Oncofusion

start / graph / concept-chromatin-and-epigenetics

Explainerestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.Q3-mechanism-core

Plain-English: chromatin and epigenetics — the sticky notes on the DNA, and which ones matter here

also called what is epigenetics · histone marks explained · DNA methylation explained · readers writers erasers

Every cell has the same DNA but reads different parts of it, because the DNA is wrapped on spools carrying chemical sticky notes that say 'read me' or 'skip me'. KMT2A-fusion leukemia is a disease of those notes, not of the DNA letters themselves.

The same thing, for a scientist

Chromatin is DNA wrapped on histone octamers. Post-translational histone modifications and DNA cytosine methylation constitute a layer of heritable regulatory information that is read by reader domains, deposited by writers and removed by erasers. In KMT2A-rearranged leukemia the relevant marks are: unmethylated CpG DNA (read by the retained KMT2A CXXC domain), H3K36me2/3 (read by LEDGF's PWWP domain), H3K79me2/3 (written by DOT1L, with no known eraser), H3K4me3 (read by the LOST KMT2A PHD3 finger), H3K27me3 (written by PRC2, read by CBX8's chromodomain) and H3K27ac/H4K16ac at enhancers. The disease is a mis-reading and mis-writing of this layer with an almost silent DNA mutational landscape.

Go deeper
The vocabulary, in the order it matters for this disease2 deeper

Two metres of DNA fit into a nucleus by wrapping around protein spools (histones). The tails of those spools carry small chemical tags, and the DNA itself can carry a methyl group on cytosine. Together these are the 'epigenetic' layer: information about how to read the genome that is not written in the genome's letters.

The grammar has three parts, and it is worth learning because the whole KMT2A mechanism is built from it:

  • Writers put tags on. DOT1L writes H3K79 methylation. PRC2 writes H3K27me3. DNMTs methylate DNA.
  • Readers recognise tags through specialised pockets. KMT2A's CXXC domain reads unmethylated CpG DNA 1. LEDGF's PWWP domain reads H3K36-methylated nucleosomes, gripping both the histone tail and both turns of DNA at once 2. ENL's YEATS domain reads acetylated histones. CBX8's chromodomain reads H3K27me3.
  • Erasers take tags off — demethylases, deacetylases. Note a crucial gap: there is no known H3K79 demethylase. That single fact explains why DOT1L inhibitors are slow (see mechanism-dot1l-h3k79-methylation): the mark can only be diluted away by cell division.

A KMT2A fusion is, in this vocabulary, a reader bolted to a writer-recruiter. Its front half reads unmethylated CpG; its back half recruits the machinery that writes H3K79 methylation and releases polymerase. Nothing about it is a new chemical activity; it is a new wiring of existing activities.

Why 'epigenetic' is the right word for this leukemia specifically

Infant KMT2A-rearranged B-ALL has an exceptionally silent DNA mutational landscape — consistent with its prenatal origin — which is the explicit reason alternative epigenetic mechanisms are thought to contribute substantially to its leukemogenesis 3. Most cancers are diseases of accumulated mutation; this one arrives with almost none. One translocation, and then a regulatory-layer disease.

That is genuinely hopeful. Mutations cannot be reversed; sticky notes can. It is also why the field's disappointments matter so much: a demethylating agent in infants was safe and demonstrably removed methylation in 87% of samples, and still did not improve survival (see mechanism-kmt2a-r-dna-methylation-signature). Being an epigenetic disease does not guarantee that changing the epigenome helps.

A warning about how these marks are measured

Most claims about histone marks come from ChIP-seq averaged over millions of cells. That method cannot tell the difference between (a) every cell having two different marks on the same spool and (b) half the cells having one mark and half the other. The first is 'bivalency'; the second is heterogeneity. They mean completely different things biologically, and the field has often reported the first when it measured something compatible with the second — see mechanism-bivalent-primed-chromatin-fusion-targets.

A second warning: marks are usually measured as relative enrichment, not absolute amounts, so 'H3K4me3 went up' can mean the genuine amount rose or that something else fell. Calibrated methods (ICeChIP, spike-in normalisation) fix this and occasionally overturn conclusions.

When reading any claim in this graph about a histone mark, the two questions to ask are: was it single-cell or population? and was it calibrated or relative?

Two textbook framings that do not hold here (added 2026-09-19)

'It is a chromatin disease, so epigenetic drugs should work.' Tested directly. COG AALL15P1 gave five days of azacitidine before chemotherapy in four post-induction courses to infants with newly diagnosed KMT2A-r ALL. It was safe (DLT in 2 of 31 evaluable, 6.5%) and pharmacodynamically on target (decreased DNA methylation by WGBS in 87% of samples tested), and event-free survival was similar to prior studies [E3 counter-evidence]. Target engagement was achieved and the disease did not care.

'A chromatin disease is a differentiation block, so release the block.' That is the PML::RARA/AML story. In KMT2A-rearranged ALL, menin inhibition produces rapid dose-dependent leukaemic cell death, not differentiation; the slow differentiating response is what KMT2A-r AML does [E3 counter-evidence]. The same drug, the same fusion class, two different mechanisms of benefit - which is a caution against reading any single 'epigenetic therapy' narrative across lineages.

Facts and figures
audiencelayman entry point
marks that matter here
[
  {
    "mark": "unmethylated CpG DNA",
    "reader": "KMT2A CXXC (retained in fusion)",
    "role": "targets the fusion to CpG-island promoters"
  },
  {
    "mark": "H3K36me2/3",
    "reader": "LEDGF PWWP",
    "role": "anchors the clamp to transcribed chromatin"
  },
  {
    "mark": "H3K79me2/3",
    "writer": "DOT1L",
    "eraser": "none known",
    "role": "sustains fusion target expression; removable only by replicative dilution"
  },
  {
    "mark": "H3K4me3",
    "reader": "KMT2A PHD3 (LOST in fusion)",
    "role": "the off-switch the fusion discards"
  },
  {
    "mark": "H3K27me3",
    "writer": "PRC2 (EED/SUZ12/EZH1/EZH2)",
    "reader": "CBX8 chromodomain",
    "role": "repressive default at loci such as PROM1"
  },
  {
    "mark": "H3K27ac / H4K16ac",
    "writers": "p300/CBP; MOF/KAT8",
    "role": "enhancer activity; H4K16ac is the critical output of wild-type KMT2A"
  }
]
measurement warnings
[
  "population ChIP cannot distinguish bivalency from cell heterogeneity",
  "relative enrichment is not absolute quantity; prefer calibrated ICeChIP or spike-in normalisation"
]
links to
[
  "mechanism-bivalent-primed-chromatin-fusion-targets",
  "mechanism-dot1l-h3k79-methylation",
  "mechanism-kmt2a-r-dna-methylation-signature"
]
How this could bite us back
  • Epigenetic marks are shared machinery: every writer, reader and eraser named here also operates in normal cells, so 'epigenetic therapy' has no intrinsic selectivity.
  • Being an epigenetic disease does not mean epigenetic drugs work — azacitidine removed methylation in 87% of treated infants without improving survival.
  • Because the marks are maintained by ongoing enzyme activity, they can be re-established after drug withdrawal, making durability rather than initial response the hard problem.
Still open
  1. Is there a reader for H3K79me2? None is firmly established, which is odd for such a functionally important mark and would be a fusion-adjacent target if one existed.
  2. Given that no H3K79 demethylase is known, is the mark removed only by replication, and does that mean slowly dividing leukemia stem cells retain it longest — making them the cells least affected by a DOT1L inhibitor?
  3. How much of the KMT2A-r epigenome is inherited from the fetal cell of origin rather than created by the fusion? Nobody has profiled the matched pre-leukemic fetal progenitor.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The KMT2A CXXC domain binds non-methylated CpG DNA and discriminates against methylated CpG; point mutations abolishing this binding also abolish transactivation and myeloid transformation by a KMT2A fusion protein.

    Binding to nonmethylated CpG DNA is essential for target recognition, transactivation, and myeloid transformation by an MLL oncoprotein · Ayton PM, Chen EH, Cleary ML · Molecular and Cellular Biology 24(23):10470-10478 2004
    PMID 15542854DOI
    What this evidence cannot show

    In-vitro myeloid immortalisation with retroviral overexpression; cited here as the canonical reader-domain example rather than for the full targeting mechanism.

    Evidence that argues against this (1)
    • CXXC reading of unmethylated CpG is necessary but on its own is not the targeting mechanism. Okuda et al. showed that *two* modules are necessary and sufficient - non-methylated CpG binding and H3K36me2/3 binding - and that an artificial two-module protein with an AF4-coactivator domain substitutes for the fusion. The single-reader picture in this explainer understates the requirement for a second, transcription-coupled mark. Okuda H et al. 2014
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Cryo-EM of the LEDGF PWWP domain bound to an H3K36-methylated nucleosome shows multivalent binding to the methylated histone tail and to both gyres of nucleosomal DNA, explaining how PWWP-domain proteins are recruited to H3K36-methylated chromatin.

    Structure of H3K36-methylated nucleosome-PWWP complex reveals multivalent cross-gyre binding · Wang H, Farnung L, Dienemann C, Cramer P · Nature Structural & Molecular Biology 27(1):8-13 2020
    PMID 31819277DOI
    What this evidence cannot show

    Reconstituted nucleosome with an isolated domain in vitro; no leukemia system present.

    Evidence that argues against this (1)
    • A cryo-EM structure of an isolated PWWP domain on a reconstituted nucleosome shows what the domain CAN do at equilibrium with purified components. In a nucleus LEDGF is in complex with menin and KMT2A, H3K36 methylation is widespread across transcribed gene bodies, and the selectivity that matters is which of many methylated nucleosomes the complex occupies — which multivalency alone does not explain. Cancer Research 2014
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Infant B-ALL shows an exceptionally silent DNA mutational landscape consistent with its prenatal origin, which is the stated rationale for alternative epigenetic mechanisms contributing substantially to its leukemogenesis; genome-wide methylome and transcriptome analysis of 69 Interfant-99/06-treated patients segregated KMT2A-rearranged from non-rearranged cases.

    Integrative methylome-transcriptome analysis unravels cancer cell vulnerabilities in infant MLL-rearranged B cell acute lymphoblastic leukemia · Tejedor JR, Bueno C, Vinyoles M, et al. (Menendez P, senior author) · Journal of Clinical Investigation 131(13):e138833 2021
    PMID 33983906DOI
    What this evidence cannot show

    'Exceptionally silent' is a comparative statement against other cancers; KMT2A-r infant ALL does carry recurrent co-mutations (notably RAS-pathway) in a substantial minority of cases.

    Evidence that argues against this (2)
    • The clinical corollary most readers take from 'a disease of the epigenetic layer' - that epigenetic drugs should work - has been tested in exactly this population and failed. COG AALL15P1 gave azacitidine priming to infants with newly diagnosed KMT2A-r ALL, demonstrated decreased DNA methylation by WGBS in 87% of samples tested, and saw event-free survival similar to prior studies. A silent mutational landscape licenses the search for epigenetic mechanism; it does not license the inference that demethylating the genome helps. Guest EM et al. 2024
    • The other textbook corollary - that a chromatin disease is a differentiation block to be released - is an AML narrative and does not describe KMT2A-r ALL. Adriaanse et al. found that KMT2A-r ALL cells show rapid, dose-dependent death under menin inhibition while KMT2A-r AML responds slowly and differentiates. Anyone importing the PML::RARA/ATRA differentiation framing into infant ALL is importing the wrong disease. Adriaanse FRS et al. 2024
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Calibrated, internally standardised ICeChIP-seq measurement of histone marks in a KMT2A::AFF1 line revealed that DOT1L inhibition depletes H3K79me2 with paradoxical increases in H3K4me3 and loss of H3K27me3 — a result demonstrating why calibrated rather than relative mark measurement matters.

    Non-canonical H3K79me2-dependent pathways promote the survival of MLL-rearranged leukemia · Richter WF, Shah RN, Ruthenburg AJ · eLife 10:e64960 2021
    PMID 34263728DOI
    What this evidence cannot show

    Single cell line under drug perturbation; cited here as a methodological illustration rather than as a general biological claim.

    Evidence that argues against this (1)
    • The calibrated-ChIP result is one experiment, and this beat cites it in three separate entries as if it were three observations. Internal calibration fixes the scale but not the biology of the spike-in: paradoxical mark movements can arise when the standard's occupancy differs between conditions, and a single cell line under one inhibitor for one duration cannot establish which direction marks move in the disease. The entry also draws a methodological moral — that calibrated measurement matters — from a result whose surprising direction is the very thing a single uncorroborated experiment cannot establish. eLife 2021
Connected to