Use SEA-AD multimodal atlas (Gabitto, Travaglini et al. 2024, Nat Neurosci) snRNA-seq + MERFISH across 84 donors staged by Braak/CERAD/Thal to: (1) extract microglia and oligodendrocyte lineage pseudobulk counts per donor per Braak stage; (2) fit scCODA/Dirichlet regression for MGnD (SPP1, GPNMB, LPL, ITGAX markers), homeostatic MG (P2RY12, TMEM119, CX3CR1), mature oligodendrocytes, and OPCs as response variables against Braak as ordinal covariate; (3) compute cross-cell-type Pearson correlation between MGnD proportion and oligo/OPC proportion per donor; (4) test causal ordering via lagged-Braak regression (does MGnD at Braak N predict oligo loss at Braak N+1 within cross-cohort replication SEA-AD × ROSMAP × Mathys 2023?). MERFISH layer-resolved co-localization of SPP1 signal with MBP/MOG signal will serve as spatial validation. In silico KO of SPP1 and GPNMB in microglia pseudobulk signatures (zeroing expression, recomputing predicted NMF program scores) will estimate directionality. Cross-cohort replication required before any causal claim.
Details
- disease
- alzheimer's disease
- target_ref
- SEA-AD:DLPFC:snRNA-seq
- identification_strategy
- in_silico_KO
Raw fields (4)
- assay_spec
Step 1 (feasibility gate): CELLxGENE Census query — pull SEA-AD DLPFC snRNA-seq (collection_id filter on SEA-AD), filter to microglia (cell_type LIKE '%microglia%') and OPC/oligodendrocyte lineage (cell_type LIKE '%oligodendrocyte%' OR '%OPC%'), join donor metadata for Braak stage (obs columns: 'Braak stage' or 'braak_stage'). Confirm ≥10 donors per Braak stratum per lineage before proceeding. Step 2 (composition): pseudobulk cell-type fraction per donor, Dirichlet regression (scCODA) of MGnD-state fraction (proxy: SPP1+/GPNMB+ cluster) and OPC/mature-oligo fraction against Braak stage as ordinal predictor; report posterior coefficient ± 95% credible interval. Step 3 (temporal ordering): RNA velocity (scVelo) on microglia subset and oligodendrocyte subset separately; assess directionality at Braak III–IV boundary. MERFISH layer co-localization of SPP1+GPNMB+ microglia with MBP-low zones as spatial validation. Kill criteria apply at each gate. All outputs are planned; no results yet exist.
- hypothesis
In late-Braak DLPFC, the compositional expansion of SPP1+/GPNMB+ MGnD-state microglia is temporally upstream of oligodendrocyte and OPC depletion, mediated by lipid-handling and phagocytic gene programs that strip myelin before remyelination capacity is exhausted.
- kill_criteria
Abort or revise if: (1) Braak stage is absent as a direct obs column in SEA-AD Census slice and cannot be recovered via donor metadata join; (2) fewer than 10 donors per stratum for either microglia or OPC/oligo lineage; (3) SPP1+/GPNMB+ MGnD subtype labels are not present in the cell-type annotation layer and cannot be recovered by marker-gene scoring; (4) cross-cohort replication in ROSMAP Census slice yields <5 donors per stratum, making composition regression underpowered.
- primary_endpoint
Planned outputs: (1) donor×stratum feasibility table (≥10 donors/stratum/lineage confirmed or denied); (2) Dirichlet regression posterior coefficient and 95% CI for MGnD fraction vs Braak stage, and for OPC/oligo fraction vs Braak stage; (3) RNA velocity directionality summary at Braak III–IV boundary for each lineage; (4) MERFISH co-localization score (Manders' M1/M2) for SPP1+GPNMB+ microglia vs MBP-low zones across Braak strata. Success criterion: MGnD fraction increase and OPC/oligo fraction decrease both reach posterior probability >0.95 with non-overlapping 95% CIs, and velocity arrows in microglia precede oligo depletion signal.