MED19 - the Mediator subunit curated against an engineered MLL-AFF4 fusion protein, and the one REST uses to deliver G9a

HeLa cells only, and a proteomic subunit census rather than a functional experiment. IntAct curates from this paper a MED19-bait / AFF4-prey anti-tag co-immunoprecipitation (interaction EBI-394580, MI:0007, MI:0914, MI-score 0.35, spoke-expanded), so the AFF4 row in the read queue is a co-purification inside a Mediator preparation and not a measured MED19-AFF4 contact. The abstract names LCMR1 as MED19 explicitly, which is how the alias was confirmed rather than recalled.

• A MudPIT census of an immunoaffinity preparation cannot distinguish a stoichiometric subunit from a co-purifying associated factor, and the same laboratory's later work showed Mediator's association with elongation factors is docking-domain-specific rather than whole-complex - so co-purification in 2004 did not establish which subunit the contact runs through. Takahashi H et al. 2011

Retrieved 2026-09-22. MI:0914 association from an anti-tag co-IP is a co-complex statement, not a binary binding event, and the MI-score of 0.35 is low. The system is HeLa, an epithelial cervical adenocarcinoma line with no KMT2A rearrangement, and the fusion is an overexpressed construct, not an endogenous t(4;11) product - it is an MLL-AFF4 chimera, and AFF4 is the 5q31 paralogue of AFF1, not AFF1. The interactor record is untyped for organism (taxName null).

• In the same paper the functional result is assigned to MED26, not MED19: the N-terminal TFIIS-like domain of MED26 carries the overlapping docking sites for the super elongation complex and for TFIID, and MED19's appearance in the co-IP is not accompanied by any mapped MED19 interface. A co-IP hit inside a Mediator preparation is fully compatible with MED19 being a passenger of MED26's contact. Takahashi H et al. 2011

Retrieved 2026-09-22. The per-interactor counts are IntAct's own interactionCount field. A findInteractions query on EBI-2620048 returns 7 rows rather than 6 because IntAct's accession search matches by SUBSTRING and EBI-26200486 contains EBI-2620048 - the spurious row is S100B-NEDD8 from PMID 32814053, identifiable because neither participant is the queried interactor; the same collision adds a SPATS1-SARS1 row to EBI-2620075. All four constructs are overexpressed laboratory chimeras, so their partners are candidates for the endogenous fusion, not members of it.

• This project has already established that a proteome-scale AP-MS interactome of the index fusion exists in the literature and has never been curated into IntAct at all, so the fusion-construct records found here are a fragment of the fusion-side evidence rather than the whole of it, and an argument built on their scarcity understates how much fusion-side data is uncurated. Skucha A et al. 2018

Retrieved 2026-09-22. Absence of an annotated domain is absence of an ANNOTATION: A0JLT2's feature table lists only disordered regions and compositional biases, so a cryptic module is unrecorded rather than excluded. Two isoforms are annotated (A0JLT2-1, A0JLT2-2) and 244 aa is the canonical isoform; a RefSeq length may differ and was not checked. ECO:0000305 is weaker than the ECO:0000269 experimental code and weaker still than a named immunofluorescence experiment.

• UniProt's cited experimental annotations frequently point at a paper that is not about the fact being annotated, and filtering on evidence code keeps the least informative comment while discarding isoform- and fragment-scoped ones - so both the presence and the absence of a coded localisation comment are weak instruments for deciding where a protein is. UniProt Consortium 2025

The system is REST-mediated neuronal-gene silencing in non-neuronal cells, not leukaemia and not a KMT2A fusion; the mark is H3K9me2 and the writer is G9a/EHMT2, neither of which MED19 catalyses. This establishes MED19 as a docking surface for a repressive methyltransferase pathway, which is an epigenetic FUNCTION and not an epigenetic ACTIVITY. No H3K9me2 measurement under MED19 perturbation in any haematopoietic cell exists. Read as abstract plus title; the mapped interaction surfaces inside the paper were not read.

• Mediator's contribution to REST silencing has also been placed on MED12 rather than on MED19/MED26, in work from the same laboratory linking Mediator-dependent G9a recruitment to X-linked intellectual disability - so which subunit carries the G9a route is contested inside one group's own output. Ding N et al. 2008

A ranked difference between 11 and 43 lines is underpowered, and this project's own null work shows membership of the selective HALF of such a scan is a ~50% base rate; rank 273 of 17,728 is not that null, but the effect is small in absolute terms (-0.19) and MED19's pan-essentiality (-0.51) means the KMT2A-rearranged mean is a shift within a gene every line needs. The stored scan records a compute DATE and NOT a DepMap RELEASE, so this figure cannot be told from quarterly drift - reported as a defect in the stored analysis rather than patched here.

• Mediator core subunits are broadly essential as a class, so an individual core subunit in the selective half of a KMT2A scan is the expected behaviour of the class rather than a property of that subunit: this repository's module split finds 25 of 26 core subunits broadly essential with a core median rank of 445, and only MED16 clears both a selectivity and a non-pan-essentiality bar. This repository (DepMap-derived) 2026

A stored analysis computed against an earlier state of the data; this repository's staleness reporter found 25 of 36 self-measuring analyses drifted at least 15%, and the rule that followed is to quote ratios and re-run before quoting absolutes. The core/kinase contrast is quoted as a ratio-like statement for that reason. Complex Portal membership is curated, not measured per cell line.

• Paralogue structure inside the kinase module makes a module-level median misleading: MED13 ranks 13,569 while its paralogue MED13L ranks 9, a spread of 13,560 places, so one median over five subunits conceals the member whose rank would put the kinase module in the selective half. This repository (DepMap-derived) 2026

Every query is written one tag per term and never as a parenthesised group, and each querytranslation was read back - "MED19"[Title/Abstract] AND "leukemia"[Title/Abstract] and so on - all clean, with no MeSH explosion and no UID filter. A PubMed zero bounds TITLES AND ABSTRACTS only; supporting figures were not searched, and the likeliest hiding place for a MED19 measurement in a KMT2A model is a supplementary mass-spectrometry table. Absence class for these zeros is measured-not-curated at best and never-measured at worst.

• A bounded literature zero on a gene symbol is silent on whether the measurement exists in an archive: this project found an MLL fusion used as a BAIT inside a PRIDE deposition while one of its own entries asserted proteomics files had not been swept, so a symbol-level zero should be read as an index limit rather than as an absence of data. Perez-Riverol Y et al. 2022

Non-small-cell lung cancer cells, not leukaemia; a single-laboratory report in a low-impact journal with a two-hybrid origin. The relevance here is by analogy only - DEK's N-terminus is the portion retained in the t(6;9) DEK::NUP214 fusion of AML - and no DEK::NUP214 cell was tested. This record was found only because the MED19 alias LCMR1 was run as a control; the approved symbol does not reach it.

• Mapping an interaction to the N-terminal region of DEK does not establish that it survives in DEK::NUP214: the fusion replaces DEK's C-terminus with a nucleoporin FG-repeat region that changes what the retained N-terminus is presented in, and retention of a domain is not retention of a contact. Kappes F et al. 2011