Oncofusion

start / graph / genome-ctcf-cohesin-hoxa-tad-boundary

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q10-genome

CTCF/cohesin boundaries and the HOXA topological domain — the 3D scaffold the fusion program runs on

also called HOXA TAD · CTCF boundary HOXA cluster · cohesin loop anchors HOXA · TAD integrity in MLL-rearranged leukemia

DNA is folded into loops and neighbourhoods, and the HOXA genes that KMT2A fusions switch on sit inside one such neighbourhood held together by proteins called CTCF and cohesin. In this leukemia one experiment shows the neighbourhood matters - but when scientists destroy these neighbourhoods everywhere at once, most genes carry on as before, so the fence is not automatically the thing holding the disease up.

The same thing, for a scientist

The HOXA cluster is organised into a topologically associating domain (TAD) whose boundaries are CTCF/cohesin-anchored. In MLL-rearranged and NPM1-mutant AML the lncRNA HOTTIP is aberrantly activated and is required to form the HOXA-driven TAD and its aberrant transcription; HOTTIP loss attenuates leukemogenesis, and re-expressing HOTTIP restores leukemic TADs, transcription and leukemogenicity in CTCF-boundary-attenuated AML cells [E1]. Mechanistically HOTTIP directly binds a subset of CTCF-binding sites and recruits CTCF/cohesin plus R-loop regulators to form R-loops that reinforce the boundary; deleting the boundary CBS or targeting RNase H to dissolve the R-loop impaired boundary activity, reduced promoter/enhancer interactions, lowered target expression and mitigated leukemogenesis in xenografts [E2]. A 2025 study adds an RNA-modification layer: METTL3 is transcriptionally activated by MLL in MLLr AML and, with YTHDC1, m6A-marks architectural RNAs to sustain CTCF-dependent TAD activity [E3]. Cohesin itself is a haematopoietic tumour suppressor whose loss reprogrammes chromatin accessibility and myeloid differentiation [E4], so the architectural machinery is both a dependency and a lesion class. **Standing caution (added cycle 35):** the acute-degron literature has substantially revised how much TAD disruption changes transcription. Cohesin degradation eliminates all loop domains without widespread ectopic gene activation [E1 counter, Rao 2017], Nipbl deletion makes TADs vanish in mouse liver with no transcriptional change [E1 counter, Schwarzer 2017], and CTCF degradation abolishes insulation while leaving compartments intact [E2 counter, Nora 2017]. Boundary disruption is therefore not in general sufficient to explain a transcriptional programme, and the HOTTIP/CTCF results here are locus-specific claims that must be read against that baseline.

Go deeper
What is genuinely demonstrated2 deeper

Three things, in increasing order of how much they surprised us.

1. The HOXA TAD is a functional requirement, not a description. Luo and colleagues did the experiment that matters: they attenuated the CTCF boundary, which collapsed the leukemic TAD, and then showed that re-activating HOTTIP restored the leukemic TAD, its transcription and leukemogenesis in those boundary-attenuated cells 1. That is a rescue experiment on a 3D-genome feature, which is rare.

2. The boundary is held by an R-loop, and dissolving the R-loop is enough. In the follow-up, targeting RNase H to the boundary CTCF-binding site of the β-catenin TAD — not deleting any protein, just removing the RNA:DNA hybrid — impaired boundary activity, reduced promoter/enhancer interactions, lowered β-catenin target expression and mitigated leukemogenesis in xenograft models 2. So an architectural element can be dismantled by an RNA-level intervention.

3. An RNA modification sits upstream of the boundary. METTL3 is transcriptionally activated by MLL in MLLr AML, forms a complex with YTHDC1 and CTCF, and its depletion disrupts CTCF-binding sites and chromatin accessibility at MYB and RUNX1; YTHDC1 loss compromises CTCF-dependent 3D organisation by Hi-C 3. This is covered in depth in genome-m6a-mettl3-ythdc1-ctcf-tad-kmt2a-r.

The honest limit: almost all of this is AML, and most is HOTTIP-centric

The HOTTIP TAD work is done in AML driven by MLL rearrangements or NPM1 mutation, and the flagship experiments in the 2022 paper are at the β-catenin TAD, not at HOXA. The infant B-ALL case the Iqbal Lab cares about — KMT2A::AFF1 in a fetal-liver-derived pro-B progenitor — has, as far as we can find, no published Hi-C, Micro-C or CTCF-degron study of its own. Nobody has asked whether the HOXA TAD in t(4;11) infant ALL has the same boundary architecture as in MLL-r AML, and given that KMT2A::AFF1 leukemias split into HOXA-high and HOXA-low groups, the assumption that they do is unsafe.

The second limit is directionality. Everything above shows the fusion program needs the existing architecture. Almost nothing shows the fusion creates new loops. That distinction matters enormously for therapy: a program that uses pre-existing architecture is attackable by removing the architecture; a program that builds its own architecture might rebuild it.

The degron results this has to survive (added cycle 35)

Between 2017 and 2018 the 3D-genome field ran the experiment that this entry's logic depends on, and got an answer that is mostly negative.

  • Cohesin degron: removing RAD21 acutely eliminates all loop domains. Compartment domains and histone marks are untouched, and the authors report that this 'does not lead to widespread ectopic gene activation', affecting only a significant minority of active genes. Many loops re-form within an hour of restoring cohesin [E1 counter-evidence].
  • Nipbl deletion in mouse liver: TADs and Hi-C peaks vanish globally in the absence of transcriptional changes, while compartments are preserved and even sharpened [E1 counter-evidence].
  • CTCF degron: insulation and CTCF-site looping are lost dose-dependently, yet compartmentalisation is unaffected; CTCF's transcriptional role is enhancer blocking, and restoring CTCF reinstates the architecture [E2 counter-evidence].

What this does and does not do to the entry. It does not refute the HOTTIP results: those are locus-specific experiments with a rescue arm, and a locus can be boundary-dependent even when the genome on average is not. It does refute the inference that would be natural to draw from them - that boundary integrity is generally what holds a leukaemic transcriptional programme up. It also raises the bar on the HOTTIP work itself, because the field-wide baseline is that TAD loss usually does nothing to transcription, so a paper reporting that it does everything at one locus needs independent replication before it carries therapeutic weight - and, as this entry already notes, no independent group has replicated the HOTTIP-CTCF interaction.

And it changes the therapeutic reading. If loops re-form within an hour of restoring the extruder, an intervention that dismantles a boundary buys a transient effect unless the intervention is continuous. That is a different, worse pharmacological problem from the one this entry originally described.

Why this is an intervention point, and why it bites back

The attraction is that the boundary is a small, discrete, sequence-defined element: a CTCF-binding site plus an R-loop. Candidate modalities are antisense oligonucleotides against HOTTIP, RNase H1 recruitment to a specific CBS, CRISPR deletion of a single CBS, and small-molecule inhibition of the METTL3 writer that maintains the architectural RNA 35.

The bite-back is severe and specific. CTCF and cohesin are architectural machinery for the whole genome, and cohesin loss in mouse haematopoiesis does not kill cells — it gives progenitors extra replating capacity, shifts chromatin accessibility and myeloid maturation genes, and produces a myeloproliferative-neoplasm-like disease in aged animals 4. In other words, partially disabling the architecture is itself leukemogenic. Any therapeutic strategy aimed at loop anchors must be locus-selective, which is precisely what an ASO or an RNase H-targeting approach could be and what a global CTCF or cohesin inhibitor could never be.

Facts and figures
disease context of evidenceMLL-rearranged and NPM1-mutant AML; no Hi-C/Micro-C or CTCF-degron study of KMT2A::AFF1 infant B-ALL identified
toxicity flagcohesin partial loss is itself leukemogenic in mouse (E4); global CTCF/cohesin inhibition is not a viable strategy
architecture elements
[
  {
    "element": "HOXA TAD",
    "anchor_type": "CTCF/cohesin boundary",
    "requirement": "HOTTIP-dependent formation; boundary attenuation collapses leukemic TAD",
    "refs": [
      "E1"
    ]
  },
  {
    "element": "boundary CTCF-binding site with HOTTIP-dependent R-loop",
    "requirement": "CBS deletion or RNase H targeting impairs boundary and promoter/enhancer interactions",
    "demonstrated_locus": "beta-catenin TAD",
    "refs": [
      "E2"
    ]
  },
  {
    "element": "CTCF-binding sites at MYB and RUNX1",
    "requirement": "METTL3/YTHDC1-dependent m6A on architectural RNAs",
    "refs": [
      "E3"
    ]
  }
]
candidate modalities
[
  "HOTTIP antisense oligonucleotide",
  "locus-targeted RNase H1 recruitment",
  "single-CBS CRISPR deletion",
  "METTL3 catalytic inhibitor (STM2457)"
]
degron baseline
{
  "note": "Acute-degron results that any locus-specific boundary claim must be read against",
  "cohesin_degron": "all loop domains eliminated; no widespread ectopic gene activation; loops re-form in <1 h (PMID 28985562)",
  "nipbl_deletion_mouse_liver": "TADs and Hi-C peaks vanish globally in the absence of transcriptional changes (PMID 29094699)",
  "ctcf_degron": "TAD insulation lost dose-dependently; compartments unaffected; architecture reinstated on CTCF restoration (PMID 28525758)"
}
How this could bite us back
  • Cohesin partial loss is leukemogenic in its own right in mouse haematopoiesis, so any systemic architectural inhibitor risks creating a second myeloid disease rather than curing the first.
  • If the fusion uses pre-existing architecture rather than building it, loops may survive successful menin inhibition and allow rapid transcriptional restart — an architectural resistance mechanism that no current assay in the menin-inhibitor trials would detect.
  • R-loops are genome-wide and protective as well as pathogenic; dissolving them globally causes replication stress and DNA damage, which in a fetal progenitor is exactly the state that generates translocations in the first place.
  • CTCF-boundary editing in a patient would be a permanent genome change at a locus whose normal developmental function (HOXA collinearity) matters in tissues other than blood.
  • Loops re-form within an hour of restoring cohesin, so any architecture-directed therapy is a continuous-exposure strategy; a pulsed or intermittent schedule would let the leukaemic architecture rebuild between doses.
Still open
  1. Does the HOXA TAD in KMT2A::AFF1 infant B-ALL have the same CTCF boundary architecture as in MLL-r AML? A single Micro-C dataset on primary t(4;11) blasts versus fetal pro-B progenitors would answer it, and none exists.
  2. Do HOXA-low KMT2A::AFF1 cases lack the boundary/HOTTIP dependency altogether? If so, boundary-directed therapy would be a biomarker-selected strategy, not a pan-KMT2A-r one.
  3. Does menin inhibition change 3D architecture, or only occupancy and transcription? If loops persist after revumenib, the architecture is a reservoir the program can restart from, which is a resistance mechanism nobody is currently measuring.
  4. Has the HOTTIP-CTCF interaction been replicated outside the Huang/Xu/So group? We could not find independent confirmation, and the whole boundary-targeting logic rests on it.
  5. Given that genome-wide TAD loss usually does not change transcription, is the HOXA/HOTTIP boundary an exception, or is the reported transcriptional effect driven by HOTTIP's non-architectural functions? A CTCF or cohesin degron in a KMT2A-r line, scoring HOXA output directly, separates the two and has never been run in this disease.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    HOTTIP is aberrantly activated in AML driven by MLL rearrangements or NPM1 mutation, alters the HOXA-driven TAD and its transcription, and re-activating HOTTIP restores leukemic TADs, transcription and leukemogenesis in CTCF-boundary-attenuated AML cells.

    HOTTIP lncRNA Promotes Hematopoietic Stem Cell Self-Renewal Leading to AML-like Disease in Mice · Luo H, Zhu G, Xu J, Lai Q, Yan B, Guo Y, Fung TK, Zeisig BB, Cui Y, Zha J, Cogle C, Wang F, Xu B, Yang FC, Li W, So CWE, Qiu Y, Xu M, Huang S · Cancer Cell 2019
    PMID 31786140DOI
    What this evidence cannot show

    AML models (human AML lines and Hottip transgenic mice), not KMT2A::AFF1 infant B-ALL. Hottip aberration alone produces an AML-like disease in mice, so HOTTIP is not merely downstream of the fusion; the causal ordering in a human t(4;11) blast is untested. Field baseline added cycle 35: acute cohesin or CTCF degradation abolishes TADs with little transcriptional consequence genome-wide, so a locus-specific boundary-dependence claim needs an expression-matched control locus and independent replication, neither of which this paper provides.

    Evidence that argues against this (2)
    • Acute degradation of cohesin (RAD21) eliminates ALL loop domains genome-wide, yet the authors state plainly that loss of loop domains 'does not lead to widespread ectopic gene activation' and affects only a significant minority of active genes; compartment domains and histone marks are unaffected. This sets the baseline any locus-specific boundary claim must beat: dissolving a TAD is, in general, NOT sufficient to collapse a transcriptional programme. The HOTTIP rescue is therefore evidence about one locus in one model, not a general licence to treat boundary integrity as the cause of the leukaemic programme. The same paper also shows many megabase loops re-form within an hour of cohesin restoration, so architecture destroyed pharmacologically is expected to rebuild. Rao SSP et al. 2017
    • Deleting the cohesin loader Nipbl in mouse liver makes TADs and Hi-C peaks 'vanish globally, even in the absence of transcriptional changes', while compartmental segregation is preserved and even reinforced. An in-vivo mammalian tissue can therefore lose its entire TAD layer with no transcriptional consequence, which directly contradicts the intuition in this entry's lay summary that breaking the fence makes the gene programme collapse. Schwarzer W et al. 2017
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    HOTTIP recruits CTCF/cohesin and R-loop regulators to form R-loops at a subset of CTCF-binding sites; deleting the boundary CBS or targeting RNase H to eliminate the R-loop impaired CTCF boundary activity, inhibited promoter/enhancer interactions, reduced target gene expression 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

    The decisive boundary experiments are at the beta-catenin TAD in AML cells, not at the HOXA cluster and not in KMT2A::AFF1 B-ALL. RNase H targeting is a tool, not a therapeutic modality. Single laboratory; the HOTTIP-CTCF interaction has not been independently replicated by another group as far as we can find. Added cycle 35: boundary activity and transcription are separable read-outs in degron experiments, so 'impaired boundary activity' does not by itself establish that the transcriptional change is caused by the architectural change.

    Evidence that argues against this (1)
    • Auxin-degron depletion of CTCF abolishes CTCF-site looping and TAD insulation dose-dependently, but active and inactive compartments stay properly segregated and CTCF's transcriptional role is specifically enhancer blocking, not a general barrier. Two consequences for this entry: (i) 'boundary activity' and 'transcriptional output' are separable read-outs and an experiment that moves one is not evidence about the other; (ii) restoring CTCF reinstates proper architecture, i.e. the boundary is instructively rebuilt, which undercuts the idea that a single CBS or R-loop is a durable therapeutic target. Nora EP et al. 2017
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In MLL-rearranged AML, METTL3 is transcriptionally activated by MLL and forms a complex with YTHDC1 and CTCF; METTL3 depletion disrupts CTCF-binding sites and chromatin accessibility at MYB and RUNX1, and YTHDC1 loss compromises CTCF-dependent 3D genome organisation by Hi-C.

    The METTL3-YTHDC1 axis mediates architectural RNA m6A modification to modulate the integrity of chromatin TADs in MLLr+ AML genome · Fu R, Yu W, Zhao R, Yang H, Cao B, Dai C, Qianjiang H, Xia Y, Wang L, Lu J, Zhou R, Shi A, Lou H, Zhang X, Luo H · Molecular Cancer 2025
    PMID 41430607DOI
    What this evidence cannot show

    Very recent, single group (overlapping senior authorship with the HOTTIP work, so not independent of E1/E2), MLLr AML cell lines rather than primary infant B-ALL, and the arcRNA claims rest on MALAT1 as the exemplar. Not yet independently replicated. Added cycle 35: the DRIP-seq/R-loop arm uses S9.6, whose specificity and cross-method concordance are actively disputed; no RNase H1/RNase T1 control is stated in the abstract.

    Evidence that argues against this (2)
    • The R-loop arm of this paper rests on DRIP-seq with the S9.6 antibody, which has substantial affinity for double-stranded and ribosomal RNA; Smolka et al. show S9.6 signal in fixed human cells is predominantly ribosomal RNA and is RNase H1-insensitive. They do find that S9.6 DRIP *sequencing* maps remain RNase H1-sensitive, so DRIP-seq is not invalidated - but an arcRNA/R-loop claim built on S9.6 needs RNase H1 and RNase T1 controls reported, and this paper's abstract does not state that they were. Smolka JA et al. 2021
    • Reprocessing 810 R-loop mapping datasets, Miller et al. report 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 - and show that a substantial fraction of published R-loop data is of poor quality. A single-group, single-modality DRIP-seq result linking m6A-arcRNAs to R-loops at CTCF boundaries is therefore not yet a settled observation. Miller HE et al. 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Inducible silencing of cohesin subunits in mice does not impair cell viability but increases haematopoietic progenitor replating capacity, alters stem-cell homeostasis and myelopoiesis with widespread changes in chromatin accessibility, and produces a myeloproliferative-neoplasm-like disease in aged animals.

    Cohesin loss alters adult hematopoietic stem cell homeostasis, leading to myeloproliferative neoplasms · Mullenders J, Aranda-Orgilles B, Lhoumaud P, Keller M, Pae J, Wang K, Kayembe C, Rocha PP, Raviram R, Gong Y, Premsrirut PK, Tsirigos A, Bonneau R, Skok JA, Cimmino L, Hoehn D, Aifantis I · Journal of Experimental Medicine 2015
    PMID 26438359DOI
    What this evidence cannot show

    shRNA knockdown in mouse, a partial-loss model chosen deliberately; the phenotype is myeloproliferative rather than acute leukemia, and no KMT2A fusion is present. It bears on the toxicity of architectural targeting rather than on KMT2A-r biology directly.

    Evidence that argues against this (1)
    • This entry uses E4 to argue that architectural machinery cannot be targeted systemically. Rao et al. sharpen and partly redirect that: complete, acute cohesin degradation removes every loop domain without widespread ectopic gene activation, so the leukaemogenic phenotype in Mullenders et al. is a chronic, partial-loss, haematopoiesis-specific effect rather than a direct consequence of losing loops. The toxicity argument survives, but the mechanism attributed to it in this entry (disabling the architecture) is not what the acute experiments show. Rao SSP et al. 2017
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A selective catalytic METTL3 inhibitor (STM2457) reduces AML growth, increases differentiation and apoptosis, and prolongs survival in mouse AML models, establishing that the writer maintaining architectural-RNA m6A is pharmacologically tractable.

    Small-molecule inhibition of METTL3 as a strategy against myeloid leukaemia · Yankova E, Blackaby W, Albertella M, Rak J, De Braekeleer E, Tsagkogeorga G, Pilka ES, Aspris D, Leggate D, Hendrick AG, Webster NA, Andrews B, Fosbeary R, Guest P, Irigoyen N, Eleftheriou M, Gozdecka M, Dias JML, Bannister AJ, Vick B, Jeremias I, Vassiliou GS, Rausch O, Tzelepis K, Kouzarides T · Nature 2021
    PMID 33902106DOI
    What this evidence cannot show

    The reported mechanism is translational (m6A on leukaemogenic mRNAs), not architectural; using this paper to support architecture-directed therapy is an inference across two literatures. Efficacy was shown in AML models, not KMT2A::AFF1 infant ALL.

    Evidence that argues against this (1)
    • Tractability of the writer is not specificity of the target. METTL3 controls myeloid differentiation of NORMAL haematopoietic cells as well as leukaemia cells, so inhibiting it acts on the compartment the patient needs, and the mouse efficacy quoted here was obtained in AML models generally rather than in KMT2A-rearranged disease specifically. Nature Medicine 2017
Connected to