Selective Th2 Suppressor Atlas

AHSA1 — mechanism dossier

AHSA1 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.11, magnitude-controlled residual = -2.53, condition = Rest, class = other

Final verdict (post–adversarial review): UNCERTAIN · investigator: uncertain (low conf) · citations: partially_supported

Hypothesis

Loss of AHSA1 (Aha1), the sole mammalian activator of HSP90 ATPase, impairs HSP90 chaperoning of the metastable Th2-defining clients STAT6 and GATA3, collapsing the GATA3/IL4/IL5/IL13 program while leaving the less chaperone-addicted T-bet/IFNG axis intact.

Mechanism

AHSA1/Aha1 stimulates HSP90's ATPase cycle and is preferentially required for maturation of a subset of "hard-to-fold," metastable clients (kinases, steroid receptors, v-Src) rather than for bulk proteostasis. STAT6 (the IL-4/STAT6 axis that induces and sustains GATA3) and GATA3 itself are dosage-sensitive, chaperone-dependent signaling proteins, so reduced HSP90 productivity from AHSA1 knockdown could lower functional STAT6/GATA3 and thereby the whole IL4/IL5/IL13 output. Because the mechanism acts by removing positive Th2 drivers rather than by inducing T-bet, it would not push cells toward Th1. Importantly, unlike HSP90 or core translation/RNA machinery, Aha1 is non-essential (viable KO cell lines and dispensable in yeast/basal conditions), so its knockdown need not globally kill or arrest the cell.

Bypasses T-bet / Th1? True — The proposed route (impaired HSP90 chaperoning of STAT6/GATA3) removes Th2-positive drivers and provides no mechanism to activate T-bet or STAT4; Aha1 loss is not known to induce a Th1 program, so it fits a 'selective suppressor' rather than a 'Th1 driver' that indirectly cross-represses Th2.

Evidence

• AHSA1/Aha1 is the only known mammalian activator of HSP90 ATPase activity; it is a stress-regulated co-chaperone that drives HSP90 into its closed, ATPase-active state.

source: Recruitment of Ahsa1 to Hsp90 is regulated by a conserved peptide that inhibits ATPase stimulation, PMC11316058

• Aha1 is dispensable under basal conditions (non-essential in yeast; viable human AHSA1-KO cell lines exist), but its loss selectively impairs activation of hard-to-fold clients such as v-Src and hormone receptors — i.e., it modulates a client subset rather than bulk proteostasis.

source: Silencing of HSP90 cochaperone AHA1 decreases client protein activation (PubMed 18281495); Abcam human AHSA1 KO HeLa line ab265043; Panaretou/Prodromou co-chaperone reviews

• STAT6 is required to induce/sustain GATA3 and the Th2 cytokine program; STAT6 or GATA3 loss/inhibition selectively abrogates IL4/IL5/IL13 and allergic inflammation without driving Th1.

source: STAT6 inhibitor AS1517499 reduces Th2 cytokines and regulates GATA3 (PMC8851827); GATA-3 promotes Th2 responses (Nature Cell Research, 7310002)

• HSP90/heat-shock chaperone machinery is physically and functionally linked to GATA3 function (e.g., HSP60 maintains GATA3-dependent Pol II pausing at the Il5 locus; Aha1 modulates GATA3-dependent phenotypes in vivo).

source: Methylation of Gata3 at Arg-261 / HSP60-GATA3 at Il5 (PMC4505565); zebrafish Aha1a modifies GATA3-mutant craniofacial phenotype

• HSP90 also chaperones Dicer1 and thereby miRNA maturation, illustrating that Aha1/HSP90 perturbation can have pleiotropic, program-shaping downstream effects.

source: HSP90 and Aha1 modulate microRNA maturation through folding of Dicer1, PMC9262616

Proposed confirming experiment

Arrayed CRISPRi knockdown of AHSA1 (plus non-targeting and a known Th2 driver e.g. GATA3/STAT6 as positive controls) in primary human CD4+ T cells cultured in parallel under Th2-polarizing (anti-CD3/28 + IL-4 + anti-IL-12) and Th1-polarizing (IL-12 + anti-IL-4) conditions. Primary readout: intracellular flow/qPCR for GATA3, IL-4/IL-5/IL-13 vs T-bet/IFN-gamma, with the key test being a Th2-specific drop and NO increase in T-bet/IFN-gamma. Mechanistic readouts: STAT6 total and phospho-STAT6 levels after IL-4, GATA3 protein half-life by cycloheximide chase, and HSP90 co-IP with STAT6/GATA3 upon AHSA1 KD (expect reduced client stability/association). Critical artifact controls: CellTrace proliferation, viability, and a global-magnitude panel (total protein synthesis / housekeeping TF such as RUNX or general activation markers) to confirm the effect is Th2-selective and not global dampening; also confirm on-target via multiple guides and rescue with guide-resistant AHSA1.

Adversarial review

• Citations check: partially_supported
• Agrees artifact call (False): True
• Notes: Cited sources are all real and I confirmed each. However support is partial and the mechanism has an unsupported core. Confirmed: PMC11316058 (Aha1 ICD autoinhibition), PMC9262616 (Aha1/HSP90-Dicer1), PubMed 18281495 (client-subset modulation), PMC8851827 (STAT6i reduces Th2), zebrafish PMC3759348. Key objections: (1) The load-bearing premise that STAT6 and GATA3 are HSP90/Aha1 clients is NOT supported by any cited source; Aha1's known client subset is kinases/v-Src/steroid receptors/CFTR, not these TFs, so AHSA1->HSP90->STAT6/GATA3 is an inference gap. (2) The 'HSP90/GATA3 machinery' citation PMC4505565 is actually about HSP60 (not HSP90/Aha1) and HSP60 REPRESSES Il5 by pausing Pol II — a misleading, arguably wrong-direction link. (3) The zebrafish paper shows Aha1a knockdown alleviates a GATA3 loss-of-function (HDR) phenotype, i.e. a genetic modifier, not that GATA3 is an Aha1 client. (4) 'Only/sole mammalian activator' is overstated — source says 'most potent stimulator'; AHSA2 paralog exists. (5) The 'bypasses T-bet' selectivity is contradicted by evidence that HSP90 blockade (17-DMAG, PMC3976086) suppresses Th1/Th17 (IFN-g, IL-17) and disrupts Lck/TCR signaling, i.e. broad T-cell dampening rather than Th2-specific collapse. Partial defense: AHSA1 KD is milder and more client-selective than HSP90 inhibition (viable KO lines), so a selective phenotype is not impossible. Net: gene is real and biology is coherent enough that the screen hit is unlikely a pure artifact (agree likely_artifact=false), but the specific selective-Th2 mechanism is speculative and under-supported; 'uncertain' is appropriate, leaning skeptical.

ARNT — mechanism dossier

ARNT — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.51, magnitude-controlled residual = -2.59, condition = Stim48hr, class = TF

Final verdict (post–adversarial review): UNCERTAIN · investigator: plausible (medium conf) · citations: partially_supported

Hypothesis

ARNT (HIF-1β) is the obligate dimerization partner for HIF-1α/HIF-2α, and because stimulation-induced HIF signaling promotes the GATA3/IL-4 Th2 program while Th1/T-bet is relatively HIF-independent, ARNT knockdown collapses HIF output and selectively lowers Th2 without inducing Th1.

Mechanism

ARNT is the shared, obligate bHLH-PAS partner that HIF-1α and HIF-2α (and AHR) require to form functional transcription factors. TCR/mTOR signaling induces HIF-1α even under normoxia during activation, and multiple studies show HIF-1α and HIF-2α positively drive Th2 differentiation, GATA3, and IL-4/IL-13 (a HIF-1α inhibitor suppresses Th2/IL-4/IL-13; HIF-2α deficiency impairs Th2 and asthmatic inflammation; a 2024 Immunity paper shows HIF-2α promotes pathogenic stem-like Th2 polarization). Losing ARNT would abrogate both HIF-1α and HIF-2α output simultaneously, removing this pro-Th2 input. Because HIF preferentially supports Th17/glycolysis and Th2, and Th1/T-bet commitment is comparatively HIF-independent, the Th1 arm can stay flat rather than being reciprocally induced.

Bypasses T-bet / Th1? True — The proposed route works by removing a positive HIF-driven input to the Th2/GATA3 program, not by relieving Th2 repression of Th1. Loss of HIF/ARNT is not known to induce T-bet or IFNG; if anything HIF drives Th17 and glycolysis, and Th1 commitment is comparatively HIF-independent. So a Th2-down/Th1-flat pattern is mechanistically consistent with selective suppression rather than Th1 induction. Caveat: the AHR:ARNT arm points the opposite way (AHR activation suppresses Th2, AHR loss raises IL-4), so the net Th2-down phenotype implies the HIF arm dominates in this stimulated normoxic context.

Evidence

• ARNT/HIF-1β is the obligate heterodimeric partner required by both HIF-1α and HIF-2α (and AHR); it is ubiquitously expressed and its total loss is embryonic lethal (a hub/essential factor).

source: Conditional Disruption of Arnt, Mol Endocrinol 2000; Arnt endothelial KO embryonic lethality, PMC1559728; ScienceDirect ARNT overview

• HIF-1α promotes Th2 differentiation and IL-4/IL-13; a HIF-1α inhibitor restrains HIF-1α and inhibits Th2 differentiation and IL-4/IL-13 expression.

source: Frontiers Immunol 2022, HIF-1α in inflammatory autoimmune diseases (10.3389/fimmu.2022.1073971)

• HIF-2α promotes pathogenic polarization of stem-like Th2 cells; HIF-2α deficiency impairs Th2 differentiation and alleviates asthmatic inflammation.

source: Immunity 2024, 'Hypoxia-inducible factor 2α promotes pathogenic polarization of stem-like Th2 cells' (Cell Press S1074-7613(24)00496-5)

• Hypoxia/HIF-1α augment Tc2/Th2 responses via JAK1/3 and GATA-3, linking HIF output to GATA3 induction and IL-4 signaling.

source: J Allergy Clin Immunol / Hypoxia enhances CD8+ Tc2 via HIF-1α, PMC11098440

• HIF-1α drives glycolysis (Glut1, glycolytic enzymes) and tips Th17 vs Treg balance; loss impairs Th17 but not classically Th1 — supporting that HIF loss is not a Th1 driver.

source: Shi et al., J Exp Med 2011, PMC3135370 (HIF1α glycolytic checkpoint for Th17/Treg)

• Via the AHR arm, ARNT loss would be expected to INCREASE Th2: AHR activation (TCDD) suppresses Th2 cytokines/allergy, and AHR deficiency elevates IL-4 — a direction opposite to the screen hit, indicating the HIF arm must dominate.

source: AhR master regulator of allergic immune responses, PMC9806217; Differential regulation of asthmatic phenotype by AhR, PMC8566992

Proposed confirming experiment

Single-gene ARNT CRISPRi knockdown in primary human CD4+ T cells under explicit Th2-polarizing vs Th1-polarizing conditions, reading GATA3, IL4, IL5, IL13 vs TBX21/IFNG by intracellular cytokine flow and qPCR. Critically distinguish selective suppression from global dampening by co-measuring: (a) proliferation/viability, total protein synthesis, and blast size; (b) glycolytic flux (ECAR, GLUT1/SLC2A1, LDHA) to confirm HIF axis engagement; (c) normoxia vs hypoxia (1% O2) and CoCl2 stabilization to test HIF-dependence. Run epistasis/rescue: compare ARNT KD to HIF1A + HIF2A double KD (should phenocopy) and to AHR KD (predicted opposite/Th2-up), and rescue ARNT KD with a HIF-target-independent GATA3 overexpression to test whether Th2 loss is downstream of GATA3. Confirm direct regulation with HIF/ARNT ChIP-seq or CUT&RUN at the GATA3/IL4 loci. Selective, HIF-dependent GATA3/IL4 loss with intact viability and flat Th1 would confirm a genuine selective Th2 suppressor rather than a magnitude artifact.

Adversarial review

• Citations check: partially_supported
• Agrees artifact call (False): True
• Notes: Confirmed: the cited primary sources are real and accurate for what they individually say. The Immunity 2024 HIF-2α paper (PMID 39609127) robustly supports HIF-2α promoting Th2 via a HIF2a-GATA3 circuit, with HIF2a deficiency lowering GATA3/IL-4/IL-5/IL-13 and alleviating asthma. Shi 2011 JEM (PMC3135370) is correctly cited but concerns Th17/Treg glycolysis and does NOT address Th1; the claim that HIF loss is not a Th1 driver is an unsupported inference, and the 'bypasses T-bet' framing is a strawman (the mechanism never engages T-bet). The core weakness: NO source demonstrates ARNT knockdown selectively lowering Th2 in CD4 cells — the entire ARNT link is inferential through HIF. ARNT is a pleiotropic hub (obligate for HIF-1a, HIF-2a, AND AHR); ARNT deletion in CD8 T cells broadly impairs effector function (perforin/granzyme/migration), which favors the competing 'broadly dampens the cell' hypothesis over Th2 selectivity. The investigator also concedes the AHR arm predicts the opposite direction (AHR loss raises IL-4), so net Th2 suppression rests on an untested assumption that the HIF arm dominates. Not likely an artifact (HIF genuinely drives Th2, so a Th2-lowering hit is mechanistically coherent), but 'plausible' overstates the selectivity case given hub pleiotropy and absent direct ARNT-Th2 evidence; downgrading to uncertain.

ELAVL1 — mechanism dossier

ELAVL1 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.68, magnitude-controlled residual = -2.35, condition = Stim48hr, class = other

Final verdict (post–adversarial review): UNCERTAIN · investigator: plausible (medium conf) · citations: partially_supported

Hypothesis

Knockdown of ELAVL1 (HuR) removes AU-rich-element-mediated stabilization of GATA3, IL4, and IL13 mRNAs, collapsing the Th2 program post-transcriptionally without needing to induce the Th1/T-bet axis.

Mechanism

HuR is an ARE-binding RNA-stabilizing protein that directly binds and extends the half-life of GATA3 mRNA (and the Th2 cytokine mRNAs IL4 and IL13), which was demonstrated by RIP-seq of Th2-polarized CD4+ T cells and actinomycin-D chase showing increased GATA3 mRNA half-life after restimulation. Because GATA3 is the master Th2 transcription factor, losing HuR-mediated GATA3 mRNA stabilization directly deflates the entire Th2 transcriptional/cytokine cascade. Critically, partial loss-of-function (heterozygous conditional KO) lowered steady-state Gata3/Il4/Il13 mRNA, which mirrors a CRISPRi partial-knockdown and matches the screen's Th2-arm signal. This is a bona fide post-transcriptional Th2 node rather than a pure global-metabolism gene.

Bypasses T-bet / Th1? True — The documented mechanism is destabilization/loss of GATA3, IL4, and IL13 transcripts, not induction of T-bet or the Th1 program. HuR acts by removing a positive post-transcriptional input to Th2 rather than by activating a Th1-driving transcription factor, so a flat Th1/IFNG arm under Th2-polarizing Stim48hr conditions is mechanistically consistent, not a red flag.

Evidence

• HuR binds AU-rich elements in the 3'UTRs of GATA3, IL4 and IL13 mRNAs and stabilizes them; GATA3 mRNA half-life increases after restimulation in human memory/Th2 cells.

source: Coordinate regulation of GATA3 and Th2 cytokine gene expression by the RNA-binding protein HuR, PMC5801757

• Partial (heterozygous) HuR conditional KO in CD4+ T cells decreased steady-state Gata3, Il4 and Il13 mRNA under Th2 polarization, revealing a gene-dosage effect on cytokine production.

source: Conditional Knockout of the RNA-Binding Protein HuR in CD4+ T Cells Reveals a Gene Dosage Effect on Cytokine Production, Mol Med 2013, PMC3960399

• HuR RIP-seq identified 271 direct HuR target transcripts in Th2-polarizing conditions, establishing GATA3/Th2 cytokine mRNAs as direct targets.

source: Transcriptomic-Wide Discovery of Direct and Indirect HuR RNA Targets in Activated CD4+ T Cells, PLOS ONE, PMC4498740

• HuR is proposed as a therapeutic node for GATA3-driven type 2 inflammation across CD4+ Th2 and ILC2; pharmacologic HuR inhibition is expected to shorten GATA3 mRNA half-life and suppress type-2 cytokines.

source: HuR Regulates GATA3-Driven Type 2 Inflammation in CD4+ T cells and ILC2, bioRxiv 2026 (2026.04.23.720195)

• Caveat / complication: HuR also controls IL-2 homeostasis and JAK-STAT signaling, and full homozygous HuR KO paradoxically INCREASED Il4/Il13 via compensatory mechanisms, and HuR is broadly essential (also strengthens Th17 differentiation).

source: The RNA-Binding Protein HuR Posttranscriptionally Regulates IL-2 Homeostasis and CD4+ Th2 Differentiation (RG 319192969); HuR Plays a Positive Role in CD4+ T Cell Activation and Th17 Differentiation, PMC8357502

Proposed confirming experiment

In primary human CD4+ T cells, perform titratable CRISPRi (or shRNA) partial knockdown of ELAVL1 and split cells into Th2-polarizing vs Th1-polarizing conditions (Stim48hr). Readouts: (1) qPCR/protein for GATA3, IL4, IL5, IL13 vs TBX21, IFNG — expect selective Th2 drop with flat Th1 and, importantly, NO compensatory rise in T-bet; (2) actinomycin-D mRNA half-life chase to show shortened GATA3/IL4/IL13 half-life under KD; (3) HuR RIP-qPCR to confirm direct binding to GATA3 mRNA; (4) a rescue with a GATA3 transgene bearing a HuR-independent 3'UTR to test whether restoring GATA3 rescues the Th2 defect. Use graded knockdown to avoid the full-KO compensatory-derepression regime and confirm the partial-KD phenotype matches the screen.

Adversarial review

• Citations check: partially_supported
• Agrees artifact call (False): True
• Notes: Confirmed real: PMC5801757 (GATA3 3'UTR AREs; HuR overexpression stabilizes GATA3/IL4/IL13 — supports core mechanism), PMC3960399 (gene-dosage KO), PMC4498740 (RIP-seq, 271 targets confirmed). Two problems weaken the dossier: (1) The RIP-seq paper does NOT list GATA3/IL4/IL13 among its direct targets — it only cites prior work — so the claim that it 'establishes them as direct targets' is overstated/contradicted. (2) The bioRxiv 2026 preprint could not be found or verified in any index (unverifiable). Strongest objection to the hypothesis: the same KO paper shows a gene-dosage PARADOX — heterozygous (partial) KO lowered Gata3/Il4/Il13, but full homozygous KO INCREASED Il2/Il4/Il13 via compensation. Since 'knockdown' is a partial/variable reduction, a strong KD could raise rather than collapse type-2 cytokines. Moreover HuR is a broadly essential RBP (IL-2, JAK-STAT, T-cell activation, Th17), so KD broadly dampens CD4 T cells rather than acting as a SELECTIVE Th2 suppressor. The 'bypasses T-bet' logic is not itself unsound (post-transcriptional collapse needn't induce Th1), but the selectivity and the directionality are doubtful. Not a statistical artifact — gene-target links are genuine — so I agree likely_artifact=false, but downgrade plausible→uncertain.

GPATCH8 — mechanism dossier

GPATCH8 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.18, magnitude-controlled residual = -1.94, condition = Rest, class = other

Final verdict (post–adversarial review): UNCERTAIN · investigator: uncertain (low conf) · citations: supported

Hypothesis

As a branchpoint splicing quality-control factor, GPATCH8 loss could selectively impair productive splicing/expression of a Th2-program transcript (e.g., GATA3 or an upstream STAT6/IL4-axis component), lowering Th2 output without engaging the T-bet/Th1 axis.

Mechanism

GPATCH8 is a G-patch protein that activates the RNA helicase DHX15 in branchpoint/intron quality control, functionally opposing SUGP1; it is required for aberrant splicing induced by mutant SF3B1. There is no direct evidence linking GPATCH8 to GATA3, IL4, or Th2 differentiation. A selective route would have to run through splicing: GPATCH8-dependent retention or correct processing of a Th2-enabling transcript (GATA3 has multiple isoforms and is dosage-sensitive), such that KD skews toward non-productive isoforms of a Th2 node while leaving Th1 transcription factors unaffected. This is mechanistically conceivable for a splicing factor but is entirely speculative for GPATCH8 specifically.

Bypasses T-bet / Th1? True — GPATCH8 is a splicing quality-control factor, not a Th1 transcription factor or T-bet inducer; there is no known route by which losing it would activate the Th1 program. The observed Th1-flat/Th2-down pattern is therefore consistent with a mechanism that reduces Th2 output directly rather than by diverting cells into Th1, so the selectivity is not explained by simple T-bet induction.

Evidence

• GPATCH8 is a G-patch splicing quality-control factor that activates DHX15 and antagonizes SUGP1 in branchpoint selection; it is required for mutant SF3B1-induced mis-splicing.

source: Benbarche et al., 'GPATCH8 modulates mutant SF3B1 mis-splicing and pathogenicity in hematologic malignancies', Molecular Cell 2024 (PMC11102302)

• Silencing GPATCH8 corrected ~one-third of mutant-SF3B1 splicing defects and IMPROVED dysfunctional hematopoiesis in mice and primary human progenitors, indicating its loss is well tolerated (not an essential/lethal gene).

source: Benbarche et al., Molecular Cell 2024; ASH Blood 2023 abstract 179848

• GPATCH8 domain architecture: N-terminal G-patch (RNA-processing) domain, C2H2 zinc finger, lysine-rich and serine-rich regions; a serine-region variant is linked to hyperuricemia — i.e., a ubiquitous RNA-processing protein, not a T-cell-restricted regulator.

source: UniProt Q9UKJ3; OMIM 614396; GeneCards GPATCH8

• The screen is the genome-scale CRISPRi perturb-seq in primary human CD4+ T cells nominating Th1/Th2 polarization regulators; GPATCH8 is not otherwise reported as a Th2 regulator in the T-cell/GATA3 literature.

source: Genome-scale perturb-seq in primary human CD4+ T cells, bioRxiv 2025.12.23.696273 (Marson lab); no GPATCH8 hits in GATA3/Th2 reviews e.g. Nat Rev Immunol nri2476

Proposed confirming experiment

CRISPRi knock down GPATCH8 in primary human CD4+ T cells, then split into Th2-polarizing and Th1-polarizing arms in parallel. Read out (i) intracellular IL4/IL5/IL13 and GATA3 vs IFNG/T-bet by flow, plus qPCR, to confirm Th2-selective loss; (ii) matched viability/proliferation (CellTrace, live-cell counts) and a global transcriptional-magnitude metric to exclude nonspecific dampening; (iii) RNA-seq with splicing analysis (rMATS/LeafCutter) focused on GATA3 and Th2-axis transcripts to test whether a specific mis-splicing/isoform shift underlies the phenotype. Epistasis test: rescue with GATA3 cDNA (splicing-insensitive ORF) — restoration of Th2 output would pin the effect on a GATA3-axis splicing dependency rather than global toxicity.

Adversarial review

• Citations check: supported
• Agrees artifact call (False): True
• Notes: Sources are real and each cited factual claim checks out. Confirmed Benbarche et al., Mol Cell 2024 (PMC11102302): GPATCH8 is a G-patch branchpoint quality-control factor that interacts with DHX15 and functionally opposes SUGP1; "Gpatch8 silencing was well-tolerated in WT mouse BM cells despite ~80% reduction" and silencing "partially rescued impaired hematopoietic growth" — so loss is not lethal/essential (supports claims 1-2). The Marson-lab genome-scale CD4+ perturb-seq preprint (bioRxiv 2025.12.23.696273) is real and does nominate Th1/Th2 polarization regulators via the screen; I could not extract the specific GPATCH8 hit line (403 on full text), but the investigator honestly concedes GPATCH8 is not independently reported as a Th2/GATA3 regulator. Strongest objection: the "bypasses T-bet / selectively suppresses Th2" hypothesis is entirely speculative and unsupported by any cited source. GPATCH8 is a ubiquitous RNA-processing factor whose ONLY characterized function is enabling MUTANT-SF3B1 mis-splicing; in WT context its loss perturbs little splicing, so a clean mechanistic link to GATA3/STAT6/IL4 splicing is unestablished and there is no data showing Th1/T-bet is spared. A splicing-QC factor scoring as a Th2 hit could equally reflect broad transcriptional/translational dampening rather than genuine Th2 selectivity. I agree likely_artifact=false only in the narrow sense that documented tolerability of GPATCH8 loss argues against a pure fitness/dead-cell artifact — but selectivity remains unproven and the mechanism is weak; "uncertain" is appropriate, arguably leaning toward the skeptical end.

MPG — mechanism dossier

MPG — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.21, magnitude-controlled residual = -2.03, condition = Stim48hr, class = other

Final verdict (post–adversarial review): UNLIKELY · investigator: uncertain (low conf) · citations: supported

Hypothesis

Loss of MPG (AAG) may selectively blunt the Th2 program by disrupting base-excision-repair/5hmC-reader activity at the demethylation-dependent IL4/IL5/IL13 Th2 cytokine locus, rather than by driving Th1.

Mechanism

MPG/AAG is a monofunctional DNA glycosylase that initiates base-excision repair (BER) of alkylated/deaminated purines, but it also has a non-catalytic role: it associates with the transcription-elongation machinery and has been proposed to act as a reader of the epigenetic marks 5-hydroxymethylcytosine (5hmC) and 5-formylcytosine, with its loss altering gene expression independent of glycosylase activity. The Th2 cytokine locus (IL4/IL5/IL13) is one of the most dynamically TET/5hmC-remodeled and actively demethylated regions during Th2 differentiation, so an AAG that reads 5hmC and couples repair to elongation could preferentially support transcription at this locus. Under this model, KD would weaken IL4/GATA3 induction without necessarily engaging the T-bet/IFNG circuit, giving apparent Th2 selectivity. This route is mechanistically thin and unproven in T cells, so it is a hypothesis rather than an established pathway.

Bypasses T-bet / Th1? True — MPG has no known role as a T-bet or Th1 driver; it is a BER enzyme/epigenetic reader. The proposed selective route acts at the Th2 cytokine locus (IL4 demethylation/5hmC and transcription-elongation-coupled repair), which is independent of T-bet induction. The observed Th1-flat/Th2-down pattern is therefore consistent with a genuinely Th2-directed effect rather than reciprocal Th1 skewing, though this remains unconfirmed.

Evidence

• MPG (AAG/ANPG) is a monofunctional DNA glycosylase that recognizes alkylated/deaminated purines (3-methyladenine, 7-methylguanine, hypoxanthine, ethenoadenine) and initiates base-excision repair.

source: DNA-3-methyladenine glycosylase, Wikipedia; GeneCards MPG (https://www.genecards.org/cgi-bin/carddisp.pl?gene=MPG)

• AAG associates with transcription elongation to coordinate DNA repair with gene expression, i.e. it has a transcription-linked function beyond canonical repair.

source: Montaldo et al., 'Alkyladenine DNA glycosylase associates with transcription elongation...', PMC6884549 (2019)

• Independent of glycosylase activity, AAG was proposed to act as a reader of epigenetic marks 5hmC/5fC, and its loss alters gene expression (shown in mouse stem cells/brain).

source: Loss of alkyladenine DNA glycosylase alters gene expression in the developing mouse brain, ScienceDirect S1568786424000089 / PMID 38280242 (2024)

• Aag/Mpg-null mice are viable, fertile and overtly phenotypically normal, indicating MPG is not an essential housekeeping gene required for cell survival.

source: Engelward et al., 'Base excision repair deficient mice lacking the Aag alkyladenine DNA glycosylase', PNAS 1997, PMID 9371804

• No primary literature directly links MPG to GATA3, IL4, Th2 differentiation, or CD4+ T-cell polarization; the T-cell/Th2 connection is absent from the published record.

source: Absence of hits across PubMed/GeneCards/OMIM searches for MPG + Th2/CD4 T cell (this review)

Proposed confirming experiment

Arrayed CRISPRi knockdown of MPG (2-3 guides + non-targeting control) in naive human CD4+ T cells, differentiated in parallel under Th2-polarizing and Th1-polarizing conditions. Primary readout: intracellular flow + qPCR for GATA3/IL4/IL5/IL13 vs T-bet/IFNG, testing the selectivity prediction (IL4 down, IFNG NOT up). Critically pair this with global-dampening controls run in the SAME cells: CellTrace-dilution proliferation, viability/apoptosis (Annexin V), total protein/RNA yield, and gammaH2AX to quantify genotoxic stress from BER loss. Selectivity is confirmed only if IL4/GATA3 fall without IFNG rising AND without proliferation/viability/global-transcription loss. To probe the proposed mechanism, add hMeDIP/bisulfite and MPG ChIP at the IL4/IL13 locus, and test rescue with catalytically-dead MPG to separate the repair function from the 5hmC-reader function.

Adversarial review

• Citations check: supported
• Agrees artifact call (False): True
• Notes: All five citations are real and accurately represented. Confirmed: Montaldo 2019 Nat Commun (PMC6884549) shows AAG couples to RNA Pol II/Elongator during transcription elongation; the 2024 brain paper (PMID 38280242) shows Aag loss alters gene expression and 5hmC signal and proposes AAG as a 5hmC reader; Engelward 1997 PNAS (9371804) confirms Aag-null mice are viable/fertile/overtly normal; and PubMed/GeneCards genuinely show no MPG–Th2/GATA3/IL4/CD4 link. So the dossier is not built on fabricated sources, and MPG being non-essential with epigenetic-adjacent functions means a screen hit is not obviously a technical artifact — I agree with likely_artifact=false.

Strongest objection: the Th2-SELECTIVE mechanism is entirely speculative and the selectivity argument is internally weak. (1) Every functional AAG study cited is in stem cells/brain — zero T-cell, CD4, or cytokine-locus evidence exists. (2) The '5hmC-reader at demethylation-dependent loci' logic actively undercuts selectivity: TET/5hmC-driven demethylation is critical at the IFNG (Th1) and FOXP3 (Treg) loci as much as at IL4/IL5/IL13, so a genuine 5hmC-reader defect would be expected to perturb Th1 too, not spare it — the 'bypasses T-bet / not driving Th1' framing has no support. (3) A monofunctional glycosylase loss more plausibly causes generic BER/genotoxic stress in highly proliferative differentiating T cells (broad dampening) rather than a clean Th2-specific effect. (4) Viable, fertile, overtly normal knockouts with no reported allergy/asthma/immune phenotype argue against MPG being a dedicated Th2 regulator. The hypothesis is a plausible-sounding but unsupported mechanistic chain; I downgrade from 'uncertain' to 'unlikely' as a selective Th2 suppressor, while agreeing it is not a technical artifact.

STK35 — mechanism dossier

STK35 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.11, magnitude-controlled residual = -1.82, condition = Stim8hr, class = kinase · druggable

Final verdict (post–adversarial review): UNLIKELY · investigator: uncertain (low conf) · citations: partially_supported

Hypothesis

Losing STK35 could lower the Th2 program indirectly by blunting the AKT/glycolysis-driven proliferative and metabolic burst that GATA3/IL4 induction disproportionately depends on, rather than by acting as a dedicated Th2 transcriptional regulator.

Mechanism

STK35 (STK35L1) is a nuclear/nucleolar Ser/Thr kinase with no reported function in T cells, Th2/Th1 differentiation, GATA3, IL4, or TCR/cytokine signaling. Its documented roles are cell-cycle control (inhibits G1-to-S via CDKN2A/p16, binds nuclear actin), endothelial migration/angiogenesis, spermatogenesis, and promotion of glycolysis plus suppression of apoptosis through AKT signaling (colorectal cancer). The only route to the Th2 program is therefore indirect: AKT-mTOR and glycolytic flux support proliferation and biosynthesis, and Th2/GATA3/IL4 induction upon stimulation is more metabolically and proliferation-demanding than the (here already flat) Th1 readout. Reduced STK35 could thus preferentially shrink the higher-dynamic-range Th2 output without any GATA3-specific mechanism.

Bypasses T-bet / Th1? True — There is no evidence STK35 drives Th1/T-bet, so its loss is not expected to induce IFNG/T-bet. The Th1-flat pattern is trivially consistent with a global/metabolic dampener (both arms suppressed or Th1 already low), not with active Th1 skewing. So it plausibly avoids the 'it's just a Th1 driver' explanation, but this bypass is trivial and does not by itself demonstrate genuine Th2-selective regulation.

Evidence

• STK35L1 is a nuclear/nucleolar Ser/Thr kinase that binds nuclear actin and regulates endothelial cell cycle (G1-to-S via CDKN2A/p16), migration and angiogenesis; no T-cell/Th2 role reported.

source: STK35L1 Associates with Nuclear Actin and Regulates Cell Cycle and Migration of Endothelial Cells, PLOS ONE 2011, PMC3024402

• STK35 promotes glycolysis and inhibits apoptosis via the AKT signaling pathway and is ubiquitinated by NEDD4L (colorectal cancer chemoresistance) — a metabolic/survival/proliferation profile.

source: STK35 Is Ubiquitinated by NEDD4L and Promotes Glycolysis and Inhibits Apoptosis Through Regulating the AKT Signaling Pathway, Front Cell Dev Biol 2020, PMC7578231

• Reviewed functions of STK35/STK35L1 span cell cycle, migration, angiogenesis, DNA damage response and spermatogenesis/gametogenesis; the locus also encodes an oxidative-stress-responsive lncRNA. No immune/Th2 function is described.

source: The Multifaceted Role of STK35/STK35L1 in Human Diseases, Kinases and Phosphatases 2025, doi:10.3390/kinasesphosphatases3020012; PMC6124569

• GATA3 is the master Th2 regulator (opens IL4/IL5/IL13 locus, represses IFNG); no upstream link to STK35 exists in the literature, so any STK35 effect on this axis would be indirect.

source: GATA-3 promotes Th2 responses..., Cell Research 2006, nature.com/articles/7310002

Proposed confirming experiment

Arrayed CRISPRi knockdown of STK35 (plus non-targeting and a known Th2-selective control such as GATA3) in primary human CD4+ T cells under matched Th2- and Th1-polarizing conditions. Primary readout: intracellular GATA3/IL4/IL5/IL13 vs T-bet/IFNG by flow at 8hr and after full polarization. Critically, co-measure the global-dampening confounders on the SAME cells: proliferation (CellTrace dilution), viability, live-cell yield, and metabolic state (glucose uptake / Seahorse ECAR, plus pAKT/pS6). Test whether the IL4/GATA3 deficit persists after normalizing per-live-cell and per-division, and whether supplementing exogenous IL-2 or restoring glycolysis/AKT rescues Th2. Genuine selective suppression = GATA3/IL4 drop that survives proliferation/viability/metabolic normalization while IFNG and total activation (CD25/CD69, blast size) are spared; a metabolic-dampener artifact = Th2 loss that tracks with reduced division/glycolysis and is rescued by metabolic support.

Adversarial review

• Citations check: partially_supported
• Agrees artifact call (True): True
• Notes: All four sources are real and verified (PMC3024402, PMC7578231, 2025 Kinases&Phosphatases review, PMC6124569, GATA3 Cell Research 2006/7310002); each broadly supports its claim. One unverified embellishment: the "(G1-to-S via CDKN2A/p16)" detail is not confirmed by the 2011 PLOS ONE abstract, hence partiallysupported. Strongest objection: the pro-proliferation mechanism is internally inconsistent — in endothelial cells STK35L1 depletion ACCELERATES G1-to-S (inhibitory role), opposite to the CRC pro-glycolysis/AKT role, so "blunting a proliferative burst" is direction-inconsistent and context-dependent. There is zero immune/T-cell evidence; a generic AKT/glycolysis/proliferation dependency is the classic proliferation/metabolic-fitness confound in Th2-selective screens (Th2 cells are highly glycolytic/proliferative), making apparent selectivity most likely an artifact rather than dedicated Th2 transcriptional regulation. The "bypasses T-bet" concern is moot since no evidence touches T-bet and the hypothesis is itself indirect/metabolic. Mild mitigation: an AKT1->GATA3 phosphorylation axis (Nat Commun 2016) gives a thin plausible indirect route, so not fully dismissable. I agree with likelyartifact=true but rate the selective-Th2-suppressor claim as unlikely rather than merely uncertain.

KDM2A — mechanism dossier

KDM2A — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.22, magnitude-controlled residual = -1.72, condition = Stim48hr, class = TF

Final verdict (post–adversarial review): LIKELYARTIFACT · investigator: likelyartifact (medium conf) · citations: supported

Hypothesis

Because KDM2A is tightly coupled to PI3K/mTOR/SGK1-driven rRNA transcription, ribosome biogenesis and proliferation, its knockdown most likely dampens the highly proliferation/autocrine-amplified Th2 program more than the low-baseline Th1 program in a Th2-polarizing assay, rather than acting as a bona fide selective GATA3/IL4 regulator.

Mechanism

There is no primary literature linking KDM2A to GATA3, IL4, or Th2 differentiation, and its known biochemistry predicts the OPPOSITE of a Th2 activator: KDM2A is an H3K36me1/2 eraser that binds unmethylated CpG islands and REPRESSES transcription, it negatively regulates NF-kB (demethylates p65) and promotes proteasomal degradation of TCF/LEF (Wnt). Losing KDM2A would raise H3K36me2 at cis-regulatory elements (opposing NSD1/2 that support enhancer H3K27ac), increase NF-kB, and stabilize TCF1/LEF1 (a pro-GATA3/pro-Th2 factor) — each of which predicts UNCHANGED or INCREASED Th2, not a decrease. The only route consistent with reduced Th2 upon KD is indirect: KDM2A is an essential gene (embryonic-lethal KO) governing PI3K/mTOR/SGK1-controlled rDNA transcription and ribosome biogenesis, so KD lowers proliferative/biosynthetic capacity and preferentially collapses the STAT6→GATA3→IL4 autocrine amplification loop while leaving a flat Th1 baseline flat. That is a global-dampening mechanism, not selective Th2 regulation.

Bypasses T-bet / Th1? True — None of KDM2A's known activities induce T-bet/Th1: loss increases NF-kB, Wnt/TCF-LEF, and H3K36me2, and impairs proliferation — so the Th1-flat readout is not driven by Th1 induction. However, the flatness is better explained by a low Th1 baseline plus generic dampening than by a genuine Th2-specific circuit, so bypassing T-bet here is not evidence of a real selective suppressor.

Evidence

• KDM2A/FBXL11 is an H3K36me1/2 demethylase that binds unmethylated CpG islands and acts as a transcriptional REPRESSOR, predicting the opposite of a Th2 activator.

source: KDM2A - Wikipedia; EMBO J 2010 (KDM2A regulator of rRNA transcription), https://en.wikipedia.org/wiki/KDM2A

• KDM2A is a negative regulator of NF-kB (demethylates p65) and its loss enhances inflammatory NF-kB signaling — loss would not lower an inflammatory/Th program.

source: Reviews of FBXL11/KDM2A NF-kB regulation (liver-specific Kdm2a deficiency enhances NF-kB), summarized via KDM2A literature

• KDM2A promotes proteasomal degradation of TCF/LEF; since TCF1/LEF1 promote GATA3 and Th2, losing KDM2A should raise, not lower, the Th2 program.

source: Lysine Demethylase KDM2A Promotes Proteasomal Degradation of TCF/LEF, PMC10670284, https://pmc.ncbi.nlm.nih.gov/articles/PMC10670284/

• KDM2A is essential (embryonic-lethal KO, represses cell-cycle regulators) and controls rDNA/rRNA transcription and ribosome biogenesis downstream of PI3K/mTOR/SGK1 — a proliferation/growth axis prone to global-dampening screen artifacts.

source: Fbxl11/Kdm2a essential in embryonic development (PubMed 25463925); KDM2A regulator of rRNA transcription (EMBO J 2010, doi:10.1038/emboj.2010.56); SF-KDM2A rDNA/proliferation (PubMed 28350064)

• H3K36me2 deposited by NSD enzymes activates cis-regulatory elements (H3K27ac); KDM2A erases H3K36me2, so its loss would increase enhancer activity — opposite of suppressing Th2 genes.

source: Systematic perturbations of SETD2/NSD1/2/3/ASH1L on H3K36 methylation, bioRxiv 2023.09.27.559313

Proposed confirming experiment

Perform arrayed CRISPRi KD of KDM2A (plus a magnitude-matched non-Th2 essential control and a scramble) in primary human CD4+ T cells under parallel Th2-polarizing and Th1-polarizing conditions at 48hr stim. Read out (a) Th2 (GATA3, IL4/IL5/IL13 by intracellular flow + qPCR) vs Th1 (T-bet, IFNG) as the selectivity test, AND critically (b) global-dampening covariates on the SAME cells: CellTrace proliferation (division index), viability, cell size (FSC), and bulk translation by O-propargyl-puromycin (OPP) incorporation, plus rRNA/45S pre-rRNA levels. Genuine selective suppression requires Th2 loss that persists after regressing out proliferation/OPP/viability and is NOT reproduced by the magnitude-matched control. Add two orthogonal checks: (i) rescue with catalytically-dead (JmjC-mutant) vs WT KDM2A to test whether the phenotype is demethylase-dependent, and (ii) CUT&RUN for H3K36me2 at the GATA3/Th2 LCR/IL4 locus to confirm direction — if KD raises H3K36me2 yet Th2 falls, the effect is indirect/global rather than a direct chromatin mechanism.

Adversarial review

• Citations check: supported
• Agrees artifact call (True): True
• Notes: All five citations are real and support their claims: KDM2A/FBXL11 is a CpG-island-binding H3K36me1/2 demethylase and repressor (EMBO J 2010, PMC2876952; NAR 2019); rRNA/ribosome-biogenesis regulator; embryonic-lethal essential gene repressing cell-cycle regulators (PubMed 25463925); NF-kB p65 negative regulator; and SCF/neddylation-dependent degrader of TCF/LEF (PMC10670284). The internal logic is coherent and cuts against a bona fide direct Th2 activator: as a repressor of NF-kB and a degrader of TCF1/LEF1 (pro-GATA3), KDM2A loss should if anything DE-repress/raise the Th2 program, so a KD-reduces-Th2 screen hit is hard to reconcile with direct GATA3/IL4 activation and is better explained by loss of an essential proliferation/rRNA/ribosome-biogenesis gene. Two caveats weaken confidence but do not overturn the artifact call: (1) NAR 2019 shows KDM2A's demethylase activity is largely dispensable and KDM2B is the dominant repressor, so the repressor framing is overstated; (2) essential/ribosome-biogenesis dampening predicts NON-selective loss of both Th1 and Th2 — the assertion that Th2 is uniquely more proliferative/autocrine-amplified than the equally proliferative Th1 effector program is unsupported hand-waving and is the strongest objection to a clean proliferation-artifact story. Critically, ALL cited mechanism is from non-T-cell systems (HEK293T, breast/gastric cancer, embryo, liver); there is zero direct CD4 T-cell evidence for a GATA3/IL4 role. Net: no support for a bona fide selective Th2 regulator; essentiality + wrong-direction repressor biology make likely_artifact the best-supported call.

MIA2 — mechanism dossier

MIA2 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.01, magnitude-controlled residual = -1.71, condition = Rest, class = other

Final verdict (post–adversarial review): LIKELYARTIFACT · investigator: unlikely (medium conf) · citations: partiallysupported

Hypothesis

If MIA2 knockdown selectively lowered the Th2 program, it would most plausibly be by degrading the ER-export/secretory capacity that autocrine IL-4/STAT6/GATA3 feedback (and the high secretory load of Th2 cells) depends on, rather than by any Th2-specific transcriptional role.

Mechanism

MIA2 has no described role in TCR signaling, STAT6, GATA3 induction, or the IL4/IL5/IL13 locus. It encodes the TANGO1-family ER-exit-site proteins TALI/cTAGE5, which partner with TANGO1 (MIA3) and Sec12/Sar1/COPII to export bulky cargo (procollagen, ApoB chylomicrons/VLDL) from the ER. The only speculative Th2 route is indirect: Th2 differentiation runs on an IL-4 -> IL-4R -> STAT6 -> GATA3 autocrine positive-feedback loop, so a cell-wide secretory-pathway defect could blunt establishment of that loop. However, IL-4/IL-5/IL-13 are small conventional-secretion cargoes that do not require TANGO1/TALI bulky-cargo machinery, so this link is mechanistically weak, and it would predict a broad secretory/fitness defect rather than a clean transcriptional Th2 collapse.

Bypasses T-bet / Th1? True — An ER-export/secretory-pathway gene has no known activity that would drive T-bet or IFNG. If MIA2 loss impairs anything, it is secretory capacity/fitness, which would if anything dampen rather than divert toward Th1. So the observed Th1-flat, Th2-down pattern is consistent with a non-Th1-driving (secretory-stress or magnitude) effect rather than active Th1 induction. This is a point in favor of it not being a mislabeled Th1 driver, but it does not establish a Th2-specific mechanism.

Evidence

• MIA2 encodes the ER-exit-site proteins TALI and cTAGE5, which cooperate with TANGO1 (MIA3), Sec12 and COPII (Sar1) to export bulky cargo such as pre-chylomicrons/VLDL and collagen from the ER; it is a secretory-pathway component, not a transcription factor.

source: Santos et al., TANGO1 and Mia2/cTAGE5 (TALI) cooperate to export bulky pre-chylomicrons/VLDLs from the ER, eLife 2016 (PMC4862334); OMIM 602132 MIA2

• MIA2 localizes to ER exit sites (transitional ER) and shows low immune-cell specificity; it is expressed most abundantly in glandular/hepatic and neuronal cells, not enriched in T cells.

source: Human Protein Atlas, MIA2 (ENSG00000150527), immune cell and tissue expression

• MIA2 transcription in hepatocytes is driven by IL-6/STAT3 and TGF-beta/SMAD, not by TCR- or Th2-associated cytokines; no literature links MIA2/TALI/cTAGE5 to GATA3, STAT6, or the IL4/IL5/IL13 program.

source: GeneCards MIA2; literature search (PubMed) returned no MIA2-Th2/GATA3 associations

• The Th2 program is controlled by an IL-4 -> STAT6 -> GATA3 autocrine feedback loop; GATA3 (not any secretory factor) opens the IL4/IL5/IL13 locus and represses IFN-gamma.

source: Zhu et al., GATA3 and the T-cell lineage, Immunol Rev 2010 (PMC2998182); Nawijn/Zhu, Cell Research 7310002

Proposed confirming experiment

Arrayed CRISPRi knockdown of MIA2 (plus non-targeting and a positive-control GATA3 guide) in naive human CD4+ T cells, split into parallel Th2-polarizing (IL-4 + anti-IFNg/anti-IL-12) and Th1-polarizing (IL-12 + anti-IL-4) cultures. Primary readout: GATA3 and IL4/IL5/IL13 vs T-bet/IFNG by intracellular flow and qPCR/RNA-seq to confirm the arm-selective transcriptional effect. Critically add specificity controls that separate a real Th2 regulator from a secretory/global-dampening artifact: (1) viability, proliferation (CellTrace), and total mRNA/blast size to catch fitness/magnitude effects; (2) a general secretion reporter (e.g., secreted Gaussia luciferase or bulk supernatant protein) and an ER-stress marker (XBP1 splicing, BiP) to test whether MIA2 KD broadly impairs secretion/induces ER stress; (3) rescue by exogenous IL-4 -> if adding saturating IL-4 restores GATA3, the effect is upstream/secretory rather than a dedicated Th2 transcriptional role. A genuine selective suppressor should lower GATA3/IL4 with intact viability, normal bulk secretion/no ER stress, and not be rescued simply by exogenous IL-4.

Adversarial review

• Citations check: partially_supported
• Agrees artifact call (True): True
• Notes: Confirmed all four claims against primary sources. MIA2 encodes TALI/cTAGE5, an ER-exit-site COPII accessory that cooperates with TANGO1 (MIA3) to export bulky cargo (pre-chylomicrons/VLDL, collagen) — a secretory component, not a TF (Malhotra et al. 2016; OMIM 602132). CITATION ERROR: this paper is in Journal of Cell Biology (213:343-354, PMID 27138255), NOT eLife as cited, though the PMC ID (PMC4862334) is correct and real. HPA confirms glandular/hepatic/neuronal expression, liver-abundant, low immune specificity. Hepatocyte MIA2 is driven by IL-6/STAT3 + TGF-beta/SMAD (Bosserhoff PMID 12586826), with no literature link to GATA3/STAT6/IL4-5-13. The task's 'bypasses T-bet' question does not apply — the dossier makes no T-bet argument. Strongest objection to a genuine hit: knocking out an essential ER-export gene in the most secretion-heavy Th subset (high autocrine IL-4 load) plausibly produces a selective-Th2 dropout as a secretory-capacity confounder, not a Th2-specific role — which is exactly an artifact signature and would also be expected to broadly impair effector/proliferative function. I agree with unlikely/likely_artifact.

RAB21 — mechanism dossier

RAB21 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.76, magnitude-controlled residual = -2.94, condition = Rest, class = other

Final verdict (post–adversarial review): LIKELY_ARTIFACT · investigator: uncertain (low conf) · citations: supported

Hypothesis

Loss of RAB21 could preferentially blunt the highly glycolysis/mTOR-dependent Th2 (GATA3/IL4) program by mis-sorting surface GLUT1 and impairing receptor recycling, without inducing T-bet/Th1.

Mechanism

RAB21 is an early-endosome Rab GTPase governing retromer-mediated recycling of GLUT1/SLC2A1, integrin endo/exocytosis, EGFR/receptor recycling, and autophagy. Its KD lowers surface GLUT1 and glucose uptake, dampening cellular energy homeostasis; because Th2 induction of GATA3 depends heavily on sustained TCR/PI3K/mTOR/glycolytic signaling, reduced glucose flux or impaired cytokine-receptor (e.g. IL-2/IL-4R->STAT5) recycling could preferentially attenuate the Th2 program. However, none of these routes is Th2-specific in the literature, and Th1 is also glycolysis-dependent, so a genuinely selective effect on GATA3/IL4 over IFNG is speculative. No published work connects RAB21 to GATA3, IL4, or Th2/Th1 fate.

Bypasses T-bet / Th1? True — RAB21's known roles are in trafficking/metabolism/autophagy, not transcriptional Th lineage priming; there is no route by which its loss would induce T-bet or drive Th1. Any Th1-flat selectivity would arise from Th1 being less sensitive to the perturbation, not from T-bet induction. So the phenotype is not explained by a Th1-driver mechanism.

Evidence

• RAB21 controls retromer-mediated recycling of glucose transporter GLUT1/SLC2A1; KD mis-sorts GLUT1 to lysosomes, lowers surface GLUT1, and disrupts cellular energy homeostasis and autophagy.

source: RAB21 controls autophagy and cellular energy homeostasis by regulating retromer-mediated recycling of SLC2A1/GLUT1, Autophagy 2022/2023, PMC10012929

• RAB21 is an early-endosome small GTPase that regulates integrin internalization, cell adhesion, migration and cytokinesis via integrin endo/exocytic traffic.

source: Small GTPase Rab21 regulates cell adhesion and controls endosomal traffic of beta1-integrins, J Cell Biol 2006, PMC2063892; Integrin trafficking regulated by Rab21 is necessary for cytokinesis, Dev Cell 2008, PMID 18804435

• RAB21 acts as a general regulator of cell-surface receptor identity (broad, non-lineage-specific trafficking role) and affects EGFR internalization/degradation.

source: Phenotypic screens for SIRPA reveal RAB21 as a general regulator of macrophage surface identity, Cell Reports 2025, S2211-1247(25)00692-8

• No primary literature links RAB21 to GATA3, IL4, or Th2/Th1 differentiation; Th2/GATA3 induction is instead driven by TCR + STAT5/STAT6 and is glycolysis/mTOR-dependent.

source: Transcriptional regulation of Th2 cell differentiation, PMC3477614; T cell receptor-dependent translational control of GATA-3 enhances Th2 differentiation, PMC3993005 (RAB21 absent from Th2 literature)

Proposed confirming experiment

CRISPRi KD of RAB21 (plus non-targeting and a known Th2-essential positive control e.g. GATA3) in naive human CD4+ T cells split into parallel Th2-polarizing and Th1-polarizing arms. Read out GATA3/IL4/IL5/IL13 vs T-bet/IFNG by intracellular flow and qPCR, with a sgRNA rescue (RAB21 re-expression). Crucially, include global-dampening controls run in the same cells: proliferation (CTV dilution), viability, surface GLUT1 and glucose uptake (2-NBDG), and Seahorse ECAR/OCR. Selective Th2 suppression is supported only if GATA3/IL4 drop disproportionately to IFNG AND is not accounted for by reduced proliferation/viability or a proportional metabolic collapse; a glucose/pyruvate add-back arm tests whether the effect is purely metabolic. Rescue restoring the phenotype confirms specificity.

Adversarial review

• Citations check: supported
• Agrees artifact call (True): True
• Notes: All four cited sources are real and accurately represented. Verified: (1) Autophagy 2022, PMC10012929 (PMID 35993307) - RAB21 depletion mis-sorts SLC2A1/GLUT1 to lysosomes, lowers glucose uptake, activates AMPK-ULK1, increases autophagy; also sensitizes cancer cells to energy stress. Correct. (2) J Cell Biol 2006 PMC2063892 (PMID 16754960) - Rab21/beta1-integrin adhesion/migration. Correct. (3) Dev Cell 2008 (PMID 18804435) - Rab21 integrin trafficking required for cytokinesis; loss causes multinucleate cells. Correct. (4) Cell Reports 2025 S2211-1247(25)00692-8 (PMID 40580479) - RAB21 as GENERAL regulator of macrophage surface identity, remodels whole surfaceome, reduces FcgR. Correct.

Strongest objection: the evidence directly refutes selectivity rather than supporting it. RAB21 is a broad housekeeping endosomal/retromer trafficking GTPase whose loss (a) globally remodels the surfaceome (paper's own framing: 'general regulator of surface identity'), (b) impairs cytokinesis producing multinucleate cells - which would blunt proliferation of ALL activated T cells including Th1, and (c) lowers global glucose uptake/energy. The 'preferentially blunts Th2 because Th2 is uniquely glycolysis/mTOR-dependent' premise is weak: Th1 (T-bet) and Th17 are equally or more glycolysis/mTORC1-dependent, so a GLUT1/energy hit would dampen Th1 as much as Th2 rather than sparing it - undercutting the 'without inducing T-bet/Th1' claim. No primary literature links RAB21 to GATA3/IL4/T-bet (investigator concedes this). A pan-proliferation/energy defect masquerading as Th2-selective is exactly the artifact signature. I agree with likely_artifact=true.

SSR1 — mechanism dossier

SSR1 — selective Th2-suppressor mechanism dossier

Screen signal: Th2-arm z = -2.58, magnitude-controlled residual = -2.13, condition = Rest, class = other

Final verdict (post–adversarial review): LIKELYARTIFACT · investigator: likelyartifact (medium conf) · citations: partially_supported

Hypothesis

Loss of SSR1 impairs ER translocation/secretion of the heavily secreted Th2 cytokines and could collapse the autocrine IL-4→STAT6→GATA3 positive-feedback loop, indirectly lowering the Th2 mRNA program more than the less autocrine-dependent Th1 program.

Mechanism

SSR1 (signal sequence receptor alpha / TRAP-alpha) is a subunit of the heterotetrameric translocon-associated protein (TRAP) complex that docks with Sec61 and promotes co-translational translocation and N-glycosylation quality control of secretory and membrane proteins, especially those with weak signal sequences (e.g., proinsulin biosynthesis is TRAP-dependent). It is not a transcription factor and has no reported role in GATA3/IL4 gene regulation. The only route to the transcriptional Th2 program is indirect: Th2 effector cytokines (IL4/IL5/IL13) are secreted glycoproteins with high ER-translocation load, so SSR1 knockdown could reduce IL-4 secretion and thereby weaken the autocrine IL-4/IL-4Ra/STAT6/GATA3 feed-forward loop that sustains the Th2 program. However, Perturb-seq reads mRNA, and this loop is weak in the non-polarizing Rest condition, making a genuinely Th2-selective transcriptional effect unlikely relative to broad secretory/ER-stress dampening.

Bypasses T-bet / Th1? True — SSR1 is a secretory-pathway housekeeping gene with no known link to T-bet/Th1 differentiation, so its knockdown would not induce a Th1 program. The Th1-flat, Th2-down signature is therefore consistent, but this pattern is equally (and more parsimoniously) explained by broad dampening of the most secretory/highly-expressed effector genes rather than by a Th2-specific regulatory function.

Evidence

• SSR1 encodes TRAP-alpha, a subunit of the ER translocon-associated protein (TRAP/SSR) complex that associates with Sec61 and mediates co-translational translocation of nascent secretory/membrane proteins.

source: GeneCards SSR1; OMIM 600868 (SIGNAL SEQUENCE RECEPTOR, ALPHA; SSR1)

• The TRAP complex is required for efficient biosynthesis of secreted proteins; TRAP deficiency limits proinsulin and insulin biosynthesis, demonstrating a general secretory-load dependence rather than lineage specificity.

source: Huang et al., FASEB J 2021, PMC8106808 'Deficient endoplasmic reticulum translocon-associated protein complex limits the biosynthesis of proinsulin and insulin'

• The TRAP complex regulates quality control of N-linked glycosylation during ER stress, tying SSR1 loss to global UPR/ER-stress phenotypes.

source: Science Advances 2020, doi:10.1126/sciadv.abc6364

• The Th2 transcriptional program is driven by STAT6 and GATA3 with an autocrine IL-4/IL-4Ra positive-feedback loop; none of these nodes is SSR1 or the secretory apparatus.

source: Transcriptional regulation of Th2 cell differentiation, PMC3477614; GATA3 review PMC3123974

• No primary literature links SSR1/TRAP-alpha to CD4+ T-cell differentiation, GATA3/IL4 regulation, or TCR/cytokine signaling; SSR1 hits in cancer datasets are as a broadly expressed secretory/proliferation-associated gene.

source: PubMed/PMC searches; SSR1 hepatocellular carcinoma biomarker study PMC11543030 (proliferation/immune-infiltration correlate, not a T-cell fate regulator)

Proposed confirming experiment

Perform arrayed CRISPRi SSR1 knockdown in primary human CD4+ T cells under explicit Th2-polarizing conditions and, in parallel, measure (a) intracellular vs secreted IL-4/IL-5/IL-13 protein (to distinguish a secretion/translocation defect from a transcriptional defect), (b) GATA3 and STAT6-phospho by flow, and (c) IFNG/T-bet, alongside a global secretory-load readout (total secretome, XBP1s/UPR markers, and viability/proliferation via CellTrace). Include a magnitude-matched panel of essential secretory-pathway controls (e.g., SEC61A1, SRP54, other TRAP subunits SSR2/3/4). If SSR1 KD lowers Th2 cytokine mRNA and GATA3 selectively while UPR markers, proliferation, and non-Th2 secreted proteins are unaffected, it is a genuine selective suppressor; if IL-4 mRNA/GATA3 are preserved but secreted IL-4 drops, or if UPR/proliferation defects and the SSR2/3/4/SEC61 controls phenocopy the Th2-down signature, it is a secretory-load/global-dampening artifact.

Adversarial review

• Citations check: partially_supported
• Agrees artifact call (True): True
• Notes: Confirmed: SSR1 = TRAP-alpha/signal sequence receptor alpha, OMIM 600868, subunit of the ER translocon-associated (TRAP/SSR) complex mediating co-translational translocation (GeneCards/OMIM verified). FASEB 2021 PMC8106808 (Huang et al.) is real and supports that TRAP deficiency limits proinsulin/insulin biosynthesis — a general secretory-load dependence, not lineage specificity (corroborated by follow-up JCI/bioRxiv Trapα-knockout work). Science Advances sciadv.abc6364 is real BUT the N-glycosylation/ER-stress QC effect is specifically attributed to SSR3/SSR4 knockouts, not SSR1/TRAP-alpha — so tying 'SSR1 loss' to that phenotype is a subunit-level overreach (hence partiallysupported). Th2 STAT6/GATA3/autocrine-IL4 program (PMC3477614, PMC3123974) verified; no primary literature links SSR1 to CD4 T-cell fate, GATA3/IL4, or TCR signaling — my own search returned zero SSR1–Th2 hits, confirming absence. Agree with likelyartifact: SSR1 is a broadly expressed housekeeping secretory-apparatus gene with no T-cell-fate role; knockdown would trigger global UPR/ER stress and broadly dampen the cell rather than selectively suppress Th2. Strongest objection to the investigator's OWN mechanistic hypothesis (which I do not need to endorse): the autocrine-collapse story is a speculative just-so rationale — Th1 cells also heavily secrete IFN-gamma through the same translocon, so a general secretory defect does not cleanly predict Th2 selectivity; the apparent selectivity is more parsimoniously explained as a viability/secretory-stress artifact than genuine lineage selectivity. Artifact call stands.

Synthesis

Selective Th2 Suppressor Atlas — synthesis

UPDATE (hardened re-analysis, supersedes the optimistic read below). Running the review's

recommended fixes changed the conclusion. (1) Competitive rank-based scoring collapses the arm

correlation +0.54 → +0.18 — most of the "global-magnitude confound" was a scoring artifact.

(2) A permutation null shows the 2-arm selective set (420 under competitive scoring) is **not

enriched over random gene sets** of the same size (expected ~526; empirical FDR ≈ 1.25) — genuine

selective suppressors are no more common than chance, because Th1/Th2 are genuinely reciprocal.

(3) GATA3 fails: with the matched Th1-vs-Th0 arm, GATA3 KD drives Th1 to +2.8 (a Th1-skewer,

as canonical biology predicts); the earlier "top selective hit" was a mean-of-z + noisy-Ota-arm

artifact. Only 7 genes pass the strict multi-arm + concordance gate and none are credible at FDR≥1.

Honest conclusion: the naive selective-suppressor atlas does not survive rigorous statistics.

The real deliverables are the calibrated negative result, the reproducible pipeline, and the

agentic review/QC infrastructure. Next step before any claim: a functional IL-4/5/13 readout and a

control-guide-based FDR. See hardenedstats.json / notebooks/10hardened.py.

Data: Zhu & Dann et al. 2025 genome-scale CRISPRi Perturb-seq, primary human CD4+ T cells

(GWCD4i.DE_stats.h5ad, 33,983 perturbation×condition, streamed from the public Virtual Cells S3).

Signature: Ota et al. 2021 Th2-vs-Th1, split into a Th2-up arm (505 measured genes) and Th1-up arm (802).

What we set out to do

Separate selective Th2 suppressors (KD lowers Th2, leaves Th1 flat — the allergy target) from

Th1-skewers (KD just flips the cell toward Th1). The paper's single bidirectional Th2−Th1 axis

cannot make this distinction; a two-arm 2D decomposition can.

Results

1. Reproduced the teammate's result from the authoritative matrix. Arm correlation +0.54 (their +0.49);

the arms move together, so the dominant confound is global magnitude (sick-cell dampening), not skewing.

The Th1 arm therefore doubles as a built-in control. Selective ≈127, true skewers ≈10 at the default gate.

1. Calibration holds. GATA3 (Th2 master regulator) is the #1 selective hit and #1 in the scored shortlist; RARA #2.
2. The known-regulator list is directionally mixed — and the 2D method resolves it. Th2 activators

(GATA3, RARA, STAT6-trending) recover as suppressors; NuRD/chromatin repressors (MTA2, CHD4, SETDB1) and ICOS

correctly appear as Th1-inducers (Th1 arm up), not suppressors. IL4/IL4R produce ~no transcriptional footprint.

This is exactly the direction the 1D axis conflates.

1. Context-dependence is the novel cut. Of selective hits, ~60 act only at Rest and ~49 only under stimulation;

even GATA3 is selective at Rest/Stim8hr but not Stim48hr. The paper's per-condition 1D coefficients cannot

express this as selectivity.

1. Two false-positive sources handled honestly. A hard KD-confirmed + real-footprint (≥10 DE genes) gate

(Haltavey's "did it actually knock down?" concern) plus magnitude-controlled selectivity (regress Th2 arm on the

Th1 arm) collapsed 127 noisy hits to 21 high-confidence genes.

The agentic mechanism layer is the differentiator

For the top 10 novel candidates, an investigator agent (web research) proposed a mechanistic route to the Th2

program bypassing T-bet + a confirming experiment; a skeptic agent adversarially verified citations and logic.

Outcome: after review, 0/10 survived as "plausible" — 4 uncertain, 2 unlikely, 4 likely-artifact. The only

non-artifact leads worth following up were ARNT and ELAVL1 (real, still-unverified Th2 links). The rest of

the top-scored list (AHSA1, RAB21, SSR1, KDM2A, MPG…) are global-effect / housekeeping false positives that the

numeric score under-penalized. The mechanism layer performs the triage the score cannot — separating real Th2

biology from sick-cell artifacts, with every claim adversarially checked.

Honest limitations

• Arm construction (unweighted signature mean) differs slightly from the teammate's; cross-check against their .npy.
• 21 high-confidence genes still contain artifacts; mechanistic triage narrows the real leads to ~2 of the top 10.
• Selectivity ≠ efficacy/safety; Th2 has protective roles (anti-helminth, barrier). This is transcriptomic screen

evidence, not in-vivo.

Artifacts produced

arms.parquet · selectivetiered.csv · planeplot.svg · recall_overlay.svg ·

conditionselectivitymatrix.csv · realvsfpshortlist.csv · mechanismdossiers/*.md · live dashboard.

Independent review report

Independent review report — Selective Th2 Suppressor Atlas

Two-part self-review: (A) a data/QC audit against the dataset's own richer QC columns, and

(B) a 6-agent parallel literature + methods audit (each finding adversarially sourced).

Part A — Data / QC self-audit (issues found in our own pipeline, now fixed)

1. We ignored the neighboringgeneKD flag. CRISPRi can silence a gene adjacent to the

intended target. Our former #1 novel hit AHSA1 is neighbor-flagged in Stim8hr (its selective

Rest row is clean, but the gene is suspect). Fix: neighbor-flagged perturbation×conditions are

now dropped from the high-confidence set.

1. We used the wrong donor-reproducibility column. The suppl-table crossdonor was NaN for 78%

of hits (median 0.15). The correct donorcorrelationhits_mean has median 0.81 on our set —

our hits are more donor-reproducible than first reported. Fix: switched the donor metric;

ARNT (0.87), ELAVL1 (0.80), GATA3 (0.82) now confirmed reproducible.

1. Single-guide hits weren't penalized. ARL1 and MPG rest on one guide (singleguideestimate).

Fix: single-guide candidates are down-weighted.

1. The selective count is threshold-soft (49 / 127 / 199 as the Th1 gate loosens). The honest

robust set is the core ~12–23 genes, not 127.

Part B — Literature & methods audit (6 parallel agents)

Foundational immunology — well supported

Holds up:

• Th1 and Th2 are reciprocal, mutually antagonistic CD4+ effector lineages: their cytokine profiles are dictated by the mutually exclusive expression of the master transcription factors T-bet (Th1) and GATA3 (Th2), which directly cross-repress (T-bet binds and sequesters GATA3, blocking it from Th2 cytokine promoters).
• GATA3 is the master regulator of Th2 differentiation: enforced GATA3 drives Th2 even under Th1 conditions, and GATA3-deficient cells fail to produce IL-4/IL-5/IL-13. This supports GATA3 emerging as the top selective Th2 suppressor in our screen.
• The IL4/STAT6 axis sits upstream of GATA3: IL-4 -> IL-4Ra -> STAT6 phosphorylation induces high GATA3, and IL-4/IL-4Ra/STAT6 signaling forms a positive feedback loop stabilizing the Th2 phenotype. STAT6 is correctly paired with GATA3 as a Th2 driver.
• Th2 markers IL4, IL5, IL13 (the IL5-IL13-IL4 cytokine locus bound/modified by GATA3) and CCR4 are validated: IL-4/IL-5/IL-13 production is restricted to the CCR4+ memory CD4+ population; CCR4 is a useful surface marker of circulating Th2 effectors.
• Th1 markers are validated: T-bet/TBX21 (T-box factor on chr17) controls IFN-gamma; IL-12 signaling via STAT4 stabilizes T-bet/the Th1 phenotype; CXCR3 marks circulating IFN-gamma-producing Th1 effectors.

Concerns:

• 'Master regulators = GATA3 + STAT6' slightly conflates two roles: GATA3 is the lineage-defining transcription factor, whereas STAT6 is the upstream signal transducer of the IL-4 cue. GATA3 can drive Th2 commitment STAT6-independently via autoactivation (Stat6-deficient cells reconstituted by GATA3), so they are not interchangeable co-master TFs.
• IL4R (IL-4Ra) is a weaker/looser Th2 marker than IL4/IL5/IL13/CCR4: IL-4Ra is broadly expressed across many hematopoietic and non-immune cells (it is the drug target precisely because it is shared and widespread), so its inclusion in the Th2-up arm may add noise or reflect general activation rather than Th2 identity specifically.
• The Th1/Th2 dichotomy is a simplification of a broader landscape (Th17, Treg, Tfh, Th9) and CD4 T cells exhibit plasticity; 'reciprocal lineages' is correct for the T-bet/GATA3 axis but the two arms are not the only fates, which is worth acknowledging given the positive Th1-Th2 arm correlation we observed.
• Type-2 disease is not purely Th2-cell-driven: ILC2s and other type-2 sources contribute in vivo, and asthma is heterogeneous with T2-high and non-type-2 (T2-low) endotypes. The in vitro CD4+ CRISPRi system captures the Th2-cell arm only, so mapping hits to 'allergy/asthma' disease relevance should be hedged.
• GATA3 has T-cell-lineage-wide roles beyond Th2 (required for early T-cell/CD4 development and expressed in Tregs and ILC2s), so a GATA3 knockdown could exert broad effects; its appearance as the top 'selective' hit is biologically expected but does not by itself prove Th2-selectivity of the assay.

Corrections to make:

• Frame STAT6 as the upstream IL-4 signal transducer feeding GATA3, not as a co-equal 'master regulator' transcription factor with GATA3.
• Consider down-weighting or annotating IL4R in the Th2-up arm, since IL-4Ra is broadly expressed and less Th2-specific than IL4/IL5/IL13/CCR4.

Sources:

• T-bet and GATA3 are mutually exclusive master regulators that directly cross-repress; T-bet binds GATA3 and bl — Nat Commun 2012, T-bet and GATA3 orchestrate Th1 and Th2 differentiation (PMC3535338); ScienceDirect T-bet overview
• GATA3 is master regulator: enforced expression drives Th2 under Th1 conditions; deficiency abolishes IL-4/IL-5 — Int Immunol 2011, Updated view on GATA3-mediated regulation of Th1/Th2 (PMC3123974)
• IL-4->IL-4Ra->STAT6 induces GATA3; IL-4/STAT6 positive feedback stabilizes Th2; GATA3 STAT6-independent autoac — Immunity 2000, Stat6-Independent GATA-3 Autoactivation (Ouyang et al.); PMC3557721 STAT6-dependent and -independent mechanisms
• Th1 requires T-bet/TBX21 controlling IFN-gamma; IL-12/STAT4 stabilizes T-bet; STAT4 in Th1 — ScienceDirect Transcription Factor T-bet overview; JEM 2018 STAT4 and T-bet (Rockefeller)
• Th1 express CXCR3/CCR5, Th2 express CCR4/CCR3; IL-4/5/13 restricted to CCR4+ and IFN-gamma to CXCR3+ memory CD — J Leukoc Biol 2000 Yamamoto et al. (PMID 11037980); Bonecchi et al. J Exp Med 1998 (PMID 9419219)
• Dupilumab is anti-IL-4Ra mAb blocking both IL-4 and IL-13 (shared IL-4Ra chain); dual blockade required to inh — JACI 2019, Dual blockade of IL-4 and IL-13 with dupilumab required to broadly inhibit type 2 inflammation (S0091-6749(19)32117-7); McCann 2024 Clin Transl Sci dupilumab MOA

Known-regulator direction reframe — mostly supported

Holds up:

• GATA3 is the master POSITIVE regulator of Th2 (induced by IL-4/STAT6, drives Th2 cytokines while suppressing Th1). Its knockdown lowers Th2 and de-represses Th1 - correctly classified as an activator.
• STAT6 is a POSITIVE regulator: IL-4/STAT6 drives GATA3 expression and Th2 commitment; STAT6-deficient cells default to a Th1-like response. Knockdown lowers Th2 - correctly an activator.
• IL4R (IL-4 receptor alpha) is a POSITIVE regulator: it transduces the IL-4 -> STAT6 -> GATA3 signal that promotes Th2. Knockdown lowers Th2 - correctly an activator.
• RARA (retinoic acid receptor alpha) is a POSITIVE regulator: ATRA/RARalpha signaling promotes IL-4/IL-5/IL-13 and increases Th2 output; RAR-alpha-selective agonists reproduce the pro-Th2 effect. Knockdown lowers Th2 - correctly an activator.
• ICOS is a POSITIVE costimulatory regulator of Th2 (ICOS deficiency impairs Th2 cytokine production); knockdown lowers Th2 - reasonably classified as an activator, though its effect is broadly costimulatory rather than Th2-specific.

Concerns:

• The clean 'activator vs repressor' dichotomy oversimplifies the NuRD biology. CHD4 is explicitly BIMODAL (Hosokawa et al., PNAS 2013): the Gata3/Chd4/p300 complex ACTIVATES Th2 cytokines (IL-4/IL-5/IL-13) while the Gata3/Chd4-NuRD complex REPRESSES Tbx21. Chd4 knockdown decreases IL-4+ cells (18.3%->7.9%) AND increases IFN-g+ cells (3.7%->13.8%). So CHD4 KD suppresses Th2 too - it is not a pure Th1-repressor, and could legitimately score on the Th2-down arm.
• SETDB1 is also mixed, not a clean Th1-repressor: Setdb1 loss enhances Th1 priming BUT also IMPAIRS Th2 commitment (via de-repression of ERVs that act as Th1 enhancers). So SETDB1 KD would push the Th2 arm down as well as the Th1 arm up - it does both, contradicting the 'induces Th1 rather than suppress Th2' phrasing.
• MBD2 is genuinely context-dependent/conflicting in the literature rather than a clean repressor: some studies show MBD2 silencing in CD4+ T cells INCREASES the Th2 ratio and IL-4 (negative regulator of Th2 / represses Th1-STAT1-IFN-g axis), while others show Mbd2-/- dendritic cells are markedly IMPAIRED in Th2 induction (positive regulator of Th2). Its direction cannot be asserted with confidence.
• Several 'activators' (GATA3, STAT6) also actively REPRESS Th1, so their KD is expected to raise the Th1 arm somewhat too - the assumption that a true selective Th2 suppressor leaves Th1 perfectly flat (|z|<1) may be too strict even for genuine positive Th2 regulators.

Corrections to make:

• TRAF3 is MISCLASSIFIED. The reframe lists TRAF3 as a negative regulator whose knockdown should induce Th1, but the direct T-cell evidence (Lu/Bishop lab, J Biol Chem 2024) shows TRAF3 ENHANCES STAT6 activation and PROMOTES Th2 skewing, and TRAF3 deficiency IMPAIRS Th2 (and Th1) differentiation. TRAF3 is therefore a POSITIVE Th2 regulator - its knockdown should LOWER Th2, behaving like the activators (GATA3/STAT6), not the repressors. The 'negative regulator' label likely derives from TRAF3's role in B-cell/alternative-NF-kB signaling, which does not transfer to the T-cell Th2 program.
• CHD4 should not be placed cleanly on the repressor side. Because it is bimodal (activates Th2 cytokines and represses Th1), CHD4 KD is expected to suppress Th2 as well as de-repress Th1; it can plausibly appear as a Th2 suppressor for legitimate reasons.
• SETDB1's expected phenotype should be stated as BOTH lowering Th2 and raising Th1 (mixed), not 'induces Th1 rather than suppress Th2.'
• MBD2's direction should be flagged as unresolved/context-dependent rather than asserted as a Th1-repressor.

Sources:

• GATA3 is the master positive regulator of Th2, induced by IL-4/STAT6, drives Th2 cytokines and suppresses Th1 — Tindemans et al., updated view on GATA3 regulation of Th1/Th2, PMC3123974; Transcriptional regulation of Th2 differentiation, PMC3477614
• STAT6 promotes GATA3 and Th2; STAT6/IL-4 knockouts give Th1-like responses — STAT6-dependent mechanisms in Th2 polarization, PMC3557721; STAT6-mediated displacement of polycomb by trithorax to maintain GATA3, JEM 207:2493
• RARalpha (RARA) mediates pro-Th2 effects; ATRA/RAR-alpha agonists promote IL-4/IL-5/IL-13 and increase Th2 out — Retinoic Acid Receptor-alpha mediates human T-cell Th2 cytokine production, BMC Immunology 9:16 / PMC2394516; ATRA and Ro415253 regulate IL-5+ Th2 cells, PMC3882882
• IL4R alpha binds IL-4 to promote Th2 differentiation — IL4R GeneCards; Transcriptional regulation of Th2 differentiation, PMC3477614
• CHD4 is bimodal: Gata3/Chd4/p300 activates Th2 cytokines, Gata3/Chd4-NuRD represses Tbx21; Chd4 KD decreases I — Hosokawa et al., Functionally distinct Gata3/Chd4 complexes establish Th2 identity, PNAS 2013, PMC3606997
• SETDB1 loss enhances Th1 priming but impairs Th2 commitment via ERV de-repression at Th1 enhancers — Adoue et al., SETDB1 controls Th cell lineage integrity by repressing ERVs, Immunity 2019, PMC/PubMed 30737147

Novelty & prior art — problematic

Holds up:

• Our core method is sound but is essentially what Zhu & Dann et al. already did. Their released analysis code (src/4polarizationsignatures/pert2statepolarization.ipynb) uses the SAME external signature we did (Ota et al. 2021 Th2-vs-Th1), splits it into an up-arm and a down-arm (signaturegsup / signaturegsdown), and scores each single cell separately as 'scoreth2' (up genes) and 'scoreth1' (down genes) with an NTC-normalized scoregenes routine (scoregenesntc_norm). This is exactly our Th2-arm / Th1-arm split.
• The 2D decomposition we treat as our angle is already in their paper. They compute per-perturbation mean scoreth2 vs scoreth1 and plot them on a 2D scatter with quadrant lines at x=0/y=0 (their Supplementary Fig 17, 'Effect of predicted polarization regulators on Th1 and Th2 genes'), and separately examine each perturbation's effect on Th1 genes vs Th2 genes. That is the same 2-arm / 2D scoring picture we built.
• GATA3 as the top Th2 hit is expected and reproduces theirs, not a novel discovery. GATA3 is the first entry in their curated th1th2known_regulators.yaml (th2 list) and is an explicitly highlighted/labeled regulator in their pert2state notebook; it is the textbook master Th2 regulator recovered by prior genome-wide CRISPR screens in T helper cells (Freimer/Marson 2022).
• Our concern about a global/magnitude confound is legitimate and shared by the authors. They guard against a closely related confound by using an independent activation signature (Arce et al. 2024, Rest-vs-Stim) as a negative control (known Th2/Th1 regulators should NOT be prioritized by activation) and K562 perturb-seq as a second negative control, and they check that polarization scores are not just tracking the activation axis.
• Using separate up/down gene-set scores for a bidirectional signature is a well-established, correct technique, so our implementation is not methodologically wrong — it is just not new (see corrections/sources).

Concerns:

• Novelty is low. The specific combination we pitched as our angle - split the Ota 2021 Th2-vs-Th1 signature into arms, score every genome-scale CRISPRi perturbation on both arms, decompose in 2D, and land on GATA3 - is a near-duplicate of the published Fig 4 / pert2state analysis in the very paper we are analyzing, using the identical discovery signature. Anyone reading their supplement + code would see this.
• The 2-arm / up-down decomposition of a bidirectional signature is a decades-old standard, not an innovation. Connectivity-Map connectivity scores are explicitly defined as an up-score plus a down-score computed on separately-supplied up and down gene lists; single-cell tools (UCell/pyUCell, scanpy score_genes) routinely score positive and negative gene sets separately. Framing this as a novel technique would be indefensible.
• Our claimed differentiator - selecting 'selective Th2 suppressors' by requiring the Th1 arm to stay flat (|z|<1) as a built-in control - is a modest reframing, not a new capability. The authors already visualize the same Th1-flat-vs-diagonal information in their 2D scoreth2/scoreth1 plot; they simply do not crown a formal 'selective vs skewing' category. Our contribution is essentially adding an explicit flatness threshold and a magnitude/'sick-cell dampening' narrative on top of their existing decomposition.
• Their regulator classification is directional-axis-based (regs_class = positive / negative / control, from signed model coefficients), i.e. an inherently 'skewing along the Th2-Th1 axis' model. So while they did not formalize 'selective suppression vs skewing' as competing categories, their whole framing is skewing-centric and our 'selective suppression' label sits inside their existing 2D data rather than beside it.
• Risk of reinventing with weaker statistics: their pipeline uses a trained, cross-dataset-validated linear model (train on Ota, reconstruct held-out genes in Hollbacher 2021, replicate across donors and culture conditions) with negative controls. Our z-score thresholding on two arm scores is a coarser version of the same idea and would be judged against their more rigorous model.

Corrections to make:

• Drop or heavily soften the novelty claim. We should not present 'split the Th2-vs-Th1 signature into arms and score both' or the 2D decomposition as our idea - Zhu & Dann et al. already do exactly this with the same Ota 2021 signature (scoreth2/scoreth1, Supplementary Fig 17). Position our work as a re-analysis/reframing, at most contributing an explicit selective-suppression flatness criterion, not a new method.
• Re-examine the GATA3-as-selective-Th2-suppressor claim; it likely conflicts with canonical biology. The classical reciprocal Th1/Th2 model holds that loss of GATA3 de-represses IFNG/T-bet and actively SKEWS toward Th1 (e.g. Gata3 loss converts Notch into a Th1 inducer; GATA3 directly represses Th1 factors). A knockdown that pushes Th2 down while leaving the Th1 arm flat (our selective definition) is biologically surprising for GATA3 and may be an artifact of the resting / short-stimulation human context, low Th1-arm expression at baseline, or our flatness threshold - flag this explicitly rather than presenting GATA3 as a clean selective suppressor.
• State the correct prior art for the technique: attribute up/down-arm decomposition to Connectivity Map (up-score + down-score) and to UCell/scanpy-style separate positive/negative gene-set scoring, so reviewers do not think we are claiming it.
• Reframe the 'sick-cell dampening / global magnitude' confound relative to their controls. The authors already control an analogous confound via an activation signature and K562/negative controls; we should either adopt an equivalent negative control (activation axis) or acknowledge that using the Th1 arm alone as the control is weaker than their approach, rather than presenting the confound insight as novel.

Sources:

• Zhu & Dann et al. split the Ota 2021 Th2-vs-Th1 signature into up/down arms, compute per-cell scoreth2 (up) a — github.com/emdann/GWTperturbseqanalysis2025 — src/4polarizationsignatures/pert2statepolarization.ipynb and polarizationsignature.ipynb (signaturegsup/down, scoreth2/scoreth1, regsclass positive/negative/control)_
• Fig 4C-H and Supplementary Figs 16-18 are the Th2/Th1 polarization-regulator analysis; Suppl Fig 17 is 'Effect — github.com/emdann/GWTperturbseqanalysis2025 — metadata/figuremap.md and metadata/th1th2knownregulators.yaml_
• Their pert2state model trains on Ota 2021, validates by held-out-gene reconstruction in Hollbacher 2021, and u — github.com/emdann/GWTperturbseqanalysis2025 — pert2statepolarization.ipynb markdown ('Train on discovery signature (Ota2021)', 'Validate ... Hollbacher2021', 'activation signature ... as a negative control')
• The paper nominates regulators of Th1 and Th2 polarization by modeling perturbation signatures against populat — bioRxiv 2025.12.23.696273 (Zhu, Dann, ... Pritchard, Marson) — abstract; biorxiv.org/content/10.64898/2025.12.23.696273v1
• Decomposing a bidirectional signature into separately-scored up and down gene lists is a standard technique: C — Connectivity Map up/down connectivity-score method; UCell and pyUCell (PMC12925249); scanpy scoregenes_
• Canonical biology: GATA3 loss de-represses Th1 (IFNG/T-bet) and drives Th1 skewing rather than neutral loss of — GATA-3 promotes Th2 and inhibits Th1-specific factors (Nature Cell Research, 7310002); Gata3 loss turns Notch into a Th1 inducer (PMC2062505); genome-wide CRISPR in T helper cells confirms GATA3/STAT6/IRF4 as Th2 regulators (PMC6370901)

Lead: ARNT — mixed

Holds up:

• There is a genuine, recent mechanistic link between ARNT-dependent HIF signaling and the Th2/GATA3 program: HIF2alpha (EPAS1) promotes Th2 differentiation and forms a HIF2alpha-GATA3 circuit, with EPAS1 directly bound by GATA-3 in Th2 cells (Cell/Immunity 2024, S1074-7613(24)00496-5; helminth paper PMC11761574).
• ARNT (HIF1-beta) is the OBLIGATE class-I dimerization partner for HIF2alpha/EPAS1 (and HIF1alpha and AhR). Because HIF2alpha is non-functional without ARNT, ARNT knockdown would be expected to abrogate the pro-Th2 HIF2alpha-GATA3 circuit, giving a coherent mechanism for ARNT KD lowering Th2/GATA3/IL4. This raises the prior above baseline that the hit is on-pathway rather than random.
• The Th2-selective (not Th1-inducing) prediction is mechanistically reasonable: HIF2alpha is the Th2-associated HIF paralog, whereas HIF1alpha is tied to Th17/glycolysis and Treg/effector-in-hypoxia biology, so loss of the ARNT/HIF2alpha axis need not push cells toward Th1.
• HIF2alpha-deficiency has been shown to IMPAIR Th2 differentiation and reduce Th2-mediated immunity / asthmatic inflammation in mouse models, matching the direction (ARNT/HIF2alpha loss -> less Th2) our pipeline nominated.

Concerns:

• Context mismatch: the HIF2alpha-Th2 link is established in HYPOXIC / tissue / intestinal-helminth settings and is hypoxia- or genetic-activation-dependent. Whether the ARNT/HIF2alpha axis meaningfully drives GATA3/IL4 in a 48hr in-vitro Stim of normoxic peripheral blood CD4+ T cells (typical Perturb-seq culture at ~21% O2, no HIF stabilization) is unproven and is the main reason to down-weight the lead.
• Pleiotropy / artifact risk is high precisely because ARNT is broadly and constitutively expressed and is the shared partner for HIF1alpha, HIF2alpha AND AhR. Its knockdown collapses multiple metabolic/stress pathways at once, so a 'selective' Th2 drop could reflect Th2 cells' higher metabolic/hypoxic-program dependence rather than a GATA3-specific effect. The flat Th1 arm mitigates a pure global sick-cell explanation but does NOT exclude a Th2-selective metabolic-fitness confound that mimics on-target selectivity.
• Countervailing pharmacology: AhR ACTIVATION suppresses Th2/GATA3, so losing the AhR arm via ARNT KD would be predicted to INCREASE Th2. The net effect depends on which ARNT partner dominates in this condition (AhR is reported to be largely ligand-dependent / functional mainly in Th17, so the HIF arm may dominate absent exogenous AhR ligand) - but this is an assumption, and it makes the sign of the ARNT effect genuinely context-dependent.
• Evidence is largely mouse and tissue-resident/pathogenic Th2; direct human primary-CD4 loss-of-function data for ARNT specifically on IL4/GATA3 is thin. No paper directly tests ARNT (as opposed to EPAS1/HIF2alpha) knockdown on the Th2 program.

Corrections to make:

• Frame the mechanistic hypothesis as ARNT-via-HIF2alpha (EPAS1), not ARNT-via-AhR or generic HIF1alpha - HIF2alpha is the Th2-relevant partner and gives the cleanest, direction-correct rationale. Any writeup should name EPAS1/HIF2alpha as the presumed effector.
• Calibrated plausibility: MODERATE, ~0.3-0.4. Above the ~random-hit baseline because a specific, recent, direction-consistent HIF2alpha-GATA3 mechanism exists, but held down by the normoxic short-stim context mismatch and ARNT's pleiotropy/fitness-confound risk. Do not present ARNT as a high-confidence GATA3-like selective suppressor.

Sources:

• HIF2alpha (EPAS1) promotes Th2 differentiation via a HIF2alpha-GATA3 circuit; EPAS1 is directly bound by GATA- — Hypoxia-inducible factor 2alpha promotes pathogenic polarization of stem-like Th2 cells, Immunity 2024 (cell.com/immunity/fulltext/S1074-7613(24)00496-5)
• EPAS1/HIF2alpha is highly expressed in intestinal CD4+ T cells during helminth infection; hypoxia or genetic H — Hypoxia-inducible factor 2alpha promotes protective Th2 cell responses during intestinal helminth infection, PMC11761574 / bioRxiv 2025.01.09.631414
• HIF-2 is a heterodimer of oxygen-sensitive HIF-2alpha and its obligate partner ARNT (HIF-1beta); ARNT is const — Bidirectional modulation of HIF-2 activity, Nat Chem Biol 2019 (s41589-019-0234-5); HIF-1 review PMC1959117
• AhR requires ARNT as heterodimer partner; AhR activation suppresses Th2 differentiation and can down-regulate — Aryl Hydrocarbon Receptor-Mediated Perturbations in Gene Expression during CD4+ T-cell Differentiation, PMC3412388; Cell Research cr200863
• Confirming experiment design rationale: ARNT is obligate partner for HIF1alpha, HIF2alpha and AhR, so epistasi — Synthesis of the above sources (ARNT pleiotropy)

Lead: ELAVL1 — mostly supported

Holds up:

• HuR/ELAVL1 directly binds AU-rich elements (AREs) in the 3'UTRs of GATA3, IL4, and IL13 mRNAs and stabilizes them, established by RNP-IP, biotin pull-down, and mRNA half-life assays (Techasintana/Atasoy et al., PMC5801757; Stellato lab).
• There is a specific, well-documented mechanistic link between HuR and the Th2 program: HuR coordinately regulates the master Th2 TF GATA3 plus its downstream cytokines IL4/IL13 as an ARE-linked regulon, so knocking down HuR is expected to lower the Th2 axis — consistent with our pipeline nominating ELAVL1 as a Th2 suppressor.
• Direct experimental evidence for Th2-over-Th1 SELECTIVITY exists: HuR silencing reduced GATA3 transcript stability and cut IL-13 mRNA half-life (5.1h to 2.9h) while IFN-gamma mRNA was reportedly unaffected; HuR overexpression raised GATA3/IL4/IL13 with NO increase in IFN-gamma in either Th1 or Th2 conditions. This convergence with our arm-split (Th2 arm down, Th1 arm flat) is a genuine point in the lead's favor.
• A gene-dosage/dose-dependent relationship is reported: even ~20% changes in HuR level significantly moved GATA3 mRNA and stability, and heterozygous conditional-KO Th2 cells showed reduced Gata3/Il4/Il13 mRNA — meaning partial CRISPRi knockdown could plausibly produce a measurable Th2-specific effect without full-cell collapse.
• Conditional HuR KO in CD4+ T cells causes defects in Th2 differentiation specifically (loss of IL-2 homeostasis, JAK-STAT/Th2 defects), so the differentiation-level phenotype is not just a cytokine mRNA-stability artifact (Chen et al., Mol Med 2013; Springer).

Concerns:

• Global-artifact risk is real and non-trivial: HuR is a pan-cellular mRNA-stability hub with hundreds of ARE targets. Full/strong HuR loss impairs core T-cell activation, proliferation, and survival, and also stabilizes Foxp3 (Treg), IL-2, and Th17 program mRNAs. Strong CRISPRi knockdown could push cells toward the 'sick-cell' global-dampening regime our own analysis flagged as the dominant confound (Th2 and Th1 arms +0.5 correlated).
• The Th2-over-Th1 selectivity claim is not clean at the mechanism level: HuR ALSO stabilizes canonical Th1/pro-inflammatory ARE-cytokine mRNAs — IFN-gamma and TNF-alpha (LFA-1/Vav1/p38 pathway, PMC3012057). So HuR is not intrinsically Th2-private; the apparent selectivity in the Stellato papers may reflect context/polarization state rather than an absolute Th2-only wiring.
• The strongest selectivity evidence comes largely from one lab (Stellato/Atasoy) and from HuR-overexpression and heterozygous/partial-loss systems, not from the strong loss-of-function regime a CRISPRi screen produces — the effect direction may hold but the Th1-sparing may not survive stronger knockdown.
• Our pipeline's Th1-arm-flat criterion could still be satisfied trivially if IFN-gamma/Th1 genes are simply less ARE-dependent at baseline in this culture condition, i.e. selectivity could be a substrate-availability quirk rather than a druggable Th2-specific dependency.

Corrections to make:

• Do not describe HuR/ELAVL1 as a Th2-specific factor; it is a global ARE-binding mRNA stabilizer that happens to have a disproportionately characterized effect on the GATA3/IL4/IL13 regulon. The lead is plausible but the mechanism is 'global regulator with a strong Th2 sub-module,' not 'selective Th2 gene.'
• Frame the Th1-arm control as necessary but not fully sufficient here: because HuR can also stabilize IFN-gamma/TNF, a flat Th1 arm in our data is informative but does not by itself exclude the global-dampening confound at high knockdown.

Sources:

• HuR binds AREs in GATA3, IL4, IL13 3'UTRs and stabilizes them; HuR silencing reduces GATA3 stability and IL-13 — https://pmc.ncbi.nlm.nih.gov/articles/PMC5801757/
• HuR coordinately regulates GATA-3 and Th2 cytokine gene expression in a dose-dependent manner — https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3643703/
• Conditional HuR KO in CD4+ T cells reveals gene-dosage effect on cytokine production and Th2 differentiation d — https://link.springer.com/article/10.2119/molmed.2013.00127
• HuR also stabilizes Th1-type cytokine mRNAs IFN-gamma and TNF-alpha via LFA-1/Vav1/Rac/p38/MKK3 signaling (evi — https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3012057/
• HuR is a global regulator: stabilizes Foxp3 in Tregs and strengthens Th17 differentiation/CD4 activation signa — https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8357502/
• Transcriptome-wide identification of direct and indirect HuR targets in activated CD4+ T cells (hundreds of AR — https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4498740/

Methods red-team — mixed

Holds up:

• Using the Th1 arm as a built-in negative-direction control and explicitly naming global 'sick-cell' dampening as the dominant confound is the right instinct; the +0.5 arm-arm correlation is real and correctly diagnosed as a magnitude artifact rather than biology.
• Preferring a residual-on-control over a raw DE-breadth penalty is defensible: a breadth penalty would wrongly punish strong bona fide regulators like GATA3 that legitimately move many genes (notebooks/04_score.py comment is correct on this specific tradeoff).
• The gate-threshold robustness grid (TX in {-1.5,-2,-2.5} x TY in {0.5,1,1.5}, gaterobustness 0..9 in 03credibility.py, and the 07qcaudit sweep) is good practice and partially mitigates the arbitrariness of the hard cutoff.
• Requiring on-target KD confirmation (ontargetsignificant, negative effect, fracsignif_guides>=0.5) plus cross-donor and cross-guide stability flags is a sound orthogonal-evidence layer that most naive Perturb-seq screens skip.
• Recovering GATA3 as the top hit and checking known-regulator recall/specificity is a reasonable sensitivity calibration, and separating directional 'activators' from 'repressor-complex' known genes (TH2_ACTIVATORS set) shows awareness that the known-gene list is directionally mixed.

Concerns:

• SCORING IS SELF-CONTAINED, NOT COMPETITIVE: the arm score is an unweighted mean of per-gene DE z-scores over signature genes. This measures absolute movement of the signature, so any global transcriptional shift (sick cell) moves both arms together. The +0.5 correlation is therefore largely manufactured by the scoring choice, not a fact of nature to be regressed away afterward. Competitive/rank-based scores (AUCell, singscore, fgsea/CAMERA-style) normalize against the rest of the genome and would remove most of this confound at the scoring step.
• NO PROPER NULL / NO FDR: z<-2 is a tail of the across-perturbation distribution, whose center is the mean over mostly-inert perturbations, not a true no-effect null. With thousands of perturbation x condition rows and a nominal ~2.3% left tail, dozens of the 172 'selective' calls are expected by chance. There is no permutation/empirical FDR and no comparison of 172 to a label-shuffled null, so the count is uncalibrated.
• 'Th1 flat' (|th1|<1) ACCEPTS THE NULL: |z|<1 conflates 'genuinely no Th1 effect' with 'Th1 estimated too noisily to reach 1'. A perturbation with a weak/underpowered Th1 measurement passes the flatness gate spuriously, inflating apparent selectivity. Flatness needs an equivalence test (TOST) with the Th1 uncertainty, not a bare threshold.
• TH1 ARM IS A BIASED, NOISY PROXY FOR GLOBAL MAGNITUDE: it is a small biased gene subset, not total DE burden. Regressing Th2 on Th1 (a) removes only the confound component that projects onto Th1 genes and (b) also removes real reciprocal Th1/Th2 biology, so the residual can MANUFACTURE selectivity for a perturbation that genuinely raises Th1. Because Th1 is measured with error, OLS slope is attenuated (regression dilution), so it under-corrects the confound it targets.
• DOUBLE-DIPPING / SELECTION BIAS: perturbations are selected for extreme Th2, then Th2 effect (residual) is ranked and reported on the same data. This is winner's-curse — top hits' effect sizes are inflated. Also the residual is fit on the full df but the gate already restricts to |th1|<1, so within the retained band there is little Th1 variance left and th2resid ~= th2arm — the 'control' does little where it is actually applied.
• PSEUDOREPLICATION: rows are perturbation x condition; the same gene appears up to 3x (Rest/Stim8/Stim48). The across-perturbation z-scoring, the OLS, and the gate all treat these as independent, ignoring within-gene clustering and the different dynamic range per condition (one global OLS line across conditions).
• SIGNATURE-COUNT TUNING: membership thresholds (adjp<0.05, |lfc|>0.5) are explicitly described as 'tunable to reproduce teammate counts' (172/4). Fitting gene-set definition to hit a target number is a garden-of-forking-paths / analytic-circularity risk.
• EXTERNAL SIGNATURE TRANSFER: Ota-2021 is bulk and from a different protocol/cohort; genes are lost in mapping and some 'Th2-up' genes are generic activation/proliferation markers that reintroduce the magnitude confound. The dataset's OWN matched polarization signature (present in data/) is the better-matched primary and was not used as such.
• nDE>=10 IS A POWER FILTER IN DISGUISE: it correlates with effect magnitude and with per-perturbation cell count / KD efficiency, so it preferentially removes subtle-but-selective regulators and biases the retained set toward strong global movers — partially reintroducing the very confound being controlled. nanto_num imputing missing gene z-scores to 0 also biases arm means toward flat/attenuated.
• BIOLOGICAL: Th2-vs-Th1 is not a clean 1D axis in CD4 cells. A knockdown can lower the Th2 arm by diverting cells to Th17/Treg/Tfh or into apoptosis/translation stress, not by specific Th2 inhibition. With only Th1 as a control, 'selective Th2 suppressor' cannot be distinguished from lineage diversion or a sick cell whose most dynamically expressed (Th2) genes fall first. No viability/proliferation or other-lineage arm rules this out; no functional (IL-4/5/13) readout confirms transcriptional -> functional suppression.

Corrections to make:

• Move the magnitude correction from a post-hoc regressor to the scoring step: replace the mean-of-z self-contained score with a competitive/rank-based enrichment (AUCell, singscore, or an fgsea/CAMERA/ROAST-style rotation test that normalizes against matched random gene sets). This removes most of the +0.5 confound a priori instead of trusting a noisy Th1-proxy regression.
• Build a real null from non-targeting / safe-harbor control guides: define arm z-scores and the selectivity threshold relative to the control-guide distribution, and compute an empirical/permutation FDR (shuffle signature labels or perturbation labels) so the 172 count is reported as q-value-controlled rather than a raw tail.
• Replace the hard AND gate with a continuous, uncertainty-aware selectivity statistic: per-perturbation Th2 and Th1 effects with SEs (from guide/donor replicates), test Th1 flatness by equivalence (TOST), and rank by a Th2-minus-Th1 contrast with confidence intervals — avoid dichotomizing continuous scores.
• If keeping a regression control, residualize on a DIRECTLY measured global-magnitude covariate (total DE burden, mean|z|, PC1 of the z matrix, and/or KD strength), fit WITHIN condition, and use total-least-squares/partial correlation to avoid regression-dilution. Report both raw and controlled ranks side by side.
• Use the dataset's own matched polarization signature as the primary readout and Ota-2021 as orthogonal cross-validation; require concordance across >=2 signatures (Ota + Hollbacher + Diff043 are already in combinedTh2vsTh1signature.csv). Freeze membership thresholds a priori and show full sensitivity instead of tuning to reproduce 172/4.
• Add cross-validation to beat the winner's curse: fit the gene set / score on half the guides (or donors) and evaluate selectivity on the held-out half; make cross-guide and cross-donor replication the PRIMARY evidence for each hit rather than a secondary flag, and report shrinkage-adjusted effect sizes.
• Add orthogonal control arms — Th17/Treg/Tfh lineage, cell-cycle/proliferation, apoptosis, ribosome/translation — and redefine 'selective' as Th2-down while Th1 AND other-lineage AND viability arms are all flat (by equivalence). This is the single biggest fix for the lineage-diversion / sick-cell biological threat.
• Add a dose-response causal check: test that Th2 suppression scales monotonically with per-guide KD efficiency (an MR-style within-perturbation gradient), which is much stronger evidence of specific causation than a single thresholded z.
• Fix pseudoreplication: model gene as a random/grouping factor (or aggregate to one estimate per gene with condition as a covariate) in any z-scoring, regression, or inference so the 3 conditions per gene are not treated as independent observations.
• Treat GATA3 recovery as sensitivity only, not validation of novel hits; report specificity as the empirical false-selective rate under the permuted null, and plan a functional confirmation (IL-4/IL-5/IL-13 cytokine or protein readout) for top novel candidates before claiming Th2-selective suppression.

Sources:

• Self-contained gene-set scores (mean of member statistics) are inflated by global shifts and inter-gene correl — Wu & Smyth CAMERA (Nucleic Acids Res 2012) and Goeman & Buhlmann (Bioinformatics 2007) on competitive vs self-contained gene-set testing
• Rank/rotation-based enrichment normalizes against the rest of the transcriptome and controls inter-gene correl — fgsea (Korotkevich et al. 2021); ROAST rotation test (Wu et al. 2010); AUCell/singscore for single-cell signature scoring (Aibar SCENIC 2017; Foroutan singscore 2018)
• Accepting the null ('flat') from a non-significant test is invalid; equivalence testing (TOST) is required to — Lakens, Two One-Sided Tests (TOST) equivalence testing, Soc Psychol Personal Sci 2017
• Selecting features on an extreme statistic then re-estimating that statistic on the same data biases effects u — Kriegesk1el 'circular analysis' / double-dipping (Nat Neurosci 2009); winner's-curse effect-size inflation literature
• Regressing on an error-laden covariate attenuates the slope (regression dilution), so a noisy proxy under-corr — Standard errors-in-variables / attenuation bias result (Fuller, Measurement Error Models)
• Perturb-seq best practice uses non-targeting control guides as the null and controls FDR across the perturbati — Replogle et al. genome-scale Perturb-seq (Cell 2022); Dixit/Adamson Perturb-seq framework (Cell 2016)

Part C — Hardened re-analysis (acting on the methods red-team)

Implemented the four highest-impact fixes and re-ran over the full matrix (notebooks/10_hardened.py):

1. Competitive rank-based scoring (singscore-style) instead of mean-of-z: arm correlation

+0.54 → +0.18. Confirms most of the confound was a scoring artifact.

1. Permutation null → empirical FDR: 2-arm selective = 420; expected under random

same-size gene sets ≈ 525 → FDR ≈ 1.25. The selective

category is not enriched over chance (Th1/Th2 reciprocity means Th2-down usually raises Th1).

1. Multi-lineage + proliferation control arms (Th1/Th17/Treg/cell-cycle, from the dataset's own

vs-Th0 signatures): requiring all flat leaves 41 (FDR ≈ 5.51).

1. Two-signature concordance (Ota + matched Th2-vs-Th0): 7 genes pass everything

(BRPF1, DALRD3, GLIS2, RABEPK, RELL2, SLC30A5, UQCR11); none of the earlier 21 high-confidence genes survive.

GATA3 under proper scoring is a Th1-skewer, not a selective suppressor (Th1-vs-Th0 arm +2.8),

consistent with canonical biology — invalidating the earlier calibration claim.

Conclusion: the naive selective-suppressor atlas does not survive rigorous statistics. Report this

as a calibrated negative result. To make any positive claim would require: competitive scoring as the

primary score, a control-guide (non-targeting) permutation FDR, equivalence (TOST) tests for arm

flatness with per-guide/donor SEs, and a functional IL-4/5/13 confirmation of top candidates.