# Tier-2 Gut-Restricted Blocker Design — NKG2D and IL-15

### Structure-guided, interface-fixed scaffold redesign against the two apex nodes of the villous-destruction axis

**Goal.** Design protein blockers for the two highest-value Tier-2 targets from the omics analysis — the IL-15/NKG2D-MIC killing machine that directly executes villous destruction — in a *gut-restricted* format aligned with the celiac community's preference for low-systemic-side-effect therapy.

---

## 1. Design rationale (from the omics evidence)

The single-cell and meta-analysis placed two nodes at the center of enterocyte killing:
- **NKG2D (KLRK1)** — the activating receptor on cytotoxic IELs/NK that engages stress ligands **MIC-A/B** (MICB up +0.70, padj ~0 in the meta-signature) on enterocytes to trigger direct epithelial lysis.
- **IL-15** — the apex cytokine (stromally trans-presented, single-cell) that drives IEL survival, NKG2D induction, and cytotoxicity; the AMG-714 (anti-IL-15) target and the leading refractory-celiac strategy.

Both act at the **terminal effector step** — blocking either protects the villi directly, and neither requires systemic immunosuppression if delivered in a gut-restricted format.

---

## 2. Design strategy (matched to available compute)

With no GPU/RFdiffusion available, the campaign used **structure-guided, interface-fixed scaffold redesign** (ProteinMPNN/SolubleMPNN, CPU) rather than de novo backbone generation — a tool-matched approach that produces stable, expressible sequences on validated natural scaffolds:

| Design | Target | Scaffold (PDB) | Interface fixed | Mechanism |
|--------|--------|----------------|-----------------|-----------|
| **MICA-decoy** | NKG2D (KLRK1) | MICA α1α2 (1HYR) | 29 NKG2D-contact residues | Soluble decoy sequesters NKG2D away from enterocyte MIC-A/B |
| **IL-15 binder** | IL-15 | IL-2Rβ D1 (4GS7) | 12 IL-15-contact residues | Binds IL-15 β/γc signaling face (spares IL-15Rα trans-presentation), neutralizes IEL survival signal |

The interfaces were mapped at 5 Å from the solved complexes; contact residues were held fixed while the non-interface surface was redesigned with **SolubleMPNN** (soluble-biased weights) at temperatures 0.1–0.2.

---

## 3. Results

**48 designs per target, all with the functional interface 100% preserved.**

- **MICA-decoy:** best ProteinMPNN score **0.988** (48 designs); ~150 surface mutations per design; net charge shifted −3→−10 and free cysteines 7→4 (solubility-favorable, aggregation-reducing) while the 29-residue NKG2D-binding face is native.
- **IL-15 binder:** best score **0.889** (48 designs); net charge −3→−8, cysteines 8→3; the 12-residue IL-15-contact face preserved.

Both retain hydrophobic-core fraction (~0.30–0.34) — the redesign optimized the surface, not the fold core.

---

## 4. Honest validation status and next steps

**What is established:** the designs preserve the exact functional interface and carry solubility-favorable physicochemistry on experimentally-validated binding scaffolds. This is a strong *sequence-design* result.

**What is NOT yet validated (requires GPU):** actual 3D fold and binding-mode confirmation. The recommended next steps, once GPU/remote compute is available:
1. **Fold each design** with ESMFold2-Fast / Boltz-2 / Chai-1 and confirm the scaffold fold (pLDDT) and that the fixed interface is presented correctly.
2. **Co-fold the complex** (decoy·NKG2D; binder·IL-15) and filter on interface confidence (ipTM) — the true binding test.
3. **Affinity maturation** via RFdiffusion partial diffusion + a second ProteinMPNN round on the top folds.
4. **Gut-restriction engineering:** fuse to a protease-resistant/mucus-anchoring domain, or reformat as an oral biologic, to realize the tolerability advantage.

**Design caveats:** (1) the MICA-decoy is a soluble-protein reformat of a natural ligand — expression and stability of the isolated α1α2 platform need experimental confirmation; (2) the IL-2Rβ scaffold carries 10 unmodeled (X) residues from the crystal structure that MPNN left unassigned — these gaps should be resolved before synthesis; (3) IL-15 blockade competes with an existing antibody program — the novelty is the gut-restricted small-scaffold format, not the target biology.

---

## 5. Deliverables

| Artifact | Contents |
|----------|----------|
| celiac_design_spec.json | Both blocking interfaces (residue-level) from 1HYR & 4GS7 |
| mica_decoy_designs.csv | 48 NKG2D-decoy sequences + scores |
| il15_binder_designs.csv | 48 IL-15-binder sequences + scores |
| celiac_tier2_best_designs.fasta | Top design per target |
| celiac_design_summary.csv | Campaign summary table |
| celiac_design_campaign.png | Score/recovery distributions + mechanism schematic |
| 1HYR.pdb, 4GS7.pdb | Target complex structures |