Which partner, in which patient: KMT2A partner frequencies across six age-lineage strata

A referral-based cohort accrued 2003-2022 at the Frankfurt DCAL from centres in 20 countries; samples are sent specifically to obtain genomic fusion sequences for MRD monitoring, so partner frequencies reflect referral behaviour rather than population incidence, and rarer partners are plausibly over-represented. 43 patients could not be age-classified and 57 lacked gender data and are excluded from stratum denominators. Percentages as published do not always sum to 100 within a stratum because of the 'other rearrangements' remainder. Counts were read from the PMC full text (PMC10169636) on 2026-09-18, including the italicised gene symbols, because a plain-text rendering of the same article had the gene names stripped.

• PEDIATRIC AML, MLLT10 and MLLT3. Two independent international karyotype-based cohorts put MLLT10 well below the DCAL figure of 23.4%: 17.4% (218/1256) and 13% (98/756), with MLLT3 correspondingly higher at 43.3% and 43% against 36.6%. The two errors are arithmetic mirrors of each other, which is the signature of a resolution difference rather than a sampling difference - when cryptic MLLT10 is recovered molecularly, the cytogenetically obvious t(9;11) share is diluted. van Weelderen RE et al. 2024
• THE MECHANISM, quantified. KMT2A::MLLT10 requires three or more breaks and is frequently invisible to karyotype: 26% of cases with conventional chromosome studies had no abnormality of chromosome 10 or 11 at all, and in a large COG AML series about 20% of MLLT10-fusion cases had a reported cytogenetic band that does not map to MLLT10. That makes the DCAL figures the molecular-complete-ascertainment estimate and the karyotype cohorts systematically shifted toward legible partners - so on this axis the DCAL numbers are probably the better ones, and this node should say so rather than leaving the disagreement unexplained. Peterson JF et al. 2019
• INFANT ALL - and this one does NOT have a cryptic-partner explanation. The prospective COG trial AALL0631 gives MLLT3 at 5% (8/146; about 6% of typed cases) against the DCAL 16%, a 2.6 to 3.2-fold relative gap and the largest single-partner discrepancy found. MLLT1 moves the other way, 34% against 24%. AFF1 is 38% against 48.3%. Because MLLT3 and MLLT1 move in opposite directions, a single global bias cannot explain it, and the 12% partner-unknown fraction cannot close the MLLT3 gap. This is the disagreement that warrants a confidence downgrade on the infant ALL stratum specifically. Brown PA et al. 2021
• ADULT ALL. The claim that adult KMT2A-rearranged ALL is 82.7% AFF1 is the single number in this node most worth flagging. An RNA-seq cohort spanning all ages puts AFF1 at 69% overall and 71% in adult plus elderly patients - about twelve absolute points lower. The denominators are not identical (321 in the cross-tabulation, and 18-54 / 55+ strata rather than a single adult class), so this is a qualification rather than a refutation, but the downstream argument in `platform-one-asset-two-populations-menin` that an adult KMT2A-r ALL trial is effectively a KMT2A::AFF1 trial is weaker at 71% than at 83%. Hartmann AM et al. 2026
• THE STRUCTURAL POINT that applies to every row. These are frequencies among TYPED cases, and the typed fraction ranges from 100% (DCAL, molecular breakpoint cloning) through 90%, 88%, 84.3% and 81% down to 70% in a FISH-heavy adult series. Untyped cases are not missing at random; they are enriched for cryptic and complex rearrangements. No two of these tables share a denominator definition, and comparing them percentage-to-percentage without saying so is the commonest way this literature is misread. Shen H et al. 2026
• AGE CUT-OFF IS NOT THE EXPLANATION. Every prospective trial checked defines infant consistently at about 12 months (Interfant 365 days, AALL0631 under 366 days), and moving to the non-infant childhood window (1-18 years, Ponte-di-Legno n=629) moves AFF1 only from 48.3% to 51.5% while MLLT10 falls to 2.6% (14/530) - again the karyotype signature. The discontinuity in this literature sits at the infant/adult boundary and at the detection method, not at the 6-versus-12-versus-24-month cut-off. Attarbaschi A et al. 2023

A 2013 educational review, therefore secondary; the percentages are of the subset of cases with an identified partner (93% of infant ALL, 66% of infant AML), not of all cases, which makes them not directly comparable to the denominators in [E1]. Included here for cross-checking, not as an independent replication — both sources ultimately draw on overlapping referral networks.

• A caveat this node should carry about its own strongest apparent corroborator: Interfant KMT2A typing was performed at the Frankfurt DCAL, with Meyer C and Marschalek R as co-authors of the Interfant MRD papers. Interfant therefore differs from the DCAL recombinome cohort in ascertainment only, not in method, and is not an independent check on the partner frequencies. The genuinely independent comparators are the karyotype-based cohorts (van Weelderen, Balgobind, Attarbaschi) and the RNA-seq cohorts (Hartmann, Abla). Brown PA et al. 2021

FULL TEXT read (PMC11225675). This is a karyotype-driven cohort: group assignment is by centrally reviewed cytogenetic band, and fusions present in fewer than ten patients were left in a 'KMT2A-other' bin (n=126, 10.0%). Karyotype cannot see cryptic KMT2A::MLLT10 insertions, so the direction of the disagreement is exactly what a detection-method difference predicts. Only 56 of 1,256 had incomplete karyotypes, so this is not a coverage problem - it is a resolution problem.

• A karyotype-defined cohort cannot see the partner it is being scored against. Transcriptome profiling of 2,080 children and young adults with acute myeloid leukaemia on COG AAML0531 and AAML1031 found 127 MLLT10 fusions (6.1%), 104 of them KMT2A::MLLT10 - a resolution central karyotype review cannot reach. The six-point MLLT10 deficit and the mirror-image MLLT3 excess in this 1,256-patient karyotype cohort are exactly what a detection difference predicts, so the disagreement is evidence about assays rather than about two populations differing. Abla O et al. 2024

FULL TEXT read (PMC2927031); note that the PubMed full-text tool returned an empty body for this PMCID and the numbers were read from the rendered PMC page. The 19% 'MLL-other' bin is the key limitation and is not missing at random - it is enriched for exactly the cryptic and complex rearrangements that drive the disagreement.

• The gap is not an epoch effect. The same LDI-PCR series, in its 2013 edition at n=1590 and therefore contemporaneous with this cohort's 1993-2005 accrual, already placed MLLT10/AF10 among the seven rearrangements accounting for about 90% of all KMT2A rearrangements. MLLT10 was not becoming commoner between the two series; what changed was how many cases were characterised molecularly rather than by conventional cytogenetics, which half this cohort was not. Meyer C et al. 2013

FULL TEXT read (PMC8763141); gene symbols were stripped from the PMC text extraction and were recovered from the rendered page, the same failure mode this node already documents for the DCAL numbers. This is a treated trial population, not a diagnostic series, so eligibility criteria shape it. 12% have no partner assigned; even if every one of those were MLLT3 the ceiling would be 17%, and the point estimate among typed cases is about 6%, so the MLLT3 gap is not an artefact of the unknowns. MLLT1 moves in the OPPOSITE direction to MLLT3, which rules out a single global bias.

• A single national trial's partner mix is shaped by who reaches that trial, and the contemporaneous international trial had a visibly different population. Interfant-06 enrolled 651 infants across 18 national and international groups over an overlapping period and reported 6-year event-free survival of 46.1%, with 6-year EFS 10-12 percentage points higher in West European and North American groups than elsewhere. Two infant trials whose outcomes differ that much by geography are not interchangeable denominators for a partner frequency, and AALL0631's 146 KMT2A-rearranged infants are a subset of one arm of that landscape. Pieters R et al. 2019

FULL TEXT read (PMC12931199); gene symbols stripped by the extractor and recovered from the rendered page. CAUTION ON DENOMINATORS: the cohort is 465 patients, 436 had fusion data, and 321 is the set used for the age-by-partner cross-tabulation, so the 71% is adult plus elderly combined within that 321, not a figure on 465. The age strata (18-54 and 55+) also do not match the DCAL definition of adult. Very recent (2026) and not yet independently confirmed.

• The numeric disagreement runs the other way too, and at least one of the two figures is not a partner frequency in the sense the other means. The DCAL adult ALL stratum is 647 patients with AFF1 in 535 of them, 82.7%, against 71% for adult-plus-elderly in the RNA-sequencing cohort - a 12-point gap between two molecularly typed series. Because the RNA-seq figure is computed on the 321 patients used for the age-by-partner cross-tabulation rather than on all 465, and its age strata (18-54, 55+) do not match the DCAL definition of adult, the comparison is between differently defined quantities before any biology is invoked. Meyer C et al. 2023

ABSTRACT ONLY (no PMC deposit). n=60, single laboratory, cases referred for a dedicated dual-fusion FISH assay, so the series is itself ascertained and the 26% is a rate within referred KMT2A/MLLT10 cases, not a population rate. Corroborated independently by E10.

• A FISH assay that scores KMT2A/MLLT10 fusion cannot see the part of the MLLT10 landscape that is not KMT2A. Among 127 MLLT10 fusions identified by transcriptome profiling in 2,080 COG AML patients, only 104 (81.9%) were KMT2A::MLLT10; 13 were PICALM::MLLT10 and 10 were other partners (DDX3X, TEC, DDX3Y, CEP164, SCN2B, TREH, NAP1L1, XPO1). A 26% cryptic rate measured inside a series referred for a dedicated dual-fusion assay is therefore a rate within a doubly selected population and does not transfer to MLLT10 rearrangements generally. Abla O et al. 2024

FULL TEXT read (PMC12896996) for the 181-patient figure; the other typed fractions are from the papers cited elsewhere in this node. Single-centre Chinese adult/AYA cohort, median age 33, so it is not directly comparable to the DCAL adult stratum on case mix. The general point is the important one: cases that fail typing are NOT missing at random - they are enriched for cryptic and complex rearrangements, which is the MLLT10 and rare-partner end of the distribution.

• Where unselected RNA sequencing is applied, the untyped fraction largely disappears and the heterogeneity moves somewhere else entirely. In 175 infants with B-cell precursor ALL typed by karyotyping, FISH and RNA-seq, 33 (17.7%) had no KMT2A rearrangement at all and were instead NUTM1-rearranged (n=9) or PAX5-rearranged (n=10). The same cohort found no survival difference by KMT2A partner gene - KMT2A::AFF1 2-year EFS 36% against 37% for all other partners, p = 0.9 - so the clinical stake in the typed-fraction problem is smaller than the emphasis on partner frequencies implies. Ilyasova K et al. 2025

ABSTRACT ONLY (all quoted numbers are in the abstract). Karyotype-based partner assignment with 15.7% of patients having no identifiable partner at all, so the 2.6% MLLT10 is the clearest karyotype-driven undercount in the lymphoid strata. Cited here to test whether the infant age cut-off explains the cross-cohort disagreements - it does not.

• The molecular series disagrees with this karyotype-based one on the same age group, in both directions. In the DCAL cohort the paediatric ALL stratum (n=530) is AFF1 215 (40.6%) and MLLT1 100 (18.9%), against 51.5% and 20.0% among the 530 typed cases here. An eleven-point AFF1 gap between two cohorts of almost identical size in the same age band, with 15.7% of this one's patients having no identifiable partner, means the low MLLT10 figure cannot be interpreted without knowing where the untyped cases went. Meyer C et al. 2023

FULL TEXT read (PMC11024924). IMPORTANT LIMIT: the paper does not state how many of the 2,080 patients were KMT2A-rearranged, so NO KMT2A::MLLT10 share of KMT2A-r can be computed from it and none is claimed here. The ~2/3 and ~20% figures are the paper's own approximations, not exact counts.

• An MLLT10-centred series measures a partner-gene population, not a KMT2A one, and the non-KMT2A fraction is biologically distinct rather than noise. CALM-AF10 (PICALM::MLLT10) T-cell ALL has its own expression programme - HOXA5, HOXA9, HOXA10 and BMI1 overexpression, with BMI1 elevated only in CALM-AF10 and not in MLL-translocated leukaemia. So the 23 of 127 MLLT10 fusions that are not KMT2A::MLLT10 are a different disease sharing a gene, and the karyotype-versus-RNA-seq concordance figures quoted here are properties of MLLT10 detection, which is what this item claims, but they cannot be converted into a KMT2A-r partner share. Dik WA et al. 2005