Oncofusion

start / graph / ana-the-potent-menin-inhibitor-collapses-what-the-weak-one-left-untouched-and-hoxa9-was-never-measured-by-either

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.

The experiment that refutes cycle 55's headline was already public: VTP-50469 collapses MEIS1 32-fold in the same line where MI-2-2 left it untouched - and HOXA9 was never actually measured by either dataset

also called GSE127506 · VTP-50469 · revumenib comparator · MI-2-2 versus VTP-50469

Last cycle we found that a menin-blocking drug left the leukaemia's key genes untouched, and said plainly that we could not rule out the drug simply being too weak. A stronger drug from the same class was already tested on the same cell line in public data, and it switches those genes off hard, so the weak-drug explanation was the right one.

The same thing, for a scientist

Cycle 55's analysis of GSE63664 found MI-2-2 (10 uM, 6 days) leaving HOXA7, HOXA10 and MEIS1 unmoved in two KMT2A::AFF1 lines, and its `bite_back` named the one thing it could not exclude: that the compound was too weak. **GSE127506** (subseries of GSE127508, PMID 31821784) settles it. Same line, **RS4;11**; **VTP-50469 at 330 nM**, well above the 20-60 nM IC50 the series quotes; days 2 and 7; **n=3 in all 12 groups**; 3'-end RNA-seq. The panel was **copied verbatim from the cycle-55 script**, which declared it before that dataset was read and before this one was known to exist. **The same cell line, the same genes, two compounds:** | gene | MI-2-2 10 uM, 6 d | VTP-50469 330 nM, d7 | detected? | |---|---:|---:|---| | MEIS1 | **+0.21** | **-5.25** | yes (log2CPM 6.68) | | MEF2C | -0.12 | **-4.35** | yes (7.53) | | PBX3 | -0.16 | **-2.15** | yes (7.96) | | HOXA10 | **+0.45** | **-1.46** | yes (6.63) | | FLT3 | -0.40 | -1.38 | yes (8.35) | | CDK6 | -0.14 | -1.34 | yes (9.41) | | CDKN1A | -0.29 | -1.13 | yes (7.52) | Housekeeping genes are flat in both (ACTB -0.04/-0.35, GAPDH -0.24/-0.42, TBP +0.15/+0.33), and the drug's own machinery is untouched (MEN1 -0.30, DOT1L -0.06, AFF1 +0.01, PSIP1 -0.06), so the collapse is specific rather than a global shutdown. MOLM13 agrees (MEIS1 -2.96, MEF2C -2.01, PBX3 -2.45). **Cycle 55's headline was a fact about MI-2-2 and not about menin inhibition, and this entry exists to say so.** **And the second finding is worse for the field than for us: HOXA9 has never been measured by either dataset.** On GPL570 it has no single-gene probe. Here its RS4;11 DMSO baseline is **log2CPM 0.76** and in MOLM13 - a canonically HOXA9-high MLL-AF9 line - every HOXA9 value is **0.00**. HOXA7 is 0.00 in RS4;11 too. So the project's canonical marker gene is **undetected in both menin-inhibitor datasets it has analysed**, and every HOXA9 number either analysis produced is noise on an absent transcript. **What survives from cycle 55: ATF5.** Flagged there as the most reproducible repression in GSE63664, and it repeats here under a different compound on a different platform - **-2.43 at day 7** on a well-expressed baseline of 8.44.

Facts and figures
scriptanalyses/vtp50469-does-it-collapse-hoxa-where-mi22-did-not/run.py
resultsanalyses/vtp50469-does-it-collapse-hoxa-where-mi22-did-not/results.json
correctsana-the-menin-inhibitor-response-is-nine-fold-fusion-selective-and-does-not-touch-hoxa-or-meis1-in-either-kmt2a-aff1-line
seriesGSE127506, subseries of GSE127508, PMID 31821784
designMOLM13 (KMT2A::MLLT3) and RS411 (KMT2A::AFF1); DMSO, EPZ5676 1 uM (pinometostat, DOT1L inhibitor), or VTP-50469 330 nM; days 2 and 7; n=3 in every one of the 12 groups; 3'-end RNA-seq htseq counts by gene symbol.
why this dataset settles itIt holds the one variable cycle 55 could not: compound potency. VTP-50469 at 330 nM is well above the 20-60 nM IC50 range the series summary quotes, so a null result here could NOT have been blamed on underdosing - and it is not null.
atf5 survivesCycle 55 flagged ATF5 as the most reproducible repression in GSE63664 (-3.02/-2.89/-2.77 on three probes in SEM, -1.09/-1.12/-0.74 in RS4;11, flat in the wild-type line). Here it is -2.43 at day 7 in RS4;11 on a baseline of 8.44, under a different compound and a different platform. That is cross-compound, cross-platform reproduction and it is the one cycle-55 positive that strengthens.
mi22 movers recheck17 of 27 MI-2-2 movers present in this matrix move the same way under VTP-50469 at day 7. Of the disagreements, 7 (IGJ, MS4A1, TPH2, LGR5, CLEC14A, INSM1, EDIL3) have an RS4;11 DMSO baseline of exactly 0.00 and are undetected rather than contradicted. Genuine cross-compound disagreements on detected genes: FCGR2B, LSP1, SDC2.
shown for fusiontrue
absence classmeasured-not-analysed
the side by side in rs411
{
  "MEIS1": "MI-2-2 +0.21 -> VTP -2.59 (d2) and -5.25 (d7); DMSO log2CPM 6.68",
  "MEF2C": "MI-2-2 -0.12 -> VTP -3.29 and -4.35; 7.53",
  "PBX3": "MI-2-2 -0.16 -> VTP -1.38 and -2.15; 7.96",
  "HOXA10": "MI-2-2 +0.45 -> VTP -0.88 and -1.46; 6.63",
  "FLT3": "MI-2-2 -0.40 -> VTP -0.48 and -1.38; 8.35",
  "CDK6": "MI-2-2 -0.14 -> VTP -0.48 and -1.34; 9.41",
  "CDKN1A": "MI-2-2 -0.29 -> VTP -0.25 and -1.13; 7.52"
}
hoxa9 was never measured
{
  "on_gpl570": "no single-gene probe; the three probes whose annotation mentions HOXA9 are multi-gene and were reported for completeness only",
  "in_this_rnaseq": "RS4;11 DMSO d7 log2CPM 0.76 - effectively undetected",
  "the_tell": "MOLM13, a canonically HOXA9-high MLL-AF9 line, returns 0.00 for HOXA9 in every arm and at both timepoints. A HOXA9-high line reading zero is an assay failure, not biology.",
  "HOXA7": "RS4;11 DMSO d7 log2CPM 0.00 - also undetected",
  "consequence": "Every HOXA9 and HOXA7 number in either analysis is noise on an absent transcript and must not be cited. The genes that carry the result are MEIS1, MEF2C, PBX3 and HOXA10, all with baselines above log2CPM 6.6."
}
global response genes moving 1 log2 of expressed
{
  "RS411 VTP d2": "688 of 12,426",
  "RS411 VTP d7": "1,686 of 12,426 (1,151 up / 535 down)",
  "RS411 EPZ d2": "416",
  "RS411 EPZ d7": "1,527",
  "MOLM13 VTP d2": "883 of 12,482",
  "MOLM13 VTP d7": "936"
}
How this could bite us back
  • This entry corrects a negative result into a positive one, which is the direction in which over-claiming is easiest. It establishes that a potent menin inhibitor collapses a target programme in two cell lines. It does not establish clinical benefit, does not speak to any patient, and says nothing about revumenib specifically - VTP-50469 is a related tool compound, not the approved molecule, and this is cultured cells at a fixed dose for 7 days.
  • Do not read the collapse as isolating the menin interface. The DOT1L inhibitor moves the same genes the same way in the same experiment, so MEIS1, MEF2C and PBX3 report on the elongation complex as a whole rather than on menin engagement specifically.
  • Do not cite any HOXA9 number from this analysis or from the cycle-55 one. HOXA9 is undetected in both datasets and the diagnostic is that a HOXA9-high line reads exactly zero.
  • The cycle-55 entry should be read with its correction attached and not deleted. Its selectivity finding - the ~9x fusion-versus-wild-type response ratio - is untouched by this work, because GSE127506 contains no KMT2A-wild-type line.
Still open
  1. Does the occupancy half of GSE127508 close the absence this project still asserts? The SuperSeries has three further subseries (GSE127507, GSE137813, GSE137814) and the sample titles include RS4;11 ChIP for MENIN, MLL1n, DOT1L and H3K79me2 under VTP-50469 versus DMSO at day 3. `absence-no-menin-inhibitor-occupancy-dataset-in-an-infant-derived-kmt2a-aff1-line` may be narrower than it looks or already refuted for the non-infant case, and that should be checked before the entry is quoted again.
  2. Is HOXA9 measurable in ANY dataset this project has analysed? If the answer is no, then every HOXA9 claim in this graph rests on other people's assays and the graph should say which. A survey of the platforms used by the entries that assert HOXA9 levels is a day's work and would bound a lot of inherited confidence.
  3. Why does ATF5 fall 5- to 8-fold under two chemically unrelated menin inhibitors on two platforms, in fusion lines and not in a KMT2A-wild-type line, while having no described connection to the complex? It is now the best cross-validated novel observation from either dataset and there is still no ATF5 entry.
  4. MI-2-2's 9x fusion-selectivity in GSE63664 remains unexplained by potency. If the compound was too weak to collapse the target programme, what was the 769-gene fusion-selective response it DID produce? A gene-level intersection of the MI-2-2 movers with the VTP-50469 movers would say whether the weak compound was doing a fraction of the same thing or something else entirely.
  5. Does the day-2 versus day-7 split in GSE127506 separate direct from secondary effects? MEIS1 is already -2.59 at day 2 and MEF2C -3.29, while most of the 1,686-gene day-7 response is absent at day 2. That is a natural early/late partition and this analysis does not exploit it.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    In RS4;11, a KMT2A::AFF1 line, the potent selective menin-MLL inhibitor VTP-50469 at 330 nM collapses the canonical menin-KMT2A target programme by day 7: MEIS1 -5.25, MEF2C -4.35, PBX3 -2.15, HOXA10 -1.46, FLT3 -1.38, CDK6 -1.34, CDKN1A -1.13 log2 against the same-line same-day DMSO arm, with effects already present at day 2 (MEIS1 -2.59, MEF2C -3.29, PBX3 -1.38). Housekeeping genes are flat (ACTB -0.35, GAPDH -0.42, TBP +0.33) and the pathway's own machinery is transcriptionally untouched (MEN1 -0.30, DOT1L -0.06, AFF1 +0.01, AFF4 +0.12, PSIP1 -0.06). MOLM13 (KMT2A::MLLT3) agrees: MEIS1 -2.96, PBX3 -2.45, MEF2C -2.01. The DOT1L inhibitor EPZ5676 at 1 uM produces the same direction more weakly in RS4;11 (MEIS1 -1.50, MEF2C -1.71, PBX3 -1.13 at day 7).

    What this evidence cannot show

    Counts are as deposited; no realignment was performed and the gene-symbol assignment is the submitter's. Library sizes are small for RNA-seq (0.46-1.14 M assigned counts per sample), which is expected for 3'-end digital gene expression but means low-expressed genes are poorly estimated - hence the log2CPM >= 1 expression filter on all global counts and the per-gene baselines quoted beside every fold change. No p-values are reported; n=3 would support a test, but the comparison this analysis exists to make is against a cycle-55 result that has none, and effect sizes are what transfer between them. 3'-end sequencing does not resolve isoforms sharing a 3' end.

    Evidence that argues against this (2)
    • This is a different platform, compound, duration and normalisation from the cycle-55 comparison, so strictly the two differ in four ways at once and only the CELL LINE and the gene panel are held constant. Compound potency is the explanation the cycle-55 entry itself nominated in advance, and the effect sizes here are far too large to be a platform artefact, but the comparison is not a controlled one and should not be described as if a single variable had been changed.
    • That the DOT1L inhibitor moves the same genes in the same direction - MEIS1 -1.50, MEF2C -1.71 at day 7 - means these genes are not a specific readout of menin-interface engagement. They report on the elongation complex as a whole. So the collapse confirms the compound works and does NOT by itself isolate the menin interface as the node responsible.
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Cycle 55's headline finding was a fact about MI-2-2, not about menin inhibition, and is corrected by E1. In the SAME cell line RS4;11, MI-2-2 at 10 uM for 6 days gave HOXA10 +0.45, HOXA7 +0.22, MEIS1 +0.21, PBX3 -0.16 and MEF2C -0.12, while VTP-50469 at 330 nM gives HOXA10 -1.46, MEIS1 -5.25, PBX3 -2.15 and MEF2C -4.35. The alternative explanation that the earlier analysis explicitly declined to exclude - that the compound was too weak to engage the target - is the correct one.

    What this evidence cannot show

    What is corrected is the INTERPRETATION, not the measurement. The MI-2-2 numbers are unchanged and were re-read from the stored results rather than retyped, and the cycle-55 observation that the MI-2-2 response was ~9x larger in fusion-positive lines than in the KMT2A-wild-type REH line stands untouched - GSE127506 contains no wild-type line, so it cannot speak to selectivity at all. The two analyses answer different questions and only one of them has been overturned.

    Evidence that argues against this (2)
    • The correction cuts in a direction that should be stated plainly: it REMOVES a result that looked like a challenge to the menin-interface hypothesis. With a potent compound the chain 'interface -> target-programme maintenance' behaves exactly as the field says it does. Anyone who saw the cycle-55 entry and treated flat HOXA as evidence against menin dependence should stop; the graph carried that reading for one cycle and it was an artefact of compound potency.
    • The cycle-55 entry did publish its own escape route - its bite_back named this exact experiment and said the two explanations were unseparated - so this is a caveat discharging as designed rather than a claim collapsing. That is the mildest possible reading of it and it should not be used to imply the entry was fine: its TITLE asserted the unqualified negative anyway, and a title is what gets quoted.
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    HOXA9, the marker gene this project's central argument runs on, was not measured by either menin-inhibitor dataset it has analysed. On Affymetrix GPL570 there is no single-gene HOXA9 probe. In GSE127506's 3'-end RNA-seq the RS4;11 DMSO day-7 baseline is log2CPM 0.76, and MOLM13 - a canonically HOXA9-high MLL-AF9 line - returns exactly 0.00 for HOXA9 in all six of its arm-by-day groups. HOXA7's RS4;11 baseline is likewise 0.00. By contrast the genes that carry the result are well expressed: MEIS1 6.68, HOXA10 6.63, MEF2C 7.53, PBX3 7.96, CDK6 9.41.

    What this evidence cannot show

    A zero count is evidence of non-detection by this assay, not of absent transcript: 3'-end sequencing depends on a well-annotated and well-captured 3' end, and HOXA cluster genes are short, closely spaced and share regulatory sequence. The MOLM13 zero is the diagnostic - an HOXA9-high line reading zero cannot be biology - but the inference that the assay is at fault rather than the annotation is not separated here. What is established is only that neither dataset supports ANY claim about HOXA9, in either direction.

    Evidence that argues against this (1)
    • This undercuts a habit of this project more than it undercuts these two datasets. Both analyses put HOXA9 at the head of a declared panel and both reported a HOXA9 number - flat probes in one case, -0.76 in the other - and in both cases the number was uninterpretable. The correct handling was available in both: check the baseline before reporting the fold change. Cycle 55 did notice the missing probe and said so; it still left HOXA9 in the headline framing.
Connected to