Oncofusion

start / graph / hyp-pentimento-ground-predicts-post-drug-destination

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q12-analogy

Hypothesis: where a blast goes when the fusion program is released is written in its methylome at diagnosis, not in its transcriptome

When a drug switches off the leukemia program, the cell goes somewhere. We think it goes wherever its original identity was set before it became cancerous, and that identity is recorded in a chemical marking on the DNA that the cancer never erased.

The same thing, for a scientist

Menin inhibition releases the differentiation block; the graph already asks where the released cells go. We add a claim about the *predictor*. Because DNA methylation is mitotically heritable and carries a recoverable record of the originating cell — the basis of methylome-based cell-of-origin modelling [E2] — while the transcriptome is the thing the drug is about to remove, we hypothesise that a cell-of-origin score computed from the diagnostic methylome predicts the post-treatment lineage destination better than any diagnostic transcriptomic signature does. Concretely: in KMT2A::AFF1 cases whose methylome-derived origin sits earlier in the hierarchy (lympho-myeloid primed progenitor rather than committed pro-B), menin inhibition should produce more myeloid-directed differentiation and a higher risk of myeloid lineage switch. This is our inference from a conservation-science frame applied to published methylome biology; it has not been tested.

Go deeper
What is new here relative to what the graph already holds

hyp-menin-release-lineage-destination asks where released cells go, with exp-menin-inhibitor-lineage-destination as its test. hyp-methylome-of-origin-sets-fusion-target-set asks whether the origin methylome sets HOXA-high versus HOXA-low at diagnosis. Neither asks which measurement, taken at diagnosis, predicts the destination under drug.

That is the contribution: a head-to-head predictor comparison — methylome-derived origin score versus diagnostic transcriptomic signature — with destination as the outcome. It is a small, specific addition, and it is bolted onto an experiment the graph already specifies rather than requiring a new one.

The competing account, which is well supported

Lineage switching in MLL/AF4 leukemia involves substantial chromatin-accessibility change and transcriptional rewiring recurrently associated with altered chromatin modifiers, notably CHD4, and perturbing CHD4 alone induces myeloid gene expression in MLL/AF4 models 3. That is active reprogramming driven by an acquired lesion, not a fixed ground being revealed.

If the CHD4 account is right, destination is set by an acquired chromatin-modifier state that could change at any point, and no diagnostic measurement will predict it. That is a serious and likely possibility, and it is the reason the prior here is below 0.5. The two accounts are not mutually exclusive — the ground could set a default that an acquired lesion can override — and a well-designed comparison should be able to apportion between them rather than declare a winner.

The practical version, if it holds

A methylation array on the diagnostic sample is cheap, already performed in many paediatric protocols for classification, and needs no new tissue. A score derived from it that flags children at elevated risk of myeloid switch under menin inhibition or under CD19-directed therapy would be usable immediately, because the mitigation is already known and cheap: full multi-lineage immunophenotyping and fusion-transcript monitoring rather than B-gated flow, concentrated in the first six months.

That is a low-cost, low-harm application of a speculative hypothesis, which is the kind worth pursuing even at a prior below 0.5.

Facts and figures
statementA cell-of-origin score computed from the diagnostic DNA methylome predicts the lineage destination of KMT2A::AFF1 blasts after release of the differentiation block (by menin inhibition, and by CD19-directed therapy in the case of lineage switch) better than any diagnostic transcriptomic signature, because methylation is a mitotically heritable record of the pre-transformation state while the transcriptome is the fusion program the therapy removes.
originanalogy
source analogyanalogy-art-pentimento-and-restoration
prior probability0.3
prior reasoningLowered deliberately from an initial intuition of about 0.5 by the CHD4 result, which shows destination can be set by an acquired chromatin-modifier lesion rather than by an inherited ground. We also lack a fetal methylation reference of the quality the method needs. The hypothesis is retained at 0.3 because the test is cheap, bolts onto an experiment already specified in the graph, and has a low-harm clinical application even if the effect is modest.
decision valueA diagnostic methylation array is already run in many paediatric protocols. If the score works, flagging children at elevated switch risk costs nothing and the mitigation (multi-lineage immunophenotyping and fusion-transcript monitoring in the first six months) is already established.
predictions
[
  "In matched diagnostic samples, a methylome-derived cell-of-origin score outperforms a transcriptome-derived signature in predicting post-menin-inhibition lineage destination by area under the curve.",
  "Cases whose methylome-derived origin sits earlier in the haematopoietic hierarchy show more myeloid-directed differentiation under menin inhibition.",
  "The same methylome score is elevated in patients who subsequently undergo lineage switch after CD19-directed therapy, measured retrospectively on banked diagnostic samples.",
  "Methylation at the relevant loci is stable between diagnosis and relapse in the same patient, as a ground should be, whereas the transcriptomic signature is not.",
  "CHD4 status at diagnosis adds independently to the methylome score, consistent with the ground setting a default that an acquired lesion can override."
]
experiments
[
  "exp-menin-inhibitor-lineage-destination"
]
analyses
[
  "ana-pentimento-methylation-ground-destination",
  "ana-lineage-switch-signature-scoring"
]
falsifiers
[
  "The transcriptomic signature predicts destination as well as or better than the methylome score, removing the reason to prefer the ground.",
  "Methylation at the informative loci changes substantially between diagnosis and relapse, disqualifying it as a stable ground.",
  "Destination is fully explained by CHD4 or another acquired chromatin-modifier lesion, with no residual methylome contribution.",
  "No fetal haematopoietic methylation reference of sufficient quality can be assembled, making the score unconstructable — a practical rather than scientific refutation, and it should be reported as such."
]
How this could bite us back
  • A score that flags high switch risk could be used to withhold CD19-directed therapy, which in this population would trade a 0-28.6% switch risk against a therapy that took 4-year disease-free survival from 44.0% to 83.3%. Any clinical use must be restricted to intensifying monitoring, not to withholding treatment.
  • If destination is set by the ground, then some children cannot be safely differentiated at all, and the entire differentiation-therapy logic of this drug class has a subgroup in which it is contraindicated and which nobody currently identifies.
Still open
  1. Does a fetal liver and fetal bone marrow haematopoietic methylation reference exist at array or bisulphite-sequencing resolution, and if not, who holds the tissue to build one?
  2. Can the ground contribution and the acquired-lesion contribution be apportioned in the same model, rather than treated as competing explanations?
  3. Is methylation at the informative loci actually stable across relapse in KMT2A-rearranged disease, or does chemotherapy move it?
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Infant B-ALL including MLL-rearranged disease has an exceptionally silent DNA mutational landscape consistent with prenatal origin, with methylome signatures in 69 uniformly treated infants mapping onto chromatin states related to enhancer and transcriptional control in normal haematopoietic cells.

    PMID 33983906DOI
    What this evidence cannot show

    Establishes that methylation carries discriminating signal in exactly this disease; it does not perform cell-of-origin inference and does not examine behaviour under therapy.

    Evidence that argues against this (1)
    • REFUTES THE PREMISE THAT THE GROUND IS FIXED. The whole force of the analogy rests on methylation being a stable, mitotically heritable record of the pre-transformation state, in contrast to the transcriptome that therapy removes. Paired diagnosis-remission-relapse analysis of paediatric T-ALL by whole-exome sequencing, targeted ultra-deep sequencing and DNA methylation array found that relapse is characterised by relapse-specific *promoter hypomethylation* alongside acquired mutations in epigenetic modulators including SUZ12, WHSC1 and SMARCA4. The methylome is therefore itself remodelled between diagnosis and relapse under therapy, which is exactly the interval over which a diagnostic cell-of-origin score is supposed to remain predictive. In conservation terms, the ground is being repainted while the overpaint is being cleaned. Caveat: T-ALL, not KMT2A-rearranged B-ALL, n=13, and array rather than whole-genome bisulfite; the equivalent paired analysis in infant KMT2A::AFF1 has not been done, which is itself the gap this hypothesis should now be built around. Kunz JB et al. 2015
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Methylome-based cell-of-origin modelling decomposes a cancer epigenome into the pattern inherited from the tumour-initiating cell and tumour-acquired changes, mapping individual patients onto a normal differentiation continuum, and is proposed by its authors as generalisable beyond CLL.

    PMID 32188505DOI
    What this evidence cannot show

    Built on a mature B-cell reference. A fetal haematopoietic methylation reference of comparable quality may not exist, and without one the method cannot simply be lifted across.

    Evidence that argues against this (1)
    • WHERE THE ANALOGY BREAKS, in the source domain. The conservator's problem is not that the ground is readable and the overpaint is not; it is that no imaging modality separates the layers cleanly. Gooch and Tumblin, working on exactly this task, note that X-ray interpretation of pentimenti is hard because X-rays superimpose features from all paint layers and because image intensity depends jointly on paint-layer thickness and each pigment's opacity, so recovering the earlier layer requires user-guided inference rather than measurement. Methyl-COOM has the same structure: it is a regression that *assumes* the observed methylome decomposes into an inherited component plus tumour-acquired changes, and it was fitted and validated in CLL, a mature, indolent, slowly proliferating B-cell malignancy with a well-mapped normal differentiation continuum to regress against. Infant KMT2A::AFF1 has neither the indolence nor an agreed normal reference continuum, and no published application of the method to an acute leukaemia was found. The decomposition may simply not be identifiable here. Verification note: Verified from the publisher's listing and abstract (title, authors, journal, volume/issue, year, DOI); the full text was not retrievable without a subscription, so the claim used here is the abstract's own statement about layer superimposition and thickness/opacity confounding. Gooch AA et al. 2007
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Lineage switch in MLL/AF4 leukemia arises from myeloid relapses sharing fusion breakpoints with matched lymphoid presentations and originating from a range of differentiation stages, and involves chromatin-accessibility and transcriptional rewiring recurrently associated with altered chromatin modifiers including CHD4, whose perturbation alone induces myeloid gene expression.

    Epigenetic regulator genes direct lineage switching in MLL/AF4 leukemia · Tirtakusuma R, Szoltysek K, Milne P, et al. · Blood 2022
    PMID 35839448DOI
    What this evidence cannot show

    The strongest evidence against this hypothesis as stated, and cited here for that purpose. The range of originating differentiation stages is, however, consistent with a ground effect operating alongside the acquired one.

    Evidence that argues against this (1)
    • Read against the hypothesis rather than for it, this study says the destination is set by lesions acquired at or near switching, not by the diagnostic ground: switching was recurrently associated with altered expression, splicing or mutation of chromatin modifiers, notably CHD4, and perturbing CHD4 alone induced myeloid gene expression in MLL/AF4 models. A diagnostic methylome cannot predict a CHD4 lesion that has not yet occurred. The hypothesis survives only in the weaker form that the diagnostic ground constrains *which* destinations are reachable, not which is taken — and that weaker form makes a much less useful clinical prediction. Tirtakusuma R et al. 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Post-immunotherapy lineage switch is concentrated in KMT2A-rearranged disease (64.3% of a 70-case cohort), occurs at a median of 1.5 months with 81.4% inside six months, and more than half of affected patients had achieved MRD-negative complete remission of their B-ALL beforehand.

    PMID 40193715DOI
    What this evidence cannot show

    Describes switch after immunotherapy rather than after menin inhibition; whether the two share a destination-setting mechanism is assumed here, not shown.

    Evidence that argues against this (1)
    • A competing predictor of lineage destination is already on record, and it is genetic rather than methylomic. Whole-genome phylogenetic reconstruction of a post-CD19 lineage switch showed the AML derived from the initial ALL and carried *distinct driver mutations that were detectable before the AML emerged*. That is a diagnostic-sample predictor of destination — the thing this hypothesis claims only the methylome can supply — obtained from DNA sequence. The hypothesis's specific claim is comparative ('better than any diagnostic transcriptomic signature'), and the comparison it actually needs to win is against pre-existing subclonal drivers, which it does not currently name as a competitor. Caveat: n=1 and KMT2A wild-type. Coorens THH et al. 2023
Connected to