HR genes are menin-bound at 1.76x and so is ribosome biogenesis at 1.78x: the pathway-specific reading does not survive its own control
also called menin DNA repair occupancy · HR genes menin · experiment H evidence · BRCA2 menin
RETRACTED IN PART, within the hour, by the control this entry itself named as required: the claim was that menin specifically binds the DNA-repair machinery. Every highly-expressed housekeeping gene set turns out to be menin-enriched to the same degree, so repair genes are part of the cell's busy machinery rather than singled out.
The same thing, for a scientist
**RETRACTED IN PART, within the hour, by the control this entry named as required.** | gene set | menin-solo ratio | |---|---:| | ribosome biogenesis | **1.78** | | homologous recombination | 1.76 | | oxidative phosphorylation | 1.61 | | DNA repair | 1.55 | | translation | 1.50 | | cell cycle | 1.28 | | glycolytic process | 0.49 | **HR is indistinguishable from generic housekeeping.** Every highly-expressed constitutive set is menin-enriched at 1.5-1.8x, which is exactly what *menin binds active promoters* predicts. **The pathway-specific claim is dead** and the "third independent line" for the PARP combination was never independent - it restates menin's general preference for active chromatin. **What survives.** The occupancy numbers themselves are unchanged and correct: HR genes are 75 of 116 menin-bound against 42.7 expected, and depleted among H2AK119ub-marked genes. **What does not survive is the inference** that this says anything specific about repair. It is one more instance of the pattern reported in `ana-menin-sits-on-housekeeping-and-polycomb-sits-on-neurons`. **One genuine outlier remains**: glycolysis is *depleted* among menin-bound genes at 0.49x (p=0.012), against every other constitutive set being enriched. That is unexplained and was not predicted.
The confound that has to be stated before the finding
Menin binds active promoters. DNA repair and homologous recombination genes are expressed in a proliferating leukaemia line. So an enrichment of HR genes among menin-bound genes is partly expected for that reason alone, and this test does not separate 'menin marks HR genes' from 'menin marks expressed genes, and HR genes are expressed'.
One internal observation cuts slightly against the pure-expression explanation and I do not want to oversell it. HR genes are depleted in the bivalent group (11 against 21.6, p=0.009). Bivalent genes are also menin-bound - so if occupancy tracked expression alone, and bivalent genes are partially repressed, this is exactly what expression would predict too. It is consistent with both explanations and discriminates neither.
The required next step is an expression-matched background: resample the null from genes matched on expression decile in OCI-AML2 rather than from the called set at large. That is a standard control in this project - analyses/depmap-null-variance-matched and expression-control-for-paralogue-splits are the templates - and it was not run here. Until it is, this is a lead with a named confound, not a result.
Why it still matters, and what it changes
Because it was not predicted by the thing it agrees with. The PanNET result (MEN1 knockout -> HR defect -> talazoparib sensitivity) came from a different disease, a different perturbation and a different laboratory, and it was found this morning while checking an unrelated ledger claim. This occupancy pattern was found separately, in a leukaemia dataset, by an unbiased scan. Neither was looking for the other.
What it changes about ledger experiment H:
- The rationale is now repair, with two independent supports, rather than the three-step HOXA9 chain the ledger used.
- The experiment should measure BRCA2 and RAD51 foci, not just viability - and this analysis names a concrete gene set to look at.
- It sharpens the prediction. If menin occupies HR promoters and displacing it reduces their expression, then a menin inhibitor should produce a measurable HR deficit before any synergy screen is run. That is a one-arm experiment.
What it does not change: knockout is not pocket blockade, this is an adult myeloid KMT2A::AFDN line, and the drug in question would be given to infants. All three objections stand exactly as they did this morning.
The control, and what it cost
The entry as first written named an expression-matched background as the required control and graded itself moderate for lacking one. A full expression match needs OCI-AML2 expression data that is not readily available, so a proxy discriminator was run instead: if the enrichment is really 'menin binds expressed genes', then other highly-expressed constitutive gene sets should show the same ratio.
They do. Ribosome biogenesis scores higher than HR. Translation, OXPHOS and DNA repair are all in the same band.
This is not a full expression match and does not need to be. The proxy answers the specific question that mattered - is HR special? - and the answer is no. A real expression-matched null would refine the ratios; it would not resurrect the specificity.
What this cost and what it bought. The claim stood for about forty minutes and was committed. It was retracted by a control chosen because the entry itself named the confound in its own caveats. That is the caveat discipline working as designed - the same mechanism that made the PRC1.1 cost correction cheap earlier in this project. The entry was written to be killed cheaply and it was.
The glycolysis outlier, resolved the same hour and deflationary
Glycolysis came back depleted among menin-bound genes (8 observed vs 16.2, p=0.012) while every other constitutive set was enriched, and was recorded as unexplained. Reading the members explains it, and the explanation removes the anomaly.
H2AK119ub-silenced 'glycolysis' genes (26): ENO2 (neuron-specific enolase), ENO3 (muscle), ALDOC (brain aldolase), ACTN3 (muscle), INS (insulin), HK3, FBP1, PPARA, MLXIPL, SLC2A6, UCP2, PFKFB3/4...
Menin-bound or bivalent (18): ALDOA, HK2, LDHA, PFKP, PGAM1, plus the regulators HIF1A, ARNT, EP300, NCOR1.
The split is by tissue, not by pathway. The silenced set is the collection of tissue-restricted paralogues - brain, muscle, liver, pancreas - and the menin-bound set is the glycolytic machinery a proliferating blood cell actually uses. So this is not a glycolysis result at all: it is the same canonical Polycomb signature reported in ana-menin-sits-on-housekeeping-and-polycomb-sits-on-neurons, surfacing as an apparent pathway depletion because the GO term pools isoforms from every tissue.
The transferable point: a GO term is a list, not a pathway. For any gene family with tissue-specific paralogues, an enrichment test silently asks a question about tissue identity rather than about the process named on the label. That is the same failure shape as the multi-mapping problem - a set that looks like the biology but is not.
DNA repair
{
"n_in_zhou": 364,
"Menin solo": {
"obs": 208,
"exp": 134.1,
"p": 0
},
"Menin/H2AK119ub shared": {
"obs": 62,
"exp": 67.9,
"p": 0.4572
},
"H2AK119ub solo": {
"obs": 94,
"exp": 162,
"p": 0
}
}homologous recombination
{
"n_in_zhou": 116,
"Menin solo": {
"obs": 75,
"exp": 42.7,
"p": 0
},
"Menin/H2AK119ub shared": {
"obs": 11,
"exp": 21.6,
"p": 0.0087
},
"H2AK119ub solo": {
"obs": 30,
"exp": 51.6,
"p": 0
}
}- This entry retracted its own headline within the hour. Cite it as evidence that menin occupies active promoters broadly, and for nothing about DNA repair specifically.
- The practical consequence is that ledger experiment H loses a support it briefly appeared to gain. Its rationale is again a single functional result in pancreatic neuroendocrine tumour, by genetic knockout of a tumour suppressor, in adults - which is a thin basis for proposing a PARP inhibitor alongside a menin inhibitor in infants, and should be described that way.
- The glycolysis depletion that looked anomalous is resolved and was never anomalous: the silenced members are tissue-restricted paralogues (ENO2 neuronal, ENO3 muscle, ALDOC brain, INS pancreatic) and the menin-bound members are the isoforms a blood cell uses. A GO term is a list, not a pathway - for any family with tissue-specific paralogues, an enrichment test is partly a question about tissue identity.
- RESOLVED 2026-09-20: the glycolysis depletion is an isoform artefact - the H2AK119ub-marked members are tissue-restricted paralogues (ENO2, ENO3, ALDOC, ACTN3, INS), which is the ordinary Polycomb signature rather than a metabolic finding.
- Would a true expression-matched null change any of the ratios materially, or only compress them all toward 1?
- Does a pharmacological menin inhibitor reduce BRCA2 or RAD51 foci in a KMT2A-rearranged line? Still the discriminating experiment, and now the only support for experiment H besides the PanNET result.
- E1weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.analysis
In OCI-AML2, an adult KMT2A::AFDN AML line, homologous recombination genes (GO:0000724) are over-represented among menin-bound genes in Zhou's locus set at 75 observed against 42.7 expected (1.76-fold, two-sided permutation p<1e-4 over 20,000 shuffles) and under-represented among both H2AK119ub-only genes (30 vs 51.6) and bivalent genes (11 vs 21.6, p=0.009); DNA repair genes overall show the same direction at 208 vs 134.1 menin-bound and 94 vs 162.0 H2AK119ub-only.
menin-dna-repair-occupancy: GO DNA repair and homologous recombination gene sets tested against Zhou supplemental Table 4 by permutation · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
RETRACTED IN PART 2026-09-20 by a housekeeping control run the same hour. The occupancy counts are correct; the pathway-specific interpretation is not. Ribosome biogenesis enriches at 1.78x against HR's 1.76x, and translation, OXPHOS and DNA repair occupy the same 1.5-1.8 band, so the HR enrichment is fully explained by menin's preference for active promoters. **Weak, and retained only as an instance of the general pattern.** All earlier caveats also stand: occupancy is not regulation, the table has no condition column, GO terms reflect research attention, and the system is adult myeloid KMT2A::AFDN rather than infant B-lineage KMT2A::AFF1 (rule 73).
Evidence that argues against this (2)
- The authors of this dataset report that reducing menin occupancy genome-wide produced comparatively few expression changes. If menin is present at HR promoters but not driving them, displacing it would create no repair deficit and the whole inferential chain to a PARP combination fails at the first step. That is the specific way this lead dies, and one transcript measurement would show it. Sparbier CE et al. 2023
- THE DECISIVE COUNTER, and it is this analysis's own control. Ribosome biogenesis (1.78x), translation (1.50x), oxidative phosphorylation (1.61x) and cell cycle (1.28x) are all menin-enriched in the same band as homologous recombination (1.76x). There is no HR-specific signal. The one unexplained result in the other direction is glycolysis, depleted at 0.49x. Oncofusion knowledge graph (internal analysis) 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.journal-article
An independent published result in a different disease and by a different perturbation reports that MEN1 loss of function impairs double-strand break repair through a homologous recombination defect with reduced BRCA2 expression and increased talazoparib sensitivity, which is the functional counterpart of the occupancy pattern reported in E1.
Loss of MEN1 function impairs DNA repair capability of pancreatic neuroendocrine tumors. · Lakiza O, Lutze J, Vogle A, et al. · Endocr Relat Cancer 2022What this evidence cannot show
Pancreatic neuroendocrine tumour, where MEN1 is a tumour suppressor and is genetically deleted - not KMT2A-rearranged leukaemia, where menin is an oncogenic cofactor displaced pharmacologically from one interface while the protein remains. **Knockout is not pocket blockade.** Read from the abstract; full text not obtained. The convergence with E1 is suggestive precisely because the two were found independently, but two studies agreeing across species of perturbation is not replication. **AND THE CONVERGENCE CLAIMED WITH E1 IS WITHDRAWN**: the occupancy pattern in E1 is not HR-specific, so it does not independently corroborate this functional result. This PanNET finding now stands alone as it did before - one study, another disease, a knockout rather than a pocket blocker.
Evidence that argues against this (1)
- Convergence can be an artefact of both observations tracking the same third thing. HR genes are highly expressed and menin binds expressed promoters; MEN1 loss is profoundly disruptive and could reduce BRCA2 through any number of indirect routes. Neither observation isolates a menin-to-HR mechanism, and agreeing about a conclusion is not the same as agreeing about a mechanism. Zhou X et al. 2024
THIRD INDEPENDENT LINE TO THE SAME PLACE: HR genes are 1.76x enriched among menin-bound promoters, found separately from the PanNET result that entry rests on.
SHARPENS ITS DNA-REPAIR OBSERVATION from a broad GO term to the specific HR machinery, and names the expression confound that both share.
A MENIN FUNCTION OUTSIDE THE KMT2A INTERFACE: occupancy over the homologous recombination machinery, converging with MEN1-loss HR deficiency in another disease.
THE PROPOSED SECOND AGENT: the rationale is now repair rather than HOXA9.
USES THE SAME PUBLISHED TABLE, with HGNC primary ids so no gene is dropped.
nothing in this hypothesis rests on that occupancy analysis, which its own author retracted in part after a housekeeping control
revives the menin-HR line by a protein-scaffold route after the occupancy-based version failed its own housekeeping control
the scaffold route to menin-HR, independent of the occupancy argument that entry retracted