CTCF - named in the id of an existing entry in this graph and having no entry of its own, and the protein whose binding sites break preferentially in KMT2A-rearranged patients

A naming fact. gene_exact:CTCF AND organism_id:9606 AND reviewed:true returned exactly one row with gene_primary CTCF. The retrieval hazard here is in PubMed rather than UniProt: the CTCF query translation expands to include 'ccn2 protein human', so a raw count on this symbol contains connective-tissue-growth-factor papers and must be title-checked.

• The paralogue CTCFL (BORIS) is a testis-specific CTCF-like protein that binds an overlapping site repertoire and is re-expressed in some cancers. It was not resolved in this session and nothing in this entry covers it; a break-mapping result attributed to CTCF sites is attributed to the motif, which CTCFL also reads. UniProt Consortium 2026

Three KMT2A-rearranged remission patients versus three non-fusion AML patients, adult, peripheral blood mononuclear cells cultured ex vivo. n=3 per arm is the single largest limitation and every number above should be read with it. The comparator is other AML patients, not healthy donors, so the result is a between-AML-subtype difference. The authors report no significant difference in genome-wide median breakage between any of the samples, so the effect is locus-specific and not a global instability phenotype. Full text read via the PMC HTML route after Europe PMC fullTextXML returned HTTP 500 and efetch db=pmc returned HTTP 429.

• A remission sample from a patient who already has KMT2A-rearranged AML cannot distinguish an inherited or pre-existing fragility from a consequence of the disease or its treatment - and every one of these patients had been treated. The design cannot deliver the susceptibility interpretation the title proposes without a pre-diagnosis or healthy-relative comparison, which does not exist. Atkin ND et al. 2021

Three patients, each with a different partner, so the partner-side analysis has n=1 per partner and the authors say so explicitly: they note it is conceivable that break-prone partner regions were not detected across three patients with three different rearrangements. This is a weak negative and is recorded as one. Read from the article full text, not the abstract.

• The negative is bounded by a filtering choice rather than by biology: the 34 regions are those passing a stringent KMT2A-to-non-fusion ratio filter, so a partner region with real but smaller preferential breakage would be excluded by construction. The absence of partners from a stringently filtered list is not the absence of partner-side fragility. Gothe HJ et al. 2019

Class-level genome-wide enrichment; it does not name the KMT2A BCR as an anchor. The polymorphism-rewiring result is the strongest causal element and it is about anchor position in general, not about this locus.

• Canela 2017 describes loop-anchor breakage as largely transcription-, replication- and cell-type-independent. If that holds at the KMT2A BCR then none of the cell-type specificity of KMT2A-rearranged leukaemia is explained by breakage geography, and the whole CTCF story explains position while explaining nothing about why this happens in a fetal progenitor. Gothe HJ et al. 2019

The predictive claim is quantitative modelling on genome-wide data in cultured cells. Note that this result attributes the requirement to COHESIN binding specifically, not to CTCF - which is the distinction the Atkin site sets do not separate.

• CTCF and cohesin occupancy are collinear at loop anchors, so a study that finds cohesin-dependence and a study that finds CTCF-site enrichment are not independent observations. Attributing the fragility to CTCF rather than to cohesin, or vice versa, requires a degron of one with the other intact, and no such experiment exists at the KMT2A locus. Canela A et al. 2017

Established as a statement about the record. Three of the four subnuclear statements in the note are HEDGED ('may translocate', 'may be preferentially excluded') and carry no attribution at all, so they cannot be cited for anything; they are recorded because rule 91 item 2 says to read the uncoded comments rather than discard them, and because a hedged curator note is a pointer to an unread literature.

• The reflex objection - that reading uncoded notes imports low-grade information into a graph that should hold measurements - has force, and this entry should say why it is overridden here rather than quietly importing them. In CTCF's case the uncoded note is the ONLY place in the entire record where the two facts a leukaemia graph most wants appear at all: a differentiation-dependent nucleolar shift, in a disease defined by differentiation block, and heterochromatin exclusion in interphase. Discarding the note would leave the entry with a nucleoplasm call whose citation is an endometrial-cancer genetics paper and nothing else. The correct handling is the one taken: record them, mark them uncoded and hedged, cite nothing to them, and convert the differentiation claim into an open question rather than a finding. Marshall AD et al. 2017

Vertebrate insulator biology (the chicken beta-globin HS4 insulator system this group works in, worked in cultured cells), not haematopoietic and not fusion-bearing; abstract read, full text not read, so the cell system is not named in the claim (rule 73). This is a statement about where CTCF puts DNA, which is one step removed from where CTCF itself is concentrated - the nucleolar tethering is inferred from the DNA's position plus the nucleophosmin interaction rather than from a CTCF immunofluorescence with a nucleolar marker. 2004, and this beat found no replication of the tethering result in a modern imaging system.

• There is a serious argument that this should not count as a CTCF localisation at all, and it deserves stating because it is the argument this entry uses against the ChIP-seq literature. What is measured is the position of a transfected INSULATOR SEQUENCE, plus a biochemical CTCF-nucleophosmin association - so a critic can say the DNA is at the nucleolus and CTCF is merely required for it to get there, which is a tethering function rather than a residence. Two things make it stronger than a ChIP-seq address nonetheless: nucleophosmin is a compartment marker rather than a sequence feature, so the readout is positional in a way a motif map is not; and the tether is CTCF-DEPENDENT, meaning removing the protein moves the DNA, which no occupancy map can demonstrate. It remains an inference about CTCF's own position, and the entry grades it as the best available rather than as decisive. UniProt Consortium 2026

Abstract read, full text not read. Cultured human cells, not haematopoietic and not fusion-bearing. This is the source of UniProt's 'Chromosome, centromere' value and, unlike the nucleoplasm attribution, it IS a paper about where CTCF is. The finding is mitosis-specific by construction, so it says nothing about interphase distribution, and a single-state localisation record for CTCF would therefore be wrong - which is why this entry carries the state alongside the compartment.

• The pericentromeric result could be read as undercutting this entry's own central distinction, and that objection should be faced rather than buried. If CTCF's centromeric position is defined by 'CTCF sites in pericentric DNA', then this too is a sequence address and the line between occupancy and compartment starts to look arbitrary. The line survives, but it moves: what makes this a subnuclear result is not that CTCF binds particular sequences, it is that those sequences are CONCENTRATED IN ONE CYTOLOGICALLY DEFINED STRUCTURE and that CTCF's occupancy there is restricted to one phase of the cell cycle and uses a different part of the protein (C-terminal rather than the usual fingers). A genome-wide loop-anchor map has none of those properties - the sites are dispersed, the timing is not restricted, and the same domain does the binding. So the distinction is about whether the DNA addresses converge on a structure, not about ChIP versus microscopy. Yusufzai TM et al. 2004

Bounded by the ABSTRACT: a mutant-versus-wild-type immunofluorescence almost certainly exists in that paper's figures - it is the standard control when arguing a missense mutation is loss-of-function - and this beat did not read the full text to find it. So the correct statement is 'the paper is not ABOUT CTCF's nucleoplasmic localisation', not 'the paper contains no such experiment'. That is a weaker claim than the equivalent one made for BICRA in this tranche, where the full text was read and the absence is measured.

• One could argue this observation is unfair and unimportant: CTCF being nucleoplasmic is not in doubt by anyone, so who cares which paper UniProt hangs it on. The reason it is worth recording is not doubt about the conclusion but the pattern, and the pattern now has a measured shape in this graph. Rule 91 found four of six proteins where the ECO:0000269 localisation citation points at the wrong paper; this tranche alone adds BICRA (full text read, zero localisation experiments), PARP1's nucleolus (abstract reports nuclear PERIPHERY) and now CTCF. That is a high enough hit rate that 'UniProt has an experimental attribution' should be treated as a pointer to check rather than as a fact, and any coverage metric that counts ECO:0000269 rows as measurements is counting something other than measurements. UniProt Consortium 2026

Abstract-level: the full text was not read, so the fraction of CTCF in the speckle pool, the identity of the complexes, the stressors used and the definition of the neuronal commitment stage are not established here. Mammary epithelial and neuronal systems only - nothing haematopoietic, nothing in a KMT2A-rearranged cell, so this does not close this entry's bounded absence. 'Largely unchanged' is the authors' own hedge about the ChIP-seq and is not a null result at any specific locus, so it cannot be used to claim CTCF occupancy is stable at KMT2A. The word 'stress' covers diverse treatments that are not enumerated in the abstract and is NOT established to include doxorubicin.

• This constrains rather than supports this graph's CTCF model, and the constraint is uncomfortable. PMID 37788669 reports that doxorubicin LOWERS CTCF occupancy at the KMT2A element; this paper reports that stress leaves genomic CTCF binding largely unchanged while destroying a speckle-localised protein pool. Both can be true only if the KMT2A effect is site-selective against a globally stable background - which is a stronger and more specific claim than the graph has been making, and nobody has tested whether the KMT2A site is one of the selective ones.
• It also raises a rival mechanism for anything this graph attributes to CTCF occupancy loss under stress. If the stress-sensitive species is defined by RNA interactions and speckle residence rather than by DNA binding, then a phenotype following stress may run through CTCF's RNA-processing role and not through insulation at all - and no experiment in this graph distinguishes those two routes.