Oncofusion

start / graph / ana-not-one-patient-phosphosite-is-in-a-folded-domain

Analysisunrated confidence 0 of 5: unrated

All 46 patient phosphosites sit outside folded domains, none of LEDGF's twelve is in the clamp that holds KMT2A, and six of KMT2A's eleven are not on the fusion protein

also called patient phosphosite domain mapping · LEDGF IBD phosphosite negative

We took every phosphorylation site measured in patients with this leukaemia and asked where on each protein it sits. Not one is inside a folded, structured part - they are all in the floppy stretches between the working parts.

The same thing, for a scientist

Mapping all 46 sites from `ana-phosphosites-observed-in-kmt2a-r-aml-patients` onto UniProt feature tables: **0 of 46 fall in a folded domain**, 2 in a named functional region, 19 in a low-complexity stretch and 23 in an explicitly disordered region. **None of LEDGF/PSIP1's twelve sites is in the integrase-binding domain (340-417)** - all twelve lie in 102-273 - so the KMT2A-LEDGF clamp is phospho-regulated from the KMT2A side only in the measured data. And **six of KMT2A's eleven sites lie beyond residue 1400**, past the breakpoint cluster, so they cannot be sites on the fusion protein at all.

Go deeper
Not one site is in a folded domain
bucketsites
folded domain (Domain, zinc finger, DNA-binding, repeat, coiled coil, motif)0
named functional region2
low-complexity stretch19
explicitly disordered region23
no annotation2
total46

Q25 found that none of ENL's 23 phosphosites lies in its YEATS domain and that all 23 are in its IDR, and asked whether that generalises. It does, across ten proteins and 46 sites, with no exceptions. Not one site is in a YEATS domain, a PWWP domain, a bromodomain, a PHD finger, a FYR module, a SET domain or an IBD.

This is what phospho-regulation of scaffolds usually looks like, so it is not a surprise so much as a constraint. Folded domains are where binding happens; disordered linkers are where charge is added to tune it. It means the experiments this graph should want are affinity and condensate measurements on phosphomimetic linkers, not structure-guided work on a domain surface - and it means a phosphosite in this family will almost never be visible in a crystal structure.

The bucketing had to be redone to get this right. The first pass counted UniProt 'Compositional bias' features - Polar residues, Pro residues, Basic and acidic residues - as named domains, and reported 21 of 46 inside one. A compositional-bias call is UniProt saying a stretch is low-complexity, which is a statement that it is not folded. Counting those as domains inverted the headline.

None of LEDGF's twelve sites is in the domain that holds KMT2A

LEDGF/PSIP1 carries twelve sites, more than any other protein in the table. Its integrase-binding domain is residues 340-417, and that is the surface that clamps the KMT2A N-terminus.

All twelve sites lie between 102 and 273. None is in the IBD, and none is within sixty residues of it. Nine of the twelve are UniProt-annotated phosphosites, so this is not a coverage problem in the source - these are well-known sites, they are simply all in the other half of the protein.

That makes the clamp asymmetric in a testable way. The CK2 work (PMID 29997176) shows the KMT2A side of this interface is a phospho-switch: S136 and S142 are CK2 substrates worth 2-10x in IBD affinity, S136A/S142A impairs transformation, and the phosphomimetic rescues it. S136 is in this patient table and is one of the five KMT2A sites that are on the fusion. So in the measured data the interface is regulated from one side only.

The honest reading is narrower than 'LEDGF is not regulated here'. Twelve sites in 102-273 is a lot of regulation - it is just not regulation of this interface. What those sites do is unknown, and 102-273 spans LEDGF's NLS (146-156) and its PWWP-adjacent charged region, which is where nucleosome and chromatin association would be tuned.

Six of KMT2A's eleven sites cannot be on the fusion protein

UniProt annotates recurrent translocation breakpoint sites in KMT2A at 1362/1363, 1406/1407 and 1444/1445. Taking 1400 as the working boundary:

  • On the fusion (retained N-terminal side): S136, S153, S504, S516, S1059 - five.
  • Wild-type allele only (beyond the breakpoint): S1837, T1839, T1845, S2098, T2857, S3036 - six.

So a table headed 'phosphosites in KMT2A-rearranged AML patients' is, for KMT2A itself, reporting the wild-type allele more often than the fusion. Every one of those six is a real measurement in a real patient; none of them can be a fusion site, because that sequence is not in the fusion protein.

This sharpens Q25's point rather than answering it. Q25 established that no phosphosite has ever been mapped on a KMT2A fusion protein, because retained-region tryptic peptides are identical between fusion and wild-type allele. This adds that for the majority of KMT2A's measured sites the question does not even arise - they are unambiguously wild-type. The five retained ones are exactly the ambiguous set, and S136 is among them.

Still open
  1. Which LEDGF isoform did the source's search database use? If it was p52, which lacks the IBD entirely, then 'no IBD phosphosite' is an artefact of the database and this entry's second finding is vacuous. Two of ten proteins in this table were already on non-canonical isoforms.
  2. Is the absence of folded-domain sites real or a detection artefact? The test is to ask the same question of a large unbiased phosphosite atlas: if folded domains are equally empty there, it is a property of phosphoproteomics rather than of this family.
  3. What do LEDGF's twelve sites at 102-273 actually do? They bracket the NLS at 146-156, which makes nuclear import or chromatin retention the obvious hypotheses, and neither has been tested in a KMT2A-rearranged context.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    Of 46 phosphosites measured in KMT2A-rearranged AML patients across ten proteins, none falls within a folded domain. Two lie in a named functional region, 19 in a low-complexity stretch, 23 in an explicitly disordered region and 2 carry no annotation.

    What this evidence cannot show

    Run 2026-09-20 against UniProt REST, with each accession's primary gene name asserted before use and the two known isoform offsets applied (MLLT10 -16, MEN1 -5). **The result is entirely dependent on UniProt's domain annotation**, which is incomplete and conservative: an unannotated folded region would be scored as disorder here. It is also a property of where phosphoproteomics finds sites - tryptic peptides from disordered, lysine- and arginine-rich linkers are easier to detect than peptides from tightly folded cores - so detection bias points in the same direction as the finding and cannot be separated from it with this data.

    Evidence that argues against this (1)
    • **The detection-bias objection is not small.** Disordered regions are over-represented in every phosphoproteomic dataset, for reasons of digestion and solubility as much as biology, so '0 of 46 in a folded domain' may be partly a statement about mass spectrometry. The claim that survives is the conditional one: *of the sites we can see, none is on a domain surface* - which still constrains which experiments are worth doing, but is not evidence that folded domains here are unphosphorylated. Oncofusion knowledge graph (internal analysis) 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    None of LEDGF/PSIP1's twelve patient phosphosites lies in its integrase-binding domain at residues 340-417; all twelve fall between 102 and 273, and nine of the twelve are independently annotated phosphosites in UniProt.

    What this evidence cannot show

    The IBD boundary 340-417 is UniProt's own Region annotation on O75475, read 2026-09-20. The source table is one 74-patient cohort, so absence of an IBD site here is absence in that cohort at that depth, not a demonstration that no IBD phosphosite exists. LEDGF has a second, shorter isoform (p52) that ends before the IBD entirely, and the source's search database isoform was not established for this protein. **The p52 risk is real and now verified**: UniProt O75475-2 (p52) replaces residues 326-333 and deletes 334-530, so it lacks the IBD entirely. A search against p52 could not report an IBD site at any depth. **Which isoform the source used has not been established.**

    Evidence that argues against this (2)
    • The p52 objection is the serious one and it is unresolved. Confirmed from the entry: **O75475-2 (p52) deletes 334-530 and therefore has no IBD**, while sharing the entire N-terminus. A search database built on p52 would place every detectable site in exactly the 102-273 window observed, making this negative an artefact rather than a finding. Two of the ten proteins in this table were already numbered on non-canonical isoforms, so assuming the canonical one here runs against the local base rate. UniProt Consortium 2026
    • Even taken at face value, a one-sided interface is not an unregulated one. The CK2 result shows the KMT2A side of this clamp is a functional phospho-switch, and an affinity switch needs only one side to be switched. Proceedings of the National Academy of Sciences 2018
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Six of KMT2A's eleven patient phosphosites - S1837, T1839, T1845, S2098, T2857 and S3036 - lie beyond residue 1400 and so beyond the breakpoint cluster region, and therefore cannot be sites on a KMT2A fusion protein. Five - S136, S153, S504, S516 and S1059 - are on retained sequence.

    What this evidence cannot show

    1400 is a working boundary, not a fact about any patient: UniProt annotates recurrent breakpoint sites at 1362/1363, 1406/1407 and 1444/1445, and an individual patient's breakpoint can fall outside that window - minor-cluster breaks at intron 20 to exon 24 would retain far more. So the split is right for the common architecture and wrong for a minority of patients, and the source cohort's breakpoints are not reported per patient. S1059's assignment is the least secure because it is nearest the boundary.

    Evidence that argues against this (1)
    • This cuts against reading the table as a fusion phosphoproteome, but it does not make the six sites uninteresting. Wild-type KMT2A is the allele CX-4945's route runs through, and it is the allele that supplies WRAD-loaded methyltransferase activity in a cell whose fusion has none - so sites on the wild-type allele bear directly on this graph's central questions. They are just not fusion sites. UniProt Consortium 2026
Connected to