ATF4 / CREB2 (P18848) - the hub this graph inferred from its targets and never wrote, and its own transcript does not move under either menin inhibitor

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

• ATF4 is required by 72% of cancer cell lines at a gene effect of -0.71 and by normal haematopoietic stem cells. A drug that lowers ATF4 output is therefore operating on a housekeeping axis in an infant whose entire marrow reserve is ahead of them - and ATF4-deficient mouse HSCs show aging-like attrition. Any attempt to deepen the ISR arm pharmacologically inherits that toxicity.
• The collapse is reproduced almost exactly by a DOT1L inhibitor (CHAC1 -5.03 vs -5.11), so if a clinical combination of menin and DOT1L inhibition is ever tried, this axis is already saturated by either agent alone and the combination buys nothing on it - while the HOX/MEIS arm, where the two compounds DO differ by 2.6-3.8 log2, is where a combination could act.
• ATF4 activation - not loss - is the published therapeutic lever in KMT2A-r AML via the DELE1/HRI mitochondrial ISR plus MCL-1 inhibition. A menin inhibitor moves this axis the other way, so the two strategies may be mutually antagonistic and nobody has tested the combination in either direction.
• CHAC1, the largest effect in the whole signature, degrades glutathione and is pro-apoptotic. Losing it is predicted to RAISE glutathione and could blunt any co-administered agent that depends on oxidative or ferroptotic stress - a menin inhibitor may be quietly protecting the cell against a partner drug.
• If the mechanism is translational rather than transcriptional, then resistance does not require a MEN1 mutation. Anything that restores eIF2alpha phosphorylation or bypasses the uORF - an ISR-activating stress, a GADD34 loss, a nutrient-poor niche - could restore ATF4 output while the drug remains fully bound to menin, and nothing in a MEN1 sequencing panel would see it.

1. Does ATF4 PROTEIN fall in RS4;11 under VTP-50469 at 330 nM by day 7, and does phospho-eIF2alpha change? This is one western blot on cells two laboratories already grow, and it converts every 'ATF4 output' sentence in this graph from an inference into a measurement.
2. Is the 0.86 log2 fall at day 7 the transcriptional component Jimenez 2021 reports, or noise? A time course at 6, 12 and 24 hours would separate a small direct transcriptional effect from a late consequence of arrest; both public datasets sample only day 2 and day 7, which are far too late.
3. Does ATF4 occupy the promoters of CHAC1, TRIB3, ATF5 and STC2 in RS4;11, and does that occupancy change under a menin inhibitor? Without an ATF4 ChIP the 'ATF4 signature' remains a gene-list argument.
4. Is ATF4 in nuclear speckles in a KMT2A-rearranged cell, and does it share that compartment with AFF1 or AFF4? The only speckle evidence is p300-dependent and from overexpression, and this graph already has an open AFF1/AFF4 speckle question.
5. Are the MLL-rearranged B-ALL lines in PMID 33741715 KMT2A::AFF1, and did that paper measure ATF4 mRNA as well as protein? The full text would settle whether this beat's transcript null contradicts them or completes them.

1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

   In RS4;11, ATF4's own transcript moves 0 of 1 GPL570 probes and 0 of 4 RNA-seq line/day contrasts past 1 log2 under either menin inhibitor (largest excursion -0.86, VTP-50469 330 nM day 7) while CHAC1 falls 5.11 log2 in the same samples; ATF4 is expressed at 13.72 log2 RMA in DMSO, level with ACTB, so the null is a measurement and not a detection failure; and EIF2AK1, EIF2AK3, EIF2AK4 and EIF2S1 are flat on both platforms.

   Q106 pre-registered ISR panel run over the cached GSE63664 (MI-2-2 / EPZ004777, SEM / RS4;11 / REH, GPL570 RMA) and GSE127506+GSE127508 (VTP-50469 330 nM / EPZ5676 1 uM, RS4;11 and MOLM13, d2 and d7, 3-prime-end RNA-seq) matrices. Script analyses/does-atf4s-own-transcript-move-under-a-menin-inhibitor/run.py, results.json, panel frozen in panel-preregistration.json BEFORE any expression value was read. ATF4 array probe 200779_at; RNA-seq row id ATF4. · NCBI GEO, re-analysed here 2026
   ▸What this evidence cannot show

   This is a count inside two named public cohorts computed by a script whose panel was frozen first, which is why it is graded established - but it is established ONLY as a statement about mRNA in these samples. THE GENERALISATION IS NOT: a flat transcript does not establish that ATF4 protein is intact, and 3-prime-end RNA-seq cannot see the 5-prime leader where ATF4's regulation lives. ONE PROBE: GPL570 carries a single ATF4 probe, so the multi-probe corroboration available for ATF5 is not available here. n=2 DMSO arrays in RS4;11 because the submitter excluded RS411-DMSO-1 for QC, so array Welch p-values there rest on 2 vs 3 and are indicative. MI-2-2 at 10 uM is a weak tool compound that does not collapse MEIS1/MEF2C/PBX3 in this dataset, so its arm is not a clean menin-inhibition arm.

   ▸Evidence that argues against this (1)

   • The paper this graph cites for the menin-ATF4 link reports that menin inhibition leads to LOSS OF ATF4 in MLL-rearranged B-ALL, and attributes menin-dependent regulation of ATF4 to transcriptional AS WELL AS post-transcriptional mechanisms. If a transcriptional component exists and this dataset does not see it, the most likely explanations are that RS4;11 is not one of the lines that paper used, that day 7 is the wrong window, or that a 0.86 log2 fall is the transcriptional component and 1.0 log2 is an arbitrary threshold that excludes it. The third reading is the most uncomfortable and cannot be excluded: -0.86 is a 1.8-fold fall. Jimenez JA et al. 2021
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   Jimenez 2021 reports that inhibition of either EWS-FLI1 or menin causes loss of ATF4 and of ATF4-dependent transcription in Ewing sarcoma, identifies an EWS-FLI1 binding site at the ATF4 promoter, states that menin-dependent regulation of ATF4 is mediated by transcriptional AND post-transcriptional mechanisms, and reports that menin inhibition similarly causes loss of ATF4 and the ATF4-dependent signature in MLL-rearranged B-cell acute lymphoblastic leukaemia.

   EWS-FLI1 and Menin Converge to Regulate ATF4 Activity in Ewing Sarcoma. · Jimenez JA, Apfelbaum AA, Hawkins AG, Svoboda LK, et al. · Molecular Cancer Research 2021
   PMID 33741715DOI
   ▸What this evidence cannot show

   ABSTRACT ONLY - the full text was not obtained, so the B-ALL experiment's cell lines, menin inhibitor, dose, duration and readout are unknown to this entry, and 'loss of ATF4' is not resolved into mRNA, protein or activity for the B-ALL arm. RULE 73: the paper's primary system is Ewing sarcoma, a non-haematopoietic soft-tissue sarcoma driven by an unrelated fusion; the MLL-rearranged B-ALL result is an extension reported in one sentence. The graph already cites this paper twice, so the conclusion is partly written down elsewhere (rule 82) - what is new here is that the B-ALL clause exists and that a prior entry characterised the paper as being about a different cancer entirely.

   ▸Evidence that argues against this (1)

   • Menin's role in this paper is scaffold-like and indirect, and the Ewing serine-biosynthesis phenotype it centres on does not reproduce cleanly in the KMT2A datasets measured here: PHGDH falls 3.24 log2 and PSAT1 2.40 under VTP-50469 in RS4;11, but PHGDH is essentially flat on the array under MI-2-2 (-0.44) and both are flat in MOLM13 (-0.20, -0.57). A serine-arm collapse that appears in one line, one compound and one platform is weaker than the paper's framing would predict. NCBI GEO, re-analysed here 2026
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   In DepMap release 26.06, ATF4 has a mean CRISPR gene effect of -0.7118 across 1,258 screens with 71.9% of lines below -0.5 and a leukaemia mean of -0.8571, deeper than MEN1 (-0.3616, 28.9%, leukaemia -0.5459) and DOT1L (-0.1474, 12.1%, leukaemia -0.3931), while CHAC1 (+0.0133), TRIB3 (+0.0937), STC2 (+0.0593), ATF3 (+0.0538) and SESN2 (-0.0199) are not dependencies in any lineage.

   DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL depMapEssentiality field; dataVersion year 26 month 06, read back from meta{dataVersion} in the same session · DepMap / Open Targets Platform 2026
   ▸What this evidence cannot show

   Release recorded per rule 83 (dataVersion year 26 month 06, read back in the same session, not a fetch date), and screen counts are gene-specific by design - ATF4 1,258, DOT1L 1,257, IKZF3 1,253 - so no corpus-wide denominator is implied. The 57 leukaemia screens are identified by a substring match on diseaseFromSource ('leukemia', 'leukaemia', 'lymphoblastic', 'myeloid') and their DepMap ids are persisted in the analysis results per rule 90. A CRISPR knockout gene effect measures loss of the gene, which is NOT what a menin inhibitor does to ATF4 in these data - the comparison is deliberately a contrast, not a model of the drug.

   ▸Evidence that argues against this (1)

   • A deep pan-lineage gene effect is exactly what a gene required for basic proteostasis looks like, and it therefore cannot distinguish an oncogenic ATF4 requirement from a housekeeping one. ATF4 knockout mice are viable but severely anaemic and ATF4 is required for normal haematopoietic stem cell function, so the DepMap signal is at least partly generic essentiality - which weakens rather than strengthens any argument that ATF4 is a fusion-specific node. Sun Y et al. 2021
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In mouse, ATF4 deficiency causes aging-like functional attrition of adult haematopoietic stem cells with defective repopulating and self-renewal capacity and myeloid bias, and significantly delays leukaemogenesis in the MLL-AF9-induced leukaemia model, acting through HIF1alpha and p16Ink4a.

   Loss of ATF4 leads to functional aging-like attrition of adult hematopoietic stem cells. · Sun Y, Lin X, Liu B, Zhang Y, et al. · Science Advances 2021
   PMID 34936448DOI
   ▸What this evidence cannot show

   MURINE, and the fusion is MLL-AF9 - a myeloid model with a different 3-prime partner from the index KMT2A::AFF1 B-lineage disease (rule 73). 'Significantly delayed' leukaemogenesis is a latency result, not abrogation, so ATF4 is a modifier and not a requirement in that model. Germline/conditional deletion for the lifetime of the experiment is not comparable to days of pharmacological output loss. Abstract read; the figures were not.

   ▸Evidence that argues against this (1)

   • The same paper shows ATF4 loss cripples normal HSC function with an aging-like phenotype, so any therapeutic reading of the leukaemogenesis delay runs straight into on-target haematopoietic toxicity - and in a disease of infants, whose entire haematopoietic reserve is ahead of them, that is a more serious objection than in adult AML. Sun Y et al. 2021
5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   Inhibition of electron transport chain complex I activates the mitochondrial integrated stress response through DELE1 and the eIF2alpha kinase HRI, causing ATF4 activation, and this creates a synthetic-lethal vulnerability to MCL-1 inhibition in AML cell lines and patient-derived xenografts, in a paper framed on KMT2A-rearranged AML as the high-risk subtype of interest.

   Mitochondrial integrated stress response activation creates a therapeutic vulnerability to MCL-1 inhibition in acute myeloid leukemia. · Brakefield-Laird L, Budhraja A, Hall PM, Brewington D, et al. · Cell Death & Disease 2026
   PMID 42350371DOI
   ▸What this evidence cannot show

   ABSTRACT ONLY. The KMT2A-r framing is in the abstract's opening and closing sentences; how many of the lines and PDX samples are actually KMT2A-rearranged is not stated there and was not checked, so 'in KMT2A-r AML' is the paper's framing and not a verified property of its panel. MYELOID, not B-lineage. 2026 publication, so it has had almost no independent replication. HRI/EIF2AK1 is the kinase here, and EIF2AK1's transcript is flat in both menin-inhibitor datasets - as expected, since kinase activation is post-translational.

   ▸Evidence that argues against this (1)

   • This result points the opposite way to the menin-inhibitor observation: here ATF4 ACTIVATION is the therapeutic lever, whereas the menin-inhibitor datasets show ATF4 OUTPUT collapsing. If both are true in KMT2A-r cells then a menin inhibitor and a mitochondrial-ISR-plus-MCL-1 strategy are antagonistic on the same axis, and combining them would be predicted to blunt one of them. Nobody has tested the combination. NCBI GEO, re-analysed here 2026
6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   PERK (EIF2AK3) and GCN2 (EIF2AK4) are metabolic-stress-sensing kinases that phosphorylate the alpha subunit of eukaryotic translation initiation factor 2 in response to unfolded protein load and to low amino-acid availability respectively, and UniProt describes ATF4 as the master transcription factor of the integrated stress response acting downstream of that phosphorylation.

   UniProtKB - Q9NZJ5, E2AK3_HUMAN, eukaryotic translation initiation factor 2-alpha kinase 3 / PERK · UniProt Knowledgebase 2026
   ▸What this evidence cannot show

   A database function line, curated from many primary papers, used here only for the textbook architecture of the pathway and not for any claim about this disease. RULE 91: the individual PubMed ids inside UniProt's function comment were not opened, so this is cited as a map rather than as a measurement. The uORF mechanism specifically is standard ISR biology stated in the same comment block; no single founding paper is cited for it here, and a reader wanting the primary source should be pointed at the ATF4 uORF literature rather than at this row.

   ▸Evidence that argues against this (1)

   • Describing ATF4 as translationally controlled through a uORF is true of the canonical ISR and is not the only mode: ATF4 transcription is itself inducible, and in Ewing sarcoma a fusion oncoprotein binds the ATF4 promoter directly. So 'ATF4 is translationally controlled' must not be used to argue that ATF4 mRNA CANNOT move under a perturbation - it demonstrably can, in a fusion-driven cancer. Jimenez JA et al. 2021
7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   UniProt P18848 is the single reviewed human ATF4 record, 351 aa, with a bZIP domain at 278-341, synonyms CREB2 and TXREB, and five subcellular-location statements of which Nucleus speckle rests on ECO:0000269 from PubMed 16219772 - a paper that reports ATF4 recruitment into nuclear speckles as a p300-binding-dependent event, ATF4 stabilisation by p300 independent of p300 acetyltransferase activity, and ATF4 as an SCF(betaTrCP) substrate.

   UniProtKB - P18848, ATF4_HUMAN, cyclic AMP-dependent transcription factor ATF-4 · UniProt Knowledgebase 2026
   ▸What this evidence cannot show

   Resolved live 2026-09-22: gene_exact:ATF4 AND organism_id:9606 AND reviewed:true returns EXACTLY ONE record (rule 62 satisfied), and the accession was re-fetched alone with protein name, length and primary gene symbol compared back (rule 1). RULE 91 APPLIED: the speckle paper WAS opened and it is genuinely an ATF4 paper - but the speckle observation is conditional on the p300-binding domain and comes from a transfection/overexpression system, so no constitutive endogenous speckle residence is claimed. Cytoplasm and cell membrane are ECO:0000250 by similarity to rat Q9ES19 and are recorded as uncoded inferences, not measurements.

   ▸Evidence that argues against this (1)

   • A compartment seen only when a co-activator binding domain is intact, in transfected cells, is weak evidence of where endogenous ATF4 sits - and the same paper's central claim is about stability, not location, so the speckle panel is a supporting observation in a paper about something else. Rule 91's warning applies in its milder form: the citation is on-topic but is not the best possible source for this fact, and no better one was found. Lassot I et al. 2005
8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   CHAC1 is activated by the ATF4-ATF3-CHOP arm of the unfolded protein response and specifically NOT by the parallel XBP1 or ATF6 pathways, established by siRNA and overexpression in human aortic endothelial cells, and CHAC1 overexpression enhances apoptosis while CHAC1 siRNA suppresses it.

   CHAC1/MGC4504 is a novel proapoptotic component of the unfolded protein response, downstream of the ATF4-ATF3-CHOP cascade. · Mungrue IN, Pagnon J, Kohannim O, Gargalovic PS, Lusis AJ · Journal of Immunology 182(1):466-76 2009
   PMID 19109178DOI
   ▸What this evidence cannot show

   HUMAN AORTIC ENDOTHELIAL CELLS, with chemically induced UPR - not leukaemia, not a fusion, not a menin inhibitor. It is cited here for branch specificity, which is the transferable part, and the branch assignment is what the menin-inhibitor data independently agree with (ATF4 output down 1.4-5.1 log2, XBP1 -1.31, ATF6 +0.14 in RS4;11 d7). The graph already cites this paper once (rule 82); the new content is the agreement with the branch-specificity prediction in a KMT2A-rearranged cell.

   ▸Evidence that argues against this (1)

   • The branch specificity was established with chemical UPR inducers in a vascular cell type. A menin inhibitor is not a UPR inducer and the direction is opposite - loss of output rather than induction - so the argument 'CHAC1 falls, therefore the ATF4-ATF3-CHOP branch is the one affected' is an inference from an induction experiment to a de-induction one, in a different lineage. Consistency is not proof of pathway identity. NCBI GEO, re-analysed here 2026
9. E9established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   PubMed contains 6 records for ATF4 jointly with any of eight KMT2A-fusion operands and exactly 1 for ATF4 jointly with a thirteen-name menin-inhibitor list, and the alias CREB2 reaches 145 records of which only 36 also mention ATF4 while TCGGCF reaches zero records in the whole index; the fusion-scoped CREB2 query returns a zero with a non-empty querytranslation and warninglist 'No items found.'

   NCBI PubMed esearch, run 2026-09-22, with querytranslation and warninglist read back from the response for every query quoted · NCBI PubMed 2026
   ▸What this evidence cannot show

   Rule 94 form used throughout (one [tiab] tag per term, never a parenthesised group carrying the tag); rule 95 observed (every cytogenetic term quoted, and "t(4;11)" translated correctly to "t 4 11"[Title/Abstract] returning 736 records as a control); rule 67 controls non-empty (menin AND ATF4 = 1, CHAC1 AND ATF4 = 66, KMT2A OR MLL = 7,685). Rule 80 clause: these zeros are bounded by TITLES AND ABSTRACTS ONLY and supporting figures were not searched, which matters here because an ATF4 western is exactly the kind of panel that never reaches an abstract. Drug name list: menin inhibitor / inhibitors / inhibition, revumenib, ziftomenib, VTP-50469, VTP50469, MI-2-2, MI-503, MI-538, bleximenib, enzomenib, SNDX-5613.

   ▸Evidence that argues against this (1)

   • A six-record fusion-scoped count is small enough that the operand list is doing most of the work, and two of the six are off-beat on title-check. Widening to MeSH, to the full text via Europe PMC, or to mouse-model vocabulary would very likely raise it; this graph's own rule 80 records a case where a mark measured in a figure was invisible to exactly this kind of query. The count should be read as a floor on what has been PUBLISHED PROMINENTLY, not on what has been measured. NCBI PubMed 2026

• related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)

  ATF4 output collapses in both KMT2A::AFF1 lines under menin inhibition while ATF4's own transcript is flat (max -0.86) and the KMT2A wild-type line REH is unaffected

• ← regulatesMEN1 / menin — the adaptor the fusion cannot work without

  menin inhibition causes loss of ATF4 and the ATF4-dependent signature in MLL-rearranged B-ALL and in Ewing sarcoma, by transcriptional AND post-transcriptional mechanisms (PMID 33741715)

• ← related toRevumenib (SNDX-5613, Revuforj) — first approved menin–KMT2A interface inhibitor

  class effect: a menin inhibitor removes ATF4 transcriptional output without lowering ATF4 mRNA past 1 log2; measured for VTP-50469 and MI-2-2, not for revumenib itself

• ← related toPinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field

  EPZ5676 1 uM reproduces the ATF4-output collapse in RS4;11 at day 7 almost exactly (ATF4 itself -0.87 vs -0.86), so the axis is NOT menin-interface-specific

• binds →EP300 — the acetyltransferase whose paralogue split is the finding: p300 is the 108th most KMT2A-selective gene, CBP is the 16,950th

  p300 stabilises ATF4 and is required for its recruitment into nuclear speckles, independently of p300 acetyltransferase activity (PMID 16219772)

• related to →Nuclear speckles — the compartment everyone assumes KMT2A::AFF1 uses, on evidence that skips AFF1 itself

  ATF4's only annotated subnuclear compartment is the nuclear speckle, and it is p300-dependent and shown on overexpressed protein; never co-imaged with a KMT2A fusion

• related to →The differentiation block — what it actually is, operationally, in KMT2A::AFF1 leukaemia

  the ATF4 output loss is measured in the arrested/differentiating state produced by 6-7 days of continuous menin- or DOT1L-inhibitor exposure, not in a steady proliferative state

• regulates →ASNS - the resistance gene for a backbone drug, never measured in this disease

  ASNS is an ATF4 amino-acid-response target; it falls 1.72 log2 under VTP-50469 in RS4;11 at day 7 while ATF4 mRNA is flat

• related to →KMT2A-rearranged acute lymphoblastic leukaemia

  menin inhibition is reported to cause loss of ATF4 and its signature in MLL-rearranged B-ALL (PMID 33741715, abstract), and this beat reproduces the transcriptional half in two t(4;11) lines

• ← depends onCHAC1 / BOTCH (Q9BUX1) - the largest single effect in the whole menin-inhibitor response (-5.11 log2) and a glutathione-degrading pro-apoptotic enzyme, named nowhere in this graph until now

  CHAC1 is activated by the ATF4-ATF3-CHOP arm and specifically not by XBP1 or ATF6 (PMID 19109178)

• regulates →CHAC1 / BOTCH (Q9BUX1) - the largest single effect in the whole menin-inhibitor response (-5.11 log2) and a glutathione-degrading pro-apoptotic enzyme, named nowhere in this graph until now

  ATF4-ATF3-CHOP cascade target; falls 5.11 log2 while ATF4's own transcript is flat at -0.86

• ← regulatesTRIB3 / TRB3 (Q96RU7) - a pseudokinase that is the ISR's own brake, falls 3.1 log2 under either inhibitor, and has a published physical link to the MLL complex in the wrong tissue

  negative feedback: TRIB3 is induced via ATF4-CHOP and then represses CHOP/ATF4 transactivation and its own induction (PMID 15775988)

• regulates →TRIB3 / TRB3 (Q96RU7) - a pseudokinase that is the ISR's own brake, falls 3.1 log2 under either inhibitor, and has a published physical link to the MLL complex in the wrong tissue

  ATF4 and CHOP cooperate at overlapping amino-acid-response elements in the TRIB3 promoter (PMID 15775988)

• regulates →DDIT3 / CHOP (P35638) - the cascade's middle node, and the symbol whose two reviewed human records and 16,893-record alias are the worst identifier trap in this beat

  ATF4-ATF3-CHOP cascade: DDIT3 is the middle node, falling 1.50 log2 against ATF4's flat -0.86 and CHAC1's -5.11

• related to →ATF5 / ATFX (Q9Y2D1) - the most reproducible repression in the earlier dataset, and the three probe values the graph quotes are SEM, not RS4;11

  two uORF-controlled ISR bZIP factors behaving differently: ATF4 mRNA flat at -0.86, ATF5 mRNA -2.43 in the same samples, which means part of the response is transcriptional and ATF5-selective

• ← regulatesPPP1R15A / GADD34 (O75807) - the eIF2alpha phosphatase adaptor, and KMT2A fusion proteins were shown to BIND it and block its apoptosis in 1999, a paper this graph has never cited

  GADD34 recruits PPP1CA to dephosphorylate eIF2alpha, terminating the translational shut-off that raises ATF4 - the OFF switch for the mechanism this beat argues is operative

• regulates →ATF3 (P18847) - the one cascade member the DOT1L inhibitor hits HARDER than the menin inhibitor, and a transcriptional REPRESSOR being counted as a stress output

  ATF4-ATF3-CHOP cascade; ATF3 falls 1.16 log2 (array, MI-2-2) and 1.84 (VTP-50469) while ATF4 mRNA is flat

• regulates →SESN2 / sestrin-2 (P58004) - the ISR's exit into mTORC1: a leucine sensor that falls 1.4 log2 under either inhibitor, and the one panel member with a real structural record

  SESN2 is an ATF4/ISR-responsive gene; it falls 1.32 and 1.21 log2 on two probes under MI-2-2 and 1.42 under VTP-50469 while ATF4 mRNA is flat

• regulates →STC2 / stanniocalcin-2 (O76061) - the second largest effect in the signature (-4.40 log2), a SECRETED glycoprotein, and its localisation rests on curator inference alone

  STC2 is an ATF4/ISR-responsive gene and the second largest effect in the signature (-4.40 log2) while ATF4's own transcript is flat

• ← regulatesEIF2AK3 / PERK (Q9NZJ5) - the kinase whose transcript is flat under both inhibitors, whose arm IS functionally required in MLL-AF9 leukaemia stem cells, and whose common name collides with phospho-ERK

  PERK phosphorylates eIF2alpha and is the key effector of the ISR to unfolded proteins, raising ATF4 translationally; its own transcript is flat under both inhibitors (max 0.35 log2) while ATF4 output falls 1.4-5.1

• ← regulatesEIF2AK4 / GCN2 (Q9P2K8) - the amino-acid-sensing eIF2alpha kinase: transcript flat, and NOTHING has ever been published about it in any KMT2A-rearranged system under either name

  GCN2 transduces low amino-acid availability into eIF2alpha phosphorylation and ATF4 induction; its transcript is the most solidly flat in the panel (max 0.38 log2) while the amino-acid-response programme it senses falls 1.09-5.11

• ← supportsATF4's own transcript does NOT move under either menin inhibitor - and the whole ISR arm is reproduced by a DOT1L inhibitor, so it is not a menin-interface readout

  ATF4's own transcript does not move: 0 of 1 array probes and 0 of 4 RNA-seq contrasts reach 1 log2, max -0.86, against CHAC1 -5.11

• ← related toThe ATF4/ISR collapse under menin inhibitors reports loss of the KMT2A-fusion transcriptional programme by ANY route, not engagement of the menin interface - and the loss is of ATF4 activity, not ATF4

  the hypothesis is about ATF4 ACTIVITY rather than ATF4: its transcript is flat at -0.86 while its targets fall up to 5.11 log2

• ← regulatesPHGDH (O43175) - the largest menin-SELECTIVE fall in the whole serine panel, and the least essential gene in it

  ATF4 modulates core SSP transcription (PMID 33741715) - but PHGDH's compound-selectivity does not match the ATF4-output arm, so the input is not ATF4 alone

• ← depends onPSAT1 (Q9Y617) - a DIRECT ATF4 target, and the gene that shows the brief's menin-selectivity claim was PHGDH's and not the pathway's

  ATF4 directly binds the PSAT1 promoter in AML (PMID 42044713); PSAT1's compound-independence (77% reproduced by EPZ5676) matches an ATF4-proximal target

• ← regulatesCTH (P32929) - gene_exact returns TWO reviewed records and the second is VSIG2, a third instance of rule 62 on the same symbol the Ensembl resolver already got wrong

  an ATF4/CTH/SQOR axis is published in renal ferroptosis (PMID 41759595, title only); CTH mid-arm margin of -0.75 neither confirms nor refutes an ATF4-proximal position