Phosphosite counts for the five 3' partners: the atlas beats UniProt 1.4-6.0x, and the retained/lost split is completely different for each partner
also called MLLT1 MLLT3 MLLT10 ELL ELL2 phosphosites · proteomics-ptm partner counts · partner phosphosite census
We counted every place a phosphate has been seen on the five partner proteins that KMT2A fuses to, using a deep proteomics resource rather than the hand-curated database most people look at. The deep resource finds several times as many spots, and - the point of the exercise - whether those spots survive into the cancer protein differs wildly between partners: essentially all of them do for ENL, and only three of nine for AF9.
The same thing, for a scientist
The EMBL-EBI Proteins API proteomics-ptm endpoint (PRIDE/PTMeXchange re-analysis served through PeptideAtlas) was fetched for MLLT1 Q03111, MLLT3 P42568, MLLT10 P55197, ELL P55199, ELL2 O00472, EAF1 Q96JC9 and EAF2 Q96CJ1, with each accession's primary gene name and sequence length read back before use. Against UniProt's curated Modified-residue table the atlas reports 23 vs 5 (MLLT1), 7 vs 5 (MLLT3), 20 vs 7 (MLLT10), 12 vs 3 (ELL), 7 vs 2 (ELL2), 6 vs 1 (EAF1) and 8 vs 3 (EAF2) phosphosites - a 3.5x mean ratio against the 5.2x the previous beat reported for KMT2A, which this session reproduced exactly (130 vs 25 on Q03164) as an arithmetic check. The resources do not nest: UniProt carries MLLT3 S412/S419 and ELL2 S580 that the atlas does not. Mapped onto the fusion architectures this graph already holds, the retained fraction is partner-specific and not predictable from the count: MLLT1 23/23 retained, ELL 12/12 retained, MLLT10 16/20 firmly retained, and MLLT3 only 3 of 9 union sites retained, because about 92% of MLL-AF9 breakpoints fall at or after AF9 residue 376 and six of the observed sites sit N-terminal of it.
The counts, fetched 2026-09-19
| protein | accession | length | UniProt curated phospho | atlas phospho | ratio | union |
|---|---|---|---|---|---|---|
| MLLT1 (ENL) | Q03111 | 559 | 5 | 23 | 4.6x | 23 |
| MLLT3 (AF9) | P42568 | 568 | 5 | 7 | 1.4x | 9 |
| MLLT10 (AF10) | P55197 | 1068 | 7 | 20 | 2.9x | 20 |
| ELL | P55199 | 621 | 3 | 12 | 4.0x | 12 |
| ELL2 | O00472 | 640 | 2 | 7 | 3.5x | 8 |
| EAF1 | Q96JC9 | 268 | 1 | 6 | 6.0x | 6 |
| EAF2 | Q96CJ1 | 260 | 3 | 8 | 2.7x | 8 |
| KMT2A (control) | Q03164 | 3969 | 25 | 130 | 5.2x | - |
The KMT2A row is the previous beat's published figure, re-derived in this session from the same endpoint by the same arithmetic as a check that the peptide-offset calculation is right; it returned 130 exactly 12. Every position reported here was additionally checked against the UniProt FASTA to confirm the residue at that index is S, T or Y. All pass.
MLLT3's ratio of 1.4x is the outlier and is the interesting number. AF9 is not a lightly studied protein; it is one whose observed modification is concentrated in a serine-rich stretch, and the atlas adds almost nothing to the curated set.
Retained or lost - the only question this beat exists to answer
Breakpoint architectures are taken from this graph's own subtraction entries, which cite Hu et al. 2023 (PMID 36435883) and Hu & Muntean 2023 (PMID 37295550) 34.
MLLT1 (ENL) - 23 of 23 retained. About 84.1% of MLL-ENL patients break at or before ENL residue 5, so the fusion contains essentially the entire partner. Every observed phosphosite, from S155 to S480, is on retained sequence. This is the cleanest retained-phosphosite set of any protein in the KMT2A system, including KMT2A itself.
MLLT3 (AF9) - 3 of 9 retained. About 92% of MLL-AF9 patients break at or after AF9 residue 376. Retained: S412, S419, S483. Lost: S288, S294, S301, S309, S321, S364. The lost six sit inside or beside the serine-rich region; the retained three sit in the AHD-proximal C-terminus.
MLLT10 (AF10) - 16 of 20 firmly retained. The PZP module (aa 22-198) is discarded and carries no observed phosphosite at all. Four sites (S215, S217, S219, S252) lie between the PZP and the approximate first-retained-exon boundary this graph records (exon 10, about aa 266) and their fate depends on the exact break; MLLT10 rearrangements are structurally heterogeneous, so this is genuinely undetermined rather than merely unmeasured.
ELL - 12 of 12 retained. MLL-ELL contains ELL aa 46-621 and the most N-terminal observed site is S126.
ELL2 - not applicable. ELL2 is not a recurrent KMT2A 3' partner. It is included because it is ELL's paralogue and the SEC/LEC literature treats the two together, not because a KMT2A::ELL2 fusion is on this graph's partner list.
What the numbers do NOT say
A phosphosite count measures how often a peptide carrying a phosphate has been observed in public mass spectrometry, and nothing more. It carries no kinase, no stoichiometry, no cell type and no functional consequence. Of the 86 union sites across these seven proteins, five have a named kinase in the published literature - MLLT1's conserved SQ sites and MLLT3 S395 for the PIKK family, and ELL T551/S561/S589 for ATM - and every one of those was found through the literature, not through the atlas.
The asymmetry runs the other way too. AF9 S395, the best-characterised phospho-switch on any of these proteins, appears in neither the atlas nor UniProt's curated table: it is a damage-induced site, and unstimulated shotgun runs do not see it. A census of steady-state proteomics is systematically blind to exactly the class of site this beat was sent to find.
| status | run |
|---|---|
| mean ratio partners only | 3.5 |
| union sites total | 86 |
| union sites with a named kinase | 5 |
| result summary | Atlas beats UniProt 1.4-6.0x across the partners. Retained fraction is partner-specific: MLLT1 23/23, ELL 12/12, MLLT10 16/20, MLLT3 3/9. Five of 86 union sites have a named kinase and all five came from the literature, not from the atlas. |
inputs
[
"https://www.ebi.ac.uk/proteins/api/proteomics-ptm/{accession}",
"https://rest.uniprot.org/uniprotkb/{accession}.json?fields=ft_mod_res",
"https://rest.uniprot.org/uniprotkb/{accession}.fasta"
]counts — the numbers behind the chart
[
{
"gene": "MLLT1",
"uniprot": "Q03111",
"length_aa": 559,
"uniprot_phospho": 5,
"atlas_phospho": 23,
"ratio": 4.6,
"union": 23,
"atlas_positions": [
155,
261,
265,
266,
267,
292,
296,
314,
315,
323,
325,
359,
361,
366,
411,
414,
440,
442,
444,
447,
475,
478,
480
],
"uniprot_positions": [
155,
292,
296,
359,
361
],
"retained_in_fusion": 23,
"lost_in_fusion": 0,
"breakpoint_rule": "break at or before ENL aa 5 in 84.1% of patients"
},
{
"gene": "MLLT3",
"uniprot": "P42568",
"length_aa": 568,
"uniprot_phospho": 5,
"atlas_phospho": 7,
"ratio": 1.4,
"union": 9,
"atlas_positions": [
288,
294,
301,
309,
321,
364,
483
],
"uniprot_positions": [
288,
294,
412,
419,
483
],
"retained_in_fusion": 3,
"retained_positions": [
412,
419,
483
],
"lost_in_fusion": 6,
"lost_positions": [
288,
294,
301,
309,
321,
364
],
"breakpoint_rule": "break at or after AF9 aa 376 in about 92% of patients"
},
{
"gene": "MLLT10",
"uniprot": "P55197",
"length_aa": 1068,
"uniprot_phospho": 7,
"atlas_phospho": 20,
"ratio": 2.9,
"union": 20,
"atlas_positions": [
215,
217,
219,
252,
278,
370,
432,
436,
444,
456,
499,
519,
527,
532,
536,
679,
680,
684,
686,
689
],
"uniprot_positions": [
217,
252,
436,
532,
684,
686,
689
],
"retained_in_fusion": 16,
"undetermined_positions": [
215,
217,
219,
252
],
"lost_in_fusion": 0,
"breakpoint_rule": "PZP module aa 22-198 discarded; first retained exon recorded as exon 10, about aa 266"
},
{
"gene": "ELL",
"uniprot": "P55199",
"length_aa": 621,
"uniprot_phospho": 3,
"atlas_phospho": 12,
"ratio": 4,
"union": 12,
"atlas_positions": [
126,
180,
308,
309,
317,
319,
401,
437,
442,
444,
519,
561
],
"uniprot_positions": [
180,
309,
561
],
"retained_in_fusion": 12,
"lost_in_fusion": 0,
"breakpoint_rule": "MLL-ELL contains ELL aa 46-621"
},
{
"gene": "ELL2",
"uniprot": "O00472",
"length_aa": 640,
"uniprot_phospho": 2,
"atlas_phospho": 7,
"ratio": 3.5,
"union": 8,
"atlas_positions": [
305,
316,
416,
420,
502,
503,
604
],
"uniprot_positions": [
503,
580
],
"retained_in_fusion": null,
"note": "not a recurrent KMT2A 3' partner; included as ELL's paralogue"
},
{
"gene": "EAF1",
"uniprot": "Q96JC9",
"length_aa": 268,
"uniprot_phospho": 1,
"atlas_phospho": 6,
"ratio": 6,
"union": 6,
"atlas_positions": [
21,
157,
158,
165,
264,
266
],
"uniprot_positions": [
165
],
"note": "ELL-associated factor, not a fusion partner"
},
{
"gene": "EAF2",
"uniprot": "Q96CJ1",
"length_aa": 260,
"uniprot_phospho": 3,
"atlas_phospho": 8,
"ratio": 2.7,
"union": 8,
"atlas_positions": [
22,
144,
146,
151,
154,
207,
256,
258
],
"uniprot_positions": [
146,
151,
154
],
"note": "ELL-associated factor, not a fusion partner"
},
{
"gene": "KMT2A",
"uniprot": "Q03164",
"length_aa": 3969,
"uniprot_phospho": 25,
"atlas_phospho": 130,
"ratio": 5.2,
"note": "control; the previous beat's figure, reproduced exactly in this session"
}
]- A kinase inhibitor chosen from a phosphosite count would be chosen from a measure of mass-spectrometry sampling depth, not of biology. Nothing in this table licenses a target choice on its own.
- The retained/lost split makes MLLT1 look like the richest phospho-regulated fusion partner, which could push a programme toward ENL for a reason that is really about breakpoint position. Breakpoint position is a property of the patient population, not of the protein, and the minority ENL architecture that excludes the YEATS domain would also move several of these sites.
- Counting sites and then inhibiting the kinase that is assumed to make them inverts the evidence: for four of these seven proteins not one site in the table has a kinase assigned, so any inhibitor rationale built on this census would be assigning kinases by motif, which this project has already had to withdraw once.
- Is MLLT3's flat atlas-to-UniProt ratio (1.4x) a real signal that AF9 is lightly phosphorylated, or an artefact of its 42-residue polyserine stretch producing peptides that behave badly in tryptic digestion and phospho-enrichment?
- The four MLLT10 sites at S215-S252 sit in the zone whose fate depends on the exact breakpoint. Mapping MLLT10 breakpoints to residue rather than exon in a real patient series would settle four sites at once.
- Do any of the 23 retained MLLT1 sites change occupancy between KMT2A::MLLT1 cells and MLLT1-wild-type cells? No phosphoproteomic comparison stratified by fusion partner exists.
- AF9 S395 is in neither resource yet is the partner's best-characterised functional site. How many other damage- or signal-induced sites on these proteins are invisible to steady-state proteomics?
- 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 23 phosphorylation sites on MLLT1 (Q03111), 7 on MLLT3 (P42568), 20 on MLLT10 (P55197), 12 on ELL (P55199), 7 on ELL2 (O00472), 6 on EAF1 (Q96JC9) and 8 on EAF2 (Q96CJ1).
Proteins API proteomics-ptm: PTMeXchange/PeptideAtlas re-analysis for Q03111, P42568, P55197, P55199, O00472, Q96JC9, Q96CJ1 · EMBL-EBI 2026What this evidence cannot show
Fetched 2026-09-19. Site positions are computed as peptide begin plus intra-peptide offset minus 1, and were validated by confirming the residue at each computed index is S/T/Y in the UniProt FASTA; the same arithmetic reproduces the previously published KMT2A figure of 130 exactly. These are presence observations pooled across many unrelated studies and tissues, with no stoichiometry, no cell type and no kinase. Coverage tracks protein abundance and tryptic peptide behaviour, so a low count is not evidence of low modification.
Evidence that argues against this (1)
- The proteomics re-analysis is not a superset of the curated table and cannot be treated as a census either way. For MLLT3 the curated UniProt table carries S412 and S419, which the PeptideAtlas/PTMeXchange re-analysis does not report, and for ELL2 it carries S580, also absent from the re-analysis. So the two resources fail to nest in BOTH directions, and a count taken from either alone understates the observed set. UniProt Consortium 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
UniProt's curated Modified-residue tables carry 5 phosphosites for MLLT1, 5 for MLLT3, 7 for MLLT10, 3 for ELL, 2 for ELL2, 1 for EAF1 and 3 for EAF2; each accession's primary gene name and sequence length were read back and match the expected protein (MLLT1 559 aa, MLLT3 568 aa, MLLT10 1068 aa, ELL 621 aa, ELL2 640 aa).
UniProtKB entries Q03111, P42568, P55197, P55199, O00472, Q96JC9, Q96CJ1: primary gene name, sequence length and Modified residue features · UniProt Consortium 2026What this evidence cannot show
Accessions were resolved by direct accession fetch rather than by gene_exact search, precisely because gene_exact matches synonyms; the primary geneName value was read back for each and all seven matched the expected symbol. Curated tables are conservative by design and lag the literature.
Evidence that argues against this (1)
- Shotgun phosphoproteomic depth tracks protein abundance and peptide behaviour, not biological modification. The same disease-context argument the KMT2A-rearranged AML patient phosphoproteome makes against itself applies here: that study detects no peptide covering KMT2A S136 although S136 has direct functional evidence, purely for coverage reasons. Absence from either resource is therefore not evidence a residue is unmodified. Casado P et al. 2023
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Approximately 84.1% of MLL-ENL patients have breakpoints at or before ENL residue 5, so the ENL YEATS domain and everything C-terminal of it is retained in the fusion; deleting the YEATS domain from MLL-ENL impairs chromatin binding, lowers Eya1 and Meis1 expression and reduces leukaemic stem-cell frequency.
The ENL YEATS epigenetic reader domain critically links MLL-ENL to leukemic stem cell frequency in t(11;19) Leukemia. · Hu H, et al. · Leukemia 2023What this evidence cannot show
Read as abstract and through this graph's existing subtraction-mllt1-yeats-retained entry, which cites the same paper for the same figure; the primary full text was not re-read in this session. The 84.1% is a breakpoint-distribution statistic, and a minority ENL exon 6-7 architecture excludes YEATS entirely, with frequency unquantified here.
Evidence that argues against this (1)
- A partner-level review from the same group frames ENL and AF9 as a pair whose YEATS domains are therapeutically comparable, which cuts against reading ENL's near-complete retention as unique. The retained/lost asymmetry reported here is a sequence fact and does not by itself establish that retention is functionally decisive. Hu H et al. 2023
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Approximately 92% of MLL-AF9 patients have breakpoints at or after AF9 residue 376, so the AF9 YEATS domain (aa 1-138) and the serine-rich region are excluded from the fusion while the AHD is retained.
The YEATS domain epigenetic reader proteins ENL and AF9 and their therapeutic value in leukemia. · Hu H, et al. · Exp Hematol 2023What this evidence cannot show
Abstract-level, plus this graph's subtraction-mllt3-yeats entry, which cites this review for the 92% figure. It is a review rather than a primary breakpoint series, the figure is a pooled distribution, and individual cases fall outside it.
Evidence that argues against this (1)
- A structural and biophysical review of the same two proteins emphasises that the intrinsically disordered AHD is the functional unit engaged by AF4, DOT1L, BCOR and CBX8, and does not treat the discarded N-terminal region as regulatory. On that reading the loss of six phosphosites N-terminal of residue 376 may be incidental rather than meaningful. Kabra A et al. 2022
23 observed phosphosites, all retained
- related to →MLLT3 / AF9 — a famous fusion partner that is not a dependency, and a hub domain that is
9 union sites, only 3 retained
20 observed sites, none in the discarded PZP module
12 observed sites, all retained
- related to →ELL2 — a SEC subunit that almost no cell needs, next to a paralogue almost every cell needs
7 observed sites; paralogue control, not a KMT2A partner
same method applied to the 3-prime partners
worked example of a well-observed site with no kinase
beat ledger