Hypothesis: YEATS-pocket occupancy disables the reader module of all three SEC-family complexes in proportion to how much of the reader pool each holds — so the drug's footprint is set by an unmeasured partition, not by the fusion
also called scaffold-blindness hypothesis · q187 reader-pool partition hypothesis
The idea to test: a drug that plugs the reading pocket cannot tell which of three machines it is sitting in, so it switches off reading in all three. How much that matters depends entirely on one number nobody has measured — what fraction of the reader proteins live in each machine.
The same thing, for a scientist
**Statement.** ENL and AF9 are subunits of SEC, SEC-L2 and SEC-L3, and the YEATS pocket is not the scaffold-binding surface (the AHD is), so a YEATS-pocket ligand engages the reader without regard to scaffold identity. The **consequence** — how much of the drug's transcriptional footprint falls outside SEC — is predicted to be governed by the **partition of the cellular ENL and AF9 pool** across SEC, SEC-L2, SEC-L3, DotCom and free protein, a quantity that has never been measured in any cell. **Prior probability 0.3 that the non-SEC footprint is large enough to matter** (defined below), because the one available proxy points the other way: AFF2 and AFF3 mRNA are roughly 10-fold lower than AFF4 in HEK293T. This hypothesis is deliberately stated so it can be **cheap to kill**: the partition experiment is one quantitative immunoprecipitation series, and the public-data version is an intersection of two GEO deposits. It exists because the corpus currently carries the scaffold-blindness claim as an unqualified bite-back sentence in two entries, with no number attached and no stated way to be wrong.
Predictions, each with a number
- Partition. In a KMT2A::AFF1 line, quantitative immunodepletion of AFF1+AFF4 from a cleared lysate removes less than 85% of total ENL and of total AF9. (If it removes more than 95%, the sibling complexes are a trace sink and this hypothesis is dead.)
- Transcriptional footprint. In published YEATS-inhibitor RNA-seq (GSE185091, GSE162592), genes defined as AFF2- or AFF3-responsive-and-not-AFF4-responsive from GSE34097 move more than a size- and expression-matched random gene set, at fold-enrichment ≥ 1.5, p < 0.01 by permutation.
- Occupancy. ENL or AF9 ChIP/CUT&RUN detects signal above an IgG control at at least one of: an AFF3-bound gametic DMR, the Meg3 enhancer, the XIST DMR.
- Scaffold retention. In GSE185093's AFF1 and AFF4 ChIP under TDI-055, scaffold occupancy at ENL target promoters falls less than Pol II S2P does at the same promoters at 2 h — i.e. the reader is an anchoring device and the scaffold stays.
Prediction 4 is the two-surface model's own prediction and can be scored from data already on disk at NCBI.
How this could be wrong, pre-registered with three outcomes not two
This hypothesis is set up so the answer can be SUPPORTED, REFUTED, or MALFORMED. The third is live and has to be named in advance: if the reader pool turns out to exchange between scaffolds on a timescale faster than any fractionation can resolve — which the AF9 AHD's intrinsic disorder and "signalling hub" behaviour make plausible — then "what fraction of ENL is in SEC-L3" is not a well-posed question and prediction 1 cannot be scored. In that case the hypothesis should be withdrawn as ill-formed rather than recorded as refuted, and replaced by a residence-time question.
The corresponding trap: an immunodepletion experiment always returns a number, and a number from a malformed question looks exactly like a number from a well-formed one. Prediction 1 must be run with an exchange control — the same depletion on lysates mixed after lysis from separately tagged cells — or its output is uninterpretable.
What is already known that bears on it
- For: composition is established in all three complexes 1; the surfaces are separate 2; the three complexes' knockdown-response sets are largely distinct, with 27.5% of the union invisible to AFF4 depletion (
ana-q187-sec-family-target-set-overlap-measured). - Against: AFF2/AFF3 are ~10-fold lower expressed than AFF4 1; ENL is required in AML while AF9, in the same complexes, is not 2, so co-membership does not equalise anything; and a YEATS probe already has an in vivo therapeutic window in the index fusion with clean haematopoietic endpoints 3.
- Neither: nothing measures the partition, and nothing measures reader dependence in SEC-L2 or SEC-L3 (
absence-q187-no-yeats-inhibitor-has-been-tested-against-sec-l2-or-sec-l3).
The prior of 0.3 is set low deliberately. The strongest single piece of evidence available is the abundance gap, and it points against.
| statement | A YEATS-pocket ligand engages ENL and AF9 without discriminating which AFF scaffold they are bound to, so its transcriptional and toxicological footprint extends beyond SEC in proportion to the fraction of the cellular reader pool held by SEC-L2 and SEC-L3 — a partition that has never been measured. |
|---|---|
| origin | literature-gap |
| prior probability | 0.3 |
| prior basis | Composition and surface separation are established, so the qualitative claim is near-certain; 0.3 is the probability that the QUANTITATIVE consequence is material, defined as prediction 1 (AFF1+AFF4 immunodepletion removing less than 85% of total ENL and AF9). The prior is set below 0.5 because the only relevant measurement anywhere — AFF2 and AFF3 mRNA roughly 10-fold below AFF4 — points against, and because a published in vivo window already exists. |
predictions
[ "AFF1+AFF4 immunodepletion from a KMT2A::AFF1 lysate removes less than 85% of total ENL and of total AF9", "AFF2/AFF3-responsive-not-AFF4-responsive genes move under a YEATS inhibitor at fold-enrichment >= 1.5 over a size- and expression-matched random set, permutation p < 0.01", "ENL or AF9 signal above IgG at at least one of: an AFF3-bound gametic DMR, the Meg3 enhancer, the XIST DMR", "In GSE185093, AFF1 and AFF4 promoter occupancy under TDI-055 at 2 h falls less than Pol II S2P at the same promoters" ]
experiments
[ "exp-q187-partition-the-enl-and-af9-pool-across-the-sec-family", "exp-q187-allele-specific-expression-under-a-yeats-antagonist" ]
analyses
[ "ana-q187-sec-family-target-set-overlap-measured", "ana-q187-the-scaffold-blindness-argument-does-not-survive-the-published-therapeutic-window", "absence-q187-no-yeats-inhibitor-has-been-tested-against-sec-l2-or-sec-l3" ]
falsifiers
[ "PREDICTION 1 FALSIFIER: in a KMT2A::AFF1 line (SEM, RS4;11 or MV4;11), sequential immunodepletion of AFF1 and AFF4 from cleared nuclear lysate removes MORE than 95% of total ENL and more than 95% of total AF9 by quantitative western against the input, with an exchange control (post-lysis mixing of differentially tagged lysates) showing under 10% redistribution. That result makes SEC-L2 and SEC-L3 a trace sink and kills the hypothesis outright.", "PREDICTION 2 FALSIFIER: the AFF2/AFF3-only gene set from GSE34097 shows fold-enrichment below 1.1 for differential expression in BOTH GSE185091 and GSE162592 against a size- and expression-matched permutation null of 10,000 draws. Note this falsifier is weakened by a cell-type confound (HEK293T sets against AML drug data) and cannot alone kill the hypothesis; it demotes it.", "PREDICTION 3 FALSIFIER: ENL and AF9 CUT&RUN in a female human cell line show no signal above IgG at the XIST DMR, the Dlk1-Dio3 IG-DMR or the Meg3 enhancer, at a depth where a positive control promoter (MYC) gives at least 10-fold enrichment. That closes the imprinting limb specifically and leaves the transcriptional limb standing.", "PREDICTION 4 FALSIFIER: in GSE185093, AFF1 and AFF4 occupancy at ENL target promoters falls by MORE than Pol II S2P at 2 h of TDI-055. That would mean pocket occupancy destabilises the scaffold-reader assembly rather than only the chromatin contact, which contradicts the two-surface model this hypothesis is built on and would require the mechanism entry to be rewritten, not just this hypothesis withdrawn.", "WITHDRAWAL-AS-MALFORMED CONDITION, pre-registered: if the exchange control in prediction 1 shows more than 30% redistribution of reader between scaffolds after lysis, the partition is not a stable property and the hypothesis should be withdrawn as ill-posed rather than recorded as refuted, and replaced by a residence-time question." ]
measurements
[
{
"quantity": "prior_probability",
"subject": "probability that AFF1+AFF4 immunodepletion leaves more than 15% of total ENL and AF9 in a KMT2A::AFF1 lysate",
"value": 0.3,
"units": "probability",
"n": null,
"universe": "subjective prior set 2026-09-24 from one relevant measurement (AFF2/AFF3 mRNA ~10-fold below AFF4 in HEK293T, PMID 22547686) and one relevant negative (a published in vivo therapeutic window, PMID 42242459); no direct measurement of the partition exists in any system",
"null": null,
"note": "Stated so that a later cycle can score whether this beat's priors were calibrated. The qualitative claim (a YEATS ligand cannot discriminate by scaffold) is near-certain and is NOT what this probability refers to."
}
]- If prediction 1 comes back at 96% depletion, the correct response is to withdraw this hypothesis and amend `mechanism-q187-target-specificity-is-a-scaffold-property-while-the-reader-is-shared` and the two bite-backs in other entries that assert scaffold-blindness as a consequence. It is not to retreat to 'but the composition claim still holds' — the composition claim was never in doubt and is not what anyone is acting on.
- This hypothesis makes a reader-directed drug look more dangerous than a scaffold-directed one, which is the opposite of where this corpus's dependency data point: `protein-ell-eaf-module` records the compositionally selective handle (ELL) as sitting on proteins that are not dependencies. Acting on this hypothesis without that context would push a programme toward a target the DepMap evidence does not support.
- The prior of 0.3 is a guess with one data point behind it, and the field's habit is to treat a stated prior as more considered than an unstated one. It should be read as 'this beat thinks the abundance gap probably wins', not as a calibrated estimate.
- What antibody pair would give a clean quantitative AFF1 and AFF4 immunodepletion? This corpus already records that the two AFF1 antibodies used in a published deposit disagree with each other more than the two patients do (ana-q142-the-two-aff1-antibodies-disagree-more-than-the-two-patients-do), so the reagent question is not trivial and the depletion experiment could fail on it.
- Is a partition even the right frame, or is the reader pool exchanging faster than fractionation can resolve? The malformed branch is live and the AF9 AHD's documented dynamic partner exchange is the reason.
- Does the answer differ between ENL and AF9? They are treated as a pair throughout this beat and they are not interchangeable — one is required in AML and the other is not, and one governs human HSC self-renewal. The partition should be measured separately for each and this hypothesis's predictions say so.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
ENL/MLLT1 and AF9/MLLT3 are subunits of SEC-L2 (AFF2) and SEC-L3 (AFF3) as well as of SEC, the three complexes control largely different gene subsets, and AFF2 and AFF3 mRNA are about 10-fold lower than AFF4 in the HEK293T cells used.
The super elongation complex family of RNA polymerase II elongation factors: gene target specificity and transcriptional output. · Luo Z, Lin C, Guest E, Garrett AS, Mohaghegh N, Swanson S, Marshall S, Florens L, Washburn MP, Shilatifard A · Molecular and Cellular Biology 2012What this evidence cannot show
HEK293/293T, over-expressed tagged baits, one laboratory, no haematopoietic cell and no fusion. The abundance figure is mRNA by RT-qPCR in one background and is offered by the authors as a possible explanation for a failed detection, not as a measurement they designed. It is nonetheless the only quantity anywhere that bears on the partition, which is why this hypothesis's prior is set against itself.
Evidence that argues against this (1)
- Lu and colleagues argue SEC composition is combinatorial rather than a set of three fixed particles, with different ELL and AFF content serving different genes. If that is right, 'the fraction of ENL in SEC-L3' is not a property of a stable object and prediction 1 is closer to the MALFORMED branch than to either of the other two. Their panel excludes AFF2 and AFF3, so they cannot settle it either way. Lu H et al. 2015
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The YEATS domain binds acetylated histone H3 and its structure-based disruption reduces Pol II recruitment to ENL target genes without removing the protein, while ENL is required for AML disease maintenance and its paralogue AF9 — resident in the same complexes — is not.
ENL links histone acetylation to oncogenic gene expression in acute myeloid leukaemia. · Wan L, Wen H, Li Y, Lyu J, Xi Y, Hoshii T, Joseph JK, Wang X, Loh YE, Erb MA, Souza AL, Bradner JE, Shen L, Li W, Li H, Allis CD, Armstrong SA, Shi X · Nature 2017What this evidence cannot show
Myeloid AML, not infant B-lineage KMT2A::AFF1. A genetic YEATS mutant is not a ligand, and nothing in this paper involves AFF2, AFF3, SEC-L2 or SEC-L3. The AHD half of the two-surface model is carried by PMID 23260655 and is cited in the sibling mechanism entry, not here.
Evidence that argues against this (1)
- The ENL-required / AF9-not-required asymmetry is the cleanest available demonstration that shared complex membership predicts nothing about functional importance, and it therefore weakens this hypothesis at its root: if two paralogues in the same three complexes differ that much, the assumption that reader occupancy matters wherever the reader sits is unwarranted. This is cited as supporting evidence for the surface separation and as counter-evidence for the inference built on it. Leach BI et al. 2013
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
A compositionally scaffold-blind YEATS probe, SGC-iMLLT, inhibits KMT2A::AFF1-driven leukaemia growth in vitro and in vivo while not altering human HSPC colony-forming potential or long-term mouse HSPC function, so the existence of a therapeutic window for this class is already established and any prediction of this hypothesis must be compatible with it.
Inhibition of MLLT1 Limits Growth of KMT2A::AFF1 Leukemias Without Killing Healthy Hematopoietic Stem Cells. · Rajhansa S, Crump NT, Khoo HM, Dopico-Fernandez A, Bozhilov Y, Brennan PE, Fedorov O, Adams C, Farnie G, Milne TA, Wilkinson AC · Experimental Hematology 2026What this evidence cannot show
Abstract only; full text not in PMC and not read by this beat. The safety endpoints are both haematopoietic, so the window bounds haematopoietic toxicity and nothing else. One study, one group overlapping the probe's originating consortium, and no on-target YEATS-mutant rescue in a KMT2A::AFF1 line.
Evidence that argues against this (1)
- A clean haematopoietic window is compatible with a large non-SEC footprint, because the sibling complexes' documented functions are imprinting and X-inactivation rather than blood. So this evidence item constrains the hypothesis less than it appears to: it rules out the strongest version (that scaffold-blindness destroys the therapeutic index) and leaves prediction 2 and prediction 3 entirely untouched. Zhang Y et al. 2019
the surface-separation mechanism this hypothesis makes quantitative
prediction 1: AFF1+AFF4 depletion removing >95% of ENL and AF9 kills the hypothesis
functional test of the imprinting limb
the bounded absence that leaves every prediction open
- related to →SEC-L2 — the AFF2-scaffolded super-elongation-like complex: P-TEFb plus ENL and AF9, and NO ELL
the entry that names this hypothesis in its prose
the second sibling whose reader share the hypothesis quantifies over
refutes the unqualified selectivity claim and leaves the partition question as the testable residue