# Eosinophilic Esophagitis — Phase 1 Target Discovery Report
### Exhaustive mining of public omics datasets

**Project:** Built with Claude Life Sciences Hackathon — EoE therapeutic target discovery
**Scope:** Systematic integration of public transcriptomic, single-cell, miRNA, and druggability data to nominate novel, tractable therapeutic targets for protein-therapeutic design.

---

## Executive summary

We assembled and mined an exhaustive inventory of **57 public EoE datasets** across GEO, ArrayExpress, PRIDE, and CELLxGENE. From **9 harmonized case-control bulk transcriptomic studies (235 samples: 152 EoE, 83 control)** we derived a reproducible cross-study meta-signature via random-effects meta-analysis, validated it at single-cell resolution in **166,420 esophageal cells (GSE218607)**, corroborated it with an independent miRNA layer, and annotated every candidate for therapeutic tractability using Open Targets.

The result is a **prioritized shortlist of therapeutic targets** dominated by secreted and cell-surface proteins — ideal substrates for antibody and engineered-protein therapeutics.

**Key findings:**
- A **567-gene high-confidence EoE signature** (383 up, 184 down), reproducible across ≥6 studies each, recovers all canonical EoE markers and is dominated by an **interferon + IL-6/JAK-STAT + NF-κB inflammatory program** superimposed on **loss of epithelial keratinization/barrier**.
- Upstream master regulators: **IRF1, STAT1/2, IRF8, RELA/NFKB1** — the IFN-JAK/STAT and NF-κB axes (JAK-druggable).
- Single-cell analysis reveals **13-fold mast-cell expansion** (0.2%→2.6%, p=1e-4), basal epithelial depletion, and **epithelial induction of IFN/antigen-presentation genes + the EoE GWAS gene CAPN14**.
- Dominant cell-cell communication: **S100A8/A9 alarmins** (epithelium/fibroblast → mast/myeloid) and mast-cell signaling to epithelium.
- **24 of the top 25 targets are secreted or surface-accessible; 9 have existing drugs/clinical candidates.**

---

## 1. Dataset landscape

Exhaustive inventory (n=57; 56 EoE-relevant after excluding GSE168883, an environmental-enteropathy study):

| Assay | Datasets |
|---|---|
| Bulk RNA-seq | 26 |
| Single-cell RNA-seq | 9 |
| miRNA / ncRNA | 7 |
| Microarray | 6 |
| Proteomics (PRIDE) | 5 |
| Methylation | 2 |
| Other | 1 |

Tiering: **Tier 1** = 12 human esophageal case-control bulk/array (primary DE); **Tier 2** = 6 single-cell resources; **Tier 3** = 38 complementary/functional (miRNA, methylation, proteomics, blood, in-vitro/mouse).

---

## 2. Harmonized bulk transcriptomics & QC

Nine Tier-1 datasets were retrieved, ID-harmonized to gene symbols, log-normalized, and QC'd (235 samples). Per-study PCA separated EoE from control in 6/9; the canonical diagnostic panel (CCL26, POSTN, ALOX15, CDH26, CPA3, TNFAIP6 up; DSG1, SPINK7, CRISP3 down) was recovered with consistent direction, validating labels and orientation. FLG showed heterogeneous behavior (up in one recurrence cohort), reflected in its high meta-analysis heterogeneity.

Per-study differential expression (moderated-t, BH-FDR):

| gene | genes_tested | sig_FDR05 | up | down |
|---|---|---|---|---|
| **GSE250595** | 22763 | 8178 | 6909 | 1269 |
| **GSE148381** | 29797 | 3582 | 3047 | 535 |
| **GSE278888** | 22103 | 0 | 0 | 0 |
| **GSE197702** | 19652 | 4985 | 4331 | 654 |
| **GSE303169** | 32640 | 3330 | 1166 | 2164 |
| **GSE234973** | 19471 | 24 | 20 | 4 |
| **GSE246323** | 21766 | 0 | 0 | 0 |
| **GSE58640** | 20477 | 6630 | 5691 | 939 |
| **GSE228083** | 10103 | 88 | 16 | 72 |


---

## 3. Cross-study meta-analysis

DerSimonian-Laird random-effects pooling across 25,654 genes (≥3 studies): **8,102 significant at FDR<0.05; 4,788 fully concordant** in direction. The **high-confidence signature (567 genes)** requires |pooled log2FC|≥1, significance, and concordance in ≥6 studies.

Top up-regulated (by significance): IFI27, SIDT1, ACSL5, ANO1, CDH26, LITAF, SCUBE2, NTRK2, IL15RA, OAS2.
Top down-regulated: CRYAB, CCNYL1, and the keratinization/barrier module.

All canonical markers significant with correct direction (CCL26, POSTN, ALOX15, CDH26 up; DSG1, CRISP3, SPINK7 down), though with high inter-study heterogeneity (I² 76–98%) — expected for magnitude-variable disease markers.

---

## 4. Single-cell resolution (GSE218607, 166,420 cells)

10x scRNA-seq of paired proximal/distal esophageal biopsies (12 Healthy + 11 EoE samples), Harmony-integrated by subject, 25 Leiden clusters → 7 cell types.

**Composition shifts (per-sample fraction, EoE vs Healthy):**

| Cell type | Healthy % | EoE % | MWU p |
|---|---|---|---|
| Mast | 0.2 | 2.6 | 0.0001 |
| Epithelial basal | 42.7 | 30.6 | 0.001 |
| Endothelial | 1.5 | 0.4 | 0.0003 |
| Epithelial suprabasal | 44.5 | 52.7 | 0.04 |
| T cell | 7.3 | 9.8 | 0.37 |

**Cell-type-resolved DE:** epithelium up-regulates IFN/antigen-presentation (IFI27, HLA-A/B/C, B2M) and the **EoE GWAS risk gene CAPN14**; mast cells express CPA3, CTSG, SIGLEC6. Eosinophils were not captured as a discrete cluster — a known limitation of droplet scRNA-seq for granulocytes.

---

## 5. Pathway, regulator & cell-cell communication

- **UP:** Interferon α/β/γ response (top, padj≈1e-28), cytokine signaling, complement, IL-6/JAK/STAT3, EMT/ECM organization.
- **DOWN:** Keratinization / cornified-envelope formation, epidermal differentiation, metallothionein / zinc response — the barrier-dysfunction signature.
- **Upstream TF regulators:** IRF1, STAT1, STAT2, IRF8, RELA, NFKB1.
- **Cell-cell communication (liana, 73k EoE cells):** S100A8/A9 alarmins (epithelium/fibroblast → mast via CD69, → myeloid via ITGB2); mast VIM–CD44 → epithelium; MHC-I antigen presentation among immune cells.

---

## 6. Multi-omics integration & miRNA cross-check

A multi-modal evidence table integrates bulk meta-analysis, cross-study reproducibility, and single-cell cell-type DE: **58 signature genes have both bulk-meta and single-cell confirmation** with cell-type localization. The miRNA layer (GSE36727, paired pre/post-steroid, n=5) is underpowered (no FDR-significant hits) but recovers canonical **miR-21↑, miR-223↑, miR-375↓** with literature-correct direction — confirmatory of the inflammatory axis.

---

## 7. Prioritized therapeutic targets

159 candidates were annotated via Open Targets (tractability, known drugs, subcellular localization, target class) and scored on a composite of effect size, reproducibility, single-cell confirmation, therapeutic accessibility (secreted > surface), antibody-tractability, clinical precedent, and existing pharmacology.

### Tier-A targets (curated for biological specificity + accessibility)

| gene | pooled_log2FC | padj | k | accessibility | n_drugs | drugs |
|---|---|---|---|---|---|---|
| **CCL26** | 4.56 | 0.00 | 9 | secreted | 0 | — |
| **IL1RL1** | 2.06 | 0.00 | 8 | secreted | 1 | ASTEGOLIMAB |
| **POSTN** | 5.55 | 0.00 | 8 | secreted | 0 | — |
| **SIGLEC6** | 2.01 | 0.00 | 8 | secreted | 0 | — |
| **TPSAB1** | 2.76 | 0.00 | 9 | secreted | 0 | — |
| **ALOX15** | 5.92 | 0.00 | 9 | surface | 0 | — |
| **CPA3** | 3.28 | 0.00 | 9 | secreted | 0 | — |
| **MSLN** | 3.00 | 0.00 | 8 | secreted | 5 | AMATUXIMAB;ANETUMAB RAVTANSINE;LMB- |
| **CDH26** | 4.98 | 0.00 | 8 | surface | 0 | — |
| **NTRK2** | 3.13 | 0.00 | 8 | surface | 13 | ALTIRATINIB;AZD-6918;AZD-7451;CENEG |


**Mechanistic rationale & suggested modality:**

- **CCL26 (eotaxin-3)** — most EoE-specific gene; master eosinophil chemoattractant. *Neutralizing Ab / chemokine trap.*
- **IL1RL1 (ST2)** — IL-33 alarmin receptor; astegolimab already in trials. *Blocking Ab / ST2-Fc decoy.*
- **POSTN (periostin)** — Th2-induced ECM/fibrosis effector. *Neutralizing Ab.*
- **SIGLEC6** — mast-cell-restricted inhibitory receptor. *Agonist Ab / ADC to deplete mast cells.*
- **TPSAB1 (tryptase) / CPA3** — mast-cell proteases driving tissue damage. *Neutralizing Ab (anti-tryptase precedent).*
- **ALOX15 (15-LOX)** — eicosanoid mediator. *Small molecule (intracellular enzyme).*
- **MSLN (mesothelin)** — secreted/surface; validated ADC target (anetumab). *ADC / CAR.*
- **CDH26 (cadherin-26)** — EoE-enriched epithelial immunomodulatory cadherin. *Surface Ab.*
- **NTRK2 (TrkB)** — epithelial; rich TRK-inhibitor pharmacology. *Small molecule.*

The broader top-20 shortlist is in `eoe_target_shortlist.csv`.

**Caveat:** a few high-scoring genes (IFI27, LITAF, SCIN, SIDT1) receive a "surface" label from broad membrane annotation but are not established antibody targets; HLA-A is surface-accessible but ubiquitously expressed. These require manual triage before design.

---

## 8. Recommendations for protein-therapeutic design (phase 2)

1. **Lead antibody/trap candidates (secreted):** CCL26, POSTN, IL1RL1/ST2, TPSAB1, MSLN — accessible to neutralizing biologics; several with clinical precedent.
2. **Mast-cell-directed:** SIGLEC6 (depleting Ab/ADC), CPA3/TPSAB1 (protease neutralization) — targets the most disease-specific cell-composition change.
3. **Epithelial surface:** CDH26 — EoE-restricted, under-explored.
4. **Structure/design pipeline:** feed Tier-A secreted/surface targets into ESM-based sequence analysis and structure prediction; prioritize CCL26 and IL1RL1/ST2 for binder design given clean secreted biology and clinical validation.

---

## Artifacts produced

Inventory & tiering, harmonized bundle, per-study + meta DE, single-cell object + composition/DE, pathway/regulator/communication tables, multi-omics evidence, druggability annotation, and this report. See project artifact tray.