Oncofusion

start / graph / gene-qki

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q132-interactome-tail-promotion

QKI - a KMT2A proximity row in K562, and a tumour suppressor whose loss concentrates in the HOXA subtype of T-ALL

also called Quaking · QK · QK1 · QK3 · KH domain-containing RNA-binding protein QKI

QKI is an RNA-reading protein that decides how other genes' messages are spliced together. It shows up once as a neighbour of KMT2A in a leukaemia cell line, and separately - and much more solidly - it is lost in childhood T-cell leukaemia, most of all in the HOXA-driven subtype that this project's fusions produce.

The same thing, for a scientist

QKI (HGNC:21100, ENSG00000112531, UniProt Q96PU8, 341 aa, 6q26) is a STAR-family KH-domain RNA-binding protein controlling splicing, circular-RNA formation, export and stability [E3]. Its interactome link here is thin: one spoke-expanded BioID2 row, **QKI bait / KMT2A prey**, IntAct IM-30059, K562, MI-score 0.27 [E1][E2]. Its disease link is not. Palhais et al. analysed **two independent paediatric T-ALL cohorts** and found QKI expression *frequently reduced in T-ALL, particularly within the HOXA subtype*, with reduction correlating with poor overall and event-free survival; QKI depletion in T-ALL lines induced widespread splicing alterations corroborated in patient samples, and QKI over-expression reduced proliferation and induced G0/G1 arrest [E4]. **The HOXA subtype is the one this project's biology produces** - KMT2A rearrangements and PICALM::MLLT10 are the canonical drivers of HOXA-activated T-ALL - so a tumour suppressor whose loss concentrates there is a co-occurrence worth measuring, and nobody has measured it against fusion status.

Go deeper
Two independent facts, and the join between them is untested

Fact one, weak. One IntAct row: MI:1314 BioID2, MI:2364 proximity, MI:1060 spoke, QKI bait / KMT2A prey, host taxid:9606(Homo sapiens K562 myeloid leukemia cell), annotation Data file S6 2. Proximity in an over-expressing CML line with no KMT2A rearrangement.

Fact two, strong. QKI is reduced in paediatric T-ALL, most in the HOXA subtype, and the reduction tracks with worse survival 4.

The join is the question and it has not been asked: is QKI loss enriched in KMT2A-rearranged or PICALM::MLLT10 T-ALL specifically, or is it a property of the HOXA transcriptional state however it is produced? Those are different claims with different consequences, and the published cohort analysis is stratified by subtype rather than by fusion as far as the abstract states.

This graph has already been through the corresponding story once: ana-five-unwritten-interactors-that-are-also-selective-dependencies nominated QKI among five unwritten interactors that looked selectively required in KMT2A-rearranged lines, and ana-the-null-kills-my-own-shortlist then found none of the five survived a null once the 254-protein search was accounted for. The DepMap selectivity claim for QKI is dead. What is written here is a different and independent line of evidence - patient cohorts, not cell-line dependency - and it should not be allowed to inherit the dead claim's credibility.

Localization and the isoform problem

UniProt Q96PU8 records nucleus, cytoplasm, cytosol and stress granule, all experimental, across isoform-scoped comments: there is a nuclear isoform acting as an indirect splicing regulator and a cytosolic isoform recognising m7G-modified mRNA and partitioning it into stress granules with G3BP1 3. The BioID row cannot say which isoform was the bait, and the paper itself warns that it used a single isoform per protein 1. A nuclear-isoform proximity to KMT2A and a cytosolic-isoform one would mean entirely different things.

No subnuclear compartment is assigned and none is asserted here.

Cell type and cell state

Interaction row: K562, BCR::ABL1 CML, no KMT2A rearrangement. Disease evidence: primary paediatric T-ALL, two cohorts, plus T-ALL cell lines 4. Those are different cells and different lineages, and the entry does not let the T-ALL result lend weight to the K562 row or the reverse.

Facts and figures
hgnc IDHGNC:21100
symbolQKI
ensembl gene IDENSG00000112531
uniprotQ96PU8
length AA341
chromosomal location6q26
identifier provenanceHGNC REST fetch/symbol/QKI and UniProtKB Q96PU8 retrieved live 2026-09-23; PMIDs 41537349 and 39633556 fetched from PubMed efetch and their titles read back before citation.
subnuclear locationnone assigned; none asserted here
cell type of the observationinteraction row: K562. Disease evidence: primary paediatric T-ALL, two cohorts, plus T-ALL lines.
epigenetic roleNone. QKI acts post-transcriptionally; the HOXA link is transcriptional-state co-occurrence, not a chromatin activity.
shown for fusionNO for the physical relation - nobody has looked in any fusion. For the disease relation: reduced QKI concentrates in the HOXA SUBTYPE of paediatric T-ALL, which is the subtype KMT2A rearrangements and PICALM::MLLT10 produce, but the published stratification is by subtype and not by fusion, so no fusion is named by the evidence.
depmap statusThis graph's null analysis (ana-the-null-kills-my-own-shortlist) found QKI's apparent selective dependency in KMT2A-rearranged lines does not survive correction for the 254 proteins searched. That claim is dead and must not be revived from this entry.
evidence classinteraction: single publication, single method, spoke-expanded BioID prey row. Disease: two independent patient cohorts plus cell-line functional assays, abstract-read only.
prev and alias symbols
[
  "QK3"
]
interaction with project baits
[
  {
    "bait": "KMT2A",
    "bait_uniprot": "Q03164",
    "partner_experimental_role": "bait",
    "scaffold_experimental_role": "prey",
    "method": "MI:1314 proximity-dependent biotin identification (BioID2)",
    "interaction_type": "MI:2364 proximity",
    "expansion": "MI:1060 spoke",
    "mi_score": 0.27,
    "imex": "IM-30059",
    "pmid": "39251607",
    "host_cell": "K562 (BCR::ABL1 CML, no KMT2A rearrangement)",
    "n_rows": 1,
    "n_publications": 1
  }
]
subcellular location uniprot
[
  "Nucleus (ECO:0000269)",
  "Cytoplasm (ECO:0000269)",
  "Cytoplasm, cytosol (ECO:0000269)",
  "Cytoplasm, stress granule (ECO:0000269)"
]
How this could bite us back
  • QKI is reported as a TUMOUR SUPPRESSOR here - it is lost, not gained. There is nothing to inhibit; any intervention would have to restore it, which is a far harder pharmacological problem than inhibition.
  • If QKI loss is a consequence of the HOXA state rather than a driver, restoring it would not reverse the state, and the survival correlation would be a marker rather than a mechanism.
  • The cell-line dependency signal and the patient tumour-suppressor signal have opposite signs. Acting on either without resolving that contradiction risks building on whichever one happens to be an artefact.
Still open
  1. Within HOXA-subtype T-ALL, is QKI loss enriched in KMT2A-rearranged and PICALM::MLLT10 cases specifically, or is it uniform across the subtype however HOXA is activated? Re-stratifying the published cohorts by fusion is a reanalysis, not a new experiment.
  2. Is QKI reduced in B-lineage KMT2A::AFF1 infant ALL at all? The entire disease evidence here is T-lineage and the index disease has never been checked.
  3. Which QKI isoform - nuclear or cytosolic - produced the KMT2A proximity row, given that the source study used a single isoform per bait?
  4. Do the splicing events QKI depletion induces overlap the transcripts a KMT2A fusion drives, or are they an independent programme that happens to co-occur?
Every source behind this entry (5)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Khoroshkin et al. generated in vivo proximity-dependent biotinylation datasets for 50 human RNA-binding proteins in K562 cells and deposited the curated bona fide pair list to IMEx as IM-30059; QKI is one of the 50 baits and KMT2A appears in its prey list.

    Systematic identification of post-transcriptional regulatory modules. · Khoroshkin M, Buyan A, Dodel M, et al. · Nat Commun 2024
    PMID 39251607DOI
    What this evidence cannot show

    Abstract read here. This graph's ana-rbp-bioid2-proximity-map-k562 records that the full text (PMC11385195) was read in an earlier cycle and that the elongation factors appear nowhere in the body - these pairs are reachable only through the supplementary data and the IMEx deposition. K562 is a BCR::ABL1 chronic myeloid leukaemia line with NO KMT2A rearrangement. The authors note that RBP over-expression, BioID2 fusion and the use of a single isoform per protein may bias the data.

    Evidence that argues against this (1)
    • No SEC component, no KMT2A-complex component and no fusion protein is a BAIT in this deposition; the scaffold proteins appear only as prey among 2,682 preys, and the median prey is recovered by five of the fifty baits. A single prey row here is weak by the deposition's own internal scale. source 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct serves one curated human QKI-KMT2A row: MI:1314 proximity-dependent biotin identification, MI:2364 proximity, MI:1060 spoke expansion, QKI experimental role MI:0496 bait and KMT2A MI:0498 prey, MI-score 0.27, host taxid:9606 Homo sapiens K562 myeloid leukemia cell, IMEx IM-30059, publication PMID 39251607, curator annotation 'Data file S6'.

    What this evidence cannot show

    Retrieved live 2026-09-23. Rule 93: PSICQUIC is a floor relative to IntAct REST and cannot reach processed-chain identifiers. MI-score 0.27 is the deposition-wide value for this screen and carries no pair-specific information.

    Evidence that argues against this (1)
    • The source deposition used a single isoform per bait, and QKI's nuclear and cytosolic isoforms have different partners and different compartments. The row therefore cannot say which QKI was near KMT2A, and a cytosolic-isoform row would carry no nuclear meaning at all. source 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt Q96PU8 is 341 aa and records QKI as an RNA reader regulating pre-mRNA splicing, circular RNA formation, mRNA export, stability and translation, with isoform-scoped experimental subcellular locations nucleus, cytoplasm, cytosol and stress granule, and a cytosolic isoform that binds m7G-modified transcripts and partitions them into stress granules with G3BP1.

    What this evidence cannot show

    Fetched live 2026-09-23. Several function comments are 'By similarity' rather than experimental and are not relied on here. Rule 91 applies to the location codes.

    Evidence that argues against this (1)
    • Several of the UniProt function statements for QKI are annotated 'By similarity' - inferred from orthologues rather than measured in human - and the retrieved record does not separate them cleanly in the summary text. Only the human-experimental subset should be relied on, and this entry has not audited which statements fall in it. source 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Palhais et al. analysed two independent paediatric T-ALL cohorts and found QKI expression frequently reduced in T-ALL, particularly within the HOXA subtype, with reduction correlating with poor overall and event-free survival; QKI depletion in T-ALL cell lines induced widespread splicing alterations, numerous events corroborated in patient samples, and QKI over-expression significantly reduced proliferation and induced G0/G1 cell cycle arrest, identifying QKI as a splicing regulator with tumour-suppressive properties in T-ALL.

    QKI dysregulation induces extensive splicing changes in T-cell acute lymphoblastic leukemia. · Palhais B, Sharma ND, Fijalkowski I, et al. · Haematologica 2026
    PMID 41537349DOI
    What this evidence cannot show

    ABSTRACT ONLY; the truncated final sentence of the abstract was not resolved. The cohorts are T-ALL, not B-lineage KMT2A::AFF1 - this is a lineage this project touches through PICALM::MLLT10 and ETP-ALL, not its index disease. The abstract stratifies by HOXA SUBTYPE, not by fusion, so nothing here says QKI loss co-occurs with a KMT2A rearrangement. A survival correlation in a retrospective cohort is not a causal claim.

    Evidence that argues against this (1)
    • This graph's own DepMap analysis found that QKI's apparent selective requirement in KMT2A-rearranged lines does NOT survive a null accounting for the 254 proteins searched - QKI had the best p-value of five candidates and it was still not significant. A cell-line dependency signal and a patient-cohort tumour-suppressor signal point in opposite directions, and both are in this entry on purpose. source 2026
  5. E5weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    A second Haematologica report, 'Depletion of the RNA binding protein QKI and circular RNA dysregulation in T-cell acute lymphoblastic leukemia' (2025), exists from an overlapping author group.

    PMID 39633556DOI
    What this evidence cannot show

    TITLE AND AUTHOR LIST ONLY - PubMed serves no abstract for this record, so nothing about its design, cohort or result is known to this entry. It shares an author with [E4] and is therefore NOT an independent replication. Cited as a pointer, not as evidence.

    Evidence that argues against this (1)
    • PubMed serves NO abstract for this record, so nothing about its design, cohort size or result is known here, and it shares an author with the 2026 paper it sits beside - it is not an independent replication and must not be counted as a second cohort. Cited as a pointer only. source 2025
Connected to