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

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.

• CHOP is the pro-apoptotic arm of the ER-stress response. A drug that lowers CHOP is lowering a death programme, which is the wrong direction for a therapy and means the ISR collapse cannot be counted as part of the drug's benefit without a rescue experiment.
• In a leukaemia project, searching this gene by its common name returns a chemotherapy regimen. Any bounded absence, coverage count or citation audit that used CHOP as a token is contaminated, and the twelve fusion-scoped CHOP records look entirely plausible until their titles are read.
• Leukaemia lines are LESS DDIT3-dependent than the average cancer line (-0.074 versus -0.156), so whatever CHOP is doing in this disease, the cells do not need it - which argues against any attempt to exploit the CHOP axis therapeutically here.

1. Does CHOP protein accumulate in the nucleus of RS4;11, and does that change under VTP-50469? The coded localisation evidence for CHOP points at an altORF annotation paper, so even the baseline compartment is unmeasured in a leukaemia cell.
2. Is the monotone ATF4 < DDIT3 < TRIB3 < STC2 < CHAC1 gradient cascade amplification or dynamic range? Normalising each gene to its own maximum induction under tunicamycin in the same cells would separate them.
3. Is the 34-aa DDIT3 uORF peptide made in KMT2A-rearranged cells, and does its level change under a menin inhibitor? In a beat about uORF-mediated control, a uORF-encoded protein on a pathway member is the most obviously untested object available.
4. Can a 1.5-fold to 2.8-fold fall in CHOP account for a 30-fold fall in CHAC1, or is a second lost input required? The obvious candidate for the second input is ATF4 protein, which nobody has measured.

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

   DDIT3 falls 0.82 log2 under MI-2-2 in RS4;11 (p 0.004, DMSO 9.90) and 1.50 log2 under VTP-50469 at day 7 (EPZ5676 -1.59), which is three to four log2 smaller than its own targets CHAC1 (-5.11) and STC2 (-4.40) in the same samples, producing a monotone effect-size gradient ATF4 -0.86 < DDIT3 -1.50 < ATF5 -2.43 < TRIB3 -3.10 < STC2 -4.40 < CHAC1 -5.11.

   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. DDIT3 probe 209383_at; RNA-seq row DDIT3; the full cascade ordering. · NCBI GEO, re-analysed here 2026
   ▸What this evidence cannot show

   The individual values are counts inside named cohorts and are established as such. THE GRADIENT'S INTERPRETATION IS NOT: the three largest effects also have the three lowest RNA-seq DMSO baselines (CHAC1 6.35, STC2 5.02, DDIT3 5.50 versus ATF4 8.68), so dynamic range is a live alternative to cascade amplification and the ordering is graded as suggestive. REH also moves -0.30 with p 0.031, so DDIT3 is not cleanly fusion-selective. RS4;11 array DMSO arm n=2. DDIT3 is not one of the 24 concordant genes precisely because it does not clear 1 log2 under MI-2-2.

   ▸Evidence that argues against this (1)

   • The same collapse is produced almost exactly by EPZ5676, a potent DOT1L inhibitor, in the same cells at the same timepoint, so the effect marks disruption of the fusion's transcriptional output by either route and not the menin-KMT2A interface specifically. NCBI GEO, re-analysed here 2026
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   DDIT3/CHOP is a multifunctional transcription factor of the ER-stress response that induces cell-cycle arrest and apoptosis, acts both as an inhibitor of C/EBP function and as an activator, is cytoplasmic under non-stressed conditions and accumulates in the nucleus on ER stress, and carries a bZIP domain at residues 99-162.

   UniProtKB - P35638, DDIT3_HUMAN, DNA damage-inducible transcript 3 protein · UniProt Knowledgebase 2026
   ▸What this evidence cannot show

   A curated database function line, cited as a map of the architecture rather than as a measurement. RULE 91 FLAG, and it is a clear one: DDIT3's cytoplasm and nucleus location calls carry ECO:0000269 pointing at PubMed 29083303 - 'Deep transcriptome annotation enables the discovery and functional characterization of cryptic small proteins' - an altORF discovery paper, NOT a CHOP localisation study. The biologically meaningful statement (cytoplasmic unstressed, nuclear on ER stress) is in UniProt's uncoded NOTE, which per rule 91 is where the condition-scoped content usually lives. Rule 62 FAILS on this symbol: two reviewed human records.

   ▸Evidence that argues against this (1)

   • If the coded localisation evidence for CHOP points at a cryptic-small-protein annotation paper, then the compartment assignment in the database is not resting on a dedicated CHOP imaging experiment, and the stress-dependent nuclear accumulation - the part that matters mechanistically - is carried only by an uncoded note. The annotation should not be treated as a measurement. Samandi S et al. 2017
3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   CHOP and ATF4 cooperate at overlapping amino-acid-response elements in the TRIB3 promoter to activate it, and CHOP is required for ER-stress-induced TRIB3 induction; separately, CHAC1 is activated by the ATF4-ATF3-CHOP cascade and not by parallel XBP1 or ATF6 pathways.

   TRB3, a novel ER stress-inducible gene, is induced via ATF4-CHOP pathway and is involved in cell death. · Ohoka N, Yoshii S, Hattori T, Onozaki K, et al. · EMBO Journal 2005
   PMID 15775988DOI
   ▸What this evidence cannot show

   Chemical ER stress in non-haematopoietic cells, with reporter, dominant-negative and knockdown methods. Cited for the direct upstream relationship to TRIB3 only. The CHAC1 half of the claim comes from a separate paper (PMID 19109178) in human aortic endothelial cells; both are mechanism papers in other systems and neither concerns leukaemia, a fusion or a menin inhibitor. Abstracts read; figures not.

   ▸Evidence that argues against this (1)

   • If CHOP is genuinely required for CHAC1 and TRIB3 induction, a 1.50 log2 fall in CHOP is a thin explanation for a 5.11 log2 fall in CHAC1 - a fivefold drop in an upstream activator producing a thirty-fold drop in its target requires either steep cooperativity or a second lost input. ATF4 protein is the obvious candidate for that second input and it was never measured. NCBI GEO, re-analysed here 2026
4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   In DepMap 26.06, DDIT3 has a mean CRISPR gene effect of -0.1557 across 1,258 screens with 2.5% of lines below -0.5 and a leukaemia mean of -0.0742 across 57 leukaemia screens.

   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; gene-specific screen count. Note the leukaemia mean (-0.074) is SHALLOWER than the pan-cancer mean (-0.156), i.e. leukaemia lines are less DDIT3-dependent than the average line - the opposite of ATF4, where leukaemia is deeper. Leukaemia screens by substring match on diseaseFromSource; ids persisted per rule 90.

   ▸Evidence that argues against this (1)

   • CHOP is a pro-apoptotic effector, so knockout should if anything help a cell grow, and a mean of -0.156 is therefore hard to interpret - it may reflect CHOP's separate role as a C/EBP-family partner in normal transcription rather than anything about stress. A gene effect near zero on a pro-apoptotic transcription factor does not license any conclusion about its role under drug. UniProt Knowledgebase 2026
5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   gene_exact:DDIT3 AND organism_id:9606 AND reviewed:true returns TWO reviewed human records: P35638 (DDIT3_HUMAN, 169 aa, DNA damage-inducible transcript 3 protein) and P0DPQ6 (DT3UO_HUMAN, 34 aa, DDIT3 upstream open reading frame protein), both carrying DDIT3 as primary gene symbol; both were re-fetched alone and their names, lengths and primary symbols compared back successfully.

   UniProtKB - P0DPQ6, DT3UO_HUMAN, DDIT3 upstream open reading frame protein · UniProt Knowledgebase 2026
   ▸What this evidence cannot show

   Measured live 2026-09-22 on the exact query form rule 62 warns about, and it is the only symbol in this beat's 22-symbol panel where that query returns more than one record. The 135-residue length difference is far outside rule 66's 16-residue discrimination floor, so an accession-plus-length checker WOULD catch a swap - that is a property of this pair, not a general defence. This is not in data/symbol-collisions.json (157 rows read).

   ▸Evidence that argues against this (1)

   • Calling this a trap slightly overstates the operational risk: any sane selection rule - take the longest, take the record whose entry name matches the symbol, take the one with a bZIP domain - picks P35638 correctly, and no automated pipeline in this repo currently queries DDIT3. The claim that survives is narrower: the query returns two, so an assertion of exactly one would fail, and the second record is a real object that deserves attention rather than suppression. UniProt Knowledgebase 2026
6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   PubMed returns 973 records for DDIT3[tiab] and 16,893 for CHOP[tiab], sharing only 160 ids (Jaccard 0.0148); fusion-scoped, DDIT3 returns 1 record while CHOP returns 12, and every one of the 12 was title-read and none concerns DDIT3 - they are CHOP-regimen and therapy-related-leukaemia papers, one of which (PMID 22845170) is titled 'Multiple clonal MLL fusions in a patient receiving CHOP-based chemotherapy'.

   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

   All twelve CHOP-scoped titles were fetched and read; none was assumed. Rule 94 form (one tag per term); rule 95 observed. This is the inverse of the usual rule-71 failure: widening the name list here ADDS only false positives, so the narrow DDIT3 query is the correct instrument and its near-zero is the correct answer. RULE 80 still applies to the DDIT3 zero-ish count: titles and abstracts only.

   ▸Evidence that argues against this (1)

   • A large alias record count is not by itself evidence of contamination - PERK[tiab] also returns far more than EIF2AK3[tiab] and there the extra records are mostly genuine uses of the gene's common name. What makes CHOP different is that the twelve fusion-scoped hits were read and none was about the gene; without that title check the count alone would not have settled it, and the same reasoning applied to GADD153 (951 records, 32 shared) would be wrong, because GADD153 IS a genuine CHOP synonym. NCBI PubMed 2026
7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   The 34-aa DDIT3 upstream open reading frame protein was characterised in a deep-transcriptome altORF annotation study showing that alternative open reading frames in annotated protein-coding genes are translated, that the resulting small proteins have orthologues across species and contain functional domains, and that thousands are detectable in proteomic datasets when the search database includes them.

   Deep transcriptome annotation enables the discovery and functional characterization of cryptic small proteins. · Samandi S, Roy AV, Delcourt V, et al. · eLife 2017
   PMID 29083303DOI
   ▸What this evidence cannot show

   ABSTRACT ONLY, and the abstract's worked examples are altMiD51 and others - it does NOT single out the DDIT3 uORF product, so this citation supports the class of object and the methodology, not a DDIT3-specific finding. The connection to DDIT3 is that UniProt's P0DPQ6 record and DDIT3's own location annotation both trace to this paper, which is itself the rule-91 observation in E2. Nothing here says the DDIT3 uORF peptide has a function.

   ▸Evidence that argues against this (1)

   • Mass-spectrometric detection of a 34-residue peptide in a reanalysis with an expanded search database is a low bar, and the field's replication record for cryptic small proteins is mixed. The existence of a UniProt record does not establish that this peptide is made at physiological levels in any cell, let alone in a leukaemia cell, and no experiment has looked. UniProt Knowledgebase 2026

• 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

  CHAC1 is activated by the ATF4-ATF3-CHOP cascade; CHOP falls 1.50 log2 while CHAC1 falls 5.11 in the same samples

• 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

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

• ← regulatesATF4 / CREB2 (P18848) - the hub this graph inferred from its targets and never wrote, and its own transcript does not move under either menin inhibitor

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

  -0.82 log2 under MI-2-2 and -1.50 under VTP-50469 in RS4;11; REH also moves -0.30, so fusion-selectivity is modest

• related to →Revumenib (SNDX-5613, Revuforj) — first approved menin–KMT2A interface inhibitor

  the alias CHOP is a chemotherapy regimen: 16,893 PubMed records against DDIT3's 973, and all 12 fusion-scoped CHOP hits are regimen papers, not gene papers