MLLT3 / AF9 — a famous fusion partner that is not a dependency, and a hub domain that is

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

• Compensation is the whole story here. MLLT3 and MLLT1 are paralogues with overlapping YEATS and AHD functions; a drug that hits AF9 alone is buffered by ENL, and the DepMap flat line is most simply explained by exactly that buffering. Any AF9-directed programme should assume it needs to cover both paralogues, which doubles the selectivity problem.
• Essential-function toxicity: gnomAD LOEUF 0.172 with 4 observed against 53.2 expected loss-of-function variants puts MLLT3 among the most constrained genes in the genome, and HPA reports group-enriched blood lineage specificity. A systemic AF9 inhibitor would engage it across every haematopoietic compartment, and the reassuring DepMap number comes from immortalised lines, not stem cells.
• Resistance is pre-loaded into the target rationale: if AF9 is already dispensable in unperturbed KMT2A-rearranged lines, a cell has no adaptation to make. Selection would act instead on whatever the drug does hit, which for a YEATS ligand means other acyl-lysine readers.
• The YEATS pocket is not a specific reader. Verified structures show it accommodating acetyl, crotonyl and butyryl marks on H3 and H4 and acyl-lysine on the non-histone protein MOZ/KAT6A. A ligand for that pocket displaces an unknown set of physiological contacts, and 9ARO/9ARR show the set extends beyond histones.
• Delivery and pharmacology: the AHD is intrinsically disordered and folds on binding, so the productive AHD ligands so far are peptidomimetics (8TLV, 8TLW, 8TLX). Peptidomimetics against a nuclear target in a paediatric population carry a permeability and exposure problem that the crystal structures do not address.

1. Do the DepMap library's MLLT3 guides fall 5' or 3' of the fusion breakpoint in MOLM-13, MOLM-14, THP-1 and NOMO-1? If 3', then the flat scores in KMT2A::MLLT3 lines are a strong negative result about the fusion itself; if 5', they say nothing about it. The gene-level API value cannot distinguish these, and nothing else in this dossier turns on a single unresolved annotation as much as this does.
2. Is the ENL-versus-AF9 asymmetry a property of the YEATS reader or of the AHD? Swapping the two YEATS domains between the paralogues, with the AHD held constant, would answer it in one experiment and would tell a medicinal chemist which domain to work on.
3. Is MLLT3 dispensable because MLLT1 compensates? A double knockout, or an MLLT1 knockout in an MLLT3-null background, has not been reported in these lines and is the obvious test of whether the pair is redundant rather than individually irrelevant.
4. Does the 10 nM AF9-YEATS binder (CHEMBL4204744) engage the domain in cells, and does it displace acyl-lysine marks at endogenous occupancy? A potent in vitro binder of a reader domain has to outcompete an abundant mark at its own local concentration, and nothing in ChEMBL addresses that.
5. What does a LOEUF of 0.172 with 4 observed loss-of-function variants against 53.2 expected actually reflect, given a completely flat cancer-cell dependency profile? Whatever MLLT3 is needed for in a developing human is invisible to every assay in this dossier.
6. Is AF9's nuclear residence chaperone-dependent in a leukaemic cell? The Hsp90 result is from kidney; if it holds in blood it would predict that an Hsp90 inhibitor partially phenocopies AF9 loss at AF9 target loci, which is cheap to test and has never been asked.
7. Does a KMT2A fusion displace wild-type AF9 in an endogenous t(4;11) or t(9;11) cell? The only positive result is a transfection experiment from 2004, and it is one of the few localisation phenotypes this disease has.

1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   MLLT3 is not DepMap common-essential; mean Chronos across 1,208 lines is -0.002 with 0.4% of lines depleted, the leukemia mean is 0.000, the 10 KMT2A-rearranged lines average +0.059 and the 4 KMT2A::AFF1 lines +0.058, giving selectivity_vs_leukemia +0.059 and selectivity_vs_all_lines +0.061; gnomAD gives pLI 1.0, LOEUF 0.172, 4 observed versus 53.2 expected loss-of-function variants and mis_z 1.87; HPA classifies MLLT3 as low tissue specificity, detected in all tissues, low immune cell specificity with group-enriched blood lineage specificity and nucleoplasmic localisation.

   DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL API; loss-of-function constraint via the gnomAD GraphQL API; normal-tissue expression via the Human Protein Atlas API · Broad Institute DepMap; gnomAD; Human Protein Atlas; Open Targets Platform · Databases 2026
   ▸What this evidence cannot show

   Means were computed by us from per-line API values rather than read from a DepMap release table, and the API does not state the release version. Only 10 KMT2A-rearranged lines (4 KMT2A::AFF1) contribute, so with n=10 any absolute selectivity under 0.15 is within noise and +0.059 should be read as indistinguishable from zero. Whole-gene knockout removes the YEATS reader and the AHD hub together and cannot separate them. Critically, the gene-level score does not tell us whether the library's MLLT3 guides fall 5' or 3' of the fusion breakpoint, so we cannot say whether the fusion allele was also cut in the four KMT2A::MLLT3 lines. Immortalised cancer lines say nothing about human haematopoietic stem cells.

   ▸Evidence that argues against this (1)

   • gnomAD constraint points the opposite way: LOEUF 0.172 with 4 observed against 53.2 expected loss-of-function variants places MLLT3 among the most constrained genes in the genome, more constrained than DOT1L or MLLT1. Human heterozygous loss is strongly selected against, so a flat DepMap profile is not a licence to treat the protein as dispensable in a whole organism. gnomAD Consortium 2026
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   Depletion of ENL, but not of AF9, suppresses leukaemic gene expression and proliferation in acute myeloid leukaemia; the requirement is tied to the ENL YEATS domain reading histone acetylation, which links acetylation to oncogenic gene expression.

   ENL links histone acetylation to oncogenic gene expression in acute myeloid leukaemia · Wan L, Wen H, Li Y, et al. · Nature 2017
   PMID 28241141DOI
   ▸What this evidence cannot show

   The paralogue contrast is a depletion experiment, not a knock-in point-mutant comparison, so residual AF9 protein or incomplete knockdown cannot be excluded as an explanation for the asymmetry. The models are AML rather than KMT2A::AFF1 infant B-ALL, and the abstract reports no affinity measurement for the YEATS-acetyl-lysine interaction. The paper's own structural anchor is 5J9S (ENL YEATS with H3K27ac), an ENL structure, not an AF9 one.

   ▸Evidence that argues against this (1)

   • This is the paralogue contrast and it cuts against the entry it sits in: ENL depletion suppresses leukaemic gene expression and AF9 depletion does not. An AF9 entry citing it is citing evidence that AF9 is the dispensable one of the pair. eLife 2021
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   The AF9 YEATS domain is a broad short-chain acyl-lysine reader with verified co-crystal and NMR structures against H3K9ac (4TMP, 2.30 A), H3K9cr and H3K18cr (5HJB 2.70 A, 5HJD 2.81 A), H3K18ac and H3K18cr (2NDF, 2NDG, NMR), H3K9bu (6MIL, 1.93 A), di-acetylated histone H4 (7EIC 1.95 A, 7EID 2.00 A), acetylated and dicrotonylated MOZ/KAT6A peptides (9ARO 2.30 A, 9ARR 2.10 A) and a cyclopeptide inhibitor (6L5Z, 3.05 A); the separate AF9 AHD (aa 500-568) has verified complexes with DOT1L (2MV7), CBX8 (2N4Q), BCOR (6B7G), AFF1 (2LM0) and peptidomimetic inhibitors (8TLV 2.66 A, 8TLW 2.11 A, 8TLX 2.10 A); and the AF9 YEATS domain links histone acetylation to DOT1L-mediated H3K79 methylation.

   AF9 YEATS domain links histone acetylation to DOT1L-mediated H3K79 methylation · Li Y, Wen H, Xi Y, et al. · Cell 2014
   PMID 25417107DOI
   ▸What this evidence cannot show

   Identifiers, resolutions and modelled residue ranges were resolved against the RCSB API in this session, but the ranges are RCSB's UniProt-mapped modelled segments rather than independently curated domain boundaries. None of these entries carries a published affinity in the abstracts we verified, so no Kd is attributed. 4TMP is a YEATS structure and is sometimes miscited as an AHD structure. The MOZ peptide complexes show the domain reads non-histone acyl-lysine too, which widens rather than narrows the selectivity problem.

   ▸Evidence that argues against this (1)

   • Counting PDB entries measures how tractable a protein was to CRYSTALLISE, which correlates with being small, ordered and well-behaved rather than with being druggable. What decides design value is whether the deposited density covers the site that matters, at a resolution that supports modelling, in the relevant state — and this graph has repeatedly found structures described in the direction that overstates what is resolved. Broad acyl-lysine reading is a liability for a reader-pocket inhibitor rather than an asset: a pocket that accepts acetyl, crotonyl and other acylations at several histone positions is a pocket whose occupancy has many possible consequences. RCSB PDB 2026
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   The degree to which DOT1L is recruited to MLL-AF9 through the AF9 AHD sets the level of H3K79 di- and tri-methylation on target genes and the transformation potential of the fusion, with AHD mutants giving graded rather than all-or-none loss; DOT1L itself is pharmacologically tractable, with pinometostat/EPZ-5676 inhibiting it at Ki 80 pM and 37,000-fold selectivity over other methyltransferases.

   Degree of recruitment of DOT1L to MLL-AF9 defines level of H3K79 di- and tri-methylation on target genes and transformation potential · Kuntimaddi A, Achille NJ, Thorpe J, et al. · Cell Reports 2015
   PMID 25921540DOI
   ▸What this evidence cannot show

   This is MLL-AF9 in murine transformation assays, not KMT2A::AFF1. The pinometostat affinity comes from a separate source (PMID 23801631) and is the only numeric affinity verified anywhere in this interactome's source set; it is a DOT1L number and must not be transferred to any AF9 interaction. The graded-recruitment result is from engineered AHD point mutants, which is a cleaner genetic experiment than a drug but does not predict the dose-response of a competitive ligand.

   ▸Evidence that argues against this (1)

   • Graded AHD mutants in MLL-AF9 are a murine myeloid model, and the graded response is to engineered recruitment rather than to drug occupancy. Non-canonical H3K79me2-dependent pathways promote survival, so the mapping from methylation level to requirement is not monotonic. eLife 2021
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   AF4/AFF-family and ENL-family proteins assemble with P-TEFb into a higher-order complex that facilitates both oncogenic and physiologic MLL-dependent transcription, which is the route by which an AFF1-containing fusion can access an ENL or AF9 AHD without any direct KMT2A-DOT1L contact.

   A higher-order complex containing AF4 and ENL family proteins with P-TEFb facilitates oncogenic and physiologic MLL-dependent transcription · Yokoyama A, Lin M, Naresh A, Kitabayashi I, Cleary ML · Cancer Cell 2010
   PMID 20153263DOI
   ▸What this evidence cannot show

   This is pan-MLL-chimera biochemistry, not a KMT2A::AFF1-specific interactome; no KMT2A::AFF1-specific published interactome paper was found this cycle. Family-level statements ('AF4 family', 'ENL family') cannot be resolved to MLLT3 versus MLLT1 or AFF1 versus AFF4 from this source, which is precisely the redundancy that makes the DepMap result hard to interpret.

   ▸Evidence that argues against this (1)

   • Complex composition is not requirement. The AEP/SEC papers establish what co-purifies and what a reconstituted complex does; they do not establish that each subunit is rate-limiting in a patient's leukaemia, and this project's own dependency data separates the members by an order of magnitude. It is also the route by which an AFF1-containing fusion works, which makes it evidence about AFF1 rather than about AF9. Nature 2012
6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   ChEMBL v34 target CHEMBL4295761 (Protein AF-9) carries 66 activity rows at pChEMBL >= 6 with a best value of Ki 10 nM / Kd 15 nM (CHEMBL4204744, binding assay) and no curated mechanism of action; the paralogue target CHEMBL4295801 (Protein ENL) carries 144 potent rows with a best Ki of 56 nM, also with no curated mechanism of action.

   ChEMBL database version 34, targets CHEMBL4295761 (Protein AF-9) and CHEMBL4295801 (Protein ENL): target_search, get_bioactivity and get_mechanism queries · EMBL-EBI ChEMBL · ChEMBL 2026
   ▸What this evidence cannot show

   Distinct-compound counts and 'best' values are page-local: the API returns 25 rows per page, so these describe one sampled page probably dominated by a single publication rather than a potency-ranked view of the whole target. Binding affinity against an isolated YEATS domain is not cellular target engagement; the domain competes with an abundant chromatin mark at its own local concentration. Querying only SINGLE PROTEIN targets systematically under-reports clinical matter, because curated mechanism records tend to live on family or protein-protein-interaction targets instead.

   ▸Evidence that argues against this (1)

   • A ChEMBL activity-row count measures how much medicinal chemistry has been DONE, not how tractable the target is, and the best reported affinity belongs to the most optimised compound in whichever series someone chose to pursue. Zero rows means nobody tried; thousands of rows means a target had a decade of industrial attention. ChEMBL, EMBL-EBI 2026
7. E_q72_mllt3_loc1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   UNIPROT PUTS AF9 IN THE NUCLEUS AND ON CHROMATIN AND DOES NOT MENTION THE CYTOPLASM. Fetched 2026-09-21, P42568 (AF9_HUMAN, primary gene MLLT3, 568 aa) carries two location values: 'Nucleus' with ECO:0000255 (PROSITE-ProRule PRU00376, a sequence-model prediction) AND ECO:0000269 to PMID 27105114; and 'Chromosome' with ECO:0000269 to PMID 25417107 and 27105114. The free-text note, also ECO:0000269, reads 'Colocalizes with acylated histone H3 ... Colocalizes with histone H3 crotonylated at Lys-18'. The only NLS motif in the feature table is residues 295-300, ECO:0000255 - a prediction with no cited publication and no experimental test.

   UniProtKB entry P42568 (AF9_HUMAN), SUBCELLULAR LOCATION comment and feature table · UniProt Consortium · UniProtKB 2026
   ▸What this evidence cannot show

   A database state on a date. Both experimental attributions for the location come from the SAME GROUP'S PAIR OF YEATS-DOMAIN PAPERS, so the 'Chromosome' assignment is really one line of work read twice - and its readout is co-localisation with a histone mark, which is a chromatin-association argument rather than a compartment measurement. The ECO:0000255 half of the 'Nucleus' call is a PROSITE rule firing on the YEATS domain, not a measurement. IDENTIFIER NOTE (brief rule 88): gene_exact:MLLT3 returns entry name AF9_HUMAN with recommended protein name 'Protein AF-9' - NEITHER STRING CONTAINS 'MLLT3', so a reader applying rule 65's read-the-name-back defence concludes a correct lookup failed. This is the same DPF2/AF17 readback class; no row existed in data/symbol-collisions.json before this beat.

   ▸Evidence that argues against this (1)

   • THE DATABASE IS INCOMPLETE IN A WAY THAT MATTERS: AF9 IS ALSO CYTOPLASMIC. In renal tubular cells AF9 is present in the cytoplasm as well as the nucleus, and Hsp90 - as part of an Hsp90-Hsp70-p60/Hop chaperone complex that binds AF9 directly - is required to keep it nuclear. UniProt annotates no cytoplasmic value at all, so an entry that reads the database and stops will assert a single-compartment protein. Lin JJ et al. 2010
8. E_q72_mllt3_loc2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   IN RENAL TUBULAR CELLS (not leukaemia), AF9/MLLT3 IS PRESENT IN THE CYTOPLASM AS WELL AS THE NUCLEUS, AND ITS NUCLEAR RESIDENCE IS CHAPERONE-DEPENDENT. Hsp90 interacts directly and specifically with AF9 as part of an Hsp90-Hsp70-p60/Hop complex; inhibiting Hsp90 with NOVOBIOCIN - but not with 17-AAG - or knocking Hsp90 down with siRNA REDISTRIBUTES AF9 FROM A PRIMARILY NUCLEAR TO A CYTOPLASMIC LOCATION, which increases ENaC-alpha expression and DECREASES AF9 OCCUPANCY AT THE ENaC-alpha PROMOTER. The authors conclude the chaperone complex is necessary for the proper SUBNUCLEAR localisation and activity of AF9.

   Hsp90 directly modulates the spatial distribution of AF9/MLLT3 and affects target gene expression. · Lin JJ, Hemenway CS · J Biol Chem 2010
   PMID 20159978DOI
   ▸What this evidence cannot show

   Abstract read in full; full text not read. RENAL TUBULAR CELLS AND ALDOSTERONE SIGNALLING - not leukaemia, not haematopoietic, no KMT2A fusion anywhere in the paper (brief rule 73), and the target gene is the epithelial sodium channel. The novobiocin-versus-17-AAG split is a real but awkward result: the two agents hit different Hsp90 sites, and a phenotype seen with only one of two inhibitors of the same chaperone needs the siRNA arm to carry it, which here it does. 'Proper subnuclear localisation' is the authors' phrase and the data behind it are a nuclear-versus-cytoplasmic partition plus a promoter ChIP, NOT a subnuclear compartment assignment. NOTHING IN THIS HAS BEEN TESTED IN A LEUKAEMIC CELL, and the fusion loses most of AF9 anyway.

   ▸Evidence that argues against this (1)

   • The result is hard to transfer to the fusion for a structural reason. The only NLS UniProt annotates on AF9 is at residues 295-300 and t(9;11) breaks occur almost exclusively downstream of MLLT3 exon 4, so that region is absent from the fusion - the chimera's nuclear address is a KMT2A property, and a chaperone dependence measured on full-length AF9 in kidney says nothing about it. localization-kmt2a-mllt3-the-address-is-a-kmt2a-property carries that reasoning and the breakpoint statistic. Erfurth F et al. 2004
9. E_q72_mllt3_loc3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   AF9'S CHROMATIN ADDRESS IS A MARK, NOT A COMPARTMENT. The AF9 YEATS domain binds H3K9ac strongly and H3K27ac and H3K18ac less so, using a serine-lined aromatic sandwiching cage, and ChIP-seq shows strong genome-wide co-localisation of AF9 with H3K9 acetylation that is important for chromatin recruitment of DOT1L. The same domain binds crotonyl-lysine with HIGHER affinity than acetyl-lysine, and in a cell-based model AF9 co-localises with crotonylated histone H3 and positively regulates gene expression in a YEATS-dependent manner.

   Molecular Coupling of Histone Crotonylation and Active Transcription by AF9 YEATS Domain. · Li Y, Sabari BR, Panchenko T, Wen H, Zhao D, Guan H, Wan L, Huang H, Tang Z, Zhao Y, Roeder RG, Shi X, Allis CD, Li H · Mol Cell 2016
   PMID 27105114DOI
   ▸What this evidence cannot show

   Abstracts read in full; full texts not read. These are the two papers UniProt uses for both the 'Chromosome' value and the note, so the database's chromatin assignment and this evidence item are the SAME observation, not two. ChIP-seq co-localisation with a histone mark is a genomic-address measurement, NOT a subnuclear-compartment measurement: it tells you which loci, not which morphological structure, and this graph's standing priority asks for both. Both papers are already cited here (25417107 in 7 entries, 27105114 in 3) - subtraction-mllt3-yeats is where the fusion-relevant reasoning lives (rule 82). Neither paper is in a KMT2A-rearranged B-lineage system.

   ▸Evidence that argues against this (1)

   • A mark-reading address does not survive into the index disease intact and is partner-specific. The YEATS domain is retained in KMT2A::MLLT3 only in the sense that the fusion keeps the C-terminal AHD and loses the N-terminal YEATS - this graph's subtraction-mllt3-yeats records that the YEATS domain (P42568 residues 1-138) is on the LOST side in about 92% of MLL-AF9 patients, whose breakpoints fall at or after AF9 residue 376, so an H3K9ac/H3K18cr address measured on full-length AF9 is not the fusion's address. Reading it across would be a partner-imported inference. Li Y et al. 2014
10. E_q72_mllt3_loc4weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    THE ONE PLACE A KMT2A FUSION HAS BEEN SHOWN TO MOVE AF9 IS A TRANS EFFECT, NOT A CIS ONE. AF4 and AF9 interact at discrete non-nucleolar nuclear foci ('AF4 bodies') that are not coincident with any known nuclear structure the authors examined; the interaction is maintained by the MLL-AF4 fusion protein; and expression of MLL-AF4 CAN ALTER THE SUBNUCLEAR LOCALISATION OF AF9. Separately, conditionally expressed MLL-AF9 localises to nucleoplasm and nucleolus in a punctate pattern indistinguishable from an N-terminal MLL fragment and from endogenous MLL.

    MLL fusion partners AF4 and AF9 interact at subnuclear foci. · Erfurth F, Hemenway CS, de Erkenez AC, Domer PH · Leukemia 2004
    PMID 14603337DOI
    ▸What this evidence cannot show

    Abstract read in full; full text not read. TRANSFECTION WITH TAGGED PROTEINS, not endogenous t(4;11) or t(9;11) cells, and the displaced protein is wild-type AF9 rather than the fusion. 'AF4 bodies' have never been reproduced, have no marker, and have not been sought in a patient sample in 22 years. Confidence is weak deliberately. This graph already carries the compartment (localization-af4-bodies-a-named-compartment-matching-nothing-known) and the whole two-paper subnuclear literature for KMT2A::MLLT3 (localization-kmt2a-mllt3-the-address-is-a-kmt2a-property) - read those, do not re-derive them.

    ▸Evidence that argues against this (1)

    • The MLL-AF9 half of this is overexpression in U937 and HeLa and is indistinguishable from endogenous MLL and from a bare N-terminal MLL fragment, so it attributes no compartment to the fusion at all - it attributes one to the KMT2A moiety. The entire published subnuclear-localisation literature for KMT2A::MLLT3 is these two papers, from 2000 and 2004, both overexpression. Caslini C et al. 2000

• ← regulatesHDAC5 - the eraser in an existing entry's own id, and the one HDAC in this graph that erases a mark on a fusion partner protein rather than on a histone

  HDAC5 deacetylates AF9 at K339, reversing the PCAF mark that monomerises AF9

• ← regulatesKAT2B / PCAF - named in an existing entry's own id and never written, and the only acetyltransferase in this graph whose substrate is a fusion partner protein

  PCAF acetylates AF9 at K339, monomerising it and reducing TFIID interaction

• ← bindsAFF1 / AF4 — domain architecture and residue coordinates

  AF4 interaction motif aa 761-775 binds the AF9 ANC1 homology domain (PDB 2LM0)

• ← supportsExpressed, required, maintains — three different claims about where a gene matters, and the graph must label which one it has

  MLLT3 is the mirror case: a normal-cell requirement with no requirement in any screened leukaemia line

• ← part ofMLLT3 runs normal stem-cell self-renewal and is required in essentially no leukaemia line — the inverted case

  the cell-context dossier for this gene

• ← depends onBone marrow haematopoietic stem cell — the postnatal comparator, and why the same fusion behaves differently here

  MLLT3 governs human HSC self-renewal and engraftment - a wild-type, normal-cell requirement

• regulates →The stem-like self-renewal state — a limited stem-cell module grafted onto a committed cell

  sustains the self-renewing state of human HSCs ex vivo; normal cells, not leukaemic

• ← related toAnti-AF9 antibodies see three proteins, not two - and the third is an internal-promoter isoform discovered in 2025

  an internal promoter in this locus makes a YEATS-devoid isoform sharing the fusion's retained epitopes

• ← bindsBlocking where DOT1L binds instead of what it does - the modality that removes the mark only at fusion targets

  AF9 AHD pocket is the drugged interface

• ← bindsAFF3 (LAF4) — the AFF-family scaffold of SEC-L3, and a KMT2A fusion partner in its own right

  SEC-L3 contains AF9/MLLT3

• binds →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)

  AF9, a famous partner that is not itself a dependency

• ← bindsBCOR — the other side of the CBX8 conflict, and a target you would be inhibiting in the direction of disease

  competes for the same AF9 surface as CBX8

• binds →H3K27ac - named in 40 entries with no entry of its own, and the strongest KMT2A::AFF1 measurement of it is invisible to every abstract search

  AF9 YEATS binds H3K27ac weakly (H3K9ac strongly) and is discarded in ~92% of KMT2A::MLLT3 patients

• binds →H3K9ac - the mark KMT2A fusions pile up on their own targets, the one the AF9 reader actually binds, and the one a DOT1L inhibitor strips off first

  AF9 YEATS binds H3K9ac with high affinity and recruits DOT1L - but is discarded in ~92% of KMT2A::MLLT3 patients

• ← related toMLLT1 / ENL — the reader that beat its own paralogue, with a degrader already in hand

  Counter-evidence (Calvanese 2019, PMID 31776511): MLLT3/AF9 governs human haematopoietic stem-cell self-renewal and engraftment; depletion prevents maintenance of transplantable human HSPCs. 'AF9 is dispensable' holds in leukaemia lines only, so paralogue cross-reactivity for an ENL ligand or degrader is a stem-cell toxicity risk, not a selectivity nicety.

• ← part ofKMT2A::MLLT1 versus KMT2A::MLLT3: paralogous partners whose chromosomes break on opposite sides of the same domain

  AF9, UniProt P42568, 568 aa, AHD 502-568

• ← related toThe KMT2A::MLLT3 chimeric protein: what survives the join on each side, in residue numbers - and the four AF9 modules that do not

  P42568 coordinates, fetched and read back 2026-09-20

• ← part ofThe retained AF9 surface in residue detail: one disordered 70-mer, three DOT1L sites, two of them high-affinity and simultaneously occupiable

  the retained AF9 AHD surface

• ← related toKMT2A::MLLT3 / t(9;11): the best-known partner prognosis claim, and the three cohorts that did not reproduce it

  clinical profile of the partner from which this graph imports most mechanism

• ← regulatesPRKDC / DNA-PKcs - the kinase that phosphorylates AF9 on the one regulatory residue the fusion keeps

  phosphorylates AF9 at S395

• ← regulatesDNA-PKcs phosphorylates AF9 at S395 and releases it from the super elongation complex - the only AF9 regulatory element the fusion keeps

  S395 phospho-switch on the retained half

• ← related toPhosphosite counts for the five 3' partners: the atlas beats UniProt 1.4-6.0x, and the retained/lost split is completely different for each partner

  9 union sites, only 3 retained

• ← part ofFour regulatory elements on AF9, and the fusion keeps exactly one - the phospho-switch that controls SEC binding

  element-level map of AF9

• ← regulatesThe AF9 acetylation arm: PCAF at K339, HDAC5 gated by CaMKII - a kinase-controlled switch the fusion throws away

  PCAF/HDAC5 acetylation switch at K339

• ← related to"ENL" returns 1,109 PubMed records of which about 30% are the protein: a measured alias-precision failure

  AF9 symbol is buried inside the MLL-AF9 disease literature

• ← bindsKMT2A::AFF1 has no mapped interface to DOT1L: it lacks both structurally defined recruitment modules, and the one complex-level route claimed for it is contested by reconstitution

  AF9's AHD is the surface the index fusion reaches in trans

• ← part ofImport class 2: AHD-dependent mechanism, which cannot transfer because KMT2A::AFF1 has no AHD

  AF9 carries the AHD

• ← bindsAFF1 AF4_int motif to AF9/ENL AHD: the index fusion's only documented elongation interface, and it competes with DOT1L

  the AF9 AHD, the receptor side (PDB 2LM0)

• ← testsDoes KMT2A::AFF1 deliver DOT1L to the AHD or displace it? A competition experiment in t(4;11) cells

  AF9 is the immunoprecipitated receptor

• ← related toPhase-2 dossier: the ENL YEATS pocket - two co-crystals, nanomolar chemistry, and a co-dependency that only appears inside leukaemia

  mandatory paralogue selectivity comparator for any YEATS ligand

• ← related toThe super elongation complex, sampled from the RNA side: ELL, ENL, CDK9 and cyclin T1 are all in the proximity record - ELL2, EAF1/2, AF9 and AF10 are at zero

  MLLT3/AF9 has never appeared in any proximity-labelling record

• ← related toThe paper's four comparison loci, marks read against FISH calls: the H3K9 mark response agrees with the cytogenetic response at two loci and contradicts it at two - the mark is neither necessary nor sufficient

  clean negative on every channel - no mark change, no CTCF change, no cytogenetic event

• ← related toHypothesis: H3K79me2 at a KMT2A fusion's own sites survives menin inhibition when the C-terminal partner contacts DOT1L directly (MLLT3/AF9) and decays when it does not (AFF1) - so the mark's response is a property of the partner and not of menin, which predicts the opposite of the direct-contact intuition and is testable in one line

  MLLT3/AF9 contacts DOT1L directly through its ANC1-homology domain, which is the menin-independent contact the hypothesis invokes

• ← recruitsDOT1L leaves chromatin before menin does when the fusion is degraded, and the ordering survives a change of instrument that destroys the magnitudes - ChIP-qPCR at selected loci, target-restricted read counts and genome-wide peak presence all rank DOT1L ahead of menin while disagreeing about every number

  AF9/MLLT3 binds DOT1L directly through three mapped binding sites whose graded disruption grades H3K79 methylation and transformation (PMID 21896721, PMID 25921540) - the architecture that predicts DOT1L leaving first when the fusion is degraded.

• ← inhibitsSGC-iMLLT - the MLLT1/MLLT3 YEATS chemical probe, and the 2026 paper that put it into KMT2A::AFF1 in vivo with a healthy-HSPC comparison

  same probe, same pocket on AF9; the probe does not distinguish the two paralogues

• ← related toSEC-L2 and SEC-L3 contain ENL and AF9 exactly as SEC does, so AFF2 and AFF3 sit in complexes built around the same acyl-lysine readers as the index fusion's partner - and both entries carry no epigenetics annotation at all

  AF9 likewise appears in all three SEC-family complexes

• part of →SEC-L2 — the AFF2-scaffolded super-elongation-like complex: P-TEFb plus ENL and AF9, and NO ELL

  AF9 is a shared subunit of SEC-L2

• part of →SEC-L3 — the AFF3-scaffolded complex, and the only SEC-family member with a known physiological job: genomic imprinting and X inactivation

  AF9 is a shared subunit of SEC-L3

• part of →SEC — the AFF1/AFF4-scaffolded super elongation complex, the particle KMT2A::AFF1 is a permanent subunit of

  AF9 is a YEATS-reader subunit of SEC, and of SEC-L2 and SEC-L3 as well

• ← depends onTwo surfaces, two jobs: the AHD picks the scaffold and the YEATS pocket reads the histone — so a YEATS ligand cannot ask which of the three SEC-family complexes it is disabling

  AF9 AHD-AF4 is the solved instance of the scaffold-binding surface

• ← testsMeasure the one number the whole YEATS-selectivity argument turns on: what fraction of ENL and of AF9 is in SEC, in SEC-L2, in SEC-L3 and free — with an exchange control, because the question may not be well posed

  AF9 pool partition, measured separately from ENL because the two are not interchangeable

• ← related toExpressed, required, maintains — three different claims about where a gene matters, and the graph must label which one it has

  the class scheme has no slot for a subunit-versus-complex problem: MLLT3 overexpression and knockdown move SEC, SEC-L2 and SEC-L3 together (PMID 22547686), so a class-2 'required' label sits on a complex-level perturbation labelled by one subunit

• related to →Where KMT2A::MLLT3 sits in the nucleus: two papers, both overexpression, both from before 2005 - and the fusion carries no AF9 import signal

  AF9 predicted NLS 295-300 is discarded by the breakpoint; the fusion address is a KMT2A property

• related to →'AF4 bodies' - a named subnuclear compartment that is non-nucleolar and coincident with no known nuclear structure, and the one place a KMT2A fusion has been shown to move a partner

  AF9 is the other half of the AF4 body, and MLL-AF4 moves it

• depends on →MLLT3 (AF9): the YEATS acetyl-lysine reader is discarded in ~92% of MLL-AF9 patients — same shape as PZP, but the causal test has never been run

  the H3K9ac/H3K18cr chromatin address sits on the YEATS domain the fusion discards

• related to →A partner's 'normal' compartment is usually read off a different tissue, and for at least two KMT2A partners that reading is wrong in blood cells

  AF9 cytoplasmic pool is measured in kidney and absent from UniProt entirely

• ← related toThe six unwritten DOT1L partners are not six findings - they are nine spokes of ONE GFP-DOT1L pulldown, and the source paper already graded them

  AF9 is the internal positive control and the counter-example: A+ and structurally validated, yet 1:172 substoichiometric - which is why stoichiometry cannot be the filter

• ← refutesAtlas-only interactome partners are predicted by how abundant the prey is, not by which bait was pulled - and if so this graph's interactome denominator is inflated

  AF9 is the standing counter-example - a structurally validated partner at the bottom of the stoichiometry ranking - and any filter built on this hypothesis must survive it

• ← related toAF9's nuclear residence is chaperone-dependent, the Hsp90 site is the SAME C-terminal region the fusion keeps, and UniProt records no cytoplasmic pool at all

  AF9's cytoplasmic pool and chaperone-dependent nuclear residence, given its own entry; carries a correction owed to that entry's subnuclear field

• ← bindsPIP4K2A - the lipid kinase at a genome-wide childhood-ALL susceptibility locus, whose product is the ligand a PHD finger reads

  PIP4K2A-bait AP-MS, BioPlex 2.0 and 3.0, HEK293T

• ← bindsEPB41L3 / DAL-1 - a membrane-skeleton tumour suppressor that binds and modulates two arginine methyltransferases, and whose own promoter methylation is a clinical test

  EPB41L3-bait AP-MS in HEK293T (BioPlex 3.0) and HEK293 (Hippo interaction proteome, PMID 24366813)

• ← related to'AF4 bodies' - a named subnuclear compartment that is non-nucleolar and coincident with no known nuclear structure, and the one place a KMT2A fusion has been shown to move a partner

  AF9 is the partner whose subnuclear position MLL-AF4 expression alters

• ← related toMLLT3 (AF9): the YEATS acetyl-lysine reader is discarded in ~92% of MLL-AF9 patients — same shape as PZP, but the causal test has never been run

  AF9 YEATS domain aa 1-138