Oncofusion

start / graph / mechanism-hoxa-high-vs-hoxa-low

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q3-mechanism-core

HOXA-high versus HOXA-low KMT2A-rearranged leukemia — the split that breaks the one-mechanism story

also called HOXA-positive vs HOXA-negative · HOXA-low t(4;11) · HOXA subgroups in KMT2A-r

Not every child with the same KMT2A fusion has the same disease. Some cases run the HOXA gene program hard; others barely express it at all, and those tend to be the youngest infants with a particular breakpoint.

The same thing, for a scientist

Even within a single fusion, t(4;11)/KMT2A::AFF1, cases split into a HOXA-high and a HOXA-low group, and the split tracks with age and with breakpoint position. In the best-characterised analysis, the HOXA-low group comprised infants with KMT2A breakpoints in introns 11/12, while the HOXA-high group comprised non-infants with breakpoints in introns 9/10. This is a direct problem for any model in which the causal chain terminates in a HOXA9/MEIS1 lock, and it is a direct problem for menin inhibition if HOXA9/MEIS1 suppression is the pharmacodynamic readout being used to judge response.

Go deeper
The primary observation1 deeper

Working within the single entity t(4;11) ALL, per-patient expression profiles were normalised against normal bone marrow and a core signature of 186 commonly deregulated genes present in every patient was defined. Linking each patient's gene set to four biological discriminators — HOXA expression, age at diagnosis, fusion transcript, and breakpoint position — split the cohort into two groups:

  • Infants with low HOXA expression and KMT2A breakpoints in introns 11/12
  • Non-infants with high HOXA expression and KMT2A breakpoints in introns 9/10 1

The authors' own framing is the important part: 'a yet homogeneous leukemia entity was further subdivided, based on distinct genetic properties.'

How this connects to breakpoint biology already in the graph

This graph already contains breakpoint-kmt2a-intron-9, breakpoint-kmt2a-intron-11, breakpoint-kmt2a-major-bcr, breakpoint-kmt2a-minor-bcr and mech-breakpoint-position-consequences from the Q2 beat. The HOXA split is the functional phenotype that those breakpoint distinctions predict: intron 11/12 breaks give a longer retained KMT2A segment than intron 9/10 breaks, and they co-segregate with infant age and with HOXA-low biology.

The causal direction is unresolved and that matters. Three readings are all compatible with this dataset:

  1. Breakpoint causes it — the extra retained KMT2A sequence changes the complex and its output.
  2. Age/cell-of-origin causes it — infant (fetal-derived) progenitors have a different methylome and chromatin state, so the same fusion reads different loci (exactly what step 1 predicts), and the breakpoint association is incidental to when the break happened.
  3. Both are downstream of the same fetal exposure window.

Reading 2 is the one this Queen finds most mechanistically natural, given that fusion targeting is a read-out of the pre-existing epigenome rather than a sequence code. It is also the most testable: the prediction is that HOXA status should be predictable from the pre-leukemic methylome/chromatin state, not from the breakpoint, once age is controlled.

Why this threatens the project's central hypothesis

The menin-interface strategy does not require HOXA9 to be the output — menin inhibition evicts the whole complex regardless of what it is transcribing. So HOXA-low disease is not by itself an argument against menin inhibition.

But it creates three concrete risks worth stating plainly:

  1. Wrong biomarker. If HOXA9/MEIS1 suppression is used as the pharmacodynamic proof of target engagement, HOXA-low cases will look like non-responders even when the drug is working, or vice versa. Response assessment must be anchored on fusion-complex eviction from chromatin, not on HOXA transcripts.
  2. Different output, different escape. A HOXA-low leukemia is sustained by some other output program. Nobody has published what that program is. Whatever it is, it is a candidate resistance pathway that is invisible in every HOXA-centric model system.
  3. Model selection bias. The workhorse models — MOLM-13, MV4-11, SEM, SEMK2, RS4;11 and the Mll-AF9/Mll-AF4 mice — are HOXA-high. The HOXA-low infant phenotype, which is the population the Leukemia Foundation grant is about, is essentially unmodelled.

It is also worth flagging honestly that the primary evidence here is one 2009 microarray study on a small rare-disease cohort. The observation has been widely repeated in review form, but this Queen did not verify an independent primary replication with modern sequencing. Treat the existence of the split as moderate confidence, and the specific intron-11/12-versus-9/10 mapping as moderate-to-weak until replicated.

Facts and figures
chainmechanism-kmt2a-fusion-causal-chain
core signature gene count186
causal directionunresolved: breakpoint-driven vs cell-of-origin-epigenome-driven vs both downstream of fetal timing
biomarker implicationHOXA9/MEIS1 transcripts are an unsafe pharmacodynamic readout across the whole KMT2A-r population
subgroups
[
  {
    "label": "HOXA-low",
    "age": "infant",
    "kmt2a_breakpoint": "intron 11/12",
    "retained_kmt2a": "longer",
    "modelled": false
  },
  {
    "label": "HOXA-high",
    "age": "non-infant",
    "kmt2a_breakpoint": "intron 9/10",
    "retained_kmt2a": "shorter",
    "modelled": true
  }
]
discriminators
[
  "HOXA expression",
  "age at diagnosis",
  "fusion transcript",
  "KMT2A breakpoint intron"
]
related nodes
[
  "breakpoint-kmt2a-intron-9",
  "breakpoint-kmt2a-intron-11",
  "mech-breakpoint-position-consequences"
]
How this could bite us back
  • Using HOXA9/MEIS1 suppression as the pharmacodynamic endpoint could cause a working drug to be judged inactive in HOXA-low infants, or an inactive drug to look active — a trial-design failure mode, not a biology one.
  • Every standard KMT2A-r model is HOXA-high, so preclinical efficacy data systematically over-represent the subgroup that is not the grant's target population.
  • If HOXA-low disease runs an unknown alternative program, that program is a pre-existing, untested escape route from any therapy validated only in HOXA-high models.
Still open
  1. What output program sustains HOXA-low KMT2A::AFF1 infant leukemia? This is, in this Queen's view, the single largest unfilled hole in KMT2A mechanism — the population the grant targets is the population with no published mechanism and no model.
  2. Is the HOXA split caused by the breakpoint or by the cell-of-origin epigenome? Age and breakpoint are confounded in the only primary dataset. Re-analysis of a modern infant-ALL cohort with both breakpoint calls and methylation arrays could separate them.
  3. Do HOXA-low cases respond to menin inhibitors? If response is preserved, that is a strong argument that the drug works upstream of the output program and a strong argument against HOXA-based pharmacodynamics.
  4. Has the intron-11/12 versus 9/10 mapping ever been replicated with RNA-seq in an independent cohort? This Queen could not verify a replication, which is surprising for a claim this widely repeated.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Within pediatric t(4;11) ALL, a core signature of 186 commonly deregulated genes and four biological discriminators divided patients into two subgroups: infants with low HOXA expression and KMT2A breakpoints in introns 11/12, and non-infants with high HOXA expression and KMT2A breakpoints in introns 9/10.

    Two independent gene signatures in pediatric t(4;11) acute lymphoblastic leukemia patients · Trentin L, Giordan M, Dingermann T, Basso G, Te Kronnie G, Marschalek R · European Journal of Haematology 83(5):406-19 2009
    PMID 19558506DOI
    What this evidence cannot show

    Small rare-disease cohort on 2009-era microarrays, with the signature derived by pairwise subtraction against only three normal bone-marrow samples. The paper's stated purpose was methodological (how to get robust signatures from small cohorts), so the biological subgrouping is a secondary finding. Age and breakpoint are confounded with each other and cannot be separated in this design. No independent primary replication with RNA-seq was verified in this cycle.

    Evidence that argues against this (1)
    • A 186-gene signature and four discriminators derived in one paediatric t(4;11) cohort is a model with far more parameters than the cohort can constrain, and the two subgroups it defines are separated by age — which correlates with cell of origin, breakpoint position and HOXA output simultaneously. No independent validation cohort is cited. Oncofusion knowledge graph (internal analysis) 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Fusion targeting is a read-out of the pre-existing DNA-methylation and chromatin landscape rather than a sequence-specific code, providing the mechanistic route by which the same fusion in an infant (fetal-derived) versus a non-infant progenitor could produce different HOXA output.

    PMID 15542854DOI
    What this evidence cannot show

    This supports the mechanism by which an epigenome-dependent target set is possible; it does not itself show that HOXA-low disease arises this way. The inference from CXXC biology to the clinical HOXA split is ours and is speculative.

    Evidence that argues against this (1)
    • A read-out of the pre-existing landscape and a sequence-specific code are not exhaustive alternatives, and the evidence does not separate them: KMT2A's own CXXC domain binds unmethylated CpG, which is sequence-dependent targeting operating THROUGH the methylation landscape. Framing it as landscape-rather-than-code understates what the retained KMT2A portion contributes. Cell Reports 2014
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    HOXA9 is required for survival of established human KMT2A-rearranged leukemia, and HOXA9 suppression co-downregulates HOXA10, MEIS1, PBX3 and MEF2C, making HOXA transcript levels the field's default pharmacodynamic readout of fusion activity.

    HOXA9 is required for survival in human MLL-rearranged acute leukemias · Faber J, Krivtsov AV, Stubbs MC, et al. (Armstrong SA, senior author) · Blood 113(11):2375-85 2009
    PMID 19056693DOI
    What this evidence cannot show

    All lines used were HOXA-high; the paper cannot speak to HOXA-low disease, which is precisely the gap this node identifies.

    Evidence that argues against this (1)
    • HOXA9 is not universally required, and using HOXA transcripts as the default pharmacodynamic readout fails in a large minority of infant cases. Multi-omics clustering of 84 infants with KMT2A-rearranged leukaemia resolved five integrative clusters of which TWO are IRX-type rather than HOXA-type (IC1 IRX-type differentiated, IC2 IRX-type undifferentiated), with different master transcription factors and fusion partners; IC2 was the most refractory subgroup and carried RAS-pathway mutations at 100% frequency. In IRX-type disease there is no HOXA signal to read down, so a menin inhibitor's standard pharmacodynamic endpoint is uninformative exactly where the disease is worst. Isobe T et al. 2022
Connected to