# De Novo Neutralizing Binders for CCL26 and POSTN in Eosinophilic Esophagitis
### Computational protein-design deliverable — Built with Claude: Life Sciences Hackathon

## 1. Summary

Two de novo protein binders were designed, one against each of two EoE targets that
sit at distinct points in the disease cascade:

- **CCL26 (eotaxin-3)** — the canonical Th2 chemokine that recruits eosinophils via CCR3.
  The binder is designed as an **anti-ligand trap** that occludes the CCR3/glycosaminoglycan-
  binding face, blocking chemokine presentation and eosinophil recruitment.
- **POSTN (periostin, FAS1-IV domain)** — a matricellular remodeling protein that remains
  **elevated in histologic remission** (PMID 39343172). The binder targets the integrin-
  interaction patch as an **anti-remodeling agent** aimed at the barrier-dysfunction loop
  that persists after inflammation resolves.

Both leads passed structural fold-back validation with high interface confidence and
verified on-target epitope engagement.

| | CCL26 lead | POSTN lead |
|---|---|---|
| Design ID | `ccl26_design_39_T0.3` | `postn_design_39_T0.1` |
| Binder length | 67 aa | 81 aa |
| **Interface ipTM** | **0.942** | **0.901** |
| Complex pLDDT | 0.779 | 0.911 |
| Buried surface | 1220.2 Ų | 842.2 Ų |
| Epitope coverage | 4/5 | 3/4 |
| On-target score | 0.848 | 0.788 |

## 2. Target selection rationale

Targets emerged from an integrated literature + omics analysis (see `eoe_target_report.md`,
`eoe_crossref_report.md`). Both are concordant across a 9-cohort EoE meta-analysis and the
literature:

- **CCL26**: meta-analysis pooled log2FC **+4.56** (padj 3e-04), upregulated in 9/9 cohorts;
  single-cell cell-of-origin = suprabasal epithelium. **Zero** interventional trials directly
  target the eotaxin-3/CCR3 axis — an open therapeutic lane. CCL26 is IL-4/IL-13-induced and
  is the single most discriminating EoE-vs-GERD transcript (PMID 17900656).
- **POSTN**: meta-analysis pooled log2FC **+5.55** (padj 3e-16), upregulated in 8/8 cohorts;
  ranks 3rd of 20 on the omics druggability shortlist; single-cell origin = basal epithelium.
  Persistently elevated in remission alongside DSG1 loss — the barrier axis that does not
  normalize when inflammation is controlled.

Full mechanistic rationale, including the barrier-restoration hypothesis, is in
`eoe_design_rationale.md`.

## 3. Design pipeline

```
Target structure (PDB) → RFdiffusion backbones → SolubleMPNN sequences → Boltz-2 fold-back → on-target selection
```

- **Targets:** CCL26 = 1G2S (mature eotaxin-3, 71 aa NMR); POSTN = 5WT7 (FAS1 domain IV, 140 aa NMR).
- **Hotspots:** CCL26 = N-loop H16/K17 + C-helix basic cluster R54/K55/K56 (CCR3+GAG face);
  POSTN = integrin patch 83/85/108/116 on FAS1-IV.
- **Backbones:** RFdiffusion (Complex_base checkpoint), 40 per target, binder length 65–90 aa.
  All 40 CCL26 and 34/40 POSTN backbones engaged the hotspot epitope.
- **Sequences:** SolubleMPNN (v_48_020), 8 sequences × 3 temperatures × 80 backbones = 1,920 total;
  30 per target selected for fold-back after a sequence-complexity filter.
- **Validation:** Boltz-2 co-folding (3 diffusion samples, MSA server), then interface-epitope
  overlap analysis to confirm on-target binding.

## 4. Fold-back validation results

Of 30 designs folded per target: **CCL26 30/30** and **POSTN 29/30** pass the community
interface thresholds (ipTM > 0.5, complex pLDDT > 0.7). Leads were chosen by a composite
**on-target score** = ipTM × epitope-coverage × pLDDT, which rewards designs that are both
confident *and* bound at the intended functional epitope rather than an arbitrary surface.

![Fold-back interface confidence and on-target ranking]({{artifact:3cefbf9d-53fe-49a8-a2dc-47a74782ac54}})

*Left: interface ipTM vs complex pLDDT for all 60 designs, marker size ∝ number of epitope
residues contacted; dashed lines are the pass thresholds; leads circled. Right: top designs
ranked by on-target score.*

Target epitope definition (per-residue solvent accessibility, hotspots highlighted):

![Target epitopes]({{artifact:743ca3f0-7a3d-421a-8409-23e7a9fee8b1}})

## 5. Lead binders

### CCL26 anti-ligand trap — `ccl26_design_39_T0.3`
Engages 4/5 hotspots (H16 N-loop + R54/K55/K56 C-helix), the surface that mediates both
CCR3 engagement and GAG-mediated chemokine presentation (PMID 35742962). Buried surface
1220.2 Ų across 25 target and 27
binder interface residues. The binder is acidic/helical — electrostatically complementary to
CCL26's basic GAG-binding face.

```
SEQLKEEEEKKKLEEERREEEEEAREEEASRKAAAQRAAVDPEGANKELEELERARKEAEELYKKAR
```

3D rotation video: **[CCL26_binder_complex_360.mp4]({{artifact:73de383b-11aa-40ec-a02d-c58734120f09}})**
· Complex structure: **[CCL26_lead_complex.pdb]({{artifact:f860127d-bf1e-4865-8659-6ef1fb767638}})**

### POSTN anti-remodeling binder — `postn_design_39_T0.1`
Engages 3/4 hotspots (83/85/108) on the FAS1-IV integrin-interaction patch. Buried surface
842.2 Ų, complex pLDDT 0.911 (highest-confidence fold of
either lead).

```
MTKEQNEAEAEAMKLMLEARRLAAEGAAGPGVEELTKKALEKLKEAIEKAKKTGKERREKLIKQAKDLEEDAKARAAAKLA
```

3D rotation video: **[POSTN_binder_complex_360.mp4]({{artifact:730b3184-3d6a-4910-a0cd-d48f1aebaf66}})**
· Complex structure: **[POSTN_lead_complex.pdb]({{artifact:6d07b628-dabb-4230-a7bb-4b273a8dd7fa}})**

## 6. Caveats and next steps

**These are in-silico designs.** ipTM and pLDDT are model confidences, not measured affinities.
Required wet-lab validation: recombinant expression, SEC + SPR/BLI binding assays, and functional
readouts (CCL26→CCR3 chemotaxis inhibition; periostin-integrin / remodeling assays).

- Both leads are helical and charged-rich (a SolubleMPNN compositional bias on these backbones).
  This favors solubility but warrants a specificity screen and, ideally, a second design round
  with tightened composition constraints.
- The POSTN binder targets one isolated FAS1 domain; full-length periostin has four FAS1 domains,
  so this is a domain-level starting point, not a validated whole-protein neutralizer.
- The POSTN → barrier-restoration link is **mechanistically indirect** (interrupting the
  IL-13→DSG1↓→POSTN↑ feed-forward loop), not direct desmosome reassembly. Its strongest support
  is the persistence of periostin + DSG1 loss into histologic remission.

## 7. Artifact index
- `eoe_binder_validation.csv` — all 60 folded designs with full metrics
- `mpnn_scored.csv` — all 1,920 SolubleMPNN sequences with scores
- `binder_backbones.tar.gz` — 80 RFdiffusion backbones (checkpoint)
- `lead_spec_sheets.md` — per-binder spec sheets
- `CCL26_lead_complex.pdb`, `POSTN_lead_complex.pdb` — lead complex structures
- `CCL26_binder_complex_360.mp4`, `POSTN_binder_complex_360.mp4` — rotation videos
- `foldback_metrics.png`, `target_epitopes.png` — figures
- `eoe_design_rationale.md`, `eoe_target_report.md`, `eoe_crossref_report.md` — supporting analyses