Oncofusion

start / graph / hyp-aff1-reaches-ptefb-only-through-aff4

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q76-spoke-provenance

AFF1 reaches P-TEFb only through AFF4, and the index fusion's elongation phenotype is therefore a claim about AFF4 availability

also called the SEC topology around AFF1 is untested · AFF1-CDK9 is 16 spokes and no contact

Everyone says the AF4 protein in the infant-leukaemia fusion grabs the machine that switches genes on. Nobody has ever shown AF4 touching that machine directly - and its close relative AF5q31, which has been crystallised holding it, may be the one doing the grabbing for both.

The same thing, for a scientist

Every IntAct record pairing AFF1 with a P-TEFb subunit is spoke-expanded: AFF1-CDK9 16 of 16, AFF1-CCNT1 8 of 8, with no MI:0407 record, no structural detection method and no shared PDB entry [E1]. AFF4, the paralogue, is crystallised directly against CDK9 and cyclin T1 in five entries including 4IMY and 6CYT [E2]. AFF1 and AFF4 heterodimerise through the AF4-homology region, and that pair is itself 8 of 8 spoke-expanded [E1]. **The hypothesis is that AFF1 has no direct P-TEFb interface at all and is recruited to the kinase by AFF4**, which would make the AFF1 arm of the SEC an adaptor-of-an-adaptor and make AFF4 stoichiometry, not AFF1 abundance, the quantity that sets elongation output in a KMT2A::AFF1 cell. The prior is deliberately low - the co-purifications are reproducible across labs and the simplest reading is that both AF4-family proteins bind P-TEFb through a shared motif - but **nothing in the curated record distinguishes the two topologies**, and the discriminating experiment is a week of purified-protein work.

Go deeper
Why this is a hypothesis and not a finding

The observation is an absence of a discriminating experiment, not a positive result. AFF1 and AFF4 are paralogues with conserved N-terminal regions; the parsimonious expectation is that both bind P-TEFb the same way, and the AFF4-P-TEFb structures then predict the AFF1 interface rather than replacing it.

What makes the alternative worth writing down is the asymmetry of the record. AFF4 has been crystallised against the kinase five times. AFF1 has one PDB entry in existence - 42 residues, and with AF9, not with P-TEFb 3. If AFF1's P-TEFb interface were equivalent and equally tractable, twenty years of structural effort on the SEC would be expected to have produced at least one AFF1-P-TEFb structure, and there is none. That is weak evidence, and it is the only evidence there is.

The hypothesis is also worth writing down because it is cheap to kill and because the two topologies make different predictions about a drug.

Predictions that separate the two topologies
AFF1 binds P-TEFb directlyAFF1 binds P-TEFb only via AFF4
purified AFF1 fragment + purified P-TEFb, no AFF4measurable Kdno binding
AFF4 knockout, then AFF1 IPP-TEFb still recoveredP-TEFb lost
AFF4 knockout in a t(4;11) linepartial loss of elongationcollapse of elongation at fusion targets
AlphaFold-Multimer AFF1 vs CCNT1, calibrated on the AFF4 complexesconfident interface at the paralogous motifno confident interface

The second row is the decisive one and needs nothing the field does not have. AFF4-null cells are viable, AFF1 antibodies exist, and the readout is a western blot.

What it would mean for the fusion

KMT2A::AFF1 retains the AFF1 C-terminal region, including the region through which AFF1-AFF4 heterodimerisation is reported. If the fusion's route to P-TEFb runs through endogenous AFF4, then:

  1. AFF4 becomes a dependency of the fusion, and one that is not itself rearranged - the kind of node a degrader could reach without touching the fusion.
  2. The fusion competes with wild-type AFF1 for a limited AFF4 pool, which would make the phenotype dose-dependent on fusion expression in a way a direct-binding model does not require.
  3. Every measurement of 'SEC recruitment by the fusion' becomes ambiguous about which subunit is being recruited, and the existing co-IP literature cannot resolve it because it is all n-ary.

If the hypothesis is false - AFF1 binds P-TEFb directly - the graph gains a defended interface where it currently has sixteen spokes, which is worth the same week of work.

Facts and figures
statementAFF1 makes no direct contact with any P-TEFb subunit; its recruitment to CDK9/cyclin T1 is mediated entirely by AFF4, and the elongation phenotype of KMT2A::AFF1 is therefore set by AFF4 availability rather than by AFF1 abundance.
originliterature-gap
prior probability0.2
predictions
[
  "Purified AFF1 fragments show no measurable binding to purified P-TEFb by ITC or SPR in the absence of AFF4, while the paralogous AFF4 fragment binds in the same assay.",
  "In AFF4-null cells, an AFF1 immunoprecipitation recovers no CDK9 or cyclin T1, while an AFF1 IP in the parental line does.",
  "AlphaFold-Multimer, calibrated on AFF4/CCNT1 as a positive control, returns no confident interface for AFF1 against CCNT1 or CDK9 at the paralogous region.",
  "In a t(4;11) line, AFF4 depletion reduces Ser2 phosphorylation and elongation at fusion target loci more than AFF1 depletion does.",
  "Overexpressing the KMT2A::AFF1 fusion titrates a limited AFF4 pool, so elongation output at fusion targets is non-monotonic in fusion dose."
]
experiments
[
  "exp-aff1-sec-binary-ladder-against-purified-ptefb"
]
analyses
[
  "ana-aff1-has-one-observed-binary-interaction-and-forty-two-residues-of-structure",
  "ana-two-thirds-of-the-curated-interactome-is-a-row-nobody-observed"
]
falsifiers
[
  "A measurable Kd between a purified AFF1 fragment and purified P-TEFb with no AFF4 present kills it outright.",
  "Recovery of CDK9 in an AFF1 IP from AFF4-null cells kills it.",
  "Any published structure, crosslink or hydrogen-deuterium-exchange footprint placing AFF1 directly against CDK9 or cyclin T1 kills it, and a full-text sweep for one has NOT yet been done - that sweep is the cheapest possible refutation and should precede any bench work.",
  "Demonstration that the AFF4 P-TEFb-binding motif is conserved in AFF1 and sufficient for binding when transplanted would make the hypothesis very unlikely without formally killing it."
]
How this could bite us back
  • If AFF4 turns out to be the real P-TEFb contact, it becomes an attractive target - and AFF4 is a general elongation factor required in normal cells, so the therapeutic index would be worse than menin's, not better. A dependency that is not rearranged is also a dependency that is not cancer-specific.
  • The hypothesis could be true and useless: if the AFF1-AFF4 heterodimer is obligate and constitutive, then 'AFF1 binds P-TEFb via AFF4' and 'AFF1 binds P-TEFb' are the same statement operationally, and nothing downstream changes. The experiment is still worth doing because the graph would then be able to say which it is.
  • Depleting AFF4 to test this in a leukaemia line will reduce global elongation, so any fusion-target-specific effect must be normalised against a matched set of non-target genes or the readout will show a collapse that means nothing.
Still open
  1. Has anyone published a binary AFF1/P-TEFb binding measurement in a supporting figure? A full-text sweep of the SEC literature is the cheapest refutation and was not done in this beat.
  2. Is the AFF4 P-TEFb-binding motif conserved in AFF1 at the sequence level? An alignment would set the prior properly and takes minutes; it was deliberately not used to set the 0.2 here, because a prior set from the same observation that motivated the hypothesis is not a prior.
  3. Does KMT2A::AFF1 retain enough of the AFF1 N-terminal region to bind AFF4 at all, and if so is the heterodimer between fusion and endogenous AFF4 or between fusion and endogenous AFF1?
Every source behind this entry (3)
  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-CDK9 is 16 records all MI:1060 spoke-expanded, AFF1-CCNT1 is 8 records all spoke-expanded, and AFF1-AFF4 is 8 records all spoke-expanded; none of the three pairs has any MI:0407 direct-interaction record, any structural or binary-biophysical detection method, or any PDB entry containing both chains.

    What this evidence cannot show

    Bounded by IntAct's curated record on one day. A binary AFF1 binding experiment published in a supporting figure of a paper about something else would not be curated and would be invisible here; this beat did not do a full-text sweep for one, and doing so is the first step before the bench experiment. The absence of a shared PDB entry is bounded the same way and is additionally biased against intrinsically disordered proteins, which AFF1 largely is.

    Evidence that argues against this (1)
    • Sixteen spoke-expanded records across several publications is reproducibility, which is a real and different kind of support from directness. If several labs with different tags in different cells all recover CDK9 with AFF1, the co-membership is not in doubt, and a hypothesis built on the absence of a binary experiment must not be read as doubting it. EMBL-EBI IntAct 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PDB 4IMY 'The AFF4 scaffold binds human P-TEFb adjacent to HIV Tat' is an x-ray structure whose three polymer entities map to CDK9 (P50750), cyclin T1 (O60563) and AFF4 (Q9UHB7); 6CYT contains the same three plus HIV-1 Tat and TAR RNA; UniProt lists five AFF4 entries shared with CDK9 and cyclin T1 in total (4IMY, 4OGR, 4OR5, 5L1Z, 6CYT).

    What this evidence cannot show

    4IMY and 6CYT were verified entity-by-entity; the other three are asserted from the cross-reference intersection only. These structures were solved in the context of HIV Tat transactivation rather than leukaemia, so the AFF4 construct and the conditions were chosen for that question - which does not weaken the interface they show but does mean the biological framing was somebody else's.

    Evidence that argues against this (1)
    • The AFF4-P-TEFb structures argue FOR the parsimonious alternative, not for this hypothesis: if the paralogue binds through a conserved N-terminal motif, the same motif in AFF1 is the obvious prediction, and the hypothesis then amounts to betting against sequence conservation. That is the main reason its prior is set low. RCSB Protein Data Bank 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    UniProt P51825 (AFF1, 1,210 aa) carries exactly one PDB cross-reference, 2LM0, chains A = residues 738-779, an NMR structure of an AF4/AF9 chimera from PMID 23260655.

    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 PubmedData/ArticleIdList/ArticleId path. The cross-reference and chain range come from UniProt fetched the same day. A PDB count is a poor proxy for structural knowledge of a disordered protein, and NMR, SAXS, crosslinking-MS and prediction can all constrain an interface without a deposition - none of which is counted.

    Evidence that argues against this (1)
    • AFF1 is largely intrinsically disordered, so the absence of structures is expected from the molecule and not diagnostic of neglect. Using a structure count as evidence about AFF1's interfaces is close to using it as evidence that AFF1 is disordered, which everyone already knows. UniProtKB 2026
Connected to