Oncofusion

start / graph / gene-atf3

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q106-isr-stress-programme

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

also called ATF3 · Cyclic AMP-dependent transcription factor ATF-3 · ATF3_HUMAN

ATF3 is a small stress-induced protein that mostly switches other genes OFF, and it sits in the middle of the chain the graph says collapses under a menin-blocking drug. Unusually, a completely different drug - one that blocks a chromatin enzyme rather than the menin contact - switches it off harder.

The same thing, for a scientist

ATF3 (P18847, 181 aa, bZIP 86-149) is the shortest member of this beat's panel and, per UniProt, a **transcriptional REPRESSOR** - it binds the cAMP response element and "represses transcription from promoters with ATF sites", possibly by stabilising inhibitory cofactors [E3]. It is the A of the ATF4-ATF3-CHOP cascade that activates CHAC1, so it is both a repressor in general and an activating component of this specific cascade. Measured on three probes in RS4;11: **-1.16 log2 on the best probe under MI-2-2 (p 0.008, DMSO 8.52), -0.76 on a second, and -0.29 on a third that sits at background** [E1]. Under the potent compounds at day 7 the ordering **reverses**: VTP-50469 **-1.84** and EPZ5676 **-2.81** - ATF3 is the one gene in the panel where the DOT1L inhibitor produces a **0.97 log2 LARGER** effect than the menin inhibitor [E1]. And the KMT2A wild-type control is clean: REH moves +0.06, +0.45, +0.27 on the three probes. **The low baseline is the entry's main limitation**: ATF3's RNA-seq DMSO level in RS4;11 is only 4.74 log2 CPM at day 7 and 3.29 at day 2, so this is a lowly expressed gene where a 1.84 log2 fall spans a small absolute count range. DepMap 26.06: **+0.0538** mean across 1,258 screens, 0.0% of lines below -0.5, leukaemia mean **+0.1022** - the most positive leukaemia gene effect in the panel [E2]. PubMed: **1** fusion-scoped record and it is off-beat on title-check; **0** menin-drug-scoped [E4].

What it looks like
Cyclic AMP-dependent transcription factor ATF-3181 aa
bZIP (basic leucine zipper)bZIP (basic leucine zipper) · residues 86–1490100
loading AF-P18847-F1…
AlphaFold AF-P18847-F1 ↗
colourshape

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.

very high (90+)confident (70–90)low (50–70)very low (<50)
I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
The reversal, which is the finding
datasetcellarmDMSOlog2FC
array 202672_s_atRS4;11MI-2-28.52-1.16 (p 0.008)
array 1554980_a_atRS4;11MI-2-26.13-0.76 (p 0.033)
array 1554420_atRS4;11MI-2-23.79-0.29 (at background)
array 202672_s_atSEMMI-2-28.15-0.36 (p 0.37, n.s.)
array 202672_s_atREHMI-2-28.75+0.06
RNA-seqRS4;11 d7VTP-504694.74-1.84
RNA-seqRS4;11 d7EPZ56764.74-2.81
RNA-seqRS4;11 d2VTP-504693.29-1.45
RNA-seqMOLM13 d7VTP-504694.38-0.74

ATF3 is the only gene in this beat's 37-gene specificity table where the DOT1L inhibitor beats the menin inhibitor by more than 0.9 log2 in the down direction, and it is classified DOT1L-INHIBITOR-SELECTIVE-LEANING in the analysis output. Two of the 37 lean that way (ATF3 and CEBPB); nineteen are shared to within 1 log2; eight are menin-selective.

What it means, carefully. The two compounds hit different surfaces of the same complex - menin's pocket versus DOT1L's SAM site - and they discriminate strongly on the HOX/MEIS arm (MEIS1 -5.25 versus -1.50, MEF2C -4.35 versus -1.71). That a mid-cascade ISR repressor falls FURTHER under the DOT1L inhibitor is therefore not a small detail: it is the clearest single datum against reading this programme as a menin-interface readout. It is also within the range where a low-count gene can be noisy, which is why the entry grades it strong for the measurement and moderate for the interpretation.

SEM disagrees again. MI-2-2 gives -0.36 in SEM against -1.16 in RS4;11, the same pattern as CHAC1 and ATF5. Three of the four array-measured cascade members show a substantial RS4;11-versus-SEM discrepancy, which is worth recording as a property of the GSE63664 experiment rather than of any one gene.

A repressor counted as an output

UniProt's function line for ATF3 is not that of an activator: it "represses transcription from promoters with ATF sites" and "may repress transcription by stabilizing the binding of inhibitory cofactors at the promoter" 3. Yet in the ATF4-ATF3-CHOP cascade that drives CHAC1, ATF3 is an activating component - Mungrue's siRNA and overexpression experiments place all three factors as required for CHAC1 induction 5.

Both are true and the resolution is promoter-context dependence, which is ordinary bZIP biology - the same factor activates at one site and represses at another depending on its dimerisation partner and the cofactors present. But it has a consequence this graph should record: "ATF3 down" cannot be read as "less repression" or "less activation" without knowing which promoters are in play. The 24-gene concordant list contains ATF3 as a down gene and reads the list as loss of an activating cascade; that reading is defensible for the CHAC1 axis and is unsupported everywhere else.

And a DepMap consequence. ATF3's leukaemia gene effect is +0.1022, the most positive in this beat's panel - knocking it out is, if anything, marginally good for leukaemia cells, which is what you would expect of a repressor and is the opposite of what you would expect of a required activator.

Facts and figures
uniprotP18847
length AA181
entry nameATF3_HUMAN
protein nameCyclic AMP-dependent transcription factor ATF-3
ensembl gene IDENSG00000162772
PDB noteUniProt P18847 cross-references NO PDB entries.
alphafold IDAF-P18847-F1
accession resolutiongene_exact:ATF3 AND organism_id:9606 AND reviewed:true returns EXACTLY ONE reviewed row, P18847, ATF3_HUMAN, 181 aa, and UniProt lists NO synonyms at all - the only symbol in this beat's 22-gene panel with zero alias exposure. Re-fetched alone with name and length compared back: PASS.
localizationNUCLEUS, and it is the thinnest localisation record in this beat's panel: UniProt P18847 carries exactly ONE location statement, Nucleus, with ECO:0000255 (sequence-model inference from the bZIP profile) plus ECO:0000269 from PubMed 12034827, which this beat did not open. No cytoplasmic pool, no isoform-resolved statement, no subnuclear qualifier. No ATF3 localisation experiment exists in any KMT2A-rearranged cell.
subnuclear localizationNOT ASSIGNED and NEVER LOOKED FOR. UniProt's single Nucleus call has no subnuclear qualifier and no nuclear body, speckle or chromatin-address observation for ATF3 exists in any cell type. Recorded as a bounded absence; no compartment is invented.
cell typeRS4;11 (t(4;11) KMT2A::AFF1, -1.16 array / -1.84 RNA-seq), SEM (t(4;11), only -0.36), REH (KMT2A wild type, +0.06 to +0.45 on three probes - a clean negative control) and MOLM13 (KMT2A::MLLT3, -0.74), measured here [E1]. The cascade work is in human aortic endothelial cells [E5]. No primary leukaemia sample, no normal haematopoietic cell.
cell stateMeasured in the day-6/day-7 drug-arrested state. ATF3 is a classic IMMEDIATE-EARLY stress-response gene, induced within minutes to hours by a very wide range of stimuli, so its level is a sensitive function of handling, medium change and confluence - none of which was controlled or reported in either public experiment. Its RNA-seq DMSO baseline rises 1.45 log2 from day 2 to day 7, so the untreated cultures are themselves becoming more stressed.
epigenetic readoutNo writer, eraser or reader activity of its own; it may repress by stabilising inhibitory cofactors at promoters, which is a chromatin-adjacent mechanism with no mark attached. THE EPIGENETIC DATUM HERE IS THE DRUG COMPARISON: EPZ5676, a DOT1L inhibitor removing H3K79me2/me3, lowers ATF3 by 2.81 log2 against VTP-50469's 1.84 - so of all the panel members ATF3's expression is the most tightly coupled to H3K79 methylation rather than to the menin interface. No histone-mark measurement at the ATF3 locus exists in any KMT2A-rearranged cell.
shown for fusionMEASURED in KMT2A::AFF1 (RS4;11 on three probes plus RNA-seq; SEM weakly at -0.36) with a clean KMT2A-wild-type REH control, and in KMT2A::MLLT3 (MOLM13 -0.74). Transcript only, and at a low baseline. No ATF3 protein, no ChIP and no functional test in any fusion context.
absence classliterature-absence-bounded
absence class noteRECLASSIFIED 2026-09-23 (cycle 76). This field held free text, which `scripts/check-absence-class.py` rejects; the prose is preserved here VERBATIM and unsummarised because it is more informative than the token that replaced it. Token chosen: literature-absence-bounded - the two-part prose says MEASURED-AND-<something> for the data arm and BOUNDED-UNPUBLISHED for the prose arm. The data arm is a RESULT, not an absence; the absence is the bounded literature, and the bound's own limits (titles and abstracts only, per rule 80) are stated in the preserved prose. ORIGINAL absence_class, verbatim: «MEASURED-AND-MODEST in transcript at a low baseline; BOUNDED-UNPUBLISHED in prose (1 fusion-scoped record, off-beat on title-check; 0 menin-drug-scoped with a valid querytranslation and 'No items found.').»
PDB IDS
[]
domains
[
  {
    "name": "bZIP (basic leucine zipper)",
    "start": 86,
    "end": 149
  }
]
the measurement
{
  "rs411_array_mi22_by_probe": {
    "202672_s_at": -1.164,
    "1554980_a_at": -0.76,
    "1554420_at": "-0.290 AT BACKGROUND (DMSO 3.79)"
  },
  "sem_array_mi22_best_probe": -0.363,
  "reh_array_mi22_by_probe": {
    "202672_s_at": 0.061,
    "1554980_a_at": 0.447,
    "1554420_at": 0.271
  },
  "rs411_rnaseq_d7_vtp_lfc": -1.843,
  "rs411_rnaseq_d7_epz5676_lfc": -2.811,
  "rs411_rnaseq_d7_dmso_log2cpm": 4.74,
  "rs411_rnaseq_d2_vtp_lfc": -1.446,
  "molm13_rnaseq_d7_vtp_lfc": -0.744,
  "menin_specificity": "DOT1L-INHIBITOR LEANING - the DOT1L inhibitor produces a 0.97 log2 LARGER fall than the menin inhibitor. One of only two genes in the 37-gene specificity table that lean that way (the other is CEBPB)."
}
essentiality
{
  "depmap_release": "26.06 (read back from meta{dataVersion})",
  "depmap_screens": 1258,
  "chronos_mean": 0.0538,
  "pct_lines_below_minus_0_5": 0,
  "leukaemia_screens": 57,
  "leukaemia_mean_gene_effect": 0.1022,
  "reading": "the most POSITIVE leukaemia gene effect in this beat's panel - consistent with a transcriptional repressor, and inconsistent with a required activator"
}
identifier traps
{
  "registry_status": "NOT in data/symbol-collisions.json as of 2026-09-22 (157 rows read) - and correctly so.",
  "no_alias_exposure": "UniProt lists NO synonyms for ATF3. It is the one symbol in this beat's 22-gene panel with zero alias-widening exposure, which makes its bounded absence the most trustworthy in the set on the name-list axis - and the least trustworthy on the effect-size axis, because of its low baseline.",
  "array": "GPL570 carries three ATF3 probes spanning DMSO 3.79 to 8.52 log2 RMA; the lowest is at background and is reported as background, not as a null."
}
How this could bite us back
  • ATF3's leukaemia DepMap gene effect is the most positive in the panel (+0.10), so its loss marginally favours leukaemia cell growth - a drug effect that removes it is doing the cells a small favour on that axis.
  • ATF3 is the panel member whose response is most tightly coupled to DOT1L inhibition rather than to menin inhibition, so counting it among the menin-inhibitor signature genes weakens rather than strengthens the menin-interface reading of the whole set - and it is one of the five genes from which the ATF4 signature was inferred.
  • If ATF3 is a repressor at some of these promoters, then losing ATF3 should RAISE some of the very genes the graph records as falling. No promoter-level experiment exists to say which, so the sign of ATF3's contribution to the observed signature is genuinely unknown.
Still open
  1. Is the DOT1L-over-menin reversal on ATF3 real, or a low-count artefact? A qPCR or a deeper RNA-seq on ATF3 in the same three arms would settle it, and it is the clearest single datum bearing on whether this programme is menin-specific at all.
  2. Does ATF3 activate or repress at the CHAC1, TRIB3 and STC2 promoters in a KMT2A-rearranged B cell? Its general annotation is repressor and its cascade role is activator, and nothing in this graph distinguishes them in this lineage.
  3. Why do RS4;11 and SEM disagree by 0.8 log2 on ATF3, by 1.0 on CHAC1 and by 1.8 on ATF5, under the same compound at the same dose? Three of four array-measured cascade members show the same discrepancy, which suggests a property of the GSE63664 experiment rather than of any one gene.
  4. Is ATF3's level in these cultures reporting the drug or the handling? An immediate-early gene measured at day 6 and day 7 with no synchronisation and a rising DMSO baseline is a fragile readout.
Every source behind this entry (5)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    ATF3 falls 1.16 log2 on its best-expressed GPL570 probe under MI-2-2 in RS4;11 (p 0.008) but only 0.36 in SEM, and rises slightly in KMT2A-wild-type REH (+0.06 to +0.45 on three probes); under the potent compounds at day 7 the ordering reverses, with VTP-50469 giving -1.84 and EPZ5676 giving -2.81, a 0.97 log2 larger effect for the DOT1L inhibitor.

    What this evidence cannot show

    LOW BASELINE is the main limitation: ATF3's RNA-seq DMSO level in RS4;11 is 4.74 log2 CPM at day 7 and 3.29 at day 2, and one of its three array probes (1554420_at, DMSO 3.79) sits below the 6.0 log2 detection floor and is reported as background rather than as a null. At those counts a 1-2 log2 difference between two treated arms is not robust, so the DOT1L-versus-menin reversal is graded strong as a measurement and moderate as an interpretation. SEM disagrees with RS4;11 by 0.8 log2, the same pattern seen for CHAC1 and ATF5. RS4;11 array DMSO arm n=2.

    Evidence that argues against this (1)
    • The reversal could be a low-count artefact rather than a biological difference: at a DMSO baseline of 4.74 log2 CPM, the difference between -1.84 and -2.81 is a handful of counts per replicate, and neither arm has enough depth on this gene to make a 0.97 log2 difference between two treatments meaningful. The conclusion 'the DOT1L inhibitor hits ATF3 harder' should not be quoted without that qualifier. NCBI GEO, re-analysed here 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    In DepMap 26.06, ATF3 has a mean CRISPR gene effect of +0.0538 across 1,258 screens with 0.0% of lines below -0.5 and a leukaemia mean of +0.1022 across 57 leukaemia screens - the most positive leukaemia gene effect in this beat's panel.

    What this evidence cannot show

    Release recorded per rule 83; gene-specific screen count. A leukaemia mean of +0.10 is small and within genome-wide screen noise; the defensible statement is 'not a dependency, and if anything its loss is mildly favourable', consistent with a transcriptional repressor. Leukaemia screens by substring match on diseaseFromSource, ids persisted per rule 90.

    Evidence that argues against this (1)
    • A positive gene effect in rich medium says nothing about a stress-induced gene under stress, and ATF3 is one of the most reliably stress-induced transcripts in the genome. The screen never applies the stimulus that makes ATF3 relevant, so this null is a statement about the assay's conditions. Mungrue IN et al. 2009
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    UniProt P18847 is the single reviewed human ATF3 record, 181 aa with a bZIP domain at 86-149 and no listed synonyms, annotated as binding the cAMP response element and REPRESSING transcription from promoters with ATF sites, possibly by stabilising the binding of inhibitory cofactors at the promoter, and localised to the nucleus.

    What this evidence cannot show

    Resolved live 2026-09-22: exactly one reviewed record (rule 62 satisfied), re-fetched alone with name, length and primary symbol compared back: PASS. The Nucleus call carries ECO:0000255 (sequence-model inference from the bZIP profile) AND ECO:0000269 from PubMed 12034827, which was NOT opened by this beat - so per rule 91 the coded citation's relevance to ATF3's compartment is unchecked. ATF3 has NO recorded synonyms in UniProt, which is unusual and makes it the one clean symbol in this beat's panel.

    Evidence that argues against this (1)
    • The repressor annotation and the cascade role are in tension: in the ATF4-ATF3-CHOP cascade that induces CHAC1, ATF3 is required as an ACTIVATING component. Quoting UniProt's repressor line as though it settled ATF3's direction of effect would be wrong in precisely the context this graph cares about. Mungrue IN et al. 2009
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PubMed returns 1 record for ATF3 jointly with an eight-operand KMT2A-fusion list - PMID 35906225, on a CDK9 activity switch associated with AFF1 and HEXIM1 in epidermal progenitors, which is off-beat for ATF3 - and 0 for ATF3 with a thirteen-name menin-inhibitor list, with a non-empty querytranslation and warninglist 'No items found.'

    What this evidence cannot show

    The single hit's title was fetched and read; it matches on AFF1 and is not an ATF3 paper in any relevant sense, so the effective fusion-scoped count for ATF3 is zero. Rule 3 satisfied on the menin-drug zero. Rule 94 form; rule 95 observed. Rule 80: titles and abstracts only. ATF3 has no aliases to widen with, so unusually there is no name-list exposure here at all.

    Evidence that argues against this (1)
    • ATF3 has 5,480-scale visibility problems in reverse: it is a very heavily studied immediate-early gene appearing in thousands of papers as a stress marker, so a fusion-scoped count of one means the KMT2A field has not looked rather than that ATF3 is irrelevant. An absence on a well-studied gene is more informative about the field than about the gene, and less informative than the same absence on an obscure one. NCBI PubMed 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    CHAC1 is activated by the ATF4-ATF3-CHOP cascade, established by siRNA and overexpression, and specifically not by the parallel XBP1 or ATF6 pathways - placing ATF3 as a required activating component of this particular cascade despite its general annotation as a repressor.

    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 chemical UPR induction - not leukaemia, not a fusion, not a menin inhibitor. Already cited once in this graph (rule 82). Cited here specifically for ATF3's position in the cascade, which is the transferable part; the cell type is not.

    Evidence that argues against this (1)
    • Whether ATF3 activates or represses is promoter- and partner-dependent, so a requirement for ATF3 in CHAC1 induction in endothelial cells does not license reading 'ATF3 down' as 'this cascade is off' in a leukaemia cell, where the available bZIP partners differ. ATF3's positive DepMap leukaemia gene effect is at least as consistent with a repressor role there. UniProt Knowledgebase 2026
Connected to