Verdict on the WRAD contradiction: the graph is right, its citation class is wrong, and its universal is wrong
also called WRAD adjudication verdict · does the graph contradict Yokoyama 2004
The graph has carried an open question for cycles: does it claim WRAD sticks to the leukemia fusion when the founding paper says it does not? Reading every affected entry, the answer is no - the entries say what the paper says. Two different things are wrong: they cite the paper for a kind of evidence it does not contain, and they say 'every fusion' when two kinds of fusion are exceptions.
The same thing, for a scientist
`ana-kmt2a-interactome-systematic-enumeration` carries the open question *'Do WRAD subunits associate with KMT2A::AFF1 by any method published after 2004? The graph asserts they do and the founding paper says they do not.'* That framing is wrong on its own terms and this entry closes it. **Finding 1: no entry in this graph asserts that WRAD associates with a KMT2A fusion.** All four subunit entries state the opposite in their summaries, and two carry `refutes` edges to `fusion-kmt2a-aff1` labelled as corrections. The perceived contradiction was that the subunits live in folders named `interactome`, which is a filing convention, not a claim. **Finding 2: the real defect is the evidence class.** Four entries carry an evidence block, `E_yokoyama2004`, whose claim text is a faithful paraphrase of the 2004 abstract but which functions in context as a measured fusion co-IP negative. It is not one: the WRAD measurement in that paper is on a SET-domain-deleted wild-type MLL construct, and no WRAD subunit was blotted on any fusion. **Finding 3: one of those four placements is out of scope entirely** - the paper does not mention DPY30, which was not placed in the human SET1-like complexes until 2007. **Finding 4: the universal is false.** 'Lost in every KMT2A fusion' holds for the direct 5' product only; reciprocal der products and NUP98::KMT2A retain the WIN motif and SET domain. **Finding 5: post-2004, the question has still never been tested directly in either direction.**
Entry-by-entry findings, to the standard the brief requires
Reported, not edited. Each row names the entry, the field, what it should say, and the citation that settles it. Yokoyama 2004 was read in full text via the PMC HTML record (PMC480881); the publisher does not release full-text XML and fullTextXML returns HTTP 500, with efetch db=pmc returning a body-less record.
| Entry | Field | Issue | Should say | Settling citation |
|---|---|---|---|---|
gene-wdr5 | evidence[E_yokoyama2004].claim | Reads as a measured fusion co-IP negative covering WDR5 | Add to the claim or caveat: the WRAD subunits were assayed only against MLL deletion mutants; no WRAD subunit was blotted on either fusion protein in this paper | PMID 15199122, full text read, Figs 4 and 6 |
gene-rbbp5 | same | same | same | PMID 15199122, read |
gene-ash2l | same | same | same | PMID 15199122, read |
gene-dpy30 | same | Out of scope. DPY30 is not mentioned in the paper and was not an identified complex component in 2004 | Restrict the block's scope: this paper predates DPY30's assignment to the human SET1-like complexes (2007) and makes no claim about it | PMID 15199122 (read; zero occurrences of 'DPY') and PMID 17500065 |
ana-kmt2a-interactome-systematic-enumeration | open_questions[2] and the section 'The structural finding' | States 'The graph asserts they do and the founding paper says they do not', and that Yokoyama 2004 'reports directly that menin, but not HCF-1, HCF-2, ASH2L, WDR5 or RBBP5, coprecipitates with fusion MLL proteins' | No entry in the graph asserts WRAD binds the fusion. And the paper reports directly only for HCF-1; for ASH2L/WDR5/RBBP5 it infers from the delta-3820 deletion mutant | PMID 15199122, read |
ana-kmt2a-interactome-systematic-enumeration | the same table row | 'WDR5 / RBBP5 / ASH2L — not in the fusion' with no qualifier | not in the direct 5'-KMT2A fusion product; retained in reciprocal der products and in NUP98::KMT2A | PMID 34489550 (read) and PMID 31558671 |
subtraction-kmt2a-win-wrad-interface | title and body | Frames the WIN/WRAD interface as absent from the fusion without qualification | Same qualification | PMID 34489550 (read) |
gene-dpy30 | edge part_of → mechanism-chromatin-elongation-hijack, label 'shared subunit of the H3K4 machinery the fusion works through' | The only edge in the graph that reads as a DPY30-to-fusion membership claim | Relabel as a wild-type-complex membership claim, or downgrade to related_to | PMID 15199122 and PMID 30270175 (DPY30 has no mapped KMT2A contact) |
None of these is a fabricated or misattributed identifier; verify-citations.py passes on all of them, which is exactly the failure mode that checker is blind to by construction.
The one place where the graph was already ahead of its own open question
Two entries in this graph already hold post-2004 evidence that WRAD subunits can be recovered with a fusion oncoprotein - just not a KMT2A one. BioID proximity labelling of NUP98-PHF23 recovers KMT2A/MLL1, KMT2B, WDR5 and RBBP5 3, and an earlier affinity purification of five NUP98 fusions recovered WDR5 and MLL1 4. The proposed mechanism in both is that the fusion recruits an intact wild-type KMT2A/COMPASS complex, and WRAD arrives attached to it. That is the indirect route nobody has looked for around KMT2A fusions, and its existence in a sister fusion family makes the untested question materially more interesting than 'probably not'.
It also falsifies a counted absence this beat generated. A PubMed title/abstract query pairing any WRAD subunit with any proximity-labelling term returns zero records. The NUP98-PHF23 paper is a proximity-labelling paper that recovers WDR5 and RBBP5, and it is invisible to that query because its abstract names 'the KMT2/MLL H3K4 methyltransferases' at complex level 3. The absence is real as stated and useless as evidence, which is the exact trap this project's alias rule warns about, caught here on this beat's own output.
What the verdict does and does not license
Does. Correcting four evidence blocks and three universal statements. Recording that the direct question - does WRAD touch KMT2A::AFF1 - has never been asked experimentally.
Does not. Re-promoting any WRAD subunit as a KMT2A-fusion target. The structural argument is unchanged and strong: the mapped contacts are 2,400-2,700 residues past the breakpoint. The functional argument is unchanged and strong: the catalysis WRAD exists to enable is genetically dispensable for MLL-AF9 leukaemogenesis 2, and the one pharmacological programme aimed at the module acts in MLL-rearranged cells through ribosome depletion and p53 rather than through KMT2A 5. The correction is about how the graph knows what it knows, not about the answer.
| verdict | The graph does not assert what the enumeration says it asserts. Three real defects: evidence class in four blocks, scope in one of them, and a false universal in at least three entries. |
|---|---|
| paper read in full | true |
| full text route | PMC HTML; fullTextXML returns HTTP 500 and efetch db=pmc returns a record with no body element for this publisher |
| post 2004 direct tests of wrad versus a kmt2a fusion | 0 |
entries needing correction
[
{
"id": "gene-wdr5",
"field": "evidence[E_yokoyama2004]",
"defect": "evidence class",
"settling_pmid": "15199122"
},
{
"id": "gene-rbbp5",
"field": "evidence[E_yokoyama2004]",
"defect": "evidence class",
"settling_pmid": "15199122"
},
{
"id": "gene-ash2l",
"field": "evidence[E_yokoyama2004]",
"defect": "evidence class",
"settling_pmid": "15199122"
},
{
"id": "gene-dpy30",
"field": "evidence[E_yokoyama2004]",
"defect": "out of scope - paper never mentions DPY30",
"settling_pmid": "15199122, 17500065"
},
{
"id": "ana-kmt2a-interactome-systematic-enumeration",
"field": "open_questions and section 'The structural finding'",
"defect": "premise false ('the graph asserts they do') and 'reports directly' overstates the paper for WRAD",
"settling_pmid": "15199122"
},
{
"id": "ana-kmt2a-interactome-systematic-enumeration",
"field": "partner table row 'WDR5 / RBBP5 / ASH2L'",
"defect": "false universal",
"settling_pmid": "34489550, 31558671"
},
{
"id": "subtraction-kmt2a-win-wrad-interface",
"field": "title and body",
"defect": "false universal",
"settling_pmid": "34489550"
},
{
"id": "gene-dpy30",
"field": "edge part_of -> mechanism-chromatin-elongation-hijack",
"defect": "label reads as a DPY30-to-fusion membership claim",
"settling_pmid": "15199122, 30270175"
}
]- Correcting the evidence class without correcting the conclusion risks being read as reopening WRAD as a target; the structural and functional arguments against that are unchanged and strong.
- The false universal is the dangerous half: a clinician or an analyst reading 'the SET domain is lost in every KMT2A fusion' would conclude that NUP98::KMT2A behaves like the others, and it is menin-inhibitor resistant.
- A counted absence produced on this beat was falsified by a paper already in this graph, which means every counted absence here should be read as a statement about titles and abstracts, not about the literature.
- How many other 'the founding paper says X' statements in this graph are deletion-mutant inferences read as measured negatives? The HCF-1 case in the same paper is a measured negative and nothing in either citation distinguishes them.
- Does the wild-type-KMT2A-recruitment route seen for NUP98 fusions operate around KMT2A fusions? If it does, WRAD is a second-shell neighbour of every KMT2A fusion and no entry in this graph would predict it.
- Should this graph record evidence class - measured, inferred, or complex-level - as a structured field? Three of the defects found here are invisible to every existing checker because every citation is correct.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Yokoyama 2004 assayed ASH2L, WDR5 and RBBP5 only against MLL deletion mutants in transfected 293 cells and blotted its two fusion-protein immunoprecipitations for HIS, MLL-C, HCF-1, menin, MLL-N and ENL only, with no WRAD subunit assayed on any fusion and no mention of DPY30 anywhere in the text.
Leukemia proto-oncoprotein MLL forms a SET1-like histone methyltransferase complex with menin to regulate Hox gene expression. · Yokoyama A, Wang Z, Wysocka J, et al. · Mol Cell Biol 2004What this evidence cannot show
A reading of the published figures and legends. Unshown data cannot be excluded. Read in full via the PMC HTML record because the publisher does not release full-text XML for this article.
Evidence that argues against this (1)
- The inference the paper drew is independently supported by every subsequent structure: the WDR5 contact is the WIN motif at approximately 3762-3773, far past the breakpoint cluster. The 2004 conclusion is very probably correct for the direct fusion despite the evidence-class problem. Patel A et al. 2008
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The histone methyltransferase activity of MLL1, which the WRAD module exists to enable, is dispensable for haematopoiesis and for MLL-AF9-driven leukemogenesis in mouse.
The histone methyltransferase activity of MLL1 is dispensable for hematopoiesis and leukemogenesis. · Mishra BP, Zaffuto KM, Artinger EL, et al. · Cell Rep 2014What this evidence cannot show
Mouse SET-domain deletion in the MLL-AF9 model only; paralogue compensation in mouse cannot be excluded and no lymphoid or t(4;11) model was tested.
Evidence that argues against this (1)
- Pharmacological disruption of the same module with MM-401 kills MLL-rearranged cells without general toxicity to normal marrow, so the genetic and pharmacological answers disagree and the disagreement is unresolved. Cao F et al. 2014
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
BioID proximity labelling of the NUP98-PHF23 oncofusion recovers KMT2A/MLL1, KMT2B, WDR5 and RBBP5, and the paper's model is that the fusion's FG repeats partition the KMT2/MLL H3K4 methyltransferases into a chromatin-bound condensate.
The phenylalanine-and-glycine repeats of NUP98 oncofusions form condensates that selectively partition transcriptional coactivators. · Ahn JH, Guo Y, Lyons H, et al. · Mol Cell 2025What this evidence cannot show
A different fusion family; nothing in it involves KMT2A as a fusion partner. A proximity hit is a neighbour, not a binding partner: recovery is abundance-weighted and compartment-structured, and this graph's own concept entry on proximity labelling applies in full. The abstract names the methyltransferases at complex level, so the WRAD subunit identities are in the paper's data rather than its abstract.
Evidence that argues against this (1)
- An unbiased AP-MS interactome of five Strep-HA-tagged NUP98 fusions at near-physiological levels in HL-60 explicitly failed to detect the previously reported NSL and MLL1 complex interactions, so the NUP98-fusion-to-MLL1 association is contested at the level of the primary assay. Terlecki-Zaniewicz S et al. 2021
- E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Five NUP98 fusions containing no KMT2A sequence bind MOF, PHF20, OGT1, HCF1, WDR5 and MLL1 by biotin affinity purification, with binding requiring the GLFG repeats, and conditional Mll1 deletion delays NUP98-HOXA9 leukaemogenesis.
NUP98 Fusion Proteins Interact with the NSL and MLL1 Complexes to Drive Leukemogenesis. · Xu H, Valerio DG, Eisold ME, et al. · Cancer Cell 2016What this evidence cannot show
Targeted western blots after affinity purification rather than unbiased mass spectrometry, and overexpression systems. The WDR5 recovery is most simply explained by recruited MLL1 carrying its own WRAD, which would make WDR5 a second-shell rather than a direct interactor.
Evidence that argues against this (1)
- The same interactions were not detected by unbiased AP-MS of the same five fusions at near-physiological levels in a human myeloid line, which is the stronger design for a composition question. Terlecki-Zaniewicz S et al. 2021
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In MLL-rearranged cancer cells, WDR5 WIN-site inhibition acts through ribosome protein gene suppression, nucleolar stress and p53/MDM4 rather than through a KMT2A programme, and the authors state their data displace the earlier haematopoietic explanation.
Ribosome subunit attrition and activation of the p53-MDM4 axis dominate the response of MLL-rearranged cancer cells to WDR5 WIN site inhibition. · Howard GC, Wang J, Rose KL, et al. · Elife 2024What this evidence cannot show
Cell-line pharmacology; the mechanism as described requires functional p53 and so cannot operate identically in TP53-altered relapsed disease.
Evidence that argues against this (1)
- Genetic Wdr5 deletion attenuates MLL-AF9 leukemogenesis in vivo, so the requirement is real even if the mechanism is ribosomal, and a mechanism argument is not a relevance argument. Liu L et al. 2023
The verdict rests on the full-text reading of the founding paper recorded there.
- refutes →First systematic enumeration of the KMT2A fusion interactome, and the diff against this graph
Closes the open question 'The graph asserts they do and the founding paper says they do not': no entry in this graph asserts WRAD binds a KMT2A fusion, and the paper's WRAD result is an inference from a deletion mutant.
The WRAD discrimination analysis is a dependency-correlation argument and is untouched by this adjudication, which concerns physical association; the two have been read as one.
The graph already held post-2004 evidence that WRAD subunits are recovered around a fusion oncoprotein - WDR5 and RBBP5 in NUP98-fusion pulldowns and BioID - without connecting it to the KMT2A WRAD question.
Adds a route note: Yokoyama 2004 fails both XML routes (fullTextXML HTTP 500; efetch db=pmc returns a body-less record) and is fully readable through the PMC HTML page, which the current unreadable rule does not require.
Supplies the search record the verdict rests on: 24 PubMed queries, 7 Europe PMC queries, 5 full texts, 15 bounded rejections, 1 unreadable.