Oncofusion

start / graph / fusion-kmt2a-mllt10

Fusionstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q11-comparative

KMT2A::MLLT10 (MLL-AF10), t(10;11)(p12;q23)

also called MLL-AF10 · MLL::AF10 · KMT2A-AF10 · MLL/MLLT10 · t(10;11)(p12;q23)

In this version the broken KMT2A gene joins AF10, and the only part of AF10 that matters is a short coiled stretch that grabs an enzyme called DOT1L and drags it onto the wrong genes. It is one of the commonest partners in infant and childhood myeloid leukemia and one of the worst, and the same DOT1L-grabbing stretch turns up in a completely different fusion, CALM-AF10, which is strong evidence that this one small module is the whole problem.

The same thing, for a scientist

KMT2A::MLLT10 accounts for 304/3401 (8.9%) of genomically characterised KMT2A rearrangements and is heavily myeloid: ~18.7% of KMT2A-rearranged AML, 27.4% of infant AML and 23.4% of paediatric AML KMT2A-r cases, with only a minor share of ALL [E1]. Paediatric cases shift the KMT2A break towards intron 9 [E1]. The partner contribution is unusually well defined: the fusion retains the AF10 C-terminal **octapeptide-motif / leucine-zipper (OM-LZ)** module, which binds the DOT1L coiled-coil and is necessary and sufficient with the KMT2A half to drive progressive H3K79 methylation and HOXA gene activation [E2]. The N-terminal AF10 **PZP** (PHD-Zn knuckle-PHD) histone-reading module is lost. Genetic or pharmacological Dot1l inactivation abrogates transformation by both MLL-AF10 and the KMT2A-independent CALM-AF10 [E3], and CALM-AF10 leukemogenesis likewise runs through hDOT1L-dependent HOXA upregulation [E5] - two independent fusions converging on one 80-residue module, which is the cleanest demonstration in the whole family that a partner contributes a single decomposable function. Clinically t(10;11)(p12;q23) was an independently adverse subgroup in paediatric KMT2A-r AML (HR 1.5, p=0.005) against an overall 5-year EFS of 44% +/- 5% [E4]. Menin inhibition applies partner-agnostically [E6].

What it looks like
loading 6JN2…
PDB 6JN2 ↗
shape

This is real experimental data — atoms placed by crystallography or cryo-EM, deposited by the people who did the experiment.

I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
What is kept and what is lost on each side1 deeper

Kept from KMT2A: AT-hooks 1-3, SNL/repression region, MBM1/MBM2, LEDGF-binding region, CXXC. Lost: PHD1-3, bromodomain, ePHD4, FYRN/FYRC, TAD, WIN, SET 1.

Kept from MLLT10/AF10: the C-terminal OM-LZ module - an octapeptide motif followed by a leucine zipper that together form the DOT1L-binding coiled-coil interface. This is the transforming element 2.

Lost from MLLT10/AF10: the N-terminal PZP module (PHD finger - zinc knuckle - PHD finger), a histone H3 tail reader, plus the AF10 nuclear-localisation and central regions upstream of the break. There are crystal structures of the AF10 PZP domain bound to an H3 tail (5DAG, 7MJU) - and they describe a part of AF10 that is not in the oncoprotein, a trap worth flagging explicitly because the structures are easy to find and easy to misattribute 2.

So the partner's entire contribution reduces to one coiled-coil docking module. That is a genuinely different situation from AFF1 (a large multi-element scaffold) and makes MLLT10 the best test case for the proposition 'if you block the partner interface, you cure the disease'.

Cryptic and complex rearrangements

MLLT10 fusions are disproportionately cryptic: because 10p12 and 11q23 rearrangements in this partner frequently involve insertions, inversions or three-way exchanges, conventional karyotyping under-detects them and FISH or RNA sequencing is often required. That matters for two reasons: partner-frequency estimates for MLLT10 are probably biased downwards in karyotype-based series, and any partner-stratified therapy depends on molecular diagnostics being performed. Paediatric KMT2A::MLLT10 additionally shifts the KMT2A break towards intron 9, into the HOXA-high group 1.

Mechanism: the DOT1L recruitment module, isolated

Class: chromatin-elongation-hijack, and the most mechanistically reduced example of it.

The shared steps are the family's: menin/LEDGF docking, CXXC-directed targeting to CpG-island promoters. The partner supplies the effector.

AF10's OM-LZ binds the DOT1L coiled-coil directly; there is a crystal structure of the human DOT1L coiled-coil domains bound to AF10 (PDB 6JN2). Deshpande and colleagues showed that AF10 (and its paralogue AF17) is required for progressive H3K79 methylation - the conversion of H3K79me1 to me2/me3 - and that loss of AF10 collapses H3K79me2/3 and HOXA expression not only in MLL-AF10 leukemia but across diverse AML subtypes including MLL-AF9 and DNMT3A-mutant AML 2. In other words the OM-LZ:DOT1L axis is a general AML dependency that this particular fusion hard-wires.

The convergence argument. MLLT10 is also the 3' partner of PICALM (CALM-AF10) in T-ALL and AML - a fusion with no KMT2A involvement at all, no menin dependence and no CXXC domain, yet the same OM-LZ module and the same output. CALM-AF10 transformation requires hDOT1L and proceeds through HOXA5/HOXA upregulation 5, and Dot1l inactivation abrogates both MLL-AF10 and CALM-AF10 transformation 3. Two structurally unrelated N-terminal halves, one shared 80-residue C-terminal module, one shared leukemic program. This is the strongest available evidence anywhere in the fusion-oncoprotein field that a partner interface can be the whole mechanism - and therefore the strongest single argument for generalising the menin lesson to partner interfaces.

And the counter-argument, in the same system. The molecule that blocks that axis pharmacologically - pinometostat, a DOT1L catalytic inhibitor - reduced H3K79 methylation in patients and produced only modest single-agent activity. The interface is obligate; blocking its enzymatic output is not curative. Any generalisation of the menin lesson has to survive that fact.

Clinical phenotype, age and prognosis

Lineage skew: strongly myeloid. ~18.7% of KMT2A-r AML versus a minor share of ALL; 27.4% of infant AML and 23.4% of paediatric AML KMT2A-r cases 1. In the infant myeloid setting MLLT10 is the single commonest partner - the mirror image of AFF1's dominance in infant lymphoid disease. Where MLLT10 appears in lymphoid disease it is disproportionately T-ALL, consistent with its behaviour as PICALM-MLLT10.

Age skew: infancy through childhood, with a strong infant-AML representation and a lower adult share than MLLT3.

Prognosis: adverse. In the 756-patient international paediatric KMT2A-r AML series, t(10;11)(p12;q23) was an independent adverse prognostic subgroup (HR 1.5, p=0.005), and the separate t(10;11)(p11.2;q23) subgroup was worse still (HR 2.5, p=0.005), against an overall 5-year EFS of 44% +/- 5% 4. Only t(6;11) (AFDN) carried a comparable or greater hazard. So the KMT2A-r AML prognostic ladder in that study runs roughly: t(1;11) best, t(9;11)/MLLT3 intermediate, t(10;11)/MLLT10 adverse, t(6;11)/AFDN worst 4.

Contrast within the family. AFF1 is the bad infant lymphoid partner; MLLT10 and AFDN are the bad myeloid partners. MLLT3, the commonest myeloid partner, is the intermediate reference. Lineage is set by partner; prognosis within a lineage is also set by partner - but by a different ranking.

Druggability and therapeutic status

Menin inhibition: clinically validated for the class, partner-agnostic (CR+CRh 22.8%, ORR 63.2% in KMT2A-translocated R/R disease) 6.

DOT1L: the best-validated partner-specific target in the family, and a clinical disappointment. The dependency is established genetically in exactly this fusion 235. Pinometostat engaged the target in patients and delivered only modest monotherapy activity, which is the field's central cautionary result.

The OM-LZ:DOT1L coiled-coil interface itself is undrugged. This is the interesting gap. Pinometostat inhibits DOT1L's catalytic site; nobody has drugged the interface by which the fusion recruits DOT1L. Given that the menin lesson is specifically about interfaces rather than active sites, and given that a crystal structure of the DOT1L coiled-coil bound to AF10 exists (6JN2), this is the most obvious untried experiment in the comparative set. A coiled-coil interface is a hard target - long, shallow, largely helical - but it is not obviously harder than menin-MBM1 looked in 2005.

Overall: emerging. No partner-specific agent; covered only by menin inhibition.

Facts and figures
gene5KMT2A
gene3MLLT10
cytobandt(10;11)(p12;q23)
cytoband noteFrequently cryptic: insertions, inversions and three-way rearrangements mean the translocation is often invisible on conventional karyotype and requires FISH or RNA sequencing. Distinct from the separate adverse t(10;11)(p11.2;q23) subgroup, which is not MLLT10.
prognosisAdverse: t(10;11)(p12;q23) was an independent adverse subgroup in paediatric KMT2A-rearranged AML (HR 1.5, p=0.005) against an overall 5-year EFS of 44% +/- 5% in 756 children; only t(6;11)/AFDN and t(10;11)(p11.2;q23) carried greater hazard.
uniprot5Q03164
uniprot3P55197
uniprot5 verifiedQ03164 = KMT2A_HUMAN, 3969 aa (UniProt REST, checked 2026-09-18)
uniprot3 verifiedP55197 = AF10_HUMAN, gene MLLT10, protein AF-10, 1068 aa, Homo sapiens (UniProt REST, checked 2026-09-18)
exon numbering noteKMT2A exon numbers follow the Meyer/Marschalek DCAL consensus. The MLLT10 exon number is indicative of the OM-LZ-preserving breakpoint region rather than an exact, universally reported junction; treat it as schematic.
PDB noteNo structure of the KMT2A::MLLT10 fusion protein exists. 6JN2 = crystal structure of the coiled-coil domains of human DOT1L in complex with AF10 - this is the interface the fusion actually uses and the obvious undrugged interface target; 7UJ4 and 4GQ6 = menin with revumenib and with an MLL peptide, the partner-agnostic drugged node. Deliberately EXCLUDED: 5DAG, 5DAH and 7MJU (AF10 PZP domain, alone and bound to histone H3) - these resolve and are genuine AF10 structures, but the PZP module is lost from the oncoprotein, so citing them as fusion structures would be misleading. Verified against the RCSB entry API on 2026-09-18.
druggability classemerging
druggability noteClinically validated only through the partner-agnostic menin node. The partner-specific dependency (DOT1L) is the best genetically validated in the family yet failed as a catalytic-site monotherapy; the OM-LZ:DOT1L coiled-coil INTERFACE - which has a crystal structure (6JN2) - has never been drugged and is the most obvious untried interface target in this comparison set.
stage of hematopoiesishaematopoietic stem cell or early myeloid progenitor in the bone marrow (and a T-lineage precursor in the minority lymphoid cases); earlier and less lineage-restricted than the GMP origin modelled for KMT2A::MLLT3, consistent with the high infant-AML share and HOXA-high signature
age peak<2 y for the infant-AML peak, extending through childhood
developmental order rank4
developmental order noteRanked 4 of 10: earlier than MLLT3 (GMP) because of the high infant-AML share and HOXA-high, intron-9 signature, but later than KMT2A::AFF1 and the infant KMT2A::MLLT1 cases, which have demonstrable prenatal origin. This is an inference from clinical and transcriptional signature, not from a lineage-tracing experiment.
domain name notedomains_retained_5 / domains_lost_5 use the exact domain names from data/graph/nodes/king/gene-kmt2a.json so the dashboard's 3D domain ruler can match them. The LEDGF/PSIP1-binding region is retained biologically but has no lane in the parent node, so it is recorded in each breakpoint's phenotype_notes instead.
frequency
all_kmt2a_r_acute_leukemi…8.900all_kmt2a_r_acute_leukemia_pct: 8.9kmt2a_r_aml_pct18.7kmt2a_r_aml_pct: 18.7kmt2a_r_paediatric_aml_pct23.4kmt2a_r_paediatric_aml_pct: 23.4kmt2a_r_infant_aml_pct27.4kmt2a_r_infant_aml_pct: 27.4n_cases_dcal_cohort304.0n_cases_dcal_cohort: 304
diseases
[
  "infant AML",
  "paediatric AML",
  "adult AML",
  "T-ALL (minority)",
  "MPAL"
]
frequency — the numbers behind the chart
{
  "all_kmt2a_r_acute_leukemia_pct": 8.9,
  "kmt2a_r_aml_pct": 18.7,
  "kmt2a_r_infant_aml_pct": 27.4,
  "kmt2a_r_paediatric_aml_pct": 23.4,
  "n_cases_dcal_cohort": 304
}
breakpoints
[
  {
    "id": "bp-kmt2a-intron-9-mllt10-omlz",
    "gene5_exon_last": 9,
    "gene3_exon_first": 10,
    "frequency_pct": 37.2,
    "population": "paediatric KMT2A::MLLT10 preferentially breaks in intron 9; frequency_pct is the share of all KMT2A breaks falling in BCR subregion A (exon 9 + intron 9), not the share of MLLT10 fusions",
    "domains_retained_5": [
      "MBM1",
      "MBM2",
      "AT-hook 1",
      "AT-hook 2",
      "AT-hook 3",
      "SNL1/2",
      "CXXC"
    ],
    "domains_lost_5": [
      "PHD1",
      "PHD2",
      "PHD3",
      "Bromodomain",
      "PHD4",
      "FYRN",
      "TAD",
      "FYRC",
      "WIN",
      "SET"
    ],
    "domains_retained_3": [
      "OM (octapeptide motif)",
      "LZ (leucine zipper)",
      "AF10 C-terminal region"
    ],
    "domains_lost_3": [
      "PZP module (PHD - Zn knuckle - PHD histone H3 reader)",
      "AF10 NLS and central region",
      "AF10 AT-hook region"
    ],
    "phenotype_notes": "HOXA-high group. Myeloid, infant- and paediatric-AML dominant. gene3_exon_first = 10 reflects the AF10 breakpoint region that preserves the OM-LZ (the same region used by CALM-AF10) and is approximate; MLLT10 rearrangements are structurally heterogeneous and frequently cryptic, so a single exon pair does not describe all cases. The retained KMT2A half also contains the LEDGF/PSIP1-binding region (no counterpart lane in the parent gene-kmt2a domain map, so it is recorded here rather than in the domain arrays)."
  }
]
PDB IDS
[
  "6JN2",
  "7UJ4",
  "4GQ6"
]
mechanism class
[
  "chromatin-elongation-hijack"
]
drugs
[
  "drug-revumenib",
  "drug-ziftomenib",
  "drug-pinometostat"
]
key interactors
[
  "gene-dot1l",
  "gene-men1",
  "gene-psip1",
  "gene-hoxa9",
  "gene-meis1"
]
trials
[
  "NCT04065399"
]
How this could bite us back
  • Resistance: menin-inhibitor resistance via MEN1 binding-site mutation applies identically, since the drugged surface is on the shared KMT2A half.
  • Compensation: AF10 has a functional paralogue, AF17/MLLT6, which also recruits DOT1L; endogenous AF17 could buffer an AF10- or OM-LZ-directed agent, and the H3K79me2/3 mark itself decays only over several cell divisions, so inhibition is inherently slow-acting against a rapidly proliferating blast.
  • Essential-function toxicity: DOT1L and H3K79 methylation are required in normal haematopoiesis, cardiac development and neural tissue, and AF10's PZP-mediated functions are needed in normal chromatin regulation. An interface-directed agent would hit endogenous AF10-DOT1L complexes as well as the fusion, because the interface is not neomorphic.
  • Proof-of-concept hazard: this fusion is the field's best argument that a partner interface is the whole mechanism, and also the site of its clearest clinical failure (pinometostat). A generalisation drawn from menin must explain why the better-validated dependency here did worse in patients.
  • Delivery and diagnosis: cryptic rearrangements mean patients may not be identified as MLLT10 at all without RNA sequencing, so a partner-specific therapy has a diagnostic-access problem before it has a pharmacology problem.
Still open
  1. The OM-LZ:DOT1L coiled-coil interface has a crystal structure and is the fusion's entire effector mechanism, yet nobody has tried to drug the interface itself - only DOT1L's catalytic site, which failed clinically. Is a helical, largely hydrophobic coiled-coil interface tractable to a stapled peptide, a macrocycle or a molecular glue, and would blocking recruitment succeed where blocking catalysis did not?
  2. CALM-AF10 shares the effector module but not the menin-binding half. It should therefore be completely refractory to menin inhibitors and fully sensitive to an OM-LZ-directed agent. That asymmetric prediction is a decisive, cheap experiment that would establish whether partner interfaces are generalisable drug targets.
  3. Why is the same partner adverse in AML (HR 1.5) when its paralogue AF17/MLLT6 is a rare partner and MLLT3 is only intermediate? Is the difference the strength of DOT1L recruitment, or the cell in which each partner locus is transcribed and spatially available?
  4. Because MLLT10 rearrangements are frequently cryptic, how much KMT2A-r AML classified as 'partner unknown' is actually MLLT10, and would reclassification change the prognostic ladder?
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    KMT2A::MLLT10 accounts for 304/3401 (8.9%) of genomically characterised KMT2A rearrangements, about 18.7% of KMT2A-rearranged AML, 27.4% of infant AML and 23.4% of paediatric AML KMT2A-r cases, and paediatric cases shift the KMT2A break towards intron 9.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    Referral-based single-centre cohort. MLLT10 rearrangements are frequently cryptic (insertions, three-way exchanges) and karyotype-based ascertainment under-detects them, so 8.9% is plausibly an underestimate. Percentages are of cases with an identified partner.

    Evidence that argues against this (1)
    • These are shares of the recombinome series rather than population frequencies. The series is what a reference-laboratory network accumulated from samples sent to it for KMT2A breakpoint characterisation between 2003 and 2022, so its denominator is referrals, not a population. It is also a long-tailed distribution: of 107 in-frame KMT2A fusions catalogued, 37 are recurrent and 63 have been seen only once, and the seven commonest partners plus partial tandem duplications account for more than 90% of recombinations. Percentages from it are shares of a referral series and should be quoted that way, especially alongside percentages from population-based registries or from prospective trial cohorts, which are computed on different denominators. Meyer C et al. 2023
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    AF10 is required for progressive H3K79 methylation (me1 to me2/me3) and for HOXA gene expression across diverse AML subtypes, and its C-terminal OM-LZ module is the DOT1L-binding element that mediates this.

    AF10 regulates progressive H3K79 methylation and HOX gene expression in diverse AML subtypes · Deshpande AJ, Deshpande A, Sinha AU, et al. · Cancer Cell 2014
    PMID 25464900DOI
    What this evidence cannot show

    Mouse models and human cell lines; the requirement is for AF10/AF17 as endogenous proteins as well as within the fusion, so the result does not cleanly isolate the fusion's own OM-LZ. Paralogue redundancy with AF17 (MLLT6) means single-gene loss may understate the pathway's robustness.

    Evidence that argues against this (1)
    • The AF10-DOT1L axis was taken into patients and the result bounds how much of the disease it accounts for. In the phase 1 study of pinometostat, a first-in-class DOT1L inhibitor, 51 adults with advanced acute leukaemia were treated across six dose-escalation and two expansion cohorts; pharmacodynamic reduction of H3K79 methylation was achieved, the maximum tolerated dose was never reached, and exactly two patients attained complete remission - both with t(11;19), neither with a t(10;11). Single-agent efficacy was described by the authors as modest. A partner module that is required for H3K79 methylation and HOXA expression in the laboratory can still be an insufficient point of attack, and for the t(10;11) group specifically there is no reported response. Stein EM et al. 2018
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Genetic inactivation or pharmacological inhibition of Dot1l abrogates transformation by both MLL-AF10 and the KMT2A-independent CALM-AF10, showing the shared AF10 module rather than the 5' partner carries the transforming requirement.

    PMID 23138183DOI
    What this evidence cannot show

    Murine transformation assays with retroviral overexpression, not endogenous-level human disease; 'abrogation' is measured as colony formation and immortalisation rather than as cure of an established leukemia, which is precisely the gap the pinometostat trial later exposed.

    Evidence that argues against this (1)
    • Clinical DOT1L inhibition with pinometostat confirmed target engagement (loss of H3K79 methylation) yet gave only modest activity, so preclinical abrogation did not translate. Stein EM et al. 2018
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 756 children with 11q23/KMT2A-rearranged AML, t(10;11)(p12;q23) was an independent adverse prognostic subgroup (HR 1.5, p=0.005) and t(10;11)(p11.2;q23) worse still (HR 2.5, p=0.005), against an overall 5-year EFS of 44% +/- 5%.

    PMID 19528532DOI
    What this evidence cannot show

    Retrospective pooling across 11 study groups and many protocols; subgroups were assigned by centrally reviewed karyotype, so cryptic MLLT10 rearrangements without a visible t(10;11) will have been misclassified, and the t(10;11)(p11.2;q23) subgroup is not molecularly identical to MLLT10. Pre-dates menin inhibitors and modern MRD stratification.

    Evidence that argues against this (1)
    • The two t(10;11) subgroups in that analysis are cytogenetic categories, and 10p23-p11 rearrangements are molecularly heterogeneous. Genomic breakpoint characterisation of 3401 acute leukaemia patients catalogued 107 in-frame KMT2A fusions, of which 37 are recurrent and 63 have been observed only once, and resolves translocation partner genes at the sequence level rather than by band. A hazard ratio computed on t(10;11)(p12;q23) and t(10;11)(p11.2;q23) as karyotypic groups pools whatever partners fall in those bands, which is a different object from KMT2A::MLLT10 defined molecularly - and the entry uses the hazard ratio as if it belonged to the molecular fusion. Meyer C et al. 2023
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Leukaemic transformation by CALM-AF10 - a fusion containing the same AF10 OM-LZ module but no KMT2A - involves hDOT1L-dependent upregulation of Hoxa5, demonstrating convergence of two unrelated fusions on one partner module.

    Leukaemic transformation by CALM-AF10 involves upregulation of Hoxa5 by hDOT1L · Okada Y, Jiang Q, Lemieux M, Jeannotte L, Su L, Zhang Y · Nature Cell Biology 2006
    PMID 16921363DOI
    What this evidence cannot show

    Mouse and cell-line work focused on Hoxa5 specifically rather than the whole HOXA cluster; CALM-AF10 disease in humans is predominantly T-ALL with a HOXA signature, so the convergence is at the level of the HOXA program rather than an identical disease.

    Evidence that argues against this (1)
    • Convergence of two unrelated fusions on one partner module is a real and elegant biological result, and it did not convert into a shared therapeutic dependency that could be exploited. DOT1L is the node both fusions were argued to share, and the first-in-class DOT1L inhibitor produced two complete remissions in 51 treated patients, both in t(11;19) - a partner class that this convergence argument does not cover. Mechanistic convergence on a module and convergence on a drug-sensitive dependency are separable, and this pair separates them. Stein EM et al. 2018
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Revumenib was studied and approved in KMT2A-translocated relapsed/refractory acute leukemia without restriction by partner gene, achieving CR+CRh 22.8% and ORR 63.2%.

    Menin Inhibition With Revumenib for KMT2A-Rearranged Relapsed or Refractory Acute Leukemia (AUGMENT-101) · Issa GC, Aldoss I, Thirman MJ, et al. · Journal of Clinical Oncology 2025
    PMID 39121437DOINCT04065399
    What this evidence cannot show

    No partner-stratified efficacy analysis; activity in KMT2A::MLLT10 specifically is inferred from the shared retained MBM module and partner-agnostic eligibility. Note that CALM-AF10, which shares the effector module but lacks the menin-binding half, would not be expected to respond - a clean clinical prediction that has not been formally tested.

    Evidence that argues against this (1)
    • The approval is partner-agnostic; sensitivity in practice appears to track treatment history at least as strongly. In patient-derived xenografts, three AMLs with high-risk fusion partners - MLLT10 and MLLT4/AFDN among them - were sensitive to menin inhibition, but serial samples from multiply relapsed patients were markedly less sensitive than earlier samples from the same individuals, with the resistance acquired in the complete absence of any menin-inhibitor exposure and with on-target repression of the canonical target genes preserved. For this partner the encouraging PDX result and the discouraging serial-sample result come from the same paper, and the variable separating them is prior therapy, not the partner gene. Mahdavi L et al. 2025
Connected to