PHF1 (PCL1) - the PRC2 accessory subunit whose Tudor domain reads H3K36me3, and the 3' acceptor shared by three different NuA4-donor fusions in two different tumour types

Fetched live 2026-09-23 via gene_exact:PHF1 AND organism_id:9606 AND reviewed:true, exactly ONE row. RULE 91 NOTE: NONE of the location statements on this record carries an evidence code - not the nucleus, not the centrosome, not the DSB localisation, not the NEK6 co-localisation. Every one is uncoded. Rule 91 item 2 says to read the uncoded comments and record them AS uncoded rather than discard them, which is what is done here: the centrosomal and DSB localisations are recorded as database statements whose primary sources this beat did NOT resolve, and no confidence above 'established for the accession itself' is claimed for them.

• Treating every uncoded UniProt statement as unestablished cuts the wrong way as often as the right one. Rule 91 item 2 says to READ the uncoded comments because they carry the condition-scoped statements that are usually the interesting ones - and PHF1's centrosome, DSB and NEK6 annotations are exactly that kind of statement. They may well rest on good papers this beat did not resolve. The honest label is 'source not checked', not 'probably not real', and this entry's framing risks the second reading.

Fetched live 2026-09-23; exactly one document returned. The alias MTF2L2 is a retrieval hazard: MTF2 is PCL2, a DIFFERENT Polycomblike paralogue with its own gene entry in HGNC, and a substring search for MTF2 reaches PHF1's alias. PHF1, MTF2 and PHF19 are the three mammalian Polycomblike proteins and their PCL1/PCL2/PCL3 numbering does not match their PHF numbering, which is exactly the shape of rule 66. No row exists for any of these in data/symbol-collisions.json today.

• The PCL1/PCL2/PCL3 versus PHF1/MTF2/PHF19 mapping asserted here is read off PHF1's own alias list, and this beat did NOT fetch the HGNC records for MTF2 or PHF19 to confirm it. That is precisely the shape of error rule 66 exists for - MLL2 and MLL4 each naming two genes with the protein names swapped - and an alias field is a weaker source for a family numbering than the family's own records. The mapping should be verified before anyone relies on it.

Abstract read this session; full text not read. The in vivo arm is knockdown, with the usual exposure to indirect effects, and the in vitro stimulation arm uses reconstituted components. Note the pleasing but unexamined coincidence that PHF1's PRC2 target loci here are the HoxA loci, which are also where NUP98::NSD1 acts in gene-nsd1 E4 and where the KMT2A programme lives - three different mechanisms converging on one locus cluster is more likely to reflect how much of chromatin biology is done at HoxA than to be a connection, and no connection is asserted.

• PHF1 has two paralogues, MTF2 and PHF19, that are also PRC2 Polycomblike accessories with overlapping function, so a PHF1 knockdown phenotype is compatible with a total-Polycomblike-dosage effect rather than with a PHF1-specific requirement. The in vitro arm is stronger on this point than the in vivo arm, and the global H3K27me3 reduction from a single accessory-subunit knockdown is a large effect for a subunit with redundancy - which is a reason to want the experiment repeated with paralogue controls.

Abstract read this session; full text not read, and the author list is the four names in the fetched efetch record plus 'et al.'. This is peptide-level binding plus structural analysis of isolated domains - the physiological consequence of the H3K36me3 contact for PRC2 targeting is the authors' proposal, not their measurement. The paper states the same is 'likely' for MTF2, which is an inference and is recorded as such. No PDB id is stated here because none was fetched in this beat.

• Peptide binding by an isolated Tudor domain is a result that frequently fails to survive nucleosome context, and the paper's own physiological claim is a proposal rather than a measurement. The abstract also extends the finding to MTF2 as 'likely', which is an inference presented alongside data. Until a nucleosome-context measurement exists, 'PHF1 reads H3K36me3' should be held as a domain property rather than as a targeting mechanism - and this entry's whole reader/writer-inversion section rests on it.

Abstract read this session; full text not read. THREE TUMOURS, typed by PubMed as 'Case Reports; Journal Article'. This is the founding description of both JAZF1-PHF1 and EPC1-PHF1 and its n should be kept in mind whenever the ESS family is discussed as though well enumerated.

• Three tumours, typed by PubMed as Case Reports. Both founding PHF1 fusions rest on this n, and one of the two partners - EPC1 - is described in that paper as 'hitherto not associated with neoplasia', so the assignment was novel in both directions at once. The family membership this entry and concept-low-grade-ess-is-a-nua4-to-polycomb-fusion-family both use is built on case series of this size throughout.

Abstract read this session; full text not read, author list is the four in the fetched record plus 'et al.'. A SINGLE TUMOUR. The fusion protein is PREDICTED from the sequenced junction - 5'-RACE, RT-PCR and sequencing - and no protein was detected or characterised. The authors' own summary of the mechanism is explicitly speculative: 'Although the specific functions of the MEAF6 and PHF1 proteins and why they are targeted by a neoplasia-specific gene fusion are not directly apparent'. This is the single most valuable detail for this graph, because the junction is exon 5 of MEAF6 to exon 2 of PHF1, i.e. essentially ALL of PHF1 is retained - so a PHF1 fusion is a PHF1 protein with an N-terminal extension, not a truncated PHF1.

• One tumour, and the 750-residue protein was never observed - it is predicted from a sequenced cDNA junction. No protein was detected, no function was tested, and the authors explicitly say the reason these two genes are targeted is 'not directly apparent'. The exon-5-to-exon-2 architecture that this entry treats as its most valuable structural fact is therefore a property of a transcript, and whether the predicted protein is made at all is unmeasured.

Abstract read this session; full text not read, author list is the four in the fetched record plus 'et al.'. n = 39 OFMT, screened by FISH with RNA-seq on two index cases and validation by FISH and RT-PCR - so the fusion-partner proportions are from a referral-centre cohort of a rare tumour and are not population frequencies. THE EP400 ASSIGNMENT IS THE LOAD-BEARING PART FOR THIS GRAPH and this beat did NOT independently verify that EP400 is a NuA4/TIP60 subunit against a primary source; gene-ep400 exists in this corpus and should be read before that inference is used. The claim as written here reports the fusion frequencies, which is what the abstract states.

• The EP400-is-NuA4 inference that makes this the strongest independent support for the family rule was NOT verified against a primary source in this beat. gene-ep400 exists in this corpus and was not read. If EP400's complex assignment turns out to be more complicated than 'NuA4/TIP60 core ATPase' - and multi-complex membership is exactly what gene-meaf6 documents for another member of this family - then the independence of this support evaporates. The frequencies in this claim are from the abstract and are solid; the inference drawn from them is not yet.

Fetched live 2026-09-23; persisted in analyses/q126-the-partners-named-but-never-written/depmap-essentiality.json (rule 90). RULE 83: meta.dataVersion {year 26, month 06, iteration null} is the OPEN TARGETS platform release, not a DepMap release. No ESS or OFMT line is known to this beat to be in the panel. A near-zero score here is unsurprising for an accessory subunit with two paralogues (MTF2, PHF19) that could buffer it, and paralogue buffering was NOT tested in this beat.

• A near-zero dependency score cannot distinguish 'not required' from 'redundant', and PHF1 has two paralogues that were not tested. It also cannot see a dosage or differentiation phenotype of the kind PRC2 accessories are expected to have, because the readout is proliferation in an immortalised line. Reading 0.16% below -0.5 as evidence that PHF1 does little would be the same category error this beat flags for NSD1.