Oncofusion

start / graph / ana-partner-phospho-enumeration

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q25-partner-phospho

Q25 enumeration: what was searched, found, rejected and could not be read across the five 3' partners

also called partner phospho enumeration · Q25 ledger · 3-prime partner kinase enumeration

The audit trail for this beat: which searches were run, what came back, which papers were set aside and why, which papers exist and could not be read, and what would change any of those decisions. It is here so that someone repeating this work next year repeats the decisions and not just the searches.

The same thing, for a scientist

Beat Q25 covered kinase-gated contacts on the KMT2A 3' partners MLLT1, MLLT3, MLLT10, ELL and ELL2, plus EAF1/EAF2. **Found: 5 kinase-site relations** across 3 partners (ATM-ENL at conserved SQ sites; DNA-PKcs-MLLT3 S395; ATM-ELL T551, S561 and S589 - counted as three sites in one relation), plus **2 kinase-gated non-phospho switches** (PCAF/HDAC5-CaMKII on AF9 K339; p300/HDAC1 with AKT and ATM inputs on ELL K355) and **1 kinase-partner contact with no mapped site** (JAK1 recruited by AF10 fusions). **Retained/lost: 4 of 5 phospho relations are on retained sequence; the AF9 acetylation arm is the only one that is lost.** **Counted absences: 5**, the sharpest being zero records linking ENL condensates or YEATS binding to phosphorylation and zero phosphosites mapped on any KMT2A fusion protein. **Bounded rejections: 11.** **Unreadable after all four routes: 3.** **Corrections owed outside this folder: 2**, both isoform-numbering resolutions.

Go deeper
Found - the kinase-site ledger
#kinasesubstratesite(s)consequencemapped inretained?
1ATM (ATR/DNA-PKcs equally able)MLLT1 / ENLconserved SQ sites; residues not in the readable record; sequence candidates S394, S398switches elongation to PRC1/BMI1 repression, H2A ubiquitinationwild-type ENLyes (~84% break at or before aa 5)
2DNA-PKcsMLLT3 / AF9S395 - the only S/T-Q motif in AF9loses SEC binding (CDK9, ELL) and TFIID bindingwild-type AF9, HEK293Tyes (~92% break at or after aa 376)
3ATMELLT551, S561, S589increased EAF1 binding, self-association, leaves SECwild-type ELL, human linesyes (MLL-ELL = ELL 46-621)
4ATM (via Sin3A-HDAC1) and AKT (via p300)ELLK355 acetylation, kinase-gated both wayspartition between SEC and LECwild-type ELLyes
5CaMKII (via HDAC5 S259/S498 export)MLLT3 / AF9K339 acetylationoligomerisation and TFIID bindingwild-type AF9, HEK293TNO - lost
-JAK1MLLT10 / AF10none mappedfusion recruits and stabilises the kinase; direction is fusion-to-kinasemouse and human AF10-fusion AMLinterface region not mapped

Nothing was mapped in a fusion protein. Every row above is wild-type protein, and four of the six are from two laboratories.

Unassigned observed sites, retained, with no kinase: MLLT1 23; MLLT3 3 (S412, S419, S483); MLLT10 16 firmly plus 4 undetermined; ELL 12. That is 58 retained observed phosphosites across the four partners with no kinase attached, against the previous beat's 51 for KMT2A.

Counted absences (5)
  1. (ENL OR MLLT1 OR YEATS1 OR LTG19)[tiab] AND (condensate/phase separation/droplet)[tiab] AND (phosphorylation/kinase/phosphosite)[tiab] -> 0.
  2. ("fusion protein")[tiab] AND (KMT2A OR MLL)[tiab] AND ("phosphorylation site" OR phosphosite OR "phospho-specific")[tiab] -> 0; and the phospho-specific-antibody variant -> 0.
  3. (MLLT10 OR AF10 OR "AF-10" OR ZFP-1)[tiab] AND (phospho* OR kinase)[tiab] -> 25 raw, 0 on topic after title check.
  4. (MLLT1 OR MLLT3 OR MLLT10 OR ELL OR ELL2 OR AF9 OR AF10)[tiab] AND (CK2 OR CSNK2 OR "casein kinase 2")[tiab] -> 1, off topic. No CK2 relation exists for any 3' partner, which is worth stating because CK2 is the previous beat's central kinase on the KMT2A side.
  5. MLLT1[tiab] AND phosphosite[tiab] -> 0; (MLLT3 OR AF9)[tiab] AND (phosphosite OR "phosphorylation site")[tiab] -> 0; EAF1[tiab] AND (phosphosite OR "phosphorylation site")[tiab] -> 0.

Aliases searched in both directions throughout: MLLT1/ENL/YEATS1/LTG19, MLLT3/AF9/AF-9, MLLT10/AF10/AF-10/ZFP-1, ELL/ELL1, ELL2, EAF1, EAF2. Alias precision was measured rather than assumed - see ana-enl-token-collision-alias-precision.

Bounded rejections (11) - found, looked on-beat, not used

Each line: identifier, one sentence of reason, and what would change the decision.

  1. PMID 41476161 (Nat Commun 2025, P70S6K phosphorylation of AFF4 in aged insulin-resistant bone). A real kinase-site relation on a SEC subunit, but AFF4 is the previous beat's territory and the system is aged murine bone. Would change if: the site were shown on AFF4 in a leukaemic context, in which case it belongs beside mechanism-cdk9-phosphorylates-aff4-sec-assembly.
  2. PMID 42748909 (Mol Cell 2026, a dual-module ECAS complex coupling elongation and RNA processing). Names ELL-family biology and is recent and relevant, but reports no phosphorylation and no KMT2A partner site. Would change if: the complex's assembly turns out to be kinase-gated.
  3. PMID 42285105 / 40766627 (Mol Cell 2026 and its preprint, RNA reinforces condensate nucleation on chromatin to amplify oncogenic transcription). Squarely on ENL condensates and squarely on-beat by title. Read as far as the abstract; it is an RNA-dependence mechanism with no phosphorylation component. Would change if: an RNA-binding surface on ENL turns out to be phospho-regulated.
  4. PMID 38655899 (Cancer Discov 2024, condensate-promoting ENL mutation drives tumorigenesis in vivo). The in vivo counterpart of the ENL condensate story; no phosphorylation. Would change if: the dynamic histone-modification regulation it reports is traced to a kinase.
  5. PMID 39327672 (Mol Oncol 2025 commentary on the above). Secondary literature; nothing the primary sources do not carry.
  6. PMID 38766219 (bioRxiv, single-cell multiomics of ENL mutation in kidney development). Preprint, kidney, no phospho. Would change if: published with a kinase axis.
  7. PMID 42523331 (bioRxiv 2026, disrupting a convergent acetylation circuit collapses leukaemic identity across AML subtypes). On-beat by title for the acetylation arm; a preprint, not read beyond the title in this session. Would change if: peer-reviewed and found to include a KMT2A partner residue. This is the item in this list most likely to matter.
  8. PMID 24127601 (PNAS 2013, ELL as a novel TFIIH partner in transcription restart after DNA repair). Directly adjacent to the ELL damage-response mechanism and cited nowhere in this beat because it reports no phosphorylation. Would change if: the TFIIH contact is shown to be phospho-gated, which the 2026 ELL condensate paper's DNA-PKcs/Ku result makes plausible. 8b. PMID 22547686 (MCB 2012, SEC family gene-target specificity). The canonical SEC-family paper; no phosphorylation of any partner.
  9. PMID 38036714 (J Assoc Res Otolaryngol 2023, GWAS of chronic dizziness implicating MLLT10). Retrieved by the MLLT10 phospho query; a locus association with no protein-level content. Would change if: a coding variant were functionally characterised.
  10. PMID 23419774 (BBA 2013, CATS/FAM64A is a substrate of Kinase Interacting Stathmin). Retrieved by the AF10 query because CATS is a CALM-AF10 interactor; the kinase acts on CATS, not on AF10. Would change if: KIS were shown to phosphorylate AF10 itself.
  11. PMID 34235154 (Front Cell Dev Biol 2021, profilin-1 Ser71). Retrieved repeatedly by ENL and SEC queries as pure token noise; unrelated.

Laboratories considered and not cited: the Wan laboratory's ENL condensate programme is cited for its methodological content (expression-dependence, FRAP) but none of its biological findings are used, because none involves phosphorylation. That is a deliberate decision and a reader should know it, since that group is the most active in ENL biology.

Unreadable after all four routes (3)

Each is on topic, is identified, and is not paraphrased beyond its abstract anywhere in this beat.

PMIDwhat the title assertsroutes tried
25921070ENL phosphorylated by ATM recruits polycomb and switches off transcription for DSB repair (Mol Cell 2015)EPMC core -> no pmcid; NCBI PMC ID converter -> 'Identifier not found in PMC'; elink dbfrom=pubmed db=pmc -> no uid; fullTextXML / efetch db=pmc / PMC HTML all unreachable without a PMCID
41424126Enhanced ELL phase separation is crucial for efficient DNA damage repair to restart transcription and cell survival (Mol Cell Biol 2026)same four; no PMCID
41843233Phosphorylation-dependent regulation of ELL uncovered by quantitative global phosphoproteomic analysis (Mol Genet Genomics 2026)same four; no PMCID

All three are marked weak or moderate where cited and their residue-level content is explicitly not asserted. The first is the costly one: it is the founding paper for the ENL/ATM switch, it names the SQ residues, and this beat has had to substitute a sequence inference for a reading.

Two papers WERE recovered by the later routes and should be recorded as successes of the four-route rule: PMC4916901 (the ENL/ATM review) returned HTTP 500 from fullTextXML and a body-less document from efetch db=pmc, and was read only from the PMC HTML page (39,741 characters); PMC10071517 (ENL YEATS hotspot condensates) returned HTTP 500 from fullTextXML and was recovered by efetch db=pmc (68,960 characters). Without routes three and four, the ENL half of this beat would have had no readable source at all.

Corrections owed outside this folder (2)

Both concern data/graph/nodes/phospho-contacts/ana-phosphosites-observed-in-kmt2a-r-aml-patients and both are isoform-numbering resolutions rather than disputes.

1. MLLT10. The table's S674 and S705 are numbered on UniProt isoform P55197-1 (1027 aa), not the canonical 1068-aa sequence. Corrected values: S674 = canonical S658, S705 = canonical S689. Twenty-one-residue windows match exactly, and canonical S689 is independently reported as a phosphosite by both PeptideAtlas and UniProt. Note that S705 passed the residue-type check and is still at the wrong canonical position, because canonical 705 happens also to be a serine in an entirely different context (QIRYDQPGNSSLENLPPVAASI versus SSPRGSLSPRSPVSSLQIRYD). Full working in ana-mllt10-patient-phosphosite-isoform-resolution.

2. MEN1. Outside this beat and offered rather than claimed. The same table's MEN1 S538 was flagged because canonical O00255 residue 538 is a proline. It resolves the same way: isoform O00255-1 (615 aa) has a serine at 538, and the 22-residue window AGTARGPEGGSTAQVPAPTASP maps it to canonical S533. So the file carries at least two non-canonical isoform numberings and the remaining entries in it should be re-checked by window matching rather than by residue type. Verified by fetching O00255 with includeIsoform=true on 2026-09-19; not written to that entry, which this beat does not own.

A premise in the dispatch that the literature partly overturned

The dispatch framed the beat around the idea that a phospho-switch on a 3' partner is on retained sequence by definition, because the 3' partner is the half the fusion keeps from the partner.

That is right for MLLT1 and ELL and wrong as a general rule. MLLT3 discards 63% of itself, six of its nine observed phosphosites and three of its four characterised regulatory elements. MLLT10 discards its entire N-terminal reader module. 'The half the fusion keeps' is a large fraction for some partners and a small one for others, and for AF9 the retained fraction is residues 376-568 out of 568. The retention question had to be answered per site, which is what this beat did, rather than assumed.

Facts and figures
statusrun
beatQ25 phospho-dependent contacts on the 3' partners
kinase site relations found5
kinase gated non phospho switches2
kinase contacts without a mapped site1
relations on retained sequence4
relations on lost sequence1
relations mapped in a fusion protein0
retained observed phosphosites without a kinase58
counted absences5
bounded rejections11
unreadable after four routes3
dispatch premise partly overturnedThe dispatch held that a phospho-switch on a 3' partner is on retained sequence by definition. True for MLLT1 and ELL; false for MLLT3, which discards 63% of itself including six of nine observed phosphosites and three of four regulatory elements, and for MLLT10, which discards its N-terminal reader module.
result summaryFive kinase-gated relations across three of five partners, four on retained sequence and none mapped in a fusion. The headline is AF9 S395: the only one of AF9's four regulatory elements the fusion keeps, and it governs SEC binding. MLLT10 has zero relations against twelve PIKK motifs. Fifty-eight retained observed phosphosites have no kinase.
partners covered
[
  "gene-mllt1",
  "gene-mllt3",
  "gene-mllt10",
  "gene-ell",
  "gene-ell2",
  "gene-eaf1",
  "gene-eaf2"
]
papers recovered only by route 3 or 4
[
  "PMC4916901 (route 4, HTML, 39741 chars)",
  "PMC10071517 (route 3, efetch db=pmc, 68960 chars)"
]
corrections owed outside this folder
[
  {
    "entry": "ana-phosphosites-observed-in-kmt2a-r-aml-patients",
    "field": "MLLT10 sites S674 and S705",
    "should_say": "canonical S658 and S689; the table numbers MLLT10 on isoform P55197-1",
    "citation": "UniProtKB P55197 includeIsoform=true, window matching, 2026-09-19",
    "read_status": "resolved in full by this beat; see ana-mllt10-patient-phosphosite-isoform-resolution"
  },
  {
    "entry": "ana-phosphosites-observed-in-kmt2a-r-aml-patients",
    "field": "MEN1 site S538",
    "should_say": "canonical S533; the table numbers MEN1 on isoform O00255-1 (615 aa)",
    "citation": "UniProtKB O00255 includeIsoform=true, window AGTARGPEGGSTAQVPAPTASP matched to canonical 533, 2026-09-19",
    "read_status": "verified by direct fetch; outside this beat, offered not applied"
  }
]
resources unreachable this session
[
  "PhosphoSitePlus (not attempted; previous beat recorded HTTP 403)",
  "iPTMnet (not attempted; previous beat recorded HTTP 503)"
]
How this could bite us back
  • The enumeration's own bias is worth stating: four of the six relations found come from two laboratories, and a beat that reports a convergent pattern assembled largely from two groups' output is reporting those groups' interests as well as the biology. The MLLT10 absence is at least partly an absence of anyone having chosen to look.
Still open
  1. PMID 25921070 names the ENL SQ residues and cannot be read. An interlibrary request or an email to the corresponding author would convert this beat's largest inference into a fact for the cost of one message.
  2. PMID 42523331, the bioRxiv acetylation-circuit preprint, is the rejection most likely to matter and was judged on its title alone. It should be read next cycle.
  3. No CK2 relation exists for any 3' partner, while CK2 is the previous beat's central kinase on the KMT2A side. Is that a real asymmetry or an artefact of who has studied what?
  4. Fifty-eight retained observed phosphosites across four partners have no kinase. A single in vitro kinase-panel screen against synthetic peptides for the best-observed ten would be the highest-yield next experiment in this beat after the fusion immunoprecipitation.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Five kinase-gated relations were found across the 3' partners: ATM on ENL at conserved SQ sites, DNA-PKcs on AF9 S395, ATM on ELL at T551/S561/S589, kinase-gated acetylation of ELL K355 by AKT/p300 and ATM/Sin3A-HDAC1, and kinase-gated acetylation of AF9 K339 via CaMKII-controlled HDAC5 export; four are on retained sequence and the AF9 K339 arm is lost.

    PMID 38750015DOI
    What this evidence cannot show

    This is a summary of the beat's own entries, each of which carries its own evidence and caveats; the cited paper supplies two of the five relations. Four of the six relations in the full table come from two laboratories, so the apparent convergence is less independent than a raw paper count suggests. No relation was mapped in a fusion protein.

    Evidence that argues against this (1)
    • The one partner with no kinase-site relation, MLLT10, has twelve S/T-Q motifs and twenty observed phosphosites, so the pattern the other four partners show may reflect which proteins two laboratories chose to study rather than a property of the partner family. Chen BR et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Phosphosite counts from the EMBL-EBI proteomics-ptm endpoint exceed UniProt's curated counts by 1.4x to 6.0x across the seven proteins examined, and the same arithmetic reproduces the previously published KMT2A figure of 130 against UniProt's 25.

    What this evidence cannot show

    Detailed in ana-partner-phosphosite-atlas-vs-uniprot. The two resources do not nest in either direction, and neither is a census.

    Evidence that argues against this (1)
    • Site counts measure sampling depth, not biology, and the beat's own headline site (AF9 S395) appears in neither resource - so the census that motivated this beat is systematically blind to the class of site the beat actually found. Casado P et al. 2023
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Three on-topic papers could not be read after all four routes of the project's unreadable test, all three because no PMCID exists; two other papers were recovered only by route three or route four after fullTextXML returned HTTP 500.

    What this evidence cannot show

    Unreadability is a statement about this session on this date. All three papers may be obtainable by institutional subscription, interlibrary loan or an email to the authors, and PMID 25921070 in particular is worth that effort because it names residues this beat had to infer.

    Evidence that argues against this (1)
    • The four-route rule earned its keep here rather than merely being followed: PMC4916901 returned HTTP 500 from fullTextXML and a body-less document from efetch db=pmc, and yielded 39,741 characters from the HTML page. A three-route rule would have logged it unreadable and left the ENL arm of this beat with no readable source. Ui A et al. 2016
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The KMT2A-rearranged AML patient phosphosite table numbers at least two proteins on non-canonical UniProt isoforms: MLLT10 on P55197-1, where its S674 and S705 map to canonical S658 and S689, and MEN1 on O00255-1, where its S538 maps to canonical S533.

    What this evidence cannot show

    The window matches are exact and unambiguous, but they establish that the numbers are CONSISTENT with those isoforms, not that the study searched them - the primary paper's methods name Mascot and Pescal without naming a sequence database, as checked in its full text. The MEN1 resolution is offered to the owner of that entry and has not been applied.

    Evidence that argues against this (1)
    • Two proteins resolving to two different non-canonical isoforms is more consistent with a mixed or older database build than with a single isoform choice, in which case other entries in the same table may be canonical and a blanket re-numbering would introduce errors. Each site needs its own window check. Casado P et al. 2023
Connected to