The KMT2A-r AML patient table numbers MLLT10 on isoform P55197-1, not the canonical sequence: S674 is S658 and S705 is S689
also called MLLT10 S674 · MLLT10 S705 · AF10 isoform numbering · phosphosite numbering correction
A published patient dataset lists two phosphate spots on the AF10 protein, and against the standard version of that protein one is not even the right kind of amino acid while the other lands on a matching letter in the wrong place. Both are explained by the study using a shorter version of the protein: once you allow for that, both spots line up exactly.
The same thing, for a scientist
This graph's `ana-phosphosites-observed-in-kmt2a-r-aml-patients` reports MLLT10 S674 and S705 from the supplementary phosphosite matrix of a 74-patient KMT2A-rearranged AML study. A canonical-sequence check flagged S674 because residue 674 of P55197 is **leucine**. This analysis resolves the flag rather than dropping the site. The table's numbering is that of **UniProt isoform P55197-1 (1027 aa)**, not the canonical displayed sequence P55197 (1068 aa). Matching 21-residue sequence windows around each position gives an exact, unambiguous mapping: **isoform-1 S674 = canonical S658** (window `NSSMAALIAQSENNQTDQDLG`) and **isoform-1 S705 = canonical S689** (window `SSPRGSLSPRSPVSSLQIRYD`). Both are serines in both numbering systems. The mapping is independently corroborated: **canonical S689 is reported as a phosphosite by both the PeptideAtlas re-analysis and UniProt's curated table for P55197**, whereas canonical S705 (context `QIRYDQPGNSSLENLPPVAASI`) is a different residue in a different context and is reported by neither.
The check that succeeds for the wrong reason
Residue 705 of canonical P55197 is a serine, so a residue-type check passes it - and passing is misleading. The window around isoform-1 position 705 is SSPRGSLSPRSPVSSLQIRYD; the window around canonical position 705 is QIRYDQPGNSSLENLPPVAASI. These are different parts of the protein; the second begins where the first ends.
This is exactly the failure mode a residue-type check is known to have: roughly one residue in eight is a serine or threonine, so a coincidental pass is common. The site that failed the check (S674 to leucine) was the one that revealed the systematic offset; the site that passed it would have been silently recorded at the wrong position.
A residue-type check detects frame errors. It does not detect offsets. Window matching does, and costs one extra line of code.
The mapping
| table value | isoform used | 21-residue window | canonical position | canonical residue | in PeptideAtlas for P55197? |
|---|---|---|---|---|---|
| MLLT10 S674 | P55197-1 (1027 aa) | NSSMAALIAQSENNQTDQDLG | S658 | S | no |
| MLLT10 S705 | P55197-1 (1027 aa) | SSPRGSLSPRSPVSSLQIRYD | S689 | S | yes |
The two isoforms are identical for residues 1-566 and diverge thereafter; the offset in this region is canonical = isoform-1 minus 16 1. The corroboration at S689 is the part that makes this a resolution rather than a guess: an independent resource, using canonical numbering, reports a phosphosite at exactly the position the window mapping predicts 2.
Retained or lost
Both sites are retained. The discarded region of AF10 in KMT2A::MLLT10 is the N-terminal PZP module (aa 22-198) and adjoining sequence; this graph records the first retained MLLT10 exon as exon 10, approximately residue 266, while noting that MLLT10 rearrangements are structurally heterogeneous and cryptic so a single exon pair does not describe all cases 3.
S658 and S689 are both several hundred residues C-terminal of that boundary, and both are N-terminal of the OM-LZ (704-779) whose retention is the defining feature of these fusions. Any breakpoint that preserves the OM-LZ and follows the recorded architecture preserves both sites.
S689 in particular sits 15 residues before the OM-LZ, the module that binds the DOT1L coiled coil and is the oncogenic element of KMT2A::MLLT10. A phosphosite immediately N-terminal of the DOT1L-binding module, observed in patient material, with no kinase assigned, is the most specific unexploited observation this beat found for MLLT10.
| status | run |
|---|---|
| gene | gene-mllt10 |
| canonical accession | P55197 |
| canonical length | 1068 |
| isoform used by source table | P55197-1 |
| isoform length | 1027 |
| identical through residue | 566 |
| offset in this region | canonical = isoform-1 minus 16 |
| retained in fusion | true |
| retained basis | both positions are C-terminal of the PZP module (22-198) and of the approximate exon-10 boundary (about aa 266), and N-terminal of the retained OM-LZ (704-779) |
| correction owed to | ana-phosphosites-observed-in-kmt2a-r-aml-patients |
| result summary | The source table numbers MLLT10 on isoform P55197-1. S674 maps to canonical S658 and S705 maps to canonical S689; S689 is independently reported by PeptideAtlas and UniProt. The site that PASSED the residue-type check was the one at the wrong canonical position. |
mapping
[
{
"table_value": "S674",
"isoform_window": "NSSMAALIAQSENNQTDQDLG",
"canonical_position": 658,
"canonical_residue": "S",
"canonical_residue_at_table_position": "L",
"corroborated_by_atlas": false
},
{
"table_value": "S705",
"isoform_window": "SSPRGSLSPRSPVSSLQIRYD",
"canonical_position": 689,
"canonical_residue": "S",
"canonical_residue_at_table_position": "S",
"canonical_705_window": "QIRYDQPGNSSLENLPPVAASI",
"corroborated_by_atlas": true
}
]- A numbering offset that lands on the correct residue type is invisible to the check this project currently runs, and the corrected number is what any reagent - a phospho-specific antibody, a synthetic peptide, a knock-in - would be built against. This correction is worth more than the site it corrects.
- Nothing here is an intervention, but the site's proximity to the DOT1L-binding module means any future phospho-directed molecule would sit beside the interface that DOT1L inhibitors already target, and would inherit that programme's on-target haematopoietic toxicity.
- Does the same isoform offset affect the other proteins in that supplementary table? MEN1 S538 was separately flagged as landing on a proline, which suggests a second non-canonical isoform somewhere in the same file.
- What kinase phosphorylates MLLT10 S689, 15 residues N-terminal of the DOT1L-binding OM-LZ? No kinase is assigned to any MLLT10 site anywhere.
- Does phosphorylation near the OM-LZ affect DOT1L coiled-coil binding? The co-crystals 6CKN/6CKO exist and a phosphopeptide competition experiment is cheap.
- Which sequence database did the source study search? Its methods name Mascot and Pescal but not the database, which is why this had to be resolved by window matching rather than by reading.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
UniProt isoform P55197-1 is 1027 residues and the canonical displayed P55197 sequence is 1068; the two are identical through residue 566 and diverge thereafter, and 21-residue windows centred on isoform-1 positions 674 and 705 match canonical positions 658 and 689 exactly, both serines.
UniProtKB P55197 with includeIsoform=true: canonical 1068 aa, isoform P55197-1 1027 aa, isoform-2 126 aa, isoform-3 179 aa; sequence windows compared 2026-09-19 · UniProt Consortium 2026What this evidence cannot show
This establishes that the table's numbers are consistent with P55197-1 and inconsistent with the canonical sequence at position 674. It does not prove the study's search database was P55197-1 specifically - any database using that isoform's numbering would give the same result - because the primary paper's methods section names Mascot and Pescal but does not state the protein sequence database or version, as checked in its full text in this session.
Evidence that argues against this (1)
- Two data points establishing an offset is thin. A third MLLT10 site in the same table would test it, and there is none, so the inference rests on one failing site plus one coincidentally-passing site. The study's own methods do not name the sequence database, so the isoform attribution is an inference from the numbers rather than a statement of what was searched. Casado P et al. 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
Canonical P55197 S689 is reported as a phosphoserine by both the PeptideAtlas/PTMeXchange re-analysis (which lists 679, 680, 684, 686 and 689 in that cluster) and UniProt's curated Modified-residue table (which lists 684, 686 and 689), while canonical S705 is reported by neither.
Proteins API proteomics-ptm P55197 (20 phosphosites) and UniProtKB P55197 Modified residue table (7 phosphosites) · EMBL-EBI 2026What this evidence cannot show
Corroboration at one of the two mapped positions. S658 is not in either resource, so only half the mapping has independent support; the other half rests on the window match alone. Absence of S705 from the resources is weak evidence given how sparse both are.
Evidence that argues against this (1)
- A cluster of five adjacent reported sites at 679-689 is itself a region where phosphosite localisation is unreliable, so finding a match inside it is a weaker corroboration than finding one at an isolated site. The window mapping remains the primary evidence. UniProt Consortium 2026
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In KMT2A::MLLT10 the discarded AF10 region is the N-terminal PZP module (aa 22-198) and adjoining sequence, with the first retained MLLT10 exon recorded as exon 10 (approximately residue 266), while the OM-LZ at aa 704-779 that binds the DOT1L coiled coil is retained.
The role of the PZP domain of AF10 in acute leukemia driven by AF10 translocations. · Klein BJ, et al. · Nat Commun 2021What this evidence cannot show
Domain coordinates are taken from this graph's protein-mllt10-domains entry, which derives the PZP boundary from crystallographic constructs and the OM-LZ boundary from the DOT1L co-crystals 6CKN/6CKO. The exon-10 call is explicitly recorded there as approximate, and MLLT10 rearrangements are structurally heterogeneous and frequently cryptic, so a single exon pair does not describe all cases. The PZP add-back experiment that establishes the loss as causal was done in CALM-AF10, not KMT2A::MLLT10.
Evidence that argues against this (1)
- Because the breakpoint is recorded as approximate, four other observed MLLT10 phosphosites (S215, S217, S219, S252) fall in a zone whose fate genuinely cannot be called. The same uncertainty that leaves those undetermined is what makes the retention call for S658 and S689 safe only by margin, not by measurement. Chen BR et al. 2021
independently reproduces the -16 offset and confirms canonical S658/S689
patient phosphosites remapped to canonical numbering
- related to →KMT2A::MLLT10 (MLL-AF10), t(10;11)(p12;q23)
S658 and S689 retained, S689 sits 15 aa before the OM-LZ
positions placed against the PZP and OM-LZ coordinates
corrects MLLT10 S674/S705 to canonical S658/S689
- ← related toAF10 fusions recruit and stabilise JAK1 - a kinase bound to the fusion with no phosphosite on the fusion
a kinase contact with no mapped site, beside two sites with no kinase
correction owed outside this folder