# Program Memo: EoE T-Cell Reprogramming — CRISPRi Perturb-Seq Follow-On Strategy **Prepared for:** Program Review / Go-No-Go Decision **Scope:** Two candidate programs derived from genome-scale CRISPRi Perturb-seq in primary human CD4+ T cells (healthy donor, anti-CD3/CD28 stim, in vitro) --- ## 0. Framing This screen operates on the **state axis** — what transcriptional/signaling program a CD4+ T cell runs (Th2 effector vs. tolerant/anergic) — not the **specificity axis** (which TCR, which epitope, which HLA restriction). That distinction governs everything below. None of these hits touch CCL26 or POSTN because those genes aren't expressed in T cells at all; they are epithelial/fibroblast outputs several steps downstream of the IL-13/IL-5/IL-4 signal this screen actually measures. Every target discussed here is a hypothesis about an upstream regulator, generated in healthy-donor, non-antigen-experienced, generically stimulated T cells. There is no allergen, no esophageal tissue, no EoE patient cell, and no evidence yet that any of this reproduces in a chronically activated, tissue-resident, antigen-specific effector clone — which is the actual disease substrate. Read every score below as "reproducible effect in an artificial stimulation system," not "validated EoE mechanism." --- ## 1. Program-Level Go/No-Go ### Program A — State Modulation (standalone Th2-dampening drug): **DO NOT PURSUE as a standalone program** Program A is mechanistically well-supported (top hits cluster on IL-2R signaling, GATA3, and Th2 transcriptional control — exactly where biology says they should) but this is precisely the problem. **IL4R is dupilumab's own target.** Dupilumab is approved for EoE. Cendakimab (anti-IL-13) is filed. Benralizumab (anti-IL5Ra) already exists in the eosinophil-depletion lane. A screen that rediscovers IL4R, GATA3 (the master Th2 TF sitting immediately upstream of IL4/IL5/IL13 transcription), and IL2RB (IL-2 signaling gating Th2 differentiation) is excellent **internal validation of screen quality** — it confirms the assay finds real biology — but it is not a new asset. Building a T-cell-intrinsic biologic or degrader against these nodes to re-hit a pathway three approved/late-stage programs already block, with no demonstrated superiority in efficacy, durability, or route of administration, does not clear a differentiation bar. The novel nodes in the top 12 (BHLHE40, TMEM87B, WAC, CFAP20, ENO1) are more interesting precisely *because* they are not canonical Th2 axis genes — but that novelty is unproven, and none has literature-grade causal validation in allergic/Th2 disease. **Verdict: Program A does not justify a standalone drug program.** The redundant top hits (IL4R, GATA3, IL2RB, SMAD4) should be retired from active consideration except as screen positive controls. A narrow, low-cost exploratory workstream on the *non-redundant* hits (principally BHLHE40) is defensible, but it should be scoped and resourced as exploratory target biology, not a program. ### Program B — pMHC-Tolerance Amenability (adjunct to tolerogenic construct): **PURSUE WITH CONDITIONS** Program B is differentiated in a way Program A is not: it isn't trying to out-compete an approved cytokine-blocker, it's proposing a **combination mechanism** — transiently de-lock committed Th2 effector clones so a tolerogenic pMHC-II construct (Navacim-class) can actually convert them to TR1/iTreg, rather than bouncing off an already-committed, terminally differentiated effector. This is mechanistically coherent for a specific, non-obvious reason: independent top hits (STK11, SIK3, PRKAR1A, ATG12) converge on the **LKB1→AMPK/SIK→CRTC-CREB and cAMP/PKA axis**, a pathway with a real, pre-existing literature connecting energy-sensing kinases to anergy susceptibility and Treg differentiation. Multiple genes in one known pathway surfacing independently as top hits is a much stronger signal than any single gene score — it argues the amenability phenotype is a real pathway effect, not multiple-testing noise. That said, "amenability" is a short-timescale in vitro proxy (TR1/anergy marker induction at Stim8hr/Rest), not demonstrated durable tolerance, and the adjunct concept has never been tested with an actual pMHC construct. **Verdict: Advance to de-risking, conditional on (a) confirming the pathway convergence isn't an artifact of shared essential-gene fitness effects, and (b) running the combination experiment specified in Section 3 before any modality investment.** --- ## 2. Lead-Target Verdicts **GATA3 — DROP (as a target; retain as pathway confirmation).** Master Th2 lineage TF sitting directly upstream of IL4/IL5/IL13; the biology is textbook and the effect size is real. But it is an intracellular, DNA-binding TF with no tractable small-molecule pocket — realistic modality is a degrader or oligonucleotide, both harder and slower than dupilumab's already-approved antibody. GATA3 is also required broadly for T-cell development and Th2 identity system-wide, so degrading it risks durable, non-tissue-restricted Th2 immunosuppression (impaired anti-helminth and mucosal immunity) for a target class that offers no efficacy or safety edge over existing biologics. Not worth the modality risk given zero differentiation upside. **IL2RB — DROP.** IL-2Rβ is shared machinery for IL-2 and IL-15 signaling and is essential for Treg survival and NK cell function, not Th2-specific. Chronic knockdown/blockade risks depleting or destabilizing the very regulatory compartment an EoE program should be trying to expand, and NK/CD8 bystander effects raise infection and surveillance concerns. High risk of broad immunosuppression for a mechanism with no clear advantage over IL-4Rα blockade. Drop. **IL4R — DROP (as new target; useful only as screen positive control).** This is dupilumab's mechanism, full stop. Any modality directed here (biologic, small molecule, or otherwise) is chasing an already-approved drug with no stated efficacy, safety, or delivery advantage. Its presence in the top 12 is valuable only as evidence the screen correctly recovers known EoE pharmacology. **BHLHE40 — ADVANCE-WITH-CONDITIONS.** Top-scoring hit and not part of the canonical, already-drugged Th2 axis — the one node in Program A with genuine differentiation potential. BHLHE40 (DEC1/SHARP2) has documented roles in Th1/Th17/Th2 balance, GM-CSF production, and tissue-resident memory formation, giving it real biological plausibility beyond this screen. Key risks: it is a bHLH transcriptional repressor (hard-to-drug, degrader/oligo modality only) with broad expression including circadian clock and metabolic tissues, so off-tissue effects are unknown and unstudied in this context. Condition for advancement: confirm Th2-selective effect (not general T-cell activation suppression) in an allergen-stimulated, patient-derived T-cell system before any modality commitment. **SMAD4 — DROP.** SMAD4 is the shared central hub of canonical TGF-β/BMP signaling across essentially every tissue, and it is a well-established tumor suppressor — germline loss-of-function causes juvenile polyposis, and SMAD4 loss is a recurrent pancreatic cancer driver. Knocking down a tumor-suppressor signaling hub in T cells for a chronic, non-life-threatening indication is an unacceptable safety profile regardless of Th2 effect size. Drop outright; the score does not change the calculus. **STK11 (LKB1) — ADVANCE-WITH-CONDITIONS.** Highest Program B score. STK11/LKB1 is a well-characterized master metabolic checkpoint kinase with published links to T-cell anergy and Treg differentiation, giving this hit unusually strong prior plausibility rather than screen-only support. Principal risk: STK11 is a bona fide tumor suppressor (Peutz-Jeghers syndrome from germline loss), so durable pharmacological inhibition is a genuine oncology safety concern. This is mitigated, not eliminated, by the proposed use case — a **transient, combination-course adjunct** alongside a pMHC tolerogenic construct, not chronic monotherapy. Advance for feasibility work under a strict transient-dosing constraint, with genotoxicity/proliferation monitoring built into any preclinical plan. **SIK3 — ADVANCE-WITH-CONDITIONS.** The most tractable node in Program B: a kinase (existing chemical matter precedent, e.g., HG-9-91-01-class SIK inhibitors), downstream of LKB1, and the only gene scoring meaningfully in *both* programs — a useful internal consistency check. Key risks: SIK1/2/3 isoform cross-reactivity in current chemical matter (existing tool compounds are not SIK3-selective) and known SIK-inhibitor phenotypes in glucose/bone metabolism from other indications, implying systemic metabolic liability if selectivity isn't solved. Advance into a medicinal-chemistry feasibility assessment specifically scoped around isoform selectivity — this is the best near-term target for the protein-design/binder stream (see Section 5). **PRKAR1A — DROP (as direct target; retain as pathway signal).** PRKAR1A is the regulatory subunit of PKA, a signaling node so central and ubiquitous (cAMP/PKA touches nearly every cell type's core physiology) that selectivity margin is essentially unachievable. Germline PRKAR1A mutations cause Carney complex, a multi-tumor syndrome — a strong independent safety flag. The hit is mechanistically informative (confirms cAMP/PKA involvement in the amenability phenotype, reinforcing the STK11/SIK3 pathway story) but should not itself become a drug target. Drop as a target; keep as supporting pathway evidence. **SIK3 (dual-node read).** Reiterating: SIK3's appearance in both A (score 0.545) and B (score 0.815) is the strongest piece of cross-program evidence in this dataset — it argues SIK3 modulation both dampens Th2 output and increases tolerization amenability, i.e., it may be doing double duty on the same axis this program actually wants to move. That makes it the single highest-value node for chemistry investment across both programs. --- ## 3. Modality & Mechanism Fit / Combination Coherence / Top De-Risking Experiment For Program B, the core hypothesis — "make the effector clone convertible, then convert it with a tolerogenic pMHC-II construct" — is scientifically coherent enough to advance, for three reasons: (1) the LKB1-AMPK-SIK-CREB/PKA convergence across independent top hits has real prior literature support for gating anergy susceptibility, not just this dataset; (2) the two-step logic (de-lock, then instruct) matches known biology of anergy induction, where a metabolically/transcriptionally "sticky" effector state actively resists tolerogenic signals regardless of how good the pMHC construct is; and (3) it is explicitly designed as an *adjunct*, which sidesteps the standalone-immunosuppression and redundancy problems dragging down Program A. The concept has **not**, however, been tested with an actual pMHC construct, and TR1/anergy marker induction at 48hr–8day timescales is a surrogate that has not been shown to predict durable, antigen-rechallenge-resistant tolerance. **Single most important de-risking experiment:** Take primary CD4+ T cells (ideally antigen-experienced/allergen-recall or EoE patient-derived, not naive healthy-donor bulk), transiently knock down or pharmacologically inhibit STK11/SIK3, then co-culture with a tolerogenic pMHC-II nanoparticle construct presenting the relevant epitope (dependent on the separate antigen/epitope pipeline identifying that epitope and its HLA-II restriction) versus construct-alone and knockdown-alone controls. Read out TR1/iTreg conversion (FOXP3, LAG3/CD49b co-expression, IL-10 secretion), functional suppressive capacity (not just marker expression), and — critically — durability under antigen rechallenge after a rest period. A positive result requires the **combination** to outperform either arm alone; if STK11/SIK3 inhibition alone produces the same TR1 signature without the pMHC construct, the "adjunct" framing collapses into generic immunosuppression, which reduces Program B to a worse version of Program A. --- ## 4. Honest Limitations & Kill Criteria **Reason 1 — Healthy-donor, generic-stim artifact.** Both programs were built on non-antigen-experienced T cells activated with a blunt anti-CD3/CD28 signal, not disease-relevant, chronically activated, tissue-resident effector clones. *Kill criterion:* if knockdown/inhibition fails to reproduce Th2 dampening (A) or amenability gain (B) in EoE patient PBMC or esophageal-biopsy-derived T cells under allergen/food-antigen recall stimulation, kill the affected target. **Reason 2 — Selectivity is unproven; broad immunosuppression risk is real.** Several top hits (IL2RB, SMAD4, PRKAR1A, and possibly GATA3, STK11) sit in pathways essential to Treg homeostasis, other T-cell subsets, or core cell physiology broadly. *Kill criterion:* if confirmatory counter-screens show comparable perturbation effects in Th1/Th17/Treg populations (selectivity ratio <2x versus Th2), the "selective effector reprogramming" thesis fails and the target reverts to generic immunosuppression — not differentiated. **Reason 3 — In vitro convertibility markers may not equal durable tolerance.** TR1/anergy induction at 48hr–8-day readouts is a short surrogate. *Kill criterion:* if converted cells revert to effector phenotype/cytokine output upon antigen rechallenge after an extended rest (2–3 week) culture, or if marker induction (FOXP3, IL-10) does not correlate with an actual suppression assay, the combination concept for Program B fails and should not advance into construct co-development. --- ## 5. Recommended Next Steps (prioritized, no GPU compute this pass) 1. **Dry-lab cross-validation (immediate, no wet lab):** Check expression/regulation of BHLHE40, STK11, SIK3, and PRKAR1A against public EoE patient biopsy/PBMC transcriptomic datasets to see whether this pathway is even perturbed in disease tissue before committing further resource. 2. **Safety literature triage (immediate):** Compile Peutz-Jeghers (STK11), Carney complex (PRKAR1A), and existing SIK-inhibitor phenotype data (metabolic/bone) into a formal target-safety dossier before any chemistry or construct investment. 3. **Design the combination de-risking assay (Section 3)** as the top wet-lab priority once resourced — define the STK11/SIK3 perturbation reagents (CRISPRi or transient tool-compound), TR1/anergy/suppression readout panel, and rechallenge design. 4. **Dependency flag — pMHC/antigen-epitope pipeline:** the de-risking experiment cannot run without a defined EoE-relevant epitope and HLA-II restriction from that separate workstream; this is a hard blocking dependency, not a nice-to-have. 5. **Protein-design/binder stream tasking:** SIK3 isoform-selective inhibitor or degrader design is the single most concrete, well-scoped near-term ask for that stream, given its dual-node status and existing (non-selective) chemical starting points. 6. **TCR-specificity thread:** convertibility work is only actionable once dominant EoE-relevant effector clonotypes are identified; flag this as the prioritization input needed before choosing which clones the combination experiment should even test. 7. **Later, compute-gated (not this pass):** GPU-based structure prediction/docking for SIK3-selective chemical matter, and multi-omic modeling of convertibility trajectories, once the above dry-lab and assay-design steps justify the investment. --- **Bottom line:** Do not pursue Program A as a standalone drug program — it re-hits ground dupilumab and cendakimab already own; pursue Program B, conditionally, as a mechanistically coherent adjunct to the pMHC tolerogenic construct, gated on the STK11/SIK3 combination de-risking experiment before any modality or construct investment.