Oncofusion

start / graph / hyp-menin-inhibition-relocates-the-fusion-onto-the-unused-ucpg-address

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q83-fusion-fate

Hypothesis: menin inhibition does not clear the fusion from chromatin but redistributes it onto the four fifths of the unmethylated CpG address it normally ignores

also called the relocation hypothesis · fusion redistribution under menin inhibition

The fusion protein can stick to a particular chemical pattern in DNA, and there is five times more of that pattern than it actually uses. Menin seems to pick which fifth. Take menin away and the fusion may not fall off at all - it may spread out over the rest, doing nothing useful at first, but sitting somewhere new.

The same thing, for a scientist

**The null, stated first and taken seriously.** The field's model is that a menin inhibitor removes the fusion from chromatin, transcription of the programme collapses, and the displaced protein is degraded or diluted. It is supported by the only in-patient observation - oncoprotein undetectable under revumenib [E4] - and by menin siRNA reducing fusion binding at target genes in SEM [E2]. **On present evidence this null is the better-supported reading, and this hypothesis is the alternative that has never been tested.** **The case for the alternative.** Three facts fit together uncomfortably. (i) The fusion binds unmethylated CpG through its own CXXC domain and occupies only **20%** of available uCpG sites, while CFP1 occupies 50% and KDM2B 89% in the same cells - so four fifths of the chemically competent address is unused [E2]. (ii) Menin marks which fifth: MLL(N) correlates with menin at r-squared 0.96 at fusion peaks, against 0.27 for menin versus CFP1 [E2]. (iii) A menin inhibitor produces global menin loss but **does not globally displace MLL1, and in a fusion-only line the loss pattern is the same** - the protein is still on chromatin [E1]. A CXXC-competent protein that remains on chromatin after its selectivity factor is removed is the textbook setting for a loss of site selection, and the only reason nobody has seen it is that nobody has looked for gained sites [E3]. **The prediction that makes this testable rather than rhetorical.** Under drug, the fusion's genome-wide profile should become **less menin-like and more uCpG-like** - broader, flatter, shifted towards the CFP1/KDM2B distribution - while its total chromatin-bound quantity, spike-in calibrated, changes much less than target-gene occupancy does. The relocated protein is predicted to be transcriptionally inert at first (no SEC, no PAFc, no menin-stabilised spreading [E2]) and therefore invisible to every RNA readout, which is why three cycles of transcriptomics found nothing suggesting it. **Prior probability: 0.2.** Low, because the mechanism's own siRNA data point at loss, not broadening; worth testing because it is cheap to test, because the retained-occupancy fact is already established, and because two of two proteins examined for gains under this drug had them.

Go deeper
What each prediction distinguishes
predictionrelocationeviction + degradationstays-and-stops-working
treated-only fusion peaks at previously unoccupied uCpGmanynonenone
spike-in calibrated total chromatin-bound fusionroughly constantfallsroughly constant
fusion share of all uCpG sitesrises above 20%fallsunchanged
correlation of fusion profile with meninfalls sharplyundefined (no signal)falls modestly
correlation with CFP1 / KDM2B profilerisesundefinedunchanged
fusion protein level by westernunchangedfallsunchanged
effect blocked in MEN1-M327I cells at the same doseyesyesyes

The third and fifth rows are the ones only this hypothesis predicts.

Why relocation would matter clinically

If the fusion relocates rather than leaves:

  • Resistance has a substrate that requires no new recruitment. Anything that supplies an activating cofactor at the new sites - a liberated JUND, a de-repressed PRC1.1 target, an LSD1-dependent enhancer - converts silent occupancy into transcription without the fusion having to find its way back.
  • Drug holidays are more dangerous than they look. A protein parked on the address re-selects instantly when menin returns, so hysteresis would be short and intermittent dosing would be a poor strategy.
  • Depth of response should correlate with how much uCpG substrate exists, which is a methylome property - and this graph already records that enasidenib, by restoring TET activity and unmethylated CpG, would expand that substrate.
  • A degrader beats an interface inhibitor. If the problem is that the protein stays, removing it is categorically better than un-addressing it, which is an argument for fusion-directed degradation over deeper menin blockade.
Facts and figures
statementUnder pharmacological menin inhibition the KMT2A fusion is not cleared from chromatin but redistributes from the menin-selected ~20% of unmethylated CpG sites onto the remaining ~80%, in a transcriptionally inert state that is invisible to expression-based readouts.
originliterature-gap
prior probability0.2
null hypothesisThe fusion is removed from chromatin and degraded or diluted; this is the field's position and is currently better supported.
shown for fusionno - proposed for KMT2A::AFF1; the retention observation that motivates it is in KMT2A::AFDN (ML2) and the uCpG occupancy figure is in KMT2A::AFF1 (SEM)
predictions
[
  "treated-only fusion peaks at uCpG sites unoccupied at baseline",
  "spike-in calibrated total chromatin-bound fusion roughly constant while target-gene occupancy falls",
  "fusion share of all uCpG sites rises above the 20% baseline",
  "correlation of the fusion profile with menin falls and with CFP1/KDM2B rises",
  "fusion protein level unchanged by western with no proteasome rescue",
  "all of the above abolished in MEN1 interface-mutant cells at the same dose"
]
experiments
[
  "exp-calibrated-early-timecourse-fusion-profiling-under-menin-inhibition",
  "exp-resistant-versus-sensitive-fusion-fate-men1-interface-mutant"
]
analyses
[
  "ana-reanalysis-of-ml2-mll1n-chip-for-gained-peaks-under-vtp-50469"
]
falsifiers
[
  "Spike-in calibrated fusion occupancy falls globally under drug with no treated-only peak population at any timepoint from 1 to 72 hours - this kills relocation outright",
  "Fusion protein level falls under drug and is rescued by proteasome inhibition, making the occupancy change a consequence of degradation",
  "Treated-only peaks appear equally in MEN1 interface-mutant cells, showing they are not caused by menin displacement",
  "The fusion's correlation with CFP1 and KDM2B profiles does not rise under drug even though total bound fusion is preserved, which would mean retention without loss of site selection - the stays-and-stops-working reading instead"
]
How this could bite us back
  • This hypothesis, if true, says an approved paediatric drug leaves its target oncoprotein on chromatin. That is a mechanistic claim, not an efficacy claim, and it must not be communicated in a way that discourages use of a therapy producing complete remissions in relapsed KMT2A-rearranged leukaemia.
  • If relocation is real, the relocated fusion is a reservoir that needs no re-recruitment, so resistance can be fast and combination partners chosen to deepen menin blockade would be the wrong choice.
  • It also predicts that a fusion-directed degrader would be categorically better than an interface inhibitor, which is a large and expensive development implication to rest on an untested hypothesis.
  • The hypothesis is unfalsifiable as usually stated - 'it went somewhere' - unless the gain analysis is pre-specified with a calibrated total. Every falsifier listed depends on spike-in calibration; without it the hypothesis should not be pursued.
Still open
  1. Is there enough unmethylated CpG substrate in an infant KMT2A::AFF1 blast, as opposed to a cell line, for relocation to be quantitatively meaningful? The 20/50/89 figures come from SEM and the primary-sample methylome may differ.
  2. Would a relocated fusion be transcriptionally silent, or would it deposit H3K4me3 and prime new loci? The fusion retains no SET domain, but it carries menin-independent partners on the 3' side, and priming without activation is the scenario with the longest-term consequences.
  3. Does the CXXC domain's affinity for uCpG DNA support residence at non-selected sites at all? This depends on the affinity dispute Q82 is resolving - 33 nM by SPR versus tens of micromolar by NMR - and at the weak end relocation is thermodynamically implausible.
Every source behind this entry (4)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    A menin inhibitor produces global loss of menin from chromatin while explicitly not globally displacing MLL1, including in ML2 where the wild-type KMT2A allele is lost and all MLL1n signal is fusion - so the fusion remains chromatin-associated under conditions of near-complete menin displacement.

    PMID 31821784DOI
    What this evidence cannot show

    TSS-anchored quantification, day 3, one concentration, depth-normalised without spike-in on the ChIP lane, no residual-fraction statistic, adult-derived lines. 'Not globally displaced' is a qualitative reading of tornado plots.

    Evidence that argues against this (2)
    • A murine MLL-AF9 system reports a significant decrease in MLL1 chromatin enrichment under MI-503, which if it holds for the fusion would remove this hypothesis's starting premise in that system. Soto-Feliciano YM et al. 2023
    • Depth normalisation can hide a global loss: if every site loses signal proportionally, a read-depth-normalised profile looks unchanged. Without spike-in calibration 'not globally displaced' and 'uniformly reduced everywhere' are the same picture, and the second is compatible with eviction. Perner F et al. 2023
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In SEM cells the fusion occupies only 20% of unmethylated CpG sites against 50% for CFP1 and 89% for KDM2B, MLL(N) signal correlates with menin at r-squared 0.96 at fusion peaks against 0.27 for menin versus CFP1, and menin siRNA reduces MLL-AF4 binding at target genes with the authors concluding mutual co-stabilisation.

    PMID 28076791DOI
    What this evidence cannot show

    One infant-derived KMT2A::AFF1 line. Peak-overlap percentages depend on calling stringency. The correlation is between two antibodies against two members of one complex at a peak set defined by one of them. Crucially the same paper's knockdown data support menin being required for fusion binding at target genes, which is evidence against this hypothesis and is recorded here rather than in the caveat because it is the central tension.

    Evidence that argues against this (2)
    • The siRNA result is the strongest single argument against relocation: if partial menin depletion reduces fusion binding at target genes, complete pharmacological displacement should reduce it further, not redistribute it. The counter is that ChIP-qPCR at target loci measures only the sites the fusion is leaving and says nothing about where the protein went. Kerry J et al. 2017
    • 'Sufficient but not necessary' - the phrasing that originally motivated this hypothesis - is a figure title in that paper with no supporting experiment; see ana-sufficient-but-not-necessary-is-a-figure-title-not-an-experiment. The hypothesis survives without it but its prior is lower than the brief assumed. Kerry J et al. 2017
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    No study in the class has tested a KMT2A fusion for gained chromatin sites under a menin inhibitor, while the same pipelines detected more than 13,000 new UTX sites and increased JUND occupancy when the gain question was asked of those proteins.

    A Molecular Switch between Mammalian MLL Complexes Dictates Response to Menin-MLL Inhibition. · Soto-Feliciano YM, Sánchez-Rivera FJ, Perner F, et al. · Cancer Discov 2023
    PMID 36264143DOI
    What this evidence cannot show

    The enumeration behind this is 8 datasets; two were read at abstract level only. UTX and JUND both had specific mechanistic reasons to gain, so two-of-two is not a base rate for arbitrary proteins.

    Evidence that argues against this (1)
    • That two other proteins gain sites is equally consistent with a model in which the fusion leaves and other factors take over the vacated address - which is Soto-Feliciano's own thesis and requires no fusion relocation at all. Soto-Feliciano YM et al. 2023
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In the only human sample profiled under revumenib, the oncoprotein and menin were reduced to undetectable levels while ENL persisted at target loci - the observation most consistent with the null of removal rather than relocation.

    MLL oncoprotein levels influence leukemia lineage identities. · Janssens DH, Duran M, Otto DJ, et al. · Nat Commun 2024
    PMID 39472576DOI
    What this evidence cannot show

    n = 1 and confounded by a concurrent lineage switch that independently lowers oncoprotein levels; undetectable CUT&RUN signal conflates loss of binding with loss of protein, and no gain analysis was run on that sample.

    Evidence that argues against this (1)
    • If the oncoprotein protein level fell in that patient, the sample is uninformative about relocation, because relocation requires the protein to exist. A paired protein-level measurement would decide whether this counts as evidence at all. Wu Y et al. 2019
Connected to