Oncofusion

start / graph / mechanism-kmt2a-partner-loci-colocalise-in-single-nuclei

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q139-single-cell

KMT2A and several of its fusion partners occupy the same spatial cluster in individual nuclei before any break occurs — single-cell diploid Hi-C, RUNX1-dependent, and 3D architecture does NOT predict which fusion type a cell will make

also called contact first hypothesis at single-cell resolution · multipartite spatial clusters of KMT2A partners · RUNX1 transcription factories and the KMT2A breakpoint · complex rearrangement gene proximity network

Before a chromosome break ever happens, the KMT2A gene and several of the genes it later fuses to are already sitting next to each other in the same spot inside the nucleus of a normal cell — and reading one nucleus at a time shows that more than one partner can be next to KMT2A in the same cell at the same time. Knocking down RUNX1, the protein that gathers them there, pulls them apart. But the arrangement does not decide which partner a cell will eventually fuse to, or whether the resulting leukaemia will be lymphoid or myeloid.

The same thing, for a scientist

The **contact-first hypothesis** — that fusion partners colocalise before fusion — was tested at the level of individual nuclei by reanalysing published single-cell diploid Hi-C datasets. Four results, and the fourth is a negative that this graph should hold on to [E1]: 1. **Partner genes fused in leukaemia show smaller spatial distances** than pairs fused in solid tumours or control pairs. 2. **Multiple partners tend to colocalise with KMT2A in the same cell**, i.e. the cluster is multipartite rather than pairwise — an observation that bulk Hi-C cannot make, because a population-average contact map is identical whether a locus contacts partner A in half the cells and partner B in the other half, or contacts both in every cell. 3. Complex-rearrangement-related KMT2A fusion genes (CRGs) form a **genome-wide mutual proximity network**, are co-expressed and co-localised, are **enriched in RUNX1 targets**, and **RUNX1 knockdown significantly reduces contacts among them**; they sit in active transcriptional regions and loop anchors and carry high levels of **TOP2-mediated DNA breakage**, which **transcription inhibition reduces**. 4. **'3D genome architecture has little association with lineage decision of KMT2A fusion types in leukemia.'** Point 4 is the load-bearing negative. This graph asks repeatedly why one progenitor makes a lymphoid t(4;11) and another makes a myeloid t(9;11); **spatial proximity is not the answer**, and a published study looked and said so. Points 1–3 give a physical mechanism for *which genes are available to break together* — a RUNX1-organised transcription factory where TOP2 activity concentrates — and leave *what the resulting fusion does to a cell* entirely to the fusion protein and the cell state it lands in. **This is the subnuclear-localization axis at single-cell resolution, and the corpus does not cite it.**

Go deeper
What single-cell resolution buys here, stated as precisely as I can

A bulk Hi-C contact map is a population average. If KMT2A contacts AFF1 in 30% of nuclei and MLLT3 in another 30%, the average map is indistinguishable from one in which KMT2A contacts both in 30% of nuclei and neither in 70%. Those are different physical models with different implications for how a complex, three-way rearrangement arises, and only a per-nucleus contact map separates them.

The finding is the second one: multiple partners tend to colocalise with KMT2A in the same cell 1. That is the model in which complex rearrangements — which are a real and clinically annotated minority of KMT2A rearrangements — have a physical explanation, because the two illegitimate ligations needed for a three-way event are both available at the same place and time.

The honest limitation: single-cell Hi-C is extremely sparse. A negative per-nucleus contact is usually a sampling failure rather than a distance measurement, so these analyses work on aggregated distance distributions and the per-cell statement is a statistical claim about the population of nuclei, not a call on any individual nucleus.

RUNX1, and why the mechanism is transcription rather than proximity per se

The chain in the paper runs: CRGs are enriched in RUNX1 targets → they are co-expressed and co-localised → RUNX1 knockdown reduces contacts among them → they sit in active transcriptional regions and loop anchors → they carry high TOP2-mediated breakage → transcription inhibition reduces that breakage 1.

So the proposal is not 'genes that are close break together' but 'transcription gathers them and transcription breaks them', with TOP2 the enzyme and RUNX1 the organiser. That is the same causal shape this corpus already holds for the KMT2A breakpoint cluster region — transcription-coupled TOP2 activity — and it is consistent with the etoposide/topoisomerase-inhibitor aetiology of therapy-related KMT2A leukaemia.

RUNX1 is the interesting new name. This corpus has gene-runx1 and links it to fusion biology downstream; here it is a candidate organiser of the break-permissive compartment, upstream of the fusion existing at all. Whether that role is specific to RUNX1 or is a generic property of any strong lineage transcription factor in haematopoietic cells is not tested by this paper.

The negative, and the corpus entries it bears on

'3D genome architecture has little association with lineage decision of KMT2A fusion types in leukemia' 1.

This corpus contains hyp-kmt2a-aff1-shared-loop-anchor-hub, exp-hic-kmt2a-aff1-proximity-fetal-liver, and a set of CTCF/loop-anchor entries from the Q105/Q111 beats. None of them cites this paper, and it is the closest published test of the proximity-explains-outcome idea. A beat proposing a Hi-C experiment in fetal liver should read it first: the proximity part has been measured at single-nucleus resolution and the lineage part came back negative.

The negative is a little association, not a demonstrated zero, and the analysis is of published scHi-C from cell types that are not necessarily the cell of origin — that is the correct way to discount it, not by ignoring it. But a hypothesis whose central prediction has been looked for and not found in the best available instrument should carry that fact in its falsifier field.

Where this sits relative to the breakpoint literature this graph already holds
claimthis corpus's existing positionwhat scHi-C adds
the KMT2A BCR is transcription- and TOP2-coupledheld, with etoposide aetiologyCRGs genome-wide show the same coupling, and transcription inhibition reduces their breakage
partners are spatially close before the breakheld as the contact-first hypothesis, from bulk/FISHtested per nucleus; leukaemia partners closer than solid-tumour and control pairs
the cell of origin decides lymphoid vs myeloidheldsupported by elimination — 3D architecture does not decide it
complex rearrangements are a distinct classheld descriptivelygiven a physical model: a multipartite mutual-proximity network, RUNX1-dependent
Facts and figures
mechanism classspatial genome organisation / break formation
systemreanalysis of published single-cell diploid Hi-C datasets, plus RUNX1 knockdown and transcription inhibition in the authors' own system
organiserRUNX1
enzymeTOP2
shown for fusionKMT2A fusions as a class, including complex-rearrangement-related fusion genes; the lineage-decision negative is stated for 'KMT2A fusion types' generally and is not resolved to KMT2A::AFF1 versus KMT2A::MLLT3 individually
negative result3D genome architecture has little association with lineage decision of KMT2A fusion types in leukaemia
localization axissubnuclear — transcription factory / multipartite spatial cluster
source pmid35945623
cited in this corpus0
assays
[
  "single-cell diploid Hi-C (reanalysis)",
  "RUNX1 knockdown",
  "transcription inhibition",
  "TOP2-mediated DNA breakage mapping"
]
How this could bite us back
  • If transcription at a locus is what makes it break-prone, then any agent that increases transcription at KMT2A or its partners — including differentiation-inducing agents, which is what menin inhibitors are — could in principle increase the rate of new rearrangements in surviving normal progenitors. This is speculative and untested, and it is the kind of long-latency risk a response-rate endpoint cannot see.
  • A RUNX1-targeting strategy aimed at dissolving the break-permissive compartment would be a prophylactic, not a treatment, and RUNX1 is essential for definitive haematopoiesis. The therapeutic index for that idea is almost certainly unworkable and it should not be pursued on the strength of this mechanism.
Still open
  1. Which cell types are in the reanalysed scHi-C datasets? Everything about how much weight the lineage negative deserves turns on this one methods detail, and it is not in the abstract.
  2. Is KMT2A's multipartite cluster present in a FETAL haematopoietic progenitor? That is the cell the infant t(4;11) forms in, and no scHi-C dataset of it is known to this beat.
  3. Is RUNX1 the organiser, or is any strongly expressed lineage transcription factor sufficient? A GATA- or EBF1-centred control would separate 'RUNX1 specifically' from 'active transcription generally'.
  4. Does the proximity network predict which complex rearrangements are observed clinically? The KMT2A recombinome enumerates the partners that actually occur at what frequency; joining that frequency table to the measured proximity network is a desk analysis and would be the sharpest test of the contact-first hypothesis available.
  5. TOP2-mediated breakage at CRGs falls under transcription inhibition. Does it also fall under a menin inhibitor, which reduces transcription at the fusion's own targets? That would make the drug a break-prevention agent as well as a transcription agent, with implications for secondary rearrangement under therapy.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Reanalysis of published single-cell diploid Hi-C datasets found that partner genes fused in leukaemia show smaller spatial distances than solid-tumour or control gene pairs; that multiple partners colocalise with KMT2A in the same cell; that complex-rearrangement-related KMT2A fusion genes form a genome-wide mutual proximity network, are co-expressed, co-localised and enriched in RUNX1 targets, with RUNX1 knockdown significantly reducing contacts among them; that these genes sit in active transcriptional regions and loop anchors with high TOP2-mediated DNA breakage which transcription inhibition reduces; and that 3D genome architecture has little association with the lineage decision of KMT2A fusion types.

    PMID 35945623DOI
    What this evidence cannot show

    This is a REANALYSIS of published single-cell diploid Hi-C datasets, so the cell types are whatever those datasets contain and are not stated in the abstract; there is no guarantee any of them is a haematopoietic progenitor, let alone the fetal cell of origin of infant KMT2A::AFF1 leukaemia. Single-cell Hi-C is very sparse, so every distance claim is a statistical statement over aggregated nuclei rather than a per-nucleus measurement. The RUNX1 knockdown and transcription-inhibition experiments are in the paper's own system, not in the scHi-C data. Confidence is 'moderate' rather than 'strong' for the cell-type reason above, and the lineage negative inherits the same limitation: an association absent in the wrong cell type is weak evidence of absence in the right one. Read at abstract level; the figures were not opened.

    Evidence that argues against this (1)
    • The lineage negative cuts against this corpus's own shared-loop-anchor hypothesis and against the proposed fetal-liver Hi-C experiment, and should be read as a warning to those rather than as a settled result. Conversely, a defender of the proximity model can point out that the datasets reanalysed almost certainly do not include the fetal haematopoietic progenitor in which the t(4;11) actually forms, so the negative may simply be measuring the wrong nucleus — which is the strongest available rebuttal and is also the reason the fetal-liver experiment remains worth doing. Oncofusion Knowledge Graph 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    This paper is cited nowhere in this corpus: a structural scan of all 2,603 node files finds zero entries carrying PMID 35945623, despite the corpus holding a shared-loop-anchor hypothesis and a proposed fetal-liver Hi-C proximity experiment on the same question.

    What this evidence cannot show

    A scan for the PMID as a `pmid` field value or as a prose 'PMID 35945623', plus a literal grep for the DOI 10.1186/s13059-022-02740-9. Both return zero once this beat's own folder (data/graph/nodes/q139-single-cell/) is EXCLUDED - the unexcluded DOI grep returns 1, and that 1 is this entry itself. Recording that because a checker that counts the thing it just wrote is the commonest way a zero becomes a one in this project. This does not prove no entry discusses the finding without citing the paper.

    Evidence that argues against this (1)
    • Absence of a citation is not evidence a beat did not consider and reject the paper. This corpus records refusals in prose and in table cells as often as in dedicated entries, and a gap list cannot see what was decided — so 'uncited' should be read as 'worth a look', not as 'overlooked'. Oncofusion Knowledge Graph coordinator 2026
Connected to