Oncofusion

start / graph / ana-interfant-breakpoint-prognostic-reanalysis

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q9-landscape

Analysis: is the KMT2A genomic breakpoint a prognostic variable? A reanalysis of data Interfant and Frankfurt already hold

also called breakpoint prognostic reanalysis · 249-patient proposal

Two groups already hold everything needed to answer whether the exact position of the DNA break changes how a child does. One holds the sequences, the other holds the outcomes, and nobody has put them together.

The same thing, for a scientist

A retrospective, computation-only analysis proposal. Interfant-06 generated patient-specific genomic KMT2A breakpoint information for 249 infants as a technical prerequisite for DNA-based MRD assays, and holds protocol-standardised outcome and MRD data for the same patients. The Frankfurt DCAL series holds nucleotide-resolution breakpoints for 3,401 patients with age and partner-gene annotation but without systematic outcome linkage. The analysis tests breakpoint position, retained/lost domain content, junction microhomology signature and partner-gene intron as predictors of event-free survival, MRD kinetics and relapse phenotype, with no new samples, no new patients and no assay development.

Go deeper
Design1 deeper

Exposure variables, derived entirely from sequence:

  • breakpoint position within the KMT2A breakpoint cluster region (continuous, and as intron 9 / intron 11 / other)
  • predicted retained versus lost KMT2A domain content (AT-hooks, CXXC, menin-binding motif), computed per patient
  • partner gene and partner intron
  • junction microhomology length and insertion content, as a proxy for NHEJ versus alt-EJ formation mechanism
  • repeat content (Alu and other) flanking the junction
  • presence or predicted absence of a reciprocal product

Outcomes, already collected: event-free and overall survival, end-of-induction and end-of-consolidation MRD level, relapse timing, and relapse lineage where recorded.

Covariates: age at diagnosis (critical, because Frankfurt reports an age-dependent breakpoint shift, so breakpoint and age are confounded by construction), white cell count, prednisone response, CNS status, treatment arm.

Primary test: does breakpoint position add prognostic information beyond age, white cell count and MRD? The age adjustment is the analysis's central methodological problem, not a footnote.

Power: 249 infants with roughly 40-60% events is adequate for a moderate effect on a binary breakpoint classification (intron 9 versus intron 11) and underpowered for fine continuous position. The Frankfurt cohort at 3,401 is far better powered but lacks outcomes, which is precisely why the two must be used together: Frankfurt to establish which breakpoint classes are biologically distinct and sufficiently frequent, Interfant to test those pre-specified classes against outcome.

Why it has not been done, and what could stop it

Not because it is hard. Because the sequences and the outcomes live in different institutions with different governance, and because the breakpoint was treated as assay plumbing rather than as data.

The three things that could stop it, in order of likelihood:

  1. The Interfant breakpoint sequences were not archived. If MRD assay design consumed them and only primer sequences were retained, primers still encode approximate breakpoint position, which is a degraded but usable exposure variable. Worth asking about explicitly.
  2. Consent and governance across 18 national study groups. A secondary analysis of existing molecular data is usually within scope, but 'usually' is doing work in that sentence.
  3. Age confounding is irreducible. If breakpoint position and age at diagnosis are too tightly coupled in infants (a narrow 0-12 month range, where the Frankfurt shift is steepest), the analysis may be unable to separate them at n=249. This should be assessed on the Frankfurt data first, before requesting anything from Interfant.
Facts and figures
scriptanalyses/interfant-breakpoint-prognostic/run.py
statusplanned
result summaryNot run. Blocked on confirming that Interfant-06 breakpoint sequences were archived and on data-access agreements with two institutions.
cost tierlow
statistical planPre-specify breakpoint classes on the Frankfurt data before seeing any Interfant outcome. Cox proportional hazards for EFS with age, WBC, prednisone response and treatment arm as covariates; ordinal logistic for MRD category. Assess breakpoint-age collinearity (VIF, and stratified analysis within narrow age bands) as a gating step. Pre-register the analysis plan.
first actionWrite to the Interfant coordinating centre at the Princess Maxima Center with one question: were the patient-specific genomic KMT2A breakpoint sequences generated for MRD assay design in Interfant-06 archived, and in what form?
inputs
[
  "Interfant-06 KMT2A-rearranged cohort: patient-specific genomic breakpoint sequences or MRD primer sequences (n=249), with EFS/OS, EOI and EOC MRD, age, WBC, prednisone response, treatment arm (access required from the Interfant consortium)",
  "Frankfurt DCAL breakpoint catalogue (n=3,401) for breakpoint-class definition and age-confounding assessment (access required from DCAL)",
  "KMT2A domain annotation (UniProt Q03164) for computing retained/lost domain content per breakpoint"
]
How this could bite us back
  • A null result is entirely plausible and would close one of the lab's better differentiating angles, though it would still be publishable and would settle a twenty-year-old question.
  • Age confounding could make the analysis uninterpretable rather than negative, which is the worst outcome: effort spent with nothing concluded.
  • If breakpoint position DOES predict outcome, the implication is that a single menin-interface drug will not serve all KMT2A-r patients, which complicates rather than supports the platform argument the lab is making to funders.
Still open
  1. Were the Interfant-06 breakpoint sequences archived, or only the derived primers? This one administrative fact gates the whole proposal.
  2. Is breakpoint position separable from age at diagnosis within a 0-12 month cohort, or are they collinear to the point of being one variable?
  3. Should the analysis be run first on the Frankfurt cohort against whatever partial outcome information exists there, as a lower-governance pilot, before approaching Interfant?
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Interfant-06 measured MRD in 249 infants with KMT2A-rearranged ALL by DNA-based PCR of rearranged KMT2A, immunoglobulin and/or T-cell receptor genes, with 6-year disease-free survival of 60.2%, 45.0% and 33.8% by end-of-induction MRD category, establishing both the cohort size and the existence of protocol-quality outcome data.

    PMID 33405950DOI
    What this evidence cannot show

    That patient-specific genomic breakpoint sequences were generated AND archived is inferred from the DNA-based KMT2A MRD methodology, not stated in the paper. This inference is the single point of failure for the whole proposal and must be confirmed before anything else.

    Evidence that argues against this (1)
    • MRD strata are an observational association, not a randomised comparison: patients with high end-of-induction MRD differ from low-MRD patients in disease biology before any treatment decision, so the 60.2/45.0/33.8% gradient measures prognosis rather than what changing treatment on MRD would achieve. DNA-PCR MRD in KMT2A-rearranged infants also often uses the fusion itself as the target, whose sensitivity differs from immunoglobulin targets, so the strata are not measured identically across patients. Statistics in Medicine 2006
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The Frankfurt DCAL series provides nucleotide-resolution breakpoints for 3,401 patients with partner-gene annotation, and reports an age-dependent shift in breakpoint position within the KMT2A breakpoint cluster region, which establishes both the reference dataset for defining breakpoint classes and the confounding structure the analysis must handle.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, ... Marschalek R · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    Referral cohort, so breakpoint class frequencies are subject to referral bias; classes defined on this dataset may not have the same frequencies in a trial cohort.

    Evidence that argues against this (1)
    • A prior attempt at this exact analysis already exists and must be the comparator, not ignored. The Brazilian Collaborative Study Group mapped genomic breakpoints in 30 MLL-rearranged infant leukaemia cases within a 545-patient cohort and reported that poor-outcome patients preferentially carried breakpoints within MLL intron 11. That both sets a prior for the effect direction and creates a multiple-testing and publication-bias problem: an intron-11 cut point chosen after seeing that paper is no longer a pre-specified hypothesis. The analysis plan should pre-register the intron-11 threshold as a confirmatory test and treat any other cut point as exploratory. Emerenciano M et al. 2013
Connected to