Oncofusion

start / graph / hyp-the-menin-inhibitor-class-splits-at-val371

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q36-menin-residue-axis

The menin-inhibitor class splits at Val371: three of four clinical compounds contact the first residue of the gamma-H2A.X motif and bleximenib does not, so DNA-repair effects should separate from KMT2A-displacement effects within the class

also called V371 contact hypothesis · bleximenib spares the HR surface · menin inhibitor class non-uniformity

Three of the four menin drugs that have been crystallised touch a menin amino acid that is also the first letter of the short stretch menin uses to recognise damaged DNA; the fourth, bleximenib, stops about ten angstroms short of it. If that contact matters, the three should interfere with DNA repair in a way bleximenib does not, at doses that block the leukaemia target equally.

The same thing, for a scientist

**Statement.** Revumenib, ziftomenib and DS-1594b each contact menin V371 at 2.3-4.3 A and E366 at 2.6-3.7 A; bleximenib's nearest approach is 10.72 A to V371 and 8.02 A to E366, and its footprint stops at residue 363 (`interface-menin-drug-pocket-revumenib-ziftomenib`) [E1]. V371 is the first residue of the Val371-Gln400 motif whose internal deletion abolishes menin-gamma-H2A.X binding and abolishes rescue of homologous recombination in Men1-knockdown cells, and whose isolated 30-mer suppresses IR-triggered HR while potentiating NHEJ [E2]. The hypothesis is that **the class is not uniform at this surface, and that a dose-matched comparison of a V371-contacting compound against bleximenib will separate a DNA-repair phenotype from the shared KMT2A-displacement phenotype.** **Prior 0.25, and the reasons it is not higher** are in the counter-evidence: a 2.3 A contact to one residue at a pocket rim is a long way from occluding a 30-residue motif whose ordered part (371-385) projects away from the cavity, and the same compounds differ in a dozen other ways. **Prior is not lower** because the experiment is nearly free - all four compounds are clinical-stage, the readouts are standard, and the graph already carries `exp-hr-competence-under-menin-inhibition-in-kmt2a-rearranged-blasts` as a single-arm version of it.

Go deeper
The four measurements the hypothesis rests on
revumenib 7UJ4ziftomenib 8VA6DS-1594b 8IG0bleximenib 9WKU
min distance to V3714.33 A2.33 A3.11 A10.72 A
min distance to E3662.90 A2.60 A3.67 A8.02 A
highest-numbered contact residue371371371363
% of footprint in the fingers lobe43.546.247.623.8

The first two rows for revumenib, ziftomenib and DS-1594b reproduce ana-q34-menin-surface-geometry; the bleximenib and DS-1594b numbers were re-measured independently in this beat from a separately written script and agree to 0.00 A 1.

What makes this scoreable rather than suggestive is that the class provides its own negative control. Any effect shared by all four compounds is a menin-pocket effect; an effect present in three and absent in the fourth, at matched occupancy, localises to what those three do and the fourth does not - and the largest single thing in that category is contact with the acidic shelf at 366-371.

Predictions, in the order they get cheaper
  1. Biochemical. In a pulldown of recombinant menin by the biotinylated Pep-MH 30-mer (menin 371-400), pre-incubation with ziftomenib or revumenib reduces recovery and pre-incubation with bleximenib does not. This is a competition assay between a compound and a peptide for the same protein and needs no cells.
  2. Cellular, DNA-damage. In an irradiated KMT2A-rearranged line at compound concentrations giving matched HOXA9/MEIS1 suppression, ziftomenib and revumenib delay gamma-H2A.X foci resolution and bleximenib does not.
  3. Cellular, HR reporter. In a DR-GFP or equivalent assay, the V371-contacting compounds reduce HR efficiency and bleximenib does not, with NHEJ reciprocally unchanged or up.
  4. Clinical, and only as a retrospective signal. If the three V371-contacting compounds carry a myelosuppression or secondary-malignancy signal that bleximenib does not, at matched response rates, that is consistent - but confounded by everything, and this prediction should not be weighted.

If prediction 1 fails, the hypothesis is dead cheaply, which is the reason it is ordered this way.

The strongest argument against, stated before the arguments for

The ordered part of the gamma-H2A.X motif points away from the cavity. In the parse of ana-q36-menin-residue-axis, residues 371-385 form a helix (_struct_conf HELX_P 370-385 in 3U84) whose solvent accessibility averages 59.5 A^2 per residue - it is an exposed helix on the outside of the fingers lobe. V371 is its N-terminal cap and is the only residue of the motif any compound comes within 4.5 A of. A ligand touching the first residue of a 30-residue exposed helix has not occluded the helix.

The counter-counter is that the motif's function is not the whole helix either: the published mapping is a 30-mer, the deletion is 30 residues, and nobody has done an alanine scan, so which residues carry the contact is unknown. If the determinants are at the N-terminal end, V371 matters; if they are in the disordered 387-400 tail, no compound is anywhere near them.

There is a second, duller explanation for everything above, and it may be the right one. Compounds differ in size, shape, charge and residence time. A single distance measurement to one residue is a weak proxy for anything, and four compounds differ in far more than one contact. This is why the prior is 0.25 and not 0.5.

Facts and figures
statementThe clinical menin-inhibitor class is not uniform at menin's acidic shelf: revumenib, ziftomenib and DS-1594b contact E366 and V371 at 2.3-4.3 A and bleximenib does not (8.0-10.7 A). Because V371 begins the Val371-Gln400 motif required for the menin-gamma-H2A.X contact and for HR rescue, DNA-repair effects should separate from KMT2A-displacement effects within the class, with bleximenib as the built-in negative control.
origindata-pattern
prior probability0.25
shown for fusionNO - nobody has looked in any fusion. All four co-crystals are menin plus compound with no fusion protein present, and no published DNA-repair readout under any menin inhibitor exists in a KMT2A-rearranged cell; the proposed test is designed in KMT2A::AFF1 (RS4;11, SEM) for exactly that reason.
absence classnever-measured
predictions
[
  "Pep-MH (menin 371-400) pulldown of recombinant menin is competed by ziftomenib and revumenib but not by bleximenib",
  "gamma-H2A.X foci resolution after 2 Gy is delayed by ziftomenib and revumenib and not by bleximenib, at concentrations giving matched HOXA9 and MEIS1 suppression",
  "HR efficiency in a DR-GFP reporter falls under ziftomenib and revumenib and not under bleximenib, with NHEJ unchanged or increased",
  "The shared effects - HOXA9/MEIS1 collapse, differentiation, MEN1 target eviction - are indistinguishable across all four compounds at matched occupancy"
]
experiments
[
  "exp-bleximenib-versus-revumenib-hr-and-gammah2ax-readout"
]
analyses
[
  "ana-q36-menin-residue-axis"
]
falsifiers
[
  "In a KMT2A-rearranged human B-lineage line (RS4;11 or SEM), at compound concentrations titrated to give matched HOXA9 mRNA suppression (within 20% of each other) after 96 h, the fraction of cells with >10 gamma-H2A.X foci at 6 h after 2 Gy differs by LESS THAN 1.3-fold between ziftomenib and bleximenib across three biological replicates. That result kills the hypothesis: the class is uniform on this readout at matched target engagement.",
  "In a biotinylated Pep-MH (menin 371-400) pulldown of recombinant menin, 10 uM ziftomenib reduces recovered menin by LESS THAN 25% relative to DMSO by densitometry, in three independent pulldowns. That kills the biochemical limb outright - the compound and the peptide do not compete for menin even at a concentration two orders of magnitude above the compound's Kd - and makes every cellular prediction an indirect one.",
  "An alanine scan of Pep-MH shows the H2A.X-contacting determinants lie C-terminal of residue 385, i.e. in the segment no compound approaches within 10 A and no structure resolves. That does not kill the hypothesis outright but removes its mechanism, and it should then be withdrawn rather than reinterpreted.",
  "A published head-to-head already exists showing no difference. Bounded 2026-09-22: no PubMed record pairs bleximenib with any gamma-H2A.X, homologous-recombination or DNA-repair readout; this falsifier is listed so that a later cycle checks again rather than inheriting the absence."
]
How this could bite us back
  • If the hypothesis is right, the compound with the best leukaemia data may be the one with the worst genomic-instability liability, and the two cannot be separated by dose because they are the same molecule in the same pocket. In a child cured of leukaemia with forty years of follow-up, a secondary-malignancy signal is the outcome that matters and is the one nobody is powered to see.
  • If the hypothesis is wrong - if all four compounds behave identically on DNA-repair readouts - that is a stronger result for the class than the hypothesis being right, and it should be published as such rather than filed as a negative. A measured null on genomic instability for a drug given to infants is worth more than most positives.
  • Choosing bleximenib on the strength of a structural argument alone, before any of the predictions is tested, would be selecting a compound on geometry rather than on outcome. The footprint difference is 5 residues and its functional meaning is currently zero.
  • The hypothesis assumes the KMT2A-displacement phenotype is the shared one. If bleximenib also displaces KMT2A less well at matched exposure, a clean 'DNA-repair minus, efficacy equal' comparison is not available and the experiment has to be occupancy-matched rather than dose-matched, which is harder and is the commonest way this design fails.
Still open
  1. Which residues inside 371-400 actually contact H2A.X? An alanine scan of the published 30-mer settles both this hypothesis's mechanism and the open question already standing in `interface-menin-h2ax-val371-gln400`.
  2. Is bleximenib's occupancy of the pocket comparable at comparable free concentration? A smaller footprint may mean lower affinity, in which case a 'matched occupancy' arm has to be defined by a target-engagement assay and not by a potency ratio.
  3. Do the four compounds differ in residence time at menin? A slow off-rate could make a transient rim contact functionally larger than the static distance suggests, and residence times are not in this graph for any of the four.
  4. Does a menin degrader, which removes the whole protein including the fingers lobe, produce the DNA-repair phenotype that this hypothesis predicts for the V371-contacting compounds? `risk-a-menin-degrader-is-not-a-pocket-inhibitor-for-the-hr-scaffold` makes the same point from the other direction and the two should be read together.
Every source behind this entry (3)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Revumenib, ziftomenib and DS-1594b contact menin V371 at 4.33, 2.33 and 3.11 A and E366 at 2.90, 2.60 and 3.67 A respectively, while bleximenib's nearest approach is 10.72 A to V371 and 8.02 A to E366 and its 4.5 A footprint ends at residue 363; 23.8% of bleximenib's footprint lies in menin's fingers lobe against 43.5-47.6% for the other three.

    What this evidence cannot show

    One crystal form per compound; rim side chains are the least reproducible part of a structure and 8IG0 is 2.60 A. The lobe fractions depend on the palm|fingers boundary placed at 317/318 by a contact-graph parse with a stated uncertainty of +/- 3 residues; they would move if the boundary moved. Independently reproduces the revumenib, ziftomenib, DS-1594b and bleximenib V371 distances reported by `ana-q34-menin-surface-geometry`, which is a check on this code and not on the deposits.

    Evidence that argues against this (1)
    • All four compounds share a 15-residue contact core and all four displace KMT2A from menin in cells, so the pharmacology that has been measured is uniform. The non-uniformity is at the rim and has no measured phenotype attached to it. Huang J et al. 2012
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Menin residues Val371-Gln400 are necessary and, as an isolated 30-mer, sufficient for the menin-H2A.X contact: an internal deletion of exactly those residues abolishes gamma-H2A.X binding and fails to rescue homologous recombination in Men1-knockdown A549 cells, and a synthetic biotinylated peptide covering 371-400 precipitates wild-type H2A.X and the S139E phosphomimetic, blocks the menin-H2A.X interaction, suppresses IR-triggered HR while potentiating NHEJ, and blocks aphidicolin-induced menin binding at a common fragile site.

    The tumor suppressor menin is a key scaffold mediator for homologous recombination repair. · Lian A, Zhu X, Jin B, Gu G, Xu B, Zhang L, et al. · Nucleic Acids Research 2026
    PMID 42578372DOI
    What this evidence cannot show

    A549 lung adenocarcinoma and Men1-knockout mouse embryonic fibroblasts, with several of the load-bearing panels in supplementary figures that were not retrieved. No leukaemia experiment, no haematopoietic cell, no KMT2A fusion anywhere in the paper, and no affinity measured by any method. The residues inside 371-400 that actually contact H2A.X are named only in Supplementary Fig. S3C.

    Evidence that argues against this (1)
    • The isolated peptide binds wild-type H2A.X and the phosphomimetic equally, so the motif is not phospho-selective in vitro even though the cellular interaction is damage-induced. Selectivity must come from somewhere other than the motif, which weakens any argument that occluding the motif's first residue would have a specific damage-dependent consequence. Lian A et al. 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Menin residues 371-385 form a solvent-exposed helix (HELX_P 370-385 in 3U84) averaging 59.5 A^2 of accessible surface per residue, and 389-397 are observed in none of the 69 menin deposits, so the motif the hypothesis is about is an exposed helix followed by a segment no structure has ever resolved.

    What this evidence cannot show

    Accessibility computed on the isolated chain from one crystal form with a peptide in the pocket. Secondary structure is as deposited by the authors, not recomputed with DSSP, and the two can disagree at helix ends by one or two residues. 'Observed in none of 69' is a statement about the deposit set, which is not 69 independent experiments - many deposits share a construct and a crystal form.

    Evidence that argues against this (1)
    • An exposed helix that is invariant across 68 deposits is also a well-ordered structural element, not a floppy peptide, so an argument that a rim-binding ligand cannot influence it has to explain why a rigid element two turns from the pocket is insulated from it. The geometry cuts both ways. Lin J et al. 2023
Connected to