Oncofusion

start / graph / hyp-q180-the-index-fusion-specific-interactome-has-never-been-measured-and-this-table-cannot-settle-it

Hypothesismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q180-the-af4-specific-interactors

KMT2A::AFF1 has no measured fusion-specific interactome: the one dataset that appears to supply one is a rank-selected tail in the wrong lineage, and the question is open rather than answered badly

also called the AF4-specific interactome does not exist yet · q180 hypothesis

It looks as though we know which proteins bind only our fusion — a published table lists 110 of them. On inspection that list is the weak end of a fixed-size ranking, from a T-cell line that does not have our fusion, and it excludes the proteins that are genuinely specific to it. The honest position is that nobody has measured this yet.

The same thing, for a scientist

**Statement:** the set of proteins that bind KMT2A::AFF1 and not other KMT2A fusions has never been measured in a system where the question is answerable, and Skucha 2018's 110 MLL-AF4 singletons are not that set. Four measurements support this: the selection is top-300 per bait so the singleton count tracks each bait's cut depth (Spearman +0.821, exact p 0.0341); within MLL-AF4 the singletons are the non-significant tail (median p 0.114, 5% below 0.05, median 2 spectra); the set is 14.5% mitochondrial against 2.4% for the shared core (Fisher p 0.042); and the definition excludes AFF1, AFF4 and ELL, the three proteins most specific to this fusion. **The hypothesis is falsifiable in two directions and both are cheap enough to be worth stating:** a reanalysis of PXD009338 at peptide level could show the tail is better than it looks, and a B-lineage endogenous-fusion interactome would settle the question outright.

Go deeper
Why this is a hypothesis and not simply a negative result

A negative result would be "the AF4-specific partners are background". This entry deliberately does not say that, for two reasons.

First, the singleton class demonstrably contains real biology elsewhere in the same table: DOT1L under MLL-AF9 (p 0.00172, 24 spectra) and four bait 3' moieties at p 1e-17 to 1e-55. The axis works. MLL-AF4 simply has no member of the strong mode, because AFF1 left the class on a technicality.

Second, this beat adjudicated 6 of 110 and declined all six. The other 104 are unexamined, and the compartment audit says 37 of the 110 are nuclear. Declining six and inferring 110 would be the exact error this beat spent its analysis entries criticising.

So the claim is about the state of knowledge: the question "what binds KMT2A::AFF1 and not its relatives" is open, and the instrument that appears to have answered it cannot.

Predictions
  1. A peptide-level reanalysis of PXD009338 will show that a substantial minority of the 110 rest on one or two PSMs, and that at least some — RGPD8 is the named case — cannot be resolved to the gene at all. If instead the tail is clean, well-scored, unique-peptide identifications, the depth argument survives but the quality argument does not.
  2. The mitochondrial members will recur in the other six baits' specific tails. If ACADM, GLUD1, ECHS1 and PDHA1 are singletons of several different baits (which cannot happen under the n_fusions definition, so the test is whether they appear in the other baits' columns below the cut), they are a preparation signature.
  3. An endogenous KMT2A::AFF1 interactome in SEM or RS4;11 will share the pan-fusion core and almost none of the 110. This is the strong prediction and the expensive one.
  4. β-catenin will be found in PXD009338 below the rank cut, or not at all. Either outcome is informative for the LEF1 row specifically.
Facts and figures
statementThe set of proteins that physically associate with KMT2A::AFF1 and not with other KMT2A fusions has never been measured in a system where the question is answerable. Skucha 2018's 110 MLL-AF4 singletons are a rank-depth artefact of a top-300-per-bait design in a Jurkat T-ALL host, enriched for mitochondrial contaminants, and defined so as to exclude AFF1, AFF4 and ELL.
origindata-pattern
prior probability0.8
shown for fusionThe hypothesis is ABOUT the index fusion, KMT2A::AFF1, and about the fact that the only dataset bearing on it is a tagged transgene in a Jurkat T-ALL host.
epigenetic readoutnone - a hypothesis about the state of proteomic evidence.
absence classnever-measured
absence class noteThere is no AP-MS or proximity interactome of the ENDOGENOUS KMT2A::AFF1 in a B-lineage host. That is wet work. The subsidiary questions - peptide evidence, beta-catenin below the cut, cross-bait recurrence of the mitochondrial set - are all measured-not-analysed and answerable from PXD009338.
predictions
[
  "A peptide-level reanalysis of PXD009338 will show a substantial minority of the 110 resting on one or two PSMs, and RGPD8 specifically will prove unresolvable from its paralogues.",
  "The sixteen mitochondrial members of the MLL-AF4 tail will be detectable in the other six baits' raw data below the rank cut, marking them as a preparation signature rather than bait-specific.",
  "An endogenous KMT2A::AFF1 interactome in a B-lineage KMT2A-rearranged line will recover the pan-fusion core and almost none of the 110.",
  "CTNNB1 will be found in PXD009338 below the rank cut, or not at all; either result is decisive for the LEF1 row."
]
experiments
[
  "exp-q180-reanalyse-pxd009338-for-peptide-level-evidence-under-the-rank-cut"
]
analyses
[
  "ana-q180-how-far-a-baits-fixed-list-runs-into-noise-predicts-how-specific-it-looks",
  "ana-q180-the-af4-specific-tail-is-mitochondrial-where-the-shared-core-is-nuclear",
  "ana-q180-the-af4-specific-definition-throws-away-aff1-aff4-and-ell"
]
falsifiers
[
  "ANY member of the 110 recovered as an MLL-AF4 partner by a second, independent method — co-IP, BioID, proximity ligation or a structure — in any system. One such result kills the strong form of this hypothesis, because it would show the tail contains at least one real partner and the question then becomes how many.",
  "A peptide-level reanalysis of PXD009338 showing that the six p<0.05 AF4-specific preys rest on three or more distinct, unique, high-scoring peptides each with no shared-peptide ambiguity. That would not make them partners, but it would move the argument from 'the identifications are thin' to 'the identifications are sound and the biology is unproven', which is a materially different claim and would require this beat's six declines to be re-reasoned.",
  "An endogenous KMT2A::AFF1 pulldown in SEM or RS4;11 recovering 20% or more of the 110 at better than p 0.01. That would show the Jurkat tail is reproducible across lineage and host, and the depth-artefact reading would be wrong.",
  "A published MLL-fusion AP-MS dataset other than PXD009338 in which MLL-AF4's specific partner count is NOT the largest of its baits despite MLL-AF4 again having the deepest cut. That would break the depth correlation and the Spearman +0.821 would be a coincidence of seven points."
]
How this could bite us back
  • If this hypothesis is right, any target proposed from the AF4-specific tail is a target with no measured interaction behind it. The graph should not carry such a target as though the interactome supported it.
  • If this hypothesis is wrong — if the tail is largely real — then the graph has declined and de-prioritised a set of genuinely index-fusion-specific partners, which is the more costly error for the PI's standing priority. The six declines in this beat are each individually reversible from the deposit, and the hypothesis says so explicitly for that reason.
Still open
  1. Is there any MLL-fusion interactome dataset in PRIDE or MassIVE other than PXD009338? This beat did not sweep either archive. If a second exists, every claim here becomes testable by replication instead of by argument.
  2. Would a proximity-labelling approach (BioID, TurboID) on KMT2A::AFF1 in a B-lineage line be a better instrument than AP-MS for this question? This corpus already holds proximity-interactome entries for other baits, and proximity labelling survives the compartment problem that sinks a nuclear-extract pulldown.
  3. How many of the 104 unadjudicated members of the tail can be declined on annotation alone, without reading anything further? The compartment audit puts 16 in the mitochondrial class and 58 in neither nucleus nor mitochondrion.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Four independent measurements over Skucha 2018 Supplementary Table 5 — cross-bait depth correlation (Spearman +0.821, exact p 0.0341), within-bait significance decomposition (singleton median p 0.114 vs 0.019), compartment enrichment (14.5% vs 2.4% mitochondrial, Fisher p 0.0417) and the definitional exclusion of AFF1, AFF4 and ELL — together establish that the 110 MLL-AF4 singletons are not a measurement of the index fusion's specific interactome.

    What this evidence cannot show

    All four measurements are on the same table, so they are not fully independent of one another - a single systematic feature of that table (its top-N-per-bait design) plausibly drives the first two, and the third and fourth are logically separate but drawn from the same 960 rows. Nothing external to this dataset was used to test any of them, and no second MLL-fusion AP-MS dataset exists to replicate against.

    Evidence that argues against this (1)
    • The strongest argument against this hypothesis is that it is unfalsifiable in the direction that matters most: if a B-lineage endogenous interactome is never done, the claim 'the index-fusion-specific interactome has never been measured' stays true forever without ever having been tested. That is a real weakness of the framing, and it is why predictions 1, 2 and 4 - all answerable from the existing deposit - matter more than prediction 3. ProteomeXchange / PRIDE, EMBL-EBI 2018
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The singleton class is not uniformly noise — it contains DOT1L under MLL-AF9 at p 0.00172 on 24 spectra and four bait 3' moieties (CREBBP, EPS15, SH3GL1, GAS7) at p 1.77e-55 to 3.13e-17 on 95-257 spectra — so the correct conclusion is that MLL-AF4 has no member of the strong mode, not that the axis fails.

    What this evidence cannot show

    A bait's own 3' moiety is present at 1:1 stoichiometry with the bait by construction, so it calibrates the axis generously; DOT1L at 24 spectra is the only non-trivial calibrator and a single point. If the true scale for a genuine sub-stoichiometric partner-specific interactor were nearer 5 spectra, the six MLL-AF4 candidates would not be excluded by this comparison.

    Evidence that argues against this (1)
    • MLLT1 under MLL-ENL carries p = 9.92e-23 on only THREE spectra, which directly contradicts the idea that a strong p requires many spectra in this GLM - the baseline term can drive an extreme p from few observations when the control is empty. If that can happen for a true positive it can happen for a false one, and it means neither p nor spectral count alone orders these candidates reliably. Nature Communications 2018
Connected to