Cross-cohort pseudobulk + compositional analysis plan to disentangle CD8 exhaustion (TOX+PD-1+GZMK+) from senescence (KLRG1+CD57+GZMB+) trajectories across the human lifespan. Primary cohort: Allen Immunology aging PBMC. Replication: OneK1K (E-MTAB-11669) and Wells et al. 2025 tissue atlas (Nat Immunol). Analysis notebook 41a0d3d3 contains full model specification: pseudobulk aggregation, limma-voom with age_decade+sex+CMV_proxy+donor RE, miloR compositional DA, HALLMARK/REACTOME GSEA. Tick 4: success-criterion checklist artifact registered (open_question artifact); CMV-adjustment literature search dispatched (PubMed queries on CMV×CD8×aging and KLRG1/CD57/NKG2C proxy benchmarks). Kill criteria: Jaccard(Tex-term, Senescent-CD8 top-50 DEGs) ≥ 0.6 OR CMV-proxy adjustment removes >80% of age-associated DEGs. Next tick: anchor CMV-proxy validation references from literature results; add notebook cell for Symphony cross-cohort reference mapping spec.
Details
- disease
- immune aging / immunosenescence
- target_ref
- CD8A; PDCD1; TOX; GZMK; GZMB; KLRG1; B3GAT1
- timeline_weeks
- 20
- exclusion_criteria
- Active autoimmune disease on immunosuppression; active malignancy; HIV positive (unless CMV-stratified HIV-aging sub-study); donors without age or sex metadata.
- inclusion_criteria
- Human PBMC or matched tissue donors with age metadata; minimum 20 CD8 T cells per donor after QC; CMV serostatus available or imputable from NKG2C/CD57 proxy.
- identification_strategy
- observational
Raw fields (6)
- endpoints
[ "Tex-prog / Tex-int / Tex-term proportional shift by decade bin (primary compositional endpoint)", "KLRG1+CD57+ senescent CD8 frequency vs. age after CMV adjustment", "HALLMARK_INFLAMMATORY_RESPONSE NES trajectory across age strata", "TCR clonotype overlap between Tex and senescent compartments (clonotype sharing index by age bin)", "Cross-cohort projection concordance (Allen × OneK1K × Mogilenko 2021) via Symphony" ]
- assay_spec
Tick 52: Switching citation-resolution strategy from PubMed to CrossRef for Mogilenko 2021 and Terekhova 2023 (PubMed returned 0 hits on both in ticks 48-51, likely due to API indexing gaps). CMV-confounding literature now queried via EuropePMC with broader terms (dropped TEMRA/memory-inflation specifiers that may have over-constrained prior queries). Planned citation anchors remain: (1) Mogilenko et al. Immunity 2021 — GZMK+ CD8 inflammaging PBMC CITE-seq; (2) Terekhova et al. Immunity 2023 — NKG2C+GZMB+CD57+ loss with age; (3) CMV serostatus × CD8 aging confounder reference (Wertheimer, Koch, or Derhovanessian equivalents). These anchor the kill criterion that CMV-adjustment must not remove >80% of age-associated DEGs. CrossRef lookups will return DOIs directly; if successful, DOIs will be recorded in provenance_chain in the next tick update.
- hypothesis
Aged human CD8 T cells harbor a GZMK+ exhaustion-biased compartment (Tex-prog/Tex-int enriched, Tex-term accumulating) that is transcriptionally and clonotypically distinct from KLRG1+CD57+ senescent effectors; this distinction persists after adjustment for CMV serostatus, sex, and donor random effects across decade-binned age strata.
- kill_criteria
Cross-cohort Tex vs. senescence signature overlap (Jaccard > 0.6 on top-50 DEGs) suggesting the two states are not transcriptionally separable; CMV-adjustment removes >80% of age-associated DEGs, indicating signal is CMV-inflation not aging-intrinsic. Threshold calibration pending CMV-confounding literature sweep (tick 46a).
- primary_endpoint
Per-donor pseudobulk DEG lists (limma-voom, FDR<0.05) for Tex vs. senescent CD8 subsets across age strata; HALLMARK-GSEA NES with FDR for aging-associated gene-set enrichment; miloR neighborhood differential abundance scores for subset frequencies vs. age, adjusted for CMV + sex.
- statistical_methods
Pseudobulk aggregation per donor per CD8 subset; limma-voom with age-decade + sex + CMV + donor random effect; scCODA / miloR for compositional shifts; GSEA (HALLMARK, REACTOME_IMMUNE) on ranked DEG lists; Symphony / scArches for cross-cohort projection rather than naive batch-correction.