Oncofusion

start / graph / gene-nudt21

Genemoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q86-pan-fusion-core2

NUDT21 / CFIm25 (O43809) - the CFIm subunit that picks poly(A) sites, in the pan-MLL-fusion core, and a paraspeckle protein

also called NUDT21 · CFIm25 · CFIM25 · CPSF5 · CPSF25 · Nudix motif 21 · cleavage factor Im 25 kDa subunit

NUDT21 is the small subunit of a protein pair that decides where a gene's message is cut and given its poly(A) tail, which sets how long the message's untranslated tail is and therefore how stable it is. It co-purifies with every MLL fusion tested, and a 2026 paper shows it drives T-cell leukaemia - the same lineage as the cell line the co-purification was done in.

The same thing, for a scientist

NUDT21 (O43809, 227 aa) is CFIm25, the RNA-binding subunit of cleavage factor Im, and the best-characterised single determinant of proximal-versus-distal poly(A) site choice in human cells. In Skucha 2018 it co-purified with all seven MLL fusions; the MLL-AF4 call is p = 0.0382 with 3|4 spectral counts and a clean 0|0 negative control. It is one of two pan-fusion members of the 3'-end-processing group (with WDR33), and this beat's reading of that group is that the AAUAAA-recognition pair and the CFIm pair are present while the CPSF scaffold and endonuclease (CPSF1/2/3) are entirely absent - a sub-module signature, not a complex dragged along whole. There IS a published link between 3'-end processing and MLL, and it is not about a fusion: Symplekin, a polyadenylation scaffold, interacts and co-localises with MOZ and MLL in immature haematopoietic cells and its loss over-recruits both to the HOXA9 promoter. Independently, NUDT21 is a paraspeckle protein by UniProt's own experimental annotation, and in 2026 was reported to drive T-ALL through APA of UBE2D3 and through promoter binding with MYB, RUNX1 and GATA3 - which is simultaneously the most interesting and the most confounding fact here, because the AP-MS host is Jurkat, a T-ALL line.

Go deeper
The Skucha row, stated in full

7 of 7 fusions. MLL-AF1p 0.0094 (6|4), MLL-AF4 0.0382 (3|4), MLL-AF9 0.0055 (8|8), MLL-CBP 0.0282 (3|5), MLL-EEN 0.0363 (7|7), MLL-ENL 0.0008 (12|13), MLL-GAS7 0.0216 (8|7); negative control 0|0 1. The MLL-AF4 p is the highest of the seven for this protein but is still below 0.05 - and that is worth saying plainly, because the dispatch brief for this beat predicted that several of these twelve would have a weak index-fusion p. They do not: all twelve of this tranche have an MLL-AF4 p below 0.05, the largest being this one at 0.0382. What IS systematically weak is the spectral depth: MLL-AF4 contributes 7 spectra to NUDT21 against 25 under MLL-ENL, and across all 42 pan-fusion proteins MLL-AF4 has the lowest total of the seven baits 34 times.

Is there any published link between 3'-end processing and a KMT2A fusion? Yes, and it is a repressive one

The honest answer to the question this beat was sent to ask is yes, but not for a fusion and not for NUDT21. PMID 23994619 identified Symplekin - the scaffold that assembles the polyadenylation machinery - as a MOZ-interacting protein, showed it interacts and co-localises with both MOZ and MLL in immature haematopoietic cells, and found that knocking it down lowers HOXA9 protein without lowering Hoxa9 mRNA while causing over-recruitment of MOZ and MLL onto the HOXA9 promoter 4. So a 3'-end-processing factor sits on the promoter of the gene that defines this disease's transcriptional programme, and it acts as a brake on MLL there. That is wild-type MLL in mouse/human immature haematopoietic cells, not KMT2A::AFF1, and SYMPK is not in the Skucha 960 network at any n (checked by this beat, 2026-09-21). But it means the co-purification of NUDT21 and WDR33 with all seven fusions is landing on a question the field has already opened once and dropped.

A second, newer line runs the other way round. The 2026 ECAS paper reports that loss of AQR - an intron-binding-complex component in an AFF4-ELL2-EAF1 elongation module - increases promoter-proximal accumulation of cleavage/polyadenylation and RNA exosome factors, and that Aqr perturbation profoundly impairs leukaemia initiation, maintenance and stem-cell function in MLL-rearranged leukaemia 5. AQR itself is not in the Skucha network; XAB2, its partner, is (5 of 7 fusions).

What NUDT21 does, and why the 3'UTR length matters

CFIm25 is a broad repressor of proximal poly(A) site usage. Depleting it shortens the 3'UTR of at least 1,450 genes - about 11% of expressed mRNAs - relieving microRNA- and ARE-mediated repression and raising the protein output of growth-promoting transcripts including cyclin D1; CFIm25 depletion increases proliferation and its re-expression reduces tumour growth in a glioblastoma model 2. In leukaemia specifically, NUDT21 knockdown in K562 inhibits proliferation and promotes apoptosis through MAPK/ERK 3, and in 2026 NUDT21 was reported as a T-ALL dependency acting through distal poly(A) site usage on UBE2D3 and, separately, through binding active promoters with MYB, RUNX1 and GATA3 to drive MYC 6.

The T-ALL confound, which is the strongest caveat in this entry

The Skucha purification was done in Jurkat, a T-ALL line. NUDT21 is reported aberrantly upregulated in T-ALL, associated with poor outcome, and localised to transcriptionally active promoters in T-ALL cells 6. A protein that is abundant, nuclear, chromatin-associated and lineage-relevant in the host cell is exactly the kind of prey an AP-MS experiment recovers regardless of the bait. That does not make the observation uninteresting - it makes the host, not the fusion, the leading alternative explanation, and it is testable in one experiment: repeat the pulldown in a B-lineage t(4;11) line.

Subnuclear address: paraspeckles, not speckles

UniProt O43809 carries an experimental comment that NUDT21 is 'in punctate subnuclear structures localized adjacent to nuclear speckles, called paraspeckles' (ECO:0000269, PubMed 15169763), plus nucleus (four experimental citations) and cytoplasm, and shuttles in an XPO1/CRM1-independent manner. This graph already holds localization-paraspeckles-neat1-kmt2a-r and localization-nuclear-speckles-kmt2a-r, and NUDT21 is the first pan-fusion interactome protein with a paraspeckle annotation rather than a speckle one. The distinction matters because the two compartments have different scaffolds and different RNA.

The complex analysis, and its own negative control

READ THE CONTROLS BEFORE THE ROW. In the same hypergeometric analysis (analyses/mll-fusion-interactome-complexes/results.json) the single strongest complex is CCT/TRiC at 8/9, p = 2.4e-10 - the canonical AP-MS contaminant, which nobody believes is a specific partner. So complex-wide enrichment is not by itself evidence of specific binding, and that caps how much any row in that table can carry. The background is the 20,000-protein proteome rather than the 4,600 proteins the paper actually recovered, which inflates every p; the ranking is the result, not the absolute value.

MODULE READ, measured by this beat from the 960-protein table on 2026-09-21. The five 3'-end-processing proteins in the network are NUDT21 (7/7 fusions), WDR33 (7/7), CPSF4 (6/7), CSTF2 (6/7) and CPSF6 (4/7). What is ABSENT is the point: CPSF1/CPSF160, CPSF2/CPSF100 and CPSF3/CPSF73 - the scaffold and the endonuclease, the structural core of CPSF - are not in the network at any n, and neither are FIP1L1, PCF11, CLP1, CSTF1 or CSTF3. So what co-purifies is the AAUAAA-recognition pair (WDR33 + CPSF4/CPSF30) and the CFIm pair (NUDT21/CFIm25 + CPSF6/CFIm68), not the holo-machine. That is a sub-module signature rather than a complex carried along whole, and it is what the hypergeometric row 'CPSF/3'-end 5/12, p = 1.5e-4' is actually made of.

Facts and figures
uniprotO43809
length AA227
accession resolutionResolved live 2026-09-21: gene_exact:NUDT21 AND organism_id:9606 AND reviewed:true returns exactly ONE reviewed row, gene_primary NUDT21, O43809, entry name CPSF5_HUMAN, 227 aa, synonyms CFIM25 CPSF25 CPSF5. Matches Skucha's table. NOTE a trap in the record itself: UniProt's alternative-name list for O43809 ends with '(Pre-mRNA cleavage factor Im 68 kDa subunit)', which is CPSF6's description, not this 25 kDa protein's - read the LENGTH, not the name string. Registry checked first per rule 88: data/symbol-collisions.json has no row for NUDT21, CFIm25 or CPSF5.
localizationNucleus (ECO:0000269, four PubMed citations) and cytoplasm (ECO:0000269); shuttles nucleus-cytoplasm in a transcription- and XPO1/CRM1-independent manner.
subnuclear localizationPARASPECKLES - UniProt O43809 comment, ECO:0000269 PubMed 15169763: 'In punctate subnuclear structures localized adjacent to nuclear speckles, called paraspeckles.' Fetched 2026-09-21.
localization shown for fusionNO - nobody has imaged NUDT21 in any KMT2A-rearranged cell. The paraspeckle annotation is from a non-leukaemic system.
shown for fusionCo-purification with all seven MLL fusions including MLL-AF4 in Jurkat (p = 0.0382); shRNA-screened against MLL-AF9 in MOLM-13 with the score unpublished. NO independent experiment ties NUDT21 to any KMT2A fusion, and no experiment of any kind ties it to KMT2A::AFF1.
selectivityUNKNOWN. Not named among the paper's non-selectivity controls (RFC1, SF3A3, CCT3), but the paper does not list which 40 of the 128 entered the selectivity re-screen, so absence from that list is uninformative. The one leukaemia knockdown in the literature is in K562 - Skucha's own MLL-WILD-TYPE comparator line - which if anything points to the non-selective class. Must not reach a target list.
functional screenFUNCTIONAL SCREEN. Supplementary Data 2 (file MOESM5) is the shRNA library for the negative-selection screen: 768 rows, 128 unique gene symbols, exactly six shRNAs per gene - counted by this beat on 2026-09-21. All 128 conserved interactors were screened in MLL-AF9-expressing MOLM-13 cells, and the paper says so in those words ('scoring of shRNA-induced effects upon knockdown of all 128 MLL-fusion interactors'). So this protein WAS functionally tested against an MLL fusion, which is more than a co-purification. But the per-gene result is published only as the heatmaps in Fig. 2c/2d, with no table of scores anywhere in the paper or its five supplementary files, so the outcome for this protein is not recoverable. Absence class for that number: measured-not-reported - an afternoon with the authors' source data would close it.
absence classmeasured-not-reported
absence class noteTwo distinct absences here. (1) Whether NUDT21 knockdown is selective for MLL-fusion cells: measured-not-reported - it was screened with six shRNAs in MOLM-13 and the score exists only as a heatmap pixel. (2) Whether NUDT21 binds the endogenous KMT2A::AFF1 fusion in a B-lineage t(4;11) cell: never-measured. PRIDE was checked: PXD009338 is the only relevant deposition and it is the Jurkat experiment itself.
graph contextNUDT21, CFIm25 and CPSF5 appear in ZERO node files in this graph before this entry (checked by id listing over all 1,900 nodes, 2026-09-21). WDR33 is written today in the same tranche; CPSF6, CPSF4 and CSTF2 remain unwritten.
skucha 2018
{
  "bait": "MLL-AF4 (= KMT2A::AFF1)",
  "p_value": 0.0382,
  "spectral_counts": "3|4",
  "n_fusions_of_seven": 7,
  "negative_control_counts": "0|0",
  "per_bait": {
    "MLL-AF1p": "0.0094 (6|4)",
    "MLL-AF4": "0.0382 (3|4)",
    "MLL-AF9": "0.0055 (8|8)",
    "MLL-CBP": "0.0282 (3|5)",
    "MLL-EEN": "0.0363 (7|7)",
    "MLL-ENL": "0.0008 (12|13)",
    "MLL-GAS7": "0.0216 (8|7)"
  },
  "af4_spectral_rank_of_seven": 7,
  "index_bait_note": "p = 0.0382, below 0.05 but the highest of the seven for this protein. Spectral depth, not significance, is the weak point: 7 spectra under MLL-AF4 against 25 under MLL-ENL.",
  "source": "analyses/skucha-mll-fusion-interactome/results.json, re-derived from the cached MOESM4 xlsx on 2026-09-21; run.py re-run the same day, both controls PASS"
}
complex membership
{
  "complex": "CFIm (cleavage factor Im), within CPSF/3'-end processing",
  "partners_in_the_960": {
    "CPSF6 (CFIm68)": 4,
    "WDR33": 7,
    "CPSF4 (CPSF30)": 6,
    "CSTF2": 6
  },
  "absent_from_the_960": [
    "CPSF1 (CPSF160)",
    "CPSF2 (CPSF100)",
    "CPSF3 (CPSF73, the endonuclease)",
    "FIP1L1",
    "PCF11",
    "CLP1",
    "CSTF1",
    "CSTF3",
    "CPSF7 (CFIm59)"
  ],
  "reading": "The recognition and CFIm sub-modules are present and the scaffold/endonuclease core is entirely absent, so this is not a holo-complex carried along by the bait.",
  "hypergeometric": "5 of 12, p = 1.5e-4 against a 20,000 proteome (analyses/mll-fusion-interactome-complexes). Read beneath the negative control in that same table: CCT/TRiC scores 8/9 at p = 2.4e-10 and is a canonical contaminant."
}
cell types and states
[
  "Jurkat (human T-ALL) - the AP-MS host, nuclear lysate, dox-induced single-copy tagged fusion",
  "MOLM-13 (human MLL-AF9 AML) - shRNA screen host, six shRNAs, score not published",
  "K562 (human BCR::ABL1 CML, MLL-wild-type) - knockdown inhibits proliferation",
  "human T-ALL patient samples and T-ALL lines - upregulated, poor outcome",
  "no B-lineage KMT2A::AFF1 cell of any kind"
]
bounded literature
{
  "query": "(NUDT21[tiab] OR CFIm25[tiab] OR CPSF5[tiab]) AND (KMT2A[tiab] OR MLL[tiab] OR \"MLL-AF4\"[tiab] OR \"MLL-AF9\"[tiab] OR \"t(4;11)\"[tiab])",
  "querytranslation": "(\"NUDT21\"[Title/Abstract] OR \"CFIm25\"[Title/Abstract] OR \"CPSF5\"[Title/Abstract]) AND (\"KMT2A\"[Title/Abstract] OR \"MLL\"[Title/Abstract] OR \"MLL-AF4\"[Title/Abstract] OR \"MLL-AF9\"[Title/Abstract] OR \"t 4 11\"[Title/Abstract])",
  "count": 0,
  "date": "2026-09-21",
  "tool": "NCBI esearch db=pubmed",
  "names_tried": [
    "NUDT21",
    "CFIm25",
    "CPSF5",
    "CPSF25",
    "CFIM25"
  ],
  "positive_control": "MEN1[tiab] AND the identical five-term clause returns 85.",
  "pathway_level_query": "The standing rule that a pathway is often named by something other than its gene symbol was applied: (\"alternative polyadenylation\"[tiab] OR \"cleavage and polyadenylation\"[tiab] OR \"3' end processing\"[tiab] OR \"polyadenylation site\"[tiab]) AND the same KMT2A clause returns 4, of which PMID 23994619 (Symplekin/MOZ/MLL/HOXA9) and PMID 42748909 (ECAS) are on topic - so the SYMBOL-level zero is a true zero and the PATHWAY-level answer is not. That difference is the finding.",
  "bound": "Titles and abstracts only; supporting figures not searched. Every cytogenetic term quoted per the standing rule; the querytranslation is recorded above and shows 't 4 11'[Title/Abstract], i.e. the quoted form parsed correctly and no [UID] term was injected."
}
functionally screened in the source paper
{
  "_added": "king, cycle 48, 2026-09-21, after Q86 flagged that cycle-47 entries understate the record. Verified by me directly from the supplementary file, not from the paper's prose.",
  "screened": true,
  "library_symbol": "NUDT21",
  "shrnas_for_this_gene": 6,
  "design": "Supplementary Table 6 / MOESM5 is 768 data rows, 128 unique gene symbols, EXACTLY 6 shRNAs each -- counted with openpyxl on 2026-09-21. Negative-selection RNAi screen in MLL-AF9-expressing MOLM-13.",
  "what_this_DOES_establish": "Every one of the 128 conserved MLL-fusion interactors, this protein included, was functionally tested against an MLL fusion. An entry saying 'nothing in the paper validates this prey' is true of BINDING and understates the record.",
  "what_this_does_NOT_establish": "A per-gene result. Scores are presented as heatmaps (Fig. 2c/2d) and no per-gene depletion value is tabulated, so for any individual gene the outcome is measured-not-reported. It is also MLL-AF9 MOLM-13, a myeloid line, not the index fusion and not B-lineage.",
  "absence_class_for_the_per_gene_score": "measured-not-reported"
}
How this could bite us back
  • CFIm is required for normal pre-implantation development and NUDT21 is broadly expressed; a 3'-end-processing intervention inherits a developmental toxicity profile in the exact age group this disease affects.
  • The only published leukaemia knockdown of NUDT21 is in K562 - an MLL-WILD-TYPE line, and Skucha's own negative comparator. If that result is the general case, NUDT21 is an essential-function target, not a selective one.
  • CFIm25's effect on tumour growth has the opposite sign in different cancers, so an inhibitor could shorten 3'UTRs and increase output of the very oncogenes it was meant to lower.
  • If the published direction is right - polyadenylation factors restraining MLL at HOXA9 - then removing NUDT21 would raise fusion-target output, not lower it.
Still open
  1. Does the endogenous KMT2A::AFF1 fusion co-immunoprecipitate NUDT21 in a t(4;11) B-lineage line (SEM, RS4;11, MV4-11)? That single experiment separates the fusion from the Jurkat host.
  2. Do KMT2A::AFF1 target genes - HOXA9, MEIS1, PROM1, BCL2 - show altered 3'UTR length on menin inhibition? APA is measurable from RNA-seq already in GEO, so this is a measured-not-analysed question with no wet work.
  3. Symplekin represses MLL at HOXA9 while NUDT21 and WDR33 co-purify with every MLL fusion. Is the 3'-end machinery a brake on the fusion rather than a passenger? The experiment is a Symplekin or WDR33 knockdown with HOXA9 protein and mRNA read separately, in a t(4;11) line.
  4. Is the paraspeckle annotation relevant? NEAT1 is already a topic in this graph. Co-staining NUDT21 with NEAT1 and with the fusion in a t(4;11) line would say whether these compartments are shared.
Every source behind this entry (6)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    NUDT21 co-purified with all seven Strep-HA-tagged MLL fusions expressed as single-copy transgenes in isogenic human Jurkat T-ALL cells; for MLL-AF4 (= KMT2A::AFF1) the combined p-value is 0.0382 with spectral counts 3|4 and a negative control of 0|0.

    MLL-fusion-driven leukemia requires SETD2 to safeguard genomic integrity. · Skucha A, Ebner J, Schmoellerl J, Roth M, Eder T, Cesar-Razquin A, Stukalov A, Vittori S, Muhar M, Lu B, Aichinger M, Jude J, Mueller AC, Gyorffy B, Vakoc CR, Valent P, Bennett KL, Zuber J, Superti-Furga G, Grebien F · Nature Communications 2018
    PMID 29777171DOI
    What this evidence cannot show

    SYSTEM: the AP-MS host is JURKAT, a human T-ALL line, while KMT2A::AFF1 disease is B-lineage infant leukaemia; a partner requiring a B-cell or infant proteome cannot appear here and a T-lineage partner can appear spuriously. The bait is a Strep-HA-tagged, SINGLE-COPY, doxycycline-induced transgene at an FRT/Flp cassette, not the endogenous fusion. Prey came from bulk nuclear lysate (buffer C420, 420 mM NaCl) after a single STREP-Tactin step, so co-purification is NOT evidence of a direct contact. INCLUSION RULE: the paper kept the top 300 interactions per bait by combined p-value rather than applying a p threshold (Methods, verbatim: "All the identified bait-prey pairs were ranked by the combined p-value. The 300 most significant interactions per bait were retained."). Measured by this beat from the cached MOESM4 xlsx on 2026-09-21: 545 of the 2,049 per-bait p-values in the table (26.6%) exceed 0.05; within the MLL-AF4 column alone it is 186 of 291 (63.9%), and that column's maximum p is 0.175. MEASURED BY THIS BEAT, 2026-09-21, from the cached MOESM4 xlsx: across all 42 pan-fusion proteins the MLL-AF4 column carries the LOWEST total spectral counts of the seven baits for 34 of 42, and is sixth or seventh for 40 of 42. That is a bait-level property (transgene expression, purification depth or the DDX5/DDX17 GLM normalisation), not a statement about any one prey, so a low MLL-AF4 count must not be read as a weaker association than the same protein's count under another bait. For NUDT21 the MLL-AF4 total (7 spectra over two replicates) is rank 7 of the seven baits, against a maximum of 25 under another bait. All seven per-bait values, p (replicate spectral counts): MLL-AF1p 0.0094 (6|4); MLL-AF4 0.0382 (3|4); MLL-AF9 0.0055 (8|8); MLL-CBP 0.0282 (3|5); MLL-EEN 0.0363 (7|7); MLL-ENL 0.0008 (12|13); MLL-GAS7 0.0216 (8|7). SPECIFIC TO THIS PROTEIN: the host cell line is Jurkat and NUDT21 has since been reported as a T-ALL dependency that binds active promoters in T-ALL cells, so host-lineage abundance is a live alternative explanation for this row and is not excluded by anything in the paper. SELECTIVITY: the paper's named non-selectivity controls are RFC1, SF3A3 and CCT3 - knockdown inhibited MLL-wild-type K562 and HL-60 as much as MOLM-13 - against MEN1 as the selective positive. Only 40 of the 128 ('the 40 candidate genes with the highest confidence') entered that MLL-r versus MLL-wild-type re-screen, and the paper does not list which 40. This protein is not named among the controls, but that is NOT evidence of selectivity: it is equally consistent with the protein never having reached the second screen. Recorded as unknown, and this entry must not be read as nominating a target. Values re-read from the cached xlsx on 2026-09-21 and cross-checked against analyses/skucha-mll-fusion-interactome/results.json, whose two controls (128 proteins shared by >=5 baits and 42 by all seven, both matching the paper; MEN1 present with all seven) both PASS on today's re-run.

    Evidence that argues against this (2)
    • IntAct has never curated this dataset, so no independent curator has re-examined these calls and no interactome-coverage metric built on IntAct can see them; the raw spectra at ProteomeXchange PXD009338 have not been reanalysed here. The authors themselves removed seven proteins from the shared set by hand as 'either frequently observed contaminants or ... not detected in human hematopoietic cell lines', which concedes that the rank cutoff lets contaminants through - and ACTB survived that filter, sitting in the pan-fusion 42 with a NON-ZERO negative control (3|3). Ten of the 42 carry a non-zero negative control: PSIP1 0|5, MCM2 0|2, ACTB 3|3, CSNK2A1 2|0, RUNX1 0|2, SF3A3 0|3, TCEB3 0|1, CHD4 2|2, SMARCC1 1|0, ZNF512 0|2 - counted by this beat from the xlsx neg_ctrl column, 2026-09-21. ProteomeXchange/PRIDE 2018
    • The paper states that 'more than one third of proteins in the network of conserved MLL-fusion protein interaction partners were identified as essential in recent genome-wide screens'. Abundance and essentiality both raise the odds of recovery in an AP-MS experiment, so membership in the conserved core is partly a statement about how much of a protein is present in a nucleus, not only about what it binds. Skucha A et al. 2018
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    CFIm25/NUDT21 is a broad repressor of proximal poly(A) site usage; its depletion shortens the 3'UTR of at least 1,450 human genes, raises expression of growth-promoting transcripts including cyclin D1, and increases cell proliferation, while its re-expression reduces glioblastoma tumour growth.

    CFIm25 links alternative polyadenylation to glioblastoma tumour suppression. · Masamha CP, Xia Z, Yang J, Albrecht TR, Li M, Shyu AB, Li W, Wagner EJ · Nature 2014
    PMID 24814343DOI
    What this evidence cannot show

    Human cell lines and a glioblastoma xenograft model, 2014; not haematopoietic and not a fusion system. The 1,450-gene figure comes from a regression model applied to standard RNA-seq, so it is an inference from read distribution rather than a direct 3'-end measurement. Read from the abstract. A published Author Correction exists (PMID 36418405) and was not read; the headline claim is unaffected as far as the correction notice's title indicates, and this entry does not depend on any specific number from the paper beyond the order of magnitude.

    Evidence that argues against this (1)
    • The direction of CFIm25's effect is not uniform across cancers - a 2019 review in Cancer Letters is titled 'CFIm25 and alternative polyadenylation: Conflicting roles in cancer', and in hepatocellular carcinoma NUDT21 has been reported as tumour-suppressive by a different mechanism. Any inference from 3'UTR shortening to a phenotype in leukaemia is therefore not transferable from glioblastoma. Tan S et al. 2018
  3. E3weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    NUDT21 knockdown in the human K562 leukaemia line inhibits proliferation and promotes apoptosis, with MAPK/ERK identified as the main pathway affected.

    PMID 30349365DOI
    What this evidence cannot show

    K562 is a BCR::ABL1 CML line, MLL-wild-type, and is one of the two lines Skucha et al. used as their MLL-WILD-TYPE comparator. So this result, if it generalises, points toward NUDT21 being a non-selective leukaemia dependency rather than an MLL-fusion-specific one - the SF3A3 shape. Two-author single-line study in a low-impact journal; no rescue experiment reported in the abstract; n not stated. Read from the abstract.

    Evidence that argues against this (1)
    • A dependency demonstrated only in an MLL-wild-type line cannot distinguish a general proliferation requirement from an MLL-fusion-linked one, and Skucha et al. used K562 precisely as the negative arm. On its own this makes NUDT21 look like the RFC1/SF3A3/CCT3 class, and the entry is written that way. Skucha A et al. 2018
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Symplekin, the scaffold of the polyadenylation machinery, interacts and co-localises with both MOZ and MLL in immature haematopoietic cells, and its inhibition lowers HOXA9 protein without lowering HOXA9 mRNA while causing over-recruitment of MOZ and MLL onto the HOXA9 promoter.

    Symplekin, a polyadenylation factor, prevents MOZ and MLL activity on HOXA9 in hematopoietic cells. · Largeot A, Paggetti J, Broseus J, Aucagne R, Lagrange B, Martin RZ, et al. · Biochim Biophys Acta 2013
    PMID 23994619DOI
    What this evidence cannot show

    This is the only published physical link this beat could find between the 3'-end-processing machinery and MLL, and it concerns WILD-TYPE MLL in immature haematopoietic cells, not any KMT2A fusion and not the index fusion. Symplekin (SYMPK) is NOT in the Skucha 960-protein network at any n, checked by this beat against the cached MOESM4 xlsx on 2026-09-21, so this paper's protein and this entry's protein are different members of the same machinery. Read from the abstract only; the co-localisation and ChIP experiments were not inspected. 2013, BBA Molecular Cell Research.

    Evidence that argues against this (1)
    • The mechanism reported is REPRESSIVE - Symplekin is a brake whose removal increases MOZ and MLL occupancy at HOXA9. If 3'-end-processing factors generally restrain MLL at HOXA cluster promoters, then their presence in a fusion purification would be evidence of a restraining interaction rather than a cooperating one, and the therapeutic direction inverts. Nothing in the Skucha data distinguishes the two. Skucha A et al. 2018
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Acute loss of AQR, a component of a U2 snRNP-associated intron-binding module coupled to an AFF4-ELL2-EAF1 elongation module (ECAS), increases promoter-proximal accumulation of cleavage/polyadenylation and RNA exosome factors, and Aqr perturbation profoundly impairs leukaemia initiation, maintenance and leukaemia stem-cell function in MLL-rearranged leukaemia in mouse haematopoietic cells.

    A dual-module ECAS complex couples transcription elongation and RNA processing. · Xiang Y, Wang H, Yin Z, Xiao R, Song Z, Lu R, et al. · Mol Cell 2026
    PMID 42748909DOI
    What this evidence cannot show

    2026 Molecular Cell; read from the abstract only. The leukaemia arm is MOUSE MLL-rearranged haematopoietic cells and the abstract does not state which fusion, so this is not an index-fusion result and by rule the claim above says 'mouse'. The link to this entry is indirect: ECAS is about AQR and the elongation-splicing interface, and its relevance here is only that perturbing that interface redistributes cleavage/polyadenylation factors at promoters, which is a mechanism by which 3'-end factors could be at chromatin at all. AQR is NOT in the Skucha 960 network; XAB2 is, at 5 of 7 fusions. Another agent in this cycle owns gene-aqr; this entry deliberately does not write that protein.

    Evidence that argues against this (1)
    • ECAS is reported as molecularly DISTINCT from the canonical P-TEFb-containing SEC, which is the complex this project's KMT2A::AFF1 mechanism rests on. So the paper is as much a warning against merging elongation assemblies as it is a bridge between them, and no experiment in it involves a KMT2A fusion protein as bait. Skucha A et al. 2018
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    NUDT21 is aberrantly upregulated in human T-ALL and required for T-ALL maintenance, acting both as a CFIm subunit promoting distal poly(A) site usage on UBE2D3 and, separately, by localising to transcriptionally active promoters and interacting with MYB, RUNX1 and GATA3 to facilitate MYC transcription.

    PMID 41926643DOI
    What this evidence cannot show

    2026, Advanced Science; read from the abstract only. T-ALL, not B-lineage infant KMT2A::AFF1, and no MLL fusion appears anywhere in the abstract. The pharmacology reported is ouabain octahydrate, a cardiac glycoside with a well-known general cytotoxicity, so 'pharmacological targeting of NUDT21' should be read as a degradation phenotype rather than as a selective tool compound. The reason this citation is in this entry is the opposite of a supporting one: it is the best available description of what NUDT21 does IN THE LINEAGE OF THE HOST CELL, and therefore the leading confound for the Skucha row.

    Evidence that argues against this (1)
    • RUNX1 is itself in the pan-fusion 42 with a non-zero negative control (0|2), and MYB is one of Skucha's own positive-control shRNAs. A NUDT21-MYB-RUNX1 promoter module in T-ALL is an alternative generator of the same co-purification pattern that does not require the fusion to be involved at all. Skucha A et al. 2018
Connected to