Oncofusion

start / graph / absence-no-experiment-tests-cxxc-reader-competition-at-a-kmt2a-fusion-target

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

CORRECTED: two of eleven CXXC readers HAVE been measured alongside a KMT2A fusion, and the result supports the menin-interface hypothesis

also called CXXC competition bounded absence · who else reads the address - the missing experiment

CORRECTED 2026-09-21: this entry originally said nobody had measured two of these DNA-reading proteins in the same cells as the fusion, and a 2017 paper we already cite 29 times had done exactly that in human t(4;11) cells. Its answer is the interesting part - the fusion sits on only a fifth of the available addresses while the other readers sit on half and nearly all, so the fusion is being placed somewhere specific rather than competing for space.

The same thing, for a scientist

**CORRECTED 2026-09-21 (cycle 45). THE COUNTED ABSENCE BELOW IS FALSE.** PMID 28076791 (Kerry 2017, Cell Rep) reports ChIP-seq for **CFP1/CXXC1 and KDM2B in SEM**, a human t(4;11) KMT2A::AFF1 line, verified from the full text at the coordinator: *"we performed ChIP-seq for CFP1 ... and KDM2B ... in SEM cells. In contrast to MLL-AF4, CFP1 and KDM2B bound more ubiquitously to uCpG sites, being found at 50% and 89% of all uCpG sites, respectively."* So the figure is **at least 2 of 11, not 0 of 11**, and the rule-of-three bound below does not apply. **None of this is in the abstract** — rule 80 exactly, and this entry's own caveat predicted the failure. The paper was already cited 29 times in this graph. **The CONCLUSION survives and is strengthened.** MLL-AF4 occupies 20% of unmethylated CpG sites against CFP1's 50% and KDM2B's 89%, and MLL(N)–menin correlate at r² = 0.96 at fusion peaks against menin–CFP1 at 0.27. The address is not a contested resource because **the fusion is not competing for it** — it leaves four fifths of the available uCpG unused and is placed by menin, not by uCpG availability. That is direct, genome-wide, index-fusion, human-cell support for the menin-interface hypothesis. The original text follows unaltered. **The counted absence: of the eleven human CXXC-domain proteins other than KMT2A itself, 0 have a published measurement of their chromatin occupancy in the same cells as a KMT2A fusion protein.** 0/11, Wilson 95% 0-25.9%, rule of three <27.3% [E1]. Eight of the eleven do have *some* published experiment touching KMT2A-fusion biology - 8/11 = 72.7% (Wilson 43.4-90.3) - so this is not a neglected family, it is a family studied with every tool except the one the competition question needs [E1]. **State the null before reading anything into that**, per rule 84: the human genome has of the order of 25,000 CpG islands and eleven nuclear CXXC readers with different partner complexes, different secondary reader modules and different cell-type expression. **The null is that they bind different subsets at different times with different partners and never interfere**, and co-occupancy of a promoter class is close to a base rate in chromatin data. Nothing in this beat refutes that null. Three observations pull weakly against it and each is qualified: **KDM2A overexpression displaces wild-type MLL and LEDGF from chromatin and demonstrably does NOT displace the fusion** [E2]; **MBD1's CXXC3 recruits TET1 and TET1's catalytic activity then displaces MBD1**, the family's one documented reciprocal-occupancy cycle, in heterochromatin and not at CpG islands [E3]; and **CXXC5 is transcriptionally down-regulated in MLL-rearranged AML** - but also in t(8;21) and GATA2-mutant AML, so it is not fusion-specific [E4]. **The best-supported reading on current evidence is the null**, with the KDM2A result as the only direct test at a fusion and its answer being *no displacement*.

Go deeper
The queries, so the absence is bounded and not just asserted

All run 2026-09-21 with querytranslation inspected and every returned title read 1:

queryrecordscompetition/co-occupancy experiments among them
(KMT2A OR MLL OR MLL-AF9 OR MLL-AF4 OR MLL-ENL)[tiab] AND (CXXC1 OR CFP1 OR KDM2A OR KDM2B OR FBXL10 OR FBXL11 OR FBXL19 OR MBD1 OR CXXC4 OR CXXC5 OR TET1 OR TET3 OR DNMT1)[tiab]530
CXXC[tiab] AND (competition OR competitive OR compete OR competes)[tiab] AND (CpG OR chromatin)[tiab]10 - the one hit is a DNMT1 specificity analysis
(ZF-CxxC OR CXXC)[tiab] AND (redundant OR redundancy OR displace OR displacement OR co-occupancy OR cooccupancy)[tiab]160 - fifteen are copper/disulphide chemistry, one is Cfp1
("CpG island" OR "CpG islands")[tiab] AND (KMT2A OR MLL-AF9 OR MLL-AF4 OR "MLL fusion")[tiab] AND (occupancy OR ChIP OR binding)[tiab]10 - the hit is the 2003 fusion-dimerisation paper
(CXXC1 OR CFP1 OR SETD1A OR SETD1B OR SET1A OR SET1B)[tiab] AND (MLL1 OR KMT2A)[tiab] AND ("CpG island" OR "CpG islands" OR CpG)[tiab]20 - both reviews

Rule-67 control. The same query shapes return non-zero counts readily: the family-by-MLL cross returns 53, CXXC[tiab] alone returns hundreds, KDM2A[tiab] returns 191. The instrument can return a one.

Rule-80 bounding, and it matters more here than usual. Every count above is bounded by titles and abstracts. An occupancy measurement of a second CXXC reader in a KMT2A-fusion cell would almost certainly appear as a supporting ChIP panel rather than in a title - which is exactly the class of evidence rule 80 says a PubMed zero cannot reach. So the honest claim is: no paper is titled or abstracted as such, and the supporting figures of the field were not searched. Two candidate full texts to open before anyone quotes this absence in a grant: the ASH1L/KDM2A paper 2 and the CXXC-swap paper 5, both of which had the reagents to run it.

Rule-70 name-list bounding. The name list used was the twelve current HGNC symbols plus the retired symbols FBXL10, FBXL11, CFP1 and CGBP. It did not include JHDM1A, JHDM1B, NDY1, LCX, CXXC6, IDAX, RINF or CXXC3-as-a-protein-name in the cross query, because each of those is separately contaminated (measured in the individual gene entries) and would have inflated the count with noise rather than reaching new papers. A future re-run should add them and title-check.

The null, written down

Rule 84 requires the base rate before the observation. For this question the null has three parts and all three are quantitative:

  1. There are of the order of 25,000 CpG islands in the human genome and eleven nuclear CXXC readers. If each reader occupied even a tenth of them, pairwise overlap would be large by construction. Co-occupancy is therefore not evidence of competition and any study reporting overlap has measured the base rate.
  2. The readers are not equivalent. The systematic family comparison classifies the human CXXC domains into several distinct DNA-selectivity groups 6, and TET3's domain prefers 5-carboxylcytosine over unmethylated CpG outright 7. Different ligands means different addresses at the same nominal element.
  3. At least two of the eleven may not be in the nucleus at all in a haematopoietic cell. CXXC4's only UniProt location term is cytoplasm, by similarity; FBXL19's experimental record is three cytoplasmic supports to one nuclear. The effective denominator for a competition question may be nine, not twelve.

Against that null, the finding of this beat is: nobody has run the measurement, and the one experiment that came closest returned no displacement of the fusion.

What would count as evidence either way

Competition is a claim about occupancy under perturbation, not about overlap. Four results would move it, in rough order of cost:

  1. Deplete one reader, ChIP the fusion. If KMT2A-fusion occupancy at shared CpG islands rises on CXXC1, KDM2A or KDM2B degradation, the address is contested. If it does not move, it is not. exp-cxxc-reader-depletion-versus-kmt2a-fusion-occupancy specifies this.
  2. Deplete the fusion, ChIP the readers. The reciprocal, and the cheaper half in a line that already has a fusion degron.
  3. Quantify the stoichiometry. Absolute copy numbers of the eleven readers and of the fusion, against the number of bound islands. If the readers are in tenfold excess of available islands the competition question is real; if islands are in excess it is not. No such measurement exists for any of the twelve.
  4. Repeat the MLL-AF9 swap across the whole family. Only 3 of 11 heterologous domains have ever been tested (DNMT1, CXXC1, MBD1; 3/11 = 27.3%, Wilson 9.7-56.6) 5. Eight domains have never been put in the fusion, and the one positive result - DNMT1 - was the least expected.
What this beat declined to claim

Recorded because a reader would reasonably expect these claims and should know they were considered and rejected:

  • 'The CXXC readers compete for the fusion's address.' Declined: no occupancy measurement exists and the one displacement experiment spared the fusion.
  • 'The fusion clears its competitors.' Declined: CXXC5 down-regulation in MLL-rearranged AML is the only supporting datum and it is shared with two non-KMT2A subtypes, so it is not fusion-attributable 4.
  • 'TET2 co-mutation contests the address.' Declined and refuted - see gene-tet2; TET2 is not a CXXC protein and TET2 mutation is significantly depleted in KMT2A-rearranged AML.
  • 'The address is a property of the methylome rather than of the fusion.' Not declined - partly supported, but through the CXXC-swap experiment's finding that CpG protection from methylation is the functional variable 5, and through the index cohort's authors concluding that the infant target cell 'may already have a chromatin state that is more permissive to transformation' 8. Neither route runs through TET2 and neither is a competition result.
Facts and figures
absence classnever-measured
denominator11
denominator definitionthe human CXXC-domain proteins other than KMT2A itself (CXXC1, CXXC4, CXXC5, DNMT1, FBXL19, KDM2A, KDM2B, KMT2B, MBD1, TET1, TET3)
null statedOf the order of 25,000 human CpG islands and at most eleven nuclear CXXC readers with different partner complexes, different secondary reader modules and different cell-type expression. Pairwise overlap of a promoter class is expected by construction; co-occupancy is therefore not evidence of competition. At least two of the eleven (CXXC4, FBXL19) may not be predominantly nuclear, so the effective denominator may be nine.
verdictThe null is the best-supported reading. The only direct displacement test at a fusion returned no displacement of the fusion.
search boundingtitles and abstracts only; supporting figures not searched; contaminated retired symbols (JHDM1A, JHDM1B, NDY1, LCX, CXXC6, IDAX, RINF) excluded from the cross query and measured separately in the individual gene entries
absence class previous freetextno experiment of the required design exists
absence class basisCorrected in cycle 45: two of eleven readers HAVE been co-measured, but no experiment of the competition design has been run.
absence class assigned2026-09-21 (cycle 47, king)
occupancy co measured with a kmt2a fusion
{
  "k": 0,
  "n": 11,
  "wilson_95": [
    0,
    25.9
  ],
  "rule_of_three_upper_pct": 27.3
}
any published experiment in a kmt2a fusion system
{
  "k": 8,
  "n": 11,
  "pct": 72.7,
  "wilson_95": [
    43.4,
    90.3
  ],
  "without": [
    "CXXC4",
    "FBXL19",
    "KMT2B"
  ]
}
tested in the mll af9 cxxc swap
{
  "k": 3,
  "n": 11,
  "pct": 27.3,
  "wilson_95": [
    9.7,
    56.6
  ],
  "tested": [
    "DNMT1",
    "CXXC1",
    "MBD1"
  ],
  "positive": [
    "DNMT1"
  ]
}
somatically mutated in a kmt2a rearranged cohort
{
  "k": 1,
  "n": 12,
  "pct": 8.3,
  "wilson_95": [
    1.5,
    35.4
  ],
  "which": [
    "TET3"
  ]
}
CORRECTION 2026 09 21
{
  "what_was_wrong": "0 of 11 CXXC readers co-measured with a KMT2A fusion",
  "truth": "at least 2 of 11 - CFP1/CXXC1 and KDM2B, ChIP-seq in SEM (human t(4;11))",
  "source": "PMID 28076791 full text, verified at the coordinator 2026-09-21",
  "why_it_was_missed": "the finding is absent from the abstract; the paper is cited 29 times in this graph for its spreading-domain result",
  "occupancy_of_unmethylated_cpg_sites": {
    "MLL-AF4": "20%",
    "CFP1": "50%",
    "KDM2B": "89%"
  },
  "menin_correlation_at_fusion_peaks": {
    "MLL(N)-menin r2": 0.96,
    "menin-CFP1 r2": 0.27
  },
  "effect_on_the_conclusion": "the conclusion is unchanged and better supported: the address is not contested because the fusion is not competing for it"
}
How this could bite us back
  • A CXXC-directed agent is being designed against a target whose functional relationship to its eleven paralogous domains has never been measured in a cell. The selectivity gate in the phase-2 dossier is therefore being set against in-vitro affinity ratios with no cellular readout to validate them.
  • If the null holds and the readers do not interfere, then a pan-CXXC agent would remove eleven activities and gain nothing in fusion selectivity - the worst case for that modality, and the case current evidence supports.
Still open
  1. Does depleting CXXC1, KDM2A or KDM2B change KMT2A-fusion occupancy at shared CpG islands? One degron line and two ChIPs.
  2. What are the absolute cellular copy numbers of the eleven readers and of the fusion, relative to the number of bound CpG islands? Without stoichiometry the competition question is not well posed.
  3. Do the ASH1L/KDM2A paper or the CXXC-swap paper contain an unreported CXXC-reader occupancy panel in a supporting figure? Both had the reagents; neither abstract mentions one.
  4. Would any of the eight untested heterologous CXXC domains support MLL-AF9 leukaemogenesis? DNMT1's did and it was the least expected of the three tested.
Every source behind this entry (9)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    Across five bounded PubMed queries with every returned title read, no study measures the chromatin occupancy of a non-KMT2A CXXC-family protein in the same cells as a KMT2A fusion protein: 0 of the 11 non-index family members, against 8 of 11 that do have some published experiment touching KMT2A-fusion biology.

    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

    Exact queries and counts are tabulated in the first section. Bounded by titles and abstracts only - per rule 80 an occupancy measurement in a supporting figure is invisible to this method, and that is precisely where such a measurement would live, so this is a floor on the literature and not a ceiling. The 8/11 numerator was assigned by this beat by reading each family member's dossier and counting any published experiment in a KMT2A-fusion system; the three without are CXXC4, FBXL19 and KMT2B. 0/11 gives Wilson 95% 0-25.9% and rule of three <27.3%; 8/11 gives Wilson 43.4-90.3%. The name list excluded several contaminated retired symbols, which is recorded in the section above.

    Evidence that argues against this (1)
    • The ASH1L/KDM2A study did measure occupancy of LEDGF, wild-type MLL and the fusion under KDM2A overexpression, which is one step from the missing experiment - so the absence is of a CXXC-reader ChIP, not of any occupancy work in this space. Zhu L et al. 2016
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    KDM2A overexpression reduces H3K36me2 and reduces chromatin occupancy of LEDGF and of wild-type MLL, but not of the MLL fusion, while antagonising MLL-associated leukaemogenesis.

    ASH1L Links Histone H3 Lysine 36 Dimethylation to MLL Leukemia. · Zhu L, Li Q, et al. · Cancer Discov 2016
    PMID 27154821DOI
    What this evidence cannot show

    Abstract read this session; the occupancy detail is carried in this graph's existing entries mark-h3k36me2-3 and mechanism-does-the-fusion-need-the-ledgf-pwwp-arm, which were read before writing. The manipulation is overexpression, the variable is H3K36me2 rather than CpG occupancy, and the system is murine MLL leukaemia. It is the closest thing to the missing experiment and it is not the missing experiment.

    Evidence that argues against this (1)
    • An overexpression experiment can saturate an address in a way physiological levels never would, so a failure to displace the fusion under overexpression is a stronger negative than a failure at endogenous levels - which cuts in favour of the null. Zhu L et al. 2016
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    MBD1's CXXC3 domain enhances TET1 localisation to heterochromatin and TET1-mediated 5mC oxidation, and TET1 recruitment and activity then displace MBD1 from methylated DNA - a reciprocal occupancy cycle between two CXXC-family members.

    PMID 28449087DOI
    What this evidence cannot show

    Mammalian cell lines, fluorescent-protein fusions and live imaging; the compartment is pericentromeric heterochromatin, not CpG islands, and displacement is inferred from imaging dynamics rather than ChIP. No leukaemia and no KMT2A fusion.

    Evidence that argues against this (1)
    • This is recruitment followed by displacement, i.e. co-operation then hand-off, which is a different relationship from competition for a limiting resource and should not be quoted as evidence for the latter. Zhang P et al. 2017
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    CXXC5 mRNA is down-regulated in AML with MLL rearrangements, and also in t(8;21) and GATA2-mutant AML, with evidence for silencing by promoter methylation.

    PMID 25805812DOI
    What this evidence cannot show

    Adult AML expression cohort; mRNA only, no protein and no occupancy. Because the down-regulation is shared with two subtypes that have no KMT2A rearrangement, it cannot be attributed to the fusion and cannot support a 'the fusion clears its competitor' reading.

    Evidence that argues against this (1)
    • A second large AML cohort reads the same prognostic direction as evidence that CXXC5 is oncogenic and anti-apoptotic, which would make its loss in MLL-rearranged AML a disadvantage to the leukaemia rather than an advantage. Astori A et al. 2013
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Only three heterologous CXXC domains - DNMT1, CGBP/CXXC1 and MBD1 - have ever been substituted for the MLL CXXC in a leukaemogenic fusion, and of the three only DNMT1's supported leukaemogenesis.

    Functional specificity of CpG DNA-binding CXXC domains in mixed lineage leukemia. · Risner LE, Kuntimaddi A, et al. · J Biol Chem 2013
    PMID 23990460DOI
    What this evidence cannot show

    Murine MLL-AF9 myeloid system with transplantation. 3 of the 11 available heterologous domains is 27.3% (Wilson 9.7-56.6%), so eight domains are untested and the sample is too small to say anything about what fraction would work.

    Evidence that argues against this (1)
    • One of the three, MBD1, did not bind DNA in the assay, so the informative sample is really two - and a two-domain sample in which one works and one does not is uninformative about the family. Risner LE et al. 2013
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The human CXXC domains fall into several distinct DNA-selectivity classes measured by multiple binding assays and supported by co-crystal structures, rather than constituting one uniform reader class.

    PMID 29276034DOI
    What this evidence cannot show

    Isolated domains, short duplexes, in vitro. A selectivity class measured on oligonucleotides need not correspond to a distinct genomic address.

    Evidence that argues against this (1)
    • Single-molecule work on a long DNA carrying a real CpG island shows spatially dependent collaboration among CpGs that oligonucleotide assays cannot see, so the class boundaries may not survive on genomic substrate. Liang L et al. 2021
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    TET3's CXXC domain binds unmethylated CpG but with highest affinity toward 5-carboxylcytosine, so at least one family member's preferred ligand is not the unmethylated CpG the KMT2A CXXC requires.

    PMID 26774490DOI
    What this evidence cannot show

    Isolated domain binding plus a co-crystal; genomic mapping in mouse neuronal cells. Whether 5caC is abundant at CpG islands in a haematopoietic cell is not addressed.

    Evidence that argues against this (1)
    • A direct MLL1-versus-TET3 comparison frames TET3-CXXC as tolerating modifications rather than preferring one, a weaker claim than a preference. Stroynowska-Czerwinska A et al. 2018
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The authors of the index whole-genome study of infant KMT2A-rearranged ALL conclude from the age difference in epigenetic-regulator mutation burden that the target cell in infants may already have a chromatin state more permissive to transformation by the MLL rearrangement.

    PMID 25730765DOI
    What this evidence cannot show

    This is the authors' interpretation of a mutation-burden difference (9/20 non-infant versus 3/22 infant, P = 0.04), not a chromatin measurement. It is quoted here because it is the field's own statement that the address may be a property of the cell, and because it does NOT run through TET2 or any CXXC gene.

    Evidence that argues against this (1)
    • A lower epigenetic-mutation burden in infants is equally consistent with a shorter time for mutations to accumulate before a potent fusion transforms the cell, which requires no permissive-chromatin hypothesis at all. Andersson AK et al. 2015
  9. E_k45_correctionstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    THE ABSENCE ASSERTED BY THIS ENTRY IS FALSE. Kerry 2017 performed ChIP-seq for CFP1/CXXC1 and KDM2B in SEM cells, a human t(4;11) KMT2A::AFF1 line, alongside MLL(N), AF4(C), menin, PAF1 and LEO1. MLL-AF4 was found at 20% of unmethylated CpG sites, CFP1 at 50% and KDM2B at 89%. MLL(N) and menin correlate at r-squared 0.96 at fusion peaks, while menin and CFP1 correlate at 0.27 and KDM2B does not correlate with menin at all.

    PMID 28076791DOI
    What this evidence cannot show

    Full text read at the coordinator 2026-09-21 and the quoted sentence verified verbatim; the percentages come from the same passage and Figure S1F, which was not itself inspected. One cell line, SEM. The occupancy percentages are of unmethylated-CpG sites genome-wide, not of the fusion's target genes specifically, so they bound availability rather than competition directly.

    Evidence that argues against this (1)
    • Co-measurement is not a competition experiment. Kerry shows three CXXC proteins occupying overlapping site sets in one cell line; it does not deplete one and measure another's redistribution, which is what the entry's companion experiment node specifies and what would actually test competition. So the narrower absence - no DEPLETION-AND-REDISTRIBUTION experiment - still holds, and only the co-measurement claim was wrong. Kerry J et al. 2017
Connected to