WDR33 (Q9C0J8) - the subunit that reads AAUAAA, in the pan-MLL-fusion core, and the plant homolog of which physically links 3' processing to H3K4 demethylation

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 WDR33 the MLL-AF4 total (9 spectra over two replicates) is rank 7 of the seven baits, against a maximum of 45 under another bait. All seven per-bait values, p (replicate spectral counts): MLL-AF1p 0.0046 (9|4); MLL-AF4 0.0214 (6|3); MLL-AF9 0.0001 (16|17); MLL-CBP 0.0006 (7|38); MLL-EEN 0.0005 (19|17); MLL-ENL 0.0000 (18|18); MLL-GAS7 0.0002 (22|16). SPECIFIC TO THIS PROTEIN: five of the seven baits are at or below 1e-3 for WDR33, so the pan-fusion pattern is among the better-supported in the 42; the MLL-AF4 value of 0.0214 is the second-weakest of its seven and its 9 spectra are the shallowest. 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.

• 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

In vitro and in vivo assays in human cells, 2014; a biochemical and crosslinking result, not a structure. The paper's framing is explicitly a correction of the prior CPSF160 assumption. Not a haematopoietic or fusion system. Read from the abstract.

• A published biophysical re-examination of PAS recognition (PMID 31462423, RNA 2019) revisits the affinities involved, and a later structure shows mPSF also recognises the variant AUUAAA signal (PMID 36130077), so 'the AAUAAA reader' is a simplification of a machine with measurable sequence tolerance. Sun Y et al. 2018

A reconstituted quaternary complex with a short RNA, not a cellular measurement; the construct boundaries and any engineered changes were not checked against the deposit, and by this project's standing rule residue spans should not be read off a deposit without aligning to a same-day UniProt sequence, which was not done here because no residue numbers are asserted in this entry. Not a haematopoietic or fusion system. Read from the abstract.

• CPSF160/CPSF1 is an essential scaffold in this structure and is ABSENT from the Skucha 960-protein network entirely, at any n - checked by this beat from the xlsx, 2026-09-21. So the assembly that this structure describes is not the assembly that co-purified with the fusions, and the entry should not be read as saying the AAUAAA-recognition machine came along intact. Skucha A et al. 2018

PLANT. Arabidopsis thaliana, at one locus (FLC), with an antisense transcript that has no established counterpart in this disease. No human, no haematopoietic cell, no fusion. This item is cited for ONE narrow thing - that a physical link between the AAUAAA reader and an H3K4-demethylating module exists somewhere in biology - and for nothing else. The mark in question is H3K4me1, not the H3K4me3 that KMT2A deposits. Read from the abstract. Treating this as evidence about KMT2A::AFF1 would be exactly the cross-system generalisation this project's standing rules exist to prevent.

• FLD is an LSD1 homolog acting on H3K4me1 at a gene body; KMT2A's activity is H3K4me3 at promoters, and the KMT2A fusion has lost the SET domain altogether, so the enzymatic arm of any analogy is missing on the index fusion. The analogy therefore cannot be carried past 'a 3'-processing factor can sit in a physical chain with an H3K4-modifying module'. Skucha A et al. 2018

Wild-type MLL, not a fusion, and not the index fusion. SYMPK is not in the Skucha 960 network at any n (this beat, from the xlsx, 2026-09-21). Read from the abstract. Cited here as the only published physical link between 3'-end processing and MLL that this beat could find, established by running the pathway-name query rather than the symbol query.

• The reported direction is repressive, so if it generalises, 3'-end factors in a fusion purification would be brakes rather than collaborators - which inverts the therapeutic reading of the whole CPSF row. Skucha A et al. 2018