Oncofusion

start / graph / ana-kmt2a-phosphosite-atlas-retained-vs-lost

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q22-phospho-contacts

56 of KMT2A's 130 observed phosphosites survive the translocation, and the retained fragment is phospho-denser than the part that is lost

also called KMT2A phosphosite atlas · retained vs lost phosphosites · Q03164 130 sites

We counted every place on KMT2A where a phosphate has actually been seen, then asked how many of those places are on the piece that leukaemia fusions keep. Forty-three percent of them are, and per unit length the kept piece is more heavily phosphorylated than the piece that is thrown away.

The same thing, for a scientist

UniProt annotates 25 phosphosites on Q03164. The PRIDE/PTMeXchange re-analysis served by the EBI Proteins API observes **130**, plus 21 acetylation, 19 ubiquitinylation and 4 SUMOylation sites. Splitting at residue 1400 (this graph's conventional protein-level stand-in for the breakpoint cluster region): **56 phosphosites retained, 74 lost**. The retained fragment is 35.3% of the chain but carries **43.1%** of the observed phosphosites - **40.0 sites per 1,000 residues retained against 28.8 lost**, or per available Ser/Thr, **19.1% against 14.7%**. **The fragment every KMT2A fusion keeps is not a phosphorylation-poor region.** That is the quantitative frame the beat needed: if only one phospho-gated contact had ever been found there, the explanation is that nobody looked, not that there is nothing to find.

Go deeper
The numbers
modificationretained (1-1400)lost (1401-3969)total
phosphorylation5674130
acetylation101121
ubiquitinylation51419
SUMOylation134
all72102174

Density, phosphorylation only:

  • retained: 56 / 1,400 aa = 40.0 per 1,000 residues; 56 / 293 Ser+Thr = 19.1% of available acceptors
  • lost: 74 / 2,569 aa = 28.8 per 1,000 residues; 74 / 502 Ser+Thr = 14.7% of available acceptors

UniProt's 25-site table is not a subset of the 130: it carries T840 and T3372, which the proteomics set does not. The union is therefore about 140 sites, and any statement of the form 'this residue is not phosphorylated' has to name which table it means 1.

Which of the retained sites have a named kinase

Five, at the time of writing, and they carry the whole beat:

sitekinasewhat it gatesentry
S136CK2LEDGF/p75 IBD affinityinterface-psip1-ibd-kmt2a-phosphoswitch
S142CK2samesame
S516ATRSCF(Skp2) binding and MLL degradationmechanism-atr-kmt2a-s516-skp2-switch
S518-observed; no kinase, no function-
S1241-observed; no kinase, no function-

51 of the 56 retained phosphosites have no assigned kinase and no assigned function. That is the size of the unworked space the interactome enumeration was pointing at when it said the enumeration is 'not at all' saturated for phosphorylation-dependent contacts.

Caveats that limit what the density comparison can carry

Shotgun phosphoproteomics detects what it can digest, ionise and localise. Three biases work in the same direction here and are not corrected for:

  1. Tryptic peptide length. Lys/Arg density varies along the chain; regions producing very short or very long peptides are under-sampled.
  2. Disorder. Phosphosites concentrate in disordered regions, and the KMT2A N-terminus is heavily disordered while the C-terminal half contains the folded PHD-bromo cassette and the SET domain. So part of the retained-fragment enrichment may be a disorder effect rather than a regulatory one.
  3. Protein processing. Wild-type KMT2A is cleaved by taspase1 into N and C fragments that persist as a complex; whether a given run sampled both halves equally is not knowable from the aggregate.

The comparison is therefore good enough to refute 'the retained fragment is phospho-poor' and not good enough to claim it is a phosphorylation hotspot.

Facts and figures
scriptanalyses/kmt2a-phosphosite-atlas/run.py
statusrun
uniprotQ03164
length AA3969
split residue1400
result summary43.1% of observed KMT2A phosphosites lie in the 35.3% of the chain that fusions retain; the retained fragment is phospho-denser per residue and per Ser/Thr than the lost fragment. 51 of 56 retained sites have no named kinase.
counts
sumo_total4.000sumo_total: 4lost_pct_of_ST14.7lost_pct_of_ST: 14.7ubiquityl_total19.0ubiquityl_total: 19retained_pct_of_ST19.1retained_pct_of_ST: 19.1acetyl_total21.0acetyl_total: 21uniprot_phospho25.0uniprot_phospho: 25lost_per_1000aa28.8lost_per_1000aa: 28.8retained_per_1000aa40.0retained_per_1000aa: 40phospho_retained56.0phospho_retained: 56phospho_lost74.0phospho_lost: 74phospho_total130.0phospho_total: 130all_sites174.0all_sites: 174
inputs
[
  "https://www.ebi.ac.uk/proteins/api/proteomics-ptm/Q03164",
  "https://rest.uniprot.org/uniprotkb/Q03164.json?fields=ft_mod_res"
]
counts — the numbers behind the chart
{
  "phospho_total": 130,
  "phospho_retained": 56,
  "phospho_lost": 74,
  "acetyl_total": 21,
  "ubiquityl_total": 19,
  "sumo_total": 4,
  "all_sites": 174,
  "uniprot_phospho": 25,
  "retained_per_1000aa": 40,
  "lost_per_1000aa": 28.8,
  "retained_pct_of_ST": 19.1,
  "lost_pct_of_ST": 14.7
}
Still open
  1. Run the same split on the 3' partners. If AFF1, MLLT1, MLLT3 and ELL are phospho-dense in their retained halves too, the fusion is a fusion of two regulatory regions, which is a different picture from the domain-centric one this graph draws.
  2. Is the retained-fragment phospho-enrichment explained by disorder? A disorder-matched null over the same protein would settle it in an afternoon.
  3. Which of the 51 unassigned retained sites lie in a strong kinase consensus? That would turn an atlas into a ranked list for a targeted phospho-IP.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The EMBL-EBI Proteins API proteomics-PTM endpoint reports 174 distinct modified residues on KMT2A Q03164 - 130 phosphorylation, 21 acetylation, 19 ubiquitinylation, 4 SUMOylation - of which 72 lie at or below residue 1400 and 102 above; phosphorylation splits 56 retained to 74 lost, giving 40.0 versus 28.8 sites per 1,000 residues and 19.1% versus 14.7% of available Ser/Thr acceptors. UniProt's own table for the same protein lists 25 phosphosites and is not a subset, carrying T840 and T3372 which the proteomics set does not.

    What this evidence cannot show

    Fetched and tabulated 2026-09-19. The residue-1400 split is a convention this graph already uses as a protein-level stand-in for the breakpoint cluster region, not a measured boundary; every count is reproducible from the endpoint so a reader can redraw the line. The density comparison is uncorrected for tryptic accessibility, for disorder, and for the fact that wild-type KMT2A is proteolytically processed into two fragments. Dataset composition is whatever PTMeXchange has reprocessed, none of it KMT2A-rearranged leukaemia.

    Evidence that argues against this (2)
    • Counting observed sites measures where mass spectrometers have looked, not where regulation happens. The same limitation cuts the other way for the retained fragment's conclusion: a dedicated KMT2A-rearranged AML phosphoproteomic study found the discriminating signal in phosphorylation of DOT1L and in CDK1 activity - partner and pathway rather than KMT2A residues - which is a reminder that the fusion's phospho-regulation may not live on KMT2A at all. Casado P et al. 2023
    • The retained-fragment enrichment may be a disorder artefact rather than a regulatory signal: KMT2A's N-terminal half is largely disordered and its C-terminal half contains folded PHD, bromodomain and SET modules, and phosphosite discovery is strongly biased towards disordered regions. No disorder-matched control was run here. UniProt Consortium 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Three of the five retained phosphosites with a named kinase have published functional consequences: S136 and S142 gate LEDGF/p75 IBD affinity and fusion transformation capacity, and S516 gates SCF(Skp2)-mediated degradation of MLL.

    PMID 29997176DOI
    What this evidence cannot show

    The S516 assignment comes from a separate paper (PMID 20818375) read at abstract level; the S136/S142 assignment is UniProt's curation of this one. 'Five with a named kinase' is a count over UniProt's kinase-attributed annotations plus the ATR literature and is a floor, not a census - a kinase named only in a paper UniProt has not curated would not appear.

    Evidence that argues against this (1)
    • Two of the five kinase-attributed retained sites are the same pair from one study, so the effective number of independent observations is three, not five. A count that treats database annotations as independent evidence overstates the coverage of this region. UniProt Consortium 2026
Connected to