Transcription factories and Pol II clusters - no KMT2A::AFF1 imaging, condensate evidence for SEC and MLL-ENL, and 3D-FISH that actively contradicts the factory story for t(4;11)

Permeabilised HeLa cells with Br-UTP/biotin-CTP run-on labelling; permeabilisation and nucleoskeleton preparation are the same extraction conditions that generate the nuclear-matrix debate, and the site count depends on labelling efficiency. Fixed cells, no dynamics.

• The 'nucleoskeleton' to which these factories are said to be attached is the contested fraction, and the same extraction chemistry produces both. Pederson's review traces the nuclear matrix from its 1948 patent and 1974 revival, and concludes that convincing evidence for its role in gene expression has been elusive and recently further challenged: the preparations use harsh extraction under precipitation-prone conditions, the measured biophysical properties of nucleoplasm in vivo do not sit comfortably with an extensive internal filament network, and scepticism about whether the fraction represents the living nucleus persists. Permeabilised HeLa cells labelled with Br-UTP are exactly such a preparation. What survives the objection is that nascent transcription is spatially clustered; the number 2,100, the 71 nm size, and the attachment to a skeleton do not. Pederson T 2000

Mouse erythroid cells and the Hbb locus, a system with unusually strong long-range architecture; fixed-cell FISH gives population frequencies, not trajectories, and 'dynamically' is inferred from state comparisons.

• The field's own critical review declines to accept this model, and lists the reasons. Sutherland and Bickmore surveyed the evidence for and against transcription factories and state plainly that the concept 'is not universally accepted', setting the experimental support for discrete, multi-polymerase, anchored sites against the evidence that argues AGAINST such a spatially structured view of transcription, and noting that the model carries specific implications for initiation and elongation control, for gene organisation in the genome, for co-regulation and for genome instability - implications that would be wrong if the model is. Colocalisation frequencies of distal active genes in fixed erythroid nuclei are among the observations they weigh; the alternative reading, that active genes share a permissive environment without a shared machine, is not excluded by frequency data. Sutherland H et al. 2009

A tagged, overexpressed Pol II subunit in a single human cell line; cluster detection depends on the localisation-density threshold chosen, and the lifetime is a population mean over a broad distribution. Not a leukemia cell and not a KMT2A context.

• A measured cluster lifetime is a property of the imaging threshold as much as of the object, and the physiological picture that emerged from the same methodological lineage is of transient binding hubs rather than of clusters at all. Chong et al. used live-cell single-molecule imaging to show that transcription-factor low-complexity domains form local high-concentration interaction hubs at synthetic and endogenous loci, that those hubs stabilise DNA binding and recruit RNA Pol II to activate transcription, that LCD-LCD interactions are highly dynamic and partner-selective, and that under physiological conditions they occur without detectable phase separation. On that account 'Pol II cluster lifetime' is measuring the dwell of a population of rapidly exchanging molecules around a locus, so the 5.1 s figure is a summary statistic of an exchange process rather than the lifetime of a structure. Chong S et al. 2018

Four cell lines, none of them the fetal-liver haematopoietic progenitor in which infant t(4;11) actually arises, and 3D-FISH after fixation distorts nuclear geometry. Gene position is cell-type specific, so a negative in these lines does not exclude proximity in the true cell of origin. It is nonetheless the only direct measurement.

• Fixed-cell FISH in a handful of lines measures an average over a stochastic arrangement, and nuclear position is now known to be reshuffled every cycle. Kind et al. showed at single-cell resolution that only about 30% of lamina-associated domains occupy the periphery in any one nucleus, that contacts are intermittent at the molecular level, and that positioning is NOT detectably inherited through mitosis but is stochastically reshuffled - with contact linked to repression and H3K9me2 cell by cell. If large-scale radial position is re-randomised each division, then an ordering of inter-locus distances measured across four fixed populations describes probability distributions with wide overlap, and a difference in mean distance can coexist with frequent proximity of the pair that translocates. The MLL/AF4 versus MLL/ENL discrepancy this item reports is the expected shape of that measurement, not necessarily a finding. Kind J et al. 2013

A review advancing a model; no primary imaging data supporting factory co-occupancy of MLL and a partner gene is presented in it or, as far as this search found, anywhere. Cited to document that the idea exists in the literature, not as evidence for it.

• 3D-FISH of the three loci gives the wrong proximity ordering to explain partner frequency. Gue M et al. 2006

Recombinant droplet assays plus transfected constructs in cell lines; no endogenous tagging, no leukaemia model, and the MLL chimera tested is MLL-ENL, not KMT2A::AFF1. Droplets were not co-imaged with RNA Pol II clusters, so this is a condensate result that bears on the factory question only if the two names describe one object.

• Whether SEC droplets and Pol II clusters are the same object cannot be settled by the assays used to call the droplets a phase. Muzzopappa et al. benchmarked LLPS diagnostics in vitro and in living cells and found that 1,6-hexanediol and classical FRAP fail to distinguish phase separation from binding to spatially clustered binding sites, that preferential internal mixing under half-bleaching does distinguish them, and that the interfacial barrier strength measured that way differs greatly between real nuclear bodies - so 'condensate' is neither binary nor uniform. Recombinant droplet assays plus transfected puncta therefore leave open that the SEC forms a clustered multivalent binding platform, which would make the 'droplet' and the 'Pol II cluster' the same phenomenon described twice, and would dissolve rather than answer the factory question. Muzzopappa F et al. 2022

ENL point mutants found in Wilms tumour, not a KMT2A fusion; condensates are scored by imaging of tagged protein and are sensitive to expression level, as the paper itself shows (above-endogenous expression gives non-functional condensates). Pol II clustering was not the readout.

• Condensate scoring by imaging of tagged protein is sensitive to fixation as well as to expression level, and the artefact is not a uniform inflation. Irgen-Gioro et al. compared the same cells live and after paraformaldehyde fixation and found that fixation can both enhance and diminish apparent phase-separation behaviour, that for specific proteins droplet-like puncta appear in fixed cells that show none when live, that co-fixation with glycine reverses the direction of the artefact, and that the outcome depends on the balance between each protein's interaction dynamics and the relative rates at which the proteins are fixed. Wild-type and hotspot-mutant ENL differ precisely in their multivalent interaction dynamics, so a mutant-versus-wild-type comparison of condensate formation in fixed cells is confounded by the variable under study - and the paper's own observation that above-endogenous expression yields non-functional condensates shows the two artefacts can act together. Irgen-Gioro S et al. 2022