Oncofusion

start / graph / localization-nuclear-speckles-kmt2a-r

Mechanismweak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.Q4b-localization

Nuclear speckles — the compartment everyone assumes KMT2A::AFF1 uses, on evidence that skips AFF1 itself

also called SC35 speckles · SON SRRM2 · interchromatin granule clusters · speckle-associated domains · SPAD · AFF family speckle localisation

Nuclear speckles are droplet-like hubs of RNA-splicing machinery, and genes that sit close to them get expressed more strongly. Three of the four AFF-family proteins have been shown to live in speckles — but not AFF1, the one that is actually fused to KMT2A in infant leukemia, and no one has ever co-stained a real KMT2A fusion with a speckle marker.

The same thing, for a scientist

The speckle hypothesis for KMT2A::AFF1 rests on a chain with a missing link. Melko et al. showed AFF2/FMR2, AFF3/LAF4 and AFF4/AF5q31 localise to nuclear speckles, bind RNA with high apparent affinity for G-quadruplexes, modulate splicing of a G4-containing minigene, and on overexpression interfere with speckle organisation or biogenesis; enlarged speckles are present in FRAXE fibroblasts [E1]. AFF1 itself was not among the proteins shown to localise there. Independently, speckle proximity is now known to be a quantitative determinant of output: TSA-seq maps a nuclear lamina-to-speckle axis correlated with a gradient of genome activity, with the most highly expressed genes, housekeeping genes and super-enhancers positioned deterministically closest to speckles [E2], SPIN formalises this into genome-wide compartmentalisation states [E3], and a transcription factor can actively drag its targets to speckles to boost nascent RNA output [E4]. Nothing in this chain has been measured in a KMT2A-rearranged cell.

Go deeper
What the AFF-family speckle evidence actually says

Melko et al. tested AFF2/FMR2, AFF3/LAF4 and AFF4/AF5q31. All three localise to nuclear speckles, bind RNA with high apparent affinity for G-quadruplex structures, and modulate splicing efficiency of a minigene carrying a G4 in an alternatively spliced exon. Overexpression of AFF2/3/4 interferes with speckle organisation or biogenesis, and FRAXE patient fibroblasts have enlarged speckles 1.

AFF1/AF4 is named in the paper as a family member but is not among the proteins shown to localise to speckles. That is the whole gap. Reading this literature as the AFF family is a speckle family, therefore KMT2A::AFF1 is in speckles skips the only member that matters for t(4;11). AFF1 and AFF4 differ markedly outside the conserved ALF/CHD/pSer architecture, and AFF4 is the SEC scaffold while AFF1 is the leukemia partner — the graph elsewhere records statistics being mis-attributed between these two paralogues, so the distinction is not pedantry.

The speckle-disruption result also comes from overexpression, which is the standard way to perturb a condensate non-specifically. Enlarged speckles in FRAXE fibroblasts is the stronger observation because it is patient material at endogenous expression, but it concerns AFF2 loss, not AFF1 gain.

Why speckle proximity would matter if it were true1 deeper

Speckles are not inert. TSA-seq, which uses tyramide signal amplification as a cytological ruler to estimate mean chromosomal distance from a nuclear structure genome-wide, revealed in K562 a clear nuclear lamina-to-speckle axis correlated with a striking spatial gradient in genome activity. Transcription hot zones that protrude furthest into the nuclear interior and position deterministically very close to speckles contain more total genes, the most highly expressed genes, housekeeping genes, genes with low transcriptional pausing, and super-enhancers 2. SPIN integrates TSA-seq, DamID and Hi-C into ten spatial compartmentalisation states relative to speckles, lamina and nucleoli, and relates them to transcription and replication timing 3.

Crucially, speckle association is regulated, not just correlated: Alexander et al. showed p53 drives its target genes into speckle association, that speckle association boosts p21 expression by raising nascent RNA amounts, and that this required an intact proline-rich domain and direct DNA binding but not p53's transactivation function — and that speckle-associating p53 targets are more robustly activated than non-speckle-associating ones 4.

So the mechanism exists in principle: a DNA-binding factor can relocate its targets to a speckle and get more RNA per locus. If KMT2A::AFF1 did that at HOXA9, MEIS1, PROM1 and the rest, it would be a genuinely new layer of the fusion's mechanism and one that is invisible to every ChIP-based assay the field currently runs.

Why this is testable today with no new imaging

K562 already has public SON TSA-seq. What does not exist is SON TSA-seq in a t(4;11) line, and that is the missing dataset. But even before generating it, one can ask the cheap version of the question computationally: do published KMT2A::AFF1 binding sites and up-regulated genes fall preferentially into speckle-proximal SPIN states in a comparable haematopoietic reference, relative to matched control gene sets? A positive result would justify the imaging; a negative would deflate the hypothesis for the price of a week of analysis. That analysis is written up separately in this beat.

What 'speckle' means in the older KMT2A literature

The 1997 SNL paper describes ALL-1 as having a nuclear punctate distribution and later literature has drifted into calling these 'speckles'. That is a vocabulary accident. A punctate immunofluorescence pattern is not a nuclear speckle in the SC35/SON/SRRM2 sense; almost every chromatin protein gives puncta. No published image co-stains any KMT2A protein or fusion with SON or SRRM2. Until that co-stain exists, the correct statement is that KMT2A fusions form nuclear puncta of unknown identity.

Facts and figures
compartmentnuclear speckle (interchromatin granule cluster)
evidence class for kmt2anone direct; indirect via AFF-family paralogues that exclude AFF1
endogenous tag datafalse
patient material datafalse
marker proteins
[
  "SON",
  "SRRM2",
  "SC35/SRSF2 epitope",
  "MALAT1 (RNA)"
]
aff members shown in speckles
[
  "AFF2/FMR2",
  "AFF3/LAF4",
  "AFF4/AF5q31"
]
aff members not shown
[
  "AFF1/AF4"
]
methods used
[
  "transfected-construct immunofluorescence",
  "TSA-seq",
  "DamID",
  "Hi-C",
  "DNA FISH with speckle immunofluorescence"
]
public datasets usable
[
  "K562 SON TSA-seq (4DN)",
  "K562 SPIN states"
]
How this could bite us back
  • If speckle proximity amplifies fusion target output, a drug that evicts the fusion from chromatin but leaves the locus speckle-proximal may allow rapid re-activation on drug withdrawal — a structural basis for the hysteresis the graph already suspects.
  • Speckle components (SON, SRRM2, SRSF factors) are broadly essential, so any intervention aimed at the compartment rather than the fusion has no therapeutic window.
Still open
  1. Does AFF1 itself localise to nuclear speckles? The paper everyone cites for the AFF family tested AFF2, AFF3 and AFF4 and not AFF1. A single co-stain of endogenous AFF1 with SON would close a 15-year-old gap.
  2. Do KMT2A::AFF1 target genes sit in speckle-proximal SPIN states more often than matched controls? Answerable today from public data with no experiment.
  3. Does the fusion move its targets toward speckles, the way p53 moves p21? DNA FISH for HOXA9 plus SON immunofluorescence, with and without fusion degradation, is a one-month experiment.
  4. Does the AFF1 G-quadruplex RNA-binding activity reported for its paralogues exist, and is it what tethers the fusion to a speckle rather than protein-protein interaction?
Every source behind this entry (4)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    AFF2/FMR2, AFF3/LAF4 and AFF4/AF5q31 localise to nuclear speckles and bind RNA with high apparent affinity for G-quadruplex structures, modulate splicing of a G4-containing minigene, and their overexpression interferes with nuclear speckle organisation or biogenesis; enlarged nuclear speckles are present in FRAXE patient fibroblasts. AFF1/AF4 is not among the family members shown to localise to speckles.

    Functional characterization of the AFF (AF4/FMR2) family of RNA-binding proteins: insights into the molecular pathology of FRAXE intellectual disability. · Melko M, Douguet D, Bensaid M, Zongaro S, Verheggen C, Gecz J, Bardoni B · Human Molecular Genetics 2011
    PMID 21330300DOI
    What this evidence cannot show

    Localisation was assessed in HeLa and other cell lines with transfected constructs; the speckle-disruption result is explicitly an overexpression phenotype, the standard non-specific way to perturb a nuclear body. AFF1 — the AFF member fused to KMT2A in t(4;11) — was not shown to localise to speckles in this work, so extrapolating to KMT2A::AFF1 is an inference across paralogues, not a finding. The FRAXE fibroblast observation concerns loss of AFF2, not gain of an AFF1 fusion.

    Evidence that argues against this (1)
    • Two problems with this evidence item that are visible in the cited paper itself, and neither is recorded above. (1) AFF1/AF4 - the actual KMT2A::AFF1 partner - was NOT among the proteins shown to localise to speckles. Melko et al. name all four AFF genes but test AFF2/FMR2, AFF3/LAF4 and AFF4/AF5q31; AFF1 appears only in the family list. The entry's inference therefore runs from three paralogues to the one that matters, and the graph's own AFF-paralogue entries elsewhere note that AFF1 and AFF4 have non-redundant complexes. (2) The same paper reports that OVEREXPRESSION of AFF2/3/4 interferes with the organisation or biogenesis of nuclear speckles, and that FRAXE patient fibroblasts have enlarged speckles. So the localisation was scored in a system where the construct being scored perturbs the compartment it is being scored against - the artefact class that `localization-condensate-evidence-standards` flags for overexpressed constructs. Read: abstract only (PubMed). The citation attached to this note is the parent paper itself: both problems are visible in Melko et al. 2011 and the note is a reading of that paper, not an import from another one. Melko M et al. 2011
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    TSA-seq acts as a cytological ruler for mean chromosomal distance from nuclear speckles genome-wide and reveals in K562 a nuclear lamina-to-speckle axis correlated with a striking spatial gradient in genome activity, with transcription hot zones positioned deterministically closest to speckles containing the most highly expressed genes, housekeeping genes, genes with low transcriptional pausing and super-enhancers.

    Mapping 3D genome organization relative to nuclear compartments using TSA-Seq as a cytological ruler · Chen Y, Zhang Y, Wang Y, Zhang L, Brinkman EK, Adam SA, Goldman R, van Steensel B, Ma J, Belmont AS · Journal of Cell Biology 2018
    PMID 30154186DOI
    What this evidence cannot show

    Ensemble-averaged population measurement in a single cell line (K562, a CML-derived erythroleukemia line); mean distances conceal cell-to-cell heterogeneity, and correlation between speckle proximity and expression does not establish direction of causality. No haematopoietic-progenitor or KMT2A-rearranged dataset.

    Evidence that argues against this (1)
    • Counter to treating speckle proximity as a load-bearing functional variable. Nakagawa et al. deleted Malat1, the abundant lncRNA that is the canonical speckle-resident architectural transcript, in mice. The knockouts were viable and fertile with no apparent phenotype, and nuclear speckle markers still localised correctly in cells lacking Malat1. A compartment whose most abundant dedicated RNA can be removed with no organismal consequence and no loss of marker organisation is a poor candidate for a rate-limiting step in leukemogenesis, and TSA-seq distance-to-speckle is best read as a correlate of expression rather than as a mechanism. Read: abstract only. Nakagawa S et al. 2012
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    SPIN integrates TSA-seq and DamID with Hi-C to assign ten genome-wide spatial compartmentalisation states relative to nuclear speckles, lamina and putative nucleolar associations in K562, and relates these states to transcription and replication timing.

    SPIN reveals genome-wide landscape of nuclear compartmentalization · Wang Y, Zhang Y, Zhang R, van Schaik T, Zhang L, Sasaki T, Peric-Hupkes D, Chen Y, Gilbert DM, van Steensel B, Belmont AS, Ma J · Genome Biology 2021
    PMID 33446254DOI
    What this evidence cannot show

    A computational integration method demonstrated as proof of principle in one cell line; states are model-inferred, the nucleolar associations are explicitly described as putative, and transferring K562 states to a B-lymphoblast context is an assumption that would need checking.

    Evidence that argues against this (1)
    • Speckle-referenced compartment states should not be read as functional requirements, because the cleanest genetic test of a speckle component came back negative. Nakagawa et al. deleted Malat1 - an abundant nuclear-speckle-localised lncRNA that interacts with SR splicing factors and had been implicated in cell migration, synapse formation, cell-cycle progression and serum response in culture - and found the knockout mice viable and fertile with no apparent phenotype, and, decisively for this item, nuclear speckle markers still CORRECTLY LOCALISED in cells lacking Malat1. The only change was reduced Neat1 in particular tissues. A ten-state computational assignment of the genome relative to speckles, lamina and putative nucleolar contacts in one cell line is therefore a description of spatial correlation; the compartment tolerates removal of a component that cell-culture work had made look essential. Nakagawa S et al. 2012
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    p53 actively drives association of its target genes with nuclear speckles, speckle association boosts p21 expression by elevating nascent RNA amounts, this requires an intact proline-rich domain and direct DNA binding but not p53 transactivation function, and speckle-associating p53 targets are more robustly activated than non-speckle-associating targets.

    p53 mediates target gene association with nuclear speckles for amplified RNA expression · Alexander KA, Coté A, Nguyen SC, et al. · Molecular Cell 2021
    PMID 33823140DOI
    What this evidence cannot show

    HEK293 and related systems with p53 activation; a demonstration that one transcription factor can do this, not that transcription factors generally do, and certainly not evidence about KMT2A fusions. DNA-FISH-plus-speckle-immunofluorescence measurements are population snapshots in fixed cells.

    Evidence that argues against this (1)
    • Searched and empty, recorded explicitly. No published study places any KMT2A fusion protein, or KMT2A-fusion target loci, at or near nuclear speckles by imaging, TSA-seq or SON CUT&RUN. PubMed searches on 2026-09-18 combining KMT2A/MLL/MLL-AF4 with nuclear speckle, SC35, SON, speckle-associated domain and TSA-seq returned nothing. The entry's speckle claim for KMT2A-r is therefore an extrapolation from paralogues plus a general genome-organisation literature, with no direct observation in the disease system; the `weak` grade it already carries is appropriate and should not be raised without such an observation. No citation is attached because there is nothing to cite: this note is a bounded absence recording searches that returned nothing, not a sourced fact. Uncited and unciteable look identical in this field, so it is said here explicitly (Q42 citation-integrity repair, 2026-09-20).
Connected to