Two chromosomes break.
A child gets leukemia.
The hybrid protein this creates has no drug-shaped pocket of its own. So we go after what it holds on to. This is everything the lab knows about that problem, with the source, the strength and the counter-argument attached to every claim.
A rare disease that has not improved in thirty years
About 160 babies a year in the United States are diagnosed with acute leukemia. Most of those with the lymphoblastic form carry a broken KMT2A gene, and fewer than four in ten are alive and disease-free three years later.
Children over one year old with lymphoblastic leukemia now survive at rates above 85%. Infants with a broken KMT2A do not. That gap, not the rarity, is the reason this project exists.
Open the full entry: How many children this affects4 sources
Leukemia in babies under one year old is rare: about 41 cases per million infants each year in the United States, which is roughly 160 babies a year. Most of those babies with lymphoblastic leukemia carry a broken KMT2A gene.
Watch the denominator: 3 per million and 41 per million are both right1 deeper
Rarity figures for this disease differ by more than tenfold depending on who is in the denominator.
| Statement | Denominator | Approximate rate |
|---|---|---|
| Infant acute leukemia | infants under 1 year | 41 per million per year 1 |
| Infant leukemia expressed across all children | all children 0-14 | low single digits per million |
| KMT2A-rearranged acute leukemia, all ages | whole population | low single digits per million |
Both the lab's working figure of roughly 3 per million and the 41 per million in the literature can be true at once, because the first spreads the same small number of cases across a much larger population. When quoting rarity in a grant or a paper, state the denominator in the same sentence as the number.
Why the number matters strategically
A disease of ~160 US infants a year will never support a conventional development programme on its own. That is the argument for a platform: the biology of a hijacked chromatin complex is shared across KMT2A partners, across NPM1-mutant AML, and across other fusion families, so one mechanism-directed asset can serve many small populations. Menin inhibition is the proof of concept, since the same drug is active in KMT2A-rearranged leukemia and in NPM1-mutant AML, and KMT2A rearrangements appear in up to 10% of acute leukemias across all ages while NPM1 mutation appears in up to 30% of AML 2.
Outcomes have not moved in decades2 deeper
Infant KMT2A-rearranged ALL has a three-year event-free survival below 40%, and most relapses happen during treatment: two thirds within one year of diagnosis and 90% within two years. Intensifying chemotherapy has not changed this 3. For comparison, children older than one year with ALL now reach long-term event-free survival above 85% 1. This is the gap the project exists to close.
The historical trial numbers
Interfant-99 reported a four-year event-free survival of 47% for infants with KMT2A-rearranged ALL, and the US CCG-1953 study reported a five-year event-free survival of 34% for KMT2A-rearranged infants versus 60% for infants with an intact KMT2A 1. Within Interfant-06, minimal residual disease after induction stratified six-year disease-free survival at 60.2% (negative), 45.0% (intermediate) and 33.8% (high) 4.
The first thing in years that moved the needle
Adding a single post-induction course of blinatumomab to the Interfant-06 backbone in 30 newly diagnosed infants produced no toxic effects meeting the primary endpoint and better outcomes than historical Interfant-06 controls 3. It is a small single-arm study, so the effect size is not established, but it shows the population is not therapeutically immovable.
Sources for this section
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review
Infant acute leukemia occurs at about 41 cases per million in the United States, roughly 160 cases per year (~90 ALL, ~70 AML); 70-80% of infant ALL and about 50% of infant AML carry a KMT2A/MLL rearrangement; Interfant-99 4-year EFS was 47% and CCG-1953 5-year EFS was 34% for KMT2A-rearranged infants versus 60% for germline KMT2A; older children with ALL exceed 85% long-term EFS.
Treatment of infant leukemias: challenge and promise · Brown P · Hematology Am Soc Hematol Educ Program 2013What this evidence cannot show
An educational review from 2013, so the incidence estimate predates the current SEER release and the survival figures predate Interfant-06 and blinatumomab. Rates are US-specific and partner frequencies come from referral cohorts. Use for order of magnitude, not for a precise current rate.
Evidence that argues against this (1)
- Later summaries put infant leukemia nearer 45 per million (~175 US cases per year), so the 41 figure should be quoted as approximate. source 2024
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A rearrangements occur in up to 10% of acute leukemias and NPM1 mutations in up to 30% of AML, and a single menin inhibitor is active across both.
The menin inhibitor revumenib in KMT2A-rearranged or NPM1-mutant leukaemia · Issa GC, Aldoss I, DiPersio J, et al. · Nature 2023What this evidence cannot show
The 10% and 30% figures are framing statements in the introduction rather than the paper's own measurements; the clinical results are from a phase 1 trial in heavily pretreated patients.
Evidence that argues against this (1)
- "Up to 10%" and "up to 30%" are upper bounds from whichever series reported highest, and the two populations barely overlap: NPM1-mutant AML is overwhelmingly an adult disease and KMT2A-rearranged leukaemia is concentrated in infants. One agent being active across both is a fact about the agent, not evidence that the two indications share a patient population or a commercial logic. Journal of Clinical Oncology 2025
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Infant KMT2A-rearranged ALL has 3-year event-free survival below 40% with most relapses on treatment, outcomes have not improved in recent decades, and adding one course of blinatumomab to Interfant-06 chemotherapy in 30 infants caused no protocol-defined toxic effects.
Blinatumomab Added to Chemotherapy in Infant Lymphoblastic Leukemia · van der Sluis IM, de Lorenzo P, Kotecha RS, et al. · New England Journal of Medicine 2023What this evidence cannot show
Single-arm, 30 patients, compared against historical Interfant-06 controls rather than a randomised comparator, median follow-up 26 months. The survival benefit is suggestive, not established.
Evidence that argues against this (1)
- Three claims of different strength in one item. That outcomes have not improved in recent decades is well supported; that adding blinatumomab caused no prohibitive toxicity rests on 30 infants, where zero events is compatible with a true rate near 10%. The survival figure quoted alongside comes from a selected historical comparator rather than a randomised control. Leukemia 2021
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In Interfant-06, end-of-induction minimal residual disease stratified 6-year disease-free survival at 60.2%, 45.0% and 33.8% for negative, intermediate and high MRD.
Clinical Implications of Minimal Residual Disease Detection in Infants With KMT2A-Rearranged Acute Lymphoblastic Leukemia Treated on the Interfant-06 Protocol · Stutterheim J, van der Sluis IM, de Lorenzo P, et al. · Journal of Clinical Oncology 2021What this evidence cannot show
MRD was measurable in 249 of the enrolled infants, so the stratification excludes patients without a trackable marker; DFS is not overall survival.
Evidence that argues against this (1)
- MRD strata are an observational gradient: patients with high end-of-induction MRD differ in biology before any treatment decision, so 60.2/45.0/33.8% measures prognosis rather than what acting on MRD would achieve. The thresholds separating the three strata are also chosen cut-points on a continuous measurement. Statistics in Medicine 2006
What a family actually sees
A baby who is pale, feverish, bruising easily, with a swollen belly and a very high white-cell count. The diagnosis usually arrives within days of the first blood test.
24 entries at this scaleopen
It starts before birth, in the liver
Blood is made in the fetal liver before it moves to the bone marrow. The break that causes infant leukemia happens there, in the womb, and can be found years later on the newborn heel-prick card.
The single most striking fact about this disease: in identical twins where one develops infant leukemia, the other almost always does too, because they shared a circulation and the broken cell crossed between them. The latency is months, not years.
71 entries at this scaleopen
Why the chromosome breaks at all
KMT2A does not break randomly. One short stretch of it is unusually fragile, carrying topoisomerase-II cleavage sites, a nuclease-sensitive region and repetitive elements that invite mistakes.
331 entries at this scaleopen
The protein that comes out keeps the grip and loses the brakes
Breaks cluster in one region, so every fusion looks broadly the same: it keeps the parts that find DNA and hold on to menin, and loses everything from the regulatory fingers through the catalytic domain.
Open the full entry: KMT2A / MLL1 — the gene that breaks19 sources
KMT2A is a very large protein that acts as a foreman on DNA: it finds the right genes and marks them to be switched on during blood development. When the gene breaks in the middle, the surviving front half keeps the ability to find genes but loses the ability to stop, and gains a partner that jams the switch on.
Correction: the bromodomain coordinates were wrong
This entry stored the KMT2A bromodomain at 1703-1809. UniProt Q03164 annotates it at 1635-1765, read directly from the REST record.
This was not cosmetic. As stored, the bromodomain spanned residues 1799-1802 — the HCF-1/HCF-2 binding motif, which is a separate short motif and not part of the bromodomain at all. Any reasoning that treated the HCF contact as bromodomain-mediated, or that drew the two as one block in the domain viewer, was working from a wrong picture.
Also corrected: FYRN now ends at 2074 rather than 2072, and a pre-PHD C2HC zinc finger at 1870-1910 has been added, which UniProt annotates immediately N-terminal to PHD4 and this entry omitted.
What is kept and what is lost1 deeper
Because breaks concentrate between exon 9 and exon 13, the fusion protein's composition is remarkably consistent no matter which partner joins.
Retained on the KMT2A side: the AT-hooks that grip DNA - described in most reviews as AT-rich minor-groove binders, but the only direct binding assay ever done on MLL's own hooks found them binding cruciform DNA, recognising structure rather than sequence 5; corrected 2026-09-18 - the subnuclear localisation signals, the two menin-binding motifs at the extreme N-terminus, and the CXXC domain that binds unmethylated CpG DNA.
Lost: PHD fingers 1-3, the bromodomain, the FYRN and FYRC regions, the transactivation domain, and the catalytic SET methyltransferase domain.
That asymmetry is the whole story. The fusion can still find its target genes but can no longer perform KMT2A's own regulated chemistry, and the partner supplies a transcription-elongation engine instead. The clinically important consequence is that the drug target is in the retained half, so a menin inhibitor works regardless of partner.
The PHD3 switch that gets thrown away
PHD finger 3 is the most interesting casualty. In intact KMT2A it binds either CYP33/PPIE or H3K4me2/3. When CYP33 binds, it isomerises a proline and converts KMT2A from an activator into a repressor by recruiting BMI1 and histone deacetylases. PHD2 and PHD3 also dock the E3 ligases that control how fast KMT2A is degraded.
So the fusion loses a brake, a self-regulating stability timer, and a context switch, all in one module. That is a candidate explanation for why the fusion program never turns itself off, and it is the mechanistic reason breakpoint position at exon 11 might matter 2. Whether it actually changes clinical behaviour is a separate and unsettled question.
Over a hundred partners, but a handful do most of the damage
Systematic sequencing of thousands of patients has catalogued more than 100 KMT2A fusion partners 1. In infant lymphoblastic leukemia the distribution is heavily skewed: AFF1 about 49%, MLLT1 about 22%, MLLT3 about 17% and MLLT10 about 5%. In infant myeloid leukemia the order changes, with MLLT10 about 27%, MLLT3 about 22% and ELL about 17% 3.
The skew is strategically useful. Four partners cover roughly nine in ten infant lymphoblastic cases, so a partner-agnostic target plus four partner-specific backup strategies would cover most of the population.
The gene is not a straightforward drug target itself
KMT2A is not DepMap common-essential, with a mean CRISPR gene-effect score of about -0.21, but it carries the tightest loss-of-function constraint in this whole target set, with a LOEUF of 0.070 4. Human heterozygotes for KMT2A loss have Wiedemann-Steiner syndrome, so removing KMT2A function is not benign.
There is also a practical obstacle for structural work: at 3,969 residues there is no AlphaFold model of full-length KMT2A, and no experimental structure of an intact fusion protein exists either. Any picture of the whole fusion is necessarily a schematic, and the graph marks it as such. Only fragments bound to partners have real coordinates.
Added cycle 89: which cells need KMT2A, and in which state
This entry carried no cell-type and no cell-state field for 71 cycles. Both are now recorded, and they do not say what a reader would guess.
Expression carries no information here. Human KMT2A RNA is Low tissue specificity, Detected in all, with low immune-cell and low blood-lineage specificity [Q127_celltype2]. Nothing about where the gene is transcribed narrows the compartment.
Requirement does. Conditional Mll deletion in mouse makes HSCs and myelo-erythroid progenitors Mll-dependent and leaves committed lymphoid and myeloid cells independent - and the direction of the cell-cycle effect REVERSES at that boundary: quiescence is maintained in HSCs, whose loss causes ectopic cell-cycle entry, while proliferation is promoted in progenitors, whose loss slows them down [Q127_celltype1]. The fetal arm of a second allele agrees on the quiescence half [Q127_cellstate1].
Cell-cycle dependence is a property of the protein and the fusion escapes it. Wild-type MLL is degraded biphasically, by SCF(Skp2) at S phase and APC(Cdc20) at late M, and the tested prevalent MLL fusions are resistant [Q127_cellstate2]. This graph already held that in cell-state-unsynchronisable-cell-cycle; it was not in this entry.
The boundary the sources support is multipotent versus committed, in mouse, and nothing finer. See mechanism-mll1-dependence-ends-at-lineage-commitment and cell-state-hsc-quiescence-maintained-by-mll1.
Added cycle 89: the epigenetics field this entry did not have
The oddest gap in the corpus: an entry for an H3K4 methyltransferase with no epigenetics field. What goes in it is less flattering than the job title.
- Scope of the requirement. In Mll1-null MEFs, Mll1 is required for H3K4me3 at fewer than 5% of the promoters carrying the mark, with the affected set enriched for Hox and other developmental regulators, and MLL3/MLL4 and Set1 complexes do not cover for it there [Q127_epi1].
- Whether the catalysis matters at all is contested. A SET-domain deletion leaves HSC maintenance and MLL-AF9 leukaemogenesis intact, and the mark that actually falls at MLL1 targets on Mll1 deletion is H4K16 acetylation, restored by SIRT1 inhibition - i.e. recruited MOF activity, not intrinsic methyltransferase activity [Q127_epi2]. Against that, MM-401, a WDR5-interface inhibitor of the same activity, selectively kills MLL cells with differentiation and apoptosis and no general marrow toxicity (PMID 24389101, in the counter-evidence of [Q127_epi2]).
- The mark oscillates. H3K4me3 at circadian promoters is rhythmic and MLL1 is required for the rhythm [Q127_epi3]. No occupancy experiment in this project records time of day.
None of this is new to the corpus. PMID 24813891 is cited in 32 other entries and PMID 17908926 in 24. What was new was putting them here. See ana-q127-the-pi-dimensions-were-absent-from-the-entries-not-from-the-corpus.
Added cycle 89: where these additions sit against what this entry already said
Three tensions, stated rather than silently resolved, because this project treats a silent overwrite as worse than a contradiction. No existing sentence in this entry was changed.
- The expert summary opens 'KMT2A ... is a histone H3K4 methyltransferase, and the mammalian Trithorax homolog, essential for HOX gene maintenance in haematopoiesis.' Every clause is defensible. Read together with [Q127_epi1] and [Q127_epi2] it overstates: the enzyme is required for the mark at under 5% of marked promoters, and the SET domain is dispensable for the HSC and MLL-AF9 phenotypes. The summary is not withdrawn; it is now bounded.
- 'KMT2A is not DepMap common-essential, with a mean CRISPR gene-effect score of about -0.21' (E4). True, and a 21-day proliferation screen in cancer lines cannot see a quiescence set-point in a normal stem cell. [Q127_celltype1] and [Q127_cellstate1] are the requirement measurement DepMap cannot make, and they point the other way. The two are not in conflict; they are different instruments and the entry now says so.
- The 'Correction: the bromodomain coordinates were wrong' section stands untouched and nothing added here revises any domain coordinate. The only coordinate touched by the new material is the SET domain, and it was re-verified against live UniProt Q03164 on 2026-09-23 as 3829-3945, which is exactly what this entry already stores.
Sources for this section
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The 2023 KMT2A recombinome analysed 3,401 acute leukaemia patients diagnosed between 2003 and 2022 and, including published data, identifies **107 in-frame KMT2A fusion partners**, with a further 16 out-of-frame rearrangements, 18 patients with no partner gene fused to 5'-KMT2A, two with a 5'-KMT2A deletion and one KMT2A insertion at an ETV6::RUNX1 breakpoint. Breakpoints concentrate overwhelmingly in the major cluster region BCR1, between KMT2A intron 7 and exon 13: **3,336 patients (98%)**, against 47 (1.4%) in the minor BCR2 between intron 20 and exon 24. The seven most frequent partners together with partial tandem duplications account for more than 90% of all KMT2A recombinations; 37 partners occur recurrently and 63 were seen once.
The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023What this evidence cannot show
Read in full from PMC10169636 on 2026-09-19; every figure above is from the abstract and the breakpoint-distribution section, not recalled. This item previously said only 'more than 100 documented fusion partners, catalogued across thousands of patients, with breaks concentrated in a major cluster region' and was surfaced by `scripts/verify-claims.py` as one of the graph's lowest-scoring established claims — vagueness scores low against a specific abstract. A referral-based cohort from one diagnostic centre, so partner frequencies reflect which centres send samples rather than a true population distribution, and the 98% BCR1 figure is conditioned on a rearrangement having been detected by that laboratory's assays.
Evidence that argues against this (2)
- 'More than 100 documented fusion partners' conflates recurrent drivers with single-patient observations, and the tail is longer than the head. This project's own non-elongation-partner enumeration found partners listed in the recombinome with ZERO supporting publications - SEPTIN3 at n=1 and ITSN1 at n=1 - alongside partners with dozens of papers. A partner count also grows with sequencing depth rather than with biology, so the number tracks method as much as it tracks the locus. Leukemia 2023
- Concentration in one cluster region is not homogeneity within it, and the same paper says so: outcome has been linked to breakpoint position inside BCR1, with patients breaking in KMT2A intron 11 faring worse than those breaking upstream, the cut point falling between intron 10 and exon 11. The 63 partners seen only once also mean the '107 partners' figure is dominated by singletons whose biology is uncharacterised. Leukemia 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A comprises 38 exons; exons 11-16 encode the PHD1-PHD2-ePHD3 module, and breaks at or beyond exon 11 destroy its fold, disabling CYP33 binding and the module's dimerisation.
The KMT2A/MLL consensus gene structure: a comprehensive update for research and diagnostic implications · Meyer C, Larghero P, Lopes BA, Marschalek R · Leukemia 2024What this evidence cannot show
The structural consequence of a break past exon 11 is inferred from domain boundaries and prior functional work on PHD3, not demonstrated on isogenic fusion proteins differing only in break position.
Evidence that argues against this (1)
- Exon numbering for KMT2A has been revised more than once — a 99-bp exon 2 and an exon inside intron 21 were added to the consensus structure — so "exon 11" means different positions in different papers and any breakpoint statement tied to an exon number has to name which numbering it uses. That breaks at or beyond exon 11 destroy the PHD fold is a structural inference from domain boundaries rather than a measured loss of CYP33 binding in patient material. Blood Cells, Molecules and Diseases 2010
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review
In infant lymphoblastic leukemia the commonest KMT2A partners are AFF1 (49%), MLLT1 (22%), MLLT3 (17%) and MLLT10 (5%); in infant myeloid leukemia MLLT10 (27%), MLLT3 (22%) and ELL (17%) dominate.
Treatment of infant leukemias: challenge and promise · Brown P · Hematology Am Soc Hematol Educ Program 2013What this evidence cannot show
A 2013 educational review; percentages are of the subset of cases with an identified partner (93% of infant ALL, 66% of infant AML) and come from referral cohorts.
Evidence that argues against this (1)
- Partner frequencies are detection-limited. Cohorts characterised by targeted RT-PCR panels built around the common partners will report those partners most, cryptic insertions can be missed entirely, and the proportions shift between eras as methods change. The ordering is robust; the percentages are not. Leukemia 2001
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
KMT2A is not DepMap common-essential (mean Chronos about -0.21) yet has the strongest loss-of-function constraint among candidate targets examined, with LOEUF 0.070 and pLI approximately 1.0.
DepMap CRISPR (Chronos) gene effect and gnomAD constraint, retrieved via the Open Targets Platform GraphQL API and the gnomAD GraphQL API · Broad Institute DepMap; gnomAD; Open Targets · Databases 2026What this evidence cannot show
Per-line values were retrieved through the Open Targets API and means computed by us; the DepMap release version is not stated by the API. Cell-line essentiality does not predict tissue tolerance, and constraint reflects selection against heterozygous carriers rather than cellular requirement.
Evidence that argues against this (1)
- A gnomAD constraint metric measures intolerance to heterozygous loss over a lifetime from conception, in a population sample. A drug gives partial reversible occupancy in an adult for weeks, so high constraint is not a safety signal for pharmacological inhibition. There is also a tension inside this item: a gene that is not common-essential in DepMap yet carries LOEUF 0.070 is showing exactly the divergence between cellular requirement and organismal constraint that makes constraint a poor proxy for on-target toxicity. gnomAD, Broad Institute 2026
- E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
The bacterially expressed MLL AT-hook domain binds cruciform DNA in a gel mobility shift assay, recognising the structure rather than the sequence of the target DNA; the same study maps a transcriptional activation domain 3' and a repression domain 5' of the MLL breakpoint region, so 11q23 translocations retain the repression domain and the AT-hook DNA-binding region while discarding the activation domain.
11q23 translocations split the "AT-hook" cruciform DNA-binding region and the transcriptional repression domain from the activation domain of the mixed-lineage leukemia (MLL) gene · Zeleznik-Le NJ, Harden AM, Rowley JD · Proceedings of the National Academy of Sciences USA 91(22):10610-4 1994What this evidence cannot show
ABSTRACT ONLY for the binding experiment's details - the wording 'recognizing structure rather than sequence of the target DNA' is the authors' own, quoted from the PubMed abstract; the PMC full text was not read, so the substrate panel, competitor set and affinities behind that conclusion were not independently inspected. 1994 gel-shift biochemistry with a bacterially expressed AT-hook fragment, not the full-length protein and not in cells. This is the ONLY direct binding assay on MLL's own AT-hooks; the much larger literature describing AT-hooks as AT-rich minor-groove binders is about the motif class (Reeves & Nissen consensus peptide, HMGA proteins), not about MLL. Structure preference and AT-richness are also confounded, because AT-rich tracts are the sequences that most readily bend and extrude cruciforms.
Evidence that argues against this (1)
- The generic AT-hook motif is defined as an A.T-rich minor-groove binder, so a class-level prior runs against the structure-recognition reading, and the 1994 result has not been independently replicated with modern methods. Reeves R et al. 1990
- E_kmt2a_structural_coverageestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
Across every PDB entry containing KMT2A (Q03164), only 629 of 3,969 residues - 15.8% - have been solved at atomic resolution, in seven disconnected islands. The breakpoint cluster region (~residues 1100-1445) is 21% covered, and that coverage is entirely the CXXC domain (1143-1214); the residues where the gene is actually broken have never been seen. The two largest unsolved stretches are 1785-2838 and 2870-3753.
PDBe SIFTS best_structures mapping for UniProt Q03164 · EMBL-EBI PDBe API 2026What this evidence cannot show
Computed by the coordinator from the live SIFTS mapping on 2026-09-19 by merging all unp_start/unp_end spans; the number will drift as new entries are deposited. Counts residues present in a deposited model, which is not the same as residues that are ordered and well-resolved - at 6.2 A the cryo-EM span in particular is backbone-level. Isoform numbering follows canonical Q03164.
Evidence that argues against this (1)
- Low coverage is not the same as low knowledge. The unsolved regions are largely predicted to be intrinsically disordered, and AlphaFold declines to model the chain at all at this length, so there may be little fixed structure there to solve. The honest reading is that the breakpoint region has no known fold, not that someone simply has not got round to it. EMBL-EBI AlphaFold DB 2026
- E_q72_kmt2a_loc1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
UNIPROT MODELS THE TWO-FRAGMENT REALITY AND ATTRIBUTES ONLY ONE THIRD OF IT. Fetched 2026-09-21, Q03164 (KMT2A_HUMAN, primary gene KMT2A, 3,969 aa) carries THREE separate SUBCELLULAR LOCATION comments, not one: the full-length protein, 'Nucleus', with three ECO:0000269 attributions (PMID 12482972, 25593309, 34850113); a comment scoped to the molecule 'MLL cleavage product N320', 'Nucleus', WITH NO EVIDENCE CODE AT ALL; and a comment scoped to 'MLL cleavage product C180', 'Nucleus', ALSO WITH NO EVIDENCE CODE, carrying the uncoded note 'Localizes to a diffuse nuclear pattern when not associated with MLL cleavage product N320'. So any entry that reads the first comment and stops has silently dropped the two fragments the protein actually exists as.
UniProtKB entry Q03164 (KMT2A_HUMAN), three molecule-scoped SUBCELLULAR LOCATION comments and feature table · UniProt Consortium · UniProtKB 2026What this evidence cannot show
A database state on a date, not a biological finding. The molecule-scoped comments are the ones that matter for this project and they are exactly the ones with no evidence code - they restate PMID 12482972's result without citing it, so they are curator prose rather than an attributed assertion, and a downstream tool that filters on ECO:0000269 will keep the least specific line and discard the two useful ones. That is the same failure shape as brief rule 69b: the field you read decides the finding.
Evidence that argues against this (1)
- The uncoded C180 note is not merely unattributed, it is CONTESTED. A 2019 study blocking taspase1 cleavage reported MLL1 left uncleaved, more degradation-resistant, with markedly increased genome-wide chromatin occupancy and SUBCELLULAR LOCALISATION UNCHANGED. Those are different perturbations - a clamp-disrupted fragment pair versus prevention of cleavage - but a reader taking UniProt's uncoded note as settled gets no warning that a peer-reviewed paper reports no localisation change in the same pathway. Zhao Z et al. 2019
- E_q72_kmt2a_loc2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
TWO OF THE THREE EXPERIMENTAL PAPERS UNIPROT CITES FOR 'KMT2A IS NUCLEAR' ARE NOT KMT2A LOCALISATION STUDIES. PMID 25593309 is a screen of human bromodomain proteins for DNA-damage-induced RELOCALISATION, in which more than a third of BRD proteins change localisation after damage and whose subject is ZMYND8 and the NuRD complex; PMID 34850113 is a KSHV latency paper about LANA recruiting MLL1 to viral terminal repeats. Both are legitimate observations of a nuclear MLL1 and neither was designed to establish where KMT2A is. The one cited paper that is a KMT2A localisation study is PMID 12482972 [E_q72_kmt2a_loc3].
Screen identifies bromodomain protein ZMYND8 in chromatin recognition of transcription-associated DNA damage that promotes homologous recombination. · Gong F, Chiu LY, Cox B, Aymard F, Clouaire T, Leung JW, Cammarata M, Perez M, Agarwal P, Brodbelt JS, Legube G, Miller KM · Genes Dev 2015What this evidence cannot show
Abstracts read in full; full texts not read, so a localisation panel inside either paper cannot be excluded (brief rule 80). This is a comment on PROVENANCE, not a challenge to the conclusion - KMT2A is nuclear and nobody disputes it. It is recorded because this graph repeatedly reads UniProt attributions as if the cited PMID were the founding experiment, and rule 82's lesson applies in reverse here: a high-profile PMID attached to a claim is not evidence that anyone reasoned about that claim in that paper. NOTE ALSO that 25593309 is a HeLa/U2OS DNA-damage system and 34850113 a herpesvirus latency system - neither is haematopoietic (rule 73).
Evidence that argues against this (1)
- A KSHV latency paper is a stranger source for a localisation code than it looks, but it is not an error: it shows MLL1 recruited to a specific chromatinised DNA template, which is a stronger nuclear demonstration than bulk staining. The criticism to keep is narrow - UniProt's citation set for this field tells you which papers a curator happened to process, not which experiment established the fact. Tan M et al. 2021
- E_q72_kmt2a_loc3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A IS NOT ONE PROTEIN IN THE NUCLEUS, IT IS TWO THAT CLIP TOGETHER. MLL is normally cleaved at two conserved sites (D/GADD and D/GVDD) into N-terminal p320 (N320) and C-terminal p180 (C180), which form a stable complex that LOCALISES TO A SUBNUCLEAR COMPARTMENT; the FYRN domain of N320 binds the FYRC and SET domains of C180; and DISRUPTING THAT INTERACTION markedly decreases N320 levels and REDISTRIBUTES C180 TO A DIFFUSE NUCLEAR PATTERN. The same paper predicts that MLL fusion proteins of leukaemia, which cannot complex with C180, have their stability conferred by the fusion partner instead.
Proteolytic cleavage of MLL generates a complex of N- and C-terminal fragments that confers protein stability and subnuclear localization. · Hsieh JJ, Ernst P, Erdjument-Bromage H, Tempst P, Korsmeyer SJ · Mol Cell Biol 2003What this evidence cannot show
Abstract read in full; full text not read. ALREADY LOAD-BEARING IN THIS GRAPH - cited in 5 entries, and localization-fyrn-fyrc-complex-the-only-module-change-with-a-measured-subnuclear-consequence already reasons about it in detail (brief rule 82: read that entry, do not restate it). The subnuclear compartment is NOT NAMED in the paper and has never been identified, so 'a subnuclear compartment' is the whole of what is known. The experiment is on transfected constructs, not endogenous protein, and not in a leukaemic cell. Crucially for this project: EVERY KMT2A FUSION LACKS C180, so the fusion is permanently in the state this experiment created artificially - which is a prediction, not a measurement, because nobody has fractionated a fusion against wild-type N320.
Evidence that argues against this (1)
- CONTESTED, and the graph must not repeat either side as settled. Preventing taspase1 cleavage altogether left MLL1 uncleaved, more resistant to cycloheximide-chase degradation and markedly INCREASED in genome-wide chromatin occupancy, with SUBCELLULAR LOCALISATION UNCHANGED and bulk H3K4 methylation unchanged. The perturbations differ - clamp disruption versus no cleavage - but the 2019 result also inverts the stability logic: cleavage destabilises MLL1 rather than maturing it. Zhao Z et al. 2019
- E_q72_kmt2a_loc4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
IN U937 AND HeLa UNDER CONDITIONAL OVEREXPRESSION, the MLL AMINO TERMINUS targets the protein to the NUCLEOLUS, the interphase NUCLEAR MATRIX and MITOTIC CHROMOSOMAL SCAFFOLDS where it colocalises with topoisomerase II, and conditionally expressed MLL-AF9 and MLL-AF4 give the same punctate nucleoplasmic-plus-nucleolar pattern as a 50 kDa amino-terminal MLL fragment and as endogenous MLL. THAT IS THE SUBNUCLEAR RECORD FOR KMT2A: a compartment set attributed to the half of the protein every fusion keeps.
The amino terminus targets the mixed lineage leukemia (MLL) protein to the nucleolus, nuclear matrix and mitotic chromosomal scaffolds. · Caslini C, Alarcòn AS, Hess JL, Tanaka R, Murti KG, Biondi A · Leukemia 2000What this evidence cannot show
Abstract read in full; full text not read. CITED 11 TIMES IN THIS GRAPH already - it is one of the corpus's load-bearing localisation papers (rule 82), and localization-nucleolus-kmt2a-fusion, localization-nuclear-matrix-mitotic-scaffold-kmt2a and localization-kmt2a-mllt3-the-address-is-a-kmt2a-property all reason about it. NOT A LEUKAEMIC B-LINEAGE SYSTEM: U937 is monocytic and HeLa is cervical carcinoma, and the fusions are CONDITIONALLY EXPRESSED constructs, not endogenous t(4;11) protein (rule 73). 'Nuclear matrix' is an operational extraction fraction. Because the pattern is indistinguishable between endogenous MLL, the N-terminal fragment and two different fusions, the experiment cannot attribute any compartment to the fusion as opposed to the KMT2A moiety - which is the paper's own point and is the reason this graph should spend no confidence on 'the fusion is nucleolar'.
Evidence that argues against this (1)
- The endogenous experiment says the picture cannot tell the species apart. Butler and colleagues resolved chimeric HRX proteins at the predicted fusion sizes on a blot from t(4;11) and t(11;19) cells - the size dimension doing the attribution - then imaged the SAME reagents and found punctate nuclear staining of wild-type and chimeric HRX alike, 'in cells with and without 11q23 translocations', concluding in their own words that this 'precludes an immunocytochemical approach for diagnosis'. So the punctate-nuclear description of KMT2A and of its fusions rests on images that their own authors said are unattributable. Butler LH et al. 1997
- E_q72_kmt2a_loc5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
THE NUCLEAR ADDRESS IS CARRIED BY THE PART EVERY FUSION KEEPS, WHICH IS WHY 'THE FUSION IS NUCLEAR' IS NOT A FUSION PHENOTYPE. Asked in the hardest case - MLL-AF6, whose partner AFDN/AF6 is a cytoplasmic Ras-binding protein - immunofluorescence AND cell fractionation showed the chimera is nuclear, and the region of the MLL N-terminus responsible was mapped to A REGION CONTAINING THE AT-HOOK MOTIFS. UniProt Q03164 annotates NO nuclear-localization-signal motif on KMT2A at all (fetched 2026-09-21); the five motifs it does annotate are MBM, IBM1, IBM2, 9aaTAD and WIN.
Chimeric MLL products with a Ras binding cytoplasmic protein AF6 involved in t(6;11) (q27;q23) leukemia localize in the nucleus. · Joh T, Yamamoto K, Kagami Y, Kakuda H, Sato T, Yamamoto T, Takahashi T, Ueda R, Kaibuchi K, Seto M · Oncogene 1997What this evidence cannot show
Abstract read in full; full text not read. 1997 deletion-construct mapping in transfected cells, not endogenous protein and not a leukaemic B-lineage system (rule 73). The mapping localises a SUFFICIENT region and does not distinguish active nuclear IMPORT from nuclear RETENTION by chromatin binding - localization-nuclear-import-determinant-maps-to-the-at-hook-region already records that ambiguity and is the entry to read. The absent NLS feature in UniProt is consistent with this: the determinant is a region, not a recognisable signal sequence.
Evidence that argues against this (1)
- One partner does move something. Expression of MLL-AF4 can ALTER THE SUBNUCLEAR LOCALISATION OF AF9 (PMID 14603337), and MLL-AF6-rearranged cells relocate endogenous AF6 from cytoplasm to nucleus (PMID 24695851, cited elsewhere in this graph). So the KMT2A half dominating the fusion's own address does not mean the fusion has no localisation phenotype - it means the phenotype to look for is in TRANS, on partner proteins, and that is where the only positive results are. Erfurth F et al. 2004
- Q127_celltype1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Committed lymphoid and myeloid cells no longer require Mll, defining the early multipotent stages of haematopoiesis as Mll-dependent; within those stages Mll maintains quiescence in HSCs, whose loss causes ectopic cell-cycle entry and depletion of the quiescent pool, and promotes proliferation in myelo-erythroid progenitors, whose loss reduces proliferation and blunts the response to cytokine-induced cell-cycle entry.
Unique and independent roles for MLL in adult hematopoietic stem cells and progenitors. · Jude CD, Climer L, Xu D, Artinger E, Fisher JK, Ernst P · Cell Stem Cell 2007What this evidence cannot show
Mouse, inducible conditional knockout of the whole Mll locus - a statement about wild-type MLL1, not about a fusion allele. ABSTRACT READ ONLY (PMC2234224 returned an empty full-text body on 2026-09-23), so no percentage, assay or n is quoted. The PubMed MCP abstract for this record silently deletes every italicised 'Mll'; the symbols here were restored by hand from the title and the PubMed web record. This paper had ZERO mentions anywhere in this corpus before cycle 89.
Evidence that argues against this (1)
- A second conditional Mll allele published in the same issue reports GROSSLY NORMAL steady-state haematopoiesis in bone marrow, thymus and spleen, against this paper's fatal bone marrow failure within 3 weeks. The quiescence direction agrees across the two; the severity does not. See concept-two-founding-mll-conditional-knockouts-disagree-about-steady-state-blood. McMahon KA et al. 2007
- Q127_cellstate1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Mll-null embryos survive to E16.5 with reduced HSC numbers and a greatly reduced quiescent fraction, and those HSCs cannot compete in transplantation; adult conditionally Mll-deleted bone marrow is highly compromised in competitive reconstitution despite grossly normal steady-state haematopoiesis.
Mll has a critical role in fetal and adult hematopoietic stem cell self-renewal. · McMahon KA, Hiew SY, Hadjur S, Veiga-Fernandes H, Menzel U, Price AJ, Kioussis D, Williams O, Brady HJM · Cell Stem Cell 2007What this evidence cannot show
Mouse; abstract read, full text not opened. 'Greatly reduced' is the authors' wording and no percentage appears in the abstract. The fetal result is a constitutive null and the adult result a conditional deletion, so the two halves are not the same experiment. ZERO mentions anywhere in this corpus before cycle 89.
Evidence that argues against this (1)
- The paper published immediately before this one in the same issue, from a different conditional Mll allele, reports FATAL BONE MARROW FAILURE within three weeks of deletion rather than grossly normal steady-state haematopoiesis. The two agree on loss of quiescence and on competitive failure and disagree on whole-animal severity. Jude CD et al. 2007
- Q127_cellstate2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Wild-type MLL protein is degraded biphasically through the cell cycle, by SCF(Skp2) at S phase and APC(Cdc20) at late M, giving two expression peaks; abolishing the peaks produces G1/S and M-phase defects and overexpression blocks S-phase progression; degradation initiates within the N-terminal approximately 1,400 amino acids, and the tested prevalent MLL fusions are RESISTANT to that degradation.
Bimodal degradation of MLL by SCFSkp2 and APCCdc20 assures cell cycle execution: a critical regulatory circuit lost in leukemogenic MLL fusions. · Liu H, Cheng EH-Y, Hsieh JJ-D · Genes Dev 2007What this evidence cannot show
HeLa and leukaemia lines; abstract read, full text not opened. Cited in 24 entries of this corpus already, and in none of them was this entry - the conclusion is not new to the graph, only to gene-kmt2a. The fusion-resistance claim is about 'tested prevalent MLL fusions' and the abstract does not enumerate them; cell-state-unsynchronisable-cell-cycle holds the detail, including that MLL-AF4 protein does not fluctuate through the cycle in SEM.
Evidence that argues against this (1)
- Cell-cycle control of MLL abundance is not the only cell-cycle circuit this protein sits in, and the two are sometimes conflated: taspase1 cleavage of MLL and MLL2 is separately required for cell-cycle progression, with MLL targeting E2Fs for cyclin expression and H3K4 trimethylation reduced at Cyclin E1 and E2 in Taspase1-null cells. An entry that records only the degradation circuit understates how much of the fusion's cell-cycle abnormality is attributable to loss of processing rather than loss of degradation. Takeda S et al. 2006
- Q127_epi1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In Mll1-null versus wild-type mouse embryonic fibroblasts, Mll1 is required for H3K4 trimethylation of LESS THAN 5% of the promoters carrying that modification; the affected genes are enriched for developmental regulators including Hox genes and show matched decreases in RNA polymerase II recruitment and expression, while loss of MLL3/MLL4 and/or the Set1 complexes has little to no effect on H3K4 methylation of Hox loci.
Global analysis of H3K4 methylation defines MLL family member targets and points to a role for MLL1-mediated H3K4 methylation in the regulation of transcriptional initiation by RNA polymerase II. · Wang P, Lin C, Smith ER, Guo H, Sanderson BW, Wu M, Gogol M, Alexander T, Seidel C, Wiedemann LM, Ge K, Krumlauf R, Shilatifard A · Mol Cell Biol 2009What this evidence cannot show
Mouse embryonic fibroblasts - not haematopoietic, no fusion. The <5% figure is a promoter-level call on 2009-era platforms and its denominator is not given in the abstract; abstract read, full text not opened. This bounds what 'KMT2A is an H3K4 methyltransferase' means in a cell; it does not say the enzyme is unimportant. ZERO mentions anywhere in this corpus before cycle 89.
Evidence that argues against this (1)
- A pharmacological experiment points the other way in the cells this project cares about: MM-401, which blocks MLL1's H3K4 methyltransferase activity via the WDR5 interface, blocks MLL-cell proliferation with arrest, apoptosis and myeloid differentiation and no general marrow toxicity. An activity required at under 5% of promoters in fibroblasts is nonetheless a viable target in MLL-rearranged cells, so the bound must not be read as 'the enzyme does not matter'. Cao F et al. 2014
- Q127_epi2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The SET domain, and thus the histone methyltransferase activity of MLL1, is dispensable for maintaining haematopoietic stem cells and for supporting MLL-AF9-driven leukaemogenesis; on Mll1 deletion it is H4K16 ACETYLATION that is selectively depleted at MLL1 target genes alongside reduced transcription, recruited MOF activity rather than intrinsic MLL1 methyltransferase activity maintains those genes, and SIRT1 inhibition is sufficient to prevent the loss.
The histone methyltransferase activity of MLL1 is dispensable for hematopoiesis and leukemogenesis. · Mishra BP, Zaffuto KM, Artinger EL, Org T, Mikkola HKA, Cheng C, Djabali M, Ernst P · Cell Reports 2014What this evidence cannot show
Mouse; abstract read, full text not opened. 'Dispensable' means a SET-deletion allele was not limiting in those two assays; a SET deletion can perturb folding as well as catalysis. This PMID is cited in 32 entries of this corpus - including subtraction-kmt2a-set-domain, mark-h4k16ac and mechanism-set-loss-neomorph-not-haploinsufficiency - and was in none of gene-kmt2a. It is the single most-cited source this entry was missing.
Evidence that argues against this (1)
- CONTESTED BY A PHARMACOLOGICAL EXPERIMENT. MM-401, which blocks the MLL1-WDR5 interaction and so MLL1 H3K4 methyltransferase activity specifically, blocks proliferation of MLL cells with cell-cycle arrest, apoptosis and myeloid differentiation and no general toxicity to normal bone marrow or non-MLL cells, with a transcriptome change resembling MLL1 deletion. A SET deletion and an assembly blockade are different perturbations, so the two results may both stand, but this graph must not state 'the activity is dispensable' unqualified. Cao F et al. 2014
- Q127_epi3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
H3K4 trimethylation at circadian promoters is rhythmic and follows the profile of circadian H3 acetylation; MLL1 is essential for circadian transcription and for cyclic H3K4 trimethylation, is in a complex with CLOCK-BMAL1, contributes to its rhythmic recruitment, and fails to interact with CLOCK-delta19.
The histone methyltransferase MLL1 permits the oscillation of circadian gene expression. · Katada S, Sassone-Corsi P · Nat Struct Mol Biol 2010What this evidence cannot show
Mouse and cultured cells, wild-type MLL1, no leukaemia and no fusion; abstract read, full text not opened; single paper with no replication identified in this beat, hence moderate. Recorded because UniProt and HPA both list 'Biological rhythms' among KMT2A's biological processes and this corpus of 2,383 entries had ZERO mentions of this paper and no circadian content about KMT2A at all.
Evidence that argues against this (1)
- Bounded by the same study that limits MLL1's H3K4me3 requirement to under 5% of promoters: a rhythm requiring MLL1 is a rhythm at a small promoter set, and nothing shows that any leukaemia-relevant locus, HOXA included, is in it. The relevance of the circadian result to this project is therefore a hypothesis, not a finding. Wang P et al. 2009
- Q127_celltype2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
For KMT2A (ENSG00000118058) the Human Protein Atlas records RNA tissue specificity 'Low tissue specificity' with distribution 'Detected in all', blood cell specificity 'Low immune cell specificity', blood lineage specificity 'Low lineage specificity' and single-cell 'Low cell type specificity'; the antibody-based data call it Tissue enriched with bone marrow the top tissue and Cell type enriched with hepatocytes the top cell type, on antibodies CAB017794, CAB024270 and HPA044910 with IH reliability 'Enhanced' and IF reliability 'Supported', subcellular main location Nucleoplasm with additional Cytosol.
Human Protein Atlas - KMT2A (ENSG00000118058), machine-readable gene record · Human Protein Atlas consortium · proteinatlas.org 2026What this evidence cannot show
Fetched 2026-09-23 from the per-gene JSON endpoint, which does NOT return the HPA release version, so this row can drift silently and must be refetched before any value is requoted. RNA calls are consensus transcriptomics and protein calls are antibody staining scored by eye - different instruments. The hepatocyte call has no support in the haematopoietic literature and points away from the bone-marrow call; two discordant calls from one instrument are an antibody signature, not a biological one. Staining intensities are arbitrary units, not concentrations.
Evidence that argues against this (1)
- HPA's own two instruments disagree for this gene. The RNA data say Low tissue specificity and Detected in all; the antibody data say Tissue enriched in bone marrow and Cell type enriched in hepatocytes. The hepatocyte call has no support in the literature this project reads, and the safest reading is that the antibody-based tissue calls should not be used as evidence of abundance at all - including the bone-marrow one that happens to agree with the genetics. Human Protein Atlas consortium 2026
- Q127_cellstate3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The Mll-dependent transcriptional network in haematopoietic stem cells extends well beyond clustered Hox genes, contains many characterised regulators of self-renewal, and comprises target genes that are both dependent on and independent of the MLL cofactor menin; PRDM16 was uniquely effective at partially rescuing Mll-deficient haematopoietic stem and progenitor cells.
An MLL-dependent network sustains hematopoiesis. · Artinger EL, Mishra BP, Zaffuto KM, Li BE, Chung EKY, Moore AW, Chen Y, Cheng C, Ernst P · Proc Natl Acad Sci U S A 2013What this evidence cannot show
Mouse; abstract read, full text not opened. The menin-dependent versus menin-independent partition is the load-bearing part for this project and the abstract does not name which genes fall on which side, so no gene-level assignment may be taken from here. Cited once in this corpus before cycle 89 (collab-ernst-colorado) and never in gene-kmt2a.
Evidence that argues against this (1)
- The menin-independent half of this network is the part that matters for the drug and this study does not enumerate it. The menin knockout itself reports that HoxA9 expression falls only MODESTLY in menin-deficient HSCs, which suggests the menin-dependent portion of the normal HSC programme is smaller than the fusion-centric account assumes. Maillard I et al. 2009
So we drug what it holds on to
The first dozen amino acids of KMT2A plug into a pocket on menin. That grip is present in every fusion whatever the partner, which makes it a target that does not care which translocation a child has.
Two drugs that fill this pocket are now approved. That is the proof the strategy works, and also the start of the harder problem, because the disease has already shown us how it escapes.
Open the full entry: The menin–KMT2A grip (MBM1), the interface the drugs actually block4 sources
The first dozen amino acids of KMT2A plug into a pocket on menin like a key in a lock. Every approved and experimental menin drug works by filling that pocket so the key cannot go in.
The numbers that matter1 deeper
Affinities measured by isothermal titration calorimetry and fluorescence polarisation 1:
| Peptide | Affinity (Kd) | Method |
|---|---|---|
| KMT2A 4-15 (MBM1) | 53 nM | fluorescence polarisation |
| KMT2A 4-15 (MBM1) | 72 nM | calorimetry |
| KMT2A 23-40 (MBM2) | 1,400 nM | fluorescence polarisation |
| KMT2A 1-46 | 6.8 nM | calorimetry |
| KMT2A 1-160 | 9.8 nM | calorimetry |
| KMT2A 1-160 with F9A | 1,550 nM | calorimetry |
Two things fall out of this table. First, Phe9 is the hot spot: a single alanine substitution costs about 160-fold in affinity, which is why the pocket that accepts it is the one the chemistry targets. Second, MBM1 and MBM2 are negatively cooperative, since the full 1-46 fragment binds at 6.8 nM rather than the roughly 3 nM two independent sites would predict.
Structures you can actually load
Every identifier here was resolved against the RCSB API rather than recalled.
- 4GQ6 menin with KMT2A 6-15, 1.55 Å. The canonical MBM1 complex.
- 3U85 menin with KMT2A 6-25, 3.0 Å; 3U88 the menin-KMT2A-LEDGF ternary complex, 3.0 Å.
- 7UJ4 (1.96 Å) and 8E90 (1.85 Å) menin with revumenib.
- 8VA6 menin with ziftomenib (1.57 Å); 8VA5 the T349M mutant (1.30 Å).
- 9Z4X menin with bleximenib, 2.95 Å.
- 9C4S / 9C4T / 9C4U / 9C4V menin-KMT2A with the G331R, M327I, T349M and G331D resistance mutants, 1.46-1.57 Å.
There is no structure of menin bound to MBM2. That motif is defined biochemically only.
A correction worth recording, because it propagates through reviews: 4GQ3 and 4GQ4 are menin with the inhibitors MI-2 and MI-2-2, not with an MBM1 peptide, and they come from Shi et al. in Blood, not from Huang et al. in Nature. 6E1A is menin with compound M-89, not revumenib. There is no AlphaFold model of full-length KMT2A, since at 3,969 residues it exceeds the database's full-chain coverage.
Why this interface is the right first target, and where the argument is thin
It is retained in every KMT2A fusion regardless of partner, it is a compact well-formed pocket rather than a flat surface, it has validated chemical matter with two approved drugs, and blocking it evicts the fusion from chromatin rather than merely slowing it.
The thin part: this pocket is not KMT2A-specific. The same menin cavity binds JUND, with opposite transcriptional output, which is captured structurally in 3U86 2. Anything that fills the pocket also displaces whatever else uses it, and the consequences of blocking menin-JUND have not been worked out.
Update: the JUND question has been answered, and the answer is unwelcome
When this entry was first written it flagged an open question: the menin pocket is not KMT2A-specific, it also binds JUND, and nobody had worked out what displacing JUND does.
It has now been worked out, and the answer is a liability. Menin inhibitors disrupt menin-JUND as well as menin-KMT2A, which increases JUND occupancy on chromatin, upregulates its targets, and contributes to resistance. Removing JUND genetically, or inhibiting JNK pharmacologically, synergises with menin inhibition 4.
This matters beyond one drug. The liberation of a second factor is a consequence of occupying a shared pocket, so it is plausibly intrinsic to this binding site rather than to any particular molecule. Better potency does not fix it. It also suggests a concrete combination: a menin inhibitor plus a JNK inhibitor, which the paper's own data supports.
The practical read for this project: when you drug a pocket that serves two masters, you should expect to find out what the other master does when it is set free.
Sources for this section
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A MBM1 (residues 4-15, core RWRFP at 6-10) binds menin with Kd 53 nM by fluorescence polarisation and 72 nM by calorimetry; MBM2 (residues 23-40) binds at about 1,400 nM; Phe9 is a hot spot whose mutation costs roughly 160-fold; MBM1 and MBM2 are negatively cooperative.
Molecular basis of the mixed lineage leukemia-menin interaction: implications for targeting mixed lineage leukemias · Grembecka J, Belcher AM, Hartley T, Cierpicki T · Journal of Biological Chemistry 2010What this evidence cannot show
Affinities are for isolated synthetic peptides against recombinant menin in vitro, not for the full fusion protein on chromatin, where avidity from CXXC-DNA binding and LEDGF contacts will dominate. Calorimetry and fluorescence polarisation values differ by roughly 35% for the same peptide, which sets the realistic precision. Citation correction 2026-09-19: the title field previously read 'Menin-MLL inhibitors reverse oncogenic activity of MLL fusion proteins in leukemia (and the associated menin-binding-motif characterisation)'. That is a descriptive composite, not a title, and it leads with the exact title of a DIFFERENT real paper. The PMID and DOI were always correct and match the claim; only the title field was wrong. Three other entries cite the same PMID under its correct title, so this was isolated to this entry.
Evidence that argues against this (2)
- A 53 nM grip that drugs demonstrably break is not the same as a grip that matters once broken. In the only human sample profiled under revumenib, the oncoprotein AND menin fell to UNDETECTABLE while ENL persisted at oncoprotein target loci - the interface was fully disengaged and the programme kept running. That is worse for the hypothesis than a resistance mutation, which at least concedes the target still matters. Nature Communications 2024
- The clinical ceiling argues the same way from the other end. In AUGMENT-101 phase 2, blocking this interface with revumenib produced CR+CRh in 22.8% of 57 efficacy-evaluable patients. Overall response was 63.2%, so the pocket is plainly engaged and the biology plainly moves - but under a quarter reach the endpoint that matters, in a disease where the interface is supposed to be the dependency. Journal of Clinical Oncology 2025
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The same menin pocket that binds the KMT2A MBM1 motif also binds JUND, with opposite transcriptional consequences, so the pocket is not KMT2A-selective.
The same pocket in menin binds both MLL and JUND but has opposite effects on transcription · Huang J, Gurung B, Wan B, et al. · Nature 2012What this evidence cannot show
Structural and biochemical work; the in vivo consequence of displacing JUND with a menin inhibitor in a patient has not been measured.
Evidence that argues against this (1)
- The shared-pocket objection predicts on-target harm from displacing menin's other partners, and that prediction has not been borne out in the way it was feared. Revumenib was approved on this cohort with a toxicity profile dominated by differentiation syndrome and QT prolongation (grade 3/4 QT 12%) - mechanism-linked and manageable - rather than by the consequences of displacing JUND or menin's tumour-suppressor functions. A non-selective pocket has proved more tolerable than its non-selectivity predicted. Annals of Pharmacotherapy 2025
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Structures of menin with revumenib (7UJ4, 8E90) and with menin interface resistance mutants show that the mutations impair inhibitor binding while preserving MLL1 binding.
MEN1 mutations mediate clinical resistance to menin inhibition · Perner F, Stein EM, Wenge DV, et al. · Nature 2023What this evidence cannot show
Structures are of menin fragments with peptides and small molecules, not of an intact fusion protein engaged with chromatin.
Evidence that argues against this (2)
- Interface resistance mutations are not the class-wide dead end this framing suggests. The mutants selected by revumenib are not uniformly cross-resistant: ziftomenib retained antileukaemic activity against T349M and low-nanomolar potency against G331R. The interface can survive a mutation that defeats one ligand, which argues the pocket is a more durable target than a single drug's resistance spectrum implies. Blood 2026
- And MEN1 mutation is not the only route out. TP53 inactivation confers de novo revumenib resistance with impaired induction of TP53 targets and deregulated BCL-2/BH3 proteins - resistance that never touches the interface at all, so no amount of interface engineering addresses it. CAVEAT: this is a bioRxiv PREPRINT and has not been peer reviewed. bioRxiv (PREPRINT, not peer reviewed) 2025
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Menin inhibitors also disrupt the menin-JUND interaction, increasing JUND occupancy on chromatin and upregulating its target genes, which contributes to resistance; inactivating JUND or inhibiting JNK synergises with menin inhibition.
Menin-MLL inhibitors enhance JUND activity in MLLr leukemic cells contributing to tumorigenesis and therapy resistance · Ozyerli-Goknar E, Nizamuddin S, Fetsch V, et al. · Blood 2026What this evidence cannot show
Cell lines and immunocompromised mouse xenografts, not patients. Whether JUND liberation contributes to clinical resistance in treated children is untested. One author is employed by the drug's sponsor, which is disclosed rather than disqualifying.
Evidence that argues against this (1)
- The JUND resistance mechanism is demonstrated in cell lines and xenografts; the clinically documented route is a different one. Acquired MEN1 mutations were described in 39% of post-revumenib relapses, and notably have NOT been identified following ziftomenib or bleximenib. Whether JUND liberation contributes measurably in treated patients, and whether it differs across the class, is untested - so this should not yet displace MEN1 mutation as the resistance mechanism to design against. Hematology Am Soc Hematol Educ Program 2025
And then the biology bites back
Patients on a menin inhibitor develop mutations in the exact pocket the drug occupies. Those mutants block the drug while keeping the natural KMT2A grip intact, so the cancer program restarts.
Anticipating this is the point of the whole graph. Every target we record carries its essentiality, its normal-tissue expression, and an explicit list of ways it could come back to hurt us.
126 entries at this scaleopen
Atlas
Every fusion, every break, side by side.
Mechanism
Why the fusion causes the disease, step by step.
Targets
Every druggable surface, ranked by whether it is safe to hit.
Hypotheses
What we think but cannot yet show, with the experiment that would settle it.
Graph
The whole knowledge graph, explorable in three dimensions.
Evidence
Every source in the graph, with its confidence and caveats.
Impact
How many children, and why a platform makes sense.
Groups
Who else works on this, and who to partner with.
How this is built
The autonomous research loop behind the graph.