Each partner prefers a different KMT2A exon: partner-specific break position in 3,401 patients

A referral-based diagnostic cohort accrued 2003-2022 at a single centre network, so both partner frequencies and any stratum-specific signal reflect which centres sent samples. The per-partner preferences are reported in supplementary tables we did not inspect directly; we took them from the Results text of the PMC full text. Significance thresholds and multiple-testing correction across ~35 partner subgroups are not stated in the text we read, and with that many comparisons some of the rarer-partner findings will be false positives. Two detection technologies were mixed (LD-PCR for 2,702 patients, targeted NGS for 696), and the BCR1/BCR2 split itself differs between them (98%/1.4% overall vs 94%/6% in the NGS subset), so technology is a further potential confounder of positional data.

• The same authors' 2018 analysis of a nested 69% subset named a DIFFERENT set of deviating partners. MLLT10 (AF10) and EPS15 were on the 2018 significant list and are absent from the 2023 one, despite the cohort growing by 45%. An effect that disappears as n increases is the signature of noise, not biology. MLLT11's assignment also moved, from 'significantly toward MLL intron 9' in 2018 to 'introns 9 and 10' in 2023. Meyer C et al. 2018
• Denominators, read directly from Table 1 of the source: SEPTIN9 n=14, SEPTIN6 n=19, MLLT6 n=27, MLLT11 n=28, EPS15 n=55, against AFF1 n=1262, MLLT3 n=612 and MLLT1 n=460. Several of the load-bearing subgroup cells are tiny - the 'SEPTIN6 prefers intron 11 in pediatric AND adult patients' claim rests on 10 pediatric AML and TWO adult AML patients; 'EPS15 shifts to intron 11 in adults' rests on 11 patients. With roughly 35 partner subgroups tested, crossed further by age class, and no multiplicity correction described in the Methods, several of these are expected to be false positives by chance alone. Meyer C et al. 2023
• The statistical comparator is circular. The paper explicitly declines a length-based null - 'We decided not to use a random distribution model (RDM) of chromosomal breakpoints, because this is only based on the length of each DNA region' - and tests each partner against the cohort's own observed mean distribution instead. Since KMT2A::AFF1 alone is 1,262 of 3,401 patients, 'deviates from the mean' substantially means 'differs from AFF1'. Under a length null the direction reverses: region C (exon 11 to exon 13) is 5,026 bp carrying 39.5% of breaks while region A (exon 9 to intron 9) is 1,761 bp carrying 37.2%, so per base pair the intron-11 region is about 2.7-fold DEPLETED relative to region A, not enriched (0.0079 %/bp against 0.0211 %/bp). Against regions A and B together - 2,440 bp carrying 57.0% - it is 3.0-fold depleted. Region lengths and the A/B/C percentages are quoted from the paper's own text, re-read from PMC10169636 on 2026-09-20: 'A) exon 9 - intron 9 = 1761 bp; (B) exon 10 - intron 10 = 679 bp; (C) exon 11 - intron 11 - exon 12 - intron 12 and exon 13 = 5026 bp', with observed mean distribution A = 37.2%, B = 19.8%, C = 39.5%. SOURCING CORRECTION 2026-09-20 (Q42 citation-integrity repair). The fold figure was 2.9 and was attributed to region A alone; recomputed from the paper's own numbers it is 2.7 against A and 3.0 against A+B. The direction and the argument are unchanged. Everything in this note is from the cited paper - the RDM sentence is a verbatim quotation from its Results - so the checker's low abstract-overlap score here reflects that the content is in the body, not that the citation is wrong. Meyer C et al. 2023
• Method conditioning, measurable inside the source itself. 2,702 of 3,401 breakpoints came from LDI-PCR, an assay whose own method paper says it works 'if located within the breakpoint cluster region'. Restricting the BCR1-versus-BCR2 analysis to the 696 NGS cases moves the split from 98%/1.4% to 94%/6% - roughly a thirty-fold difference in apparent BCR2 rate between the two technologies in the same study. That method effect is larger than any per-partner positional effect the paper reports. Meyer C et al. 2005
• There is no independent replication of ANY partner-stratified KMT2A intron distribution. A systematic search found no cohort outside the Frankfurt DCAL series that reports breakpoint position by partner. The disagreement documented here is internal (2018 versus 2023), because no external comparison exists to make. That single-source status is itself the most important fact about this table. Meyer C et al. 2018

A Letter, not a full paper. These percentages come from the same DCAL cohort as [E1] and are not an independent replication. The 2024 paper explicitly argues that public databases do not faithfully represent the KMT2A gene structure and proposes a 38-exon consensus; exon numbers quoted from older literature may not correspond, and the paper also describes a previously unrecognised 'exon 22' inside intron 21. Anyone matching an exon number across sources must check which numbering is in use.

• The BCR1 definition is not stable across the source lab's own three papers, and the headline percentage tracks the definition rather than new biology. 2018: 'the major breakpoint cluster region is localizing between MLL exon 9 and MLL intron 11, where the majority of patients (93.5%) had their individual breakpoints'. 2023: 'BCR1 can be mapped between KMT2A intron 7 and KMT2A exon 13 ... (n = 3336; 98%)'. 2024: 'the core of BCR1 lies between KMT2A exons 9 and 12 (90,85%)'. The rise from 93.5% to 98% between 2018 and 2023 is a widening of the window. This node previously described BCR1 as 'exons 9 to 13', which matches none of the three exactly - and which its own table contradicts, since SEPTIN9 at 'introns 7-9' and MLLT6 at 'introns 8-9' place breaks outside it. Meyer C et al. 2018
• Exon renumbering is not hypothetical and it moves coordinates inside the window at issue. The 2024 consensus records that the presence of KMT2A exon 2 varies between public databases - an annotation lacking it shifts every downstream exon by minus one, turning intron 11 into intron 10 - and that a newly recognised 'exon 22' inside intron 21 shifts everything downstream by plus one, which is exactly why BCR2 moved from 'intron 20 to exon 24' in 2023 to 'exons 21-25' in 2024. Exon ordinals are therefore not safe currency across papers or across annotation sources; hg38 coordinates are. Meyer C et al. 2023

The six-month switch point is described qualitatively from a scatter of individual patients against age; no changepoint statistic or confidence interval is given in the text we read. The authors' interpretation — that infant ALL under six months derives from fetal-liver pro-B cells while other subgroups derive from bone-marrow HSPCs — is an inference from breakpoint distribution plus prior literature, not a lineage-tracing result.

• CONTRADICTED BY THE SAME AUTHORS. Meyer 2018, on 69% of the same patients, states verbatim: 'The cross-over points were 10-14 years in ALL patients and 75 years in AML patients.' It also gives the AML direction as the opposite of the 2023 account - 'young AML patients usually prefer MLL intron 9 breakage, while older patients have more breaks in MLL intron 11' - where 2023 reports intron 9 throughout with a slight decline and no AML cross-over at all. A cross-over that moves from 10-14 years to about 6 months on a nested cohort is a cut-point read off a curve, not a biological boundary. The fetal-liver-pro-B interpretation built on it is downstream of an unstable number. Meyer C et al. 2018
• There is a competing and older explanation for the intron-9-versus-intron-11 axis that does not involve cell of origin at all: chromatin context and exposure history. Within the 8.3 kb BCR, de novo breaks fall centromeric and therapy-related breaks fall telomeric, into the half containing a high-affinity scaffold attachment region and six topoisomerase II consensus sites (74% versus 25%, P = 0.02). The 2023 paper concedes the adult MLLT3 intron-11 shift is 'clearly linked to therapy-induced leukemia' and that therapy-induced leukaemias as a class 'shifted significantly to KMT2A intron 11 breakpoints'. No published analysis removes that variance before attributing the residual to partner or to age. Broeker PL et al. 1996

FULL TEXT read via PMC5808070, with the load-bearing sentences re-fetched and quoted verbatim on 2026-09-19. This is not an independent cohort: the 2,345 patients are approximately 69% of the 3,401 in the 2023 paper, same centre, same pipeline, same authors. That is precisely why the disagreement matters - it is not two labs disagreeing, it is one lab's own analysis moving. The 2018 paragraph is additionally self-inconsistent as printed: it says young patients prefer intron 11 and that 'this is true for all investigated subgroups', then two sentences later says the opposite for AML.

• The lists move every edition, which is what a growing series does, so a 2018-versus-2023 difference is not by itself a contradiction. The 2013 edition of the same LDI-PCR series, at n=1590, already named a different seven-member dominant set - AFF1/AF4, MLLT3/AF9, MLLT1/ENL, MLLT10/AF10, ELL, KMT2A partial tandem duplications and MLLT4/AF6 - covering about 90% of rearrangements. ELL and MLLT4/AF6 are in that list and EPS15 is not; the 2023 membership differs again. Three editions with three memberships over nested cohorts is drift with sample size, and separating drift from a real change needs a variance estimate that none of these papers reports. Meyer C et al. 2013

ABSTRACT ONLY. The original validation was on 40 patients. The restriction is a property of the assay design, not a criticism of its execution, but it means a breakpoint-position distribution measured with it is conditioned on the window the assay covers, and cannot by itself establish how often breaks fall outside that window.

• The assay-window effect is measurable inside the same cohort, and it is about fourfold. Breaks in the minor breakpoint cluster region are 47 of 3,401 cases (1.4%) overall but about 6% of the subset characterised by targeted next-generation sequencing rather than LD-PCR, which the authors read as historical under-detection. That both confirms the conditioning this item describes and bounds it: the method demonstrably changes the answer, and it also shows LDI-PCR-based series are not strictly confined to BCR1, since a minor BCR is reported at all. Meyer C et al. 2023

ABSTRACT ONLY. Small numbers - 31 de novo and 8 t-AML patients - and 1996 Southern-blot mapping to BamHI halves rather than to introns. The topoisomerase II sites were defined by in vitro consensus criteria, not measured in vivo. Cited here for the direction and the mechanism, not for precise percentages.

• Sequence features inside the 8.3 kb breakpoint cluster region have repeatedly failed to predict who breaks there. When the whole region was sequenced in 22 patients who developed topoisomerase-II-inhibitor-related acute myeloid leukaemia, 37 who did not, 8 infants and 20 normal individuals, the three Alu poly-A tract polymorphisms occurred at equal frequency in leukaemic patients and controls and were explicitly 'not predictors of risk'; the one trinucleotide-repeat association was called suggestive and not statistically significant. A scaffold attachment region plus six in-vitro topoisomerase II consensus sites is a description of the sequence, not a demonstrated aetiological axis. Echlin-Bell DR et al. 2003

ABSTRACT ONLY, n = 6. Far too small to estimate a distribution. It is cited for a qualitative point that the per-partner tables cannot show: with partner and aetiology both held constant, breaks still split across intron 9 and intron 11, so intra-partner heterogeneity is not obviously smaller than the inter-partner 'preferences' being reported.

• Within-partner heterogeneity is structured, not random, which is the opposite of what six cases can show. Holding both partner and translocation constant across 40 t(4;11) patients, the breakpoints did not scatter: they resolved into two distinct hotspot areas that differ significantly between patients under and over one year of age, pointing to different age-linked mechanisms. A six-case series that finds breaks in intron 11 and intron 9 has not shown that intra-partner heterogeneity is unstructured; it has shown that six cases cannot resolve the structure. Reichel M et al. 2001

ABSTRACT ONLY. n = 30 for the breakpoint-outcome analysis. Not independent of the source of the frequency data in this node: Meyer C and Marschalek R are co-authors and the breakpoints were called at the same Frankfurt DCAL by the same pipeline. It should not be treated as external corroboration.

• No prospective trial has ever entered breakpoint position into a prognostic model, and the models that exist are full without it. Interfant-06 randomised 651 infants across 18 national and international groups and found KMT2A rearrangement itself to be the strongest prognostic factor for event-free survival, followed by age, white cell count and prednisone response. Breakpoint position appears nowhere in that analysis. Until it is tested against those four covariates in a trial cohort, a 30-patient retrospective association is not evidence that it adds anything. Pieters R et al. 2019