Oncofusion

start / graph / concept-kmt2-family-attribution-problem

Explainerstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q44-kmt2-family

When a family shares a complex, the result is family-selective, not member-selective - the protein version of this graph's inhibitor-class rule

also called KMT2 family attribution · which KMT2 is this result about · COMPASS family attribution problem

This graph already knows that a drug aimed at one protein usually hits its whole family. The same is true of the proteins themselves: six human enzymes share the same helper module, and a result credited to one of them is often a result about several. Eighty-five entries in this graph name a KMT2 family member other than KMT2A, and until this cycle not one of those genes had an entry.

The same thing, for a scientist

This graph carries a standing generalisation for chemistry - *when a family shares a conserved catalytic pocket, the inhibitor class is family-selective, not member-selective* - and applies it to WDR5 WIN-site agents, DOT1L, CK2 and HDAC. **The same generalisation applies to the proteins and this graph has not been applying it.** The six human SET1/MLL enzymes share the WRAD module in its entirety; the KMT2A/KMT2B pair additionally shares menin and LEDGF; the KMT2C/KMT2D pair shares PTIP, PA1 and NCOA6 [E1]. A subunit-level or module-level result therefore attributes to a branch, not to a gene. Three hard cases make this concrete in this disease: **(a)** endogenous MLL1 is dispensable for MLL-rearranged leukaemia while Mll2/KMT2B deletion alone impairs survival, so the graph's wild-type-KMT2A premises are attributions that a paralogue experiment overturned [E2]; **(b)** fusions built on KMT2B **cannot transform**, and the determinant is the CXXC/post-CXXC address rather than any shared subunit, so family-sharing does not imply functional equivalence where it matters most [E3]; **(c)** KMT2E carries a structurally dead SET domain, so any 'KMT2 methyltransferase' claim sourced from an MLL5 paper is an indirect effect wearing the family's name [E4]. An audit of this repository found **85 entries naming KMT2B, KMT2C, KMT2D, KMT2E, MLL2, MLL3, MLL4 or MLL5**, headed by `resistance-mll34-utx-tumour-suppressor-failure` at 55 mentions, while **none of those genes existed as an entry before this beat** [E5].

Go deeper
The rule, stated for proteins

A result obtained with a shared module attributes to every protein that shares the module. Concretely, for this family:

shared thingshared byso a result about it attributes to
WRAD (WDR5, RBBP5, ASH2L, DPY30)KMT2A, KMT2B, KMT2C, KMT2D, SETD1A, SETD1Ball six
the WDR5 WIN pocketKMT2A, KMT2B, SETD1A, SETD1Bfour
menin and LEDGFKMT2A, KMT2Btwo
PTIP, PA1, NCOA6KMT2C, KMT2Dtwo
the CXXC/post-CXXC addressKMT2A and KMT2B - and they behave differentlyneither, without a cell-based test

The last row is the important one and it is why this is a rule about evidence rather than a rule about sequence. The CXXC/post-CXXC modules of KMT2A and KMT2B bind DNA almost indistinguishably in vitro and take the two proteins to largely non-overlapping genes inside a cell 3. Sharing a module predicts a shared in-vitro result and does not predict a shared cellular one.

This is the exact mirror of the chemistry rule the graph already holds. A WIN-site inhibitor displaces WDR5 from four enzymes at once; a WRAD co-IP reports on six. Both are family measurements presented in member vocabulary.

The three claims in this graph that a family attribution changes

1. 'The fusion needs the wild-type KMT2A allele.' Several entries build a therapeutic-window argument on pre-bound wild-type MLL1 complexes. 2 found endogenous MLL1 dispensable in independent conditional models and Mll2/KMT2B the larger dependency, with the two acting on distinct pathways rather than redundantly. This graph has already recorded the correction in target-kmt2a-cxxc-dna-interface, subtraction-kmt2a-set-domain and others - but it recorded it as a counter-note pointing at a gene that did not exist. The entry it points at now exists (gene-kmt2b).

2. 'A menin inhibitor disrupts the menin-KMT2A interface.' It disrupts menin-KMT2A and menin-KMT2B, in biochemical and structural assays, by the admission of the study that otherwise argues the therapeutic contribution is fusion-directed 6. Every menin entry in this graph is written as if the drug had one wild-type target. In a normal tissue with no fusion, it has two.

3. 'MLL3/4-UTX is the activating arm.' resistance-mll34-utx-tumour-suppressor-failure names 55 instances of KMT2C, KMT2D, MLL3 and MLL4. Those are two genes with different lengths (4,911 and 5,537 aa), different cytobands (7q36.1 and 12q13.12), different leukaemia mechanisms (haploinsufficient 7q tumour suppressor versus mTOR/ribosome-biogenesis) and different published countermeasures (JQ1 or CDK4/6 versus CX-5461). Writing them as one unit is the family attribution in its purest form, and the field writes them that way too - including the paper that found them in distinct ASC-2 complexes rather than one 1.

The operational test, four questions

Before this graph records a KMT2-family result, ask:

  1. What was measured - a gene, a subunit, or a module? A WRAD subunit pulldown is a six-enzyme measurement. A menin pulldown is a two-enzyme measurement. A KMT2A antibody ChIP is a one-gene measurement only if the epitope is unique and the antibody is validated, which for KMT2B the field says it is not 7.
  2. Which name does the paper use, and can it be resolved? MLL2 and MLL4 each denote two genes; see concept-mll2-and-mll4-each-name-two-genes. Resolve by accession, by length, by phenotype or by the paper's own grouping - and if none of those works, record it as unresolved.
  3. Is the enzyme even catalytic? KMT2E is not 4. A methylation claim from an MLL5 paper is an indirect effect.
  4. Would the result survive if the paralogue were the actor? This is the question 2 answers no to for the whole wild-type-KMT2A recruitment literature, and it is the cheapest of the four to ask.
What the audit found

Every node file in this repository outside this beat's folder was scanned for the tokens KMT2B, KMT2C, KMT2D, KMT2E, MLL2, MLL3, MLL4 and MLL5 5. 85 entries match. The ten heaviest:

mentionsentry
55resistance-mll34-utx-tumour-suppressor-failure
41gene-kdm6a
21subtraction-kmt2a-lost-from-protein-not-from-cell
20mechanism-set-loss-neomorph-not-haploinsufficiency
16mark-h3k4me3
16interface-wdr5-win-kmt2a-arg3765
15ana-unstated-system-kdm4c-prmt1-resolved
13resistance-adaptive-nongenetic-menin
12subtraction-kmt2a-phd2-e3-ligase
12mechanism-h2bub-compass-h3k4-arm

None of the genes named had an entry. This is the fourth consecutive beat in this project to find the same shape - HDAC in 81 files with 0 entries, EZH2/PRC2 in 63 with 0, TOP2A/TOP2B in 90 with 0, and now this. The pattern is not that the graph misses proteins; it is that the graph writes mechanisms and does not write the proteins named inside them, and the resulting entries carry attributions no reader can check because the target of the attribution does not exist.

Facts and figures
kindmethodological rule
parent rulewhen a family shares a conserved catalytic pocket, the inhibitor class is family-selective, not member-selective
this rulewhen a family shares a complex, a subunit- or module-level result is branch-selective, not member-selective
shared modules
{
  "WRAD": [
    "KMT2A",
    "KMT2B",
    "KMT2C",
    "KMT2D",
    "SETD1A",
    "SETD1B"
  ],
  "WDR5 WIN pocket": [
    "KMT2A",
    "KMT2B",
    "SETD1A",
    "SETD1B"
  ],
  "menin + LEDGF": [
    "KMT2A",
    "KMT2B"
  ],
  "PTIP + PA1 + NCOA6": [
    "KMT2C",
    "KMT2D"
  ]
}
audit 2026 09 20
{
  "entries_naming_a_non_kmt2a_family_member": 85,
  "gene_entries_that_existed_before_this_beat": 0,
  "top_entries": [
    [
      "resistance-mll34-utx-tumour-suppressor-failure",
      55
    ],
    [
      "gene-kdm6a",
      41
    ],
    [
      "subtraction-kmt2a-lost-from-protein-not-from-cell",
      21
    ],
    [
      "mechanism-set-loss-neomorph-not-haploinsufficiency",
      20
    ],
    [
      "mark-h3k4me3",
      16
    ],
    [
      "interface-wdr5-win-kmt2a-arg3765",
      16
    ],
    [
      "ana-unstated-system-kdm4c-prmt1-resolved",
      15
    ],
    [
      "resistance-adaptive-nongenetic-menin",
      13
    ],
    [
      "subtraction-kmt2a-phd2-e3-ligase",
      12
    ],
    [
      "mechanism-h2bub-compass-h3k4-arm",
      12
    ]
  ]
}
four questions
[
  "was a gene, a subunit or a module measured?",
  "which name does the paper use, and can it be resolved to an accession?",
  "is the enzyme catalytic at all?",
  "would the result survive if the paralogue were the actor?"
]
How this could bite us back
  • Applied honestly, this rule weakens several of this graph's own arguments, including the wild-type-allele therapeutic window and the KMT2A-selectivity of a CXXC-directed ligand. That is the point of writing it down rather than an objection to it.
  • It also weakens the counter-arguments. A family attribution cuts both ways: a negative result obtained with a shared module is equally unable to exonerate a single member, so 'WDR5 was never blotted on a fusion' and 'menin-KMT2B was never measured' are the same kind of hole.
  • A rule that says most results are family-level invites paralysis. The corrective is that the rule names a measurement to make - endogenous tagging, a paralogue-null background, a selective degrader - not a reason to stop citing the literature.
Still open
  1. How many of this graph's 85 family-naming entries carry a claim that would change under this rule? The scan counts mentions, not claims. Reading the ten heaviest would take an afternoon and is the obvious next step.
  2. Is there any experiment in this literature that is genuinely member-specific for KMT2A? Endogenous tagging and degron knock-in at the KMT2A locus are the only designs that qualify, and the graph's own `exp-endogenous-tag-and-degron-knock-in-kmt2a-aff1` is the entry that should be read against this rule.
  3. What supplies the cellular difference between the KMT2A and KMT2B CXXC addresses, given that they bind DNA the same way in vitro? Whatever it is, it is the real member-specific determinant in this family and it is unidentified.
Every source behind this entry (7)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Quantitative label-free AP-MS of all six human SET1/MLL complexes showed that WRAD is shared across the family, that menin and PSIP1 are specific to MLL1/2 and absent from SET1A/B and MLL3/4, and that PA1 and PTIP are specific to MLL3/4 - with the MLL3 and MLL4 enzymes reported elsewhere to sit in distinct ASC-2-containing complexes rather than one common complex.

    Quantitative dissection and stoichiometry determination of the human SET1/MLL histone methyltransferase complexes. · van Nuland R, Smits AH, Pallaki P, Jansen PW, Vermeulen M, Timmers HT · Mol Cell Biol 2013
    PMID 23508102DOI
    What this evidence cannot show

    Full text read via the PMC HTML route (PMC3647974) on 2026-09-20 after Europe PMC fullTextXML returned HTTP 500 and efetch db=pmc returned no body. HeLa, GFP-tagged overexpressed baits, isoform-fragment accessions for MLL1, MLL3 and MLL4. The distinct-ASC-2-complexes finding is from the 2006 specificity paper, not from this one.

    Evidence that argues against this (1)
    • A single cell type cannot establish that subunit assignments are fixed. If a signalling state or a lineage redistributed menin or PTIP between branches, this measurement would not see it, and the whole branch-attribution scheme would be softer than stated. Lee S et al. 2006
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Using rigorous independent conditional animal models, endogenous MLL1 was dispensable for MLL-rearranged leukaemia; deletion of the closest paralogue Mll2 alone had a significant impact on survival of MLL-AF9-transformed cells, additional Mll1 loss further reduced viability, and the MLL1/MLL2 collaboration operated through regulation of distinct pathways rather than redundancy.

    MLL2, Not MLL1, Plays a Major Role in Sustaining MLL-Rearranged Acute Myeloid Leukemia. · Chen Y, Anastassiadis K, Kranz A, et al. · Cancer Cell 2017
    PMID 28609655DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20. Mouse, MLL-AF9, myeloid; conditional deletion permits adaptation. The 'MLL2' is KMT2B, resolved from the paper's own description of it as MLL1's closest paralogue. This is already cited in ten entries in this graph, all of them as a counter-note; it was re-fetched for this entry rather than copied from any of them.

    Evidence that argues against this (1)
    • The same senior author's follow-up shows that the therapeutic inference does not carry: Mll1;Mll2 co-deletion has different consequences from Men1 deletion, and Mll1;Mll2-deficient cells are MORE sensitive to menin inhibitors. A paralogue being the larger genetic dependency does not make it the route through which a drug acts. Chen Y et al. 2019
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Fusion proteins based on Mll2 (KMT2B), the closest relative of MLL, are incapable of transforming haematopoietic cells; domain-swap experiments identified the small CxxC zinc-binding region and an adjacent post-CxxC basic stretch as the essential determinants of the difference, and although the combined CxxC and post-CxxC domains of MLL and Mll2 have almost indistinguishable DNA-binding properties in vitro, within the cellular environment they guide the two proteins to a largely non-overlapping target gene repertoire.

    Alterations of the CxxC domain preclude oncogenic activation of mixed-lineage leukemia 2. · Bach C, Mueller D, Buhl S, Garcia-Cuellar MP, Slany RK · Oncogene 2009
    PMID 19060922DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20; no PMCID in the fetched record so routes 1-4 were unavailable. Engineered constructs in murine haematopoietic cells - a negative transformation result in one assay with one construct set, which is the weakest form of negative. The target-repertoire evidence is nuclear localisation, reporter assays and homeobox gene levels, not genome-wide occupancy.

    Evidence that argues against this (1)
    • A negative transformation result with engineered constructs cannot exclude that a naturally occurring KMT2B rearrangement with a different partner or breakpoint would transform. No KMT2B rearrangement has been reported in leukaemia, but nobody has looked for one with the diligence applied to KMT2A. Larson JK et al. 2023
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The 2.1 Angstrom crystal structure of the unliganded SET domain of human MLL5/KMT2E shows that missing key residues and an unusually large SET-I loop preclude interaction with cofactor and substrate, and MLL5 was shown to be devoid of in vitro methyltransferase activity on full-length histones and H3 peptides - so a family member carries the family's catalytic name without the family's catalytic activity.

    PMID 27812132DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20. An isolated-domain structure and an in vitro assay on histone substrates only; activity requiring a partner subunit or on a non-histone substrate is not excluded.

    Evidence that argues against this (1)
    • KMT2E nonetheless has a large leukaemia literature in which its expression is an independently validated favourable prognostic marker, so treating it as irrelevant to this disease because it cannot methylate would be the opposite error to the one this entry warns about. Damm F et al. 2011
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    A scan of every node file in this repository outside the kmt2-family folder, run 2026-09-20, found 85 entries containing at least one of the tokens KMT2B, KMT2C, KMT2D, KMT2E, MLL2, MLL3, MLL4 or MLL5, headed by resistance-mll34-utx-tumour-suppressor-failure with 55 mentions, gene-kdm6a with 41 and subtraction-kmt2a-lost-from-protein-not-from-cell with 21 - and none of the genes named had an entry in the graph before this beat.

    What this evidence cannot show

    A token scan counts strings, not claims: an entry that mentions MLL3 once in a counter-note and an entry whose whole argument is about MLL3 both match. The count excludes the seven gene entries written by this beat. It also excludes KMT2F/KMT2G (SETD1A/SETD1B), which do have entries, and it would miss a claim expressed only as 'the MLL family' or 'COMPASS'.

    Evidence that argues against this (1)
    • Mentioning a protein without writing an entry for it is not automatically an error - a graph cannot have an entry for everything it names, and the standing granularity rule is about interactors of the index fusion rather than about every gene in a citation. The argument for these seven is that the graph makes load-bearing claims about them, not merely that it names them. Chen Y et al. 2017
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Biochemical and structural studies show that compounds disrupting the menin-MLL fusion-protein interaction also inhibit menin-MLL1, menin-MLL2 and other menin-interacting proteins, even though the same study concludes the predominant antileukaemic action is direct inhibition of the fusion.

    Distinct pathways affected by menin versus MLL1/MLL2 in MLL-rearranged acute myeloid leukemia. · Chen Y, Jones KL, Anastassiadis K, et al. · Exp Hematol 2019
    PMID 30315824DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20; full text (PMC6472707) not obtained. The statement about compound polypharmacology is the authors' summary of prior biochemical and structural work rather than a measurement reported in this paper, so the primary sources for the menin-MLL2 inhibition are not identified here.

    Evidence that argues against this (1)
    • If the compounds' engagement of menin-MLL1/MLL2 contributed materially to efficacy, Mll1;Mll2-deficient cells should be less sensitive to them. They are MORE sensitive, which is the strongest available argument that the family attribution, while true biochemically, does not carry into the therapeutic mechanism. Sparbier CE et al. 2023
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Commercially available reagents to monitor KMT2B binding at chromatin are inadequate, so analyses based only on KMT2A accumulation likely underestimate the consequences of liberating KMT2A/B from active genes after Menin loss - a stated, technical reason why a two-gene complex is routinely measured as one gene.

    Targeting Menin disrupts the KMT2A/B and polycomb balance to paradoxically activate bivalent genes. · Sparbier CE, Gillespie A, Gomez J, et al. · Nat Cell Biol 2023
    PMID 36635503DOI
    What this evidence cannot show

    Full text read via Europe PMC fullTextXML (PMC7614190) on 2026-09-20. The statement concerns chromatin-grade reagents; the same study used a KMT2B antibody for immunoblot. It is one group's assessment of the commercial landscape in 2022-2023 and could be out of date.

    Evidence that argues against this (1)
    • A reagent limitation is a reason the measurement is missing, not a reason the attribution is wrong. If KMT2B contributed nothing at these loci, measuring only KMT2A would be adequate - and the paper's own genetic data (KMT2B knockout alone modestly impairs the phenotype) suggest the contribution is real but smaller than KMT2A's. van Nuland R et al. 2013
Connected to