Oncofusion

start / graph / ana-phosphosites-observed-in-kmt2a-r-aml-patients

Analysisunrated confidence 0 of 5: unrated

Which of our phospho-sites are actually seen in patients

also called patient phosphosites · MOESM3 phosphosite table · KMT2A-r AML phosphoproteome sites

Most of what this project knows about phosphorylation of these proteins comes from cell lines, but a study of 74 patients measured it directly and left the table in a supplementary file. Some of our sites are there, one we had ruled out is still absent, and two are numbered on the wrong version of the protein.

The same thing, for a scientist

The supplementary phosphosite matrix of the 74-patient KMT2A-rearranged AML phosphoproteomics study, parsed directly. **This is the only disease-specific, patient-derived phosphosite evidence this graph has** — everything else in the phospho beat comes from cell lines or aggregated atlases. It confirms three sites this project relies on, kills one it had already doubted, and supplies the 3' partner sites a concurrent beat was searching for.

Go deeper
What is observed

Sites recovered from the table, by protein:

proteinn sitesresidues
KMT2A (MLL)11S136, S153, S504, S516, S1059, S1837, T1839, T1845, S2098, T2857, S3036
PSIP1 (LEDGF)12S102, S105, S106, T115, T122, S129, S208, T267, T270, S271, T272, S273
AFF47S178, S549, S703, S706, S1048, Y1069, S1077
MLLT1 (ENL)4S359, S361, S366, S444
AFF13S199, S216, S588
MLLT3 (AF9)3S412, S419, S429
MLLT10 (AF10)2S674, S705
ELL2S308, S437
MEN11S538
DOT1L1S374
The three things it settles

1. KMT2A S136 is phosphorylated in patients. S136 is one of the two CK2 acceptors in the LEDGF/p75 IBM affinity switch — the second handle on the retained fragment that cost this project the claim that menin was very nearly the only one. That switch was established in cell lines and primary cells; this is the site observed in a 74-patient KMT2A-rearranged AML cohort. S142, its partner acceptor, is not in this table, which is a limit of coverage rather than a negative.

2. S821 is dead in the right disease. The 14-3-3 motif scan produced exactly one candidate on the retained fragment, S821, and the aggregated PRIDE atlas did not contain it. That atlas is not leukaemia-specific and the caveat said so. This table contains no KMT2A site anywhere in the 790–860 window. The candidate is now negative in patient material from the actual disease, which is the test it was missing.

3. KMT2A S516 and AFF1 S588 are both present. S516 is the ATR site; S588 is the residue through which 14-3-3θ binds AFF1, and the basis of this graph's hypothesis that KMT2A's reproduced 14-3-3 association is bridged by the AFF family rather than direct. Both are observed in patients, which raises that hypothesis from plausible to testable.

And what it hands to the beat running now

The concurrent partner-phospho beat was dispatched to map kinase-gated contacts on MLLT1, MLLT3, MLLT10 and ELL, because the previous phospho beat covered only KMT2A, AFF1 and AFF4 and said so. This table supplies patient-observed sites for all four: MLLT1 S359, S361, S366, S444; MLLT3 S412, S419, S429; MLLT10 S674, S705; ELL S308, S437.

Those are observations, not mechanisms — no kinase is assigned to any of them here, and whether each lies in the portion of the partner the fusion retains is exactly the question that beat exists to answer.

PSIP1/LEDGF has twelve observed sites, more than any other protein in this list. The IBM switch paper reports phosphorylation regulating the partner side of the interface; LEDGF's own phosphorylation state in this disease is an axis nobody here has looked at.

Residue numbering checked, and two sites do not survive it

This entry's own second open question was whether the table's numbering matches UniProt, since it depends on whichever isoform the study's search database used. Checked directly against the canonical sequences on 2026-09-20 — is the residue at that position actually a serine or threonine?

Eighteen of twenty confirm. KMT2A S136, S142, S153, S504, S516, S1059 and S2098 are all serines in Q03164. AFF1 S199/S216/S588 in P51825, MLLT1 S359/S361/S366/S444 in Q03111, MLLT3 S412/S419/S429 in P42568, ELL S308/S437 in P55199 and MLLT10 S705 — all serines.

That last one is wrong, and the check that passed it is why. MLLT10 S705 sits on the same non-canonical isoform as S674 and maps to canonical S689; canonical 705 is a serine in an unrelated context. A residue-type test passes a wrong position about one time in nine, so eighteen passes are not eighteen confirmations - see ana-the-numbering-check-that-passed-was-the-wrong-check.

Two do not.

reported sitecanonical residueverdict
MLLT10 S674 (P55197)L — leucineresolved: isoform offset −16, true site canonical S658
MEN1 S538 (O00255)P — prolineresolved: numbered on O00255-1, true site canonical S533 (offset −5)

Both proteins have isoforms, so the likeliest explanation is that the study's search database used a different one rather than that the sites are fabricated. But neither can be used at face value against canonical numbering, and anything built on MLLT10 S674 or MEN1 S538 without resolving the isoform is built on a residue that is not there.

The headline survives. KMT2A S136 is a serine at position 136 of Q03164, so the site this entry reports as the patient-observed CK2 acceptor is the same residue the IBM affinity-switch work identifies. That was the claim most at risk from a numbering mismatch and it holds.

The check took one API call and found what looked like a 10% error rate in data I had already sent to another beat. Writing the caveat was not a substitute for running it - and running the weak version of it was not a substitute for running the strong one. Both offsets were settled afterwards by peptide-window matching, which is unambiguous where a residue-type test is not.

Still open
  1. Resolve the isoform for MLLT10 S674 and MEN1 S538. Both positions are non-phosphorylatable on the canonical sequence, so the study used a different isoform or the sites are wrong, and neither is usable until that is settled.
  2. Get the per-patient values rather than the site list. The table is a matrix across 74 patients; this analysis read only its row labels, so which patients carry S136 and whether it tracks the MLLGA/MLLGB split is unread and would be a genuine finding.
  3. The cohort is AML and the index disease is infant B-ALL. Is there an equivalent phosphoproteomics dataset in KMT2A-rearranged ALL, and if not, that is a bounded absence worth recording.
  4. PSIP1/LEDGF carries twelve observed sites, more than anything else here. Its own phosphorylation state in this disease has never been looked at by this project, and the IBM switch makes it immediately relevant.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The supplementary phosphosite matrix of a 74-patient KMT2A-rearranged AML phosphoproteomics study contains 11 KMT2A sites (S136, S153, S504, S516, S1059, S1837, T1839, T1845, S2098, T2857, S3036), 12 PSIP1 sites, 7 AFF4, 4 MLLT1 (S359, S361, S366, S444), 3 AFF1 (S199, S216, S588), 3 MLLT3 (S412, S419, S429), 2 MLLT10 (S674, S705), 2 ELL (S308, S437), MEN1 S538 and DOT1L S374.

    What this evidence cannot show

    Parsed by the coordinator on 2026-09-20 from the shared-strings table of `41392_2022_1288_MOESM3_ESM.xlsx`, retrieved through the Europe PMC supplementaryFiles endpoint for PMC9968719. **Sites were read from the table's own site-label strings** (of the form `MLL(S136);`) rather than from per-patient intensity values, so this records which sites the study detected and quantified **somewhere in the cohort**, not in how many patients, at what stoichiometry, or with what localisation probability. The cohort is **AML**, and this project's index disease is infant **B-ALL**.

    Evidence that argues against this (2)
    • Detection in a phosphoproteomics matrix is coverage-limited in both directions. Absence of S142 next to a present S136 is far more likely to reflect which tryptic peptide was observed than a real difference in phosphorylation, so no site absent from this table can be called unphosphorylated on that basis alone - which limits how hard the S821 result below can be pushed. EMBL-EBI / PRIDE 2026
    • The gene labels in this table are the study's own, and 'MLL' is used where modern annotation would say KMT2A while 'MLL2' appears for KMT2D. Residue numbering depends on which isoform the search database used, and that was not checked against UniProt Q03164 here. A site number from a supplementary table is not the same object as a site number from a curated database. EMBL-EBI / Signal Transduction and Targeted Therapy 2023
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    KMT2A S136 - one of the two CK2 acceptors in the LEDGF/p75 IBM affinity switch, the phospho-gated contact retained by every KMT2A fusion - is observed in this 74-patient KMT2A-rearranged AML cohort. KMT2A S516, the ATR site, and AFF1 S588, the 14-3-3theta binding residue, are also observed.

    What this evidence cannot show

    Observation in a patient cohort establishes that the site is phosphorylated in the disease, and nothing about function, stoichiometry or which kinase is responsible in that context. The IBM switch's functional evidence remains the cell-line and primary-cell work of PMID 29997176; this adds in-patient presence to it and no more.

    Evidence that argues against this (1)
    • A site being detectable in patient tumour lysate is weak evidence that it is doing the thing the mechanism requires. The IBM switch works by MODULATING affinity between low and high states, so what matters is the fraction of KMT2A phosphorylated at S136 and how that fraction differs between conditions - none of which a presence/absence table reports. Proceedings of the National Academy of Sciences 2018
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    No KMT2A phosphosite anywhere in residues 790-860 appears in this patient cohort, so S821 - the single retained-fragment 14-3-3 motif candidate produced by this project's motif scan - is absent from disease-specific patient material as well as from the aggregated PRIDE atlas.

    What this evidence cannot show

    This is the disease-specific test the earlier PRIDE result explicitly lacked, and it agrees. But it is still an absence in one cohort with unknown peptide coverage over that region, and the same table is missing S142, a site with direct functional evidence. **Absence here bounds the claim; it does not close it.**

    Evidence that argues against this (1)
    • The region may simply not be covered. Phosphoproteomics depends on generating and detecting a specific tryptic peptide, and a stretch with unfavourable cleavage sites can be invisible across an entire study regardless of biology. Without the study's peptide-level coverage map - which was not retrieved - a regional absence cannot be distinguished from a regional blind spot. EMBL-EBI / PRIDE 2026
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Checking every reported site against the canonical UniProt sequence, 18 of 20 are a serine or threonine at the stated position and 2 are not: MLLT10 residue 674 in P55197 is leucine and MEN1 residue 538 in O00255 is proline. KMT2A S136, the patient-observed CK2 acceptor this entry's headline rests on, is confirmed as a serine at position 136 of Q03164.

    What this evidence cannot show

    Checked by the coordinator on 2026-09-20 against the canonical sequence of each accession. **A matching residue type is necessary, not sufficient** — serines are common, so a coincidental match at the wrong position is entirely possible and this check cannot detect it. Both mismatching proteins have isoforms, so the likely explanation is isoform choice in the study's search database rather than error.

    Evidence that argues against this (2)
    • Residue-type matching is a weak test and its pass rate flatters the data. Roughly one residue in eight is a serine or threonine, so 18 of 20 matching is only a few times better than chance for a list of that length - it rules out gross frame errors and nothing subtler. UniProt Consortium 2026
    • Two failures out of twenty is also a caution about the sites that passed. If the study used a non-canonical isoform for MLLT10 and MEN1, it may have used one for other proteins too, and those numbering offsets would be invisible wherever the shifted position happens to land on another serine. UniProt Consortium 2026
Connected to