Oncofusion

start / graph / ana-aff1-has-one-observed-binary-interaction-and-forty-two-residues-of-structure

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q76-spoke-provenance

The index fusion's 3' partner has ONE observed binary interaction in IntAct and forty-two residues of solved structure - every SEC contact attributed to AFF1 is an AFF4 result wearing AFF1's name

also called AFF1 has no observed SEC contact · the SEC is an AFF4 structure · 2LM0 is 42 residues

AF4 is the partner gene in the commonest infant leukaemia fusion, and everything this field says about how it assembles the elongation machine comes from experiments where a whole complex was pulled out at once. Of its forty-three recorded partners, exactly one has ever been shown to touch it one-on-one, and the only piece of AF4 anyone has ever solved a structure of is forty-two amino acids long.

The same thing, for a scientist

Of AFF1's 43 distinct partner pairs in IntAct, 33 (76.7%, Wilson 95% CI 62.3-86.8%) rest only on spoke-expanded rows, and exactly **one** partner - MLLT3/AF9 - has any MI:0407 direct-interaction record [E1]. Every canonical super-elongation-complex contact attributed to AFF1 is spoke-only with no shared structure: AFF1-AFF4 8 records all spoke, AFF1-CDK9 16 all spoke, AFF1-CCNT1 8 all spoke, AFF1-MLLT1/ENL 9 all spoke, AFF1-ELL2 2 all spoke [E1]. The records come overwhelmingly from two n-ary purifications, PMID 20153263 and PMID 21729782 [E2][E3]. AFF1 has **one PDB entry in existence**, 2LM0, an NMR structure of an AF4(738-779)/AF9 chimera - 42 residues, 3.5% of a 1,210-residue protein [E4][E5]. AFF4, by contrast, carries 9 PDB entries including 4IMY and 6CYT, in which AFF4 is crystallised directly against CDK9 and cyclin T1 [E4]. **So the SEC architecture this project reasons from is an AFF4 structure, transferred to AFF1 by paralogy**, and the transfer has never been tested binary. This is `concept-partner-import` at the level of the interaction rather than the fusion, and it is a sharper version of the same problem: the import here is not from another fusion but from another protein.

Go deeper
The measurement

From the IntAct REST record fetched 2026-09-21, restricted to pairs in which AFF1 (P51825) is one partner 1:

distinct partner pairs43
resting only on spoke-expanded rows33 (76.7%), Wilson 95% CI 62.3-86.8%
with at least one observed binary record8
partners with any MI:0407 direct-interaction record1 (MLLT3/AF9)

The canonical elongation contacts, one by one:

pairrecordsspoke-expandedstrongest non-spoke typeshared PDB
AFF1-AFF488nonenone
AFF1-CDK91616nonenone
AFF1-CCNT188nonenone
AFF1-MLLT1 (ENL)99nonenone
AFF1-ELL222nonenone
AFF1-MLLT3 (AF9)71direct interaction2LM0

For comparison, the same table for AFF4: AFF4-CDK9 13 records all spoke-expanded but sharing PDB 4IMY, 4OGR, 4OR5, 5L1Z and 6CYT; AFF4-CCNT1 8 records, one not spoke-expanded, same five structures; AFF4-ELL2 6 records all spoke-expanded but sharing 5JW9 14.

Forty-two residues, and they are a chimera

UniProt P51825 carries exactly one PDB cross-reference: 2LM0, chains A = residues 738-779 5. The entry is "Solution structure of the AF4-AF9 complex", solution NMR, and its single polymer entity is described as an "AF4/FMR2 family member 1 / Protein AF-9 chimera" mapping to both P42568 and P51825 4 - the AF4 fragment and AF9 were expressed as one tethered chain, which is standard practice for a low-affinity intrinsically disordered complex and is stated plainly in the source 6.

So the total solved structure of AFF1 is 42 of 1,210 residues, 3.5%, in a chimera, of the region that binds a protein that is not its fusion partner. Nothing is solved of the ALF domain, the pSER region, the AF4-homology/CHD region through which the AFF1-AFF4 heterodimer is proposed to form, or the transactivation domain retained in KMT2A::AFF1.

The contrast with AFF4 is the whole point. AFF4's nine entries include 4IMY, "The AFF4 scaffold binds human P-TEFb adjacent to HIV Tat", an x-ray structure whose three polymer entities are CDK9, cyclin T1 and AFF4 4. The SEC scaffold has been crystallised against its kinase. The scaffold that was crystallised is not the one in the fusion.

Why this is not the same claim as partner import, and is worse

concept-partner-import measures how much of this graph's mechanism was obtained in a different fusion - 54% of fusion-bearing load-bearing evidence, with 0 of 24 sampled resting on a dedicated human KMT2A::AFF1 experiment. That is an import across fusions.

This is an import across paralogues, inside a claim about the index fusion's own partner, and it is invisible to the fusion-provenance test because AFF1 and AFF4 are both AF4-family proteins and the source papers are about both at once. PMID 20153263 purifies a higher-order complex containing AF4 and ENL family proteins with P-TEFb 2; PMID 21729782 shows MED26 docking elongation factors 3. Both are excellent, focused papers - they are not atlases, and brief rule 92's warning about miscalling a focused purification a screen applies - and neither is a binary binding experiment, because neither set out to be.

What follows is narrow and should be stated narrowly. The SEC is real, AFF1 is in it, and the co-purifications are strong evidence of co-membership. What has never been observed is which subunit AFF1 touches. If AFF1 reaches P-TEFb only through AFF4, then in a cell where AFF4 is limiting, or where the fusion's retained AFF1 fragment cannot heterodimerise, the recruitment argument changes shape - and nothing in the curated record distinguishes that from direct AFF1-CDK9 contact.

What would settle it, cheaply

The reagents exist and the experiment is small. Recombinant AFF1 fragments against purified P-TEFb, with AFF4 present and absent, read out by a method that cannot be satisfied by a bridging subunit - ITC, SPR or a two-component pulldown from purified components. Written up as exp-aff1-sec-binary-ladder-against-purified-ptefb.

The negative result is as useful as the positive: if AFF1 does not bind P-TEFb without AFF4, the fusion's elongation phenotype becomes a claim about AFF4 availability, which is a druggable dependency of a different kind, and hyp-the-secs-contact-with-the-index-fusion-is-aff4-mediated-and-untested records that prediction and its falsifiers.

Facts and figures
scriptanalyses/spoke-expansion-provenance-census/run.py
statusrun
result summaryAFF1 has 43 IntAct partner pairs, 76.7% spoke-only, exactly one partner (MLLT3/AF9) with an MI:0407 direct-interaction record, and exactly one PDB entry in existence (2LM0, a 42-residue AF4/AF9 chimera). Every SEC contact attributed to AFF1 is spoke-only with no shared structure, while AFF4 is crystallised directly against CDK9 and cyclin T1.
sec contacts attributed to aff1
AFF1-ELL22.000AFF1-ELL2: 2AFF1-MLLT37.000AFF1-MLLT3: 7AFF1-AFF48.000AFF1-AFF4: 8AFF1-CCNT18.000AFF1-CCNT1: 8AFF1-MLLT19.000AFF1-MLLT1: 9AFF1-CDK916.0AFF1-CDK9: 16
inputs
[
  "IntAct REST ws/interaction/findInteractions/P51825 and Q9UHB7, 2026-09-21",
  "UniProt REST P51825 and Q9UHB7 PDB cross-references",
  "RCSB PDB entry and polymer-entity records for 2LM0, 4IMY, 6CYT"
]
aff1
{
  "uniprot": "P51825",
  "length_aa": 1210,
  "gene_symbol": "AFF1",
  "intact_pairs": 43,
  "pct_spoke_only": 76.7,
  "wilson_95": [
    62.3,
    86.8
  ],
  "partners_with_a_direct_interaction_record": [
    "MLLT3"
  ],
  "pdb_ids": [
    "2LM0"
  ],
  "solved_residues": "738-779 (42 of 1210, 3.5%), as a chimera with AF9"
}
aff4 comparator
{
  "uniprot": "Q9UHB7",
  "length_aa": 1163,
  "intact_pairs": 87,
  "pct_spoke_only": 75.9,
  "pdb_ids_count": 9,
  "structures_with_ptefb": [
    "4IMY",
    "4OGR",
    "4OR5",
    "5L1Z",
    "6CYT"
  ]
}
sec contacts attributed to aff1 — the numbers behind the chart
[
  {
    "pair": "AFF1-AFF4",
    "records": 8,
    "spoke": 8,
    "shared_pdb": []
  },
  {
    "pair": "AFF1-CDK9",
    "records": 16,
    "spoke": 16,
    "shared_pdb": []
  },
  {
    "pair": "AFF1-CCNT1",
    "records": 8,
    "spoke": 8,
    "shared_pdb": []
  },
  {
    "pair": "AFF1-MLLT1",
    "records": 9,
    "spoke": 9,
    "shared_pdb": []
  },
  {
    "pair": "AFF1-ELL2",
    "records": 2,
    "spoke": 2,
    "shared_pdb": []
  },
  {
    "pair": "AFF1-MLLT3",
    "records": 7,
    "spoke": 1,
    "shared_pdb": [
      "2LM0"
    ]
  }
]
How this could bite us back
  • If AFF1 reaches P-TEFb only through AFF4, then AFF4 becomes the node to drug and the fusion's own 3' partner becomes a passenger in the recruitment step - which would be convenient, and is exactly the kind of conclusion that should not be drawn from an absence of binary experiments. The honest position is that the topology is unknown, not that it is AFF4-mediated.
  • Reading this entry as 'the SEC evidence is weak' would be a misuse. Co-purification of AF4-family proteins with P-TEFb has been reproduced by several labs with different tags in different cells, and reproducibility across labs is a different and real kind of support that a directness grade does not see. What is missing is the topology, not the complex.
  • A programme that prioritises interactions by structural corroboration will systematically deprioritise AFF1, because AFF1 is largely intrinsically disordered and disordered proteins do not crystallise. The index fusion's own partner would drop to the bottom of the queue for a reason that has nothing to do with its importance.
Still open
  1. Does the AFF1 ALF/CHD region bind P-TEFb at all in the absence of AFF4? No experiment in the curated record separates the two, and the answer changes what the fusion's elongation phenotype is a phenotype of.
  2. Is the AFF1-AFF4 heterodimer obligate or exchangeable in a KMT2A::AFF1 cell? The fusion retains the AFF1 C-terminal region including the reported CHD, but the stoichiometry of fusion-AFF1 against endogenous AFF1 and AFF4 has never been measured in a t(4;11) line.
  3. Would AlphaFold-Multimer, run over AFF1 against each SEC subunit with the AFF4 equivalents as positive controls, discriminate between direct and AFF4-bridged contact well enough to prioritise which binary experiment to do first? The AFF4/P-TEFb structures give a calibration set the method can be scored against.
  4. AFF1's 43 partner pairs is the smallest of the six core proteins. How much of the 76.7% is under-observation and how much is under-study? A citation count per pair would separate them and is one query away.
Every source behind this entry (6)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    In the IntAct REST record fetched 2026-09-21, AFF1 (P51825) has 43 distinct partner pairs of which 33 (76.7%) rest only on MI:1060 spoke-expanded rows; MLLT3 is the only partner with any MI:0407 direct-interaction record (six records from PMID 23260655, detected by NMR, circular dichroism, molecular sieving and fluorescence spectroscopy, none spoke-expanded); and AFF1-AFF4 (8 records), AFF1-CDK9 (16), AFF1-CCNT1 (8), AFF1-MLLT1 (9) and AFF1-ELL2 (2) are spoke-expanded on every record.

    What this evidence cannot show

    Bounded by IntAct's curated record on one day, and IntAct does not index every co-IP in the literature - a binary AFF1 experiment shown in a supporting figure of a paper about something else is invisible here, which is brief rule 80's shape one database over. The same beat found that the PSICQUIC route used by this project's earlier interactome analyses reaches only 83.5% of AFF1's record, so this count is taken from the fuller REST route; a still fuller route would only add partners, and adding spoke-expanded partners would raise the 76.7%. Unit is a pair, not an experiment.

    Evidence that argues against this (1)
    • Low curation depth and low real connectivity are indistinguishable from a database count. AFF1 has 43 pairs against AFF4's 87 and KMT2A's 146, so AFF1 is under-studied as well as under-observed, and an unstudied protein will look badly evidenced whatever its true interactome. The comparison to AFF4 is informative precisely because both are AF4-family scaffolds studied by the same labs in the same papers - but AFF4 also carries the HIV Tat literature, which is a large body of structural work AFF1 has no equivalent of and never could. EMBL-EBI IntAct 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    PMID 20153263 (Yokoyama A, Lin M, Naresh A, Kitabayashi I, Cleary ML, Cancer Cell 2010, 'A higher-order complex containing AF4 and ENL family proteins with P-TEFb facilitates oncogenic and physiologic MLL-dependent transcription') is the source of the majority of AFF1's spoke-expanded SEC rows in IntAct.

    PMID 20153263DOI
    What this evidence cannot show

    Title, five-author list, journal, year and DOI copied out of a live NCBI efetch XML record fetched 2026-09-21 using the anchored PubmedData/ArticleIdList/ArticleId path (brief rule 81). This PMID is already cited in 31 entries of this graph (brief rule 82) and its conclusions about the AF4/ENL/P-TEFb complex are written in several of them; it is used here ONLY to identify the experimental source of the IntAct rows, and this entry makes no new claim about the paper's findings. It is a focused complex purification and NOT a screen - brief rule 92 names it explicitly as a case where a screen-like heuristic misfires - so nothing here should be read as grading it a low-quality source.

    Evidence that argues against this (1)
    • A focused purification by a lab that had already characterised the complex biochemically is much stronger evidence of composition than a spoke-expansion flag suggests, and the flag makes no distinction between this paper and a proteome-scale atlas. If the census is read as putting the two on a level, it understates this source badly. Yokoyama A et al. 2010
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    PMID 21729782 (Takahashi H, Parmely TJ, Sato S, et al., Conaway JW, Cell 2011, 'Human mediator subunit MED26 functions as a docking site for transcription elongation factors') supplies the remaining AFF1 and AFF4 elongation-factor spoke rows in IntAct, including AFF1-ELL2 and AFF4-EAF1.

    Human mediator subunit MED26 functions as a docking site for transcription elongation factors. · Takahashi H, Parmely TJ, Sato S, et al., Conaway JW · Cell 2011
    PMID 21729782DOI
    What this evidence cannot show

    Title, seventeen-author list, journal, year and DOI from a live NCBI efetch XML record fetched 2026-09-21 with the anchored article-id path. Cited once elsewhere in this graph (brief rule 82: low reuse risk). Again a focused purification, not a screen, and named as such in brief rule 92. The abstract was read; the identification of which IntAct rows derive from it comes from the publication identifier on each record, not from the paper's own text.

    Evidence that argues against this (1)
    • The MED26 paper's design deliberately captures elongation factors as a group, so its rows are informative about co-occupancy of a docking site and were never intended to assert pairwise contacts. Counting them as spoke-expanded is right and grading AFF1 down for them is arguably unfair to the paper, which answered a different and well-posed question. Takahashi H et al. 2011
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PDB 2LM0 'Solution structure of the AF4-AF9 complex' is a solution-NMR entry with a single polymer entity described as an 'AF4/FMR2 family member 1/Protein AF-9 chimera' mapping to P42568 and P51825; PDB 4IMY 'The AFF4 scaffold binds human P-TEFb adjacent to HIV Tat' is an x-ray entry whose three polymer entities are CDK9 (P50750), cyclin T1 (O60563) and AFF4 (Q9UHB7); PDB 6CYT adds HIV-1 Tat and TAR RNA to the same three.

    What this evidence cannot show

    Read through the RCSB data API entry and polymer_entity endpoints rather than from the HTML page, and every entity's UniProt mapping was fetched individually rather than inferred from the title, because a PDB title names what the depositors called it and not what is in the coordinates (brief: four ways the PDB will give a confident wrong answer). A chimeric construct is a legitimate and common way to solve a weak intrinsically disordered complex and is not a defect in 2LM0; it does mean the entry does not establish an affinity between separate chains, which the source paper measures separately.

    Evidence that argues against this (1)
    • Counting PDB entries is a poor proxy for how well a protein is structurally understood. AFF1 is largely intrinsically disordered, and for a disordered protein the absence of crystal structures is a property of the molecule rather than of the field's effort - NMR, SAXS, crosslinking-MS and AlphaFold-Multimer predictions can all constrain an interface without producing a PDB deposition, and none of them is counted here. UniProtKB 2026
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt P51825 (AFF1_HUMAN, primary gene name AFF1, 1,210 aa, Homo sapiens) carries exactly one PDB cross-reference, 2LM0, annotated chains A = 738-779, which is 42 residues or 3.5% of the protein.

    What this evidence cannot show

    The accession was resolved by reading the PRIMARY gene name back, not by trusting a symbol search, per this project's standing accession rule. A UniProt PDB cross-reference list is curated and can lag a new deposition by weeks; an entry deposited in the last month would be missing. The 3.5% figure treats the annotated chain range as the solved region, which is the cross-reference's own statement and not a measurement of ordered residues in the coordinates.

    Evidence that argues against this (1)
    • One PDB entry is not the same as one experiment's worth of structural knowledge. The AF4 ALF domain's interaction with SIAH1/2, the pSER region's phosphorylation-dependent contacts, and the CHD heterodimerisation have all been mapped by deletion and mutation even though none has been crystallised, and a structure-count framing erases that work. UniProtKB 2026
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    PMID 23260655 (Leach BI, Kuntimaddi A, Schmidt CR, Cierpicki T, Johnson SA, Bushweller JH, Structure 2013, 'Leukemia fusion target AF9 is an intrinsically disordered transcriptional regulator that recruits multiple partners via coupled folding and binding') is the source of both PDB 2LM0 and all six of AFF1's non-spoke-expanded IntAct records.

    PMID 23260655DOI
    What this evidence cannot show

    Title, six-author list, journal, year and DOI copied out of a live NCBI efetch XML record fetched 2026-09-21 using the anchored article-id path. The abstract was read in full; the mapping from this PMID to the six IntAct records and to 2LM0 comes from the publication identifiers on those records and from the PDB entry's own citation, not from the paper's text. Note the paper's own framing - AF9 is the intrinsically disordered partner being characterised, and AF4 is one of several ligands - so it is a study of AF9's promiscuity rather than a study of AF4.

    Evidence that argues against this (1)
    • That the single observed binary interaction of AFF1 comes from a paper about AF9 rather than about AF4 is itself the argument of this entry, and it cuts the other way too: the one well-characterised AFF1 contact exists because somebody was studying the other protein. It would be wrong to read it as evidence that the AF4-AF9 contact is especially important to AF4. Leach BI et al. 2013
Connected to