# EoE T-cell Reprogramming — Dual-Program Target Brief

**Source data:** Genome-scale CRISPRi Perturb-seq in primary human CD4⁺ T cells (Zhu, Dann et al. 2025, *bioRxiv*; CZI Virtual Cells Platform, `genome-scale-tcell-perturb-seq`). Differential-expression resource `GWCD4i.DE_stats.h5ad`: **33,983 perturbation × culture-condition profiles × 10,282 measured genes**, with per-gene log₂FC / z-score / FDR layers (DESeq2), across three conditions (Rest, Stim8hr, Stim48hr).

**Prepared for:** Therapeutic Program Specialist — to develop two reprogramming programs side by side.

---

## 0. Scope and honest caveats (read first)

This analysis was **reframed** after an initial mismatch was caught. Three points bound how these results should be used:

1. **CCL26 and POSTN are not in this dataset — by design, not omission.** Both are the canonical EoE effector genes, but CCL26/eotaxin-3 is produced by esophageal **epithelium** and POSTN/periostin by **fibroblasts**, each induced by IL-13 from Th2 cells. Neither is expressed in CD4⁺ T cells, so neither is measured or perturbed here. What this dataset *does* contain is the **upstream T-cell regulatory wiring** of the Th2 cytokines (IL13, IL5, IL4) that drive the CCL26/POSTN axis. We target the T-cell signal, not the epithelial/fibroblast effectors.

2. **This is a healthy-donor, in-vitro screen — not an EoE phenotype.** Cells are peripheral CD4⁺ T cells under generic anti-CD3/CD28 stimulation. There is no allergen, no esophageal microenvironment, no patient cell. Every hit here is a **hypothesis about a regulatory relationship**, to be validated in a disease-relevant system. IL13/IL5/IL4 are readouts of a polyclonal activation program, not of a tissue-resident pathogenic clone.

3. **This is the state axis, not the specificity axis.** These programs modulate T-cell *state*. They are orthogonal to — and do **not** substitute for — the pMHC platform's antigen/epitope/HLA specificity work, which is handled by the separate antigen-epitope pipeline. Program B is explicitly scoped as *combination* biology: co-targets that make a cognate clone more amenable to pMHC-induced tolerance.

---

## 1. The two programs

| | **Program A — State modulation** | **Program B — pMHC-tolerance amenability** |
|---|---|---|
| **Question** | Which T-cell knockdowns *dampen* pathogenic Th2 output? | Which knockdowns make an effector clone *more convertible* to a tolerogenic/anergic fate? |
| **Mechanism** | Reduce IL13/IL5/IL4 (the CCL26/POSTN drivers) | Induce TR1/iTreg/anergy programs + release the Th2 state-lock |
| **Readouts (↓ desired)** | IL13, IL5, IL4, IL4R, IL17RB, CCR4/8; Th2 master-TFs GATA3/STAT6/IRF4/BATF | — |
| **Readouts (↑ desired)** | — | IL10, LAG3, HAVCR2, CTLA4, PDCD1, ENTPD1 (TR1); FOXP3, IL2RA, IKZF2 (iTreg); EGR2/3, NR4A1-3, CBLB, DGKA, TOX (anergy) |
| **Modality use** | Standalone drug / biologic to blunt Th2 program | **Adjunct** to an anergy/iTreg pMHC construct |
| **Candidates found** | **122** selective, reproducible dampeners | **320** co-targets (**151** stim-context-relevant) |

**Screen quality (QC):** ~11,300 gene perturbations per condition; median ~540 cells/perturbation; on-target knockdown significant in 61–64% of perturbations at a median ~6 log₂FC reduction. All 47 readout-panel genes are both measured and perturbed. A well-powered screen.

![Readout coverage]({{artifact:art_fdea887f-d53a-4915-8a00-9406397e45a7}})

---

## 2. Program A — dampen pathogenic Th2 output

**Method.** For each perturbation, a Th2-dampening score = mean(−z) over IL13/IL5/IL4 (self-knockdown excluded). Required: ≥2/3 core cytokines significantly down (FDR<0.1), effective on-target knockdown, and positive **Th2-selectivity** (dampening beyond any IFNG/global-shutdown collateral). Because IL13/IL5/IL4 are stimulation-induced (IL13 baseMean 9 → 452 from Rest → Stim48hr), scoring is on the Stim conditions. Composite = 0.5·magnitude + 0.3·selectivity + 0.2·cross-donor reproducibility.

**Top targets (of 122):**

| Rank | Target | Score | Th2 damp | Best cond | Selectivity | Donor r |
|---|---|---|---|---|---|---|
| 1 | **BHLHE40** | 0.98 | 6.05 | Stim48hr | 5.20 | 0.84 |
| 2 | **GATA3** | 0.83 | 5.42 | Stim8hr | 3.87 | 0.84 |
| 3 | **IL2RB** | 0.73 | 5.21 | Stim48hr | 2.68 | 0.81 |
| 4 | TMEM87B | 0.71 | 4.05 | Stim8hr | 4.05 | 0.93 |
| 5 | CFAP20 | 0.70 | 4.83 | Stim48hr | 2.88 | 0.81 |
| 6 | WAC | 0.70 | 4.69 | Stim48hr | 3.04 | 0.83 |
| 7 | ENO1 | 0.70 | 4.25 | Stim48hr | 3.50 | 0.91 |
| 8 | **SMAD4** | 0.68 | 4.37 | Stim8hr | 3.00 | 0.89 |
| 9 | HEXIM1 | 0.67 | 3.91 | Stim8hr | 3.91 | 0.87 |
| 10 | **STAT5A** | 0.66 | 3.93 | Stim8hr | 3.93 | 0.80 |
| 12 | **IL4R** | 0.66 | 3.81 | Stim48hr | 3.81 | 0.86 |

**Biological read.** The ranking recovers known Th2 biology as internal validation: **GATA3** (master Th2 TF) is the #2 hit; **STAT5A** and **IL2RB** are the IL-2→STAT5 axis that licenses Th2 differentiation; **SMAD4** and **IL4R** are established Th2 signaling nodes; **BHLHE40** is a validated cytokine-program regulator. IL4R is the target of dupilumab, an approved EoE therapeutic — its appearance here is a strong positive control. Full table: `programA_th2_dampening_targets.csv`.

![Program A dampeners]({{artifact:art_a542d424-ec90-4821-a1e4-6c4496de7bcd}})

---

## 3. Program B — co-targets for pMHC-induced tolerance

**Rationale.** A tolerogenic pMHC construct (anergy- or iTreg-inducing) can only re-tolerize a cognate clone that is *convertible*. An epigenetically committed Th2 effector may resist. Program B nominates knockdowns that pre-bias cells toward that convertible state, lowering the barrier to pMHC-induced tolerance.

**Method.** Amenability score = tolerogenic/TR1 induction + iTreg induction + anergy induction + Th2 state-lock release (each an aggregate z over its panel; self-KD excluded). Required: ≥3 tolerogenic/anergy markers significantly up (FDR<0.1), effective on-target knockdown. A **stim-relevance** flag marks hits that retain amenability under stimulation (the state you are trying to convert).

**Top targets (of 320; 151 stim-relevant):**

| Rank | Target | Score | Amenability | Best cond | TR1 ind. | Anergy ind. | Stim-rel. |
|---|---|---|---|---|---|---|---|
| 1 | **ATP2A2** | 0.98 | 8.12 | Stim8hr | 3.78 | 0.13 | ✓ |
| 2 | **STK11** (LKB1) | 0.91 | 8.19 | Stim8hr | 2.68 | 3.58 | ✓ |
| 3 | **SIK3** | 0.82 | 7.05 | Rest | 1.64 | 2.37 | ✓ |
| 4 | MTA2 | 0.77 | 6.42 | Stim8hr | 2.34 | 1.38 | ✓ |
| 5 | **PRKAR1A** | 0.77 | 7.15 | Rest | 3.19 | 1.35 | ✓ |
| 6 | KPNA1 | 0.74 | 7.27 | Rest | 4.05 | 0.55 | ✓ |
| 8 | CISD1 | 0.67 | 7.47 | Rest | 4.81 | 1.68 | ✓ |
| 10 | CPSF6 | 0.65 | 4.80 | Stim8hr | 0.77 | 2.24 | ✓ |
| 11 | **PTPRC** (CD45) | 0.64 | 4.52 | Stim48hr | 1.51 | 1.50 | ✓ |

**Biological read.** The top of the list forms a coherent, mechanistically-grounded **metabolic / cAMP tolerance axis**: **STK11/LKB1** and **SIK3** (LKB1→salt-inducible-kinase signaling) and **PRKAR1A** (PKA regulatory subunit; cAMP–PKA is a canonical Treg/anergy pathway) — all documented to bias T cells toward regulatory/anergic states when disrupted. A second cluster is chromatin-repressive machinery (**MTA2/CHD4** NuRD, **DNTTIP1** MiDAC) whose removal increases plasticity. **STAT6** appears in both programs (dampens Th2 *and* releases the lock). Note: FOXP3 induction specifically is the scarcer arm (only ~55 knockdowns raise it), so most hits reach convertibility via the TR1/anergy routes rather than the classical iTreg-FOXP3 route. Full table: `programB_tolerance_amenability_cotargets.csv`.

![Program B amenability]({{artifact:art_7e35b675-2f03-4849-b926-43785a509b43}})

---

## 4. Dual-program nodes and orthogonal corroboration

**33 targets score in both programs** — they dampen Th2 output *and* increase tolerance-amenability. The standout is **SIK3** (A=0.55, B=0.82, high cross-donor robustness, corroborated by the authors' independent Th2/Th1 polarization model) — and it is a **kinase**, the most pharmacologically tractable class. Other dual nodes: HEXIM1, CPSF6, METTL1, IMPDH2 (the last is the target of mycophenolate, an established immunosuppressant).

Program A hits were cross-checked against the authors' orthogonal Th2/Th1 polarization regulator model (`pert2state`); high concordance for the top-ranked regulators supports that these are genuine polarization controllers, not assay artifacts.

![Integration]({{artifact:art_a01a532e-8f49-4d27-9227-013334f0f582}})

---

## 5. Druggability / modality lanes

Each candidate was classed by protein family (repo gene-lists) + UniProt localization.

**Program A — biologic lane (antibody-tractable surface/secreted):** **IL2RB**, **IL4R**, **LAT** — IL4R being the validated dupilumab target.
**Program B — kinase lane (small-molecule-tractable):** **STK11, SIK3, CSK, IKBKB, CHUK (IKK), LCK, WNK1** — the tolerance nodes are unusually druggable, with existing chemical matter for several.
**Both programs** also carry large intracellular/TF fractions (degrader / oligonucleotide territory).

![Druggability]({{artifact:art_1e914102-c8c9-49be-8c30-464cf32bee61}})

---

## 6. Condition-specificity and robustness

Effects are annotated per culture condition and per cross-donor robustness tier. Program A dampening is **stimulation-dependent** (IL13/IL5/IL4 columns light up under Stim, not Rest — consistent with cytokine induction requiring activation). Program B tolerogenic induction is broader across conditions, with the metabolic/kinase hits driving the most consistent signal.

![Program A heatmap]({{artifact:art_b739f81c-2d6e-4c3c-b82a-64171c579dec}})

![Program B heatmap]({{artifact:art_77bb05e1-ecad-4864-919e-1d2cb3abb1f2}})

---

## 7. Hand-off questions for the Therapeutic Program Specialist

**Program A (dampen Th2 output):**
1. Do we pursue the **biologic lane** (IL2RB/IL4R/LAT — surface, antibody-tractable, IL4R already clinically validated) or the **selective novel regulators** (BHLHE40, SMAD4, STAT5A) as small-molecule/degrader targets?
2. GATA3 is the strongest selective hit but a classic "undruggable" TF — is a GATA3-directed degrader or antisense (cf. SB010 GATA3 DNAzyme, prior asthma trials) in scope?

**Program B (tolerance amenability):**
3. The metabolic/cAMP axis (**STK11/SIK3/PRKAR1A**) is the most druggable and mechanistically coherent tolerance lever. Which do we prioritize for a pMHC-combination hypothesis?
4. **SIK3** is the single best dual-program node *and* a kinase — is this the lead co-target to pair with an anergy/iTreg pMHC construct?
5. For the combination concept: is the intended pMHC modality **anergy-inducing** or **iTreg-converting**? That choice reweights Program B (anergy arm: STK11, CPSF6, DNTTIP1; TR1/iTreg arm: ATP2A2, KPNA1, CISD1, PRKAR1A).

**Cross-cutting / validation:**
6. All hits are healthy-donor in-vitro hypotheses. What is the validation path — EoE patient CD4⁺ cells, an allergen-recall assay, or an esophageal organoid co-culture reading out CCL26/POSTN?
7. Separately: the **TCR→specificity thread** (which clone to reprogram) remains open and needs different data (a public EoE repertoire or VDJdb allergen-specific TCRs). Recommend scoping it in parallel.

---

## Artifacts

- `reprogramming_panels.csv` — 47-gene readout panel definitions (6 axes, program mapping, coverage)
- `de_qc_summary.csv` — per-condition screen QC
- `de_readout_long.parquet` — full perturbation × readout stats (checkpoint; 1.6M rows)
- `programA_th2_dampening_targets.csv` — 122 ranked Program A dampeners
- `programB_tolerance_amenability_cotargets.csv` — 320 ranked Program B co-targets
- `candidates_annotated.csv` — master table with polarization corroboration + robustness
- `target_druggability_classification.csv` — modality lanes + UniProt localization
- Figures: `fig1_readout_coverage.png`, `fig2_programA_dampeners.png`, `fig3_programB_amenability.png`, `fig4_integration.png`, `fig5_druggability.png`, `programA_heatmap.png`, `programB_heatmap.png`