Oncofusion

start / graph / mech-breakpoint-position-consequences

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q2-atlas

Breakpoint anatomy, and an audit of the prognosis claim

also called breakpoint position · intron 9 vs intron 11 · breakpoint cluster region · breakpoint evidence audit

This is the source audit behind the breakpoint question: exactly where KMT2A breaks, how the popular version of the age story got reversed in retelling, and why the claim that one break predicts worse survival rests on only thirty patients.

The same thing, for a scientist

Breaks concentrate in a major breakpoint cluster region spanning KMT2A exon 9 to exon 13 (98% of 3,401 cases) with a minor region at introns 20-24. Position correlates with age, lineage and therapy exposure in large univariate analyses, and there is a mechanistic reason to care because a break at or past exon 11 destroys the PHD1-3 module. The prognostic claim, however, rests on a single unreplicated cohort of 30 patients, and position has never been separated statistically from partner-gene identity.

Go deeper
The anatomy of the break2 deeper

In the largest series, 3,401 patients, the major breakpoint cluster region runs from KMT2A intron 7 to exon 13 and accounts for 98% of breaks. A second, minor region at intron 20 to exon 24 accounts for 1.4% 1.

Within the major region, breaks split into three subregions with a remarkably even spread:

SubregionSpanShare of breaks
A, exon 9 to intron 91,761 bp37.2%
B, exon 10 to intron 10679 bp19.8%
C, exon 11 to exon 135,026 bp39.5%

The functional dividing line is between A/B and C, because a break from exon 11 onward cuts through the PHD1-3 module 1.

A technical caveat with real consequences: the split between the major and minor regions is 98% to 1.4% when measured by the older PCR method, but 94% to 6% when measured by targeted sequencing on the same platform. Every pre-sequencing estimate of minor-region frequency is an underestimate 1.

Why exon 11 is the meaningful cut point

KMT2A exons 11-16 encode the PHD1-PHD2-ePHD3 module. PHD3 binds either CYP33/PPIE, which flips KMT2A from an activator into a repressor by recruiting BMI1 and histone deacetylases, or H3K4me2/3. PHD2 and PHD3 also bind the E3 ligases CDC34 and ASB2 that set KMT2A's steady-state stability.

A break upstream of exon 11 leaves this module intact in the reciprocal product. A break at or past exon 11 destroys the domain's fold, disabling both CYP33 binding and the module's self-dimerisation 1. So there is a real, specific molecular difference between the two break classes. What has not been shown is that this molecular difference translates into a different clinical course.

A numbering trap that will bite anyone reading the old literature

Papers before roughly 2000 describe the cluster region as exons 5-11 and place infant junctions in intron 8. Modern work uses exon 9 to intron 11. The offset appears to be three exons, so old intron 6/7/8 corresponds to current intron 9/10/11.

We could not find a published conversion table, including in the 2024 consensus gene-structure update, so treat that mapping as inference rather than fact and never mix the two systems in one analysis. A further wrinkle: the 2024 update added a previously unrecognised exon inside intron 21, so minor-region exon numbers published before 2024 may be off by one 2.

The classic age claim is usually stated backwards1 deeper

The widely repeated version is that infant breaks cluster toward the 5' end and adult breaks toward the 3' end. The literature says the opposite. Infants and therapy-related cases both cluster 3', in intron 11. Older patients with de novo disease cluster 5', in intron 9 31.

The error has a traceable origin. A 1996 study compared de novo cases, a mixed cohort of both lineages and all ages, against therapy-related AML and found de novo breaks were 5'. A 1997 study then split that de novo group by age and showed that infants actually behave like therapy-related cases, which was the whole etiological point: infant breaks look like topoisomerase-II-associated damage 3. The modern claim is the 1996 result with the word infant silently substituted for de novo.

The direction has since replicated across cohorts of 70, 1,530, 2,345 and 3,401 patients, so the association itself is solid. The 2017 series even quantified where the pattern flips: around 10-14 years in lymphoblastic leukemia and 75 years in myeloid leukemia 4.

Three reasons to distrust the age story anyway

Age and partner gene are badly confounded. Infant cases are dominated by AFF1 and MLLT1; adult myeloid cases by partial tandem duplications, AFDN and MLLT3. Nobody has ever fitted a model separating position from partner.

The effect reverses inside some partner strata. MLLT10 shows the opposite age pattern, described in the source as a completely contrary situation. AFDN prefers intron 9 in both lineages. Adult MLLT3 also runs contrary 4.

The same group's own numbers do not agree across editions. The 2017 series puts the lymphoblastic crossover at 10-14 years; the 2023 series puts it at about 6 months. Neither paper reconciles the two 41.

And there is a geographic shift of comparable size to the age shift that failed significance even at n of roughly 1,500, which is a useful calibration for how much apparent structure is noise at these sample sizes 4.

The prognostic claim is one small study, and the field has quietly stopped asserting it2 deeper

The statement that an intron 11 break predicts worse survival traces to a single report. Its clinical cohort was 545 children, but breakpoints were localised in only 30 cases 5. A multivariable model on 30 patients with several covariates is severely overfitted, the finding has never been independently replicated, and it is confounded with both partner gene and age, since infants are simultaneously the intron-11-enriched group and the worst-prognosis group.

Three successive large recombinome papers repeat the claim, but none of them performs its own survival analysis. There is no event-free survival curve, hazard ratio or log-rank test in any of them. They are citing the same 30 patients. By 2023 the same authors hedge it explicitly: whether these findings have an impact on clinical outcome is yet unclear 1.

The loudest silence is in the trials. Interfant-99 (n=482), Interfant-06 (n=651) and the Interfant-06 minimal-residual-disease analysis (n=249) all built multivariable outcome models, and none included breakpoint position, despite the MRD study using the genomic breakpoint sequence as its PCR target 6. They had the sequences in hand and never tested the variable. That is a real, citable gap and an opportunity.

The one breakpoint-adjacent finding that does hold up

Whether the reciprocal fusion is expressed is a direct consequence of breakpoint architecture, and it does predict outcome. In 124 infants treated uniformly on Interfant protocols, the reciprocal AFF1::KMT2A transcript was expressed in only 45% of t(4;11) cases, and those patients had four-year event-free survival of 62.4% versus 11.7%, holding up in a Cox model adjusted for age, white-cell count and prednisone response 7.

This connects to breakpoint position on the partner side: about half of KMT2A::MLLT1 cases break upstream of that partner's exon 1, producing a spliced fusion with no reciprocal allele (for AFF1 the figure is 1 in 1,262) 1. So the causal chain is partner-side breakpoint, then reciprocal expression, then outcome. This is a transcript-expression marker, not a KMT2A-intron-position marker, and the two should not be conflated.

What is simply untested

No published study tests breakpoint position against immunophenotype (CD10, NG2/CSPG4, CD133), against mixed-phenotype acute leukemia, or cross-tabulates the MLLT3 partner-side breakpoint regions against lineage with adequate numbers. Reported immunophenotype associations attach to KMT2A-rearranged status, not to where the gene broke. Treat all of these as open, not negative.

Some partners can only be reached by a complicated route

MLLT10 is transcribed toward the centromere while KMT2A is transcribed toward the telomere, so a simple reciprocal exchange cannot produce a working fusion. It takes at least three breaks, via an inversion followed by a translocation or an insertion. In a single-centre series screening 1,897 AML samples, 14 carried a 10;11 rearrangement producing this fusion, and five distinct mechanisms were observed 8.

This is a hard genomic constraint rather than a statistical tendency, and it has a diagnostic consequence: these rearrangements are often cryptic and can be missed by banding alone. A different gene, NEBL, sits about 570 kb away from MLLT10 and is indistinguishable from it by G-banding, so any 10;11 call needs molecular confirmation.

Correction: I attributed the spliced-fusion figure to the wrong partner

This entry originally stated that roughly half of KMT2A::AFF1 cases break upstream of AFF1 exon 1, producing a spliced fusion with no reciprocal allele. That is wrong, and wrong about the partner this whole project is built on.

Verified against the primary text: "Spliced fusions are rare events except for MLLT1 gene fusions (n = 460), where about 50% of the breakpoints (n = 241) localize upstream of MLLT1 exon 1." The full list, with denominators:

Partnerspliced / totalshare
MLLT1241 / 46052%
EPS1512 / 5124%
ELL10 / 1198%
PRPF193 / 3100%
AFDN2 / 1381.4%
MLLT32 / 6120.3%
AFF11 / 1,2620.08%

275 cases in total use this mechanism.

For AFF1 the figure is 1 in 1,262, essentially never. The ~50% belongs to MLLT1.

The error also conflated two distinct phenomena that happen to both concern reciprocal products:

  • A spliced fusion (Meyer): the break lies upstream of the partner's first exon, so no reciprocal allele is created. Common in MLLT1, vanishingly rare in AFF1.
  • A reciprocal allele that exists but is not expressed (Agraz-Doblas): in t(4;11), the reciprocal AFF1::KMT2A transcript is detected in only about 45% of patients, and its presence predicts markedly better survival.

The second finding stands and is the clinically important one. The first does not apply to AFF1. Reasoning downstream of this entry that treated the two as the same mechanism should be re-read.

Facts and figures
major bcrKMT2A intron 7 to exon 13
minor bcrKMT2A intron 20 to exon 24
cohort n3401
functional cut pointexon 11; breaks at or downstream destroy the PHD1-3 module and its CYP33/BMI1 switch
subregions
B679.0B: 679A1,761A: 1761C5,026C: 5026
subregions — the numbers behind the chart
[
  {
    "name": "A",
    "span": "exon 9 to intron 9",
    "bp": 1761,
    "share_pct": 37.2
  },
  {
    "name": "B",
    "span": "exon 10 to intron 10",
    "bp": 679,
    "share_pct": 19.8
  },
  {
    "name": "C",
    "span": "exon 11 to exon 13",
    "bp": 5026,
    "share_pct": 39.5
  }
]
evidence grades
{
  "position_vs_age": "B+ association, replicated across four cohorts, univariate only, partner-confounded",
  "position_vs_therapy_related": "B",
  "position_vs_lineage": "B-",
  "position_vs_outcome": "D, single study with 30 localised breakpoints, unreplicated",
  "position_vs_immunophenotype": "not tested",
  "position_vs_mpal": "not tested",
  "mllt3_partner_breakpoint_vs_lineage": "not established",
  "reciprocal_expression_vs_outcome": "B, adjusted Cox model in 124 uniformly treated infants"
}
confounded with
[
  "partner gene",
  "age",
  "geography",
  "detection method",
  "ascertainment"
]
How this could bite us back
  • If breakpoint position really does change biology, a single drug aimed at one break class could underperform in the others, and no current trial stratifies on it.
  • Building a stratification story on the intron-11 outcome claim would rest the programme on 30 patients from one unreplicated study.
Still open
  1. Interfant-06 holds genomic breakpoint sequences for 249 patients that were used as MRD targets but never modelled as a prognostic variable. Running that analysis would settle a 13-year-old claim and is arguably the single cheapest high-value study in this whole area.
  2. Can breakpoint position be separated from partner-gene identity at all, or are they so entangled that the question is unanswerable in existing cohorts?
  3. Does the exon-11 break class, which destroys PHD1-3, actually produce a different transcriptional output in matched cell models? Nobody appears to have compared isogenic fusions differing only in break position.
  4. What explains the 2017 versus 2023 disagreement about where the age crossover sits, 10-14 years versus 6 months?
  5. Does the reciprocal-fusion prognostic effect replicate in an independent cohort, and is it mediated by the partner-side breakpoint as the mechanism implies?
Every source behind this entry (9)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 3,401 patients, 98% of KMT2A breaks fall in a major cluster region (intron 7 to exon 13) and 1.4% in a minor region (intron 20 to exon 24), with subregion shares of 37.2%, 19.8% and 39.5%; the functional cut point is exon 11 because a break there destroys the PHD1-3 module; targeted sequencing raises the minor-region share to 6%; spliced fusions - a break upstream of the partner's exon 1, producing no reciprocal allele - occur in 1 of 1,262 KMT2A::AFF1 cases and in 241 of 460 KMT2A::MLLT1 cases; the authors state that whether breakpoint position affects clinical outcome is yet unclear.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    A referral-based cohort from a single diagnostic centre, so partner and breakpoint frequencies carry ascertainment bias. All breakpoint comparisons are univariate chi-square tests against a pooled mean distribution, never multivariable. Two detection methods with different sensitivities are pooled. The paper's own whole-cohort test found age class and disease subtype NOT significant, which sits awkwardly beside its within-lineage age modelling.

    Evidence that argues against this (1)
    • The same paper reports that apart from therapy-induced leukemia, none of gender, age class or disease subtype deviated significantly from the overall breakpoint distribution in the crude three-region test, which cuts against a simple age rule. Meyer C et al. 2023
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The KMT2A consensus gene structure comprises 38 exons, adding a 99-bp exon 2 and a previously unidentified exon inside intron 21, with the core of the major breakpoint cluster region between exons 9 and 12 (90.85% of breaks) and the minor region mostly between exons 21 and 25 (1.85%).

    PMID 38678092DOI
    What this evidence cannot show

    Contains no mapping between historical and current exon numbering, and no breakpoint-versus-age, lineage or outcome analysis. The newly added exon means minor-region exon numbers in earlier papers may be offset.

    Evidence that argues against this (1)
    • The 90.85% figure is computed over the breakpoints that were CHARACTERISED, which is a set assembled by assays designed around the major cluster region. Breaks outside it are systematically harder to find, so the concentration is an upper bound on itself. The consensus exon structure it uses also post-dates much of the literature it is being used to interpret. Blood Cells, Molecules and Diseases 2010
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Infant acute leukemias show the same 3'-biased distribution of KMT2A breaks as therapy-related leukemias, with 71.5% of infant cases in the telomeric region versus 33% of older de novo cases and 100% of therapy-related cases.

    PMID 9230194
    What this evidence cannot show

    Seventy patients by Southern blot and RT-PCR with only five therapy-related cases, using historical exon numbering. The shared distribution is circumstantial evidence for a shared cause; no exposure data were collected. A separate study found the therapy-related 3' clustering even in patients who had never received etoposide, which weakens the simple topoisomerase-II story.

    Evidence that argues against this (1)
    • In eleven t(9;11) cases, five secondary leukemias showed near-identical 3' KMT2A breakpoints despite no etoposide exposure during treatment of the primary malignancy, arguing the hot spot is not specifically drug-induced. Langer T et al. 2003
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 2,345 patients, breaks in intron 11 associate with lymphoblastic leukemia and younger age while intron 9 predominates in myeloid leukemia and older patients, with crossover at 10-14 years in ALL and 75 years in AML; the pattern reverses for MLLT10 and for adult MLLT3; a geographic shift of similar magnitude failed significance.

    The MLL recombinome of acute leukemias in 2017 · Meyer C, Burmeister T, Gröger D, et al. · Leukemia 2018
    PMID 28701730DOI
    What this evidence cannot show

    Univariate analysis only, same referral cohort lineage as the 2023 edition, and the reported crossover ages conflict with the later edition from the same group.

    Evidence that argues against this (1)
    • Age and lineage are confounded in this dataset, and both are confounded with era of diagnosis and with which assay characterised the breakpoint. A crossover at 10-14 years in ALL and 75 years in AML is derived from cell counts in age bins that get very thin at the extremes. Statistics in Medicine 2006
  5. E5weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    Patients with poor outcome preferentially displayed KMT2A breaks within intron 11, proposed as a novel outcome predictor.

    The distribution of MLL breakpoints correlates with outcome in infant acute leukaemia · Emerenciano M, Meyer C, Mansur MB, Marschalek R, Pombo-de-Oliveira MS · British Journal of Haematology 2013
    PMID 23432364DOI
    What this evidence cannot show

    This is the single source for the entire prognostic claim. Although the clinical cohort was 545 children, breakpoints were localised in only 30 cases, so any multivariable model is badly overfitted. Confounded with partner gene and with age. Never independently replicated; a search restricted to 2018 onward for KMT2A genomic breakpoint and outcome in infants returns only this paper. We could not access the full text, so the survival percentages, hazard ratios and model specification remain unverified.

    Evidence that argues against this (1)
    • The authors of the recombinome series, who propagate this claim, state in 2023 that whether breakpoint position affects clinical outcome is yet unclear, and perform no survival analysis of their own in any edition. Meyer C et al. 2023
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The Interfant-06 multivariable analysis (n=651) identified KMT2A rearrangement, age, white-cell count and prednisone response as prognostic factors, and did not include breakpoint position.

    PMID 31283407DOI
    What this evidence cannot show

    This is an argument from absence: the variable was not modelled, which is not the same as having been tested and found null. The trial was not designed to answer the breakpoint question.

    Evidence that argues against this (1)
    • A multivariable analysis not INCLUDING breakpoint position is not evidence that breakpoint position does not matter. Interfant-06 did not collect it in a form suitable for the model, so this is an absence of testing rather than a negative result, and citing it in a breakpoint entry risks being read as the latter. Statistics in Medicine 2006
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 124 infants with B-cell acute lymphoblastic leukemia treated on Interfant protocols, the reciprocal AFF1::KMT2A transcript was expressed in only 45% of t(4;11) cases, and its expression predicted four-year event-free survival of 62.4% versus 11.7%, retaining significance in a Cox model adjusted for risk stratification.

    PMID 30679323DOI
    What this evidence cannot show

    Only 43 t(4;11) patients were evaluable for the reciprocal transcript, so the survival split rests on modest numbers despite the adjusted model. This is a transcript-expression marker and not the same variable as KMT2A intron position.

    Evidence that argues against this (1)
    • Reciprocal transcript expression in 45% of cases is measured by RT-PCR, so it is presence of a TRANSCRIPT, not of a protein or an allele: a reciprocal allele can exist and not be expressed, or be expressed below the assay's threshold. The 62.4% versus comparison also divides 124 infants into two groups on a binary derived from a continuous readout. Statistics in Medicine 2006
  8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Because MLLT10 and KMT2A are transcribed in opposite orientations, a simple reciprocal translocation cannot generate the fusion; five distinct mechanisms were seen among 14 of 1,897 screened AML samples, and complex rearrangements are characteristic of this partner.

    PMID 12810254DOI
    What this evidence cannot show

    Single centre, and 14 positive cases is a small numerator even against a 1,897-sample denominator. No per-partner complexity rate has ever been published, so the claim that MLLT10 is the most complexity-prone partner rests on this series plus qualitative statements elsewhere.

    Evidence that argues against this (1)
    • Fourteen cases out of 1,897 screened. Five distinct mechanisms among 14 cases means most mechanisms rest on one or two patients, and complex rearrangements are exactly the class most likely to be mischaracterised by any single technique. Blood Cells, Molecules and Diseases 2010
  9. E_corrstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Spliced fusions lacking a reciprocal allele occur in 241 of 460 KMT2A::MLLT1 cases but in only 1 of 1,262 KMT2A::AFF1 cases.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    Referral-based cohort; denominators reflect which centres send samples.

    Evidence that argues against this (1)
    • The contrast is striking and the denominators are not comparable: 460 KMT2A::MLLT1 and 1,262 KMT2A::AFF1 cases were accumulated by different laboratories with different reasons for characterising the reciprocal allele. A 241/460 versus 1/1262 split that large is unlikely to be pure ascertainment, but part of it will be. Blood Cells, Molecules and Diseases 2010
Connected to