The strongest published case that condensates are load-bearing in KMT2A-fusion leukemia — and it is AKAP95, not AFF1

The condensing protein is the partner AKAP95, not the fusion; the fusion contribution is 'partial co-condensation'. The fusion tested is MLL-AF9, not KMT2A::AFF1. JD-PI95 recruits HSP70, a protein-quality-control chaperone, so loss of function could reflect degradation rather than condensate dissolution — the paper's own design does not fully separate these. Published January 2026 and not yet independently replicated.

• In a closely related system, NUP98-NSD1 condensate formation was shown to be necessary but not sufficient: SMARCA5 is an essential condensate constituent yet SMARCA5 inhibition did not affect the condensates, indicating functional activity rather than condensate formation per se is what matters. Jevtic Z et al. 2022

Recombinant protein droplet assays plus transfected constructs in cell lines; no endogenous tagging, no leukemia model, and the MLL chimera tested is MLL-ENL, not MLL-AFF1. The requirement for rapid induction is inferred from perturbations that are not condensate-specific.

• The two assays this study leans on are the two that were later shown not to discriminate. Muzzopappa et al. benchmarked LLPS diagnostics against controlled in vitro and in-cell systems and found that 1,6-hexanediol treatment and classical FRAP both fail to separate phase separation from molecules binding to spatially clustered binding sites without phase-separating; preferential internal mixing measured by half-bleaching does separate them, and the interfacial barrier it probes varies substantially between real nuclear bodies. In-vitro droplets plus transfected puncta plus FRAP recovery is therefore consistent with the SEC forming a multivalent clustered binding platform on chromatin that is not a separate phase, which predicts different pharmacology - you would target the binding valency, not the phase. Muzzopappa F et al. 2022

Saccharomyces cerevisiae protein and yeast complexes; the extension to AF9 is a statement about the shared domain class, not a demonstration with human AF9, and certainly not in a leukemia cell. Cited here only to mark that the AF9 condensate idea has a structural rationale and no experiment.

• This is the evidence class the field's own critical review says is weakest, and the entry's own framing concedes the extension to human AF9 is unmade. McSwiggen, Mir, Darzacq and Tjian surveyed the in vivo LLPS literature across cellular processes and concluded that the evidence is often phenomenological and inadequate to discriminate phase separation from other mechanisms, that causal relationships and functional consequences in vivo are more elusive still, and that quantitative measurement of proteins in their ENDOGENOUS state and at physiological abundance is the missing requirement. A yeast reconstitution showing that partner binding promotes Taf14 phase separation, generalised to AF9 by domain class, is two inferential steps from any measurement in a human leukaemia cell, and should be held at the same standard this graph applies to the hexanediol literature. McSwiggen DT et al. 2019

A review, used as a map of which claims the field does and does not make. Its asymmetric treatment of ENL and AF9 is evidence about the state of the literature, not primary data.

• At physiological concentration the behaviour these claims describe has been measured directly, and it was not phase separation. 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 genomic loci, that those hubs stabilise DNA binding, recruit RNA Pol II and activate transcription, that LCD-LCD interactions are highly dynamic and selective for particular partners, and that under physiological conditions these rapid reversible interactions occur between transcription factors and the Pol II machinery WITHOUT detectable phase separation - while being differentially sensitive to hexanediols. A review that attributes 'phase separation' to ENL and withholds it from AF9 is therefore reporting which proteins have been imaged above endogenous levels, not which ones demix. Chong S et al. 2018

ABSTRACT ONLY - the PMC full text was not read, so the specific microscopy controls behind 'do not compromise subnuclear localization' were not inspected; the imaging in that era was fluorescence microscopy of transfected constructs in COS cells, which is a coarse readout of compartment and cannot exclude changes in condensate number, size or dynamics. In-vitro myeloid immortalisation of mouse progenitors with an MLL fusion construct. The result separates gross subnuclear position from transformation; it does NOT separate chromatin occupancy from transformation - on the contrary, CXXC mutants lose CpG binding, so locus-level address and transformation stay coupled.

• 'Gross subnuclear position is preserved while transformation is lost' is exactly what a composite locus-level recruitment model predicts, so this result is weaker support for separating position from function than it looks - and it points at a mechanism the entry does not name. Milne et al. identified an essential MLL1 recruitment domain spanning CXXC and the PHD fingers, controlled by direct interactions with the PAF elongation complex and H3K4me2/3, and showed that fusion proteins lacking the PHD fingers require prebinding of a wild-type MLL1 complex together with CXXC-DNA recognition for stable HoxA9 association. A CXXC point mutant that abolishes CpG binding therefore loses one contact of a multivalent recruitment module while the rest of the protein still partitions normally into the nucleus. Nothing in that requires, or excludes, a condensate. Milne TA et al. 2010