Oncofusion

start / graph / concept-cytogenetic-band-versus-molecular-partner-misassignment

Explainermoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q19-partner-outcome

One karyotype band, two partner genes: why t(11;19) and t(10;11) can cancel a real partner effect

also called 19p13.1 ELL versus 19p13.3 MLLT1 · 10p12 MLLT10 versus 10p11.2 ABI1 · band-level partner misassignment

Two different partner genes can produce a chromosome picture that looks the same. If a study records only the picture, it mixes the two genes together, and if they point in opposite directions the effect disappears.

The same thing, for a scientist

Several recurrent KMT2A translocations are band-ambiguous. t(11;19) resolves to KMT2A::MLLT1 at 19p13.3 or KMT2A::ELL at 19p13.1; t(10;11) resolves to KMT2A::MLLT10 at 10p12 or KMT2A::ABI1 at 10p11.2, and a cytogenetically similar t(10;11) can also be PICALM::MLLT10 with no KMT2A involvement at all. The two members of each pair are not prognostically equivalent. In the 756-patient international paediatric AML series, t(10;11)(p12;q23) and t(10;11)(p11.2;q23) carried different independent hazard ratios, 1.5 (P=.005) and 2.5 (P=.005). In the I-BFM classification MLLT1 at 19p13.3 is high-risk while ELL at 19p13.1 is not, and in two adult cohorts KMT2A::ELL is the one partner with a survival advantage (3-year OS 59.8% versus 39.3%, P=0.023). A series that records only t(11;19) therefore averages an adverse partner with a favourable one. A null partner result from such a series is uninformative, and it looks exactly like a real null.

Go deeper
The concrete failure mode

Suppose ELL is favourable and MLLT1 adverse, as the paediatric AML classification and the adult ELL cohorts jointly suggest 2, 3. A study that codes both as t(11;19) will report no effect for t(11;19), and the effect it failed to find is the effect it created by pooling. This is not hypothetical arithmetic: the two largest adult KMT2A-r AML series report partner nulls 4 without stating whether their t(11;19) cases were sub-banded.

PICALM::MLLT10 does not contain KMT2A at all

In a 2,080-patient COG analysis, 13 of 127 MLLT10-fused patients had PICALM::MLLT10 and 10 had other 5-prime partners including XPO1, DDX3X and NAP1L1 1. Clinically these behave like KMT2A::MLLT10 1. A t(10;11) recorded by karyotype alone can therefore be a fusion with no KMT2A N-terminus and no menin-binding motif - which matters for eligibility, not just for prognosis. The 1998 report that first separated CALM/PICALM from MLL rearrangement in t(10;11) made the same point morphologically 5.

Practical rule for this graph

When an entry in this graph records a partner-outcome number, it should record whether the partner was assigned by karyotype band, by sub-band, by RT-PCR panel or by RNA sequencing, because those four methods have different failure modes. A targeted RT-PCR panel returns negative for every partner it was not designed around; karyotype cannot resolve sub-bands reliably; RNA-seq finds novel partners that the other methods push into an 'other' bin.

Facts and figures
confoundpartner assignment method
sourcePMIDs 38306602, 19528532, 41681872, 34588432, 9616163
ambiguous bands
[
  {
    "karyotype": "t(11;19)(q23;p13.3)",
    "gene": "KMT2A::MLLT1",
    "risk": "I-BFM high-risk in childhood AML; best-outcome cell in childhood T-ALL"
  },
  {
    "karyotype": "t(11;19)(q23;p13.1)",
    "gene": "KMT2A::ELL",
    "risk": "non-high-risk; survival advantage in two adult cohorts"
  },
  {
    "karyotype": "t(10;11)(p12;q23)",
    "gene": "KMT2A::MLLT10",
    "hr": 1.5,
    "p": 0.005
  },
  {
    "karyotype": "t(10;11)(p11.2;q23)",
    "gene": "KMT2A::ABI1",
    "hr": 2.5,
    "p": 0.005
  },
  {
    "karyotype": "t(10;11) without KMT2A",
    "gene": "PICALM::MLLT10",
    "note": "no KMT2A N-terminus, so no menin-binding motif"
  }
]
assignment methods
[
  "karyotype band",
  "karyotype with central review and sub-band",
  "targeted RT-PCR panel",
  "FISH split signal",
  "RNA sequencing"
]
method failure modes
{
  "karyotype band": "pools two genes at one band",
  "targeted RT-PCR": "negative for any partner not on the panel",
  "FISH split signal": "confirms KMT2A rearrangement, identifies no partner",
  "RNA sequencing": "finds novel partners others bin as 'other', changing denominators"
}
How this could bite us back
  • AUGMENT-101 confirmed KMT2A-rearranged status centrally by FISH split signal, a method that establishes rearrangement and identifies no partner; a quarter of its patients have an unknown partner, so partner-stratified response cannot be recovered from that trial even in principle for those patients.
  • A PICALM::MLLT10 patient has no KMT2A N-terminus and therefore no menin-binding motif, yet presents with a t(10;11) that a band-level eligibility screen would pass; class-level eligibility can enrol patients the mechanism cannot reach.
Still open
  1. In each published adult partner null, were t(11;19) cases resolved to 19p13.1 versus 19p13.3? If not, those nulls should be reanalysed before being used as evidence for the class abstraction.
  2. How many KMT2A-rearranged patients in menin-inhibitor trials had their partner assigned by FISH split signal only, which identifies no partner at all?
  3. Does RNA-seq-era partner assignment change the partner frequency denominators enough to alter published subgroup outcome estimates?
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Among 127 MLLT10-fused paediatric AML patients, 104 (81.9%) were KMT2A::MLLT10, 13 (10.2%) PICALM::MLLT10 and 10 (7.9%) had other 5-prime partners (DDX3X, TEC, DDX3Y, CEP164, SCN2B, TREH, NAP1L1, XPO1), and the non-KMT2A fusions had outcomes not significantly different from KMT2A::MLLT10 (EFS 19.5% versus 12.7%, P=.628) despite carrying a distinct NUP98::NSD1-like hypermethylation signature.

    PMID 38306602DOI
    What this evidence cannot show

    Within-MLLT10 comparisons rest on 13 and 10 patients, so the outcome equivalence is an unpowered null. Fusion detection used transcriptome profiling and/or karyotyping, so detection method varies within the cohort - which is itself an instance of the problem this entry describes.

    Evidence that argues against this (1)
    • In the 756-patient international series the two t(10;11) sub-bands did separate, with independent hazard ratios of 1.5 for 10p12 and 2.5 for 10p11.2, showing that sub-band resolution can recover a difference that band-level coding would hide. Balgobind BV et al. 2009
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Sub-band resolution changes the prognostic answer: t(10;11)(p12;q23) and t(10;11)(p11.2;q23) were separately identified as independent prognostic predictors with hazard ratios 1.5 (P=.005) and 2.5 (P=.005) in 756 children with 11q23-rearranged AML.

    PMID 19528532DOI
    What this evidence cannot show

    Retrospective, pooled across 11 groups. Central karyotype review was performed, which is why sub-bands could be distinguished here and cannot be in series relying on local reporting. Subgroup sizes are not given in the abstract.

    Evidence that argues against this (1)
    • A later analysis of the same collaboration found that additional cytogenetic aberrations carry independent prognostic weight comparable to the partner, so the difference between the two t(10;11) sub-bands may partly reflect differing rates of co-occurring aberrations rather than the partner gene itself. Coenen EA et al. 2011
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The two genes at 19p13 differ in assigned risk: KMT2A::MLLT1 (19p13.3) is in the I-BFM high-risk group in childhood AML, while KMT2A::ELL (19p13.1) is not and shows a survival advantage in adult cohorts (3-year OS 59.8% versus 39.3%, P=0.023).

    PMID 41681872DOI
    What this evidence cannot show

    The ELL advantage rests on 24 patients in one retrospective adult cohort with median follow-up of 17.5 months against a 3-year endpoint; the risk assignment for MLLT1 comes from a different study in a different age group. The comparison across the two is a juxtaposition, not a head-to-head test.

    Evidence that argues against this (1)
    • In 1,130 children with KMT2A-r AML, MLLT1 is only one of five partners inside a grouped high-risk arm and is never reported as its own curve, so the claim that MLLT1 and ELL differ has not been tested directly in a single cohort. van Weelderen RE et al. 2023
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The two largest adult KMT2A-r AML series report partner nulls (n=172, translocation subtype not predictive in multivariable analysis; n=205, OS similar across translocations including t(9;11), P=0.756), and neither abstract states whether 19p13 or 10p sub-bands were resolved.

    PMID 34588432DOI
    What this evidence cannot show

    Absence of a statement about sub-banding in an abstract is not evidence that sub-banding was not done; the full methods may address it. This evidence item is a reason to check, not a finding that the studies were flawed.

    Evidence that argues against this (1)
    • The HARMONY study does specify t(9;11)(p21.3;q23.3)/KMT2A::MLLT3 at sub-band and gene level in its abstract, which indicates at least some molecular resolution in that cohort and weakens the inference that adult series are band-level throughout. Hernández-Sánchez A et al. 2024
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The distinction between CALM/PICALM and MLL rearrangement in cytogenetically similar t(10;11) leukaemias was established by interphase FISH in 1998, which separated morphologically distinct subsets: CALM rearrangement in immature AML and ALL versus MLL rearrangement in myelomonocytic AML, with poor prognosis in both.

    PMID 9616163
    What this evidence cannot show

    Ten cases total, of which only 7 uniformly treated AML patients contributed the survival comparison; a 1998 single-centre FISH study on older protocols. Cited for the diagnostic point, not for its prognostic estimate. No DOI is available for this record.

    Evidence that argues against this (1)
    • The much larger COG analysis found that PICALM::MLLT10 and KMT2A::MLLT10 have statistically indistinguishable outcomes, so the 1998 claim that the two subsets are prognostically distinct has not been borne out at scale. Abla O et al. 2024
Connected to