Oncofusion

start / graph / absence-kdm2b-and-the-kmt2a-fusion-are-both-nucleolar-and-nobody-has-put-them-together

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q79-cxxc-readers

Three CXXC-pathway activities converge on rDNA - KDM2B represses Pol I, KMT2A and SETD1A bind the whole locus, MLL-AF4 is predominantly nucleolar - and no paper contains two of them

also called the nucleolar CXXC convergence · KDM2B rDNA versus the fusion

The nucleolus is the factory that builds ribosomes, and its DNA is full of unmethylated CG letters. Three separate groups have shown that KDM2B shuts that factory down, that KMT2A sits all over its DNA, and that the leukaemia fusion piles into the nucleolus and disrupts it. No single study has ever looked at more than one of those at a time.

The same thing, for a scientist

The rDNA locus is a CpG-rich, CXXC-readable address, and three independent lines place CXXC-pathway proteins on it. **(1)** JHDM1B/FBXL10 - KDM2B - is a nucleolar protein that preferentially binds the transcribed region of ribosomal DNA and represses rRNA transcription, JmjC-dependently, by demethylating H3K4me3 in the nucleolus [E1]. **(2)** KMT2A and KMT2F (SETD1A) **bind entire rDNA loci** on customised rDNA genome assemblies, mirrored by H3K4me2 and H3K4me3; KMT2F is the primary depositor of H3K4me3 on rDNA and promotes RNA Pol I pre-initiation-complex formation, and loss of H3K4me3 represses the locus [E2]. **(3)** The MLL amino terminus targets MLL, MLL-AF9 and MLL-AF4 to the nucleolus and nuclear matrix [E3], and human MLL-AF4 is **predominantly nucleolar**, interferes with Pol I transcription and ribosome biogenesis and switches on the nucleolar stress pathway [E4]. **Put together, that is a CXXC reader repressing Pol I and a CXXC-containing fusion occupying the same compartment and also interfering with Pol I - at a locus where the wild-type protein is known to bind and to deposit the activating mark. It is the most concrete shared address in the whole family, and no publication contains two of the three.** A bounded search confirms it: `(KDM2B OR FBXL10 OR JHDM1B OR NDY1)[tiab] AND (nucleolus OR nucleolar OR rDNA OR "ribosomal DNA")[tiab]` returns 7 records, none mentioning KMT2A; `(MLL OR KMT2A)[tiab] AND (rDNA OR "ribosomal DNA" OR nucleolar)[tiab]` returns 26, of which **10 are microbiology noise matching mL/L** and only 2 place KMT2A or a fusion in the nucleolus, neither naming any other CXXC protein [E5]. **This entry extends `localization-nucleolus-kmt2a-fusion`, which holds the fusion half of the story and does not mention KDM2B; it does not duplicate it.**

Go deeper
The three lines, and what each actually shows

KDM2B at rDNA. Frescas et al. identified human JHDM1B as a nucleolar protein, showed it preferentially binds the transcribed region of rDNA and represses rRNA transcription, and showed the repression is JmjC-dependent and works by demethylating H3K4me3 in the nucleolus, with a JmjC-dependent negative effect on cell size and proliferation 1. This graph's gene-kdm2b already carries it. What has never been asked is whether KDM2B's ZF-CxxC is what puts it there - the paper attributes the address to nucleolar localisation and the activity to JmjC, and rDNA is CpG-rich.

KMT2A and SETD1A at rDNA. Lone et al. used customised rDNA genome assemblies - a methodological point, because standard assemblies exclude the rDNA repeats and every earlier ChIP was therefore blind here - and found KMT2A and KMT2F/SETD1A binding entire rDNA loci, with KMT2F the primary H3K4me3 depositor, promoting Pol I pre-initiation-complex formation; losing H3K4me3 represses rDNA 2. This is the activating counterpart of the KDM2B result at the same locus, published nineteen years later, and the two are not cited together anywhere this beat could find.

The fusion in the nucleolus. Caslini et al. placed MLL, MLL-AF9 and MLL-AF4 in the nucleoplasm and at the nucleolar periphery and identified the MLL amino terminus - the part every fusion keeps, carrying the AT-hooks and the CXXC - as the common targeting element 3. Siemund et al. reported human MLL-AF4 as predominantly nucleolar, interfering with Pol I transcription and ribosome biogenesis and activating nucleolar stress 4. Both used exogenously expressed constructs, a limitation localization-nucleolus-kmt2a-fusion states carefully and this entry does not weaken.

Why the intersection is worth running

The three results make one falsifiable prediction that costs one experiment: if the fusion and KDM2B occupy overlapping rDNA territory, then the fusion's reported interference with Pol I transcription should be modified by KDM2B levels. Concretely - in a KMT2A-rearranged line, degrade KDM2B and measure 47S pre-rRNA, H3K4me3 at rDNA, and fusion occupancy on an rDNA-containing assembly. Three outcomes and all are informative:

  • Fusion occupancy at rDNA rises on KDM2B loss - the address is contested, at a named locus, and this is the first such demonstration in the family.
  • Nothing moves - the two proteins share a compartment and not a site, which is the null and is worth writing down.
  • rRNA output moves without fusion occupancy moving - the interaction is through the mark, not through the DNA, which is a different and more tractable mechanism.

The methodological prerequisite is non-trivial and is the likely reason nobody has done it. rDNA repeats are absent from the standard human assembly, so every ChIP-seq in the KMT2A field has discarded these reads. Lone et al. had to build custom assemblies to see the signal at all 2. A re-analysis of existing KMT2A-fusion ChIP-seq against an rDNA-containing assembly is a computational experiment with no new wet work, and it could be done this cycle.

Bounding, and the query trap found while doing it

(MLL OR KMT2A)[tiab] AND (rDNA OR "ribosomal DNA" OR nucleolar)[tiab] returns 26 records. Ten of the twenty-six are microbiology, plant pathology and environmental engineering - biohydrogen production from Clostridium, melanin from a halotolerant black yeast, lipase from arctic soil - because MLL matches as a free-text token in those fields. Every title was read and the contaminated ones excluded 5. This is the alias-precision failure this project has logged for RD and U19, appearing for the project's own central symbol: MLL[tiab] is not a clean token outside a leukaemia-scoped query, and it is proposed as a registry row in the memo.

Of the sixteen on-topic records, exactly two place KMT2A or a fusion at the nucleolus 34 and one places KMT2A and SETD1A on rDNA 2. None names a second CXXC protein. On the KDM2B side, seven records connect KDM2B to the nucleolus or rDNA and none mentions KMT2A 5. A weaker fourth line exists - combined KDM2A and KDM2B expression correlates with nucleolar size and prognosis in breast carcinoma 6 - which is a different tissue and is cited only to show the nucleolar connection is not a single-paper artefact.

Facts and figures
absence classmeasured-not-analysed
intersection publications0
methodological prerequisiterDNA repeats are absent from standard human genome assemblies, so every published KMT2A-fusion ChIP-seq has discarded rDNA reads. Seeing this at all required custom rDNA assemblies (PMID 42096476). A re-analysis of existing KMT2A-fusion ChIP-seq against an rDNA-containing assembly is a computational experiment with no new wet work.
query trap foundMLL[tiab] matches as a free-text token in microbiology and environmental-engineering abstracts; 10 of 26 records in the rDNA cross query were biohydrogen, melanin, lipase and wastewater papers. Proposed for data/symbol-collisions.json.
extendslocalization-nucleolus-kmt2a-fusion, which holds the fusion half and does not mention KDM2B or SETD1A
absence class previous freetextthree independent published observations, no publication containing two of them
absence class basisThe entry records that standard assemblies omit rDNA and that every KMT2A ChIP-seq discarded those reads. The reads exist in deposited data; the re-map was never run. ACCESSION NAMED 2026-09-21: GSE127508 carries 79 SRA experiments (SRX5444450 verified live), so the rDNA re-map has raw reads to work from. This closes the checker's 'claims data exists but names no accession' warning on this entry.
absence class assigned2026-09-21 (cycle 47, king)
the three lines
[
  {
    "what": "KDM2B binds the transcribed region of rDNA and represses rRNA transcription by demethylating nucleolar H3K4me3",
    "pmid": "17994099",
    "year": 2007
  },
  {
    "what": "KMT2A and KMT2F/SETD1A bind entire rDNA loci; KMT2F deposits the H3K4me3 that activates Pol I",
    "pmid": "42096476",
    "year": 2026
  },
  {
    "what": "The MLL amino terminus targets MLL, MLL-AF9 and MLL-AF4 to the nucleolus; MLL-AF4 is predominantly nucleolar and interferes with Pol I",
    "year": "2000 and 2022",
    "pmids": [
      "11069025",
      "35468859"
    ]
  }
]
raw data availability
{
  "raw_reads_confirmed_available": true,
  "checked": "2026-09-21 (king)",
  "accession": "GSE127508 (PMID 31821784), SuperSeries of GSE127506, GSE127507, GSE137813, GSE137814",
  "evidence": "GEO SOFT family file lists 78 samples and 79 SRA experiment ids. SRX5444450 verified live in SRA: GSM3635632, MOLM13_RNA_DMSO_d2_REP1, Illumina, 6,230,066 spots, 467,254,950 bases, Homo sapiens.",
  "route_that_works": "https://ftp.ncbi.nlm.nih.gov/geo/series/GSE127nnn/GSE127508/soft/GSE127508_family.soft.gz -- the SOFT family file, which is authoritative and is not behind bot detection.",
  "routes_that_gave_a_FALSE_NEGATIVE": [
    "elink dbfrom=gds db=sra on the SuperSeries UID returns NO LINK; SuperSeries do not carry sample-level SRA links, their subseries do.",
    "esearch db=sra term=PRJNA524881[BioProject] returns 0, while the SRX ids under that same BioProject resolve individually. The field query is a false zero.",
    "Fetching the GEO acc.cgi HTML returned a reCAPTCHA challenge page, which parsed cleanly as '0 subseries, no SRA' -- a bot-detection page is not a negative result."
  ],
  "consequence": "The rDNA re-map and any gained-peak reanalysis CAN be done from raw reads for this dataset. An earlier read of this record as coverage-tracks-only (TDF/WIG) was wrong; those are merely the supplementary files listed on the SuperSeries."
}
How this could bite us back
  • Ribosome biogenesis is the shared endpoint of several menin-adjacent observations in this graph, including the WDR5 WIN-site inhibitor result where ribosome subunit attrition and p53-MDM4 activation dominate the response. An intervention that moves rDNA transcription in a KMT2A-rearranged cell may be acting on the nucleolus and being read as chromatin biology.
  • Any CXXC-directed molecule would engage both KDM2B and the fusion in the same compartment, with opposite signs on Pol I output. The net effect on ribosome biogenesis is unpredictable and is a plausible source of on-target toxicity in a growing child.
Still open
  1. Does KMT2A-fusion occupancy at rDNA change when KDM2B is degraded? One degron line, one ChIP against an rDNA-containing assembly.
  2. Is KDM2B's nucleolar address CXXC-dependent? The 2007 paper attributes targeting to nucleolar localisation and activity to JmjC, and never tests the ZF-CxxC - which is the question that would make this a CXXC story rather than a compartment coincidence.
  3. Can existing published KMT2A-fusion ChIP-seq be re-mapped against an rDNA-containing assembly? Every such dataset currently throws these reads away, so the answer may already be in the raw data.
  4. If MLL-AF4 interferes with Pol I and KDM2B represses Pol I, do they have additive, redundant or opposing effects on rRNA output in a KMT2A-rearranged cell?
Every source behind this entry (6)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Human JHDM1B/FBXL10 (KDM2B) is a nucleolar protein that preferentially binds the transcribed region of ribosomal DNA and represses rRNA transcription, JmjC-dependently, by demethylating H3K4me3 in the nucleolus, with a JmjC-dependent negative effect on cell size and proliferation.

    PMID 17994099DOI
    What this evidence cannot show

    Human somatic cell lines, 2007, with immunofluorescence, ChIP at rDNA and nuclear-run-on-type readouts. Nothing haematopoietic and nothing leukaemic. The paper attributes the address to nucleolar localisation and does not test whether the ZF-CxxC domain is what targets it there.

    Evidence that argues against this (1)
    • KDM2B's better-known address is as the DNA-targeting subunit of non-canonical PRC1.1 at CpG islands genome-wide, which is a different compartment and a different activity from nucleolar rDNA repression; which dominates in a given cell is not established. Blackledge NP et al. 2010
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Using customised rDNA genome assemblies, KMT2A and KMT2F (SETD1A) were shown to bind entire rDNA loci with matching H3K4me2 and H3K4me3, KMT2F is the primary depositor of H3K4me3 on rDNA, KMT2-specific subunits interact with the RNA Pol I transcriptional machinery, and loss of H3K4me3 represses the rDNA locus.

    PMID 42096476DOI
    What this evidence cannot show

    Mammalian cells, 2026, and the result depends on a custom assembly - standard human genome assemblies omit the rDNA repeats, so this is not comparable to and cannot be recovered from existing ChIP-seq processed against the reference. Wild-type KMT2A, not a fusion. Abstract read this session; full text not read.

    Evidence that argues against this (1)
    • KDM2B was shown nineteen years earlier to demethylate H3K4me3 at rDNA and repress it; if both are true, rDNA H3K4me3 is under active push-pull and neither paper accounts for the other. Frescas D et al. 2007
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The MLL amino terminus targets MLL to the nucleolus, nuclear matrix and mitotic chromosomal scaffolds, and conditionally expressed MLL-AF9 and MLL-AF4 chimeras show the same nucleoplasm-plus-nucleolus pattern, indicating a common address conferred by the MLL amino terminus.

    PMID 11069025DOI
    What this evidence cannot show

    Conditional expression of exogenous constructs in U937 and HeLa; immunofluorescence, confocal and immuno-electron microscopy. Overexpressed basic arginine-rich sequences partition into the nucleolus as a recognised artefact class, a limitation this graph's localization-nucleolus-kmt2a-fusion entry states and this entry does not weaken.

    Evidence that argues against this (1)
    • No endogenously tagged KMT2A fusion has ever been imaged against a nucleolar marker in a t(4;11) line, so the nucleolar address of the fusion at physiological levels remains unestablished. Siemund AL et al. 2022
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Human MLL-AF4 localises predominantly inside the nucleolus, interferes with Pol I transcription and ribosome biogenesis, and turns on the nucleolar stress pathway, while a murinised pSer variant localises more to the nucleoplasm.

    PMID 35468859DOI
    What this evidence cannot show

    Exogenous expression in mammalian cells with gene-expression profiling and localisation studies; not endogenous fusion in a patient-derived t(4;11) line. The species comparison is the paper's main point and complicates reading the human result on its own.

    Evidence that argues against this (1)
    • Nucleolar accumulation of an overexpressed basic protein is a known artefact class, and the same limitation applies here as to the 2000 imaging study. Caslini C et al. 2000
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    Bounded searches return seven records connecting KDM2B to the nucleolus or rDNA, none mentioning KMT2A, and twenty-six records for MLL or KMT2A with nucleolar or rDNA terms, of which ten are microbiology noise matching MLL as a free-text token and only three are on-topic nucleolar or rDNA studies, none of which names a second CXXC protein.

    PubMed E-utilities esearch record counts with querytranslation inspected and every returned title read · Oncofusion knowledge-graph project · NCBI E-utilities, queries run 2026-09-21; exact query strings given in the caveat of each item that uses this source 2026
    What this evidence cannot show

    Queries `(KDM2B OR FBXL10 OR JHDM1B OR NDY1)[tiab] AND (nucleolus OR nucleolar OR rDNA OR "ribosomal DNA")[tiab]` (7 records) and `(MLL OR KMT2A)[tiab] AND (rDNA OR "ribosomal DNA" OR nucleolar)[tiab]` (26 records), run 2026-09-21 with every title read. Title/abstract only - per rule 80 a nucleolar co-localisation panel inside a paper is invisible here. The contamination count of 10/26 is a title-level judgement by this beat.

    Evidence that argues against this (1)
    • `nucleolus[tiab] AND (KMT2A OR MLL)[tiab] AND leukemia[tiab]` returns 6 and a further query on `(nucleolus OR nucleolar)[tiab] AND (CXXC OR ZF-CxxC)[tiab]` returns 5, all off-topic selenoproteins and plant proteins - so the absence survives three query shapes, but three shapes over one index is still one index. Oncofusion knowledge-graph project 2026
  6. E6weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    Combined KDM2A and KDM2B expression levels correlate with nucleolar size and prognosis in primary breast carcinomas.

    PMID 32901907DOI
    What this evidence cannot show

    Immunohistochemistry and nucleolar morphometry in breast carcinoma - a different tissue and disease, and a correlation. Used only to show the KDM2-nucleolus connection is not a single-paper artefact.

    Evidence that argues against this (1)
    • A combined-expression correlation cannot attribute the nucleolar phenotype to either paralogue, and the primary rDNA-binding result is for KDM2B alone. Frescas D et al. 2007
Connected to