The ELL/EAF activation module (ELL2, EAF1, EAF2) — a general elongation stimulator, not a KMT2A-fusion dependency

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.

• This module is general polymerase machinery by construction. ELL2-EAF1 occupies the same Pol II lobe site as TFIIF, a factor used at every promoter, so any agent blocking that interaction perturbs general transcription. There is no version of this interface that is fusion-selective, and the structure itself is the argument.
• Paralogue compensation cuts both ways and defeats the obvious experiments. EAF1/EAF2 and ELL/ELL2 are redundant pairs, so single-gene knockouts under-report the module's importance (our negative result may be a false negative) while a pan-family inhibitor would inherit all four genes' normal functions at once (the toxicity would be a true positive). Neither the screen nor the drug can isolate the piece we want.
• Essential-function toxicity in the wrong compartment. EAF1 (bone marrow 53.9 nTPM) and EAF2 (bone marrow 67.8 nTPM, lymphoid 50.0 nTPM) are enriched in exactly the tissue the disease occupies, and ELL2 is immune-cell-enhanced. An agent against this module would be maximally exposed in normal haematopoiesis while, on the genetics, buying nothing in the tumour — the worst of both.
• The expression data actively mislead, and a pipeline could act on them. A target-selection process weighting haematopoietic enrichment would nominate this module; the loss-of-function data say the most marrow-enriched member (EAF2) is the most dispensable gene in the batch, with LOEUF 1.345 and zero structures. Any automated scoring that does not let DepMap override HPA will make this mistake.
• Delivery and tractability are unresolved even if the biology were right. EAF2 has no structure at all, no verified ChEMBL record exists for any of the three, and the one high-resolution interface available (5JW9) is between ELL2 and AFF4 — the two proteins in this cluster that KMT2A::AFF1 cells need least. There is no starting point here that points at the fusion.

1. Do the three module members buffer each other? EAF1 and EAF2 are paralogues, and ELL and ELL2 are paralogues, so a single-gene knockout screen is the worst possible design for detecting a module-level requirement. A combinatorial EAF1/EAF2 and ELL/ELL2 double knockout in KMT2A::AFF1 lines is the experiment that could still overturn this node's negative conclusion — and nothing in the current data rules it out.
2. Does the ELL occludin-like domain bind the AFF-family ELLBow at the same site and with the same geometry as the ELL2 domain does in 5JW9? If the two paralogues present distinguishable surfaces, a paralogue-selective agent becomes conceivable and this module's structures become the template for it. If they are identical, any ELL-directed PPI agent will also hit ELL2.
3. Why does EAF1 have a real baseline requirement (mean -0.409, 34.7% of lines depleted) that gets progressively weaker in leukemia, then in KMT2A-r lines, then in KMT2A::AFF1 lines? Something about the KMT2A-fusion state relieves an EAF1 requirement present in other lineages, and identifying it might be more informative than the dependency itself.
4. Does EAF2 do anything at all in this compartment? It is the most bone-marrow-enriched member (67.8 nTPM), immune-cell-enhanced, and simultaneously the most dispensable gene in the batch with zero structures and the loosest constraint in the interactome. Either the marrow enrichment is functionally inert, or its function is fully redundant with EAF1 — and no experiment distinguishes these.
5. Is the TFIIF-site occupancy competitive in cells? If ELL2-EAF1 and TFIIF genuinely compete for the same Pol II lobe surface, then the ratio of the two should tune the initiation/elongation balance globally — and a fusion that raises local SEC concentration would shift that ratio at its target genes. That is a testable, fusion-relevant prediction that the structure implies and nobody has measured.

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

   None of ELL2, EAF1 or EAF2 is DepMap common-essential, and all three score positively or neutrally on selectivity in KMT2A-rearranged lines. ELL2 (O00472, 640 aa): mean Chronos -0.064 across 1,208 lines, 1.0% depleted, leukemia -0.033, KMT2A-r -0.043, KMT2A::AFF1 +0.003, selectivity_vs_leukemia -0.010, selectivity_vs_all_lines +0.021; pLI 0.9999, LOEUF 0.416, 21 observed versus 72.8 expected LoF, missense z 3.14; HPA low tissue specificity, detected in all, immune cell enhanced. EAF1 (Q96JC9, 268 aa): mean -0.409, 34.7% depleted, leukemia -0.180, KMT2A-r -0.119, KMT2A::AFF1 -0.040, selectivity_vs_leukemia +0.061, selectivity_vs_all_lines +0.290; pLI 0.0, LOEUF 1.047, 19 observed versus 26.6 expected LoF, missense z 1.76; HPA tissue enhanced with bone marrow 53.9 nTPM, detected in all, low immune cell specificity. EAF2 (Q96CJ1, 260 aa): mean -0.064, 0.9% depleted, leukemia -0.040, KMT2A-r +0.074, KMT2A::AFF1 +0.108, selectivity_vs_leukemia +0.114, selectivity_vs_all_lines +0.138; pLI 0.0, LOEUF 1.345 — the loosest in this interactome — with 29 observed versus 29.4 expected LoF and missense z 0.48; HPA tissue enhanced with bone marrow 67.8 and lymphoid tissue 50.0 nTPM, detected in many, immune cell enhanced.

   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 every magnitude here except EAF1's +0.290 against all lines is inside our declared +/-0.15 noise band and the strict reading is 'no selective requirement detected' rather than 'protective'. What carries the argument is the unanimity of sign across three independent genes, which we have not formally tested. Cell lines are not normal tissue, and a CRISPR knockout of one module member may be buffered by the others within the same complex — which is precisely the redundancy an inhibitor would also face. HPA nTPM values are RNA, not protein, and enrichment predicts exposure rather than requirement.

   ▸Evidence that argues against this (1)

   • All three genes are enriched in the bone marrow compartment (EAF2 67.8 nTPM, EAF1 53.9 nTPM) and two of three are immune-cell-enhanced, so an expression-driven pipeline would nominate this module as haematopoietically selective. That would be an error: EAF2 is the most marrow-enriched member and also the single most dispensable gene in the batch by both DepMap and gnomAD. Enrichment and requirement are orthogonal. Human Protein Atlas 2026
2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   Cryo-EM structures of an RNA polymerase II-DSIF active elongation complex containing ELL2 (1-640) and EAF1 (1-268) — 7OKX at 3.30 A and 7OKY at 4.14 A — show that the ELL2-EAF1 pair binds the polymerase lobe at the same site as TFIIF and stimulates transcription allosterically rather than by contacting the catalytic centre. ELL2's entry into the super elongation complex is separately defined by 5JW9 (X-ray, 2.00 A), the ELL2 C-terminal occludin-like domain (519-640) bound to the AFF4 ELLBow (301-351), and an isolated ELL2 ELL_N2 domain (200-292) exists as 2E5N (NMR). EAF2 (Q96CJ1) has ZERO PDB entries.

   Allosteric transcription stimulation by RNA polymerase II super elongation complex · Chen Y, Vos SM, Dienemann C, et al. · Molecular Cell 2021
   PMID 34265249DOI
   ▸What this evidence cannot show

   7OKX/7OKY are reconstituted in vitro complexes containing no KMT2A fusion protein and no chromatin, so nothing in them is fusion-specific; at 3.30 A and 4.14 A the interface is defined at backbone rather than side-chain precision, especially for 7OKY. The ELL-family subunit is ELL2, not ELL, so applying this mechanism to ELL is a paralogue extrapolation. The EAF partner is EAF1 only; EAF2 is absent from this and every other structure. The 5JW9 and 2E5N identifiers, resolutions and residue ranges were resolved against the RCSB APIs; 5JW9 pairs ELL2 with AFF4 rather than with AFF1, so it is also a paralogue proxy for the fusion setting. The EAF2 zero-structure statement is an absence-of-evidence claim scoped to an RCSB API search in this session.

   ▸Evidence that argues against this (1)

   • The structures contain ELL2 and EAF1. This module's relevance to KMT2A fusions runs through ELL, a paralogue, so the architecture is a template rather than an observation of the complex in question. RCSB PDB 2026
3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   No ChEMBL target record for ELL2, EAF1 or EAF2 was retrieved and verified in this cycle, so no chemical matter can be attributed to any member of this module. For context within the same dossier, the closely related ELL (CHEMBL5725004) has only 3 potent rows, all traceable to the BET inhibitor molibresib as a super-elongation-complex co-enrichment artefact rather than direct binding.

   ChEMBL database version 34: no verified target record obtained for ELL2, EAF1 or EAF2 in this cycle; comparator target CHEMBL5725004 (RNA polymerase II elongation factor ELL) queried by target_search, get_bioactivity and get_mechanism · EMBL-EBI ChEMBL · ChEMBL 2026
   ▸What this evidence cannot show

   This is explicitly a statement about what we verified, not about what exists: unlike MED26, ZFP64 and MEAF6 — for which we positively confirmed that no ChEMBL target entity exists — we simply did not obtain a verified ELL2, EAF1 or EAF2 record this cycle. A target entity for one or more of them may well exist. Confidence is therefore 'moderate' and the claim should be re-run before it is relied on. Nothing here should be read as a positive finding of unliganded status.

   ▸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. Recording that no target was retrieved AND VERIFIED this cycle is the right epistemic form and it is not the same as absence: a target may exist and not have been reached by the queries run. ChEMBL, EMBL-EBI 2026
4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   The comparable reconstituted activated Pol II elongation complex without the ELL/EAF module — Pol II-DSIF-PAF1c-SPT6, PDB 6GMH at 3.10 A — establishes the general architecture of an activated elongation complex and the experimental genre within which 7OKX/7OKY sit.

   Structure of activated transcription complex Pol II-DSIF-PAF-SPT6 · Vos SM, Farnung L, Boehning M, et al. · Nature 2018
   PMID 30135578DOI
   ▸What this evidence cannot show

   Cited here only to establish the experimental genre and the general architecture of an activated elongation complex. It contains PAF1c and SPT6, not the ELL/EAF module, and contains no KMT2A fusion protein; no claim about ELL, ELL2, EAF1 or EAF2 should be drawn from it.

   ▸Evidence that argues against this (1)

   • 6GMH is the comparator without the module, so it establishes the general architecture and by construction says nothing about what ELL/EAF adds. Using it as context risks reading the ELL2/EAF1 difference as larger or smaller than the two reconstitutions can support, since they differ in more than the module. RCSB PDB 2026

• ← related toSeparation-of-function test: does KMT2A::AFF1 need its C-terminal homology domain to dimerise?

  Counter-evidence to the design's prior (Luo 2001, PMID 11463848): in the closest published analogue, MLL-ELL, the partner's canonical ELONGATION domain was dispensable for immortalisation while its EAF1-interaction domain was essential, and a heterologous MLL-EAF1 fusion caused AML with shorter latency. If AFF1 behaves like ELL, the arm that separates is the 761-775 AF9/ENL motif this design lists as a control.

• ← related toELL2 — domain architecture and residue coordinates

  paralogue caution: ELL's OCEL domain is 507-617, ELL2's is 526-636; coordinates are not interchangeable

• ← part ofELL2 — a SEC subunit that almost no cell needs, next to a paralogue almost every cell needs

  ELL2-EAF1 is resolved in an active Pol II elongation complex (7OKX)

• ← related toThe little elongation complex (ELL + ICE1 + ICE2 + ZC3H8) — the other place ELL works, and why ELL's pan-essentiality is probably not about the fusion

  the second complex ELL-family factors work in, outside SEC

• ← part ofEAF1 — the interactor with an actual subnuclear phenotype: an MLL fusion evicts it from Cajal bodies, and the bodies themselves fall apart

  the EAF half of the ELL/EAF activation module the graph already carries

• ← part ofEAF2 (U19) — the EAF paralogue whose ELL-binding domain is a mapped tumour-suppressor module, described entirely in prostate cancer

  the second EAF-family partner of the ELL/EAF activation module

• part of →Mechanism class 1 — chromatin and elongation hijack (KMT2A-r, NUP98-r)

  general elongation activation module carried into the fusion complex

• ← part ofELL — rank 7 on selectivity and the deepest common-essential in the top ten, with one NMR fragment to its name
• ← related toATM phosphorylates ELL at T551, S561 and S589 - inside the OCEL domain this graph already calls essential for MLL-ELL immortalisation

  kinase input to the ELL-EAF module

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

  the ELL/EAF activation module is in SEC and is the module NOT detected in SEC-L2 or SEC-L3

• ← related toELL: the discarded N-terminal EAF-binding site, and a second subtraction the graph had not recorded — the fusion delocalises endogenous EAF1

  refines the EAF module picture: two docking sites, only one retained

• supports →KMT2A::ELL (MLL-ELL), t(11;19)(q23;p13.1)

  ELL OCEL/occludin module (aa 507-617) carries the EAF1 contact that Luo 2001 showed is essential for MLL-ELL leukemogenesis