Menin-bound genes are the RNA-processing and DNA-repair machinery, H2AK119ub-only genes are neuronal and developmental, and 'menin target' is therefore a much weaker label than this graph treats it as
also called Zhou group function scan · menin occupancy specificity · Polycomb developmental signature · GO enrichment menin H2AK119ub
A published dataset sorts about nine thousand genes into menin-bound, silenced, or both, and nobody had asked what kinds of genes fall in each bin. The menin bin is full of the cell's basic housekeeping machinery, which means calling a gene a menin target says much less than it sounds like.
The same thing, for a scientist
Unbiased GO:BP enrichment for each of the three groups of Zhou's locus set, **against a custom background of Zhou's own 9,292 called genes** (so every statement is conditional on being called), g:SCS-corrected. | group | n | top processes | best p | |---|---:|---|---| | **Menin solo** | 3,423 | RNA processing, mRNA splicing, **DNA repair**, chromosome organization, ribonucleoprotein biogenesis | **1.6e-32** | | **Bivalent** | 1,733 | generic *positive regulation of...*, intracellular signal transduction | 1.0e-08 | | **H2AK119ub solo** | 4,136 | **cell-cell signalling, chemical synaptic transmission, trans-synaptic signalling**, morphogenesis, cell adhesion | **1.2e-17** | **The third row is a positive control nobody planted.** Polycomb's canonical job is silencing lineage-inappropriate developmental genes, and in a myeloid cell line the H2AK119ub-only set comes back neuronal and developmental. The analysis reproduced a textbook result it was not testing for, which is the strongest available evidence that the partition means something. **The first row is the finding and it is a caution.**
Why 'menin target' is a weaker label than this graph treats it as
The menin-only group's enrichment is not for HOX genes, leukaemia programmes or lineage transcription factors. It is for RNA processing (348/555, p=1.6e-32), splicing, DNA repair (226/388) and chromosome organization - the constitutive nuclear machinery of a dividing cell.
Three consequences worth stating plainly:
- Menin occupancy is not, by itself, an oncogenic signature in this dataset. If menin is over the housekeeping apparatus at large, then 'gene X is menin-bound' carries little specificity, and entries in this graph that lean on menin occupancy as evidence of relevance are leaning on something broad.
- It does not weaken the menin-interface hypothesis; it relocates the selectivity. That hypothesis is about the KMT2A-menin contact and the CXXC/post-CXXC address, not about menin's own distribution. Broad menin occupancy is consistent with, and arguably required by, an argument in which the FUSION supplies the specificity and menin supplies the anchor. But it means menin occupancy cannot be cited as evidence for selectivity.
- The DNA-repair enrichment is a live thread. This graph proposed today that menin inhibition might sensitise to PARP inhibitors, on evidence that MEN1 loss impairs homologous recombination in pancreatic neuroendocrine tumour. Independently, menin-bound genes here are enriched for DNA repair (226 of 388 repair genes in the called set). Two unrelated observations pointing the same way is worth more than either alone - though occupancy is still not regulation.
What it says about today's class I result
Earlier today this project reported, at moderate confidence and domain level, that the MHC class I locus sits in the bivalent group. The obvious worry was that the bivalent group might be broadly immune-flavoured, making the class I observation a trivial instance of a general pattern.
It is not. The bivalent group's enrichment is generic regulatory language - positive regulation of biological process, intracellular signal transduction - with no immune, interferon, antigen-processing or MHC term reaching significance at all. The best bivalent p-value (1.0e-08) is also four to twenty-four orders of magnitude weaker than the other two groups', meaning the bivalent set is the least functionally coherent of the three.
So the class I locus is not swimming with the tide. That makes the earlier observation more interesting and not less - and it also means it cannot be explained away as immune genes being generally bivalent, because they are not.
The internal validity check, and why it matters more than the p-values
The H2AK119ub-only group came back neuronal. Chemical synaptic transmission (214/334), trans-synaptic signalling, regulation of membrane potential, cell adhesion, morphogenesis - in a myeloid leukaemia line.
This is the single most reproduced observation in Polycomb biology: PRC1 and PRC2 hold down the developmental programmes a cell is not using, and in a blood cell those are neural and morphogenetic. Nothing in this analysis was aimed at finding it. The background was Zhou's own called set, the source was all of GO:BP, and the term list was not curated.
An instrument that recovers a textbook result it was not looking for has earned some trust - which in this project is worth saying, because the running count of instruments that found their own construction before they found a defect reached ten this month.
group sizes — the numbers behind the chart
{
"Menin solo": 3423,
"Menin/H2AK119ub shared": 1733,
"H2AK119ub solo": 4136
}enrichment
{
"tool": "g:Profiler gost API",
"source": "GO:BP",
"background": "custom, Zhou's 9,292 called genes",
"correction": "g:SCS",
"threshold": 0.05
}top terms
{
"Menin solo": [
[
"RNA processing",
1.576176671984091e-32,
348,
555
],
[
"nucleobase-containing compound metabolic process",
9.958016979181751e-19,
1335,
3033
],
[
"nucleic acid metabolic process",
9.473624465343582e-18,
1211,
2726
],
[
"ribonucleoprotein complex biogenesis",
1.1494771671997677e-17,
170,
256
],
[
"macromolecule metabolic process",
2.62636706036941e-17,
1984,
4781
],
[
"primary metabolic process",
1.4990667734407958e-16,
2145,
5235
]
],
"Menin/H2AK119ub shared": [
[
"positive regulation of biological process",
1.0422198593539562e-8,
755,
3338
],
[
"positive regulation of cellular process",
0.000001356462562237806,
708,
3161
],
[
"positive regulation of metabolic process",
0.0000030921121429021784,
463,
1941
],
[
"intracellular signal transduction",
0.00000837350295601369,
414,
1714
],
[
"positive regulation of macromolecule metabolic process",
0.00004741126471312041,
426,
1793
],
[
"regulation of macromolecule metabolic process",
0.0001244312005256258,
686,
3114
]
],
"H2AK119ub solo": [
[
"cell-cell signaling",
1.146797004409896e-17,
375,
588
],
[
"multicellular organismal process",
2.0861274991982104e-16,
1716,
3370
],
[
"system process",
1.5410099641871426e-14,
482,
811
],
[
"anatomical structure morphogenesis",
1.152200910585184e-10,
752,
1387
],
[
"system development",
1.3406148208870025e-9,
1057,
2038
],
[
"chemical synaptic transmission",
9.579617992524753e-9,
214,
334
]
]
}n significant terms — the numbers behind the chart
{
"Menin solo": 60,
"Menin/H2AK119ub shared": 42,
"H2AK119ub solo": 60
}- This weakens a label this graph uses often. 'Menin-bound' or 'a menin target' appears throughout these entries as though it were a specific finding; in this dataset menin sits over the RNA-processing and DNA-repair machinery at large. Anyone citing menin occupancy as evidence that a gene matters should say what the background rate is.
- It does NOT refute the menin-interface hypothesis and should not be quoted as doing so. That hypothesis concerns the KMT2A-menin contact and the CXXC address, not menin's own selectivity - and a broadly distributed anchor is compatible with a narrowly distributed fusion. What it forbids is using menin occupancy as evidence FOR selectivity.
- Does the DNA-repair enrichment among menin-bound genes survive as a regulatory claim? It converges with the MEN1-loss homologous-recombination result from pancreatic neuroendocrine tumour, and the two arrived independently.
- Where is the FUSION, as opposed to menin? Zhou's table maps menin; the specificity argument needs the oncoprotein's own distribution, which this cannot supply.
- Is the menin-over-housekeeping pattern a property of this cell line, of AML, or of menin? One dataset, one line.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
In OCI-AML2, an adult KMT2A::AFDN AML line, genes called as menin-bound without H2AK119ub in Zhou's locus set are enriched against that set's own background for constitutive nuclear machinery - RNA processing (348 of 555 annotated genes, g:SCS p=1.6e-32), mRNA splicing, DNA repair (226 of 388) and chromosome organization - rather than for lineage, HOX or leukaemia programmes.
zhou-group-function-scan: GO:BP enrichment of each menin/H2AK119ub group against Zhou's own called set as custom background, g:Profiler g:SCS corrected · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Run 2026-09-20. **Background is Zhou's CALLED set**, so this describes what distinguishes the groups from each other, not from the genome. **GO annotation is itself biased toward well-studied processes**, so enrichment partly measures research attention - RNA processing and DNA repair are among the most heavily annotated processes in GO. **Occupancy is not regulation**: menin sitting at a promoter says nothing about whether it controls that gene, and the same dataset's authors report that reducing menin occupancy genome-wide changed comparatively little expression. System is adult myeloid KMT2A::AFDN, not infant B-lineage KMT2A::AFF1 (rule 73).
Evidence that argues against this (1)
- Menin occupancy being broad does not make it uninformative, because the leukaemia-relevant question is where the FUSION and menin are co-resident, not where menin is. Zhou's table cannot separate the two - it maps menin, not KMT2A::AFDN - so a broad menin distribution is fully compatible with a narrow, specific fusion distribution on top of it, and this analysis cannot distinguish those. Neville D et al. 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Genes called as H2AK119ub-marked without menin in the same dataset are enriched for cell-cell signalling (375 of 588, p=1.2e-17), chemical synaptic transmission (214 of 334), trans-synaptic signalling, anatomical structure morphogenesis and cell adhesion - the canonical Polycomb signature of lineage-inappropriate developmental programmes, recovered in a myeloid line without being sought.
Epigenetic regulation of noncanonical menin targets modulates menin inhibitor response in acute myeloid leukemia · Zhou X, Zhang L, Aryal S, et al. · Blood 2024What this evidence cannot show
Same background and correction as E1, same annotation-bias and occupancy-is-not-regulation caveats. **This is offered as an internal validity check rather than as a discovery** - that PRC1 silences neural and developmental genes in non-neural cells is long established, and the value here is that an uncurated, unaimed scan reproduced it from this particular table.
Evidence that argues against this (1)
- Recovering an expected result validates the partition without validating any inference drawn from it. A table can be correct about which genes carry a mark and still be a poor guide to what the mark does under a drug - which is exactly the disagreement about whether menin-inhibitor-associated H2AK119ub is pre-existing or newly deposited. Neville D et al. 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
The bivalent menin/H2AK119ub group shows no significant enrichment for any immune, interferon, antigen-processing or MHC biological process, its strongest enrichment being generic regulatory terms at p=1.0e-08, which is four or more orders of magnitude weaker than the other two groups and makes it the least functionally coherent of the three.
zhou-group-function-scan results: bivalent group GO:BP terms and corrected p-values · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
**Moderate because this is an argument from a negative**: no immune term reached g:SCS significance, which at this set size is weak evidence that none is enriched. The bivalent set is the smallest of the three (1,733) and generic parent terms dominating its list is itself a signature of low coherence rather than a measurement of it. This bounds the earlier class I observation rather than supporting it.
Evidence that argues against this (1)
- The functional literature reports that menin inhibition raises MHC class I and class II and drives an interferon-stimulated gene programme. If those genes were broadly bivalent one would expect an immune signature here and there is none - which is evidence either that the drug response is not read off baseline bivalency, or that the relevant genes are the MHC ones this scan's background treats as ordinary members. Sparbier CE et al. 2023
SHARPENS ITS DNA-REPAIR OBSERVATION from a broad GO term to the specific HR machinery, and names the expression confound that both share.
GIVES THAT RESULT ITS BACKGROUND: the bivalent group carries no immune enrichment at all, so the class I locus is not one instance of a general pattern.
SECOND, INDEPENDENT POINTER TO DNA REPAIR: menin-bound genes are enriched for DNA repair (226/388), arrived at separately from the MEN1-loss homologous-recombination result.
WHAT MENIN OCCUPANCY ACTUALLY MARKS OUT in this dataset: constitutive nuclear machinery, not a lineage or leukaemia programme.
USES THE SAME TABLE, unbiased and genome-wide this time, with Ensembl ids taken directly so no symbol resolution is involved.
CONFIRMS ITS CENTRAL CAUTION from the other direction: two transactivators are not a pathway, and the bivalent group is the least functionally coherent of the three.
THE PARTITION IS MEANINGFUL: the H2AK119ub-only group returns the canonical Polycomb developmental signature, unsought.