USP8 (UBPY) - the second deubiquitinase KMT2A partner, three patients, same UCH logic

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

• USP8 is required for internalisation of liganded receptor tyrosine kinases in normal cells. Any systemic USP8 inhibitor would perturb receptor turnover broadly, and USP8 gain-of-function mutations already cause Cushing disease, so this is a gene with a demonstrated phenotype when its activity is altered in humans.
• Three patients cannot support a development programme. USP8's role here is to corroborate a mechanism hypothesis, and treating it as a target in its own right would be over-reach.
• The t(11;15)(q23;q21) in the one detailed case was visible cytogenetically, so unlike KMT2A::USP2 this fusion is not reliably cryptic. Assuming it hides the same way would send a laboratory to the wrong assay.

1. Does the KMT2A::USP8 junction fall 5' or 3' of the WW domain? The answer decides whether the fusion carries a protein-interaction module in addition to the catalytic domain, and no junction sequence is published for any of the three patients.
2. USP8's dimerisation domain is N-terminal and therefore probably lost in the fusion. Is that right, and does it rule out forced oligomerisation as the mechanism, leaving catalysis as the only candidate contribution?
3. Is there a third deubiquitinase among the KMT2A partners not yet recognised as one? The recombinome's list of 107 in-frame fusions has not been systematically screened for UCH-domain content, and that screen is a morning's work against Pfam.

1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   USP8 is UniProt P40818, 1118 aa, with a UCH catalytic domain at residues 777-1106 (Pfam PF00443), a rhodanese-like domain at 184-305, a USP8 dimerisation domain at 8-115 and a WW domain at 648-680.

   UniProtKB entry P40818 (UBP8_HUMAN), Ubiquitin carboxyl-terminal hydrolase 8, with InterPro/Pfam domain boundaries · UniProt Consortium; InterPro (EMBL-EBI) · UniProtKB / InterPro 2026
   ▸What this evidence cannot show

   Database annotation on the canonical isoform. Which USP8 residue begins the fusion protein is not published for any of the three patients, so the retained/lost domain assignment is positional inference. Accession resolved by fetching the record and reading back the PRIMARY gene name (USP8; synonyms KIAA0055, UBPY) and length (1118 aa) rather than trusting a gene_exact synonym match.

   ▸Evidence that argues against this (1)

   • Pfam boundaries are model matches, not experimentally determined domain limits, and the region between the WW domain and the catalytic domain is long and poorly annotated. A junction falling there would produce a fusion whose composition this entry cannot predict. Meyer C et al. 2019
2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   USP8 was identified as a new KMT2A fusion partner at 15q21.2 in the minor-BCR discovery cohort; the KMT2A::USP8 case was one of four balanced translocations breaking in BCR2; USP8 and USP2 rearrangements alike fused the conserved UCH domain to an extended 5'-MLL portion; and USP8 normally regulates endosome morphology, receptor-tyrosine-kinase internalisation and ESCRT stability.

   Human MLL/KMT2A gene exhibits a second breakpoint cluster region for recurrent MLL-USP2 fusions. · Meyer C, Lopes BA, Caye-Eude A, Cavé H, Arfeuille C, et al., Marschalek R · Leukemia 2019
   PMID 30899083DOI
   ▸What this evidence cannot show

   One patient in this cohort. The functional description of USP8 is a background paragraph citing prior literature not read here, not a measurement. The UCH-retention statement covers USP2 and USP8 together and no junction sequence is given for either. Full text read (PMC6756029).

   ▸Evidence that argues against this (1)

   • USP8's normal biology is endosomal and cytoplasmic, which makes it a poor fit for any model in which the partner contributes to a nuclear chromatin complex. If the UCH domain's contribution is not catalytic, there is currently no account at all of what USP8 adds. Blackburn PR et al. 2020
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   The 2023 KMT2A recombinome counts 3 USP8 patients among 3401, against 30 for USP2; USP2, AFDN and SEPTIN6 dominate the 47 minor-BCR cases.

   The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, Burmeister T, Gröger D, et al., Marschalek R · Leukemia 2023
   PMID 37019990DOI
   ▸What this evidence cannot show

   Three patients at one referral centre. A count of 3 cannot distinguish a genuinely rare partner from a partner that assays miss, and USP8 shares USP2's cytogenetic invisibility. Counts read from the published article in this session.

   ▸Evidence that argues against this (1)

   • AFDN contributes 9 minor-BCR patients with no deubiquitinase domain, so UCH-partner enrichment in BCR2 is a majority pattern rather than a requirement, and a three-patient partner cannot strengthen a rule that a nine-patient partner breaks. Meyer C et al. 2019
4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   An independent KMT2A-USP8 case in a 16-month-old boy with B-ALL had trisomy 8 and t(11;15)(q23;q21), an atypical KMT2A FISH result, and a focal 15q deletion with a breakpoint within USP8 on chromosomal microarray; the rearrangement was resolved by mate-pair sequencing.

   Cryptic and atypical KMT2A-USP2 and KMT2A-USP8 rearrangements identified by mate pair sequencing in infant and childhood leukemia. · Blackburn PR, Smadbeck JB, Znoyko I, Webley MR, Pitel BA, Vasmatzis G, Xu X, Greipp PT, Hoppman NL, Ketterling RP, Baughn LB, Lindsey KG, Schandl CA, Wolff DJ, Peterson JF · Genes, Chromosomes & Cancer 2020
   PMID 32196814DOI
   ▸What this evidence cannot show

   One patient. Trisomy 8 is a co-occurring lesion whose contribution cannot be separated. The prognostic descriptors the paper applies (high-risk, CNS involvement, poor prednisone response, poor EFS) are restated from the Frankfurt USP2 series and not derived for USP8. Abstract read in full.

   ▸Evidence that argues against this (1)

   • This case carried a focal deletion inside USP8 detectable by microarray, which is the opposite of cryptic for that particular assay. Generalising 'cryptic' from USP2 to USP8 is not safe: the two have different fusion geometries and therefore different detection profiles. Meyer C et al. 2019
5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

   USP8 shows no dependency in KMT2A-rearranged leukaemia lines in either lineage half (z +0.39 myeloid, +0.24 B-lineage) in DepMap Chronos data accessed through Open Targets, in the same run in which MEN1 scored -2.29 and -2.28.

   DepMap Chronos gene-effect scores for USP2, MAML2, USP22, USP8, MEN1 and KMT2A in KMT2A-rearranged leukaemia lines, lineage-split (analyses/depmap-usp2-maml2-lineage-split) · Broad Institute DepMap; Open Targets Platform; analysis run for this graph · DepMap / Open Targets Platform 2026
   ▸What this evidence cannot show

   Eight myeloid and five B-lineage KMT2A-rearranged lines; a single line moves a mean. No KMT2A::USP8 line exists, so this measures wild-type USP8 in cells carrying other KMT2A fusions. Ensembl ENSG00000138592 was asserted to resolve to approvedSymbol USP8 by the query, which aborts on mismatch.

   ▸Evidence that argues against this (1)

   • A dependency screen run on a panel that contains none of the genotype in question is close to uninformative about that genotype. The result bounds nothing about the fusion protein's own requirement for its UCH domain. Blackburn PR et al. 2020
6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   PubMed retrieval for this partner is under-complete: (MLL OR KMT2A) AND USP8 returns exactly one record - the mate-pair case series, not the letter that named USP8 as a KMT2A partner - and the alias UBPY returns zero.

   PubMed E-utilities esearch counts for KMT2A partner symbol and alias queries, run 2026-09-19 · National Center for Biotechnology Information (NCBI), PubMed · PubMed (NCBI E-utilities) 2026
   ▸What this evidence cannot show

   A one-day snapshot of the index. The under-retrieval is caused by the founding paper being a Letter with no indexed abstract, so its gene symbols are absent from the searchable fields; this is a property of that record, not of the symbol USP8, and it would be fixed if PubMed ever indexes the letter's body.

   ▸Evidence that argues against this (1)

   • The same query shape works for USP2, which is named in the title of the same letter. So the failure is title-dependent rather than general, and an enumeration that searched titles as well as abstracts would partly recover it - which is an argument for always reading the discovery paper rather than trusting any query. Meyer C et al. 2019
7. E_q95_usp8_loc1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   USP8's NUCLEAR POOL DOES NOT SURVIVE OPENING ITS SOURCE, and the single source argues the opposite way. Human P40818 carries 'Nucleus' on one ECO:0000250 attributed to mouse Q80U87; the mouse record's 'Nucleus' carries one ECO:0000269, PubMed 16944949, which is a large-scale 14-3-3-epsilon interaction and phosphorylation screen from EMBRYONIC MURINE BRAIN, and whose only statement about USP8 is that phosphorylation at serine 680 is essential for its interaction with 14-3-3-epsilon AND FOR MAINTAINING USP8 IN THE CYTOSOL. So the sole experimental citation behind a nuclear annotation reports a mechanism for cytosolic retention.

   UniProtKB P40818 (UBP8_HUMAN) and Q80U87 (UBP8_MOUSE), SUBCELLULAR LOCATION comments with evidence codes and sources, both fetched live 2026-09-22 · UniProt Consortium · Databases 2026
   ▸What this evidence cannot show

   A statement about two database records. USP8's cytoplasmic and endosomal calls are by contrast solidly coded - four ECO:0000269s for cytoplasm and two for endosome membrane - so this is not a weak record overall; it is one weak line inside a strong one, which is the harder case to notice. Mouse Usp8 is 1,080 aa against human 1,118, a 38-residue gap, so the length assertion does discriminate this pair.

   ▸Evidence that argues against this (1)

   • THE FAIR OBJECTION IS THAT A CYTOSOLIC-RETENTION MECHANISM IMPLIES A NON-CYTOSOLIC ALTERNATIVE, AND IT DOES NOT RESCUE THE ANNOTATION. If Ser680 phosphorylation keeps USP8 in the cytosol then dephosphorylated USP8 goes somewhere else, and a curator could reasonably infer a nuclear pool from that. But an inference from a retention mechanism is ECO:0000305 reasoning presented as ECO:0000269 measurement one species removed, and the paper images nothing. Two of USP8's four CYTOPLASM citations are also non-localisation papers on inspection - an NBR1/LC3 autophagic-turnover study about NBR1, and the paediatric Cushing disease somatic-mutation genetics paper - so the record's problem is not confined to the nuclear line. This entry's pre-existing data.subcellular_localization, ['endosome', 'cytoplasm'], and its data.subnuclear_localization, 'Not known for USP8 or for KMT2A::USP8', are both CORROBORATED by this audit rather than contradicted, and were written before it. Ballif BA et al. 2006
8. E_q95_usp8_loc2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   USP8's real, well-evidenced address is the ENDOSOME and it is signal-mediated: the MIT domain of UBPY constitutes a CHMP-binding and endosomal-localisation signal required for efficient EGF-receptor degradation, and UBPY regulates endosomal ubiquitin dynamics and is essential for receptor down-regulation, with four independent ECO:0000269 citations for the cytoplasmic pool and a peripheral-membrane topology assignment for the endosomal and plasma-membrane pools.

   The MIT domain of UBPY constitutes a CHMP binding and endosomal localization signal required for efficient epidermal growth factor receptor degradation. · Row PE, Liu H, Hayes S, Welchman R, Charalabous P, Hofmann K, Clague MJ, Sanderson CM, Urbé S · J Biol Chem 2007
   PMID 17711858DOI
   ▸What this evidence cannot show

   Non-haematopoietic cell lines and receptor-trafficking assays throughout; nothing in this literature concerns leukaemia or the KMT2A::USP8 fusion. The topology calls ('peripheral membrane protein') are ECO:0000305 curator inferences rather than measurements.

   ▸Evidence that argues against this (1)

   • AN ENDOSOMAL ADDRESS IS A PROBLEM FOR THIS ENTRY'S OWN FUSION LOGIC AND THE ENTRY SHOULD SAY SO RATHER THAN LEAVE IT IMPLICIT. KMT2A::USP8 retains KMT2A's N-terminal CxxC and AT-hooks - the nuclear, CpG-targeting module - and fuses them to USP8's C-terminal UCH catalytic domain. The MIT domain that carries USP8's endosomal-localisation signal is UniProt-annotated at residues 33-116, N-terminal to the UCH domain at 777-1109, and is therefore NOT in the fusion - and that domain is explicitly dispensable for catalytic activity while being essential for endosomal localisation, so the fusion protein has lost the very determinant that puts wild-type USP8 on endosomes while keeping the determinant that puts KMT2A on chromatin. That predicts a nuclear fusion protein carrying a deubiquitinase activity whose normal substrates are endosomal, which is a substrate-relocation hypothesis nobody has tested - and it means wild-type USP8 localisation data cannot be transferred to the fusion at all. Row PE et al. 2006
9. E_q95_usp8_loc3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   Bounded 2026-09-22: USP8 has no subnuclear measurement in any system and its two apparent subnuclear hits are a registry-documented false positive. The gene clause over USP8, UBPY and KIAA0055 returns 474 records and intersects the nine-term KMT2A/MLL fusion clause at 1; the endosome sub-query returns 59 and the nuclear sub-query 26; and the subnuclear clause returns exactly 2 records - both of which are long-noncoding-RNA papers (SNHG1 in bladder cancer, SNHG12 in non-small-cell lung cancer) that match only because SNHG expands to 'small NUCLEOLAR RNA host gene'. Two of two off-topic.

   NCBI E-utilities esearch db=pubmed; every count stored with the querytranslation read back, in data.localization_bounded_absence · NCBI E-utilities · Databases 2026
   ▸What this evidence cannot show

   Titles and abstracts only; supporting figures not searched (rule 80). The 2-of-2 contamination is a live re-demonstration of symbol-collisions registry row 105 - the 'nucleolar' token's snoRNA/SNHG contamination, measured by tranche 5 at 7 of 7 on IGF2BP2 and 3 of 4 on JAK1 - on a gene in a different beat, which is what makes it a rule rather than an anecdote.

   ▸Evidence that argues against this (1)

   • THE CONTAMINATION MEASUREMENT IS WORTH MORE THAN THE ABSENCE IT BOUNDS, AND ITS DIRECTION IS THE DANGEROUS ONE. Across this tranche the same token fired three more times: 2 of 2 off-topic for USP8, roughly 19 of 20 for CDK4 (where the hits are SNHG1/4/10/12/17, SNORD9 and nucleolar-stress signalling rather than any CDK4 compartment), and it also dominates the ARID1A and WEE1 subnuclear hit lists. Against that, the SAME token on CDC73 in this tranche returned 6 of 8 genuinely on topic - because a gene with a real nucleolar literature has enough true positives to outweigh the nomenclature noise. So the registry prescription should be sharpened: the contamination rate is a function of whether the gene HAS a nucleolar literature, which means a low-count nucleolar query on a gene with no such literature is not merely unscreened, it is expected to be entirely noise. Faucz FR et al. 2017

• ← fuses withKMT2A / MLL1 — the gene that breaks

  t(11;15)(q23;q21), 3 patients in 3401; minor BCR

• ← part ofKMT2A::USP8, t(11;15)(q23;q21) - three patients, and the second instance of a deubiquitinase partner

  conserved UCH catalytic domain (USP8 777-1106); dimerisation domain probably lost

• related to →USP2 (ubiquitin carboxyl-terminal hydrolase 2) - the ninth commonest KMT2A partner and the only one that is a deubiquitinase

  the two UCH-domain KMT2A partners; different chromosomes, lengths and normal functions, same catalytic fold

• ← testsLineage-split DepMap: neither USP2 nor MAML2 is a dependency in KMT2A-rearranged lines, and USP22 is

  no dependency in either lineage half (z +0.39 / +0.24)

• related to →A partner's 'normal' compartment is usually read off a different tissue, and for at least two KMT2A partners that reading is wrong in blood cells

  USP8's nuclear pool does not survive its source: human 'Nucleus' is ECO:0000250 from mouse, and the mouse record's sole ECO:0000269 is a 14-3-3-epsilon screen from embryonic murine brain whose only USP8 statement is that Ser680 phosphorylation maintains USP8 IN THE CYTOSOL (PMID 16944949). The single experimental citation behind a nuclear call argues for cytosolic retention (Q95 2026-09-22)

• related to →Where wild-type PICALM actually is: coated pits, the Golgi, coated vesicles - and a nuclear pool nobody has explained

  The same question as PICALM's and with a sharper structural answer: USP8's endosomal-localisation signal is the N-terminal MIT domain (PMID 17711858), which is NOT retained in KMT2A::USP8, while KMT2A's CxxC and AT-hooks are. Wild-type USP8 compartment data therefore cannot be transferred to the fusion - the fusion is predicted nuclear with an endosomal enzyme's substrate preference (Q95 2026-09-22)

• related to →KMT2A::USP8, t(11;15)(q23;q21) - three patients, and the second instance of a deubiquitinase partner

  Localisation gap recorded for the fusion specifically: 1 record for USP8/UBPY/KIAA0055 against the nine-term KMT2A/MLL fusion clause and 0 with a localisation clause added; and the domain analysis says the wild-type endosomal address cannot be inherited because the MIT signal (UniProt 33-116) lies 5-prime of the retained UCH domain (777-1109) (Q95 2026-09-22)