LUZP1 / fimbacin (Q86V48) - the weakest index-fusion call of the nine, and the strongest non-index call: p = 0.127 against MLL-AF4 and 2.6e-10 against MLL-ENL

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 after a single STREP-Tactin step, so co-purification is NOT evidence of a direct contact and is not evidence that the two proteins are ever in the same complex. INCLUSION RULE: the paper kept a fixed number of top-ranked interactions per bait rather than applying a p threshold. Methods, read verbatim from the PMC full text by this beat on 2026-09-21: "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: within the MLL-AF4 column 186 of 291 listed p-values (63.9%) exceed 0.05, and that column's maximum p is 0.175. BAIT DEPTH: for all nine proteins in this tranche the MLL-AF4 spectral total is rank 7 of the seven baits - measured protein by protein - so a low MLL-AF4 count is a bait property (transgene expression, purification depth, or the GLM normalisation) and must not be read as a weaker association than the same protein's count under another bait. HOW TO READ THE NEGATIVE CONTROL, measured by this beat on 2026-09-21 over all 960 rows of MOESM4: the Spearman correlation between a protein's neg_ctrl spectral total and its maximum per-bait spectral total is +0.461, and the median maximum per-bait spectral count is 42 for the 145 proteins (15.1%) with a non-zero negative control against 8 for the 815 with 0|0. A dirty negative control in this table is therefore substantially a statement about protein ABUNDANCE, not about specificity, and a clean 0|0 is substantially a statement that two control replicates were not deep enough to see the protein at all. Two consequences. (1) PSIP1/LEDGF - the one pan-fusion partner with a solved co-crystal against the KMT2A N-terminus - carries 0|5, the second-dirtiest control of the 42, so a non-zero control cannot be used to call a protein a contaminant. (2) A 0|0 control is not purity; it is an unestimated baseline. The paper's GLM estimates each prey's baseline abundance A0,j from the control purifications, so a prey the controls never saw has its baseline set by a floor rather than by data, which is the direction that inflates apparent specificity. ABOVE THE CONVENTIONAL LINE, AND THAT IS NOT A DISQUALIFIER HERE. This protein's MLL-AF4 p-value is 0.127, above 0.05. Three things make that interpretable rather than fatal. (1) The table is RANK-SELECTED, not threshold-selected - the Methods read 'The 300 most significant interactions per bait were retained', 186 of the 291 listed MLL-AF4 p-values (63.9%) exceed 0.05 and the column's maximum is 0.175 - so 0.05 was never the inclusion criterion and applying it retrospectively imposes a cutoff the experiment did not use. (2) Measured by this beat on 2026-09-21, the three WEAKEST MLL-AF4 calls in the entire pan-fusion 42 - SMC3 0.174, NOLC1 0.134, TRIM28 0.129 - ALL already have entries in this graph, and gene-nolc1 and gene-trim28 carry confidence_overall 'weak' with those exact p-values recorded. TRIM28's 0.129 is WORSE than this protein's 0.127. (3) The index bait is the shallowest of the seven for all nine proteins in this tranche. SPECIFIC TO THIS PROTEIN, measured by this beat from the cached MOESM4 xlsx on 2026-09-21: the spread between the weakest and strongest bait for this one prey is EIGHT ORDERS OF MAGNITUDE (MLL-AF4 0.127, 6 spectra; MLL-ENL 2.63e-10, 69 spectra), the largest in this tranche, and it is the cleanest single demonstration that index-column weakness is a bait property. Four of the seven baits are below 0.005, so unlike DHX16 this is NOT weak everywhere. MODULE: LIMA1/EPLIN, LUZP1's named direct partner, IS present at n = 2 (neg 0|0, no MLL-AF4 value); FLNA n = 1; ACTR2, ACTR3, ARPC2 n = 1-2; LUZP2, MYO5A and WASL absent. LOCALISATION: UniProt Q86V48's 'Nucleus' line is ECO:0000250 by similarity to mouse Q9ESV1 - there is NO human experimental nuclear annotation, while there are eight experimental cytoplasmic/cytoskeletal ones. SELECTIVITY: not named among the paper's controls (RFC1, SF3A3, CCT3 non-selective; MEN1 selective), not known to be among the unpublished 40 re-screened. Recorded as unknown. run.py re-run 2026-09-21, both controls PASS.

• The clean 0|0 control is the base rate for this set - 32 of 42 pan-fusion proteins have one - and over the 960 rows control counts track abundance at Spearman +0.461. Coiled-coil, elongated, self-associating proteins such as LUZP1 are a recognised hazard in affinity purification, and the CRAPome authors state that a study's own limited negative controls do not capture a complete background protein set. No contaminant-repository frequency for Q86V48 was retrieved by this beat and it is the cheapest external check available. IntAct has never curated this dataset. Mellacheruvu D et al. 2013
• The eight-order-of-magnitude bait spread cuts both ways. It shows the index column's weakness is not about LUZP1 - but a prey whose recovery varies that much between baits purified in the same laboratory from the same cell line is also a prey whose recovery is dominated by something other than its affinity for the shared KMT2A N-terminus. Note that MLL-ENL, the bait with the 2.63e-10 row, is also the deepest bait for ACTB (316 spectra), VIM (73) and MAP7D1 (55) - the whole cytoskeletal group peaks on the same bait, which is a pattern about the purification and not about any one protein. The raw spectra at PXD009338 have not been reanalysed here. ProteomeXchange/PRIDE 2018

Human cultured cells (RPE-1-type ciliogenesis systems), proximity-dependent biotinylation plus depletion phenotypes; not haematopoietic, not a fusion system, and nothing in it is nuclear. It is cited to establish what LUZP1 is and where it lives, and for one cross-check: EPLIN/LIMA1, the partner named here, is present in the Skucha 960 at n = 2, which makes LUZP1 the only protein in this tranche whose named direct partner appears in the network. Read from the abstract.

• EPLIN's presence is weak corroboration at best: n = 2 of 7 baits, no MLL-AF4 value at all, and the partner is itself an abundant actin-associated protein with the same background profile. A shared background source produces exactly this pattern - two members of one cytoskeletal neighbourhood, both weak, neither in the index column - and nothing here distinguishes that from a real co-recovery. Skucha A et al. 2018

In vitro biochemistry, proteomics plus in silico motif screening, and overexpression in tissue culture cells; the octamer estimate is from size-exclusion chromatography, which reports hydrodynamic behaviour rather than stoichiometry for an elongated coiled-coil protein. Not haematopoietic, not nuclear, not a fusion system. Cited here for one specific property that bears directly on the AP-MS question - LUZP1 self-associates through an N-terminal coiled-coil and cross-links actin - which is the biophysical profile most likely to produce non-specific co-purification. Read from the abstract.

• This is the most damaging citation in the entry for the interaction claim, and it is placed here deliberately: an elongated, self-oligomerising, actin-cross-linking protein is close to a worst case for affinity purification specificity. Against that, the one thing it cannot explain is the negative control - if LUZP1 were co-sedimenting non-specifically it should appear in the mock purifications, and it gives 0|0. That is weak evidence, because 32 of the 42 pan-fusion proteins are also 0|0 and the control tracks abundance, but it is not nothing. Skucha A et al. 2018