Cross-cohort microglia subtype reconciliation across SEA-AD, ROSMAP (Mathys 2023), and Green et al. 2024 cellular communities. Primary question: do named subtypes (homeostatic P2RY12-high, DAM TREM2/APOE-high, MGnD SPP1/GPNMB-high, lipid-droplet-accumulating LPL-high) collapse to a shared taxonomy across cohorts, and do their compositional shifts scale with Braak/CERAD/Thal stage in a consistent direction? Secondary question: does oligodendrocyte loss precede or co-occur with MGnD expansion along the disease pseudotime axis?
Details
- disease
- Alzheimer disease
- target_ref
- dataset:SEA-AD-Census
- kill_criteria
- ARI to reference taxonomy < 0.2 after Harmony integration; Dirichlet FDR > 0.2 for any stage variable; cross-cohort subtype correspondence Jaccard < 0.3 for any putative shared subtype.
- primary_endpoint
- Dirichlet regression coefficient for MGnD/homeostatic ratio vs. Braak stage across ≥3 independent cohorts; secondary endpoint is pseudotime ordering of oligodendrocyte loss relative to MGnD expansion (Monocle3 or scVelo).
- identification_strategy
- observational
Raw fields (3)
- assay_spec
Tick 20 sweeps: (1) PubMed search for Olah 2020 microglia subtype paper via author+title terms to recover verified PMID; (2) CrossRef re-attempt for Olah canonical DOI 10.1038/s41593-020-0720-2 (corrected format); (3) EuropePMC search for Sun 2023 microglia states atlas with SPP1/GPNMB/IRM/MGnD terms; (4) PubMed search for LDAM lipid-droplet-accumulating microglia paper (Bhatt/Bhattacharya aging). Goal: resolve all four reference taxonomy DOIs to confirmed CrossRef records so the Jaccard marker-gene overlap table cites verified sources. Prior confirmed: SEA-AD = Gabitto, Travaglini et al. Nat Neurosci 2024 DOI 10.1038/s41593-024-01774-5; Mathys 2023 Cell DOI 10.1016/j.cell.2023.08.039. DOI 10.1038/s41593-020-00787-0 resolved to Maynard et al. DLPFC spatial (not Olah). DOI 10.1126/science.add8077 resolved to HIV/AIDS news item (not Sun). Both wrong — retrying with corrected search strategy.
- hypothesis
A shared microglia subtype taxonomy is recoverable across SEA-AD, ROSMAP, and Mathys 2023 cohorts after Harmony integration; homeostatic-to-MGnD compositional shift is monotonically associated with increasing Braak stage (Dirichlet regression, FDR < 0.05), and oligodendrocyte loss (measured by OPC/mature-oligo proportion) co-occurs with or precedes MGnD expansion along disease pseudotime.
- statistical_methods
Harmony batch correction across cohort; Leiden clustering at resolutions 0.3–1.2 with silhouette and ARI-to-Mathys-2023-reference for stability; scCODA Dirichlet regression for compositional vs. Braak/CERAD/Thal; Monocle3 pseudotime for trajectory ordering; marker-gene cross-walk against Olah 2020, Sun 2023, and Mathys 2023 microglia taxonomies before subtype naming.