# C7 decoupling analysis record (03) — 2026-10-30
## Design
- Arm 1, hyperglycemia effect: GSE263483 GDM-AGA vs Control (n=8v8, PyDESeq2 recomputation)
- Arm 2, GDM-macrosomia: GSE267340 GDMM1+GDMM2 pseudobulk vs CONTROL (n=1, **descriptive LFC, no
  statistical test**)
- Pseudobulk: 10x raw matrices aggregated per sample → summed by gene symbol → DESeq2
  median-of-ratios size factors → CPM

## Check correction (v1→v2)
- v1 had no expression filter; low-expression genes produced spurious extreme LFCs
  (e.g. CGB2 LFC_mac=22.8 with near-zero control expression)
- v2 filter: baseMean_hyp>=30 & pseudobulk CPM>=20 → 13,443 genes retained; results rationalised

## Results (v2, |LFC|>=0.58)
- Hyperglycemia arm significant: 2,391 genes; macrosomia arm |LFC|>=0.58: 4,048
- Reversed by "overgrowth": 1,172 (opposite directions in the two arms)
- Concordant between arms: 500 (genome-wide direction-agreement rate 44%)
- Key anchor: 13 mtDNA genes mean LFC_mac = +0.34 (no down-regulation in the macrosomia arm; none
  in the AGA arm either; down-regulation only in the mixed GSE249311 cohort)

## Enrichment (Enrichr, GO_BP+KEGG, adj.P)
- Concordant genes: NK-cell-mediated cytotoxicity, transplant rejection, cell adhesion, defence
  response to virus (immune-activation core)
- Reversed genes: ECM organization/collagen fibril/focal adhesion/AGE-RAGE (diabetic complications)
  — hyperglycemia arm up on ECM while the macrosomia arm is opposite
- Macrosomia-specific: cell migration/motility, DNA replication, axon guidance (growth programs)

## Biological interpretation (preliminary, pending spatial/single-cell confirmation)
1. Shared GDM response = maternal-fetal interface immune activation (CXCL8/9, IL1R2, PRL, SLPI)
2. ECM remodelling is the main axis of the "hyperglycemia vs overgrowth" divergence
3. Mitochondrial respiratory-suppression signature (mtDNA down) appears only in the mixed
   GSE249311 cohort → possibly related to treatment/batch/clinical composition; must not be
   interpreted as an obligatory GDM effect — state in Discussion and limitations

## Limitations (must be declared in the paper)
- Arm 2 control n=1: no statistical inference; GDMM fully confounded with sample; some extreme
  GSE263483 LFCs (SLC1A6 8.6) pending re-check
- Follow-up: GSE173193 pseudobulk as a third-party directional reference
