Oncofusion

start / graph / interface-psip1-utp3-declined-psip1-is-prey-to-nine-of-fifty-rbp-baits

Interfaceweak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.Q129-the-never-discussed-partners-of-kmt2a-and-psip1

PSIP1-UTP3 is one spoke-expanded BioID2 proximity row in K562 with UTP3 as bait and PSIP1 as prey at miscore 0.27; PSIP1 is recovered by 9 of the 50 baits in that screen and only 2 of the 9 are nucleolar, which is the panel's own base rate - declined

also called PSIP1-UTP3 · LEDGF-UTP3 · SAS10 · CRL1 · EBI-56918260

A database lists UTP3, a nucleolus protein that helps assemble ribosomes, as a partner of the nuclear protein LEDGF/PSIP1. In the screen that produced it, fifty different bait proteins were tagged and nine of them picked up LEDGF, only two of those nine being nucleolus proteins - the same proportion as in the bait panel overall.

The same thing, for a scientist

**Verdict: DECLINED, on a quantitative test inside the source dataset. Detection method is BioID2 proximity-dependent biotin identification (PSI-MI MI:1314), spoke-expanded, with UTP3 as bait and PSIP1 as prey** [E1]. One IntAct record, EBI-56918260, from PMID 39251607 (Khoroshkin et al., Nat Commun 2024), host K562, **intact-miscore 0.27 - the lowest of any record in this beat**, annotated to Data file S6 [E1]. UTP3 is UniProt Q9NQZ2, 479 aa, **Nucleus/nucleolus [ECO:0000269]**, a small-subunit (SSU) processome component [E3]. **This was the beat's best candidate for the pre-registered nucleolar hypothesis: it is the only one of the three nucleolar factors arriving on PSIP1 rather than KMT2A, and the only one by a non-AP-MS method.** The pre-registered discriminators resolve against it. **PSIP1 is recovered by 9 of the screen's 50 baits, and only 2 of those 9 - UTP3 and SBDS - are nucleolar or ribosome-biogenesis proteins. The panel itself is 9 of 50 nucleolar, so 2 of 9 is 22% against an expected 18%: exact binomial p = 0.50** [E4]. **PSIP1's proximity neighbourhood in this screen is not nucleolar-enriched**, so the UTP3 hit is what the panel composition predicts. This extends, on the PSIP1 side, what this graph already established on the KMT2A side - that the same census recovers KMT2A from 25 of 50 baits, a figure `ana-kmt2a-proximity-promiscuity-is-explained-by-size-97th-percentile-becomes-91st-among-proteins-its-own-length` showed is explained by KMT2A's size and which it concluded cannot support individual proximity partners [E5].

Go deeper
The pre-registered test, its result, and the condition under which it would have been uninformative

Registered before any record was pulled (scratchpad q129_prereg.md, 2026-09-23), the hypothesis was that ABT1, KRR1 (on KMT2A) and UTP3 (on PSIP1) - three nucleolar ribosome-biogenesis factors on two baits - indicate (a) genuine nucleolar biology, (b) AP-MS contaminant background, or (c) coincidence at n=3. Four discriminators were fixed in advance: D1 publication identity, D2 method orthogonality, D3 contaminant prior, D4 the source paper's own controls. A fifth clause specified what would make D1 and D2 uninformative: if the whole list of thirteen traces to one or two publications and one method, then a shared screen says nothing about nucleolar specificity, because it is a fact about the list's composition.

That clause fired, partially, and it must be reported. The thirteen partners come from four publications and four detection methods, so the list is not homogeneous - D1 and D2 retain some power. But BioPlex 3.0 alone supplies 6 of 13, including ABT1 and KRR1, so observing that those two share a screen is close to the base rate for the list and carries little information. D1 is therefore weak as applied to ABT1 and KRR1, exactly as pre-registered, and the nucleolar argument had to be made per-protein instead.

What carried information was D3 and D4, applied inside this dataset. Not pre-registered in this exact form and therefore labelled post-hoc: of the 50 BioID2 baits, 9 are nucleolar or ribosome-biogenesis proteins (DDX21, DDX52, GNL3, NOLC1, NPM1, RPS3, RPS11, SBDS, UTP3). PSIP1 is recovered by 9 baits, of which 2 are in that set. Expected 1.6; observed 2; exact binomial p = 0.50 4.

Robustness to where the boundary is drawn. A generous classification adding DDX55 and EXOSC5 gives 11 of 50 = 22%, and the observed 2 of 9 is then exactly 22%. No boundary this entry tried produces an enrichment.

And the honest limit, stated because n = 9 is small. With 9 draws at an 18% base rate, this test would have needed 5 of the 9 baits to be nucleolar to reach p < 0.05. It can exclude a large enrichment and cannot exclude a modest one. A negative here is a bound, not a proof, and the bound is: PSIP1's proximity neighbourhood is not dominated by the nucleolus.

What the record is, and how it sits against what this graph already knows about this dataset

Khoroshkin et al. fused BioID2 to 50 human RBPs in K562, captured each neighbourhood by streptavidin pull-down and mass spectrometry with matched controls processed without a biotin pulse, and deposited the pair list to IMEx 2. IntAct serves it as 19,380 rows resolving to 50 baits and 2,682 distinct preys, median 5 baits per prey, mean 7.23, maximum 40 - reproduced independently here 4.

Two instrument notes worth recording. ana-rbp-bioid2-proximity-map-k562 records 19,403 rows for this deposition against the 19,380 counted here on 2026-09-23 through IntAct REST rather than PSICQUIC; the bait count, prey count, median, mean and maximum all reproduce exactly. A 23-row difference across endpoints and dates is not an error in either, and it is recorded rather than silently harmonised. Separately, the IntAct REST endpoint returns 0 for the field-scoped form pubid:39251607 and 19,380 for the bare PMID as free text - a clean, well-formed false zero of the same shape as rules 85 and 95.

The authors' own treatment of background is the D4 answer 2. Their functional annotation converts each bait-prey target score to a Z-score across baits before any enrichment analysis - an explicit correction for preys that come up with everything. Their integrated map then thresholds at the 0.25 quantile of integrated distance to give 1,001 RBP-RBP pairs. The IntAct rows are the raw bait-prey layer beneath both of those steps, which is the layer the authors themselves treat as needing correction.

And PSIP1 is not an RBP bait in this screen. It appears only as prey, as does KMT2A. The dataset contains no bait relevant to the menin-KMT2A complex 5.

Facts and figures
partner agene-psip1
partner a uniprotO75475
partner a length AA530
partner bUTP3
partner b uniprotQ9NQZ2
partner b length AA479
verdictDECLINED
adjudicated byQ129, cycle 92, 2026-09-23
source publication pmid39251607
source datasetKhoroshkin 2024 50-RBP BioID2 proximity map, K562
detection methodBioID2 proximity-dependent biotin identification (PSI-MI MI:1314)
experimental role partnerbait
experimental role psip1prey
psip1 was never the baittrue
expansion methodspoke expansion
intact miscore0.27
n publications1
n methods1
orthogonal method existsfalse
compartment mismatchfalse
partner nucleolar evidence codeECO:0000269 (experimental)
intact row count partner295
dataset n baits50
dataset n preys2682
dataset rows intact rest 2026 09 2319380
dataset rows recorded in ana rbp bioid2 proximity map k56219403
psip1 prey percentile68
kmt2a baits recovering it25
kmt2a prey percentile97
druggabilityNot assessable. No residue, domain or region is mapped on either side; BioID2 reports a labelling radius, not a contact, so this method cannot in principle produce an interface.
absence classmeasured-not-searched
absence class noteThe per-pair target score, fold change and p-value that the authors computed for UTP3-PSIP1 are in Data file S6 of PMID 39251607 and were not opened; the raw proteomics is in PXD041608 and was not queried. Both are re-analysis, not wet work. The separate never-measured absence is that no proximity or affinity experiment anywhere has used PSIP1, KMT2A, menin or any KMT2A-complex component as a bait in this or any comparable modern proximity map - a point ana-rbp-bioid2-proximity-map-k562 already records.
shown for fusionNO - nobody has looked in any KMT2A fusion. K562 carries BCR::ABL1 and no KMT2A rearrangement, so the entire dataset is a wild-type control for this project's disease.
epigenetic readoutnone - no histone mark, rDNA or occupancy measurement exists for this pair, although UTP3's UniProt function comment names a role in the structure of silenced chromatin.
nucleolar hypothesis roleThe third and strongest of the three nucleolar candidates and the only one on PSIP1; adjudicated with ABT1 and KRR1 in ana-q129-the-nucleolar-triple-survives-none-of-the-four-pre-registered-discriminators.
host cell lines
[
  "K562 myeloid leukaemia (BCR::ABL1, no KMT2A rearrangement)"
]
records
[
  "EBI-56918260"
]
psip1 baits recovering it
[
  "AGGF1",
  "FASTKD2",
  "HNRNPC",
  "ILF3",
  "PPIL4",
  "RBM15",
  "SBDS",
  "U2AF1",
  "UTP3"
]
nucleolar baits in panel
[
  "DDX21",
  "DDX52",
  "GNL3",
  "NOLC1",
  "NPM1",
  "RPS3",
  "RPS11",
  "SBDS",
  "UTP3"
]
nucleolar enrichment test
{
  "observed": 2,
  "of": 9,
  "panel_base_rate": "9/50 = 18%",
  "expected": 1.62,
  "exact_binomial_p": 0.5,
  "threshold_for_significance": "5 of 9",
  "robustness": "a generous 11/50 = 22% classification makes observed 2/9 = 22% an exact match; no boundary tried produces enrichment",
  "prespecified": false,
  "label": "POST-HOC"
}
interface residues a
[]
structures
[]
How this could bite us back
  • PSIP1/LEDGF is the chromatin tether for the menin-KMT2A complex and the target of the LEDGIN allosteric inhibitor class, so a genuine nucleolar arm for PSIP1 would mean that displacing it from chromatin has consequences in a compartment nobody assays.
  • If the nucleolar reading were right, the intervention it implies - targeting SSU-processome assembly - has no therapeutic window: ribosome biogenesis is required by every proliferating cell including normal haematopoietic progenitors, and this is a disease of infants.
  • Declining this pair does not close the nucleolus for this project. The correct statement is that PSIP1's proximity neighbourhood in one K562 screen is not nucleolar-enriched, which bounds one measurement and leaves the compartment itself unmeasured.
Still open
  1. What is PSIP1's baits-per-prey percentile WITHIN its own length decile in this map? KMT2A's 97th percentile fell to the 91st once size was held constant, and PSIP1 at 530 aa sits in a much lower decile, so the same correction would push its 9 baits UP rather than down. That single computation is the one that could reopen this row and it was not run here.
  2. What target score did Khoroshkin et al. assign to the UTP3-PSIP1 pair in Data file S6, and does it survive their own Z-scoring across baits? The file is public, it was not opened, and it is the authors' own verdict on this row.
  3. Does any deposited nucleolar-fraction proteomics from a KMT2A-rearranged leukaemia line exist in PRIDE? Whether KMT2A or PSIP1 has a nucleolar pool at all is unmeasured in this graph, and no archive query was run in this beat.
Every source behind this entry (5)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct curates one record of a PSIP1-UTP3 proximity interaction, EBI-56918260, from PMID 39251607. Detection method 'proximity-dependent biotin identification' (PSI-MI MI:1314), interaction type 'proximity', expansion method SPOKE EXPANSION, host Homo sapiens K562 myeloid leukemia cell. UTP3 (Q9NQZ2) is the BAIT and PSIP1 (O75475) is the PREY. intact-miscore 0.27, the lowest of any of the thirteen records adjudicated in this beat. Annotation: 'figure legend (Data file S6)'. Within this publication PSIP1 is recovered as prey by 9 distinct baits - AGGF1, FASTKD2, HNRNPC, ILF3, PPIL4, RBM15, SBDS, U2AF1 and UTP3 - and KMT2A by 25.

    What this evidence cannot show

    Read live from IntAct REST on 2026-09-23, not PSICQUIC (rule 93). Data file S6 was not opened, so the target score, fold change and p-value the authors computed for this bait-prey pair are unknown here - those are the numbers that would grade the record, and the miscore of 0.27 is IntAct's aggregate, not theirs. Spoke expansion: the binary PSIP1-UTP3 pair was never measured, it was inferred by joining every prey in the UTP3 neighbourhood to the bait.

    Evidence that argues against this (1)
    • K562 is a human myeloid leukaemia line, so unlike every other record in this beat this one was obtained in a haematopoietic cell. That is a genuine advantage and it is the reason this pair was the beat's best nucleolar candidate. K562 carries BCR::ABL1 and no KMT2A rearrangement, so it is a wild-type control for this project's disease rather than a model of it. Khoroshkin M et al. 2024
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    In human K562 cells, Khoroshkin et al. fused the BioID2 system to 50 RBPs, validated expression of each fusion by western blot, and captured each protein neighbourhood by streptavidin pull-down and mass spectrometry, including for each RBP a matched control processed without a biotin pulse. For functional annotation they estimated target scores as -log10(P) times the sign of log2 fold change for every bait-prey pair, then converted target scores to Z-scores across baits for each prey before gene-set enrichment. Their integrated RBP interaction map thresholds at the 0.25 quantile of integrated distance, yielding 1,001 RBP-RBP pairs of which 776 are not in STRING, an average of 22 contacts per RBP.

    Systematic identification of post-transcriptional regulatory modules. · Khoroshkin M, Buyan A, Dodel M, Navickas A, Yu J, Trejo F, et al. (22 authors), Goodarzi H · Nature communications 2024
    PMID 39251607DOI
    What this evidence cannot show

    Full text read 2026-09-23 through Europe PMC fullTextXML (PMC11385195, 144,676 stripped characters). The strings 'UTP3', 'PSIP1', 'LEDGF' and 'KMT2A' each appear ZERO times in the paper's text - the row exists only in Data file S6, which was not opened. So the paper does not report this interaction in prose. The no-biotin control is a control for endogenous biotinylation and non-specific streptavidin binding; it does not control for the labelling radius of BioID2, which is the mechanism by which an abundant nearby protein becomes a prey without contacting the bait.

    Evidence that argues against this (1)
    • The authors' Z-scoring across baits is evidence that they took prey promiscuity seriously and corrected for it in their own analysis, and their integrated map overlaps significantly with STRING, OpenCell, hu.MAP and Zanzoni by permutation test. The dataset is careful. The objection is that IntAct exposes the raw bait-prey layer beneath those corrections, not that the study is weak. Khoroshkin M et al. 2024
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProtKB Q9NQZ2 is 479 aa, primary gene name UTP3, recommended name 'Something about silencing protein 10'. Its only subcellular-location comment is Nucleus, nucleolus [ECO:0000269]. Function: essential for gene silencing with a role in the structure of silenced chromatin, a role in the developing brain by similarity, and part of the small subunit (SSU) processome, the first precursor of the small eukaryotic ribosomal subunit assembled in the nucleolus on nascent pre-rRNA. PSIP1 (O75475) is 530 aa with a single subcellular-location comment, Nucleus [ECO:0000269], and is a transcriptional coactivator. UTP3 carries 295 interaction records in IntAct, against 464 for KRR1 and 761 for ABT1.

    What this evidence cannot show

    UniProt location comments are curated pointers, not measurements (rule 91); the cited papers were not opened. UTP3's nucleolar annotation is experimentally coded, so unlike ABT1 this protein's nucleolar identity is not itself an inference - the pre-registered 'annotation artefact' condition does NOT apply to UTP3, and that is recorded here rather than left implicit.

    Evidence that argues against this (1)
    • UTP3's function comment opens with 'Essential for gene silencing: has a role in the structure of silenced chromatin' - a chromatin function, not only a ribosome-biogenesis one. A nucleolar protein with a curated chromatin-silencing role is a more plausible neighbour for a chromatin tether than a pure ribosome assembly factor would be, and this is the strongest single argument in the record for the pair. UniProt Consortium 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    IntAct REST serves 19,380 rows from PMID 39251607, resolving to 50 distinct baits and 2,682 distinct preys, with median 5 baits per prey, mean 7.23 and maximum 40; 19,376 rows are proximity-dependent biotin identification and 4 are clip-seq, and every row is bait-prey. PSIP1 is recovered by 9 of the 50 baits, placing it at the 68th percentile of preys (851 of 2,682 preys are recovered by 9 or more baits); KMT2A is recovered by 25, in the top 2.9%. Of the 50 baits, 9 are nucleolar or ribosome-biogenesis proteins - DDX21, DDX52, GNL3, NOLC1, NPM1, RPS3, RPS11, SBDS and UTP3 - i.e. 18% of the panel. Of PSIP1's 9 baits, 2 are in that set (UTP3, SBDS), i.e. 22%. Exact binomial with n = 9 and p = 0.18 gives an expectation of 1.62 and P(X >= 2) = 0.50; reaching P < 0.05 would have required 5 of the 9.

    What this evidence cannot show

    The classification of 9 of 50 baits as nucleolar or ribosome-biogenesis is THIS ENTRY'S, made from UniProt function and location comments, and another reader could draw the boundary differently. A generous version adding DDX55 and EXOSC5 gives 11 of 50 = 22%, against which the observed 2 of 9 = 22% is an exact match, so the conclusion is not sensitive to the boundary. The test is POST-HOC - it was not among the four discriminators pre-registered before the data were pulled - and is labelled as such. n = 9 is small: this test excludes a large nucleolar enrichment and cannot exclude a modest one. Row count differs by 23 from the 19,403 recorded in ana-rbp-bioid2-proximity-map-k562, which used PSICQUIC on an earlier date; bait count, prey count, median, mean and maximum all reproduce exactly. Instrument note: the field-scoped form pubid:39251607 returns 0 on this endpoint while the bare PMID returns 19,380, so a beat using the field-scoped form would record a false zero.

    Evidence that argues against this (1)
    • A base-rate test over bait identity assumes each bait is an independent draw on PSIP1's neighbourhood, which it is not: the 50 baits are all RBPs, they share compartments and complexes, and several of the nine nucleolar ones are in the same machinery. Under that dependence the effective n is smaller than 9 and the test is weaker still than the stated bound. IntAct (EMBL-EBI) 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    This graph already holds two analyses of this same dataset. ana-rbp-bioid2-proximity-map-k562 records that no super elongation complex component, no KMT2A-complex component and no fusion protein is a bait in it, and that AFF1, AFF4, ELL, MLLT1, CDK9, CCNT1, KMT2A, MEN1 and DOT1L appear only in the prey column. ana-kmt2a-proximity-promiscuity-is-explained-by-size-97th-percentile-becomes-91st-among-proteins-its-own-length records that KMT2A's recovery by 25 of 50 baits puts it at the 97.1st percentile of all preys but only the 90.7th among the 268 proteins in its own 1,462-5,890 aa length decile, where the median prey is hit by 9 baits rather than 5, and concludes that this census cannot support individual proximity partners for KMT2A.

    What this evidence cannot show

    These are this project's own entries, not external sources, and are cited to establish what is already known rather than as independent evidence - an internal citation cannot corroborate an internal claim. The size-decile correction was performed for KMT2A and NOT for PSIP1; PSIP1 is 530 aa and sits in a much lower length decile, so the equivalent correction would run in the opposite direction and was not computed in this beat. That is a real gap in the argument here and is listed as an open question rather than glossed.

    Evidence that argues against this (1)
    • The size explanation for KMT2A could cut the other way for PSIP1. If large proteins are recovered by roughly twice as many baits as typical ones, then a 530 aa protein recovered by 9 baits may be MORE promiscuous than its size predicts, not less - which would make PSIP1's 9 baits a stronger signal than the raw 68th percentile suggests. That computation was not done and it is the one that could reopen this row. Oncofusion Knowledge Graph (internal) 2026
Connected to