This project characterises chromatin accessibility and transcriptional programs of putative disease-associated microglial (DAM) substates in human Alzheimer's disease (AD) dorsolateral prefrontal cortex (DLPFC) using single-nucleus multiome sequencing (snRNA-seq + snATAC-seq). The analysis builds on Anderson et al. 2023 (Cell Genomics), who identified AD-specific cis-regulatory elements genome-wide and nominated MAFB as a microglial AD-associated regulator. This project focuses specifically on microglial substates, testing whether MAFB and NR4A2 motif activity programs separate homeostatic-like from DAM-like microglia at the chromatin level — a microglial-substate-focused reanalysis that was not the primary focus of the original study.
- Source: GSE214979 — Anderson et al. 2023, Cell Genomics
- Title: Single nucleus multiomics identifies ZEB1 and MAFB as candidate regulators of Alzheimer's disease-specific cis-regulatory elements
- Content: 105,332 nuclei from human DLPFC, 7 AD donors + 8 unaffected controls
- Technology: 10x Genomics Multiome (matched snRNA-seq + snATAC-seq per nucleus)
- Files used:
GSE214979_filtered_feature_bc_matrix.h5(1.3 GB),GSE214979_cell_metadata.csv.gz - Note: Fragment file (59.3 GB) not downloaded — coverage plots not generated by design
Do MAFB and NR4A2 motif activity programs distinguish homeostatic-like and putative disease-associated microglial substates in human AD DLPFC, and are DAM-associated genes supported by concordant changes in gene expression and chromatin accessibility?
-
Three putative microglial substates identified after Harmony batch correction and WNN joint embedding: DAM-like (64 nuclei), Homeostatic-like (1,220 nuclei), Intermediate (1,695 nuclei). Substates were assigned based on published DAM and homeostatic gene signatures (Keren-Shaul et al. 2017).
-
74% of Homeostatic-like nuclei were from unaffected controls; 67% of Intermediate nuclei were from AD donors — consistent with, but not proving, a possible homeostatic-to-disease-associated continuum.
-
42 chromatin peaks significantly more accessible in putative DAM-like microglia (Wilcoxon rank sum test, adjusted p < 0.05) — near genes including TNFRSF9, DOCK10, NRXN3, PCDH9, and NEDD4L. Most peaks are intergenic, representing candidate distal regulatory elements.
-
15 significant peak-to-gene links at DAM and homeostatic marker loci (Pearson correlation, p < 0.05). Among tested loci, LPL and TYROBP showed the highest positive peak-to-gene correlations (r = 0.14 and r = 0.08 respectively), suggesting concordant chromatin opening and gene expression at these DAM marker loci.
-
MAFB and NR4A2 motif activity did not significantly differ between putative DAM-like and Homeostatic-like substates (MAFB proxy p = 0.938, NR4A2 p = 0.819, Wilcoxon rank sum test). This is attributed to the small DAM-like population (64 nuclei) and the absence of a MAFB-specific position weight matrix in JASPAR2020 and JASPAR2022 — a MAF family proxy (MA1520.1) was used instead.
-
Monocle3 pseudotime placed the Intermediate population at the highest mean pseudotime (4.19), supporting a possible transitional interpretation between homeostatic and DAM-like states.
Microglial substates — WNN Harmony UMAP
Volcano plot — DAM-like vs Homeostatic-like (DESeq2)
Peak-to-gene links at DAM and homeostatic loci
Monocle3 pseudotime — homeostatic to DAM
01_metadata_filter.R Filter metadata to microglia nuclei
subset_microglia.py Slice H5 file to microglia RNA + ATAC matrices
02_load_object.R Build Seurat object with RNA and ATAC assays
03_qc.R QC filtering
04_normalisation.R Log normalisation (RNA), TF-IDF (ATAC)
05_dimreduction.R PCA (RNA), LSI (ATAC), DepthCor
06_clustering_umap.R Clustering and UMAP per modality
07_harmony_wnn.R Harmony batch correction, WNN joint embedding
08_dam_scoring.R DAM/homeostatic module scoring, substate assignment
09_differential_rna.R DESeq2 pseudobulk differential expression
10_differential_atac.R Differential chromatin accessibility
11_chromvar.R chromVAR TF motif activity (633 JASPAR2020 motifs)
12_peak_gene_linkage.R Peak-to-gene linkage at DAM and homeostatic loci
13_trajectory.R Monocle3 pseudotime trajectory analysis
- R >= 4.4
- Python >= 3.9 with conda environment containing:
h5py,scipy,pandas,numpy - R packages: Seurat, Signac, harmony, DESeq2, chromVAR, JASPAR2020, monocle3, SeuratWrappers
# Clone repository
git clone https://github.com/Varun-U-Gowda/ad-microglia-multiome.git
cd ad-microglia-multiome
# Download data from GEO (not included in repo — large files)
# Place in data/raw/
# GSE214979_filtered_feature_bc_matrix.h5
# GSE214979_cell_metadata.csv.gz# In R, from project root
source("run_pipeline.R")Python subsetting step runs automatically via system() call in run_pipeline.R. Requires conda environment path set correctly in run_pipeline.R.
| Tool | Purpose |
|---|---|
| Seurat v5 | RNA processing, WNN integration, clustering |
| Signac | ATAC processing, differential accessibility, peak-to-gene linkage |
| Harmony | Donor batch correction |
| chromVAR | TF motif activity per nucleus |
| JASPAR2020 | TF motif database (633 human motifs) |
| DESeq2 | Pseudobulk differential expression |
| Monocle3 | Pseudotime trajectory analysis |
| BSgenome.Hsapiens.UCSC.hg38 | Human genome reference |
| EnsDb.Hsapiens.v86 | Gene annotations |
- Small putative DAM-like population — 64 nuclei across 15 donors limits statistical power for pseudobulk DE, differential ATAC, and chromVAR analyses. Results should be interpreted cautiously.
- MAFB absent from JASPAR2020 and JASPAR2022 — MAF family proxy (MA1520.1) used; future analysis should use ChIP-seq derived MAFB motifs from ENCODE or ReMap.
- Fragment file not downloaded — coverage and genome track plots not generated (fragment file = 59.3 GB).
- LSI component 2 retained — moderate depth correlation (r = -0.539) noted; sensitivity analysis with dims 3:20 is a future direction.
- Observational data only — all findings are associative. Causal TF function cannot be inferred from snATAC-seq data.
- Substate assignment is score-based — DAM-like and Homeostatic-like labels are derived from published module scores, not validated by orthogonal methods in this project.
Anderson et al. identified AD-specific cis-regulatory elements genome-wide across all brain cell types and nominated MAFB as a microglial AD-associated regulator. This project builds on that result by performing a focused microglial substate analysis — subclustering microglia, scoring for published DAM signatures, and testing whether MAFB and NR4A2 motif activity separate putative DAM-like from homeostatic-like substates. The primary focus of Anderson et al. was atlas-level CRE discovery across cell types; the microglial substate regulatory axis tested here was not their primary question.
See CITATIONS.md for full citations of the dataset, biological references, and all tools used.
February 2026 – June 2026
Varun U Gowda