (*p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001)

Fig. 7

Aβ oligomer treatment of co-cultured iMGL partially recapitulates ST glia response to Aβ plaques. A Schematic of in vitro co-culture experiments. Graphics created with BioRender.com (B) Representative immunofluorescence for DCX (neuronal), IBA1 (iMGL) and GFAP (astrocyte) proteins in co-culture models, scale bar: 100 µm, inset: 50 µm. C UMAP visualization of cells from control and Aβ oligomer-treated cultures, colored by annotated cell type clusters. D Dotplot of cluster/cell-type enriched genes. E GSEA enrichment of Aβ-induced genes for ST glial response by cell type. Red dotted line corresponds to FDR-threshold. F Boxplots of genes differentially expressed in iMGL upon Aβ treatment. G UMAP visualization of Aβ oligomer-treated iMGL, colored by subtype cluster. H Heatmap of the top 5 differentially expressed gene markers per iMGL subtype cluster. I GSEA enrichment of Aβ-treated iMGL cluster DEGs (Supplementary Table 11) for ST glial response. Red dotted line corresponds to FDR-threshold. J GSEA enrichment of Aβ-treated iMGL cluster (vs. control) DEGs (Supplementary Table 12) for reported MG state genesets in human AD brains[8] (*p < 5e-2, **p < 5e-3, ***p < 5e-4)