Two hypotheses in this graph make the same prediction on every lesion that exists, and the experiment queued to test them picks the one arm where neither speaks
also called polarity versus MBM discrimination · the two menin hypotheses are confounded
This project wrote two separate explanations for why some cancers respond to menin drugs and others do not, one from leukaemia and one from sarcoma, and never checked whether they actually disagree. They do not disagree about any real tumour, and the experiment already queued to settle the question was aimed at the single tumour where one of them makes no prediction at all.
The same thing, for a scientist
**What was compared.** Cycle 51 wrote `hyp-the-menin-interface-requirement-is-polarity-dependent-not-kmt2a-intrinsic` from KMT2A-rearranged sarcoma: menin dependence tracks rearrangement POLARITY, KMT2A 5-prime needs menin and KMT2A 3-prime does not. An earlier cycle had written `hyp-menin-dependency-tracks-available-mbm` from NUP98 fusions: it tracks availability of the menin-binding motif, KMT2A aa 6-25, in cis or in trans. The two were written from different diseases by different beats, were never linked, and nobody asked whether they differ. **They do not differ in prediction, only in reach.** Scoring all 11 lesions in this graph that carry either a menin-response measurement or a structural breakpoint record (`analyses/polarity-versus-mbm-discrimination/run.py`): the two hypotheses both speak on 5 lesions and **disagree on 0**. In every KMT2A-containing lesion in the corpus, 5-prime polarity arrives WITH the MBM and 3-prime or internal polarity arrives WITHOUT it, because every leukaemic 5-prime breakpoint lies downstream of exon 1 and every 3-prime or internal arrangement discards exon 1. **The variables are perfectly confounded, so no naturally occurring lesion can separate these two hypotheses.** **Where they differ is coverage, and it is not close.** Of the 7 lesions with a measured menin response, the polarity hypothesis classifies **2** and the MBM hypothesis classifies **7**, each correct on everything it classifies. Polarity is structurally SILENT on 5 measured lesions -- NUP98::NSD1, NUP98::KDM5A, NUP98::HOXA9, NPM1c and UBTF-TD -- because polarity is undefined when the lesion contains no KMT2A at all, and all five are menin-sensitive, which the MBM hypothesis absorbs through its trans limb [E4]. **The polarity hypothesis is therefore the MBM hypothesis restricted to KMT2A-containing lesions, not a rival to it**, and the sarcoma arms are new evidence for an existing hypothesis rather than a new hypothesis. The MBM entry mentions sarcoma zero times. **The consequence for the queued experiment, which is the actionable part.** `exp-menin-pulldown-on-yky-versus-kmt2a-aff1` uses YAP1::KMT2A::YAP1. YKY inserts only KMT2A exons 4-5/6 and carries no SET domain [E5], so it supplies essentially no MLL1 moiety, and whether its programme borrows endogenous wild-type MLL1 in trans is genuinely open. **The MBM hypothesis makes no determinate prediction for YKY** -- it is the single arm in the table scored INDETERMINATE. Meanwhile **VIM::KMT2A retains KMT2A aa 145-3969, losing MBM and IBM1 while keeping SET** [E3], which makes it the closest natural structural twin of NUP98::KMT2A, the one lesion in this graph with a MEASURED menin resistance (retains ~169-3969, loses MBM + IBM1 + IBM2, resistant to MI-2-2 at 3-12 uM) [E2]. Same KMT2A moiety, different 5-prime partner, different lineage. **The sharp arm was available and the indeterminate arm was chosen.** **One structural fact sharper than cycle 51 recorded.** VIM::KMT2A is the only lesion in this corpus that loses the MBM while RETAINING IBM2 (147-152); NUP98::KMT2A loses all three of MBM, IBM1 and IBM2. VIM::KMT2A therefore separates the menin tether from part of the LEDGF tether, which no other lesion here does, and cycle 51's log described the sarcoma arms as discarding 'the menin and LEDGF tether' as a unit.
| script | analyses/polarity-versus-mbm-discrimination/run.py |
|---|---|
| results | analyses/polarity-versus-mbm-discrimination/results.json |
| lesions scored | 11 |
| lesions with measured menin response | 7 |
| polarity coverage of measured | 2 of 7, correct on 2 |
| mbm coverage of measured | 7 of 7, correct on 7 |
| cases where both hypotheses speak | 5 |
| dissociating cases | 0 |
| why zero dissociations | Every leukaemic 5-prime KMT2A breakpoint lies downstream of exon 1, so 5-prime polarity always retains the MBM; every 3-prime or internal arrangement discards exon 1, so it always loses it. Polarity and MBM-in-cis cannot come apart in nature, which is why only an engineered construct can separate the two hypotheses. |
| sharpest natural test | fusion-vim-kmt2a |
| indeterminate arm | fusion-yap1-kmt2a-yap1 |
| the structural detail cycle 51 missed | VIM::KMT2A retains IBM2 (147-152) while losing MBM (6-25) and IBM1 (123-134). NUP98::KMT2A loses all three. VIM::KMT2A is the only lesion in this corpus that separates the menin tether from part of the LEDGF tether; cycle 51 treated the two tethers as lost together across all sarcoma arms. |
| the assumption this all rests on | Every resistant prediction in the table sets mbm_trans=False on the reasoning that a lesion supplying its own MLL1 moiety does not borrow endogenous MLL1. That is the MBM hypothesis's load-bearing step and it has never been measured directly in any of these lesions. If it is wrong -- if an MBM-less MLL1-supplying fusion still recruits endogenous MLL1 -- then NUP98::KMT2A, VIM::KMT2A and LDB1::KMT2A all become indeterminate and the MBM hypothesis's 7-of-7 record collapses to 5-of-7 measured on lesions with no KMT2A at all. |
| what this does not show | It does not show the polarity hypothesis is wrong. It shows it is unfalsifiable by natural variation and strictly narrower in reach. Both hypotheses remain correct on everything they classify. |
| shown for fusion | false |
| the sarcoma arms not in this table | SUPERSEDED THE SAME CYCLE, and the correction is recorded rather than the field rewritten. This field originally said the 16 non-recurrent partners have no published breakpoints and so could not be scored. They do: Supplemental Table 2 of PMID 32461620 names all of them with exon-level breakpoints, and it was opened later the same cycle. See ana-seven-sarcomas-retain-the-menin-binding-motif-and-three-carry-a-leukaemia-type-fusion. The split is 9 with KMT2A 3-prime and 7 with KMT2A 5-prime, and all 7 of the 5-prime arms break at KMT2A exon 3 or later, so all 7 RETAIN exon 1 and the MBM. |
| what the sarcoma breakpoints did to this analysis | They STRENGTHENED the confounding result rather than breaking it. Seven more KMT2A-containing lesions now have a polarity call and an MBM call, and all seven are 5-prime WITH the MBM -- exactly the coupling this analysis reported. The count of KMT2A-containing lesions where polarity and MBM-in-cis agree goes from 5 to 12 with still 0 disagreements. What was wrong was the narrower claim, inherited from cycle 51, that no sarcoma arm retains exon 1. Seven do, three of them with breakpoints inside the leukaemic breakpoint cluster region, and those three are now the sharpest available falsifier for the MBM hypothesis's 'if and only if': both hypotheses predict they are menin-sensitive, and they are sarcomas. |
polarity silent on
[ "fusion-nup98-nsd1", "fusion-nup98-kdm5a", "fusion-nup98-hoxa9", "lesion-npm1c", "lesion-ubtf-td" ]
- This analysis argues that a queued bench experiment was aimed at the less informative arm. It is a design critique from a coordinate derivation, and the counter-argument in E5 is strong enough that the honest recommendation is to ADD a VIM::KMT2A arm rather than to redirect the experiment away from YKY. No sarcoma KMT2A fusion protein has ever been observed by any method, so both arms are constructs and neither is a measurement of a patient tumour.
- Nothing here should be read as a claim about whether menin inhibitors help sarcoma patients. Menin is a tumour suppressor in mesenchyme, the only non-haematologic menin trial with a result was terminated for lack of efficacy signals, and this graph already carries an explicit risk entry about a menin inhibitor being prescribed off a KMT2A gene name in a solid tumour. A resistant prediction here is a mechanistic expectation, not clinical guidance.
- Does an MBM-less but MLL1-supplying fusion recruit endogenous wild-type MLL1? Every resistant prediction here assumes it does not, and nobody has measured it in NUP98::KMT2A, VIM::KMT2A or LDB1::KMT2A.
- Is the VIM arm genuinely HOXA-high? If so it dissociates HOXA output from MBM availability and is the closest thing to a natural falsifier this framework has. The claim is second-hand from a paper this project cannot read.
- Can a construct with 5-prime KMT2A polarity and a deleted MBM be built? That is the only configuration that separates the two hypotheses, and it does not occur in nature.
- Does LDB1::KMT2A, which loses the AT-hooks as well as the MBM, behave like VIM::KMT2A or like YKY?
- ANSWERED 2026-09-21: the 7 non-recurrent sarcoma partners with KMT2A 5-prime are CPT1A, DOCK10, C2CD3, ELP4, AHDC1, TMPRSS13 and RALGAPA2, and all 7 retain exon 1 and the MBM. So no naturally occurring dissociation was hiding there -- the confounding is now 12 cases with 0 disagreements. The remaining question is whether any of them is menin-INSENSITIVE, which is the direction that would break the MBM hypothesis.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Over the 11 lesions in this graph carrying a menin-response measurement or a structural breakpoint record, the polarity and MBM-availability hypotheses both speak on 5 and disagree on 0; among the 7 lesions with a measured response the polarity hypothesis classifies 2 and the MBM hypothesis classifies 7, each correct on everything it classifies, with polarity structurally silent on the 5 menin-sensitive lesions that contain no KMT2A.
What this evidence cannot show
The lesion set is what this graph holds, not the full published space of menin-sensitivity measurements, so the coverage figures are properties of this corpus. Rows are hand-scored rather than parsed, which makes them auditable in the diff but means a mis-scored polarity or MBM call would propagate silently; each row therefore carries the entry it was read from. The accuracy figures are not impressive on their own -- 7 of 7 on a set where 6 of 7 observations are 'sensitive' is close to the base rate, and the informative content is the single resistant case plus the coverage gap, not the hit rate.
Evidence that argues against this (2)
- A defender of the polarity hypothesis can reasonably say that being silent on lesions with no KMT2A is not a defect, because the hypothesis was written as a claim about KMT2A rearrangements specifically and a narrower hypothesis is not a worse one. On that reading the two are not competing at all and the coverage comparison is unfair. The reply is that the MBM hypothesis explains the KMT2A cases with the same accuracy AND the non-KMT2A cases, so nothing is lost by preferring it; but the objection is legitimate and the word 'subsumes' is doing contested work.
- The zero-dissociation result is partly an artefact of how few lesions carry a measured response. With 7 measured lesions and only 1 resistant, the corpus has very little power to separate any two hypotheses, so 'no natural lesion can distinguish them' is a statement about exon-1 structural coupling AND about a small sample. The structural argument is the load-bearing half.
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
NUP98::KMT2A deletes KMT2A exon 1 and with it the menin-binding motif, expresses reduced rather than elevated Hoxa/Meis1/Pbx3, and is resistant to the menin inhibitor MI-2-2 at 3-12 micromolar and to JQ1 at 0.05-0.5 micromolar, doses at which KMT2A-AF9 cells arrest.
Transforming activities of the NUP98-KMT2A fusion gene associated with myelodysplasia and acute myeloid leukemia · Fisher JN, et al. · Haematologica 2020What this evidence cannot show
Fusion expression is molecularly confirmed in only 2 AML patients, and the drug data are from an inducible transgenic mouse rather than from patient cells. Resistance is shown at the doses tested and is not a demonstration that no menin inhibitor can act on this lesion at any exposure. Verified live by esummary on 2026-09-21: title, journal, first author Fisher JN and DOI all match, and the publication year is 2020 rather than the 2021 carried in an earlier king log.
Evidence that argues against this (1)
- The same paper's cells are also resistant to JQ1, so the phenotype is not menin-specific and could reflect a generally less drug-dependent transcriptional programme rather than a missing menin tether in particular. A lesion that is HOXA-low may simply not depend on the HOXA machinery either drug targets.
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
VIM::KMT2A joins VIM exon 4 to KMT2A exon 2, retaining KMT2A residues 145-3969 -- AT-hooks, CXXC, PHD1-3, bromodomain, FYRN, FYRC, SET and IBM2 (147-152) -- while losing the menin-binding motif (6-25) and IBM1 (123-134).
Pan-sarcoma genomic analysis of KMT2A rearrangements reveals distinct subtypes defined by YAP1-KMT2A-YAP1 and VIM-KMT2A fusions · Massoth LR, et al. · Modern Pathology 2020What this evidence cannot show
The residue span is a coordinate derivation from the reported exon-level breakpoint against KMT2A ENST00000534358, not a directly sequenced protein boundary, and no VIM::KMT2A fusion protein has ever been observed by any method. Independently reported in two new cases by PMID 36959690 (Almohsen SS, Genes Chromosomes Cancer 2023, doi 10.1002/gcc.23138). Both PMIDs verified live by esummary on 2026-09-21.
Evidence that argues against this (1)
- The VIM arm is reported to UPREGULATE HOXA genes, which if true dissociates HOXA output from MBM availability -- the two travel together in NUP98::KMT2A, which is both MBM-less and HOXA-low. That report reaches this graph second-hand, stated by the YAP1-arm authors citing the founding series, at weak confidence, and the founding paper PMID 31136005 is unreadable by every route tried here. So the one observation that could break the framework is the one this project cannot read.
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
NUP98-rearranged AML depends on the menin-MLL1 interaction with no KMT2A lesion present: NUP98-KDM5A IC50 84 nM against 517 nM for NUP98-HOXA9, genetic dependence on Men1 and Kmt2a comparable with the essential-gene control Rpa3, and menin inhibition evicts both MLL1 and the NUP98 fusion protein itself from target genes.
The menin-MLL1 interaction is a molecular dependency in NUP98-rearranged AML · Heikamp EB, Henrich JA, Perner F, et al. · Blood 2022What this evidence cannot show
This is the direct evidence for the trans limb, and it is the limb that gives the MBM hypothesis its coverage advantage. PMID verified live by esummary on 2026-09-21. The coordinator first recalled this paper as PMID 35271699, which live lookup shows is 'Exosomes in multiple myeloma: from bench to bedside' (Menu E, Blood 2022) -- an entirely unrelated paper. The correct identifier came from the graph's own stored value and was then confirmed live, not from recall.
Evidence that argues against this (1)
- MEIS1 re-expression abrogates menin-inhibitor sensitivity in NUP98-fusion cells, and the workhorse NUP98-HOXA9 in vivo model co-expresses Meis1, so the weaker end of the partner-graded sensitivity spread may be a model artefact rather than a property of the fusion. A Meis1-free in vivo arm against a menin inhibitor has never been run.
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
YAP1::KMT2A::YAP1 inserts only KMT2A exons 4-5/6, carries no SET domain, and the tumours show no upregulation of HOXA9 or MEIS.
Recurrent YAP1 and KMT2A Gene Rearrangements in a Subset of MUC4-negative Sclerosing Epithelioid Fibrosarcoma · Kao YC, et al. · American Journal of Surgical Pathology 2020What this evidence cannot show
The HOXA9/MEIS result is n=3 by RNA sequencing. The claim that YKY therefore supplies no functional MLL1 moiety is this analysis's inference from the retained-domain list, not a measurement: nobody has asked whether the YKY insert recruits endogenous MLL1. PMID verified live by esummary on 2026-09-21.
Evidence that argues against this (1)
- Calling YKY indeterminate is itself a judgement that could be wrong in the informative direction. A ~49-residue zinc finger with no MBM and no SET is the cleanest possible case of a KMT2A rearrangement that cannot be using the menin tether, so a positive menin pulldown on YKY would be far more surprising -- and far more damaging to this graph's coordinate reasoning -- than a negative one on VIM::KMT2A. The queued experiment is the lower-probability, higher-information choice, which is a defensible design and not simply the wrong arm.
NARROWS IT RATHER THAN BREAKING IT: this hypothesis is correct on both KMT2A-containing lesions it classifies, but it is structurally silent on the 5 menin-sensitive lesions in this graph that contain no KMT2A, and it cannot be separated from the MBM hypothesis by any naturally occurring lesion because 5-prime polarity always retains exon 1 and 3-prime polarity always discards it. It is the MBM hypothesis restricted to KMT2A rearrangements.
SUPPLIES A SECOND DISEASE THIS HYPOTHESIS NEVER MENTIONED: the entry contains zero references to sarcoma, and KMT2A-rearranged sarcoma independently re-derives its cis limb in a non-haematopoietic lineage -- every sarcoma arm WITH A PUBLISHED BREAKPOINT discards KMT2A exon 1 and the YAP1 arm shows no HOXA9/MEIS output. The qualifier is load-bearing: 7 of the 16 unique non-recurrent partners in the pan-sarcoma screen have KMT2A 5-prime, and none of those has a published breakpoint, so whether they retain exon 1 is unknown rather than settled. Coverage on measured lesions is 7 of 7 against the polarity hypothesis's 2 of 7.
RECOMMENDS ADDING AN ARM: YKY inserts only KMT2A exons 4-5/6 and carries no SET, so it supplies essentially no MLL1 moiety and the MBM hypothesis is INDETERMINATE for it -- the one arm in an 11-lesion table where the framework makes no prediction. A VIM::KMT2A construct is the determinate arm. The counter-argument is that a positive pulldown on YKY would be the more surprising and more damaging result, so the recommendation is to add the VIM arm rather than to redirect the experiment.
THE OTHER OPEN ARM, AND THE ONE THAT COULD FALSIFY THE CIS LIMB: MLL/EB1 retains the whole KMT2A N-terminus including the MBM, so both hypotheses predict menin sensitivity, yet it is measurably weaker than MLL/AF10. An MBM-bearing, menin-INSENSITIVE lesion would break the 'if and only if' in the MBM statement, and no menin inhibitor, MEN1 knockdown or MBM-mutant experiment has ever been run on it.
SAME OBJECTION REACHED FROM A THIRD DIRECTION: that entry notes that UBTF-TD and NPM1c AML share the menin dependency with no fusion and no partner at all, which undermines a partner-supplied obligate interface. This table quantifies it -- 5 of the 7 measured menin-sensitive lesions here contain no KMT2A lesion whatsoever.
SHARPENS THE ABSENCE INTO A RANKED ONE: the missing measurement is not equally missing across the three sarcoma arms. VIM::KMT2A is the determinate and structurally informative arm, LDB1::KMT2A is intermediate, and YKY is the arm where the framework makes no prediction at all.
IDENTIFIES WHY THIS EXPERIMENT IS NECESSARY RATHER THAN MERELY USEFUL: polarity and MBM-in-cis are perfectly confounded across every naturally occurring lesion (5 cases where both hypotheses speak, 0 disagreements), so no patient tumour can separate the two hypotheses. An engineered MBM addback or deletion is the ONLY configuration that can.
SUPERSEDES ONE OF ITS FIELDS AND STRENGTHENS ITS CONCLUSION: that entry recorded, earlier the same cycle, that the 7 KMT2A 5-prime non-recurrent arms had no published breakpoints and were the last place a natural dissociation between polarity and MBM availability could hide. They have breakpoints, and all 7 are 5-prime WITH the MBM, so the dissociation is not there either -- 12 cases, 0 disagreements.