AKAP95 binds the Pol II CTD directly and nascent RNA gates the condensate - a feed-forward loop where transcription recruits the factor that sustains transcription
also called AKAP95 Pol II CTD condensate · RNA-gated AKAP95 condensation
AKAP95 sticks directly to the tail of RNA polymerase, and the RNA the polymerase is making feeds back to make the droplet bigger. More transcription makes more RNA, which recruits more AKAP95, which supports more transcription.
The same thing, for a scientist
**The mechanism underneath the leukaemia phenotype, and the cleanest experiment in the 2026 paper.** Purified recombinant AKAP95 binds purified **Pol II CTD (aa 1356-1767, 23 heptapeptide repeats)** directly by pull-down, and the two co-condense in vitro [E1]. Two separation-of-function mutants dissect it. The **PS-mutant** (six tyrosines in AKAP95 aa 101-210 to serine) condenses poorly alone, co-condenses with CTD weakly, and is largely diffuse in the nucleus. The **ZF-mutant** (zinc fingers, which are the RNA-binding module) condenses **normally in the absence of RNA** and forms discrete nuclear foci, but **adding RNA fails to enlarge either its own condensates or its co-condensates with Pol II CTD**, whereas RNA enlarges both for wild type [E1]. In cells, both mutants are more readily released to the cytoplasm on hypotonic treatment and are already released from chromatin before MNase; and the ZF-mutant's chromatin partition, uniquely, **does not respond to RNase** [E1]. Functionally, both mutants raise **total Pol II at promoters** while lowering **Ser2P across gene bodies** and reducing PRO-seq on-going transcription - a **pause-release** defect [E1]. **And the dose-response is the best internal control in the study: the magnitude of the Ser2P loss correlates positively with AKAP95 ChIP density at the gene** [E1].
Why the ZF-mutant is the informative one
The PS-mutant tells you that a condensate is needed, which is the expected result and the weakest kind.
The ZF-mutant is the one that distinguishes hypotheses. It still phase-separates. It still makes foci. What it cannot do is respond to RNA - its condensates do not grow when RNA is added, its co-condensates with Pol II CTD do not grow, its chromatin partition is RNase-insensitive, and its foci do not overlap Pol II Ser2P. So the requirement is not 'AKAP95 must form a droplet'; it is 'AKAP95's droplet must be tuned by nascent RNA'.
That converts a static scaffolding model into a feed-forward loop: the fusion drives transcription; transcription makes nascent RNA; nascent RNA enlarges the AKAP95/Pol II CTD condensate at that locus; the condensate supports pause release; more transcription. Actinomycin D for 4 h impairs AKAP95 binding to its genomic targets, which is the loop's own closure test and it passes 1.
This is also why the requirement is fusion-selective without AKAP95 being fusion-specific. A feed-forward loop amplifies wherever transcription is already high and abnormal. It does not need to recognise the fusion; it needs the fusion to have made the locus loud.
What is not shown, and the one number that should not be over-read
- All of the mechanism is in HeLa. The fractionation, the Pol II ChIP-seq, the PRO-seq, the CLIP and the in vitro biochemistry are HeLa or purified protein. The leukaemia arm is a proliferation rescue in THP-1 and MOLM-13. No Pol II ChIP-seq or PRO-seq in any leukaemia line appears in the main text except the JD-PI95 PRO-seq in THP-1, whose metagene comparison against the scrambled control was not significant.
- The mutants are pleiotropic. The ZF-mutant loses RNA binding, but the zinc fingers are also the GC-rich-DNA-binding module (SELEX, from the nucleolar paper) and the chromatin-binding region 387-450 overlaps ZF1. A ZF mutant that cannot bind DNA either is not a clean RNA-binding mutant.
- The PS-mutant's six tyrosines are in 101-210, inside the DPY30-interaction region as well as inside IDR1, so the PS-mutant is not route-specific - see
concept-the-two-akap95-routes-into-the-kmt2a-axis. - n = 414 is the PRO-seq target set around TSSs; it is a subset of the 5,656 ChIP sites and the selection criterion is not stated in the main text.
| mechanism class | RNA-gated transcriptional condensate; pause release |
|---|---|
| readouts | total Pol II ChIP-seq at promoters (up in mutants), Pol II Ser2P across gene bodies (down in mutants), PRO-seq on-going transcription (down), cell fractionation +/- RNase and MNase, in vitro droplet size +/- RNA |
| shown for fusion | NO for the mechanism itself - every Pol II, PRO-seq, CLIP and fractionation experiment is in HeLa or purified protein. The leukaemia link is that the same two mutants fail to rescue AKAP95-knockdown THP-1 and MOLM-13 growth. |
direct partners
[ "Pol II CTD (RPB1 aa1356-1767)", "MLL1 translocated fragment", "DPY30" ]
modules
{
"phase_separation": "IDR1 101-390; six tyrosines in 101-210 (PS-mutant)",
"rna_binding": "C2H2 zinc fingers 392-414 and 481-504 (ZF-mutant) - also the GC-rich DNA-binding module, so not clean"
}measurements — the numbers behind the chart
[
{
"quantity": "count",
"subject": "AKAP95 target genes in the PRO-seq TSS metagene",
"value": 414,
"units": "genes",
"n": 5656,
"universe": "a subset of the 5,656 AKAP95 ChIP-seq sites; the selection criterion is not stated in the main text, PMID 41501053 Fig 5e",
"null": null,
"note": "7.3% of the ChIP sites; the subsetting rule is the unstated part"
},
{
"quantity": "share",
"subject": "AKAP95 CLIP top RNA-binding sites located in introns",
"value": 61.26,
"units": "percent",
"n": null,
"universe": "AKAP95 CLIP in HeLa, PMID 41501053 Supplementary Fig 7e; promoter 13.47%, exon 9.33%",
"null": null,
"note": "intron-dominant, consistent with the 2016 splicing characterisation and in tension with a purely transcriptional model"
},
{
"quantity": "share",
"subject": "AKAP95 CLIP RNA targets that also carry AKAP95 ChIP binding at the DNA level",
"value": 25,
"units": "percent",
"n": null,
"universe": "PMID 41501053 Supplementary Fig 7f, reported as 'a quarter'",
"null": null,
"note": "so three quarters of the RNA binding happens where there is no detected DNA binding"
},
{
"quantity": "length",
"subject": "Pol II CTD construct used for the direct binding and co-condensation experiments",
"value": 412,
"units": "aa",
"n": null,
"universe": "RPB1 aa1356-1767, stated to include 23 heptapeptide repeats, PMID 41501053 Methods",
"null": null,
"note": "a partial CTD; human RPB1 has 52 heptads, so less than half the repeat array"
}
]- A feed-forward loop driven by nascent RNA and Pol II CTD is not a leukaemia mechanism; it is a transcription mechanism. If AKAP95 dependency tracks transcriptional output rather than fusion identity, then every highly transcribing normal cell - activated lymphocytes, regenerating epithelium, erythroid precursors - is in scope, and the germline-knockout window measured under homeostasis would not have seen it.
- Both mutants used to establish the mechanism are pleiotropic, and one of them (PS) sits inside the DPY30-interaction region as well. A therapeutic hypothesis built on 'block RNA-gated condensation specifically' has no reagent that does only that.
- The direct Pol II CTD binding uses a 23-heptad fragment, less than half the human array. CTD condensation is famously repeat-number dependent, so the in vitro droplet behaviour may not represent the full-length CTD at physiological concentration.
- Is the ZF-mutant's defect RNA binding or DNA binding? The zinc fingers do both, and the whole RNA-gating model rests on attributing the phenotype to the RNA side. An RNA-binding-selective point mutant, or an in vitro droplet assay with DNA instead of RNA, would separate them and neither is reported.
- Does the pause-release defect exist in a KMT2A-rearranged cell? Every Pol II and PRO-seq experiment is in HeLa. A Ser2P ChIP-seq in AKAP95-depleted THP-1 or MV4-11 would convert the leukaemia argument from a proliferation rescue into a mechanism.
- Is the feed-forward loop what makes the requirement fusion-selective? The model predicts that AKAP95 dependency should scale with transcriptional output at a locus rather than with fusion identity - testable by asking whether the AKAP95-sensitive genes in a non-MLLr line are simply its most highly transcribed ones, with no MLL involvement at all.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Purified recombinant AKAP95 binds purified Pol II CTD (aa1356-1767, 23 heptapeptide repeats) directly by in vitro pull-down and co-condenses with it. The AKAP95 PS-mutant (six tyrosines in aa101-210 to serine) phase-separates poorly alone and co-condenses weakly with CTD; the ZF-mutant phase-separates normally without RNA but its condensates, and its co-condensates with CTD, do not enlarge when RNA is added, whereas wild-type condensates do. In HeLa Flp-In-Trex cells, ZF- and PS-mutants are released more readily into the cytoplasmic fraction on hypotonic treatment and are already largely released from chromatin before MNase, and the ZF-mutant's nuclear partition alone fails to respond to RNase; 4 h of actinomycin D impairs AKAP95 binding to genomic targets by ChIP-seq. In endogenous-AKAP95-depleted cells, both mutants show higher total Pol II occupancy at promoters and lower Ser2P across gene bodies than wild type, and reduced PRO-seq on-going transcription at AKAP95 targets (n = 414 around TSSs); the degree of Ser2P reduction correlates positively with AKAP95 ChIP-seq read density and with mRNA down-regulation. ZF-mutant foci hardly overlap Pol II Ser2P; PS-mutant is largely diffuse. Neither mutant rescues the growth defect of AKAP95-knockdown THP-1 or MOLM-13. AKAP95 CLIP top binding sites are 61.26% intron, 13.47% promoter, 9.33% exon, and a quarter of CLIP targets also have DNA-level ChIP binding.
AKAP95 condensates regulate transcription and can be targeted in MLL-fusion driven oncogenesis. · Yan X, Luan Y, Zhu Y, Cai Y, Ye F, Yang L, Zhu K, Hu J, Luo S, Lin J, Wang Y, Huang Y, Sun P, Yang Z, Shi B, Jiang H, et al. · Nat Commun 2026What this evidence cannot show
Full text read via PMC12886973; supplementary figures (8b, 8d-h, 9, 10a-e) and Source Data were NOT available, and a large fraction of this mechanism lives there - including the in vitro pull-down, the PS-mutant phase-separation comparison, the RNA-dependence of condensate size, and the Ser2P gene-body signal. Those are the main text's descriptions of its own supplement. Every mechanistic experiment is in HeLa or with purified protein; the expected_absent_terms record that no Pol II ChIP-seq or PRO-seq in a leukaemia line supports this mechanism, only a proliferation rescue.
Evidence that argues against this (1)
- The ZF-mutant is not a clean RNA-binding separation-of-function reagent. AKAP95's zinc fingers are also its GC-rich DNA-binding module by SELEX, and UniProt's chromatin-binding region 387-450 overlaps zinc finger 1 at 392-414. A ZF mutant that has lost DNA binding as well as RNA binding would produce the same reduced chromatin association, the same loss of Pol II Ser2P overlap, and the same rescue failure, by a mechanism that has nothing to do with nascent RNA. Marstad A et al. 2016
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
AKAP95's zinc fingers were independently established as its RNA-binding module: AKAP95 binds preferentially to proximal intronic regions of pre-mRNAs across the human transcriptome and this binding requires the zinc-finger domains; it interacts with hnRNP H/F and U through its N-terminal region and mainly promotes exon inclusion.
AKAP95 regulates splicing through scaffolding RNAs and RNA processing factors. · Hu J, Khodadadi-Jamayran A, Mao M, Shah K, Yang Z, Nasim MT, Wang Z, Jiang H · Nat Commun 2016What this evidence cannot show
Abstract read; full text not opened. Cited to establish that the ZF-mutant's RNA-binding deficiency is an independently characterised property and not a claim invented for the 2026 paper. It does not resolve the DNA-versus-RNA ambiguity raised in the counter-evidence above, because the zinc fingers are documented to do both.
Evidence that argues against this (1)
- The predominant AKAP95 RNA target is intronic pre-mRNA and the predominant documented function of that binding is SPLICING, not transcription. A model in which nascent RNA gates a transcriptional condensate has to explain why the same RNA-binding activity was previously characterised as a splicing function, and the 2026 paper's own G-293T reporter control - an intronless reporter, chosen precisely so that the effect must be transcriptional - is an acknowledgement of that problem rather than a resolution of it in the genomic setting. Yan X et al. 2026
the mechanism underneath the leukaemia phenotype: direct Pol II CTD binding, condensate size gated by nascent RNA
the same PS- and ZF-mutants that break RNA-gated condensation are the ones that fail to co-localise with MLL1 N-terminal fragments and fail to rescue MLLr growth
- related to →Plain-English: what transcription elongation is, and why a traffic light explains this leukemia
the defect is pause release - total Pol II up at promoters, Ser2P down across gene bodies
the same zinc fingers carry the GC-rich DNA preference behind the nucleolar pool, which is why the ZF-mutant is not a clean RNA-binding reagent
the mechanism the peptide is designed to break