Oncofusion

start / graph / mechanism-h3k79-demethylation-contested

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q3b-dot1l-h3k79

Nobody has convincingly found the H3K79 eraser - and that constraint sets the clock on every DOT1L inhibitor

also called H3K79 demethylase · H3K79 eraser · KDM2B H3K79 demethylase claim · absence of an H3K79 demethylase

Most chemical tags on histones have an enzyme that removes them; for H3K79 nobody has convincingly found one. If the tag can only be lost by being diluted out as cells divide, then a drug that blocks the enzyme has to be kept on board for several cell divisions before the tag disappears - which is exactly what the DOT1L trial did not do.

The same thing, for a scientist

As of the only papers that make the claim, the field's own position is that the H3K79 demethylase **remains unknown despite intensive research** [E1]. One laboratory has reported KDM2B - canonically an H3K36 demethylase and the CxxC-targeting subunit of non-canonical PRC1.1 - as a di- and tri-methyl H3K79 demethylase, first in a transcriptional-repression context [E1] and then in a DNA-replication context [E2]. The literature is strikingly thin: PubMed returns **3 records** for `KDM2B[tiab] AND H3K79[tiab]` and **2** for the phrase `"H3K79 demethylase"[tiab]`, both from the same group, so this is a single-laboratory claim that has not been independently replicated in eight years. A separate line of work finds that pharmacological inhibition of KDM7A/B raises H3K79me in chondrocytes and protects mouse joints [E3], which establishes that H3K79me levels are **pharmacologically raisable** without establishing a direct demethylase. Against all of this sits a measurement that complicates the tidy dilution story: in synchronised human cells with stable-isotope labelling, **pre-existing histones continue to acquire H3K79 mono- and di-methylation at a rate equal to newly synthesised histones**, implying positional scrambling of the mark through the cell cycle [E4]. So the resolution mechanism is neither a known eraser nor pure replicative dilution, and this uncertainty is load-bearing for inhibitor scheduling.

Go deeper
Why an absent eraser is a therapeutic constraint, not a curiosity

If H3K79 methylation is removed only by replication-coupled dilution, then the time to clear the mark is set by the cell cycle, not by the drug. Each division halves the parental histone pool at a locus; a leukaemic blast dividing every 24-48 hours needs several divisions for the mark to fall to a level where the fusion's target genes shut down. A DOT1L inhibitor must therefore maintain target coverage across that whole window, and any gap lets the rheostat refill (mechanism-h3k79-nonprocessive-deposition).

Pinometostat was given by continuous intravenous infusion precisely because of this reasoning, and it still under-delivered - with the maximum tolerated dose never reached 5. That combination is the load-bearing caveat for the whole beat: the trial tested a schedule chosen for the right reason at an exposure that was never pushed to its limit.

The isotope-labelling result makes the clock less predictable rather than more 4. If old histones keep being methylated at the same rate as new ones, then dilution is not a clean first-order decay and the mark can redistribute, so the number of divisions required to clear a locus is not simply log2 of the starting occupancy.

The KDM2B claim, and how much weight it can carry

The claim is specific and testable. KDM2B, a Jumonji-C family protein known for H3K36 demethylase activity, is reported to demethylate di- and tri-methyl H3K79; to repress HOXA7 and MEIS1 by occupying their promoters and demethylating H3K79; to raise genome-wide H3K79 methylation when depleted; and to act via displacement of SIRT1 from chromatin 1. A follow-up from the same group reports PCNA binding chromatin in an H3K79me-dependent manner and KDM2B timing its dissociation 2.

Three things argue for taking it seriously in this disease specifically:

  • the reported targets are HOXA7 and MEIS1 - the exact programme a KMT2A fusion runs;
  • KDM2B is the CxxC-domain targeting subunit of PRC1.1, and PRC1.1 has just been shown to be the mediator of the cellular response to both DOT1L and menin inhibitors, with H3K79 methylation and PRC1 activity in direct biochemical antagonism 6. An enzyme that both recruits PRC1.1 and removes H3K79me would be the single point at which that antagonism is executed;
  • KDM2B is independently implicated in leukaemia (14 PubMed records for KDM2B AND leuk*).

And three that argue for holding it at weak: two papers, one laboratory, no independent replication, and a substrate assignment that is intrinsically hard because H3K36 and H3K79 methyl-lysine peptides are both plausible JmjC substrates and the readouts are largely antibody-based.

What a bounded absence claim here is worth

This entry states a counted absence, so it states its symbol set. The searches run for it were KDM2B[tiab] AND H3K79[tiab] (3 records), "H3K79 demethylase"[tiab] (2), and H3K79[tiab] AND demethylase[tiab] (14). The wider query's 14 records include the KDM7A/B osteoarthritis work 3 and papers about demethylases of other residues that mention H3K79 incidentally.

What this cannot exclude: a demethylase reported under a complex name rather than a subunit symbol; an activity described as "H3K79me2 erasure" without the word demethylase; and any enzyme whose H3K79 activity was found but framed as an H3K36 result. This graph has been caught twice by complex-name unreachability, so the absence is stated as not independently replicated rather than as does not exist.

Facts and figures
mechanism classmark removal / eraser biology
established erasernone
default modelreplication-coupled dilution of the parental histone pool
complication to default modelold histones continue to acquire K79 mono- and di-methylation at the same rate as new ones (PMID 20699226)
therapeutic consequenceif loss is replication-coupled, clearing the mark takes several cell divisions and inhibitor coverage must span them; pinometostat used continuous IV infusion for this reason and never reached its maximum tolerated dose
mark
[
  "mark-h3k79me1",
  "mark-h3k79me2",
  "mark-h3k79me3"
]
claimed erasers
[
  {
    "enzyme": "KDM2B (FBXL10)",
    "states": "me2, me3",
    "labs": 1,
    "papers": 2,
    "replicated_independently": false,
    "pmids": [
      "29763382",
      "33029857"
    ]
  },
  {
    "enzyme": "KDM7A/B",
    "evidence_type": "pharmacological and knockdown, levels only; no direct demethylation assay reported in the abstract",
    "pmids": [
      "36927643"
    ]
  }
]
counted absence
{
  "queries_run": [
    {
      "query": "KDM2B[tiab] AND H3K79[tiab]",
      "records": 3
    },
    {
      "query": "\"H3K79 demethylase\"[tiab]",
      "records": 2
    },
    {
      "query": "H3K79[tiab] AND demethylase[tiab]",
      "records": 14
    },
    {
      "query": "KDM2B[tiab] AND (leukemia[tiab] OR leukaemia[tiab])",
      "records": 14
    },
    {
      "query": "KDM2B[tiab] AND PRC1[tiab]",
      "records": 32
    }
  ],
  "run_date": "2026-09-19",
  "cannot_exclude": "an eraser reported under a complex name rather than a subunit symbol, an activity described without the word demethylase, or an H3K79 activity framed as an H3K36 result"
}
How this could bite us back
  • An inhibitor whose effect requires several cell divisions to appear is at its weakest against the quiescent leukaemia-initiating cells that cause relapse, because those cells dilute the mark most slowly - the opposite of the desired selectivity.
  • If an H3K79 demethylase does exist and is inducible, a leukaemia could resist a DOT1L inhibitor faster than the dilution model predicts, and nothing in the current pharmacodynamic toolkit would distinguish induced erasure from failure to inhibit.
  • Conversely, if no eraser exists, then normal haematopoietic cells also cannot re-set the mark quickly after drug withdrawal, so the recovery of normal marrow is on the same slow clock as the leukaemic response.
Still open
  1. Does KDM2B demethylate H3K79me2 in a defined in vitro system with recombinant nucleosomes and mass-spectrometric readout, run by a laboratory other than the one that reported it? That single experiment would move this from weak to settled in either direction.
  2. How many cell divisions does H3K79me2 at HOXA9 actually take to clear under saturating DOT1L inhibition in a KMT2A-rearranged line, measured by calibrated ChIP rather than by bulk western blot? The number is the schedule.
  3. If PRC1.1 antagonises H3K79 methylation biochemically, and KDM2B targets PRC1.1 to CpG islands, is locus-level H3K79me loss competitive rather than enzymatic - and does that make KDM2B a targeting factor mistaken for an eraser?
  4. Why does H3K79me decay rapidly in injured mouse cartilage, a minimally proliferative tissue, if the mark is lost by dilution?
Every source behind this entry (6)
  1. E1weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    The demethylase of H3K79 remains unknown despite intensive research; KDM2B (FBXL10), a Jumonji-C family protein known as an H3K36 demethylase, is reported to be a di- and tri-methyl H3K79 demethylase that represses HOXA7 and MEIS1 by occupying their promoters and demethylating H3K79, with H3K79 methylation rising genome-wide on KDM2B depletion and SIRT1 displaced from chromatin in stable KDM2B-knockdown lines.

    PMID 29763382DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Single laboratory; PubMed returns only 3 records for KDM2B AND H3K79 and 2 for the phrase 'H3K79 demethylase', both from this group, with no independent replication as of this cycle. Substrate assignment for a JmjC enzyme already known to act on H3K36 is difficult and the readouts described in the abstract are antibody-based. No leukaemic model, although the named target genes (HOXA7, MEIS1) are the KMT2A-fusion programme. Confidence is set to weak on the replication ground, not because the experiments described are poor.

    Evidence that argues against this (1)
    • The kinetic mechanism predicts that state-specific demethylation is intrinsically hard to achieve on this residue: because Dot1 methylates non-processively the states co-occur, which the original authors said 'is expected to limit the possibilities for regulation by putative demethylases or binding proteins'. That is a prior against, not a refutation, and it is the reason a two-paper claim should not be promoted. Frederiks F et al. 2008
  2. E2weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    PCNA binds chromatin in an H3K79me-dependent manner during S phase, shown by immunoprecipitation, peptide pull-down, isothermal titration calorimetry and ChIP; KDM2B is responsible for the timely dissociation of PCNA from chromatin, and KDM2B depletion aberrantly enriches chromatin with PCNA, slows residual PCNA dissociation and impairs cell proliferation.

    PMID 33029857DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Same laboratory as E1, so the two papers are not independent. HEK293T cells and a histone H3-K79R mutant; no haematopoietic or leukaemic context. The paper assumes rather than re-establishes KDM2B's H3K79 demethylase activity, so it inherits E1's uncertainty rather than corroborating it.

    Evidence that argues against this (1)
    • An equally consistent reading of the same data is the reverse causality: if H3K79 methylation is lost by replicative dilution, PCNA and the replication fork are what removes it, and an enzyme that controls fork behaviour would change H3K79me levels without being a demethylase at all. Sweet SMM et al. 2010
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In cartilage, inhibition of the Jumonji-C demethylase family raises H3K79 methylation in human chondrocytes; the KDM2/7 subfamily inhibitor daminozide increases H3K79me and glycosaminoglycans, H3K79me decays rapidly in mouse articular cartilage after joint destabilisation, intra-articular daminozide enhances H3K79me and protects against osteoarthritis, and individual silencing shows KDM7A/B mediate those effects.

    PMID 36927643DOI
    What this evidence cannot show

    Structured abstract read in full from the eutils record; full text not read. Cartilage and chondrocytes, not blood. The design shows that H3K79me levels are RAISED by inhibiting KDM7A/B, which is consistent with direct demethylation and equally consistent with an indirect route through any of the many genes those demethylases regulate - no in vitro demethylation assay on an H3K79me substrate is described in the abstract. Daminozide is a KDM2/7 subfamily inhibitor and therefore not selective for KDM7A/B; the silencing experiments are what narrow it.

    Evidence that argues against this (1)
    • The rapid decay of H3K79me observed in injured cartilage is the opposite of what the leukaemia field assumes about this mark, where slow replication-coupled loss is the standard model. Chondrocytes in adult cartilage divide very little, so a rapid decay there is hard to attribute to dilution and hard to reconcile with the leukaemia model - one of the two settings is being described wrongly, and this entry does not resolve which. Sweet SMM et al. 2010
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Using stable-isotope labelling of both histones and their methyl groups in synchronised human cells, resolved for the H3.1, H3.2 and H3.3 isoforms, pre-existing 'old' histones continue to be K79 mono- and di-methylated at a rate equal to newly synthesised histones, implying some degree of positional scrambling of K79 methylation through the cell cycle; the rate of establishing K79 methylation is the same for all three isoforms although H3.3K79me1 levels are higher.

    Kinetics of re-establishing H3K79 methylation marks in global human chromatin. · Sweet SMM, Li M, Thomas PM, et al. · J Biol Chem 2010
    PMID 20699226DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Global chromatin in a synchronised human cell line, measured by mass spectrometry - no locus resolution, so it cannot say whether the same scrambling happens at a fusion-target gene body where the enzyme is tethered. Trimethylation kinetics are not reported in the abstract. 'Scrambling' is the authors' interpretation of equal labelling rates on old and new histones, not a directly observed movement of methyl groups.

    Evidence that argues against this (1)
    • Continued methylation of old histones is also exactly what a non-processive enzyme acting on an unsaturated substrate pool looks like, with no scrambling required: if most nucleosomes are below their steady-state methylation level, old and new histones will both gain methyl groups at similar rates. The redundancy/rheostat model therefore explains the observation without positional redistribution. Frederiks F et al. 2008
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Pinometostat was administered as a continuous intravenous infusion and reduced H3K79 methylation in patients, with 2 complete remissions among 51 adults with acute leukaemia.

    PMID 29724899DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Adult phase 1; this entry uses it only for the route of administration and the outcome counts, and the inference that scheduling was chosen for kinetic reasons is this graph's reading rather than a claim in the abstract.

    Evidence that argues against this (1)
    • Resistance to DOT1L inhibition in KMT2A-rearranged ALL can be acquired through adaptation rather than through failure to deplete the mark, so attributing the clinical shortfall to schedule alone is not supported; a longer or deeper exposure may select the same adaptation sooner. Schneider P et al. 2023
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Enhanced PRC1.1 activity in leukaemia cells arises specifically from the progressive loss of DOT1L-mediated H3K79 methylation, independent of MLL-fusion displacement or transcriptional repression, and is driven by direct biochemical antagonism between H3K79 methylation and PRC1 activity.

    DOT1L provides transcriptional memory through PRC1.1 antagonism. · Neville D, Ferguson DT, Heikamp EB, et al. · Nat Cell Biol 2026
    PMID 41634341DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Cited here for the H3K79me-PRC1 antagonism and the word 'progressive', which is the load-bearing detail - the PRC1.1 response tracks the slow loss of the mark rather than the fast displacement of the fusion. KDM2B is the CxxC-targeting subunit of PRC1.1 but the abstract does not name KDM2B, so the link drawn in this entry between the KDM2B demethylase claim and PRC1.1 is an inference of this graph and not of the paper.

    Evidence that argues against this (1)
    • If H3K79 methylation and PRC1 antagonise each other biochemically at the nucleosome, then PRC1.1's apparent role in clearing the mark may be competitive exclusion rather than enzymatic erasure, which would make an H3K79 demethylase unnecessary to explain locus-level loss and further weaken the case for KDM2B acting as one. Kang JY et al. 2018
Connected to