# EoE Food-Trigger Diagnostic — Design Specification

**A personalized, antigen-specific T-cell assay to name the food(s) driving eosinophilic esophagitis from a single blood draw — replacing empiric elimination + serial endoscopy.**

*Computational / translational design built on the Phase 1–4 EoE platform. Hypothesis-generating; not a validated clinical test. See §9.*

---

## 1. The clinical problem

Identifying the food(s) that drive an individual's EoE is today done by **empiric elimination + repeat endoscopy**: remove a food (or the "six-food" group), scope 6–12 weeks later to check histologic remission, reintroduce, re-scope, and iterate. The process is slow (often a year+), burdensome, requires multiple sedated endoscopies, and is frequently inconclusive — reintroduction identifies a trigger only when a single food is removed and added back cleanly, and concurrent aeroallergen drive can confound the readout (Spergel 2012; Lucendo 2017; Furuta 2007). There is **no blood test that names the causal food.**

## 2. The scientific premise

EoE is a **food-antigen-driven, CD4+/Th2-mediated** disease (Phase 1–2 evidence: IFNγ+ T-cell infiltrate, Th2–mast correlation, elimination-diet remission; tissue Th2 cells in EoE, Wen 2019). The disease-driving lymphocytes are **allergen-specific, pathogenic effector Th2 (Th2A/CRTH2+CD161+) CD4+ T cells** (Mitson-Salazar 2017; Huang 2022). If we can **directly detect and quantify the food-antigen-specific Th2 CD4+ T cells** in a patient's blood, the food whose peptides elicit the strongest Th2 response *is* the driving antigen — read out in days, from one blood draw, without elimination or endoscopy.

> ### ⚑ Prior art & attribution — this is not a novel concept
>
> **The central idea of this assay was pioneered by David A. Hill, Jonathan M. Spergel, and colleagues, and it is patented.** This design should be read as building on their platform, not proposing it de novo.
> - **Cianferoni, Ruffner, Spergel et al. 2017** (*Ann Allergy Asthma Immunol*, doi:10.1016/j.anai.2017.11.006) first showed **elevated activated TH2 and milk-specific TH2 cells in the blood of milk-induced EoE patients** — establishing that circulating food-antigen-specific Th2 cells track the driving food.
> - **Dilollo, Spergel, Hill et al. 2025** (*JACI*, doi:10.1016/j.jaci.2025.01.008) identified and functionally validated **the first food-antigen-specific TCR in EoE**: `eoeTCR-4`, a clonally expanded peripheral-blood memory CD4+ cell with a **pathogenic effector-TH2 (peTH2)** signature, recognizing **β-casein (aa 59–78), HLA-DRB1\*07:01-restricted**, confirmed by TCR transduction into HLA-matched cells and **MHC-II tetramer** staining. Their pipeline (blood scRNA-seq + paired TCR-seq → peTH2 clonotype → tetramer validation) is the experimental realization of the workflow below.
> - **The authors disclose a patent on the use of functional T-cell assays to identify EoE-causal foods.** Any commercial development of the assay described here would need to engage that IP.
>
> What this design adds on top of that prior art is narrow and should be framed as such: a **personalized HLA-II-matched peptide-pool front end** (compute each patient's presentable cores per food, rather than whole-protein stimulation) and an explicit **aeroallergen comparator** arm. Both are incremental engineering choices on top of the Hill/Spergel concept — and their tetramer/TCR readout is a more definitive confirmation step than the AIM flow proposed here.

## 3. Assay architecture

**Input:** one peripheral blood draw (PBMC; optional esophageal biopsy for tissue-resident confirmation in equivocal cases).

**Step 1 — HLA class-II typing + personalized peptide pools.** Type the patient's HLA-II (as for the index case: DRB1, DQA1/DQB1, DPA1/DPB1). Computationally generate **per-food peptide pools** of the predicted strong-binding 15-mer cores presented on *that patient's own* MHC-II molecules (the personalized-panel engine, `patient_epitope_panel.csv`). One pool per candidate food; one or more **aeroallergen comparator pools** (e.g. dust mite, pollen) to separate food drive from environmental drive; plus positive (CEFX/anti-CD3) and negative (DMSO) controls.

**Step 2 — Antigen stimulation + AIM/cytokine readout.** Stimulate PBMC with each pool (~18–24 h for surface AIM; parallel well for intracellular cytokine). Read out by flow cytometry:
- **Activation-induced markers (AIM):** CD4+ OX40(CD134)+ CD25+, and/or CD4+ CD154(CD40L)+ — the established antigen-specific CD4 detection method, no prior knowledge of epitope-MHC needed and HLA-agnostic in execution (Poloni 2023; Grifoni 2020 methodology).
- **Th2 qualifier (the disease-relevant part):** intracellular IL-5 / IL-13 (± IL-4), and/or surface CRTH2+CD161+ Th2A phenotype on the AIM+ cells. This distinguishes a *pathogenic Th2* response (drives EoE) from a tolerized/Th1/Treg response to the same food.

**Step 3 — Call the trigger.** For each food pool compute the **Th2-AIM index** = background-subtracted % of CD4+ that are AIM+ *and* Th2-polarized. A food scoring above the positivity threshold (and above the aeroallergen comparator) is called a **driving food**; the assay names the **specific elimination to trial**. Optional confirmation: MHC-II tetramer/dextramer staining with the patient's dominant cores for the called food (Bégin 2015; Akdis 2015).

## 4. Why this beats the status quo

| | Empiric elimination + endoscopy | This assay |
|---|---|---|
| Time to answer | 6–12 months+ | days |
| Endoscopies | multiple, sedated | none (blood) |
| Confounding by aeroallergens | not separable | explicit comparator pools |
| Multi-food triggers | one-at-a-time, ambiguous | all foods tested in parallel |
| Output | "removing X helped" (indirect) | named antigen-specific Th2 response (direct, mechanistic) |

## 5. Worked example — the index case

Using the index-case patient's real personalized panel, the **presentation-based pre-test priority** (how many strong epitopes each food presents across the patient's 4 MHC-II molecules; `dx_peptide_pool_spec.csv`), with the milk pool completed to the full casein family (§5.1), is:

| Food pool | Strong epitopes | Proteins in pool | Ground truth |
|---|---|---|---|
| **Milk** (complete casein family) | **720** | 6 | **TRIGGER ✓** |
| **Soy** | 526 | 2 | **TRIGGER ✓** |
| Wheat | 204 | 1 | tolerated ✗ (false positive) |
| Fish | 49 | 1 | tolerated |
| Egg | 39 | 1 | tolerated |
| Peanut | 39 | 1 | tolerated |
| Tree nut | 37 | 1 | tolerated |
| Shellfish | 12 | 1 | tolerated |

This ranking sets **which pools to build and test first**. The **assay's actual call** comes from the functional Th2-AIM readout on the patient's cells (illustrated in the figure, panel C — *not measured here*): a food is named a trigger only if its pool elicits a Th2-polarized AIM+ CD4 response above threshold and above the aeroallergen comparator. The presentation panel is the *prior*; the functional assay is the *evidence*.

### 5.1 Calibration against this patient's blinded elimination-diet ground truth

The patient's true triggers were withheld during panel development and revealed only afterward: **milk and soy are the confirmed triggers; wheat and the rest of the Top-8 are tolerated** (the patient stays in histologic remission eating wheat). The patient has a **celiac family history but negative tTG-IgA** (no celiac disease). Calibrating the panel against this ground truth (full analysis: `index_case_groundtruth_calibration.md`, figure `index_case_groundtruth_calibration.png`):

- **Both confirmed triggers are the top-two foods** after correction (sensitivity 2/2 in the top-2); the five tolerated non-wheat foods all rank low (specificity 5/6). 
- **Milk was initially under-ranked (#3, 181) because my pool omitted β-casein.** Reading Dilollo/Spergel/Hill 2025 exposed this: their validated epitope is **β-casein aa59–78 / DRB1\*07:01**. Completing the casein family (β-, αS2-, κ-casein + α-lactalbumin) added **539 strong binders → milk to #1 (720)**. My mhcnuggets model independently calls the exact Hill β-casein aa59–78 window as a strong DR7 binder (best IC50 50 nM) — a fix *and* an external validation of the predictor.
- **Wheat is the one false positive, and it is the central lesson.** Wheat ranks #3 by *presentation* (88 gliadin binders on the patient's celiac-risk DQ2.2), yet the patient tolerates it. Presentation capacity ≠ pathology: the patient's DQ2.2 can display gluten, but no pathogenic effector-Th2 clone against it has expanded — exactly consistent with negative tTG-IgA despite celiac family history. **A binding-only test would wrongly send this patient into a gluten elimination; the functional Th2/peTH2 step is what prevents that.** The HLA finding is reported as *risk, not disease* — gluten-presentation capacity present, no evidence of gluten-driven pathology → monitor, do not eliminate.

## 6. Commercial / product form

- **Format:** CLIA/CAP reference-lab send-out. Blood tube → central lab. Turnaround target <10 days.
- **Reagent kit:** patient-specific synthesized peptide pools (generated on-demand from HLA type — the personalized engine is the moat), pre-titrated AIM/cytokine antibody panel, lyophilized controls.
- **Readout & report:** standardized flow panel + analysis pipeline; clinician report names foods above threshold with a confidence/effect-size per food and an explicit aeroallergen-vs-food statement.
- **Companion positioning:** (i) diagnostic to direct **targeted** (not empiric) elimination; (ii) response/monitoring biomarker (does the food-specific Th2 signal fall on elimination?); (iii) patient-selection + PD readout for the **pMHC anergy therapeutic** — the same peptide pools that diagnose the trigger define the therapeutic reagent, so diagnostic and therapy share one engine (theranostic).

## 7. Analytical & clinical validation plan

1. **Analytical:** reproducibility (intra/inter-assay CV), spike-in of expanded antigen-specific lines, LoD (minimum AIM+ frequency), cross-reactivity vs aeroallergen pools, HLA-coverage QC.
2. **Clinical concordance:** prospective EoE cohort with the assay run *blinded* against the gold standard (elimination + endoscopy-confirmed remission/relapse on reintroduction). Primary endpoint: sensitivity/specificity of the assay-named food vs the endoscopically-confirmed causal food. Include aeroallergen-driven and multi-food cases.
3. **Threshold locking:** set the Th2-AIM positivity cutoff on a training cohort, lock, validate on a held-out cohort.
4. **Special populations:** rare HLA types (predictor coverage), pediatric vs adult, PPI-treated vs active.

## 8. Key risks & mitigations

- **Peripheral blood may under-sample tissue-resident drivers.** Mitigate with the optional biopsy arm and by validating blood-vs-tissue concordance in the clinical study.
- **Sensitization ≠ causation** (a patient can have food-specific Th2 cells without that food driving EoE). This is exactly why the readout is **Th2-polarized effector** cells (not IgE, not any-cytokine) and why the clinical endpoint is endoscopic remission on the assay-named elimination.
- **Predicted pools may miss a true epitope.** AIM detects response to whole-protein or overlapping-peptide pools too; the personalized prediction concentrates the pool but a whole-protein-lysate backup pool de-risks coverage.
- **Aeroallergen / cross-reactive confounding.** Built-in comparator pools; PR-10/profilin cross-reactivity flagged in the report.

## 9. Scope and disclaimer

This is a **research/hackathon design specification**, not a validated diagnostic and not medical advice. Epitope pools rest on **computational MHC-II binding predictions** (mhcnuggets), and the functional readout in §5/figure-C is **illustrative, not measured** — no patient cells were assayed. Analytical and clinical validation (§7) is required before any clinical use. Test results and food-elimination decisions for any real patient must be made by qualified clinicians. Patient HLA/genotype data used here were user-provided for this exercise.

---

## Key references (DOI-verified, retrieved via OpenAlex this session)
- **Dilollo J, Spergel JM, Cerosaletti K, Hill DA et al. 2025, *J Allergy Clin Immunol* — a molecular basis for milk allergen immune recognition in EoE; first validated food-specific TCR (eoeTCR-4, β-casein aa59–78, DRB1\*07:01, tetramer-confirmed). doi:10.1016/j.jaci.2025.01.008** *(foundational prior art; patented platform)*
- **Cianferoni A, Ruffner MA, Spergel JM et al. 2017, *Ann Allergy Asthma Immunol* — elevated activated TH2 and milk-specific TH2 cells in milk-induced EoE. doi:10.1016/j.anai.2017.11.006** *(foundational prior art)*
- Burk CM & Hill DA 2026, *Ann Allergy Asthma Immunol* — precision medicine in EoE (review). doi:10.1016/j.anai.2026.04.025
- Poloni C et al. 2023, *Immunol Cell Biol* — T-cell activation-induced marker (AIM) assays in health and disease. doi:10.1111/imcb.12636
- Grifoni A et al. 2020, *Cell* — AIM-based CD4 target mapping (methodology). doi:10.1016/j.cell.2020.05.015
- Mitson-Salazar A et al. 2017, *Front Med* — pathogenic effector Th2 cells in allergic eosinophilic inflammation. doi:10.3389/fmed.2017.00165
- Huang Z et al. 2022, *Front Immunol* — Th2A cells as pathogenic players in allergic disease. doi:10.3389/fimmu.2022.916778
- Wen T et al. 2019, *J Clin Invest* — scRNA-seq of inflammatory tissue T cells in EoE. doi:10.1172/jci125917
- Spergel JM et al. 2012, *J Allergy Clin Immunol* — identifying causative foods in EoE. doi:10.1016/j.jaci.2012.05.021
- Lucendo AJ et al. 2017, *United European Gastroenterol J* — EoE evidence-based guidelines. doi:10.1177/2050640616689525
- Furuta GT et al. 2007, *Gastroenterology* — EoE systematic review / diagnostic standard. doi:10.1053/j.gastro.2007.08.017
- Bégin P et al. 2015, *J Allergy Clin Immunol* — peanut-specific T-cell clonotypes with immunotherapy (tetramer). doi:10.1016/j.jaci.2015.03.010
- Akdis CA & Akdis M 2015, *World Allergy Organ J* — mechanisms of allergen-specific tolerance. doi:10.1186/s40413-015-0063-2