Every MS-evidenced KMT2A modification site, flagged retained or lost against the breakpoint - and the retained portion is NOT enriched for acetylation once you use the right null
also called KMT2A PTM retained lost table · 174 sites · which KMT2A modifications survive the fusion
We listed all 174 chemical tags that mass spectrometers have ever found on the MLL1 protein and asked, for each one, whether it survives into the cancer fusion protein or is thrown away. The surprise is a negative: none of the four SUMO tags survives, while the tags that do survive are simply the ones that happen to sit on the part of the protein that is kept - there is nothing special about them.
The same thing, for a scientist
174 MS-evidenced site/modification calls on Q03164 were classified against the KMT2A breakpoint cluster: RETAINED (<1338), VARIABLE (1338-1607, inside the cluster, depends on the individual breakpoint), LOST (>1607). 71 are retained, 4 are variable, 99 are lost. The rule-84 test is the point of this entry. Against a LENGTH null (33.7%) every modification class looks enriched in the retained portion; against the correct residue-composition null it does not. The retained third of KMT2A holds 45.75% of its lysines, so acetylation at 10/21 = 47.6% retained is exactly the base rate (exact two-sided p = 1.00). Phosphorylation is modestly enriched (56/130 = 43.1% vs 34.3% S/T/Y, p = 0.042) and rests on a single deposit. SUMOylation is 0/4 retained, which at n = 4 bounds nothing (rule of three: < 75%).
The boundaries, and why three classes not two
From ana-kmt2a-exon-to-domain-map (Ensembl ENST00000534358 × UniProt Q03164): exon 7 = aa 1212-1338, exon 8 = aa 1338-1362, and the leukaemia breakpoint cluster spans exons 8-14 ≈ aa 1338-1607.
A two-way retained/lost flag is wrong inside the cluster, because whether a residue at 1400 survives depends on which intron broke in that patient. So:
| class | range | meaning |
|---|---|---|
| RETAINED | < 1338 | in the fusion protein of every KMT2A rearrangement |
| VARIABLE | 1338-1607 | inside the cluster — retained in some breakpoints, lost in others |
| LOST | > 1607 | absent from the fusion protein in every KMT2A rearrangement |
⚠️ This is not the first pass at this table - read `ana-kmt2a-phosphosite-atlas-retained-vs-lost` first
The phospho-contacts beat already pulled this same endpoint and reported the same totals - 130 phospho, 21 acetylation, 19 ubiquitinylation, 4 SUMOylation, 174 in all - and already did the phospho split and its composition null correctly (56/74 at a residue-1400 cut, 19.1% vs 14.7% per available Ser/Thr). That entry is the prior art and nothing here supersedes it.
What is new here, and only this:
| prior entry | here | |
|---|---|---|
| split residue | 1400, a "conventional protein-level stand-in" | 1338, the actual exon 7/8 boundary, with a third VARIABLE class for 1338-1607 |
| acetylation split | not performed | 10 retained / 1 variable / 10 lost |
| acetylation null | not performed | 0.458 lysine null, exact p = 1.00 - no enrichment |
| ubiquitination, SUMOylation splits | not performed | 5/2/12 and 0/1/3 |
| per-site identities persisted | phospho only | all 174, with PRIDE accessions |
| lysine density of the retained region | not reported | 45.75% of lysines in 33.7% of the length |
Does the 1400-versus-1338 difference change that entry's numbers? No, and it was checked rather than assumed. Exactly one site falls in 1338-1399: K1364 SUMOylation. It is not a phosphosite, and that entry reported no SUMO split, so every figure in it stands. Recorded because a split-residue disagreement is the kind of thing that later reads as a silent contradiction between two entries.
And a refinement to both. UniProt's own SITE features annotate real KMT2A translocation breakpoints, and the earliest is 1334..1335 (KMT2A-ZFYVE19) - four residues before the exon 7/8 boundary. So a strictly conservative RETAINED cut is < 1334, not < 1338. No site in this table moves: the nearest retained site is K1243, 91 residues clear.
The table
| modification | total | RETAINED | VARIABLE | LOST |
|---|---|---|---|---|
| Phosphorylation | 130 | 56 | 0 | 74 |
| Acetylation | 21 | 10 | 1 | 10 |
| Ubiquitinylation | 19 | 5 | 2 | 12 |
| SUMOylation | 4 | 0 | 1 | 3 |
| all | 174 | 71 | 4 | 99 |
Full per-site identities, with PRIDE deposit accessions and peptide sequences, are persisted in analyses/q165-the-retained-acetylation-site/ptm-retained-or-lost.json 2.
The retained acetylation sites, in order: K269, K277, K326, K365, K636, K928, K1130, K1133, K1235, K1243. Four of those ten are the Aguilar-Arnal/Choudhary set.
Retained ubiquitination sites: K277, K326, K439, K985, K1130. K1130 is on both lists.
Sites at or just past the boundary, which need naming individually because a two-way flag hides them: K1364 SUMOylation (exon 9, inside the cluster) and K1529 acetylation (inside the cluster). Both are VARIABLE, not lost. K1517, K1574 ubiquitination are also in the cluster.
🔴 Rule 84: the enrichment is an artefact of the wrong null
The retained portion is 1,337 of 3,969 residues = 33.7% by length. Against that null, every class looks enriched and the obvious story writes itself — the fusion keeps the regulated part.
That null is wrong. An acetylation site can only occur on a lysine. The correct question is: what share of KMT2A's lysines are retained?
45.75%. The retained third of the protein holds 140 of its 306 lysines. The AT-hook and CXXC region is lysine-rich, so a length null overstates lysine-modification enrichment by 12 percentage points before any data are looked at 3.
| modification | observed retained | correct null | exact two-sided p |
|---|---|---|---|
| Acetylation | 10/21 = 0.476 | 0.458 (lysines) | 1.000 |
| Ubiquitinylation | 5/19 = 0.263 | 0.458 (lysines) | 0.108 |
| SUMOylation | 0/4 = 0.000 | 0.458 (lysines) | 0.130 |
| Phosphorylation | 56/130 = 0.431 | 0.343 (S/T/Y) | 0.042 |
🟠 So: acetylation sites in KMT2A are distributed exactly as its lysines are. There is no signal that the fusion preferentially keeps acetylation sites. The interesting fact about K1130/K1133 is its position — 17 residues from the CXXC — not the mere fact of its retention, which is what 47.6% of KMT2A acetyl-lysines do.
🟡 Phosphorylation is the one class that survives its null, at p = 0.042. Treat that carefully: all 130 phosphosites come from a single deposit (PXD012174), so this is one experiment's view, not a consensus, and a single-deposit result at p = 0.042 across four tests performed is not a finding.
🟢 SUMOylation 0/4 retained is the most interesting row and the least conclusive. Rule of three puts the true retained rate below 75% at n = 4 — which excludes essentially nothing. It is a lead, not a result. But it is a lead with a mechanism behind it: K2528, the best-supported SUMO site on KMT2A (8 independent deposits), is deep in lost sequence, and if SUMOylation is a regulatory layer KMT2A has and the fusion does not, that is a subtraction nobody has written down. It needs a properly powered SUMO proteomics dataset, not more inference from n = 4.
Coverage caveat that limits every row
This is a census of MS-evidenced sites in reprocessed PRIDE deposits. It is not a census of known KMT2A modifications. Anything characterised by low-throughput methods and never seen by a mass spectrometer in a reprocessed submission is absent from the table — including, notably, the taspase1 cleavage sites, which are proteolysis and not in scope, and any modification described only by antibody. A zero in this table is a zero in one pipeline.
| script | analyses/q165-the-retained-acetylation-site/ptm-retained-or-lost.py |
|---|---|
| status | run |
| result summary | 174 sites classified: 71 RETAINED, 4 VARIABLE, 99 LOST. Acetylation retention is at its lysine base rate (p=1.00). SUMOylation 0/4 retained but n=4 bounds nothing. All four SUMO sites and the best-supported one (K2528, 8 deposits) are at or past the breakpoint. |
| classification rule | <1338 RETAINED; 1338-1607 VARIABLE; >1607 LOST |
| n sites | 174 |
| n retained | 71 |
| n variable | 4 |
| n lost | 99 |
| relationship to existing entries | ana-kmt2a-phosphosite-atlas-retained-vs-lost (phospho-contacts beat) pulled the same endpoint and reported the same 174-site totals with a correct phospho split and composition null at a residue-1400 cut. This entry adds: the exon-derived 1338 cut with a third VARIABLE class, the acetylation/ubiquitination/SUMOylation splits which that entry did not perform, the lysine-composition null showing acetylation retention is AT base rate, and persisted per-site identities for all four modification classes. It does not supersede that entry and its 1400-vs-1338 difference was checked to change none of that entry's figures (only K1364 SUMOylation falls in the gap). |
| strictest retained cut | 1334 |
inputs
[ "EBI Proteins API /proteomics-ptm/Q03164", "UniProt Q03164.fasta (3,969 aa)", "ana-kmt2a-exon-to-domain-map exon boundaries" ]
breakpoint cluster AA
[ 1338, 1607 ]
retained acetylation sites
[ 269, 277, 326, 365, 636, 928, 1130, 1133, 1235, 1243 ]
retained ubiquitination sites
[ 277, 326, 439, 985, 1130 ]
variable sites
{
"K1364": "SUMOylation",
"K1517": "Ubiquitinylation",
"K1529": "Acetylation",
"K1574": "Ubiquitinylation"
}sumoylation sites all
{
"K1364": "VARIABLE",
"K2461": "LOST",
"K2528": "LOST (8 deposits - best-supported PTM site on the protein)",
"K3804": "LOST"
}dual modified lysines
{
"K277": [
"Acetylation",
"Ubiquitinylation"
],
"K326": [
"Acetylation",
"Ubiquitinylation"
],
"K1130": [
"Acetylation",
"Ubiquitinylation"
],
"K3804": [
"SUMOylation",
"Ubiquitinylation"
]
}measurements — the numbers behind the chart
[
{
"quantity": "retained share of lysines",
"subject": "Q03164 aa 1-1337 vs full length",
"value": 0.4575,
"units": "fraction",
"n": 306,
"universe": "all 306 lysines in Q03164",
"null": 0.3369,
"note": "the LENGTH share is 0.3369; using it as the null for a lysine modification overstates enrichment by 12 points before any data"
},
{
"quantity": "retained share of acetylation sites",
"subject": "21 MS-evidenced acetylation calls on Q03164",
"value": 0.4762,
"units": "fraction",
"n": 21,
"universe": "EBI reprocessed-PRIDE acetylation calls on Q03164",
"null": 0.4575,
"note": "exact two-sided binomial p = 1.000 against the lysine null - indistinguishable from base rate"
},
{
"quantity": "retained share of phosphosites",
"subject": "130 MS-evidenced phosphorylation calls on Q03164, all from PXD012174",
"value": 0.4308,
"units": "fraction",
"n": 130,
"universe": "EBI reprocessed-PRIDE phosphorylation calls on Q03164",
"null": 0.3432,
"note": "exact two-sided p = 0.042; does NOT survive Bonferroni across the 4 tests performed (0.0125); single deposit"
},
{
"quantity": "retained SUMOylation sites",
"subject": "4 MS-evidenced SUMOylation calls on Q03164",
"value": 0,
"units": "sites",
"n": 4,
"universe": "EBI reprocessed-PRIDE SUMOylation calls on Q03164",
"null": 0.4575,
"note": "rule of three: true retained rate below 0.75 at 95% - excludes almost nothing. A lead, not a result."
}
]uniprot annotated breakpoints
{
"1362": "KMT2A-FRYL",
"1334..1335": "KMT2A-ZFYVE19",
"1362..1363": "KMT2A-AF3P21 / KMT2A-KNL1 / KMT2A-CENPK",
"1406..1407": "KMT2A-AFF4",
"1444..1445": "KMT2A-GAS7 / KMT2A-LPP"
}taspase1 sites
{
"CS1": "2666..2667",
"CS2": "2718..2719",
"chains": {
"N320": "1..2718",
"C180": "2719..3969"
}
}uniprot curated mod res
{
"n_total": 34,
"n_retained_below_1334": 14,
"n_lost": 19,
"retained": {
"S136": "Phosphoserine; by CK2 (ECO:0000269 PubMed 29997176)",
"S142": "Phosphoserine; by CK2 (ECO:0000269 PubMed 29997176)",
"S153": "Phosphoserine",
"S197": "Phosphoserine",
"K239": "N6-acetyllysine",
"K373": "N6-acetyllysine",
"S518": "Phosphoserine",
"K636": "N6-acetyllysine",
"S680": "Phosphoserine",
"T840": "Phosphothreonine",
"S926": "Phosphoserine",
"S1056": "Phosphoserine",
"K1130": "N6-acetyllysine",
"K1235": "N6-acetyllysine"
},
"note": "UniProt curates K239 and K373 as acetylated and the MS endpoint does not; the MS endpoint reports acetyl-K1133 and UniProt does NOT curate it. The two layers are complementary and neither is a superset - which is why an absence in either is not an absence."
}- The most actionable-looking row - phosphorylation enriched in retained sequence - is the one resting on a single deposit and failing multiplicity correction. It is exactly the row a reader would carry away, and it is the weakest.
- Every retained site is retained on wild-type KMT2A too. Nothing in this table identifies a fusion-specific surface; by construction, a retained modification is shared with the normal protein in every cell in the body.
- Is KMT2A SUMOylation lost in the fusion? 0 of 4 MS-evidenced SUMO sites are retained and the best-supported one (K2528, eight deposits) is deep in lost sequence - but n = 4 bounds nothing. A dedicated SUMO-proteomics deposit, or a SUMO-IP western on a KMT2A-rearranged line versus a control, would settle it, and it would be a whole regulatory layer the fusion subtracts that nobody has named.
- What is at K1364? It is a SUMOylation site inside the breakpoint cluster, which means it is retained in some patients' fusion proteins and not others. If SUMOylation matters, that is a per-breakpoint difference in fusion biology that would be invisible to any analysis that treats KMT2A-r as one class.
- Do K277, K326 and K1130 - the three retained lysines that carry both acetylation and ubiquitination - behave as switches? Three dual-modified retained lysines is more than chance would put in 140 retained lysines and the pattern has never been noted for this protein.
- Is the phosphorylation enrichment real or an enrichment-chemistry artefact of PXD012174? A second independent phosphoproteome of KMT2A would answer it, and one deposit cannot.
- Why is K1133 absent from UniProt's curated feature table when K636, K1130 and K1235 are all there, and the MS layer supports K1133 better than K1130? Either UniProt's curation has a gap or the K1133 MS calls are localisation artefacts of the shared tryptic peptide. The discrepancy is small but it points at the one place the two independent layers disagree.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
The KMT2A leukaemia breakpoint cluster corresponds to exons 8-14, with exon 7 ending and exon 8 beginning at aa 1338 in UniProt Q03164 coordinates.
ana-kmt2a-exon-to-domain-map (this graph) - Ensembl ENST00000534358 CDS exon boundaries intersected with UniProt Q03164 · this project; Ensembl REST; UniProt REST 2026What this evidence cannot show
The 1607 upper bound of the cluster is taken from this graph's existing '~1338-1607 (exons 8-14)' statement and was NOT independently recomputed here from Ensembl exon 14's end coordinate. It is used only to separate VARIABLE from LOST; every RETAINED call depends solely on the 1338 boundary, which was verified in the exon map. If 1607 is wrong the VARIABLE/LOST split near the boundary shifts and nothing else does.
Evidence that argues against this (1)
- Rare KMT2A breakpoints fall outside the classical cluster, including 5' breaks in intron 6 and 3' breaks as far as intron 24 in therapy-related cases. For those, the RETAINED class is not universal. The flag should be read as 'retained under a classical exon 8-14 breakpoint', which covers the overwhelming majority of cases but not all of them.
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
174 distinct MS-evidenced site/modification calls exist on Q03164 in the EBI reprocessed-PRIDE set: 71 retained, 4 variable, 99 lost against the breakpoint cluster; SUMOylation is 0/4 retained.
EBI Proteins API - proteomics-ptm, Q03164; classified in analyses/q165-the-retained-acetylation-site/ptm-retained-or-lost.py · EMBL-EBI Proteins API 2026What this evidence cannot show
The script asserts the sequence length is 3,969, that K1130/K1133 are lysines, and that the immunising peptide APPIKPIKPVTR sits at 1126-1137, and exits non-zero if any fails - so a silent coordinate shift cannot pass. It does NOT verify the endpoint's peptide-to-protein assignments. Per-site identities and deposit accessions are persisted rather than only counts.
Evidence that argues against this (1)
- All 130 phosphosites come from one deposit (PXD012174). Any statement about the phosphorylation distribution is a statement about that one experiment's coverage, which will be biased by its enrichment chemistry and its cell type - phosphopeptide enrichment favours some sequence contexts strongly.
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
The retained portion of KMT2A (aa 1-1337) is 33.69% of the protein by length but contains 45.75% of its lysines and 34.32% of its serines, threonines and tyrosines; acetylation is not enriched in it against the lysine null (10/21, exact two-sided p = 1.00) while phosphorylation is modestly enriched against the S/T/Y null (56/130, p = 0.042).
analyses/q165-the-retained-acetylation-site/retained-enrichment-null.py · this project 2026What this evidence cannot show
Exact binomial against a fixed composition null. Four tests were performed with no multiplicity correction; at four tests the p = 0.042 for phosphorylation does not survive Bonferroni (0.0125) and should not be called significant. The composition null itself assumes every lysine is equally available for modification, which is false - surface accessibility, disorder and sequence context all matter, and the retained portion is more disordered than the C-terminal half. A disorder-matched null would be better and was not built.
Evidence that argues against this (1)
- This analysis refutes the intuitive framing that the fusion 'keeps the regulated part of KMT2A'. On the best available data it keeps the part of KMT2A that has the lysines. If a later, deeper acetylome changes 10/21 to, say, 25/35, the conclusion could flip in either direction - the current n is small enough that this test has little power to detect real enrichment.
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
UniProt's curated feature table for Q03164 annotates 34 modified residues, of which 14 lie below residue 1334; it lists N6-acetyllysine at K239, K373, K636, K1130 and K1235 but NOT at K1133, and annotates S136 and S142 as CK2 phosphosites with experimental evidence.
What this evidence cannot show
K1133's absence from UniProt curation is a genuine discrepancy with the MS layer, not a statement that the site does not exist - UniProt curates acetylation conservatively and the 2015 paper's antibody evidence may simply not have been processed. Per this project's rule 91, the presence of ECO:0000269 on S136/S142 means a curator read AN experiment in PMID 29997176, not that the paper is necessarily the founding or best observation for those residues.
Evidence that argues against this (1)
- UniProt curates acetyl-K239 and acetyl-K373, which the EBI MS endpoint does not report at all. So the MS census used for the main table here is NOT a superset of known KMT2A modifications, and the counts in this entry are a floor.
uses its exon 7/8 boundary at aa 1338 as the retained/lost cut
same endpoint pull, classified against the breakpoint
- part of →KMT2A / MLL1 — the gene that breaks
71 retained, 4 variable, 99 lost PTM sites on Q03164
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
which KMT2A modification sites survive into the fusion protein
one of 71 retained MS-evidenced sites
one of 71 retained MS-evidenced sites
one of 10 retained acetylation sites
one of 5 retained ubiquitination sites
extends the same endpoint pull to acetylation, ubiquitination and SUMOylation with a composition null; does not supersede it
one of 10 retained acetylation sites
the best-supported site on the protein, and LOST
one of 4 VARIABLE sites inside the cluster
same question asked of NUP98; KMT2A retains 71 sites, NUP98 retains none