# EoE Phase-2 Report: Mechanistic Validation, Novelty Grounding & Therapeutic Design

## Executive summary

Phase-2 subjected the Phase-1 target hypotheses to mechanistic validation, literature grounding,
genetic cross-referencing, cross-cohort reproducibility testing, and protein-design specification.
The prioritized targets survive scrutiny, one novel target (SIGLEC6) was independently confirmed,
one proposed axis (S100A8/9) was corrected by the data, and all leads were mapped to patient-selection
biomarkers and differentiated from dupilumab.

## 1. Mechanistic validation (single-cell GSE218607)

**1.1 Epithelial differentiation arrest.** Diffusion pseudotime across 142,749 epithelial cells shows
EoE epithelium is developmentally arrested: only 5.4% of EoE cells reach terminal differentiation vs
17.9% in healthy (p≈0). Barrier genes (FLG, DSG1, SPINK7, CRNN) are blunted even at matched maturation
stage — the barrier defect is a failure to complete differentiation, not just reduced expression.

**1.2 Interferon signature is type-II/IFN-γ.** Deconvolution resolves the Phase-1 IFN dominance as a
T-cell→epithelium IFN-γ axis: IFNG expressed by ~22% of T cells; type-I IFN cytokines absent; epithelium
is the primary responder (drives HLA/antigen presentation). IFN-γ program trends with mast burden
(r=0.59, p=0.057). Positions a JAK/STAT node in EoE mechanism.

**1.3 EoE-specific cell-cell communication.** Three-way comparison (EoE vs GERD vs healthy) using the
held-out GERD arm (91,752 cells) finds 53 EoE-specific ligand-receptor axes vs only 5 shared with GERD.
Dominant EoE-specific themes: mast TIMP1→CD63 remodeling, VIM→CD44, BMP1→BMPR1A, S100A4/ANXA1→EGFR.
Confirms signaling is EoE biology, not generic esophagitis. GERD independently confirms mast expansion
is EoE-specific (0.41% vs 2.6%).

**1.4 SIGLEC6 marks the disease-associated mast state.** Mast subclustering resolves a resting state
(mixed) from three disease-associated states (~99% EoE). SIGLEC6 is expressed on 84.5% of EoE mast cells
vs 34.2% healthy (p=3e-31) — validating it as a mast-targeting/depletion marker.

## 2. Patient stratification & genetics

**2.1 Reproducible endotypes.** 110 EoE patients across 6 cohorts stratify into a mild→intermediate→severe
inflammatory gradient reproducible in 6/6 cohorts. Th2/mast/IFN modules co-vary and anti-correlate with
barrier. Broad-utility targets (CCL26/ALOX15/CDH26) are addressable in ~all patients; precision targets
(SIGLEC6/CPA3/NTRK2) scale with severity.

**2.2 PPI-refractory persistence.** Using recovered PPI responder/non-responder labels (GSE303169), all
lead targets normalize toward control after PPI in responders but persist in non-responders (e.g. ALOX15
8.2→8.2, POSTN 7.1→6.8). The prioritized targets remain active precisely in the treatment-refractory
population that most needs a novel targeted agent.

**2.3 GWAS overlap.** 7 of 567 signature genes carry EoE genetic risk (CAPN14 rs143457388 OR 1.77;
DSG1 rs7236477 OR 2.22; DDAH1, MT2A, IFFO2, NRXN1, ST6GAL1). Genetic support maps to susceptibility
genes (epithelial barrier: CAPN14, DSG1), not effectors (CCL26, ALOX15 are downstream amplifiers) — the
expected risk-gene-vs-effector split.

## 3. Novelty ledger & independent confirmation

Of 12 findings classified against retrieved literature: 7 confirmatory, 1 confirmatory-with-novel-emphasis
(IFN-γ resolution), 3 novel (SIGLEC6 mast state, S100 alarmin axis, PPI-refractory triage), 1 contradictory
(FLG up in recurrence).

Independent-dataset confirmation:
- **SIGLEC6 → CONFIRMED** (up 8/8 bulk cohorts, independent of scRNA derivation).
- **S100A8/9 axis → CORRECTED**: ligands are DOWN in bulk (track differentiation loss); the mast remodeling
  signal is S100A4, not the S100A8/9 alarmin. Real self-correction forced by the data.
- **PPI-refractory → CONFIRMED** within GSE303169 (single cohort; needs replication).
- **FLG contradiction → COHORT-SPECIFIC**: canonical downregulation holds in 8/9 cohorts; recurrence reversal
  isolated to GSE278888.

## 4. Target–biomarker pairing & dupilumab differentiation

| Target | Biomarker | Selection | Differentiation vs dupilumab |
|--------|-----------|-----------|------------------------------|
| CCL26 | serum eotaxin-3 | universal | downstream-specific; avoids broad Th2 immunosuppression |
| IL1RL1/ST2 | soluble ST2 | severity-scaling | upstream of IL-4/13 (hits initiating alarmin) |
| POSTN | serum periostin | severity-scaling | anti-fibrotic/remodeling angle |
| SIGLEC6 | SIGLEC6+ mast fraction | mast-high; most homogeneous | mast depletion (dupilumab doesn't deplete mast) |
| ALOX15 | 15-HETE lipidomics | universal | oral SM; epithelial effector arm |
| NTRK2 | BDNF | severity-scaling | neurotrophin/dysmotility axis; 13 existing TRK drugs |
| MSLN | serum mesothelin | subset | orthogonal biology; mature ADC toolbox |

## 5. Protein-therapeutic design specifications

Real UniProt sequences + AlphaFold v6 structures + ESM-2 analysis for 5 leads (POSTN structure/sequence
also fetched but not given a dedicated modality brief):
- **SIGLEC6**: depleting mAb/ADC targeting Ig-V domain (28-123); internalizing → ADC-compatible.
- **IL1RL1/ST2**: blocking mAb at IL-33 interface (Ig1-2, well-structured); astegolimab-class.
- **CCL26**: neutralizing mAb / CCR3 antagonist (small epitope).
- **ALOX15**: oral small-molecule active-site inhibitor (intracellular, not Ab-tractable; best structure pLDDT 95).
- **MSLN**: ADC (anetumab-class); avoid glycan epitopes; validate EoE surface expression.

## 6. Revised prioritized targets (Phase-2 integrated)

Top by integrated Phase-2 score:
1. **CCL26** (score 6) — canonical effector, universal, secreted, validated.
2. **SIGLEC6** (score 5) — confirmed novel, mast-specific, most homogeneous, depletion modality.
3. **IL1RL1/ST2** (score 4) — upstream alarmin, clinical precedent, companion biomarker.
4. **POSTN, ALOX15, TPSAB1, CPA3, CDH26, NTRK2** (score 4) — confirmatory, tractable, biomarker-paired.
5. **MSLN** (score 3) — novel/orthogonal, needs protein-level validation.

## Key caveats
- Eosinophils absent from droplet scRNA-seq (granulocyte dropout) — eosinophil-intrinsic biology inferred from bulk.
- PPI-refractory finding rests on a single paired cohort (GSE303169).
- MSLN addressable fraction not computed (absent from bulk panels at threshold); requires targeted validation.
- Endotypes derived from transcriptomics; prospective clinical linkage (histology, symptoms) not available in these datasets.
- SIGLEC6 marks a broad mast activation program, not one discrete subcluster — favorable for depletion but confirm sparing of protective mast functions.