Oncofusion

start / graph / hyp-r-loop-landscape-kmt2a-r

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q10-genome

Hypothesis: the R-loop landscape of a KMT2A::AFF1 blast is remodelled by the fusion and is both a vulnerability and a source of ongoing genome instability

also called R-loops in KMT2A-r · RNA:DNA hybrid landscape MLL-AF4 · transcription-associated genome instability KMT2A-r

When a gene is read, the RNA can stay stuck to the DNA and form a three-stranded knot called an R-loop. Fusion proteins force genes to be read very hard, so these knots should pile up - but they pile up in any hard-read gene in any cell, the main way of measuring them is disputed, and nobody has measured them in the infant B-cell form of this leukemia at all.

The same thing, for a scientist

KMT2A fusions hijack transcriptional elongation (SEC/DOT1L, P-TEFb) and drive hyper-elongation at a defined target set. Hyper-elongation and high transcriptional output are the canonical conditions for R-loop accumulation. Two independent lines make the prediction concrete rather than vague: in leukemia, HOTTIP-dependent R-loops at CTCF-binding sites are required for boundary activity, TAD integrity and promoter/enhancer interaction, and RNase H targeting to one boundary was enough to impair leukemogenesis [E1]; and R-loop accumulation drives transcription-associated double-strand breaks, replication stress and an inflammatory transcriptome, with germline loss of the R-loop resolvase DDX41 predisposing to AML [E2]. **Correction and downgrade, cycle 35.** This node previously asserted that no DRIP-seq, R-ChIP or S9.6 map of any KMT2A-rearranged leukaemia had been published. That is wrong: Fu et al. (PMID 41430607) report DRIP-seq in MLL-rearranged AML cells alongside Hi-C, m6A-seq and CTCF ChIP-seq. The surviving, narrower gap is that **no R-loop map of a KMT2A::AFF1 B-ALL line or of any primary KMT2A-rearranged infant ALL sample has been published that we can find**, and none of the KMT2A-r breakpoint literature measures R-loops at the KMT2A breakpoint cluster region. Two further cautions now sit on the hypothesis. R-loops are a ubiquitous physiological feature occupying up to 5% of mammalian genomes and forming co-transcriptionally at conserved promoter and terminator hotspots [E3 counter, Sanz 2016], so nothing follows from their mere presence in a blast. And the assays disagree with each other: S9.6-based and dRNH-based maps diverge starkly in size, location and factor colocalisation, a large fraction of published R-loop data fails quality control, and S9.6 imaging signal is predominantly ribosomal RNA [E1, E2 counters]. This remains a hypothesis, and it is now a hypothesis with a method problem as well as a data gap.

Go deeper
Why the prediction is sharp rather than hand-waving1 deeper

Three specific, falsifiable predictions follow.

P1 — Target-set enrichment. R-loops should be enriched over KMT2A-fusion spreading targets (the gene-body-spreading class of genome-enhancer-promoter-contacts-kmt2a-r) relative to matched non-target genes of equal expression. If R-loops simply track expression level with no fusion-target enrichment, the hypothesis is uninteresting.

P2 — Boundary dependence. R-loops should be present at the CTCF boundaries of the HOXA domain in KMT2A::AFF1 cells, as they are in MLL-r AML 1, and dissolving them should collapse HOXA expression.

P3 — DOT1L/menin dependence. If the R-loops are a consequence of fusion-driven elongation, DOT1L inhibition or menin inhibition should reduce them at target loci. If they persist after the fusion program is switched off, they are a property of the cell of origin, which would be a different and arguably more interesting result.

The instability arm, and its uncomfortable implication

If KMT2A-r blasts carry a high R-loop burden, they carry an ongoing source of double-strand breaks 2. That cuts two ways for the lab's central hypothesis.

In favour of therapy: a cell already loaded with R-loop-associated replication stress should be unusually sensitive to ATR, CHK1 or PARP inhibition, and to anything that blocks R-loop resolution (a synthetic-lethal angle that would be very welcome in a disease with almost no co-mutations to target, see comut-low-mutation-burden-infant-kmt2a-r).

Against: an ongoing break-generating state in a cell whose defining lesion is a translocation means secondary rearrangement is a live possibility during treatment, including the acquisition of new fusions or of the lineage-switch-associated changes seen at relapse. Treating a genome-instability phenotype with a genotoxic backbone, which is what current infant ALL protocols are, may be actively selecting for that.

Status of the evidence: honest grading
  • Direct KMT2A-r evidence: thin, and not in the disease of interest. DRIP-seq in MLL-rearranged AML cell lines exists (PMID 41430607), from the same group as E1. There is no R-loop map of a KMT2A::AFF1 B-ALL line and none of any primary infant KMT2A-r sample.
  • Indirect leukemia evidence: yes, but single-source. HOTTIP-dependent R-loops maintain CTCF boundaries and TAD integrity in MLL-rearranged and NPM1-mutant AML, with a functional RNase H experiment 1; the DRIP-seq above is from the same laboratory group. DDX41, an R-loop-opposing helicase, is a germline AML predisposition gene 2.
  • Mechanistic plausibility: high but non-specific. R-loop formation is a known consequence of the transcriptional state KMT2A fusions create - and of high transcription in any proliferating cell. R-loops occupy up to 5% of mammalian genomes [E3 counter]. Enrichment relative to an expression-matched control set is the only informative measurement.
  • Methodological status: contested. S9.6-based and dRNH-based R-loop maps diverge substantially; much published R-loop data does not pass quality control; S9.6 imaging is dominated by rRNA, although S9.6 DRIP-seq remains RNase H1-sensitive [E1, E2 counters].

Revised cheapest decisive experiment. Not a single DRIP-seq. Two orthogonal modalities in parallel in two KMT2A::AFF1 lines - S9.6-based DRIP-seq with an RNase H1-treated control library, and a dRNH-based method (R-ChIP, MapR or qDRIP) - scored against an expression-matched non-target gene set, with DOT1L-inhibitor and menin-inhibitor arms. A signal seen by only one modality is not a result.

Facts and figures
statementKMT2A::AFF1-driven transcriptional hyper-elongation remodels the R-loop landscape of the blast, producing (a) R-loops at fusion target gene bodies and at HOXA-domain CTCF boundaries that are functionally required for the leukemic program, and (b) a chronic transcription-associated DNA-break burden that is both a synthetic-lethal vulnerability and a driver of secondary rearrangement.
originliterature-gap
prior probability0.4
method cautionAny R-loop measurement in this project must report an RNase H1-treated control and, where the claim is load-bearing, a second modality of a different chemical class.
planned experiments noteMoved out of a reference field on 2026-09-19. It named a node that does not exist, which made the pointer read as a real cross-reference and, for hypotheses, silently satisfied the 'needs at least one experiment or analysis' rule. The intent is preserved here; the node still needs writing.
predictions
[
  "Two orthogonal R-loop modalities (S9.6 DRIP-seq with an RNase H1-treated control library, and a dRNH-based method such as R-ChIP/MapR/qDRIP) AGREE in calling R-loop enrichment at KMT2A-fusion spreading targets in KMT2A::AFF1 lines",
  "That enrichment exceeds what an expression-matched non-target gene set predicts; enrichment that merely tracks transcription level is not a result",
  "R-loops are detectable at HOXA-domain CTCF boundaries in a KMT2A::AFF1 B-ALL line as they are in MLL-r AML, and their removal by targeted RNase H1 collapses HOXA expression",
  "DOT1L inhibition (EPZ-5676) or menin inhibition (revumenib) reduces target-locus R-loops; failure to reduce them would indicate a cell-of-origin rather than fusion-driven landscape",
  "KMT2A::AFF1 lines are more sensitive than KMT2A-wild-type leukemia lines to ATR or CHK1 inhibition, and this sensitivity is rescued by RNase H1 overexpression",
  "Senataxin (SETX), DDX41 or RNASEH2 knockdown is selectively toxic in KMT2A::AFF1 lines relative to matched KMT2A-wild-type lines"
]
analyses
[
  "ana-noncoding-gap-ledger-kmt2a-r"
]
falsifiers
[
  "S9.6-based and dRNH-based maps of the same KMT2A::AFF1 cells disagree about which loci carry R-loops - in which case the hypothesis is untestable with current methods rather than false, and should be parked",
  "R-loop enrichment at fusion target loci is fully explained by transcription level in expression-matched controls",
  "RNase H1 overexpression in KMT2A::AFF1 cells does not change HOXA9/MEIS1 expression or proliferation",
  "KMT2A::AFF1 lines are no more sensitive to ATR/CHK1 inhibition than matched KMT2A-wild-type leukemia lines in DepMap or in a head-to-head dose-response",
  "R-loop levels at target loci are unchanged by DOT1L or menin inhibition, which would decouple the landscape from the fusion program"
]
evidence grade by class
{
  "direct_kmt2a_r": "weak (DRIP-seq in MLL-r AML cell lines only, PMID 41430607, same group as E1)",
  "direct_kmt2a_aff1_b_all": "none",
  "indirect_leukemia": "moderate (single laboratory group)",
  "mechanistic_plausibility": "high but non-specific; R-loops occupy up to 5% of mammalian genomes",
  "method_status": "contested; S9.6 vs dRNH maps diverge starkly (PMID 35758606) and S9.6 imaging is dominated by rRNA (PMID 33830170)"
}
prior probability history
[
  {
    "cycle": 1,
    "value": 0.55,
    "basis": "strong indirect leukaemia evidence plus high mechanistic plausibility"
  },
  {
    "cycle": 35,
    "value": 0.4,
    "basis": "downgraded: R-loops are a general property of transcribing cells (PMID 27373332), and the only available read-outs are contested and mutually discordant (PMID 33830170, PMID 35758606). The fusion-specific arm is what lost probability; the existence of R-loops in these cells is not in doubt."
  }
]
corrections
[
  {
    "cycle": 35,
    "date": "2026-09-19",
    "withdrawn": "'no DRIP-seq, R-ChIP or S9.6 map of any KMT2A-rearranged leukemia primary sample or KMT2A::AFF1 line has been published that we can find'",
    "replaced_with": "no R-loop map of a KMT2A::AFF1 B-ALL line or of any primary KMT2A-rearranged infant ALL sample has been published; DRIP-seq in MLL-rearranged AML cell lines does exist",
    "source": "PMID 41430607 (Fu 2025, Molecular Cancer)"
  }
]
planned experiments
[
  "exp-noncoding-dependency-crispr-screen-kmt2a-aff1"
]
How this could bite us back
  • If the blast already carries a high break burden, genotoxic therapy may select for secondary rearrangements and lineage switch rather than eradicating disease.
  • R-loop-resolving enzymes are required in normal cells; inhibiting resolution as a therapeutic strategy risks genome instability in normal fetal and infant progenitors, the worst possible tissue to destabilise.
  • A synthetic-lethal ATR/CHK1 strategy would add myelosuppression to a population already heavily transfused, and ATR inhibitors have never been given to infants.
  • Reliance on a single group's RNase H boundary result means the central mechanistic premise could fail replication.
  • The whole hypothesis is currently readable only through an assay whose antibody cross-reacts with rRNA and whose calls disagree with the orthogonal method; a positive result from one modality would be the kind of finding this field has repeatedly failed to replicate.
Still open
  1. The MLL-r AML DRIP-seq now exists (PMID 41430607). Is there any R-loop map of a KMT2A::AFF1 B-ALL line or of a primary infant KMT2A-r sample? We could find none; this narrowed claim of absence should be rechecked each cycle, including preprints and supplementary datasets.
  2. Are R-loops present at the KMT2A breakpoint cluster region in fetal haematopoietic progenitors, which would connect this hypothesis to fusion genesis rather than only to fusion maintenance?
  3. Does the infant KMT2A-r genome show a mutational signature of transcription-associated damage, which would be detectable in existing whole-genome data without any new experiment?
  4. If R-loops maintain the boundary, does the standard genotoxic chemotherapy backbone transiently increase or decrease them, and could sequencing of agents matter?
  5. Do S9.6-based and dRNH-based R-loop maps of the same leukaemia cells agree at all? Miller et al. say they diverge starkly in general. If they diverge in a KMT2A-r line, this hypothesis is not testable with current methods and the honest action is to park it rather than to run one assay and believe it.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In MLL-rearranged and NPM1-mutant AML, HOTTIP recruits CTCF/cohesin and R-loop regulators to form R-loops at a subset of CTCF-binding sites; deleting the boundary site or targeting RNase H to dissolve the R-loop impaired boundary activity, inhibited promoter/enhancer interactions and mitigated leukemogenesis in xenografts.

    HOTTIP-dependent R-loop formation regulates CTCF boundary activity and TAD integrity in leukemia · Luo H, Zhu G, Eshelman MA, Fung TK, Lai Q, Wang F, Zeisig BB, Lesperance J, Ma X, Chen S, Cesari N, Cogle C, Chen B, Xu B, Yang FC, So CWE, Qiu Y, Xu M, Huang S · Molecular Cell 2022
    PMID 35180428DOI
    What this evidence cannot show

    AML, not KMT2A::AFF1 B-ALL; the decisive experiments are at the beta-catenin TAD, not HOXA; single group and not independently replicated.

    Evidence that argues against this (2)
    • The assay is contested. S9.6, the antibody underlying DRIP and its derivatives, has substantial affinity for double-stranded RNA; Smolka et al. show that S9.6 immunofluorescence signal in fixed human cells comes predominantly from ribosomal RNA, is insensitive to RNase H1 and is sensitive to RNase T1. They do also show that S9.6 DRIP *sequencing* maps stay RNase H1-sensitive and RNase T1-insensitive, so DRIP-seq survives - but only with those controls reported. Any R-loop claim in this beat, including the boundary R-loop in E1, must state its RNase H1 control or be graded down. Smolka JA et al. 2021
    • This also corrects a claim of absence in the node's own body. Fu et al. performed DRIP-seq in MLL-rearranged AML cells as part of a multi-omics study, reporting that METTL3/YTHDC1-dependent m6A on architectural RNAs stabilises R-loops that sustain CTCF-dependent TAD boundary activity. An R-loop map of a KMT2A-rearranged leukaemia therefore exists. It is AML, it is cell lines, and it is from the same laboratory group as E1, so the KMT2A::AFF1 infant B-ALL gap is real - but the broader statement 'no R-loop map of any KMT2A-rearranged leukaemia' was wrong and has been withdrawn. Fu R et al. 2025
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    R-loop accumulation causes transcription-associated double-strand breaks, replication stress and an inflammatory transcriptome; the tumour suppressor DDX41, which is enriched at promoters and unwinds RNA-DNA hybrids in vitro, opposes this, and germline loss-of-function DDX41 mutations predispose to adult AML.

    R-loop proximity proteomics identifies a role of DDX41 in transcription-associated genomic instability · Mosler T, Conte F, Longo GMC, Mikicic I, Kreim N, Möckel MM, Petrosino G, Flach J, Barau J, Luke B, Roukos V, Beli P · Nature Communications 2021
    PMID 34916496DOI
    What this evidence cannot show

    Worked in HEK293 and cancer cell lines, not in leukemic blasts; the AML link is a genetic inference from DDX41 predisposition rather than a demonstration in leukemia cells. Says nothing about KMT2A fusions.

    Evidence that argues against this (1)
    • R-loop mapping has a documented reproducibility problem. Reprocessing 810 datasets, Miller et al. built a data quality-control method, revealed 'the extent of poor-quality data within previously published studies', and found a 'stark divergence' between R-loop regions called by S9.6-based and by dRNH-based methods in size, location and colocalisation with RNA-binding factors. A hypothesis whose only read-out is a contested assay cannot be resolved by one modality; the diagnostic experiment in this node has been changed accordingly. Miller HE et al. 2022
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    KMT2A-fusion binding spreads into the gene body at a subset of targets with increased H3K79me2/3 and disrupted H3K36me3, defining the hyper-elongated transcriptional state in which R-loops would be predicted to accumulate.

    MLL-AF4 Spreading Identifies Binding Sites that Are Distinct from Super-Enhancers and that Govern Sensitivity to DOT1L Inhibition in Leukemia · Kerry J, Godfrey L, Repapi E, Tapia M, Blackledge NP, Ma H, Ballabio E, O'Byrne S, Ponthan F, Heidenreich O, Roy A, Roberts I, Konopleva M, Klose RJ, Geng H, Milne TA · Cell Reports 2017
    PMID 28076791DOI
    What this evidence cannot show

    This paper does not measure R-loops at all. It is cited to establish the transcriptional state, and the R-loop inference from that state is entirely ours.

    Evidence that argues against this (1)
    • R-loops are a general property of transcribing cells, not a special feature of hyper-elongation. Sanz et al. profiled R-loops at high resolution in human and mouse cells and found them 'prevalent, collectively occupying up to 5% of mammalian genomes', forming co-transcriptionally over conserved promoter and terminator hotspots of poly(A)-dependent genes and turning over dynamically. Any R-loop enrichment observed at KMT2A-fusion targets is therefore expected from their transcription level alone, and only enrichment *beyond* an expression-matched control is informative - a control the proposed experiment did not originally specify. Sanz LA et al. 2016
Connected to