Oncofusion

start / graph / hyp-obligate-partner-interface-generalises

Hypothesismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q11-comparative

Hypothesis — the menin lesson generalises: every non-enzymatic fusion has an obligate partner interface that is more druggable than the fusion

also called obligate-partner-interface hypothesis · menin-analogy hypothesis

Menin is a protein that KMT2A fusions must grab hold of, and blocking that grip became a drug that works for two different leukaemias at once. We propose that most other pocketless fusion proteins also have one partner they cannot do without, and that this is where their drugs will come from.

The same thing, for a scientist

The menin-KMT2A precedent has two remarkable properties: the interface is required by every KMT2A fusion regardless of partner, and it is shared by a genetically unrelated disease (NPM1-mutant AML), so one drug serves both. We hypothesise this is an instance of a general rule: a fusion oncoprotein that lacks catalytic activity must recruit a wild-type host complex to execute its program, and the recruitment interface is (a) non-redundant for the fusion, (b) frequently a peptide-in-groove interaction with real pocket character, and (c) often shared across several diseases that converge on the same program. If true, the systematic search for such interfaces — rather than for fusion-intrinsic pockets — is the correct research programme for the entire non-enzymatic half of the fusion atlas. Tested against this project's own accumulated evidence (cycle 33), the existence half of the claim holds and the yield half does not: menin remains the only such interface to produce an approved drug, the closest replica (CBFbeta-SMMHC:RUNX1) has no registered trial eleven years after AI-10-49, the most chemically advanced alternative (the WDR5 WIN site) acts through ribosome depletion and p53 rather than through the interface, a validated-mechanism case (GFI1 SNAG:LSD1) has not become standard of care, and seven partner-side candidates nominated inside this project were struck on re-examination.

Go deeper
What the precedent actually shows

Menin is not required because KMT2A::AFF1 is special; it is required because the retained KMT2A N-terminus docks menin and menin is needed for chromatin occupancy. The proof that this is about program rather than genotype is NPM1-mutant AML: a CRISPR domain screen showed NPM1-mutant AML is exceptionally dependent on the menin-binding site in MLL1, and menin-MLL1 inhibition had profound antileukaemic activity in human and murine NPM1-mutant models 1. In NPM1-mutant knock-in mice the same chemistry reversed pre-leukaemic progenitor self-renewal 2. Clinically, one drug produced responses in both KMT2A-rearranged and NPM1-mutant disease in the same trial 3.

So the drug target was never the fusion. It was the host adaptor that two different lesions both conscript. That is the pattern we propose to look for elsewhere.

The candidate analogues, ranked by how much evidence already exists1 deeper

1. CBFB::MYH11 and the RUNX1 interface (strongest). CBFB-SMMHC outcompetes wild-type CBFB for RUNX1; AI-10-49 binds CBFB-SMMHC, disrupts RUNX1 binding, restores RUNX1 activity, kills primary inv(16) blasts and delays murine leukaemia 4. This is a fully validated, structurally defined, chemically drugged obligate partner interface. It is the closest existing replica of the menin story and it has stalled short of the clinic — which makes it the most informative case study the lab can examine. How far short: a ClinicalTrials.gov API query in September 2026 returns no registered interventional study of any CBFbeta-SMMHC:RUNX1 interface inhibitor, eleven years after the Science paper 7. And the mechanism is not the clean sequestration the entry assumed: inv(16) and t(8;21) cells require wild-type RUNX1 to survive, an active RUNX1 allele is preserved in patients, and RUNX1 knockdown kills them 9. The interface drug is tuning a set point, not reversing a blockade.

2. RUNX1::RUNX1T1 and the ETO NHR2 self-association interface. Oligomerisation through NHR2 is required for the differentiation block, and interfering peptides dissolve the complex and restore target-gene transcription 5. Here the obligate interface is self-association rather than partner recruitment, which is a distinct and possibly better sub-class (mutating it destroys the oncoprotein, so resistance is harder). Handled separately in hyp-oligomerisation-interfaces-as-mutation-resistant-targets.

3. NUP98 fusions and their partner-supplied reader/writer modules. Handled separately in hyp-nup98-partner-obligate-interface.

4. EWSR1::FLI1 and BAF/SMARCA4. The EWSR1 prion-like domain retargets BAF complexes to GGAA microsatellites, an activity absent from wild-type FLI1 and required for EWS-FLI1 function 6. If a discrete PrLD-BAF contact surface exists, it is a menin-analogous target; if the interaction is purely multivalent and diffuse, it is not, and that distinction is exactly what the hypothesis predicts will determine druggability.

The rule we are actually proposing

Not merely 'fusions need partners' — that is trivially true. The specific, falsifiable claim is: for a non-enzymatic fusion, at least one recruited-partner interaction is (i) a compact, structured, peptide-in-groove or coiled-coil contact of under about 1,000 square angstroms buried surface, (ii) required for the fusion program but dispensable or tolerated in normal haematopoiesis, and (iii) drugged by a small molecule with at least 10-fold selectivity for the fusion-dependent state. The menin case satisfies (i) and (iii); (ii) is only partly true, since menin has normal functions, and that is the weak point of the whole hypothesis.

Does the generalisation survive its own graph? (cycle-33 counter-pass)

The hypothesis was written as a generalisation from one success. Tested against the evidence this project has accumulated since, it survives as a search strategy and fails as a prediction about yield. Scoreboard:

Partner interfaceInterface real?Chemical matter?Phenotype runs through the interface?Clinic
menin:MBM1 (KMT2A-r, NPM1-mut)yesyesyesapproved
menin:MBM1 (NUP98-r)yesyesyes (VTP50469 evicts both MLL1 and the NUP98 fusion)preclinical [see hyp-nup98-partner-obligate-interface]
CBFbeta-SMMHC:RUNX1yesAI-10-49partly — the cells also need wild-type RUNX1 9no registered trial in 11 years 7
WDR5 WIN siteyespicomolarno — ribosome/p53 8phase 1 only
GFI1 SNAG:LSD1yesclinical-stageyes, patient blasts differentiateno standard of care
ETO NHR2 self-associationyespeptides onlyyesnone
EWSR1 PrLD:BAFdiffuse/multivalentnon/anone

And on the negative side of the ledger, the graph's cycle-21/22 work struck seven partner-side candidates (PRMT1, RBBP5, ASH2L, CTR9, CDC73, CCNT1, ELL2), mostly for pan-essentiality or for having no primary test in KMT2A-fusion disease — see ana-target-list-rewrite-audit.

Verdict. The base rate implied by this table is: obligate partner interfaces are common and findable, small molecules against them are achievable, and the step from there to a drug has succeeded once. The prior of 0.45 was a prior on 'such an interface exists'. It should not be read as a prior on 'such an interface yields a drug', which the table puts far lower. The hypothesis has been re-scoped accordingly: data.statement now claims existence and tractability-in-principle, and data.what_the_graph_shows_does_not_generalise records what does not.

Facts and figures
statementFor every non-enzymatic recurrent fusion oncoprotein there exists at least one obligate recruited-partner or self-association interface that is compact and structured enough to be drugged by a small molecule, is non-redundant for the fusion program, and is more tractable than any surface on the fusion itself.
origindata-pattern
prior probability0.45
counter argument the lab must answerMenin may be tractable not because obligate interfaces are generally druggable but because menin is an unusually pocket-rich protein whose fold happens to present an enclosed cavity at the very site where the MLL peptide binds. If pocket character is the rare ingredient, the hypothesis reduces to 'look for lucky partners', which is a screening strategy rather than a principle. A second and sharper version has emerged from this graph's own work: the WDR5 WIN site satisfies every structural criterion the hypothesis lists, and its inhibitors act through ribosome depletion and p53. If a textbook-perfect peptide-in-groove partner interface can produce a phenotype that has nothing to do with the partner, then 'find the obligate interface' does not by itself constrain what a drug against it will do, and every candidate needs an interface-mutant separation-of-function test before any chemistry is started.
prior probability note0.45 is the prior that a compact obligate interface EXISTS per non-enzymatic fusion, which the evidence supports. The prior that such an interface yields a clinical molecule is much lower: of the interfaces this graph has examined, one (menin) is approved, one (CBFB-SMMHC) has no registered trial 11 years on, one (WDR5 WIN) has a phenotype that does not run through the interface, and one (GFI1 SNAG:LSD1) has validated mechanism and no standard-of-care result.
predictions
[
  "Across the fusion atlas, mechanism classes with a discrete partner interface (chromatin-adaptor, TF-corruption) will have chemical matter in ChEMBL against the partner, while classes without one (enhancer-hijack, pure condensate) will have none — and that is what the current atlas shows.",
  "A CRISPR domain-resolution screen in any non-enzymatic fusion line will nominate a partner-binding domain rather than a fusion domain as the top selective dependency.",
  "If a partner interface is genuinely obligate, resistance will arise at the drug-binding surface rather than by loss of the interaction, because losing the interaction would inactivate the oncoprotein — exactly the pattern seen for MEN1.",
  "Interfaces of this type will be shared across diseases that converge on the same transcriptional program, so drugs will show cross-genotype activity (menin in KMT2A-r and NPM1-mut; predicted for NUP98-r).",
  "If the hypothesis is right, then for any fusion the structured retained interface should be the functional dependency. Two counterexamples are already on record in this beat - the HLF bZIP (PMID 9742120) and the PBX1 homeodomain in TCF3::PBX1, which is dispensable for transformation - so the hypothesis now predicts something checkable and uncomfortable: that KMT2A fusions are unlike TCF3 fusions in this respect, and the test is whether the menin-binding motif is dispensable in the way those interfaces were, measured with the same design (disabling point mutations in the interface, read out on the fusion's own phenotype)."
]
experiments
[
  "exp-cross-family-interface-disruption-panel"
]
analyses
[
  "ana-comparative-interface-druggability-scan"
]
falsifiers
[
  "A domain-resolution CRISPR screen across several non-enzymatic fusion lines nominates only fusion-intrinsic or broadly essential domains, with no selective partner-binding-domain dependency in any of them.",
  "Structural or crosslinking analysis shows that the key fusion-partner contacts are diffuse multivalent IDR interactions with buried surface areas far outside the range that small molecules can address, in most families rather than a minority.",
  "A candidate obligate interface is disrupted genetically (interface point mutants, not knockout) in a fusion-driven model and the leukaemic program persists, showing the interface is redundant.",
  "Partner interfaces identified this way prove to be required equally by normal haematopoietic stem cells, so that no therapeutic window exists in any family — menin, whose inhibition is tolerated, would then be the exception rather than the rule.",
  "A systematic audit of partner-interface small molecules shows that, in most cases where a compound engages the interface, the cellular phenotype is attributable to an off-interface mechanism (as WDR5 WIN-site inhibition proved to be). That would make the hypothesis unfalsifiable in practice, because 'obligate interface' would stop predicting 'interface-mediated drug effect'.",
  "Ten years pass with menin remaining the only obligate partner interface to yield an approved drug, while further candidates accumulate chemical matter and stall -- i.e. the rate-limiting step is shown empirically to be downstream of target identification, making the hypothesis true but strategically empty.",
  "ALREADY OBSERVED, in a modality-specific form: at the WDR5 WIN site - the textbook compact, structured, peptide-in-groove partner interface this hypothesis predicts will be productive - occupancy-based inhibitors were modest in vitro and had no in vivo efficacy, while a degrader of the same protein worked in MLL-rearranged AML patient-derived models (PMID 34586829). If that generalises, the hypothesis is true about targets and false about small-molecule occupancy, which is the form in which it is being used to justify programmes.",
  "ALREADY OBSERVED, in the clearest form yet: TCF3::HLF retains the HLF bZIP - compact, structured, constitutively dimeric and partner-supplied - and disabling mutations within that bZIP, including loss of DNA binding and of dimerisation, leave the chimera's antiapoptotic activity intact so long as the E2A AD1/AD2 transactivation domains remain (PMID 9742120). The dependency sits in the DISORDERED half. The hypothesis requires structural obligateness and functional non-redundancy to travel together, and here they do not."
]
what the graph shows does not generalise
[
  "Interface engagement does not entail interface-mediated mechanism: WDR5 WIN-site inhibitors displace WDR5 from chromatin yet act through ribosome protein gene suppression, nucleolar stress and p53/MDM4 (PMIDs 30865883, 38682900). Nothing in the hypothesis distinguishes these cases a priori.",
  "Validated mechanism does not entail clinical benefit: LSD1 inhibitors disrupt the GFI1 SNAG:LSD1 interaction and differentiate blasts in patients with MLL translocations (PMID 36171271), and are not standard of care in KMT2A-r leukaemia.",
  "Validated chemistry does not entail a trial: no registered interventional study of any CBFbeta-SMMHC:RUNX1 inhibitor exists eleven years after AI-10-49 (ClinicalTrials.gov, queried 2026-09-19).",
  "The 'restore the sequestered partner' logic is wrong for core-binding-factor disease: t(8;21) and inv(16) cells require wild-type RUNX1 to survive and preserve an active RUNX1 allele (PMID 24055056; PMID 23979164).",
  "Seven partner-side candidates nominated inside this project were struck on re-examination (PRMT1, RBBP5, ASH2L, CTR9, CDC73, CCNT1, ELL2; see ana-target-list-rewrite-audit)."
]
How this could bite us back
  • Every interface identified this way is on a wild-type host protein with normal functions, so the entire strategy buys selectivity from dependency ratios rather than from molecular specificity, and each drug will carry on-target normal-tissue toxicity.
  • The one fully replicated analogue (AI-10-49 at the CBFB-SMMHC:RUNX1 interface) stalled before the clinic despite excellent biology, which suggests the rate-limiting step is chemistry and pharmacokinetics rather than target identification.
  • If obligate interfaces are genuinely obligate, the leukaemia will escape by mutating the drug contact rather than the interface, and a single chemical series can then be exhausted by one clone.
  • Engaging an interface is not the same as acting through it. The WDR5 WIN site is a structurally ideal instance of this target class whose inhibitors kill MLL-rearranged cells by depleting ribosomes and inducing p53, which also predicts failure in TP53-mutant relapse -- exactly the setting where new drugs are needed.
  • The class's conversion rate from 'validated interface' to 'registered trial' is currently one in several. If the rate-limiting step is chemistry and pharmacology rather than target identification, a programme organised around finding more interfaces is optimising the step that is not binding.
Still open
  1. Which partner in the whole atlas has the best ratio of fusion-dependency to normal-tissue essentiality, and can that ratio be computed systematically from DepMap plus GTEx plus gnomAD constraint?
  2. Is pocket character at an interface predictable from structure alone, so that candidate interfaces could be triaged computationally before any wet-lab work?
  3. Does the cross-disease sharing seen for menin (KMT2A-r plus NPM1-mut) recur elsewhere, for example a single drug covering NUP98-r and KMT2A-r, or CBFB::MYH11 and RUNX1::RUNX1T1 through RUNX1?
  4. Before any chemistry is started on a candidate partner interface, what is the minimum sufficient separation-of-function test -- interface point mutants rather than knockout, with the drug's phenotype required to phenocopy the mutant -- that would have caught the WDR5 WIN-site case early?
  5. Does the WDR5 result generalise: at a partner interface where occupancy fails, does degradation of the partner succeed? Concretely, is a menin degrader more effective than a menin inhibitor in the settings where revumenib responses are shallow or short, and does it defeat the non-genetic resistance that MEN1-interface mutants do not explain?
Every source behind this entry (9)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A CRISPR domain screen showed NPM1-mutant AML is exceptionally dependent on the menin-binding site in MLL1, and pharmacological menin-MLL1 disruption had profound antileukaemic activity in human and murine NPM1-mutant models.

    PMID 27535106DOI
    What this evidence cannot show

    Cell lines and mouse models; the mechanistic reason NPM1-mutant cells need the menin-MLL1 interaction is inferred from shared HOXA/MEIS1/FLT3 output rather than from a structural requirement.

    Evidence that argues against this (2)
    • The most chemically advanced non-menin partner interface in this whole class turns out not to act through the partner biology it was chosen for. WIN-site inhibitors displace WDR5 from chromatin but do so at a small cohort of loci dominated by ribosome protein genes, producing translational inhibition, nucleolar stress and p53 induction. Aho ER et al. 2019
    • Followed up directly in MLL-rearranged cells: the response to WDR5 WIN-site inhibition is dominated by ribosome subunit attrition and p53-MDM4 activation, and the authors state that the HSC-expansion mechanism previously proposed for peptidomimetic WIN-site inhibitors is not what their data show. So a small molecule at a bona fide, structurally defined, obligate-looking partner interface produced a phenotype that was not about the partner's chromatin biology at all. Interface engagement does not entail interface-mediated mechanism, and the hypothesis has no way to tell the two apart in advance. Howard GC et al. 2024
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Oral menin-MLL1 inhibition reversed extended self-renewal of myeloid progenitors in NPM1-mutant knock-in mice, showing the dependency exists before overt leukaemia.

    PMID 32001657DOI
    What this evidence cannot show

    A mouse knock-in model with a specific cooperating genotype; preventative treatment of humans at risk is untested and the durability after drug withdrawal was limited.

    Evidence that argues against this (1)
    • A dependency demonstrated in a clean knock-in mouse before overt leukaemia is the most favourable possible test of this hypothesis, and the patient version of the same measurement disagrees. Serial samples from two children with multiply relapsed KMT2A::AFF1 ALL showed that heavily pretreated cells were much less sensitive to menin inhibition than earlier samples from the same patient, that this resistance was acquired with NO prior menin-inhibitor exposure, and that the drug's on-target repression of its canonical targets was fully preserved in the resistant cells. So the obligate interface can remain engaged, and engaged productively at the level of transcription, while the cell no longer dies. 'Obligate for the program' and 'obligate for survival' are different claims and this hypothesis needs the second. Mahdavi L et al. 2025
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A single menin inhibitor produced clinical remissions in both KMT2A-rearranged and NPM1-mutant acute leukaemia, demonstrating that one host-adaptor interface can serve genetically unrelated diseases.

    The menin inhibitor revumenib in KMT2A-rearranged or NPM1-mutant leukaemia · Issa GC, Aldoss I, DiPersio J, et al. · Nature 2023
    PMID 36922593DOI
    What this evidence cannot show

    Single-arm phase 1; response rates in the two genotypes are not formally compared and the combined efficacy population is small.

    Evidence that argues against this (1)
    • Cross-genotype coverage cuts against this hypothesis as much as for it. The hypothesis asserts that each fusion has its own obligate recruited-partner interface; a single drug working in both KMT2A-rearranged leukaemia and NPM1-mutant AML - which has no fusion, no partner and no rearrangement - shows that what is being targeted is a shared menin-dependent HOXA/MEIS1 transcriptional state rather than a partner interface belonging to any fusion. The depth of that coverage also needs stating: in the NPM1-mutant phase 2 cohort the CR+CRh rate was 23.4%, the median duration of CR+CRh was 4.7 months, and 5 of 30 responders reached transplant. A host-adaptor interface serving unrelated genotypes for under five months is a different proposition from an obligate fusion interface. Arellano ML et al. 2025
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    AI-10-49 disrupts the CBFB-SMMHC:RUNX1 interface, restores RUNX1 transcriptional activity, selectively kills primary inv(16) blasts and delays leukaemia in mice — an independent, fully chemically validated obligate partner interface.

    PMID 25678665DOI
    What this evidence cannot show

    Leukaemia was delayed rather than eradicated, and the compound has not progressed to registration, so this analogue supports the hypothesis biologically while warning about its translational ceiling.

    Evidence that argues against this (2)
    • Eleven years after the Science paper, a ClinicalTrials.gov API query for CBFB-SMMHC and for AI-10-49 returns no registered interventional study of any CBFbeta-SMMHC:RUNX1 interface inhibitor. The one fully replicated analogue of the menin story has produced zero trials, which is the strongest available measurement of how far 'validated obligate interface' is from 'drug'. US National Library of Medicine 2026
    • The mechanistic framing behind AI-10-49 is also incomplete in a way that matters for the hypothesis. CBFbeta-SMMHC is described here as acting by outcompeting wild-type CBFbeta, i.e. as a dominant negative. But inv(16) ME-1 and t(8;21) Kasumi-1 cells REQUIRE wild-type RUNX1 for survival: RUNX1 knockdown attenuates the mitotic checkpoint and causes apoptosis, knockdown of AML1-ETO rescues it, and an active RUNX1 allele is usually preserved in inv(16) and t(8;21) patients. The pathology is a balance, not a simple sequestration, so an interface drug that 'restores RUNX1 activity' is tuning a set point rather than reversing a blockade. Ben-Ami O et al. 2013
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Oligomerisation through the ETO NHR2 domain is required for RUNX1/ETO oncogenic activity, and interfering peptides dissolve the complex, restore target-gene transcription and revert the differentiation block.

    PMID 17332359DOI
    What this evidence cannot show

    Peptide expression in cell lines with no in vivo efficacy and no drug-like molecule; a peptide-accessible interface is not necessarily a small-molecule-accessible one.

    Evidence that argues against this (2)
    • A partner interface whose mechanism IS validated still delivers only a modest clinical result. LSD1 inhibitors work at GFI1-occupied sites by disrupting the GFI1-LSD1 protein-protein interaction (and the HMG20B-LSD1 interaction), and they induce morphologic and molecular differentiation of blasts in patients with MLL translocations -- yet no LSD1 inhibitor has become a standard of care in KMT2A-rearranged leukaemia. Mechanism validation at a partner interface is necessary and demonstrably not sufficient. Maiques-Diaz A et al. 2022
    • A sibling case tested this hypothesis's core prediction on a textbook example of the interface class and the prediction failed. TCF3::HLF retains the HLF basic leucine zipper - a compact, structured, constitutively dimeric, partner-supplied interface, exactly the shape this hypothesis says should be productive. A panel of E2A-HLF mutants expressed in IL-3-dependent murine pro-B cells showed that different combinations of DISABLING MUTATIONS WITHIN THE HLF bZIP DOMAIN had little effect on the chimera's antiapoptotic activity, so long as the amino-terminal portion of E2A remained intact; on a bZIP defective in DNA binding, mutants retaining EITHER of the two E2A transactivation domains still extended survival after growth-factor withdrawal. The authors state that neither DNA binding nor protein dimerisation through the HLF bZIP is required for the effect, and locate the dependency in the disordered E2A AD1/AD2 regions instead. An interface can therefore be compact, structured, obligate in the structural sense and partner-supplied, and still not be the functional dependency - which is the conjunction this hypothesis asserts cannot come apart. Bound on the refutation: the readout is survival of cytokine-deprived murine pro-B cells, not human leukaemogenesis, so it refutes the necessity of the interface for the antiapoptotic arm rather than for the whole disease. Found by the lymphoid arm of this cycle's counter-sweep; PMID and DOI re-fetched and checked independently before recording here. Inukai T et al. 1998
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    EWS-FLI1 recruits BAF (mSWI/SNF) complexes to tumour-specific enhancers through the EWSR1 prion-like domain, a neomorphic activity required for its function and absent from wild-type FLI1.

    Cancer-specific retargeting of BAF complexes by a prion-like domain · Boulay G, Sandoval GJ, Riggi N, et al. · Cell 2017
    PMID 28844694DOI
    What this evidence cannot show

    The PrLD-BAF interaction may be multivalent and diffuse rather than a discrete interface, in which case it is a counterexample to the hypothesis rather than a supporting case; no structure of the contact exists.

    Evidence that argues against this (1)
    • This is the hypothesis's hardest case and the evidence has moved against it. The hypothesis requires the obligate interface to be 'compact and structured enough to be drugged by a small molecule'; the EWSR1 prion-like domain is neither, and quantitative single-molecule and single-cell imaging in patient-derived Ewing cells shows its relationship to transcription is not even monotonic. EWS::FLI1 needs a narrow optimum of low-complexity-domain interaction: increasing those interactions toward putative phase separation REPRESSES its GGAA-microsatellite targets, and sequestering the fusion into the nucleolus inhibits both its program and transformation. An interface with an optimum rather than a threshold cannot be drugged by occupancy in the way this hypothesis assumes, and partial disruption could increase output. Chong S et al. 2022
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.trial_registry

    No registered interventional trial of any CBFbeta-SMMHC:RUNX1 interface inhibitor exists, eleven years after AI-10-49 was published.

    What this evidence cannot show

    A registry-absence claim: it shows no trial is registered under these terms, not that no programme exists. Queried 2026-09-19 via the ClinicalTrials.gov v2 API with query.term=CBFB-SMMHC and query.term=AI-10-49; the only hit for the latter was an unrelated myeloma study matched on a text fragment.

    Evidence that argues against this (1)
    • The absence of a CBFbeta-SMMHC interface trial is real (re-checked against the ClinicalTrials.gov v2 API on 2026-09-19: query.term=CBFB-SMMHC and query.term=AI-10-49 both return no matching interventional study), but reading it as neglect is the wrong inference. The field IS running an interventional trial in core-binding-factor AML - aimed at the acquired KIT co-lesion rather than at the fusion interface. NCT05821738, 'Avapritinib in Relapsed Refractory or MRD-positive CBF-AML With KIT Mutations', is a phase 2 study that started on 2022-06-01, sponsored by the First Affiliated Hospital of Soochow University, with avapritinib as the sole intervention (registry status UNKNOWN as of this check). Given that roughly half of t(8;21) patients carry an activating KIT mutation acquired after the fusion, the eleven-year gap is better explained as the field choosing a validated kinase pocket on a secondary lesion over an unvalidated interface on the primary one. That is a competing explanation this evidence item cannot exclude, and it is the one the hypothesis should worry about. The First Affiliated Hospital of Soochow University 2022
  8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Small-molecule inhibitors of the WDR5 WIN site -- a structurally defined, peptide-in-groove partner interface of the type this hypothesis predicts will be productive -- act in MLL-rearranged cells through ribosome protein gene suppression, nucleolar stress and p53/MDM4, not through the partner chromatin biology for which the interface was selected.

    PMID 38682900DOI
    What this evidence cannot show

    Abstract only for both WDR5 papers (PMIDs 30865883 and 38682900); cell lines, one laboratory's chemical series, and p53 dependence would predict failure specifically in TP53-mutant relapse. The graph's subtraction beat reached the same reading independently.

    Evidence that argues against this (1)
    • This item is used to argue that a structurally ideal partner interface can be chemically tractable and still be the wrong target, and that reading is too pessimistic: the WDR5 failure was a failure of MODALITY, not of target. In the same system, occupancy-based WDR5 protein-protein-interaction inhibitors gave modest cancer-cell killing and lacked in vivo efficacy, while a structure-designed VHL-recruiting WDR5 degrader (MS67, with two ternary-complex crystal structures) was more effective than the PPI inhibitors at suppressing WDR5-regulated transcription, at reducing the chromatin-bound fraction of MLL complex components and c-MYC, and at inhibiting proliferation, and it suppressed MLL-rearranged AML patient cells in vitro and in vivo while being well tolerated. The on-mechanism effect at this interface is reachable; blocking a peptide-in-groove contact by occupancy is not the way to reach it. This is the most consequential warning in this entry for the lab's own programme, because menin inhibitors are occupancy drugs at a peptide-in-groove partner interface. Yu X et al. 2021
  9. E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Wild-type RUNX1 is required for the survival of t(8;21) and inv(16) leukaemia cells, and an active RUNX1 allele is usually preserved in those patients, so the core-binding-factor fusions are not simple dominant negatives and an interface drug at that interface is adjusting a balance rather than lifting a block.

    Addiction of t(8;21) and inv(16) acute myeloid leukemia to native RUNX1. · Ben-Ami O, Friedman D, Leshkowitz D, et al. · Cell Reports 2013
    PMID 24055056DOI
    What this evidence cannot show

    Abstract only. Two cell lines (Kasumi-1, ME-1) plus a clinical-genetics inference from allele preservation; not tested in primary cells or in vivo in that paper. Independently supported by Goyama 2013 (PMID 23979164), which also found that RUNX1 is required for MLL-AF9 cell survival via BCL2 and that RUNX2 can compensate.

    Evidence that argues against this (1)
    • 'Non-redundant for the fusion program' is necessary for this hypothesis but not sufficient, because the same requirement holds in the host. Conditional deletion of Runx1 in adult mice blocks common lymphoid progenitor production, impairs platelet output, perturbs B- and T-cell maturation, and produces a mild myeloproliferative phenotype with expanded myeloid progenitors and extramedullary haematopoiesis. A CBF-directed agent is therefore not exploiting a fusion-specific dependency at all; it is exploiting a shared one and hoping for a dose window. The hypothesis's own criterion - 'more tractable than any surface on the fusion itself' - is about chemistry and says nothing about selectivity, and for the core-binding-factor row selectivity is where the argument fails. Growney JD et al. 2005
Connected to