H3K27me3 - the repressive mark this graph argued about for seventy-seven cycles without ever writing an entry for it
also called H3K27me3 · H3K27 trimethylation · histone H3 lysine 27 trimethylation · trimethylated H3K27
H3K27me3 is the chemical tag that means 'keep this gene switched off', and it is the direct opposite number to the H3K79 tag that this leukaemia is defined by. It is unusual among histone marks because the machine that writes it also reads it, so once a patch exists it spreads itself.
The same thing, for a scientist
H3K27me3 is the trimethyl state of lysine 27 on the histone H3 N-terminal tail, written by PRC2 through EZH2 or EZH1 [E1]. It is a **self-propagating** mark: EED binds H3K27me3 and allosterically stimulates EZH2, and cryo-EM on dinucleosomes shows one PRC2 engaging an *activating* and a *substrate* nucleosome simultaneously [E2]. It is also **switched off by transcription**: H3K36me2/3 allosterically inhibits PRC2 and unmodified H3K36 occupies a critical position in the EZH2-DNA interface, so active chromatin is refractory to the mark [E3]. **In KMT2A-rearranged leukaemia its behaviour is the mirror image of H3K79me2, and the graph should hold three specific observations.** (i) Side-by-side profiling of H3K79me2, H3K4me3, H3K27me3 and H3K36me3 in MLL-AF9 leukaemia stem cells found abnormal profiles **only** for H3K79me2 - H3K27me3 was not aberrant [E4]. (ii) But a **subset** of KMT2A oncofusion binding sites do carry bivalent H3K4me3/H3K27me3, with single-cell heterogeneity suggestive of lineage plasticity [E5]. (iii) Where the mark matters is not at the fusion's activated targets but at the **restraints** PRC2 keeps off: removing PRC2 de-represses *Cdkn2a*, *Cdkn2b*, *Gata2* and *Egr1* in mouse MLL-AF9 [E6], and *CDKN1C* and *TP53INP1* in human B-ALL [E7]. **H3K27me3 is therefore not the fusion's antagonist at HOXA. It is the fusion's collaborator at the genes that would otherwise stop the cell.** The mark has an approved pharmacology (tazemetostat, valemetostat) and is already used as a clinical pharmacodynamic endpoint [E8]; the reciprocal mark on the same residue, H3K27ac, rises when it falls [E1][E9].
Why this entry did not exist, and what its absence cost
Before this beat the graph held exactly five mark- entries - mark-h2bk120ub, mark-h2bk34ub, mark-h3k79me1/2/3 - all on the H3K79/H2Bub axis. EZH2, PRC2 and H3K27me3 appeared in the text of 63 node files with no entry behind any of them, and mechanism-polycomb-prc1-prc2-cbx8 named PRC2 in its own id while no PRC2 subunit existed.
The cost was not that a fact was wrong. It was that the graph could not state the comparison it is built around. mark-h3k79me2 says H3K79me2 has one writer and no established eraser, so it is lost only by dilution. H3K27me3 is the opposite in every respect that matters:
| H3K79me2 | H3K27me3 | |
|---|---|---|
| writers | DOT1L only | EZH2 or EZH1 (redundant) |
| erasers | none established | KDM6A/KDM6B (gene-kdm6a exists in this graph) |
| reader that feeds back | none in the writer complex | EED, inside PRC2 - read-write loop |
| switched off by | - | H3K36me2/3 and H3K4me3, allosterically |
| location | gene bodies, broad domains, some enhancers | promoters and CpG islands of silent developmental genes |
| role in this disease | the fusion's activating output | the fusion's repressive output, via menin -> EZH2 |
| drug consequence | one writer, so one target | two writers, so an EZH2-selective drug fails |
The last row is the one with a clinical consequence, and it is only visible when both marks are entries.
Is H3K27me3 excluded from fusion targets, or is PRC2 excluded?
This is the question the dispatch brief asked, and the answer the literature supports is the marks exclude the enzyme, and the mechanism is known for H3K36 and H3K4 but not for H3K79.
- H3K36me2/3 inhibits PRC2 allosterically, and unmodified H3K36 is structurally required in the EZH2-DNA interface; H3K36A/R mutation reduces H3K27me3 and defective Polycomb repression of HOX genes in flies 3.
- H3K4me3 is the other established active-mark inhibitor of PRC2, and the fusion's chromatin carries aberrant gene-body H3K4me3 that is menin-inhibitor sensitive 5.
- No published allosteric relationship exists between H3K79 methylation and PRC2. The anticorrelation is observed - in calibrated ICeChIP-seq, H3K79me2 depletion at pinometostat-downregulated targets is accompanied by paradoxical H3K4me3 gain and H3K27me3 loss (this graph's
mechanism-bivalent-primed-chromatin-fusion-targets) - but the direction of causation is not established and should not be written as if it were.
So the graph's existing PRC1-side claim, mechanism-dot1l-prc1-antagonism, does not have a symmetric PRC2 counterpart. PRC1 and PRC2 behave differently here, and the honest statement is: DOT1L antagonises PRC1; PRC2 is not antagonised by DOT1L but by the active marks that accompany transcription, and it is recruited by the fusion's own anchor to a different set of genes.
| mark | H3K27me3 |
|---|---|
| histone | H3 |
| residue | K27, N-terminal tail |
| writer complex | protein-prc2-complex |
| reader in writer complex | gene-eed (aromatic cage; allosterically stimulates EZH2 - the read-write loop) |
| competing state same residue | H3K27ac - rises when H3K27me3 falls under PRC2 inhibition (PMID 25395428, PMID 40825195) |
| status at fusion targets | not aberrant in bulk four-mark profiling of MLL-AF9 LSCs (PMID 21741597); a SUBSET of KMT2A oncofusion sites are bivalent H3K4me3/H3K27me3 by single-cell CUT&Tag (PMID 34663924) |
| pharmacodynamic use | H3K27me3 change from baseline was the pharmacodynamic endpoint of the MAK683 first-in-human phase 1/2 (PMID 39793445, NCT02900651) |
| contrast with h3k79me2 | one writer vs two; no established eraser vs KDM6A/KDM6B; no feedback reader vs EED inside the writer complex; the fusion's activating output vs its repressive output |
writers
[ "gene-ezh2", "gene-ezh1" ]
erasers
[ "KDM6A (gene-kdm6a exists in this graph)", "KDM6B" ]
allosteric inhibitors of deposition
[ "H3K36me2/3 (PMID 33211010)", "H3K4me3" ]
functionally important loci in this disease
[ "CDKN2A", "CDKN2B", "GATA2", "EGR1 (mouse MLL-AF9)", "CDKN1C", "TP53INP1 (human B-ALL)", "HDAC7 (t(4;11) infant B-ALL, named but not ChIP-confirmed in the abstract)" ]
- Depleting H3K27me3 raises H3K27ac at the same residue and increases transcription at the affected loci; in B cells with high AID activity or DNA-repair deficiency this increases chromosomal translocation frequency. In a translocation-defined B-lineage leukaemia in infants, a therapy whose pharmacodynamic signature IS the hazard signature needs an explicit argument, not an inference from a monotherapy-safe headline.
- A bulk H3K27me3 pharmacodynamic endpoint sums the EZH1- and EZH2-deposited pools, which have opposite therapeutic meanings in this disease - the same defect this graph already records for total H3K79 methylation in the pinometostat trial.
- Loss of H3K27me3 de-represses growth restraints, which is the therapeutic mechanism, but the mark's normal job in a pro-B cell includes controlling Igh rearrangement (PMID 12496962). Removing it in the compartment that has to rebuild a B-cell repertoire is a developmental as well as an oncogenic intervention.
- H3K27me3 is not unique to the leukaemia at any locus. Unlike the menin-KMT2A interface there is no fusion-specific version of this mark, so every intervention against it is a therapeutic-window argument and never a selectivity argument.
- Is H3K27me3 present at HDAC7 in t(4;11) infant blasts? PMID 40722046 names EZH2 as a silencer of HDAC7 in the index disease but the abstract does not report an H3K27me3 measurement at the locus, so the mark's presence at the single most index-relevant gene in this beat is inferred rather than observed.
- Is the bivalency at a subset of fusion sites real at the nucleosome, or two cell populations each carrying one mark? Single-cell CUT&Tag shows heterogeneity, which is evidence for the second reading as much as the first, and only a sequential or co-immunoprecipitation readout separates them.
- Does menin inhibition change H3K27me3 in patients? If menin drives EZH2 expression (PMID 23349306), it should, and no revumenib or ziftomenib trial has measured it - a free pharmacodynamic that would test the menin-EZH2 axis in humans.
- What sets the boundary of an H3K27me3 domain in a leukaemia blast? The read-write loop predicts unbounded spreading and the H3K36/H3K4 inhibition predicts boundaries at transcribed genes, but in a cell whose fusion creates broad ectopic active domains the boundary rules should be measurably different - and nobody has looked.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
EZH2 and EZH1 control gene expression by methylating histone H3 at lysine 27; the dual inhibitor UNC1999 suppresses global H3K27 trimethylation and dimethylation, de-represses Polycomb targets including Cdkn2a, and gene de-repression correlates with a decrease in H3K27me3 and a concurrent gain in H3K27 acetylation, with effects preferentially at distal regulatory elements such as enhancers.
Selective inhibition of EZH2 and EZH1 enzymatic activity by a small molecule suppresses MLL-rearranged leukemia. · Xu B, On DM, Ma A, Parton T, Konze KD, Pattenden SG, et al. · Blood 2015What this evidence cannot show
Abstract read from the eutils record; full text not read. The H3K27me3-to-H3K27ac exchange is a correlation across de-repressed genes, not a demonstrated causal sequence. Mouse and human MLL-rearranged leukaemia cells with a chemical probe; the in vivo arm is murine MLL-AF9. 'Global' suppression is a bulk measurement and does not describe locus-level redistribution.
Evidence that argues against this (1)
- The same methyl-to-acetyl exchange at K27 is the mechanism by which EZH2 inhibition increases chromosomal translocation frequency in B cells with high AID activity or DNA-repair deficiency - so the pharmacodynamic signature that reads as target engagement is also the signature of a genomic-instability hazard in the exact lineage this disease occupies. Tao J et al. 2025
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
PRC2 is stimulated upon recognition of its own catalytic product and is more active on dinucleosomes than on H3 tails or single nucleosomes; cryo-EM shows a single PRC2 positioning the H3 tails of an activating and a substrate nucleosome to interact with EED and with the EZH2 SET domain respectively - properties that probably facilitate local H3K27me2/3 spreading and heterochromatin formation.
Cryo-EM structures of PRC2 simultaneously engaged with two functionally distinct nucleosomes. · Poepsel S, Kasinath V, Nogales E · Nat Struct Mol Biol 2018What this evidence cannot show
Abstract read from the eutils record; full text not read. Reconstituted complexes on designed dinucleosome substrates in vitro. The spreading model is the authors' inference from geometry; no cellular measurement of spreading is in this work. No residue numbers or spans are asserted.
Evidence that argues against this (1)
- A self-propagating mark whose enzyme reads its own product is expected to spread indefinitely, and it does not - which means the boundaries are set by something else. The established boundary mechanism is inhibition by the active marks H3K36me2/3 and H3K4me3, so the read-write loop is only half the model and the half that determines where the mark is NOT is the other half. Finogenova K et al. 2020
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
At transcriptionally active genes di- and tri-methylated H3K36 inhibit PRC2; unmodified H3K36 occupies a critical position in the EZH2-DNA interface, mutation of H3K36 to arginine or alanine inhibits H3K27 methylation by PRC2 on nucleosomes in vitro, and Drosophila H3K36A and H3K36R mutants show reduced H3K27me3 and defective Polycomb repression of HOX genes.
Structural basis for PRC2 decoding of active histone methylation marks H3K36me2/3. · Finogenova K, Bonnet J, Poepsel S, Schäfer IB, Finkl K, Schmid K, et al. · Elife 2020What this evidence cannot show
Abstract read from the eutils record; full text not read. Cryo-EM plus in vitro methyltransferase assays on reconstituted dinucleosomes; the in vivo arm is Drosophila. No leukaemia, no KMT2A fusion, no mammalian haematopoietic system. The H3K36A/R mutants remove the residue and so cannot separate loss of the inhibitory methyl mark from loss of the unmodified lysine's structural contribution, which the same work shows is also required.
Evidence that argues against this (1)
- The HOX-repression defect is in Drosophila, where the Polycomb response elements are sequence-defined and mammalian PRC2 targeting works differently. The mammalian equivalent is contested enough that a separate study had to define three distinct classes of response element for human PRC2 and MLL1/2-Trithorax complexes, so a fly HOX phenotype does not license a statement about HOXA in human leukaemia. Du J et al. 2018
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Side-by-side epigenetic profiling of H3K79me2, H3K4me3, H3K27me3 and H3K36me3 in haematopoietic progenitors and MLL-AF9 leukaemia stem cells found abnormal profiles only for H3K79me2 at fusion target loci; the H3K27me3 profile was not aberrant.
MLL-rearranged leukemia is dependent on aberrant H3K79 methylation by DOT1L. · Bernt KM, Zhu N, Sinha AU, Vempati S, Faber J, Krivtsov AV, et al. · Cancer Cell 2011What this evidence cannot show
Abstract read from the eutils record; full text not read. The conclusion 'only H3K79me2' is bounded by the four marks profiled and by 2011-era ChIP without spike-in calibration - an uncalibrated ChIP normalises away global changes by construction, so a uniform shift in H3K27me3 would not have been detected. Mouse MLL-AF9, myeloid. This entry cites it for the negative result at fusion target loci, which is compatible with PRC2 mattering elsewhere in the genome.
Evidence that argues against this (1)
- A subset of KMT2A oncofusion binding sites DO carry bivalent H3K4me3/H3K27me3 signatures with cell-to-cell heterogeneity, found by automated CUT&Tag across diverse oncofusions in cell lines and patient samples. So 'H3K27me3 is not aberrant at fusion targets' is a bulk statement that a single-cell-resolved method contradicts for a subpopulation of sites. Janssens DH et al. 2021
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Automated CUT&Tag profiling of diverse KMT2A oncofusion binding sites in cell lines and patient samples shows that a subset are marked by bivalent H3K4me3/H3K27me3 signatures, that single-cell CUT&Tag reveals cell-to-cell heterogeneity at these sites suggestive of lineage plasticity, and that aberrant enrichment of H3K4me3 in gene bodies is sensitive to menin inhibitors.
Automated CUT&Tag profiling of chromatin heterogeneity in mixed-lineage leukemia. · Janssens DH, Meers MP, Wu SJ, Babaeva E, Meshinchi S, Sarthy JF, et al. · Nat Genet 2021What this evidence cannot show
Abstract read from the eutils record; full text not read. 'A subset' is unquantified in the abstract. Bivalency calls depend on antibody specificity for H3K4me3 and H3K27me3 and on how co-occurrence is defined; a bulk bivalent signal can arise from two populations of cells each carrying one mark, which is precisely why the single-cell arm matters and why a claim of bivalency AT A NUCLEOSOME is not established by this design.
Evidence that argues against this (1)
- The better-powered bulk comparison in the same disease found abnormal profiles ONLY for H3K79me2 among four marks including H3K27me3, and calibrated ICeChIP in MV4;11 finds H3K4me3 and H3K27me3 moving in OPPOSITE directions on DOT1L inhibition rather than being co-gained. Bivalency in this disease is a minority-site phenomenon at best, and this graph's mechanism-bivalent-primed-chromatin-fusion-targets already records that tension. Bernt KM et al. 2011
- E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Inactivation of Eed in mouse MLL-AF9 leukaemia de-represses Cdkn2a, Cdkn2b, Gata2 and Egr1, and Cdkn2a inactivation only partially rescues growth of Eed-inactivated cells - so H3K27me3 in this disease is maintained over growth restraints rather than over fusion targets.
Inactivation of Eed impedes MLL-AF9-mediated leukemogenesis through Cdkn2a-dependent and Cdkn2a-independent mechanisms in a murine model. · Danis E, Yamauchi T, Echanique K, Haladyna J, Kalkur R, Riedel S, et al. · Exp Hematol 2015What this evidence cannot show
Abstract read from the eutils record; full text not read. De-repression is measured as RNA change after Eed inactivation; the abstract does not report H3K27me3 ChIP at these loci, so the inference that the mark was present and was lost there is one step beyond the stated result. Mouse MLL-AF9, myeloid.
Evidence that argues against this (1)
- The genes de-repressed by dual EZH1/EZH2 inhibition in human B-ALL are different - CDKN1C and TP53INP1, confirmed as direct targets by CRISPR knockout - so the specific loci carrying the functionally important H3K27me3 are lineage- or species-dependent, and no single biomarker locus transfers. Ito J et al. 2023
- E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In B-cell ALL, dual EZH1/EZH2 inhibition up-regulates CDKN1C and TP53INP1, confirmed by CRISPR/Cas9 knockout to be direct targets of EZH1/2 responsible for the antitumour effect, and the inhibitor works in a B-ALL patient-derived xenograft harbouring MLL-AF4.
Dual inhibition of EZH1/2 induces cell cycle arrest of B cell acute lymphoblastic leukemia cells through upregulation of CDKN1C and TP53INP1. · Ito J, Yamagata K, Shinohara H, Shima Y, Katsumoto T, Aikawa Y, et al. · Int J Hematol 2023What this evidence cannot show
Abstract read from the eutils record; full text not read. The CDKN1C/TP53INP1 mechanism is established in B-ALL lines generally rather than specifically in the MLL-AF4 PDX, so mechanism and index-fusion efficacy are two separate claims from one paper. No H3K27me3 ChIP at these loci is described in the abstract.
Evidence that argues against this (1)
- CDKN1C is imprinted and its regulation involves DNA methylation at the KCNQ1OT1 locus as well as H3K27me3, so a CDKN1C de-repression readout is not a clean PRC2 pharmacodynamic - and compensatory DNA methylation is a reported resistance route to dual EZH1/2 inhibition. Yamagishi M et al. 2026
- E8established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
H3K27me3 change from baseline has been used as the pharmacodynamic endpoint in a first-in-human phase 1/2 study of an EED-directed PRC2 inhibitor (MAK683) in 139 patients with advanced malignancies.
A first-in-human phase 1/2 dose-escalation study of MAK683 (EED inhibitor) in patients with advanced malignancies. · Ribrag V, Iglesias L, De Braud F, Ma B, Yokota T, Zander T, et al. · Eur J Cancer 2025What this evidence cannot show
Abstract read from the eutils record; full text not read. The abstract states that pharmacodynamic profiles were measured by H3K27me3 change from baseline but does not state the tissue, the assay or whether the endpoint tracked response. No leukaemia cohort was enrolled. Adults with advanced solid tumours and lymphoma.
Evidence that argues against this (1)
- A bulk H3K27me3 endpoint measures the sum of the EZH1- and EZH2-deposited pools, which in this disease have different therapeutic meanings - the same defect this graph already records for 'total H3K79 methylation' as a pinometostat endpoint. A PRC2 pharmacodynamic that cannot separate the two paralogues' contributions will report success for depleting either. Neff T et al. 2012
- E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
EZH2 inhibition in B cells depletes H3K27me3 while concurrently enhancing H3K27 acetylation, selectively increasing transcriptional activity and facilitating chromosomal translocation formation in the presence of high AID activity or DNA ligase IV deficiency; monotherapy did not significantly change AID expression or translocation frequency, but EZH2 inhibition combined with PI3K-delta inhibition significantly increased translocation frequency in both mouse CH12F3 and human MEC-1 B cells.
Epigenetic changes by EZH2 inhibition increase translocations in B cells with high AID activity or DNA repair deficiency. · Tao J, Alessandri L, Gasparetto A, Zhao L, Zhang X, Alt FW, et al. · Blood 2025What this evidence cannot show
Abstract read from the eutils record; full text not read. The systems are a mouse B-cell lymphoma line (CH12F3) and a human CLL line (MEC-1), not KMT2A-rearranged leukaemia or infant pro-B blasts, and the sensitising conditions are high AID activity or Ligase4 deficiency - neither of which is an established feature of infant t(4;11) ALL. The authors explicitly conclude that EZH2 inhibitors are genetically safe as MONOTHERAPY, and this entry must not be read as saying otherwise.
Evidence that argues against this (1)
- The authors' own headline is reassurance, not alarm: monotherapy did not change translocation frequency in either system. The hazard is specific to combination with an agent that raises AID, or to a repair-deficient background, so treating this as a general contraindication would misrepresent the paper. It belongs in bite_back as a conditional risk requiring a specific argument, which is how risk-ezh2-inhibition-b-lineage-translocations frames it. Tao J et al. 2025
THE MARK ITSELF: the only pharmacological-menin-inhibitor measurement of it is negative and non-leukaemic.
H3K27me3 falls at bivalent MHC-I promoters after MEN1 knockout, without a change in total H3K27me3
THE MARK BOTH SIDES MOVE, IN OPPOSITE DIRECTIONS: the combination experiment would also supply the H3K27me3 measurement under menin inhibition that no study in any system has reported.
mutually exclusive on the same H3 tail
the only H3.3 reagent used in a KMT2A-dependency experiment is a K27M oncohistone that blocks H3K27me3 spreading
dosage largely uncoupled (PMID 36332031) despite the classical hierarchical model
writes H3K27me1/2/3
sustains locus-specific H3K27me3 in Ezh2-null leukaemia
aromatic cage reads H3K27me3, closing the PRC2 read-write loop
required for PRC2 assembly and H3K27me3 establishment on target chromatin (mESC)
- ← causesPRC2 - the complex, not the subunit, is the unit every experiment in this disease actually resolves
the complex writes H3K27me1/2/3 and is stimulated by its own product
- analogous to →H3K79me2 - the di-methyl state the leukaemia actually runs on, and the one menin turns out to govern only half of
the repressive counterpart: two writers vs one, erasers vs none, a feedback reader vs none
- part of →Plain-English: chromatin and epigenetics — the sticky notes on the DNA, and which ones matter here
layman explainer for the mark layer
the repressive half of the bivalent H3K4me3/H3K27me3 signature reported at a subset of oncofusion sites
- ← related toGiving a translocation-promoting epigenetic drug in a translocation-defined B-lineage leukaemia, to infants
the hazard is a property of the mark's removal, not of any one molecule
BAH module (2513-2633) reads H3K27me3 via a trimethyl-lysine cage
fusion-bound loci are DEVOID of H3K27me3; the proposed mechanism is an acetylation-plus-H3K36me3 state refractory to PRC2 invasion (MOUSE ChIP-seq)
in multipotent human HSCs/HPCs the bivalent genes that LOSE H3K27me3 and activate are the ones already carrying H4K20me1 with H2A.Z, H3K4me1, H3K9me1 and Pol II at the stem-cell stage
the fusion protein strongly decreases H3K27me2/3 in bulk chromatin when expressed ectopically; the expected spatial signature is loss of broad domains with CpG-island retention
down-regulates H3K27me2/3; the spatial signature is loss of broad domains with CpG-island retention
the mark both fusions are proposed to deplete, by opposite mechanisms
H3K27me3 immunohistochemistry, scored by the PFA ependymoma convention, with the redistribution caveat fixed in advance
second silencing readout, first binning of this arm
first binning of GSE260742's H3K27me3 NTvKO arm
the second repressive mark, largest measured effect
85 bulk libraries plus one of the three single-cell CUT&Tag libraries
H3K27me3 is LOST at MLL-fusion targets under pinometostat, and downregulation of polycomb components accounts for part of the low-dose proliferation defect. Class-level, not MEIS1-locus-resolved.
CDKN2A is a TARGET of H3K27me3: EZH2 enriched at the p16 TSS with H3K27 methylation in murine MLL/ENL cells and absent in E2A/HLF cells
the only H3K27me3 measurement near menin, and it is genetic rather than pharmacological