# Cross-Reference: Literature Candidates × EoE Omics Meta-Analysis

*Validating the literature-derived target list against an existing 9-cohort EoE bulk
transcriptomic meta-analysis (`eoe_meta_signature.csv`, random-effects pooled log2FC with
heterogeneity I²) and the omics-derived druggability shortlist (`eoe_target_shortlist.csv`)
already in this project. Purpose: separate targets with **real, measured, cross-cohort EoE
dysregulation** from those riding on publication attention alone — before committing design effort.*

## The one-line result

The two independent lines of evidence **converge**. Candidates I surfaced from the literature
sweep all appear on an omics-built 20-gene druggability shortlist that I had no hand in —
POSTN near its top (rank 3/20, priority 92), and CCL26 (11/20), ALOX15 (15/20), CPA3 (16/20)
in its lower half but still selected — and the meta-analysis confirms their EoE direction
across 9 cohorts. (The shortlist is itself a filtered top-20, so appearing on it at all means
passing an omics + druggability screen; but only POSTN ranks near the top — the others are
mid-to-lower on that list, carried more by literature + direction concordance than by the
shortlist's own priority score.) The cross-reference also **demotes** several lit-popular
targets and flags a critical interpretation caveat for the cytokines.

## Verdict by target

| Target | Lit (abs, up/dn) | Trials | Meta log2FC (padj) | k, I² | Verdict |
|---|---|---|---|---|---|
| *CCL26* | 45 (22/4) | 0 | +4.56 (3e-04) | 9, 98% | CONFIRMED + open |
| *POSTN* | 9 (6/2) | 0 | +5.55 (3e-16) | 8, 72% | CONFIRMED + open |
| *CAPN14* | 14 (5/0) | 0 | +1.85 (8e-05) | 9, 93% | CONFIRMED + open |
| *CPA3* | 4 (3/1) | 0 | +3.28 (3e-05) | 9, 94% | CONFIRMED + open |
| *ALOX15* | 3 (3/0) | 0 | +5.92 (2e-03) | 9, 98% | CONFIRMED + open |
| *DSG1* | 10 (2/5) | 0 | -2.73 (4e-04) | 9, 96% | CONFIRMED + open |
| *SPINK7* | 5 (0/3) | 0 | -2.46 (1e-04) | 9, 87% | CONFIRMED + open |
| *FLG* | 7 (1/3) | 0 | -1.68 (2e-02) | 9, 88% | CONFIRMED + open |
| *IL15* | 5 (3/1) | 0 | +0.98 (9e-07) | 8, 66% | Confirmed (modest effect) |
| *CCR3* | 4 (3/0) | 0 | +0.59 (6e-06) | 7, 41% | Confirmed (modest effect) |
| *IL13* | 110 (59/9) | 3 (PHASE2) | +0.66 (2e-03) | 9, 87% | Confirmed (has trials) |
| *IL5* | 72 (38/6) | 6 (PHASE3) | +0.26 (2e-03) | 9, 39% | Confirmed (has trials) |
| *TSLP* | 61 (33/3) | 1 (PHASE3) | +0.31 (6e-02) | 8, 33% | Bulk n.s. (cell-specific?) |
| *IL4* | 34 (19/3) | 0 | +0.04 (7e-01) | 5, 0% | Bulk n.s. (cell-specific?) |
| *IL33* | 11 (10/0) | 0 | +0.25 (4e-01) | 8, 62% | Bulk n.s. (cell-specific?) |
| *STAT6* | 7 (2/0) | 0 | +0.35 (3e-02) | 8, 75% | Confirmed (modest effect) |
| *SIGLEC8* | 11 (1/1) | 3 (PHASE3) | +0.86 (2e-03) | 9, 90% | DISCORDANT — caution |
| *IL18* | 4 (4/0) | 0 | -1.46 (2e-05) | 8, 78% | DISCORDANT — caution |
| *TGFB1* | 13 (6/1) | 2 (PHASE2) | +0.02 (9e-01) | 9, 33% | Bulk n.s. (cell-specific?) |

## What survived, what didn't

**Confirmed + open pipeline (design-ready).** These are strongly, consistently dysregulated in
EoE tissue across cohorts AND have no dedicated therapeutic program:
- **CCL26 (eotaxin-3)** — pooled +4.56 log2FC, significant, up in all 9 cohorts (I²=98% reflects
  magnitude variation, not direction — every cohort agrees on UP). Secreted, antibody-tractable,
  0 drugs. The single strongest "biology + designability + open" candidate. Confirmed in the
  independent shortlist (priority 80, ligand-trap modality).
- **POSTN (periostin)** — pooled +5.55, padj 3e-16, up in 8/8; top-3 in the omics shortlist
  (priority 92). Remodeling matrix protein, secreted, ligand-trap/neutralizing-Ab.
- **CAPN14** — pooled +1.85, up in 9/9. The EoE GWAS gene, now confirmed dysregulated in the
  meta-analysis. Intracellular protease → NOT an antibody target; small-molecule or a diagnostic/
  biomarker role, not a binder. Flagged so Stream 3 doesn't design an impossible binder.
- **CPA3, ALOX15** — mast-cell (CPA3) and eosinophil (ALOX15) effector enzymes, both strongly up
  and in the shortlist. Effector-cell abundance markers more than clean drug targets.

**Barrier-loss axis (restoration logic, distinct modality).** DSG1 (−2.73), SPINK7 (−2.46),
FLG (−1.68) — all significantly DOWN across cohorts, concordant with the literature. These are
not blockade targets; they argue for a barrier-restoration or upstream-driver-blockade strategy.

**The cytokine caveat — do NOT read bulk log2FC as importance.** IL-13 (+0.66), IL-5 (+0.26),
TSLP (+0.31), IL-4 (+0.04), IL-33 (+0.25) all show small or non-significant **bulk-tissue**
effects. This is expected and does not demote them: secreted cytokines are low-abundance and
produced by rare infiltrating cells, so bulk RNA-seq dilutes them badly. IL-13/IL-5 are the
validated, approved-drug core of EoE — their modest bulk signal is a limitation of the *assay*,
not the biology. Single-cell DE (per-cell-type) is the correct resolution for these, and the
project's sc data should be the tiebreaker.

**Discordant — caution flags:**
- **IL-18** — literature calls it up, meta says significantly DOWN (−1.46, 7/8 cohorts down).
  Worth a direct look before any claim.
- **SIGLEC8** — lit direction mixed; meta shows the *tissue* signal is up (eosinophil-driven).
  Not a contradiction so much as a cell-composition effect. Already Ph3 (lirentelimab) so moot
  for novel design.

## Cell-of-origin (single-cell DE, EoE esophageal biopsies)

The project's single-cell DE (`sc_celltype_de.csv`; cell types: epithelial basal/suprabasal,
mast, myeloid, T cell, endothelial) localizes the lead candidates to their source cell — which
matters for both mechanism and delivery:
- **CCL26** → suprabasal epithelium (logFC +6.8) — epithelial-secreted, the eotaxin driving eosinophil recruitment.
- **POSTN** → basal epithelium (+8.9) — epithelial/remodeling.
- **CAPN14** → basal + suprabasal epithelium (+5.2 / +3.0) — epithelial, IL-13-inducible, confirming the GWAS-gene mechanism.
- **CPA3** → mast cells (+1.7) — mast-cell effector, a mast-cell abundance marker.
- **ALOX15** → suprabasal epithelium + myeloid (+10.8 / +8.6) — eosinophil/epithelial 15-LOX.

This resolves the bulk cytokine caveat in the right direction: the epithelial-secreted alarmin/
chemokine axis (CCL26, POSTN) shows strong, cell-localized signal even where bulk is noisy, and
CCL26's role as the epithelial eosinophil-recruiting chemokine is confirmed at single-cell resolution.

## Revised design priority (post-cross-reference)

1. **CCL26** — confirmed both ways, secreted, open pipeline, antibody/trap-tractable. **Lead candidate.**
2. **POSTN** — confirmed both ways, top of omics shortlist, secreted. **Strong second.**
3. **Barrier axis (DSG1)** — novel mechanism, higher-risk, restoration logic. Differentiated bet.
4. **IL-33** — keep as a cytokine-axis option BUT validate on single-cell first (bulk n.s.).
- **De-prioritize for de-novo binder design:** CAPN14 (intracellular), IL-13/IL-5 (already approved/late-stage core).

## Caveats
- Meta-analysis is **bulk** tissue — cell-type-specific and low-abundance signals (cytokines) are
  systematically underpowered here; use the project single-cell DE to resolve those.
- I²>90% on CCL26/ALOX15 reflects effect-size heterogeneity across cohorts, not directional
  disagreement (n_up shows unanimity) — but it does mean the pooled magnitude is a rough estimate.
- Druggability/accessibility annotations are inherited from the existing shortlist; spot-check
  before committing.