Oncofusion

start / graph / mechanism-decitabine-degrades-dnmt1-and-lowers-kmt2a-protein

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q81-methylome-axis

Decitabine degrades DNMT1 - and in ALL cells the same treatment lowers KMT2A protein, with the shared CXXC-family DNA-binding module offered as the reason

also called DNMT1 degradation · decitabine mode of action · KMT2A degradation by decitabine

Decitabine gets built into DNA and causes the cell to destroy DNMT1, the enzyme that maintains methyl marks. A 2025 study found that in leukaemia cells the same treatment also lowers the amount of KMT2A protein - and only the cells that lost KMT2A actually responded to the drug.

The same thing, for a scientist

**Two protein-loss events, one drug, and the second is new to this graph's reasoning.** The established half: 5-aza-deoxycytidine triggers **selective proteasomal degradation of DNMT1** and of neither DNMT3A nor DNMT3B, in the nucleus, blocked by proteasome inhibitors, requiring a functional ubiquitin-activating enzyme and the DNMT1 KEN box, BAH domain and NLS, **occurring even in the absence of DNA replication**, and - contrary to the usual textbook account - **not requiring covalent enzyme-DNA bond formation**, since a catalytic-site cysteine-to-serine mutant is still degraded. 5-fluorodeoxyuridine and 5-fluorocytidine do not do it. Overexpressing the APC/C substrate-recognition subunit Cdh1 lowers DNMT1 and Cdh1 is dephosphorylated on drug treatment, implicating APC/C-Cdh1 [E1]. The new half: in ALL cell lines and xenografts, decitabine raised DNMT1 and KMT2A **transcript** through demethylation while **KMT2A protein fell in every line that degraded DNMT1**, and **only the lines with reduced KMT2A protein showed biological effects** - KMT2A wild-type cells arresting in G2 and KMT2A-rearranged cells in G1, attributed to p27/p16. The authors report the effect can act synergistically with menin inhibition, and propose as rationale the **structural similarity between the DNA-binding domains of DNMT1 and KMT2A** [E2]. **In this graph's vocabulary that shared module is the CXXC zinc finger, and DNMT1 and KMT2A are two of exactly twelve human proteins that carry one** [E3]. **If the mechanism is what the authors propose, decitabine is an accidental, low-potency, CXXC-directed agent - and that is a hypothesis with a clean falsifier, not a finding.** Three things are unresolved and all three matter: whether the degraded species is full-length KMT2A, the taspase1 N320/C180 fragments, or the fusion protein; whether the loss is degradation or reduced translation; and whether the synergy with menin inhibition depends on the KMT2A arm at all.

Go deeper
What would distinguish the proposed mechanism from the obvious alternative

The proposal: decitabine-modified DNA traps or destabilises CXXC-containing proteins, DNMT1 and KMT2A alike, because they engage the same kind of DNA.

The obvious alternative: DNMT1 loss causes global demethylation and replication stress; KMT2A protein falls as a downstream consequence of that stress, or of the cell-cycle arrest, with no direct engagement at all. The paper's own observation that transcript goes up while protein goes down is consistent with either.

experimentproposed mechanism predictsalternative predicts
CXXC point mutant KMT2A (C1188D abolishes DNA binding)no loss on decitabineloss unchanged
decitabine in a non-replicating cellKMT2A loss preserved (DNMT1 loss is replication-independent)loss reduced or absent
a non-CXXC chromatin protein of similar abundanceunaffectedmay also fall
thermal proteome profiling for direct engagementKMT2A shiftsno shift

The fourth row already exists as data and was not examined here. PMID 36380143 performed thermal proteome profiling of azacitidine and decitabine in infant KMT2A-rearranged ALL cell lines specifically to define "the target protein-binding landscape of these agents" 4. Whether KMT2A appears in that landscape is answerable today from a published supplementary table, and it is the single cheapest test of the whole proposal.

Which KMT2A species, and why the graph cannot skip the question

This project argues constantly about three KMT2A species: full-length Q03164, the taspase1 cleavage products N320 and C180, and the fusion protein. UniProt models all three; rule 91 records that filtering its location comments on evidence code discards both fragment lines.

The 2025 abstract says only 'KMT2A protein'. The three possibilities are not therapeutically equivalent:

  • Full-length wild-type KMT2A lost - this is a toxicity, not a therapy. It is the allele the patient needs, and germline haploinsufficiency causes Wiedemann-Steiner syndrome.
  • The N320 fragment lost - N320 carries the CXXC and the menin-binding motifs, so this would be mechanistically the most interesting and would overlap directly with what a menin inhibitor does.
  • The fusion lost - this would make decitabine a fusion-directed agent and would be the most important result in the beat.

The entry records the question as open rather than guessing, because the abstract does not say and the figures were not read. A western blot molecular weight in that paper answers it.

Facts and figures
drugdrug-decitabine
primary targetgene-dnmt1
proposed secondary targetgene-kmt2a
species of kmt2a affectedUNKNOWN - full-length, taspase1 N320/C180 fragments or fusion protein not distinguished in the source abstract
loss mechanism establishedDNMT1: proteasomal, nuclear, ubiquitin-dependent, KEN-box/BAH/NLS-dependent, replication-independent, covalent-trapping-independent
loss mechanism proposedKMT2A: inferred from shared DNA-binding-domain similarity; transcript rises while protein falls; not distinguished from a downstream consequence of DNMT1 loss
systemDNMT1 arm: generic cell lines, 2005. KMT2A arm: ADULT ALL cell lines and xenografts, 2025, mixed KMT2A status. Nothing in infant KMT2A::AFF1.
cheapest discriminating testInterrogate the existing thermal proteome profiling of decitabine in infant KMT2A-rearranged ALL lines (PMID 36380143) for KMT2A; and repeat the decitabine exposure in a CXXC point-mutant (C1188D) KMT2A background.
shown for fusionNOT DISTINGUISHED. KMT2A-rearranged lines were included and arrested in G1, but the abstract does not say whether the fusion protein itself was among the species lost.
retracted citation flaggedPMID 15899874 is RETRACTED (RetractionIn PMID 35412342); demoted to weak and replaced as support by PMID 41460614.
retraction hygiene 2026 09 21check-retracted-sources.py, after being fixed this cycle to read retraction warnings in prose as well as in key names, found that this entry cited PMID 15899874 at evidence[1].counter_evidence[0] with no retraction flag in scope, while flagging the same paper correctly at E1 and in the summary. An entry can warn about a retracted paper in one place and lean on it in another. Now annotated in the note itself and marked citation.retracted = true.
How this could bite us back
  • If decitabine lowers KMT2A protein, it lowers WILD-TYPE KMT2A too - the allele the patient needs for normal haematopoiesis and, in germline haploinsufficiency, for neurodevelopment. A mechanism celebrated as on-target for the fusion is simultaneously an on-target toxicity for the healthy allele, in infants with decades of life ahead.
  • The mechanism is attractive enough to be adopted before it is established. It rests on one 2025 abstract, on an inference from domain similarity, and on a protein-down-transcript-up observation that a generic stress response would also produce. This graph should carry it as a proposal with a named falsifier, and any entry citing it as established should be corrected.
Still open
  1. Which KMT2A species falls - full-length, N320, C180 or the fusion? A western blot molecular weight in PMID 39754404 answers it and this beat did not open the figures.
  2. Does KMT2A appear in the published thermal proteome profiling of decitabine in infant KMT2A-rearranged ALL lines (PMID 36380143)? Answerable today from a supplementary table.
  3. Does a CXXC point mutant of KMT2A resist decitabine-induced protein loss? C1188D abolishes CXXC-DNA binding and is the clean falsifier of the proposed mechanism.
  4. Is the reported synergy with menin inhibition dependent on the KMT2A arm, or would it appear equally with any DNMT1-degrading agent? No experiment separates them and the SAVE triplet cannot.
Every source behind this entry (4)
  1. E1weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    5-aza-deoxycytidine induces selective proteasomal degradation of DNMT1 and not DNMT3A or DNMT3B; the degradation is nuclear, blocked by proteasome inhibitors, requires a functional ubiquitin-activating enzyme and the DNMT1 KEN box, bromo-adjacent homology domain and nuclear localization signal, occurs in the absence of DNA replication, and does NOT require covalent enzyme-DNA bond formation since a catalytic-site cysteine-to-serine mutant is still degraded; APC/C-Cdh1 is implicated as the ubiquitin ligase.

    PMID 15899874DOI
    What this evidence cannot show

    *** THIS CITATION IS RETRACTED. *** PMID 15899874 (Mol Cell Biol 2005) carries PublicationType 'Retracted Publication'; RetractionIn Mol Cell Biol 2022;42(5):e0054621, PMID 35412342. Found at the coordinator 2026-09-21 by running scripts/check-retracted-sources.py, which the beat did not run. Retained only as the historical source of the KEN-box/BAH/NLS dissection; it must not be cited as support. The surviving claim - that hypomethylating agents deplete DNMT1 - rests on PMID 41460614 (Peng 2025, Cell Oncol), which is independent and not retracted. Cell-line biochemistry with ectopic mutant and deletion constructs, 2005; no leukaemia model and nothing KMT2A-related. The APC/C-Cdh1 arm rests on overexpression and siRNA in the same system. Read from abstract only. The abstract states the DNMT1 degradation result for both 5-azacytidine and 5-aza-deoxycytidine but reports the mutational dissection for the deoxy compound.

    Evidence that argues against this (2)
    • A retracted paper cannot support a mechanistic claim, and the detailed domain dissection drawn from it is exactly the content a retraction calls into question. The claim survives in narrower form on an independent source. Peng XM et al. 2025
    • The replication-independence and catalytic-mutant results contradict the standard covalent-trapping account of this drug, so the mechanism most widely cited for decitabine is at best incomplete - and any inference from 'decitabine works by trapping CXXC-like DNA-binding proteins' inherits that weakness. Brock L et al. 2025
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In adult acute lymphoblastic leukaemia cell lines and xenograft models, decitabine raised DNMT1 and KMT2A expression through demethylation while KMT2A protein was diminished in every line that experienced DNMT1 degradation; only cells with reduced KMT2A protein showed biological effects (G2 arrest in KMT2A wild-type and G1 arrest in KMT2A-rearranged cells, attributed to p27/p16); the authors attribute the effect to structural similarity between the DNA-binding domains of DNMT1 and KMT2A and report that it can act synergistically with menin inhibition.

    PMID 39754404DOI
    What this evidence cannot show

    Read from abstract only in this beat; the figures were NOT opened. Three things the abstract does not resolve and that this entry therefore does not assert: (i) which KMT2A species falls - full-length, the taspase1 N320/C180 fragments, or the fusion; (ii) whether the loss is degradation or reduced translation, since 'degradation' is inferred from protein down against transcript up rather than from a turnover measurement; (iii) what the menin-inhibition synergy experiment was, which lines, which inhibitor and what combination index. ADULT ALL cell lines and xenografts, mixed KMT2A status; nothing infant and nothing in KMT2A::AFF1 specifically.

    Evidence that argues against this (1)
    • DNMT1 loss causes global demethylation, replication stress and cell-cycle arrest, any of which could lower KMT2A protein downstream with no direct engagement - and the paper's own observation that KMT2A transcript RISES while protein falls is equally consistent with that alternative. **CAVEAT ON THIS NOTE'S OWN SOURCE, added 2026-09-21: the paper cited here, PMID 15899874, is a RETRACTED PUBLICATION** (Mol Cell Biol 2005; retraction confirmed by NCBI esummary pubtype and by Europe PMC PUB_TYPE on 2026-09-21). This entry flags that retraction at E1 and in its summary, but this counter-evidence note was citing it with no flag in scope, which is how a retracted source survives inside an entry that elsewhere warns about it. The mechanistic ALTERNATIVE argued here -- that DNMT1 loss could lower KMT2A protein indirectly through demethylation, replication stress or arrest -- does not depend on the retracted paper and stands on general grounds; what is withdrawn is the specific demonstration of decitabine-induced DNMT1 degradation. Treat the alternative as reasoning, not as a cited result. Ghoshal K et al. 2005
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    DNMT1 and KMT2A are both members of the human CXXC zinc-finger family, whose reviewed membership is exactly twelve proteins, so the 'structural similarity between the DNA-binding domains' proposed as the mechanism is in this graph's terms a shared CXXC module that reads unmethylated CpG.

    What this evidence cannot show

    The twelve-member census is this graph's own prior result (ana-cxxc-domain-family-census-human, written cycle 44 from Pfam PF02008, InterPro IPR002857 and the UniProt CXXC-type zinc-finger feature query, all three agreeing); it was NOT re-derived in this beat and the UniProt URL given here is the accession this beat actually fetched, not the family query. That the proposed shared module IS the CXXC rather than some other shared feature is an interpretation of the 2025 paper's phrase 'DNA-binding domains', not a statement the paper makes in those words.

    Evidence that argues against this (1)
    • DNMT1 and KMT2A share more than a CXXC: both are large nuclear chromatin proteins with BAH-like and PHD-type modules, and the 2025 paper's phrase 'DNA-binding domains' may not mean the CXXC at all - reading it as a CXXC claim is this entry's inference and could be wrong. Brock L et al. 2025
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Thermal proteome profiling of azacitidine and decitabine has already been performed in infant KMT2A-rearranged ALL cell lines to reveal the target protein-binding landscape of these agents, so whether KMT2A is directly engaged by decitabine is answerable from an existing published dataset rather than requiring a new experiment.

    PMID 36380143DOI
    What this evidence cannot show

    The existence of the thermal proteome profiling dataset is stated in the abstract; THE DATASET ITSELF WAS NOT EXAMINED IN THIS BEAT and it is not known whether KMT2A appears in it, whether the profiling covered the relevant cell lines, or whether the data are deposited. The claim here is about what is answerable, not about what the answer is. Thermal proteome profiling detects thermal-stability shifts, which are a proxy for engagement and can be produced by indirect complex changes.

    Evidence that argues against this (1)
    • A thermal-stability shift is not binding, and a protein that binds modified DNA rather than the drug itself need show no shift at all - so a negative result in that dataset would not refute the proposed mechanism, which weakens its value as the cheap test this entry recommends. Brock L et al. 2025
Connected to