Oncofusion

start / graph / hyp-the-isr-collapse-is-a-fusion-programme-readout-not-a-menin-interface-readout

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

The 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

also called ISR is not a menin-specific readout · ATF4 activity not ATF4 transcript

This graph has been treating a stress-response shutdown as the signature of blocking menin. The same shutdown happens when you block a completely different enzyme in the same complex, which means it reports the fusion protein losing its grip on the genome rather than anything specific to menin - and the master switch itself is untouched, so what is lost is the switch being read, not the switch.

The same thing, for a scientist

**Two linked claims, both falsifiable, both derived from a pre-registered re-analysis of public data.** (1) **Mechanism:** the observed collapse of the ATF4 output programme under menin inhibitors is loss of ATF4 **activity**, not of ATF4; ATF4's own transcript does not move (max -0.86 log2 anywhere, against CHAC1 -5.11 in the same samples, on a transcript expressed at the array ceiling) and neither does any of the four eIF2alpha kinases or eIF2alpha itself [E1]. In the ISR, ATF4 activity is set translationally by a 5-prime uORF whose bypass requires phospho-eIF2alpha - and a 3-prime-end RNA-seq assay cannot see that layer even in principle. (2) **Attribution:** the collapse is **not** a menin-interface readout. In the same RS4;11 day-7 samples, EPZ5676 (a DOT1L inhibitor, a different surface of the same complex) reproduces every ISR gene to within 0.02-0.24 log2 - CHAC1 -5.03 vs -5.11, TRIB3 -3.07 vs -3.10, ATF5 -2.36 vs -2.43, DDIT3 -1.59 vs -1.50, SESN2 -1.53 vs -1.42 - while the menin-dependent HOX/MEIS/MEF2C arm discriminates the two compounds by 1.0 to 3.8 log2 [E2]. **So the assay has the resolution to see menin specificity and on the ISR arm it sees none.** **The hypothesis is therefore that the ISR-output programme sits downstream of the fusion complex's H3K79-methylation-dependent elongation output, and reports fusion-programme disruption by either route.** That is simultaneously good news for the lab's fusion-dependence hypothesis - the programme does not move in KMT2A-wild-type REH - and a refutation of menin-interface specificity for this particular readout. **Prior probability 0.6**, because claim 1 is close to a direct observation while claim 2 rests on one cell line, one timepoint and two single doses.

Go deeper
What the hypothesis replaces

The corpus currently states that "the compound-independent core of the menin-inhibitor response in this line is loss of the integrated stress response / ATF4 output". Every word of that is defensible as written and the reading it invites is not. Specifically:

the corpus sayswhat the data support
"loss of the integrated stress response"loss of the ISR's transcriptional output. Nothing upstream moves: four eIF2alpha kinases and eIF2alpha itself are flat within 0.38 log2
"an ATF4 signature" collapsescorrect, and ATF4's own transcript is flat at -0.86, so it is loss of ATF4 activity
"compound-independent core of the menin-inhibitor response"compound-independent across two menin inhibitors - and also reproduced by a DOT1L inhibitor the entry never examined
the signature "tracks the fusion"supported: nothing moves in KMT2A-wild-type REH. This survives intact

So the correction is narrow and the surviving claim is strong. The programme is fusion-dependent; it is not menin-specific; and the lesion is at the activity layer, not the transcript layer.

Predictions, each with the arm that would show it

From claim 1 (activity, not transcript):

  1. ATF4 protein falls substantially in RS4;11 under VTP-50469 while ATF4 mRNA does not. This is the central prediction and it is a single western blot.
  2. Phospho-eIF2alpha falls (or, if the mechanism is degradation or cofactor loss rather than translational, it does not - which is why the blot must carry both lanes).
  3. Ribosome profiling or polysome fractionation shows reduced ATF4 uORF bypass, i.e. ribosomes redistributed from the ATF4 main ORF to the uORF, with no change in ATF4 mRNA in the input.
  4. An ISR activator (tunicamycin, halofuginone, a proteasome inhibitor) partially rescues CHAC1, TRIB3 and STC2 expression in drug-treated cells without restoring MEIS1 or MEF2C. If the block is at eIF2alpha phosphorylation, forcing that phosphorylation should reopen the arm.

From claim 2 (fusion programme, not menin interface):

  1. A dose-response of VTP-50469 and EPZ5676 on CHAC1, ATF5 and MEIS1 shows the ISR genes moving together at all doses while MEIS1 separates. This is the prediction that most directly tests the dose-artefact alternative.
  2. Any other perturbation that disrupts the fusion complex's elongation output - AFF1 or AFF4 knockdown, a CDK9 inhibitor, a BRD4 inhibitor, ENL-YEATS inhibition - also collapses the ISR arm, and does so more completely than it collapses HOXA/MEIS.
  3. In a KMT2A-wild-type line under a POTENT menin inhibitor, the ISR arm does not move. GSE127506/GSE127508 contains no wild-type line at all, so this has never been tested with an adequately engaged compound, and it is the decisive fusion-dependence experiment.
  4. ChIP of KMT2A, menin and H3K79me2 at the CHAC1, TRIB3, ATF5 and STC2 loci in RS4;11 shows fusion-complex occupancy and H3K79me2 that is lost under either drug. If these loci are direct fusion targets, claim 2 is mechanistically complete; if they are not, the collapse is indirect and the hypothesis needs an intermediate.
What would kill it, and what the weakest joint is

The weakest joint is dose, and it is stated first because it is the alternative most likely to be right. The comparison is 330 nM VTP-50469 against 1 uM EPZ5676 - two arbitrary single doses in one deposited experiment. If EPZ5676 at 1 uM is further above its own IC50 than VTP-50469 at 330 nM is above its, then "shared" could mean "both arms saturated on a sensitive gene" and the HOX/MEIS discrimination could mean "one arm saturated, one not". A dose-response in both compounds on the same genes is required and does not exist.

One feature of the data pushes back on that alternative: VTP-50469 produces the LARGER effect on the discriminating genes (MEIS1 -5.25 vs -1.50), so if either arm is under-engaged it is the DOT1L arm - which makes the DOT1L arm's ability to fully reproduce the ISR collapse harder to dismiss, not easier.

The second weakest joint is n. One cell line, one timepoint, three replicates per arm, no dose-response, and the RS4;11 array DMSO arm is n=2 because the submitter excluded a chip for QC.

And a joint that is not weak but is worth naming: the mechanism half of the hypothesis is unfalsifiable from transcript data by construction. "ATF4 mRNA flat, targets collapsed" is compatible with translational control, with increased ATF4 degradation, with loss of a co-activator, and with the targets falling for reasons unrelated to ATF4. The transcript result narrows the space and does not choose within it; only the protein-level experiment does.

Facts and figures
statementThe collapse of the ATF4 output programme under menin inhibitors in KMT2A-rearranged cells is (a) loss of ATF4 ACTIVITY rather than of ATF4, set at the translational layer that a transcript assay cannot see, and (b) a readout of KMT2A-fusion transcriptional-programme disruption by ANY route rather than of menin-interface engagement - because a DOT1L inhibitor reproduces it to within 0.24 log2 while failing to reproduce the HOX/MEIS/MEF2C arm by 1.0-3.8 log2.
origindata-pattern
prior probability0.6
why that priorClaim (a) is close to a direct observation - the transcript null is measured on a gene at the expression ceiling - so on its own it would sit near 0.85. Claim (b) rests on one cell line, one timepoint and two arbitrary single doses with no dose-response, so on its own it would sit near 0.5. The joint hypothesis is graded between them and closer to the weaker half.
what survives from the corpus claimThe FUSION-DEPENDENCE survives intact: none of the 24 concordant genes moves by 1 log2 in the KMT2A-wild-type line REH, and several are expressed there at high level (TSC22D3 11.37, TRIB3 10.31, SESN2 9.99, CHAC1 8.95). What does NOT survive is menin-interface specificity for this programme.
consequence if trueThe ISR signature cannot be used as a pharmacodynamic marker of menin engagement, because either drug produces it; MEIS1 and MEF2C can, because they discriminate by 2.6-3.8 log2. And a menin/DOT1L combination buys nothing on this axis, because either agent saturates it alone.
localizationNOT APPLICABLE - this hypothesis makes no positional claim. It concerns the LAYER at which a regulatory loss occurs (translational rather than transcriptional) and the ROUTE by which it is produced (fusion programme rather than menin interface), neither of which is a compartment. The one localisation-adjacent prediction is that ATF4's p300-dependent nuclear speckle pool (PubMed 16219772) has never been co-imaged with a KMT2A fusion.
subnuclear localizationNOT APPLICABLE AND NOT ASKABLE from the evidence this hypothesis rests on, which is bulk transcript abundance. Recorded explicitly so that no later reader mistakes the H3K79-methylation argument - which is inferred from drug concordance, not from any occupancy measurement - for a positional result. NO ChIP or occupancy experiment at any panel locus exists in any KMT2A-rearranged cell.
cell typeRS4;11 (infant t(4;11) KMT2A::AFF1 B-ALL) is the only cell in which the attribution claim was tested, because it is the only line in which both drug arms and a matched DMSO coexist. Supporting transcript data in SEM (t(4;11), array only), REH (KMT2A wild type, array only) and MOLM13 (KMT2A::MLLT3 AML, RNA-seq only). NO primary sample, NO xenograft, NO normal human haematopoietic cell, NO infant material. The hypothesis is a cell-line hypothesis and its most important untested prediction (prediction 7) requires a cell line that does not exist in the potent-compound dataset.
cell stateThe state is load-bearing and adverse. Everything rests on day 6-7 of continuous drug exposure - the arrested, differentiating state - in cultures whose density and medium exhaustion differ between arms because the drug arrests growth, with NO reported medium composition, amino-acid supply or feeding schedule in either deposited experiment. Several DMSO baselines drift up 1.5-3.1 log2 between day 2 and day 7, so part of every day-7 effect is failure-to-induce. The ISR is definitionally the programme most sensitive to exactly these variables, so of all the hypotheses in this graph this is the one most exposed to culture state, and prediction 4 (ISR-activator rescue) is partly a control for it.
epigenetic readoutThe attribution claim IS an epigenetic claim and it is entirely inferential: EPZ5676 acts by inhibiting DOT1L and thereby removing H3K79me2/me3, and it reproduces the ISR arm to within 0.02-0.24 log2 while failing to reproduce the HOX/MEIS/MEF2C arm - so the hypothesis places the ISR programme downstream of H3K79-methylation-dependent elongation output and the HOX arm downstream of the menin interface. NO H3K79me2, H3K4me3, H3K27me3, KMT2A or menin occupancy measurement at ANY of the 72 panel loci exists in any cell, KMT2A-rearranged or otherwise. Prediction 8 is the experiment that would convert this inference into a measurement.
shown for fusionThe MECHANISM claim (a) is tested only in KMT2A::AFF1 (RS4;11) and KMT2A::MLLT3 (MOLM13), transcript only. The ATTRIBUTION claim (b) is tested only in KMT2A::AFF1 (RS4;11, day 7), because that is the only condition with both drug arms. The FUSION-DEPENDENCE both claims rest on comes from KMT2A-wild-type REH under MI-2-2 only - a weak compound that moved just 80 probes in that line - because GSE127506/GSE127508 contains NO wild-type line. So the decisive fusion-dependence test has never been run with an adequately engaged compound.
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: «This hypothesis exists because of a MEASURED-AND-NULL result (ATF4's flat transcript) and an UNEXAMINED-ARM result (the EPZ5676 comparison, present in public data and never reported). Its central prediction rests on a BOUNDED-UNPUBLISHED absence: no phospho-eIF2alpha, ATF4-protein, ribosome-profiling or polysome measurement exists under a menin inhibitor in any cell, bounded by titles and abstracts only.»
predictions
[
  "ATF4 protein falls substantially in RS4;11 under VTP-50469 at day 7 while ATF4 mRNA does not (the central prediction; one western blot)",
  "phospho-eIF2alpha changes - and if it does NOT, the mechanism is degradation or cofactor loss rather than translational control, which the same blot distinguishes",
  "ribosome profiling or polysome fractionation shows ribosomes redistributed from the ATF4 main ORF to its uORF with unchanged ATF4 mRNA in the input",
  "an ISR activator (tunicamycin, halofuginone, a proteasome inhibitor) partially rescues CHAC1, TRIB3 and STC2 in drug-treated cells WITHOUT restoring MEIS1 or MEF2C",
  "a dose-response of both compounds on CHAC1, ATF5 and MEIS1 shows the ISR genes moving together at all doses while MEIS1 separates - the direct test of the dose-artefact alternative",
  "any other perturbation of the fusion complex's elongation output (AFF1 or AFF4 knockdown, a CDK9 inhibitor, ENL-YEATS inhibition) also collapses the ISR arm, more completely than it collapses HOXA/MEIS",
  "in a KMT2A-WILD-TYPE line under a POTENT menin inhibitor the ISR arm does not move - never tested, because GSE127506/GSE127508 contains no wild-type line at all",
  "ChIP of KMT2A, menin and H3K79me2 at the CHAC1, TRIB3, ATF5 and STC2 loci in RS4;11 shows fusion-complex occupancy and H3K79me2 lost under either drug"
]
falsifiers
[
  "ATF4 PROTEIN is unchanged in RS4;11 under VTP-50469 while its targets collapse. That kills claim (a) outright: the lesion would then be downstream of ATF4 entirely, in a co-activator, in chromatin access at the target loci, or in something with no ATF4 involvement at all.",
  "A dose-response shows the ISR genes separating between the two compounds at sub-saturating doses. That kills claim (b): the shared effect would be a saturation artefact and menin specificity would be restored.",
  "A potent menin inhibitor collapses the ISR arm in a KMT2A-WILD-TYPE line. That kills the fusion-dependence that both claims rest on, and would make the whole programme an off-fusion drug effect.",
  "ATF4 mRNA falls by more than 1 log2 in an adequately powered experiment at an earlier timepoint (6-24 h). That would make claim (a) a late-timepoint artefact: this analysis samples only day 2 and day 7.",
  "KMT2A, menin and H3K79me2 ChIP show NO fusion-complex occupancy at CHAC1, TRIB3, ATF5 or STC2. Claim (b) would then require an intermediate transcription factor between the fusion and the ISR arm, and the hypothesis as stated would be incomplete rather than wrong.",
  "An ISR activator fails to rescue CHAC1/TRIB3/STC2 in drug-treated cells. That argues the block is not at eIF2alpha phosphorylation and pushes claim (a) toward a chromatin-access explanation."
]
experiments
[
  "exp-phospho-eif2alpha-and-atf4-protein-under-vtp-50469-in-rs4-11"
]
analyses
[
  "ana-does-atf4s-own-transcript-move-under-a-menin-inhibitor"
]
How this could bite us back
  • If this hypothesis is right, an ATF4-signature readout cannot confirm menin engagement in a trial - either a menin or a DOT1L inhibitor produces it, and so might any other agent that disrupts the fusion's elongation output. MEIS1 and MEF2C are the markers that discriminate.
  • A menin/DOT1L inhibitor combination gains nothing on this axis, because either agent alone saturates it. If the combination has value it is on the HOX/MEIS arm - and that reframes the rationale for a combination trial.
  • The hypothesis predicts that anything restoring eIF2alpha phosphorylation restores ATF4 output while the drug remains fully bound to menin. That is a resistance route requiring no MEN1 mutation, invisible to a sequencing panel, and plausibly supplied by the marrow niche itself - hypoxia, nutrient limitation and asparaginase all activate the ISR.
  • And the uncomfortable converse: a published result has PERK inhibition RESCUING leukaemia stem cells in an MLL-AF9 model. If a menin inhibitor lowers the same axis, it may be protecting the stem-cell compartment while killing the bulk - and nobody has scored the two fractions separately under a menin inhibitor.
Still open
  1. Does ATF4 protein fall while its mRNA does not? One western blot decides claim (a), and the cells are grown by two laboratories already.
  2. Is the shared VTP-50469/EPZ5676 effect a dose artefact? A dose-response in both compounds on CHAC1, ATF5 and MEIS1 is the single most important missing control in this beat.
  3. Does a potent menin inhibitor spare the ISR arm in a KMT2A-wild-type line? The fusion-dependence that both claims rest on has only ever been tested with a weak compound.
  4. Are CHAC1, TRIB3, ATF5 and STC2 direct fusion-complex targets? If not, an intermediate transcription factor is required and the hypothesis is incomplete as stated.
  5. Can an ISR activator reopen the arm in drug-treated cells without restoring MEIS1? That would separate the two arms pharmacologically and is the cleanest single test of the whole structure.
Every source behind this entry (3)
  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 array probes and 0 of 4 RNA-seq line/day contrasts past 1 log2 with a maximum excursion of -0.86, on a transcript expressed at 13.72 log2 RMA (level with ACTB) and 8.68 log2 CPM; and all four eIF2alpha kinases plus eIF2alpha itself are flat (EIF2AK1 max 0.26, EIF2AK3 max 0.35, EIF2AK4 max 0.38, EIF2S1 max 0.31), while the ATF4 output arm falls 1.42 to 5.11 log2 in the same samples.

    What this evidence cannot show

    A count inside two named public cohorts, established as a statement about mRNA in these samples only. EIF2AK2/PKR is excluded from the 'flat' list because it moves -1.88 at day 7 and sign-flips to +1.40 at day 2 while being flat on a well-expressed array probe - reported as noise, not as a null. Kinase mRNA is a weak proxy for kinase activity, which is post-translational, so the kinase result bounds the transcriptional response only. 3-prime-end RNA-seq cannot see ATF4's 5-prime leader, which is where the proposed mechanism lives.

    Evidence that argues against this (1)
    • Jimenez 2021 reports that menin inhibition causes loss of ATF4 and its signature in MLL-rearranged B-ALL by transcriptional AS WELL AS post-transcriptional mechanisms. If a transcriptional component exists, -0.86 log2 - a 1.8-fold fall - may BE it, excluded here by a pre-registered but arbitrary 1.0 log2 threshold. The hypothesis's mechanism claim would then be half wrong. Jimenez JA et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    At RS4;11 day 7 in the same experiment and platform, EPZ5676 (1 uM DOT1L inhibitor) reproduces every ISR panel gene to within 0.02-0.24 log2 of VTP-50469 (330 nM) - CHAC1 -5.03 vs -5.11, STC2 -4.16 vs -4.40, TRIB3 -3.07 vs -3.10, ATF5 -2.36 vs -2.43, DDIT3 -1.59 vs -1.50, SESN2 -1.53 vs -1.42, ASNS -1.47 vs -1.72, ATF4 -0.87 vs -0.86 - while the menin-dependent target programme discriminates them by 1.0 to 3.8 log2 (MEIS1 -1.50 vs -5.25, MEF2C -1.71 vs -4.35, PBX3 -1.13 vs -2.15); 19 of 37 panel genes are shared and only 8 are menin-selective, none of them an ISR gene.

    What this evidence cannot show

    GRADED STRONG, NOT ESTABLISHED, BECAUSE THE INFERENCE EXCEEDS THE NUMBERS: the individual fold changes are established, the attribution claim built on them is a generalisation from ONE cell line, ONE timepoint and TWO arbitrary single doses. The EPZ5676 arm was present in the public data all along and unreported, so this is a re-analysis of an unexamined arm, not a new experiment. No dose-response exists in either compound, so a potency mismatch cannot be excluded as the explanation for the HOX/MEIS discrimination. Three of the 37 verdicts are floor collisions and are flagged.

    Evidence that argues against this (1)
    • The dose alternative is the serious one and it is not excluded: if 1 uM EPZ5676 sits further above its IC50 than 330 nM VTP-50469 does above its own, 'shared' may mean 'both saturated' on sensitive genes and the HOX/MEIS separation may mean 'one saturated, one not'. The partial defence is that VTP-50469 produces the LARGER effect on the discriminating genes, so the DOT1L arm is if anything the less-engaged one - but that is an argument, not a control. NCBI GEO, re-analysed here 2026
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Menin inhibition is reported to cause loss of ATF4 and of the ATF4-dependent transcriptional signature in MLL-rearranged B-cell acute lymphoblastic leukaemia, with menin-dependent regulation of ATF4 attributed to transcriptional and post-transcriptional mechanisms, in a paper whose primary system is Ewing sarcoma where EWS-FLI1 binds the ATF4 promoter directly.

    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 B-ALL experiment's lines, inhibitor, dose, duration and readout are unknown to this entry, and 'loss of ATF4' is not resolved into mRNA, protein or activity for that arm. RULE 73: the primary system is a non-haematopoietic soft-tissue sarcoma driven by an unrelated fusion. This paper both SUPPORTS the hypothesis (post-transcriptional mechanisms named explicitly) and QUALIFIES it (transcriptional mechanisms named too), which is why it is carried as evidence for the hypothesis rather than as a counter-note against it.

    Evidence that argues against this (1)
    • This paper is about MENIN specifically and says nothing about DOT1L, so it cannot support the attribution half of the hypothesis and could be read against it: if menin regulates ATF4 through a menin-specific mechanism in MLL-rearranged B-ALL, the DOT1L inhibitor reproducing the effect needs an explanation the paper does not supply - most plausibly that both drugs converge on the same output, which is the hypothesis, but equally that the DOT1L result is a dose artefact. NCBI GEO, re-analysed here 2026
Connected to