Oncofusion

start / graph / concept-partner-lineage-confound-in-kmt2a-outcome

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

The lineage confound: partner predicts whether the leukaemia is myeloid or lymphoid before it predicts anything else

also called KMT2A partner-lineage confound · AFF1 to B-ALL, MLLT3 to AML

Some partner genes almost always give B-cell leukemia and others almost always give myeloid leukemia. Since those two diseases are treated differently and have different cure rates, a partner difference can simply be a disease difference.

The same thing, for a scientist

KMT2A partner identity is strongly associated with lineage: KMT2A::AFF1 is overwhelmingly B-lymphoid, KMT2A::MLLT3 and KMT2A::AFDN predominantly myeloid, and KMT2A::MLLT1 and KMT2A::MLLT10 appear in both. Because ALL and AML are treated on different backbones with different cure rates, a partner comparison that pools lineages is measuring disease assignment. Two observations show the confound is not merely theoretical. First, the partner-based risk assignment differs by lineage: KMT2A::AFF1 and KMT2A::MLLT1 sit in the I-BFM high-risk group in childhood AML, while in non-infant childhood ALL t(4;11)-positive B-ALL reaches 64.8% (+/-3.0) 5-year EFS and t(11;19)-positive T-ALL reaches 91.2% (+/-4.9). Second, the cleanest partner comparison available holds lineage constant and still separates: within T-ALL in the same 629-patient cohort, t(11;19) gives 91.2% (n=34) and t(9;11) gives 41.7% (+/-17.3, n=9). Holding lineage constant does not abolish the partner effect - it reverses its sign relative to AML.

Go deeper
The within-T-ALL comparison is the key piece of evidence

Both extremes of the reported Ponte-di-Legno range are T-lineage 1. That is the one place in this literature where lineage is held constant and two partners are compared, and they differ by roughly 50 points. If the partner effect were entirely a lineage effect, that comparison should have been flat. It is the widest gap in the cohort. The n of 9 in the worse cell means the magnitude is unestimated, but the existence of a within-lineage partner effect is supported.

MPAL and lineage switch make the confound worse, not better

KMT2A-rearranged leukaemias are notable for ambiguous-lineage presentation and for lineage switch under pressure. A partner assigned to a lineage at diagnosis may not stay there, and studies typically classify by diagnostic immunophenotype. Any partner-by-lineage table is therefore built on a variable that is itself unstable in this disease.

How to state a partner result honestly

Report the partner-by-lineage cell with its n, never the partner main effect pooled across lineage. The Ponte-di-Legno abstract does this and it is why its numbers are usable 1; most other cohorts report a partner main effect within a single disease, which is fine, or across diseases, which is not.

Facts and figures
confoundlineage
interaction testedfalse
sourcePMIDs 36256911, 36996387, 38306602
partner lineage association
{
  "KMT2A::AFF1": "predominantly B-lymphoid",
  "KMT2A::MLLT3": "predominantly myeloid",
  "KMT2A::AFDN": "predominantly myeloid",
  "KMT2A::MLLT1": "both B-lymphoid and T-lymphoid and myeloid",
  "KMT2A::MLLT10": "both myeloid and T-lymphoid"
}
within lineage test
{
  "lineage": "T-ALL",
  "cohort_pmid": "36256911",
  "t_11_19_efs_5y_pct": 91.2,
  "t_11_19_n": 34,
  "t_9_11_efs_5y_pct": 41.7,
  "t_9_11_n": 9,
  "verdict": "partner effect persists with lineage held constant, but n is small"
}
separating designs
[
  "report partner-by-lineage cells with n, never a pooled partner main effect",
  "fit a partner-by-lineage interaction term in a pooled ALL+AML individual-patient dataset",
  "compare the same partner across lineages within one protocol era"
]
How this could bite us back
  • Menin-inhibitor eligibility is written as KMT2A-rearranged acute leukaemia, which pools lineages by design. A single response rate across that population is an average over diseases whose baseline cure rates differ by 40 points.
  • If the partner effect is real within lineage, then a class-level drug will have partner-specific incremental value, and the lineage-pooled trial design being used now cannot detect it.
Still open
  1. What are the t(11;19) B-ALL and t(9;11) B-ALL 5-year EFS values? Those two missing cells would complete the only partner-by-lineage table that exists.
  2. Does a partner-by-lineage interaction term reach significance in any pooled dataset? None has been fitted.
  3. How often does lineage switch occur within each partner, and does it differ by partner? If it does, the lineage variable is partner-dependent and the confound is not adjustable in the usual way.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Holding lineage constant within T-ALL in a 629-patient cohort of non-infant childhood KMT2A-r ALL, t(11;19)-positive T-ALL had 5-year EFS 91.2% (+/-4.9, n=34) and t(9;11)-positive T-ALL had 41.7% (+/-17.3, n=9), against a whole-cohort 69.1% (+/-1.9); t(4;11)-positive B-ALL (n=266) was 64.8% (+/-3.0).

    PMID 36256911DOI
    What this evidence cannot show

    The t(9;11) T-ALL cell has nine patients and a standard error of 17.3 percentage points, so its interval is very wide. Retrospective, 17 study groups, 1995-2010, not pre-specified. Lineage assigned by diagnostic immunophenotype, which is unstable in KMT2A-rearranged disease. Full text not open access; read from the abstract.

    Evidence that argues against this (1)
    • No t(11;19) B-ALL or t(9;11) B-ALL comparator is reported in the abstract, so an alternative reading is that T-lineage KMT2A-r ALL does well or badly for reasons unrelated to the partner and the two cells differ by chance at n=34 and n=9. Attarbaschi A et al. 2023
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The same partners carry opposite risk assignments in the two lineages: KMT2A::AFF1 (4q21) and KMT2A::MLLT1 (19p13.3) are both in the I-BFM high-risk fusion-partner group for childhood AML, whose 5-year EFS is 30.3% versus 54.0% for non-high-risk partners.

    PMID 36996387DOI
    What this evidence cannot show

    Cross-study comparison; no analysis has fitted a partner-by-lineage interaction. The AML high-risk group pools five partners, so neither AFF1's nor MLLT1's individual contribution is visible.

    Evidence that argues against this (1)
    • Two adult AML cohorts found no partner effect on survival at all, so the AML side of the lineage contrast is not established across age and the apparent sign flip may be confined to childhood myeloid disease. Issa GC et al. 2021
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    An MLLT10-centred analysis of 2,080 children shows the mirror-image result: varying the 5-prime partner while holding MLLT10 fixed did not change outcome (KMT2A::MLLT10 EFS 19.5% versus 12.7% for other MLLT10 partners, P=.628), even though the methylation signatures differed markedly.

    PMID 38306602DOI
    What this evidence cannot show

    The within-MLLT10 comparison runs on 13 PICALM::MLLT10 and 10 X::MLLT10 patients, so the null has essentially no power. Cited as an instructive contrast rather than as evidence of equivalence.

    Evidence that argues against this (1)
    • In the same disease, the 3-prime partner does separate outcome strongly across 1,130 children, so partner-independence within MLLT10 is a local result about one adverse partner. van Weelderen RE et al. 2023
Connected to