# EoE Target Landscape — Literature × Pipeline Sweep

*Batch mining of PubMed (591 abstracts with text, 2015–2026, from 1,688 hits) and
ClinicalTrials.gov (157 interventional EoE trials). Structured extraction on 459
abstracts via Claude Haiku batch (~$2.23 in API credit). This is a hypothesis-generating
map, not a validated ranking — every candidate needs confirmation against the omics
data (Stream 1) and primary literature.*

## How to read this

- **Literature evidence** = number of abstracts naming a target as dysregulated/mechanistic in EoE.
- **up/down** = count of abstracts stating the target is up- or down-regulated in EoE tissue.
- **Clinical maturity** = furthest trial phase of any agent hitting that target in EoE.
- **Opportunity score** = log-scaled literature evidence (weighted toward upregulated targets)
  minus clinical maturity. High score = strong, consistent biology that the clinical pipeline
  has *not yet* exploited. It is a triage heuristic, not an endpoint.

## Top opportunities — strong biology, thin/no pipeline

| Rank | Target | Abstracts | Up / Down | Trials (max phase) | Opportunity |
|---|---|---|---|---|---|
| 1 | *CCL26* | 45 | 22/4 | 0 (—) | 6.95 |
| 2 | *IL4* | 34 | 19/3 | 0 (—) | 6.49 |
| 3 | *IL13* | 110 | 59/9 | 3 (PHASE2) | 6.35 |
| 4 | *TSLP* | 61 | 33/3 | 1 (PHASE3) | 5.61 |
| 5 | *CAPN14* | 14 | 5/0 | 0 (—) | 4.96 |
| 6 | *IL33* | 11 | 10/0 | 0 (—) | 4.74 |
| 7 | *IL5* | 72 | 38/6 | 6 (PHASE3) | 4.47 |
| 8 | *POSTN* | 9 | 6/2 | 0 (—) | 3.84 |
| 9 | *STAT6* | 7 | 2/0 | 0 (—) | 3.47 |
| 10 | *CCL11* | 5 | 4/0 | 0 (—) | 3.23 |
| 11 | *DSG1* | 10 | 2/5 | 0 (—) | 3.0 |
| 12 | *IL1B* | 5 | 4/1 | 0 (—) | 2.99 |
| 13 | *IL18* | 4 | 4/0 | 0 (—) | 2.9 |
| 14 | *IL15* | 5 | 3/1 | 0 (—) | 2.87 |
| 15 | *IFNG* | 4 | 3/0 | 0 (—) | 2.82 |

## What the map shows

**The validated core is crowded.** IL-13, IL-4Rα, IL-5/IL-5Rα, TSLP and Siglec-8 carry
both the heaviest literature evidence *and* the densest trials — dupilumab (IL-4Rα, approved),
cendakimab (IL-13, Ph3), benralizumab/mepolizumab (IL-5 axis), tezepelumab (TSLP),
lirentelimab (Siglec-8). Competing here means competing with approved or late-stage agents.

**The opportunity sits in the lower-right of the landscape** — targets with strong,
directionally-consistent EoE biology but *no* dedicated therapeutic program:

- **CCL26 (eotaxin-3)** — the single most consistently upregulated chemokine in EoE epithelium;
  the CCR3 axis it signals through has no approved EoE agent. Ligand trap / anti-CCR3 territory.
- **IL-33 / (alarmin axis)** — 10/10 abstracts upregulated, epithelial-alarmin upstream of Th2;
  no EoE-specific trial in this set despite validated anti-IL33 biology in adjacent allergy.
- **CAPN14** — the EoE GWAS gene, esophagus-specific, induced by IL-13; genetically anchored
  and completely unexploited pharmacologically.
- **POSTN (periostin), STAT6, CCL11** — remodeling / signaling nodes with clean upregulation
  and no dedicated program.
- **DSG1, FLG, SPINK7** — barrier genes, consistently *down* in EoE; restoration (not blockade)
  logic — a different therapeutic modality worth flagging for Stream 3.

**Cell-type context:** eosinophil (338), epithelial
(136), Th2 (69), mast cell
(39) dominate the implicated populations — consistent with an
epithelial-alarmin → Th2 → effector-eosinophil/mast-cell axis.

## Most-studied drugs in the corpus
dupilumab (40), budesonide (13), fluticasone (10), mepolizumab (9), reslizumab (7), benralizumab (7), omalizumab (5), tezepelumab (4), cendakimab (4), qax576 (3), esomeprazole (3), omeprazole (3)

## Suggested next moves for the hackathon
1. **Cross-check against Stream 1 omics.** Confirm CCL26, IL33, CAPN14, POSTN directionality in
   your own DE / single-cell data before committing design effort.
2. **Pick 1–2 unexploited targets for Stream 3 protein design.** CCL26 (secreted chemokine — a
   binder/trap is tractable) and IL-33 (alarmin — neutralizing binder) are the cleanest
   "strong biology + designable + open pipeline" candidates.
3. **Barrier-restoration angle (DSG1/FLG/SPINK7)** is mechanistically distinct from every
   trial in the pipeline — higher-risk, higher-novelty.

**Important reading note on IL-4 / IL-13:** these rank high because agents are mapped to the
*receptor* (IL-4Rα → dupilumab), so the ligands appear "untargeted." In reality dupilumab
blocks IL-4 **and** IL-13 signaling through IL-4Rα, and cendakimab targets IL-13 directly.
These two are *not* unexploited — they are the validated core. The genuinely open targets are
the no-trial rows: CCL26, IL-33, CAPN14, POSTN, STAT6, CCL11, and the barrier genes.

*Caveats: extraction covers 78% of the abstract corpus (per-frame token ceiling truncated the
tail); target frequency reflects publication attention, which is biased toward already-popular
targets — the opportunity score partly corrects for this but does not eliminate it. Drug→target
mapping is a curated dictionary and may miss newer agents. Treat as a triage layer.*