Oncofusion

start / graph / mechanism-3d-genome-proximity-partner-choice

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q1-fusion-genesis

Why KMT2A picks AFF1: spatial proximity, shared transcription hubs, and the unsettled proximity-versus-damage question

also called translocation partner choice · contact-first model · transcription factory model · MLL AF4 nuclear proximity

Chromosomes 11 and 4 are separate, but inside the nucleus the KMT2A gene and the AFF1 gene can end up in the same small neighbourhood because both are switched on and busy at the same time. Ends that are already next to each other are the ones that get glued together.

The same thing, for a scientist

Partner choice in translocation is strongly influenced by spatial proximity at the time of breakage (the contact-first model), but the extent to which proximity DOMINATES is disputed for recurrent, damage-driven translocations of the kind KMT2A-r represents. Direct evidence in the KMT2A system: 2-3% of KMT2A alleles undergoing transcription do so in close proximity to one of its recurrent translocation partner genes, AF9/MLLT3 or AF4/AFF1, consistent with sharing transcription factories. Genome-wide, recurrently translocating leukaemia genes are highly transcribed and enriched at chromatin loop anchors, where they accumulate TOP2-mediated DSBs at recurrent hotspots; fusion formation at these hotspots requires both TOP2 isoforms and transcriptional elongation. The general principle was established by combining Hi-C with HTGTS in mouse pro-B cells, where normalising DSB numbers by irradiation revealed translocation frequency tracking pre-existing spatial proximity. The honest caveat is that no study has measured KMT2A-AFF1 proximity in the human fetal liver progenitor that actually gives rise to infant KMT2A::AFF1 ALL; the 2-3% figure comes from transformed and immortalised human cell lines. Counterweight: measuring contact frequency and break frequency in the same system, Hakim et al. found that non-targeted rearrangements reflect nuclear organisation whereas break formation rate governs the location and frequency of recurrent translocations, including those driving B-cell malignancies - which places KMT2A-r on the break-frequency side of the partition.

Go deeper
The measured number: 2-3%1 deeper

Cowell et al. used FISH to ask how often a transcribing KMT2A allele sits close to AF9 or AF4, and found 2-3% of transcribing KMT2A alleles do, consistent with the two genes sharing transcription factories 1. Their proposed model is that TOP2B-introduced DSBs in two genes occupying a common transcription factory become stabilised by anti-TOP2 drugs, providing the opportunity for illegitimate end joining and translocation. 2-3% is small in absolute terms but enormous relative to a random-encounter expectation for two loci on different chromosomes, and it is the only direct measurement of the geometry for this fusion that I could find.

What a transcription factory is, and the caveat

The transcription-factory concept holds that multiple actively transcribed genes cluster at shared RNA-polymerase-rich foci. The concept has been contested, and a more modern framing is transcriptional condensates and loop-anchor clustering. Either way, the empirical observation, that two translocation partners are more often adjacent when both are transcribed, survives the change in vocabulary.

Loop anchors as the structural explanation1 deeper

Gothe et al. showed TOP2-mediated DSBs are enriched at chromatin loop anchors with high transcriptional activity, with recurrent DSBs at CTCF- and cohesin-bound loop bases and DSB frequency positively correlated with transcriptional output and directionality. Genes that recurrently translocate to drive leukaemias, KMT2A included, are highly transcribed and enriched at loop anchors, and accumulate DSBs at recurrent hotspots that give rise to chromosomal fusions dependent on both TOP2 isoforms and on transcription elongation 2. This converts the fuzzy factory idea into a concrete, mappable structural feature.

Why this is elegant and also dangerous for our hypothesis

The loop-anchor model says fragility and partner choice are both consequences of the same thing: being a highly transcribed gene at a loop base. KMT2A and AFF1 are both highly transcribed in haematopoietic progenitors and both are transcription-elongation genes. The uncomfortable implication is that the fusion pairs two genes that were already co-regulated and co-located, which means the resulting oncoprotein assembles machinery that was partly pre-assembled. That is consistent with, but does not require, the menin-interface model.

The general proof: Hi-C plus HTGTS1 deeper

In G1-arrested mouse pro-B cells, a high-resolution Hi-C map was combined with genome-wide translocation sequencing from target DSBs. RAG-cleaved antigen-receptor loci dominated as partners when present, reflecting both high break frequency and colocalisation. To isolate the geometry term, the authors normalised DSBs genome-wide with ionising radiation; under those conditions translocations were highly enriched in cis and within subchromosomal domains in direct relation to pre-existing spatial proximity 3. That is the cleanest available demonstration that proximity, not a special affinity between partner sequences, drives partner choice.

The result that cuts the other way

This is the strongest single challenge to the contact-first framing and the entry should not be read without it. Hakim et al. measured contact frequency and break frequency in the same mouse B-cell system and found the two terms apply to different classes of event: without targeted damage, translocation frequency tracks Hi-C proximity; with recurrent site-directed damage, it tracks the rate of break formation, and the authors extend that conclusion specifically to the translocations that drive B-cell malignancies [E3 counter, Hakim 2012].

KMT2A-r leukaemia sits in the second class, not the first. The KMT2A BCR is a documented cleavage hotspot; breaks there are not random. The Zhang 2012 design normalised break number genome-wide with irradiation precisely in order to isolate the geometry term - a legitimate and elegant move, but it removes the variable that Hakim found dominates recurrent oncogenic translocations. Read together, the honest statement is: proximity sets the background rate of partner choice; break frequency and subsequent selection decide which pairs become recurrent disease. That is weaker than 'proximity determines partner choice' and it is what the title of this entry should be understood to mean.

Where this is inferential for KMT2A::AFF1

Being explicit, because the PI asked for honesty here: (1) the 2-3% proximity figure is from human cell lines, not fetal progenitors 1; (2) the loop-anchor work used etoposide-treated transformed human lines and measured fusion frequency in culture 2; (3) the proximity-equals-translocation proof is mouse pro-B cells with irradiation doses far above physiological 3; (4) I found no Hi-C or Capture-Hi-C dataset of human fetal liver HSC/LMPP that has been interrogated for a KMT2A-AFF1 contact. The model is well-supported as a general principle and unverified for the specific fusion in the specific cell. That gap is the basis of hyp-kmt2a-aff1-shared-loop-anchor-hub and exp-hic-kmt2a-aff1-proximity-fetal-liver.

Facts and figures
mechanism class3D genome architecture / spatial proximity
modelcontact-first: pre-existing proximity of two transcribing loci influences partner choice once both are broken. Disputed for recurrent damage-driven translocations, where break frequency has been shown to dominate (Hakim 2012, PMID 22314321).
measured proximity kmt2a partners PCT3
proximity methodFISH in human cell lines (Cowell 2012)
evidence gapno Hi-C/Capture-Hi-C measurement of KMT2A-AFF1 contact in human fetal liver HSC/LMPP
disputeproximity vs break-frequency vs selection: unresolved for KMT2A-r. Zhang 2012 (contact-first) normalised break number away by irradiation; Hakim 2012 measured both terms and found break frequency dominant for recurrent malignancy-driving translocations.
structural correlates
[
  "CTCF/cohesin chromatin loop anchors",
  "high transcriptional output and directionality",
  "transcription elongation dependence"
]
How this could bite us back
  • This step is essentially undruggable. Altering 3D genome architecture means perturbing CTCF or cohesin, which is globally toxic and itself oncogenic.
  • If the fusion partners were already co-regulated and co-located, the fusion protein may inherit an existing cofactor network rather than building a new one, which would mean fewer fusion-specific surfaces than the menin-interface model hopes for.
  • If break frequency rather than proximity decides which KMT2A partner becomes recurrent (Hakim 2012), then the project's interest in fetal-liver Hi-C is partly misdirected: the informative fetal measurement would be a break map at the KMT2A and AFF1 loci, not a contact map, and break mapping in primary fetal progenitors is the harder experiment.
Still open
  1. Is the KMT2A-AFF1 contact frequency higher in human fetal liver HSC/LMPP than in adult bone marrow HSC or in mature lymphocytes? If yes, 3D architecture would explain the age distribution of this specific fusion, which nothing else currently does.
  2. Do the most common KMT2A partners (AFF1, MLLT1, MLLT3, MLLT10, ELL, AFDN) rank by contact frequency in the same order as by clinical frequency? That is a directly testable prediction and, as far as I can tell, has never been tested.
  3. Does cohesin or CTCF perturbation change the KMT2A breakpoint distribution within the BCR?
  4. Does the KMT2A partner distribution (7 partners plus PTD = >90% of 3401 recombinations, 63 partners seen once) rank-correlate with contact frequency in any haematopoietic cell type at all? Until somebody runs that correlation, 'proximity explains partner choice' for this fusion is an analogy from other loci, not a measurement.
Every source behind this entry (4)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    2-3% of KMT2A alleles undergoing transcription do so in close proximity to AF9/MLLT3 or AF4/AFF1, consistent with shared transcription factories, supporting a model in which TOP2B-induced DSBs in co-located transcribing genes are joined illegitimately.

    PMID 22615413DOI
    What this evidence cannot show

    FISH-based colocalisation in human cell lines with a resolution limit around 200-500 nm, which is larger than the distance needed for ligation. Colocalisation at this resolution is necessary but not sufficient evidence of ligatable proximity. Not measured in fetal haematopoietic cells.

    Evidence that argues against this (1)
    • Proximity has never been tested against the distribution it is supposed to explain. The KMT2A partner distribution, as catalogued in 3401 patients, is extraordinarily skewed and long-tailed: 107 in-frame fusion partners are known, the seven commonest partners plus PTD account for more than 90% of all recombinations, 37 partners are recurrent and 63 have been seen exactly once. No study has asked whether contact frequency in a relevant cell type rank-orders those partners; the only proximity measurement for this system is a single FISH experiment covering two partners in cell lines [E1]. A steep, long-tailed, referral-cohort-derived frequency distribution is equally consistent with selection for a transforming combination, and the entry's own open question concedes the ranking test has never been run. Meyer C et al. 2023
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    TOP2-mediated DSBs are enriched at highly transcribed CTCF/cohesin loop anchors; recurrently translocating leukaemia genes are enriched there and generate fusions at recurrent hotspots requiring both TOP2 isoforms and transcriptional elongation.

    PMID 31202576DOI
    What this evidence cannot show

    Transformed human lines with induced TOP2 poisoning; break maps (BLESS/END-seq-class methods) have their own biases toward accessible chromatin, which risks circularity when the conclusion is that breaks are in accessible chromatin.

    Evidence that argues against this (1)
    • The loop-anchor TOP2 account does not describe the KMT2A breakpoint it is invoked to explain. When etoposide-induced double-strand breaks were mapped directly across the MLL breakpoint cluster region, they colocalised with the RNA polymerase II binding site and the transcription initiation region of a gene-internal MLL promoter and NOT with a nearby topoisomerase II consensus sequence. A genome-wide enrichment of TOP2-mediated breaks at transcribed loop anchors can therefore be true as a population statistic while being the wrong causal account of the specific cut that makes this fusion, because transcription and TOP2 activity are collinear at exactly these positions and the one locus-level experiment that separates them came down on transcription. Scharf S et al. 2007
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    With DSB numbers normalised genome-wide by irradiation, translocation frequency in mouse pro-B cells is directly related to pre-existing Hi-C spatial proximity, in cis and within subchromosomal domains.

    PMID 22341456DOI
    What this evidence cannot show

    Mouse, G1-arrested, engineered DSB bait, supraphysiological irradiation. Mouse and human differ in translocation joining pathway, so mouse rules should not be assumed to transfer wholesale.

    Evidence that argues against this (1)
    • The companion study from the same year reached the opposite conclusion for exactly the class of translocation this entry cares about. Hakim et al. measured nuclear architecture and DNA-damage frequency simultaneously in cultured mouse B lymphocytes and separated the two terms: in the ABSENCE of recurrent DNA damage, translocations between Igh or Myc and all other genes were directly related to their contact frequency - the proximity result - but translocations associated with recurrent site-directed DNA damage were proportional to the RATE OF DNA BREAK FORMATION, measured as RPA accumulation at the site. Their conclusion is explicit: non-targeted rearrangements reflect nuclear organisation, whereas DNA break formation governs the location and frequency of recurrent translocations, INCLUDING THOSE DRIVING B-CELL MALIGNANCIES. KMT2A-r leukaemia is a recurrent, damage-driven translocation at a locus with a documented cleavage hotspot, so on Hakim's partition the applicable model is break frequency, not contact frequency. The Zhang 2012 irradiation design deliberately normalised away the term Hakim found to be dominant. Hakim O et al. 2012
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.review

    Non-random spatial genome organisation, DNA repair pathway choice, chromatin features and the motion of broken chromatin jointly determine translocation frequency and partner selection.

    The biogenesis of chromosome translocations · Roukos V, Misteli T · Nature Cell Biology 2014
    PMID 24691255DOI
    What this evidence cannot show

    Review; the relative contribution of each determinant is not quantified for any human translocation.

    Evidence that argues against this (1)
    • 'Jointly determine' obscures an ordering that has been measured. With Hi-C and genome-wide translocation sequencing in the same G1-arrested mouse pro-B cells, RAG-cleaved antigen-receptor loci dominated as translocation partners for target breaks IRRESPECTIVE of genomic position, driven by high-frequency breakage; only after normalising DSBs genome-wide with ionising radiation did proximity-dependent enrichment become visible, and it was strongest in cis and within subchromosomal domains. For a locus that breaks frequently, partner choice is set mostly by where else the breaks are, and 3D proximity is a second-order modifier — which is the opposite emphasis from the one a review sentence gives. Zhang Y et al. 2012
Connected to