Half the PRC1.1 intersection was free all along: Zhou's locus set is a publisher supplementary table, and 159 of this graph's gene entries are in it
also called Zhou supplemental Table 4 · menin H2AK119ub target genes · locus set join · publisher supplement
I reported that settling a disagreement between two papers would need a large computing job, having checked the public data archive but not the journals' own supplementary files, and said so. Checking them now, one of the two published its list there - 9,292 rows, free - so half the job was already done and only the other half is still expensive.
The same thing, for a scientist
Cycle 31 established from the GEO filelists that PMID 39158067 (Zhou) and PMID 41634341 (Neville) deposited **coverage tracks, not locus calls**, and costed the intersection as a cluster job. That entry's caveat flagged the untested gap: *the papers' own supplementary tables were not checked and are the likeliest place a locus list would live.* **They are.** **Zhou supplemental Table 4** (`BLOOD_BLD-2023-023644-mmc4.xlsx`, 356 KB, via the Europe PMC `supplementaryFiles` endpoint for PMC11561541) is `chr | start | end | gene id | group`, 9,292 rows: **Menin solo 3,423 / Menin+H2AK119ub shared 1,733 / H2AK119ub solo 4,136**. The first two reproduce the counts in the paper's own body text exactly, which is the check that the file is the one the text describes. **Neville has no equivalent.** Its supplements are MAGeCK CRISPR gene summaries and statistical source data. **The asymmetry is the finding: one half of the intersection is published and free, the other half is the cluster job.**
What the table is, and what it is not
The intervals are 2 bp wide. Every one of the 9,292 rows is a TSS anchor point carrying a gene id, not a peak interval. So this is a gene-level target list with a representative coordinate - it supports a gene-level intersection and not a base-resolution one, and it is not a substitute for peak calls.
The system is two steps from this graph's index disease. The figure this table accompanies describes OCI-AML2 cells under a 6-day treatment with DMSO or 1 uM VTP50469. OCI-AML2 is CVCL_1619, an adult AML line carrying KMT2A::AFDN (MLL-AF6) - verified against Cellosaurus rather than recalled. Different lineage, different partner gene, different age group from infant B-lineage KMT2A::AFF1. Brief rule 73 says that belongs in the claim, and it is in the claim.
The table carries no condition column. The legend covers both the DMSO and the VTP50469 arms, and the table itself does not say which defined the partition. That is a real limit on reading group membership as a baseline property, and it is exactly the point Neville presses.
The join: 159 of this graph's gene entries are in the list
Of 300 gene entries, 223 resolved to an Ensembl gene id and 159 appear in Zhou's set. Symbols were resolved live against MyGene.info requiring an exact symbol echo, and the project's own data/symbol-collisions.json blocked CSNK2A1 and HDAC7 from automated resolution - the registry doing the job it was built for. Twenty-two symbols mapping to more than one Ensembl id were left unresolved rather than guessed.
Menin + H2AK119ub shared (n=45) - the bivalent, noncanonical group Zhou's whole argument is about:
AFF1, AFF4, ALKBH5, ARID1A, ASF1A, BRD3, BRD4, CCND3, CDK13, CEBPA, CHAF1A, CREBBP, CSNK2A2, DDX6, DOT1L, FLT3, FLT3, FUT4, GFI1, H2AX, HDAC2, HEXIM1, HOXA9, IKZF1, JAK1, KAT2B, KDM1A, KHDRBS1, MED26, MSL1, MSL2, MYB, MYC, PIK3R1, RUNX1, SETD1B, SIAH2, SKP1, SMARCA2, SMARCD2, SPI1, UBN1, USP7, WEE1, gene-myo1f
Menin solo (n=73):
AFF3, ANP32E, ARID2, ASH2L, ASNS, ATM, ATR, BCL11A, BCL2, BOD1L1, CCNT1, CCNT2, CD38, CDC73, CDKN2A, CHEK1, CLEC12A, CTCF, CTNNB1, CTR9, EAF2, EP300, FTO, H2AZ1, HDAC9, HIRA, HNRNPC, IFNGR1, KANSL2, KAT6A, KDM2B, KDM6A, KRAS, KRAS, LARP7, LEO1, MED23, MEF2C, MEIS1, METTL14, METTL3, MLLT10, MLLT3, NIPBL, NLRC5, NR3C1, PARP1, PRKDC, PSIP1, RAD21, RAD51, RBBP5, RBM15, RNF20, RTF1, SET, SIAH1, SIRT1, SKIC8, SMC3, SMNDC1, SS18, SUPT16H, SUPT5H, TDP2, TOP2A, TOP2B, WAC, WTAP, YTHDF2, ZC3H13, ZEB2, gene-usp8
H2AK119ub solo (n=41):
AGGF1, ATXN7L3, BCOR, BCORL1, BICRA, CBX8, CD22, CD7, CIITA, CSPG4, CXXC1, DNMT3A, ELL, ELL3, EP400, HDAC1, HDAC3, HDAC5, HDGFL2, HEXIM2, IGF2BP2, IGF2BP3, KAT2A, KDM4C, KIT, KMT2A, MEN1, MLLT1, NELFB, NPM1, PAX5, PCGF1, PRMT1, PTBP1, RBM15B, RNF2, TRRAP, USP22, WDR5, gene-gas7, gene-usp2
Sixty-four resolvable entries appear in none of the three groups, which is not evidence of anything: it is absence from Zhou's called set, under Zhou's thresholds, in Zhou's cell line.
Five things in the join worth arguing about
1. MEN1 and KMT2A are both H2AK119ub-solo. Neither the gene encoding menin nor the wild-type parent of the fusion is itself menin-bound in this dataset; both carry the Polycomb mark alone. A graph that treats menin occupancy as the organising axis should notice that the axis does not include its own two namesake loci.
2. PRC1.1 does not sit in one group. PCGF1, BCOR, BCORL1 and RNF2 are H2AK119ub-solo - the complex's own subunit genes carry the mark the complex deposits - while KDM2B is Menin-solo and SKP1 is shared. The complex this graph writes about as a unit is split three ways by its own mark.
3. MYC is in the shared group. Zhou's headline is that PRC1.1 loss reactivates noncanonical bivalent targets including MYC. That is now confirmed from the primary table rather than from the abstract, and it independently corroborates this graph's gene-myc entry.
4. HOXA9 and MEIS1 are in different groups - HOXA9 shared, MEIS1 menin-solo. The canonical target pair this field names in one breath is separated by H2AK119ub status, which is a testable claim about why they might respond differently to menin inhibition.
5. CIITA is H2AK119ub-solo, and this bears directly on the immune beats. The master regulator of MHC class II is Polycomb-marked and menin-unbound here. If the reported route from menin inhibition to MHC class II runs through CIITA, this says it cannot be direct menin loss at the CIITA locus - consistent with an indirect route, and a constraint the immune entries should carry rather than assume.
| n gene entries | 300 |
|---|---|
| n in locus set | 159 |
| source file | analyses/zhou-locus-set-join/BLOOD_BLD-2023-023644-mmc4.xlsx |
zhou group counts — the numbers behind the chart
{
"Menin solo": 3423,
"Menin/H2AK119ub shared": 1733,
"H2AK119ub solo": 4136
}cell line
{
"name": "OCI-AML2",
"cellosaurus": "CVCL_1619",
"fusion": "KMT2A::AFDN (MLL-AF6)",
"age_group": "adult"
}blocked by collisions registry
[ "CSNK2A1", "HDAC7" ]
ambiguous symbols
[ "ASF1B", "B2M", "BRD2", "CABIN1", "DAXX", "GPKOW", "HLA-A", "HLA-B", "HLA-DRA", "HLA-DRB1", "IL3RA", "KANSL1", "MEPCE", "MLLT6", "NELFE", "PRDM9", "RING1", "RYBP", "SMARCB1", "SYNCRIP", "TRIM27", "YTHDC1" ]
- This corrects an entry I wrote one cycle ago. 'Neither study deposited a list' was checked against GEO and was true of GEO; it was asserted of the studies. The caveat on that entry did bound it correctly and named the publisher supplements as the untested gap, which is the only reason the correction took twenty minutes instead of never happening. **The bounding caveat is what made the error cheap** - that is the argument for writing them.
- Do not read a group label off this join into an entry about infant B-lineage KMT2A::AFF1 without carrying the system with it. The assignments come from an adult KMT2A::AFDN myeloid cell line, which differs from the index disease in lineage, partner gene and age at once.
- And do not read 'H2AK119ub solo' as 'not a menin target'. It means not called as menin-bound at a TSS anchor in OCI-AML2 at Zhou's threshold. MEN1 itself lands in that group, which is a good reminder of how little the label claims.
- Does the coverage-track comparison at Zhou's 1,733 bivalent genes, using Neville's bigWigs, show the same direction? That is now a few hundred megabytes rather than 25 GB, because one side of the intersection no longer needs computing.
- Neville's code-availability statement offers analysis scripts on request. Would a request also yield the peak calls, making the expensive half free too?
- Which condition defined Zhou's partition - DMSO, VTP50469, or both? The table does not say, and Neville's deposition claim makes the answer load-bearing.
- CIITA is H2AK119ub-solo and menin-unbound in Zhou's OCI-AML2 data, and Fetsch et al (PMID 41144759) report the menin-inhibitor route to MHC class II running HERV -> interferon-stimulated genes -> CIITA - an INDIRECT route. The two agree, from different data. Does the join's menin-unbound call hold in a B-lineage KMT2A::AFF1 system, where MHC class II is constitutive and may have no headroom?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
In OCI-AML2, an adult KMT2A::AFDN (MLL-AF6) AML cell line treated for 6 days with DMSO or 1 uM VTP50469, Zhou et al published a gene-level menin and H2AK119ub target list as supplemental Table 4: 9,292 genes partitioned into 3,423 menin-specific, 1,733 bivalently menin/H2AK119ub-cobound and 4,136 H2AK119ub-only, with 2 bp TSS anchor coordinates rather than peak intervals; the first two counts reproduce the figures stated in the paper's body text.
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
File retrieved 2026-09-20 via the Europe PMC supplementaryFiles endpoint for PMC11561541 and parsed directly. **Established for what the table contains**, not for what it means. Three limits. (1) **The table has no condition column** and the figure legend spans both the DMSO and VTP50469 arms, so which condition defined the partition is not stated in the file. (2) **2 bp intervals are TSS anchors**, so this supports gene-level intersection only. (3) **The system is adult myeloid KMT2A::AFDN** - verified as CVCL_1619 against Cellosaurus - which differs from infant B-lineage KMT2A::AFF1 in lineage, partner gene and age.
Evidence that argues against this (1)
- Neville et al argue that menin inhibition INDUCES PRC1.1-dependent H2AK119ub deposition rather than revealing pre-existing marks. If that is right, a static three-way partition describes a state the drug itself changes, and 'H2AK119ub solo' is not a fixed property of a locus but a snapshot of one condition - which the table's missing condition column makes impossible to check from the file alone. This is the very disagreement the intersection was meant to settle, and it limits how much the join can settle by itself. Neville D et al. 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Neville et al's publisher supplementary files contain MAGeCK CRISPR screen gene summaries, plasmid GenBank records, a reagents table and statistical source data, and no peak, coordinate or target-locus list, so no published locus set exists on that side of the comparison in either GEO or the publisher's supplement.
DOT1L provides transcriptional memory through PRC1.1 antagonism. · Neville D, Ferguson DT, Heikamp EB, et al. · Nat Cell Biol 2026What this evidence cannot show
All five supplementary spreadsheets and the file legends were retrieved and opened 2026-09-20 via the Europe PMC supplementaryFiles endpoint for PMC12904788. **Strong rather than established** because this is a negative claim about a specific file set: the Peer Review File and Reporting Summary were not read line by line, and the paper's code-availability statement says analysis scripts are **available on request**, so the locus set may exist and be obtainable without any compute - by email rather than by download.
Evidence that argues against this (1)
- The same paper's data-availability statement points to GSE260456 and GSE260742 for processed as well as raw data, so 'processed' files may exist at the sample level that the series-level filelist and the publisher supplement both miss. The absence established here is of a PUBLISHED TABLE, not of a derivable locus set. Zhou X et al. 2024
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
Joining this graph's 300 gene entries onto that list resolves 223 to an Ensembl gene id and places 159 in Zhou's target set - 45 bivalent, 73 menin-specific and 41 H2AK119ub-only - with MEN1, KMT2A, PCGF1, BCOR, BCORL1, RNF2 and CIITA all in the H2AK119ub-only group, MYC and HOXA9 bivalent, and MEIS1 and KDM2B menin-specific.
zhou-locus-set-join: this graph's gene entries joined onto Zhou supplemental Table 4, with symbol resolution gated on the project's symbol-collisions registry · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Run 2026-09-20. **Every row inherits E1's system limit** - these are adult myeloid KMT2A::AFDN assignments being attached to entries mostly about infant B-lineage KMT2A::AFF1, and the group labels travel with that caveat or not at all. **Absence from the table is not absence of binding.** 77 of 300 entries did not resolve to an Ensembl id at all: 22 symbols map to multiple ids and were left unresolved, 2 are blocked by the collisions registry, and the remainder carry no symbol or Ensembl field. **The first version of this join was wrong** - it let a sibling `ensembl_verified` provenance string overwrite the clean `ensembl` id on 7 entries and reported them unresolvable; the graph was correct and the reader was not.
Evidence that argues against this (1)
- A gene-level membership label is a weak object. Zhou's partition is drawn at TSS anchors in one cell line at one threshold, and genes near the boundary between 'shared' and 'solo' will move under a different peak caller - which is the same methodological objection that made the full intersection expensive in the first place. The split of PRC1.1's own subunits across three groups should be read as suggestive, not as structure. Neville D et al. 2026
EXTENDS THAT JOIN AND BOUNDS IT: the same table, now with a null, and with the reason the MHC layer never appeared in it.
USES THE SAME PUBLISHED TABLE, with HGNC primary ids so no gene is dropped.
THE JOIN THIS REPAIRS: that entry's 22 unresolved symbols were mostly this same alt-haplotype problem.
falsifier F1 needs Zhou's 1,733 Menin+H2AK119ub shared loci, and that entry is where this graph established they are published as coordinates in BLOOD_BLD-2023-023644-mmc4.xlsx rather than deposited in GEO
USES THE SAME TABLE, unbiased and genome-wide this time, with Ensembl ids taken directly so no symbol resolution is involved.
CORRECTS ITS CENTRAL CLAIM: that entry said neither study deposited a locus set, having checked GEO only. Zhou's is a publisher supplementary table, free and 356 KB. Its own caveat named this as the untested gap.
THE DISAGREEMENT THE JOIN CANNOT SETTLE ALONE: Zhou's table has no condition column, so a static partition cannot distinguish pre-existing marks from drug-deposited ones.
MAKES HALF ITS TEST FREE: one side of the proposed intersection is now a published gene-level list; only Neville's side still needs peak calling.
CONFIRMED FROM THE PRIMARY TABLE: MYC is in the menin/H2AK119ub bivalent group, which is the group Zhou's reactivation argument is about.
RULE 73 APPLIED AT WRITING TIME: the system - adult KMT2A::AFDN myeloid OCI-AML2, verified as CVCL_1619 - is named in the first clause of the claim, not only in the caveat.
SPLIT FROM ITS OWN COMPLEX: KDM2B is menin-specific here while PCGF1, BCOR, BCORL1 and RNF2 are H2AK119ub-only.
THE COMPLEX CARRIES ITS OWN MARK: PCGF1 is in the H2AK119ub-only group.
MEN1 IS NOT MENIN-BOUND HERE: it lands in the H2AK119ub-only group, which bounds how much any group label claims.
CONSTRAINS THE MHC-II ROUTE: CIITA is H2AK119ub-marked and menin-unbound in this dataset, so a menin-inhibitor effect on class II cannot be direct menin loss at CIITA.
- related to →HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it
SEPARATED FROM MEIS1: HOXA9 is bivalent and MEIS1 is menin-specific, a testable reason the canonical pair might respond differently.