Erasing H2Bub: three deubiquitinases share two adaptors, and removing the enzyme the field names lowers the mark instead of raising it
also called SAGA DUB module · H2Bub deubiquitination · USP22/USP27X/USP51 · ATXN7L3-ENY2 DUB adaptors
The ubiquitin tag on H2B is taken off again by a small module inside the SAGA complex, and the level on any gene is the balance between putting it on and taking it off. Confusingly, deleting the enzyme most people name — USP22 — makes the mark go down, not up, because two other enzymes compete for the same helper proteins.
The same thing, for a scientist
H2Bub1 is removed by the SAGA deubiquitinase module: the catalytic USP22 with the adaptors ATXN7/ATXN7L3/ENY2. The field's simple model — USP22 erases H2Bub, so losing USP22 should raise it — is **wrong as stated**. Ablating USP22 *reduces* global H2Bub, whereas depleting the non-enzymatic ATXN7L3 or ENY2 *increases* it; the resolution is that two further DUBs, **USP27X and USP51**, deubiquitinate H2B independently of SAGA and compete with USP22 for ATXN7L3 and ENY2 [E1]. Mouse genetics agrees: global H2Bub is only slightly affected in *Usp22*-null embryos but strongly increased in *Atxn7l3*-null embryos and cells, and *Atxn7l3*-null embryos are delayed by E7.5 while *Usp22*-null embryos survive to E14.5 [E2]. The same study found that the large H2Bub increase in *Atxn7l3*-null cells produced only modest changes in Pol II occupancy and **no change in Pol II elongation** — a direct challenge to the idea that H2Bub level controls elongation globally [E2]. Chemically, potent macrocyclic peptide inhibitors of the human SAGA DUB module exist, with high selectivity against a panel of 44 other DUBs and demonstrated cell entry and H2Bub elevation [E3]. This is the **only** chemical matter anywhere in this beat, and it goes in the direction of *raising* H2Bub.
What the graph's existing USP22 entry does not say
The graph already holds gene-usp22, and it is a good entry: it establishes USP22 as a KMT2A-rearranged dependency acting on MEIS1 protein stability, deliberately classified as a dependency rather than an interactor, with a DepMap selectivity of -0.211 (rank 226 of 17,728). That entry mentions H2B deubiquitination only once, in bite_back, as a reason a USP22 inhibitor would not be MEIS1-specific.
This entry is the missing half and does not contradict it. Both can be true: USP22 can stabilise MEIS1 protein and participate in H2Bub turnover. What this entry adds is that the H2Bub arm of USP22's biology is not clean — the enzyme's loss lowers rather than raises the mark 1 — and that anyone reasoning from 'USP22 inhibitor raises H2Bub at fusion target genes' is reasoning from a model that the primary literature has already complicated.
The practical consequence for gene-usp22's central hypothesis — that a USP22 inhibitor would survive a MEN1 gatekeeper mutation because it acts on MEIS1 protein rather than transcription — is that the chromatin arm gives a second, independent reason the compound might work, and also a second, independent reason it might have effects nobody predicted.
The adaptor, not the enzyme, is the rheostat
| Perturbation | Effect on global H2Bub | Source |
|---|---|---|
| USP22 ablation | decreased | 1 |
| ATXN7L3 depletion | increased | 1 |
| ENY2 depletion | increased | 1 |
| Usp22-null mouse embryo | only slightly affected | 2 |
| Atxn7l3-null mouse embryo / MEFs / mESCs | strongly increased | 2 |
The consistent reading across both studies is that ATXN7L3 (with ENY2) is the node that sets H2Bub levels, because it licenses three different catalytic subunits — USP22, USP27X and USP51 — and is limiting for all of them 1. Removing USP22 alone frees ATXN7L3/ENY2 for USP27X and USP51, which is why the mark can fall rather than rise.
This makes ATXN7L3 the more interesting target of the two, and its DepMap profile is at least as good as USP22's: difference -0.188, rank 290 of 17,728, on a pan-essentiality of -0.336 4. See gene-atxn7l3.
The one chemical series in this entire beat, and it pushes the wrong way
Morgan et al. screened a massive combinatorial library of cyclic peptides and found potent inhibitors of USP22; the top hit was highly specific for USP22 against a panel of 44 other human DUBs, entered cells, and increased H2B monoubiquitination 3.
Everything else in this beat is genetic. This is chemical matter, it is selective, and it is cell-active. But note what it does: it raises H2Bub. If the KMT2A-fusion logic is 'H2Bub licenses DOT1L, so lower it', this compound class does the opposite. The only coherent way to want it in this disease is if the gene-usp22 MEIS1-stability axis dominates, or if the gene-rnf40 differentiation logic is right and more H2Bub pushes a blocked cell forward. Those are two different, testable theories of the same compound, and a KMT2A-rearranged cell line plus these macrocycles would separate them in a fortnight.
What changed the calculus while this repair was being written. Broechtel et al. report that in KMT2A-rearranged leukaemia, genetic or chemical USP22 inhibition polyubiquitinates MEIS1 and sends it for proteasomal degradation, downregulating its target genes (PMID 42418681, the same paper the graph's gene-usp22 is built on; counter-evidence on 3). On the MEIS1 axis the two perturbations agree; on H2Bub they disagree. That is good news for the compound and bad news for the experiment: whatever a USP22 macrocycle does to a KMT2A-rearranged cell will be dominated by MEIS1 loss, so the growth and differentiation read-outs cannot be attributed to the direction H2Bub moved.
| counterintuitive result | USP22 ablation lowers global H2Bub; ATXN7L3 or ENY2 depletion raises it |
|---|---|
| rheostat | ATXN7L3/ENY2, because they are limiting for all three catalytic subunits |
| chemical matter | macrocyclic cyclic-peptide USP22 inhibitors, selective against 44 DUBs, cell-active, raise H2Bub |
| direction of chemical matter | raises H2Bub - the opposite of what a DOT1L-licensing model would want |
| global elongation caveat | a large H2Bub increase in Atxn7l3-null cells produced no change in Pol II elongation |
| confound in kmt2a | USP22 inhibition, genetic or chemical, degrades MEIS1 in KMT2A-rearranged cells (PMID 42418681), so phenotype cannot be attributed to H2Bub direction |
erasers
[ "USP22 (SAGA DUB module)", "USP27X (SAGA-independent)", "USP51 (SAGA-independent)" ]
shared adaptors
[ "ATXN7L3", "ENY2", "ATXN7" ]
graph cross references
[ "gene-usp22", "gene-atxn7l3" ]
- The available chemical matter raises H2Bub, so if the licensing model is right it would be expected to **support** the fusion programme rather than oppose it. A USP22 inhibitor given in this disease could in principle amplify H3K79 methylation at fusion targets.
- Three redundant catalytic DUBs sharing two adaptors means single-enzyme inhibition has a built-in bypass, and the one published attempt to remove the canonical enzyme genetically moved the mark the wrong way.
- ATXN7L3 loss is embryonic-lethal earlier than USP22 loss in mice, so the more selective-looking node is also the more developmentally essential one.
- The graph's existing gene-usp22 rationale is MEIS1 protein stability; the chromatin arm described here is a second, independent mechanism for the same compound, and a compound with two mechanisms in opposite directions is hard to dose rationally.
- What do the USP22 macrocycles do to a KMT2A-rearranged cell? The compounds exist, are selective and are cell-active, and the experiment has not been reported in this disease.
- Does H2Bub at HOXA9 and MEIS1 rise, fall or stay flat when USP22 is inhibited chemically rather than deleted genetically? The genetic and chemical perturbations are predicted to move the mark in opposite directions, which is an unusually clean discriminating experiment.
- Is ATXN7L3's DepMap selectivity (rank 290) real, or an artefact of the small AML-weighted panel? ENY2, its obligate partner, shows nothing, which is the sort of internal inconsistency that usually means noise.
- Do USP27X and USP51 rise in KMT2A-rearranged cells under USP22 inhibition? If the adaptors are limiting, they should not - but nobody has measured it.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Ablation of USP22, the enzymatic centre of the SAGA deubiquitination module, leads to a reduction rather than an increase in global H2Bub levels, whereas depletion of the non-enzymatic components ATXN7L3 or ENY2 increases H2Bub; two further H2Bub deubiquitinases, USP27X and USP51, function independently of SAGA and compete with USP22 for ATXN7L3 and ENY2 for activity; like USP22, USP51 and USP27X are required for normal cell proliferation and their depletion suppresses tumour growth.
ATXN7L3 and ENY2 Coordinate Activity of Multiple H2B Deubiquitinases Important for Cellular Proliferation and Tumor Growth. · Atanassov BS, Mohan RD, Lan X, Kuang X, Lu Y, Lin K, McIvor E, Li W, Zhang Y, Florens L, Byrum SD, Mackintosh SG, Calhoun-Davis T, Koutelou E, Wang L, et al. · Mol Cell 2016What this evidence cannot show
Abstract read in full; full text not read. Cell lines, not leukaemia and not KMT2A-rearranged. 'Global' H2Bub is a bulk western measurement and says nothing about locus-level behaviour, which is where the KMT2A-relevant biology is. The competition model is the authors' interpretation of the depletion phenotypes.
Evidence that argues against this (1)
- The chemical arm contradicts the genetic arm, and this entry holds both. Morgan et al.'s macrocyclic USP22 inhibitors - selective against a 44-DUB panel and cell-permeable - RAISE cellular H2B monoubiquitination, which is what the textbook model predicts and the opposite of what USP22 ablation does here. Only one of the two can be a clean report of USP22's contribution to H2Bub level. The adaptor-competition model reconciles them only if removing the USP22 protein and blocking its catalysis have opposite net effects on ATXN7L3/ENY2 availability, which neither paper tests. Morgan M et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
USP22 and ATXN7L3 are each essential for mouse embryonic development but with different requirements: Atxn7l3-/- embryos show developmental delay by E7.5 while Usp22-/- embryos are normal at that stage and die at E14.5; global H2Bub levels are only slightly affected in Usp22-null embryos but strongly increased in Atxn7l3-null embryos and derived cell lines; and the strong H2Bub increase in Atxn7l3-/- mESCs and MEFs does not correlate with more than modest changes in Pol II occupancy and shows no change in Pol II elongation, indicating H2Bub deubiquitylation does not directly regulate global Pol II transcription elongation.
Histone H2Bub1 deubiquitylation is essential for mouse development, but does not regulate global RNA polymerase II transcription. · Wang F, El-Saafin F, Ye T, Stierle M, Negroni L, Durik M, Fischer V, Devys D, Vincent SD, Tora L · Cell Death Differ 2021What this evidence cannot show
Abstract read in full; full text not read. Mouse embryos, mESCs and MEFs - no leukaemia. The negative elongation result is a **global** statement; the authors themselves note the ATXN7L3-related DUB activity regulates only a subset of genes, and a subset is exactly what would matter for a fusion programme. This is therefore a caution against a global model, not evidence that H2Bub is unimportant at specific loci.
Evidence that argues against this (1)
- Different embryonic lethality timing for Atxn7l3 and Usp22 nulls establishes non-equivalence in development, not redundancy or its absence in a blast. The same paper reports that H2Bub1 deubiquitylation is essential for development WITHOUT regulating global Pol II transcription, which is awkward for any entry chaining DUB activity to transcriptional output. Cell Death & Differentiation 2021
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A screen of a massive combinatorial library of cyclic peptides identified potent inhibitors of USP22, the catalytic subunit of the human SAGA DUB module; the top hit was highly specific for USP22 compared with a panel of 44 other human DUBs, and cells treated with the peptide had increased levels of H2B monoubiquitination, demonstrating cell entry and inhibition of H2B deubiquitination - to the authors' knowledge the first reported inhibitors of USP22 or the SAGA DUB module.
Potent macrocycle inhibitors of the human SAGA deubiquitinating module. · Morgan M, Ikenoue T, Suga H, Wolberger C · Cell Chem Biol 2022What this evidence cannot show
Abstract read in full; full text not read. Cyclic peptides with demonstrated cell entry are tool compounds, not drugs; no pharmacokinetics, no in vivo data and no leukaemia data are described in the abstract. Selectivity was assessed against 44 DUBs, which is a good panel but not the whole family, and no proteome-wide selectivity is claimed. No potency value is quoted here because the abstract gives none.
Evidence that argues against this (2)
- In this disease the compounds' dominant readout is not H2Bub at all. Broechtel et al. found, from a genome-wide CRISPR screen in a KMT2A-rearranged B-ALL line, that USP22 regulates MEIS1 protein stability, and that GENETIC OR CHEMICAL inhibition of USP22 leads to polyubiquitination and proteasomal degradation of MEIS1 with downregulation of its target genes. On that axis the two perturbations AGREE, while on H2Bub they disagree - so a USP22 compound in a KMT2A-rearranged cell is acting through a non-histone substrate whose effect is large, and any H2Bub-level reasoning about it is confounded. Brochtel S et al. 2026
- Chemically raising H2Bub is not established to change transcription. Wang et al. showed that the strong global H2Bub increase in Atxn7l3-null mESCs and MEFs produced no more than modest changes in Pol II occupancy and NO change in Pol II elongation. If raising the mark is transcriptionally near-silent, the direction argument built on these compounds may have no readout to move. Wang F et al. 2021
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In the project's genome-wide DepMap KMT2A-selectivity scan, ATXN7L3 has a KMT2A-rearranged mean gene effect of -0.446 against -0.258 in other leukaemia lines, a difference of -0.188 and a rank of 290 of 17,728, on a pan-essentiality of -0.336; USP22 is -0.211 (rank 226) on +0.089; ENY2 is +0.015 (rank 11,080) on -0.313; USP51 -0.057 (rank 2,992) and USP27X +0.028 (rank 12,583).
DepMap CRISPR (Chronos) genome-wide KMT2A-selectivity scan of 17,728 protein-coding genes, 11 Cellosaurus-annotated KMT2A-rearranged leukaemia lines vs 43 other leukaemia lines · Broad Institute DepMap; Cellosaurus; Open Targets Platform; analysis run by this project (analyses/depmap-kmt2a-selective-scan-full) · Database and derived analysis 2026What this evidence cannot show
Eleven KMT2A-rearranged lines, AML-weighted, none infant-derived. ENY2's lack of signal despite being a shared adaptor with ATXN7L3 is unexplained and is a reason to treat the ATXN7L3 rank cautiously. Single-gene knockouts cannot detect a requirement that is shared across three redundant catalytic subunits.
Evidence that argues against this (1)
- SCAN CONFOUND: this number comes from the project's genome-wide KMT2A-selectivity scan, whose index arm is MYELOID-ENRICHED (8 AML lines against 5 B-ALL). A gene whose requirement differs by lineage is mis-scored in BOTH directions by that design — CEBPA ranks 25 genome-wide with no dependency at all in KMT2A-rearranged B-ALL, and conversely a gene required in one lineage and not the other can average to apparent non-selectivity. See `ana-lineage-confound-myeloid-tf-dependencies`. ATXN7L3 at -0.446 against -0.2 is the kind of modest margin most likely to be created or erased by lineage composition, so this is among the least safe numbers in the beat to quote without the split. Oncofusion knowledge graph (internal analysis) 2026
Counter-evidence (Morgan 2022, PMID 34936860 against Atanassov 2016, PMID 27132940): selective cell-active USP22 macrocycles RAISE H2Bub while genetic USP22 ablation LOWERS it, so only one of the two can be a clean report of USP22's contribution to the mark, and the adaptor-competition model reconciles them only on an assumption neither paper tests
Counter-evidence (Broechtel 2026, PMID 42418681): in KMT2A-rearranged leukaemia, genetic or chemical USP22 inhibition polyubiquitinates MEIS1 and degrades it, impairing growth - so in this disease the compound acts dominantly through a non-histone substrate and H2Bub-level reasoning about it is confounded
Counter-evidence (Wang 2021, PMID 33731875): the strong global H2Bub increase in Atxn7l3-null mouse ES cells and MEFs changed Pol II occupancy only modestly and Pol II elongation not at all, so raising the mark may be transcriptionally near-silent and leave the direction argument with no readout
- related to →What this beat looked at and did not write: rejected candidates, with counts for both halves of every search
Counter-evidence (Atanassov 2016, PMID 27132940): USP27X and USP51 are redundant with USP22 for the same two adaptors, so their flat single-gene DepMap scores are exactly what redundancy predicts and cannot justify rejecting them as candidates - the same argument applies to the three disputed E2 candidates
erases the mark; steady-state level is a writer/eraser balance
ATXN7L3/ENY2 are the rheostat rather than the catalytic subunit
shared adaptor of the SAGA DUB module