AFDN / AF6 — domain architecture and residue coordinates
also called AF6 · MLLT4 · afadin · P55196
AFDN is a cell-junction protein that becomes a KMT2A fusion partner, and unusually for a partner it normally lives outside the nucleus. This entry pins down where its parts sit, and the surprise is that the small module at its very front end — not anything larger — is the piece that makes the fusion cancerous.
The same thing, for a scientist
AFDN (UniProt P55196, 1,824 aa) is afadin, an F-actin-binding adherens-junction scaffold and the commonest cytoplasmic KMT2A fusion partner. UniProt and InterPro agree on a well-resolved multi-domain architecture: two Ras-associating domains (RA1 39-133, RA2 246-348), an FHA domain (426-492), a dilute/cargo-binding domain (668-908), a PDZ domain (1007-1093) and three coiled-coil segments. Liedtke and colleagues showed that RA1 alone is the minimal AF6 region sufficient for MLL-AF6 myeloid immortalisation and short-latency leukemogenesis, and that it works by mediating self-association rather than by binding Ras.
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.
I have never read one of these pictures before
A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.
Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.
Anything in yellow sticks is a bound small molecule, which is usually the drug.
Coordinates, and how firm each one is
AFDN is the easy case in this beat: a genuinely folded multi-domain protein where UniProt curation and InterPro HMM hits land within a few residues of each other 12.
| Domain | UniProt | InterPro/Pfam | Spread |
|---|---|---|---|
| RA1 | 39-133 | 20-131 (IPR061275); PF00788 40-133 | N-terminal edge differs by 19 aa |
| RA2 | 246-348 | 247-347 | 1 aa |
| FHA | 426-492 | 384-502 (IPR061279); PF00498 428-492 | model-dependent |
| Dilute | 668-908 | 611-932 cargo-binding (IPR037977); PF01843 DIL 785-887 | large, nested models |
| PDZ | 1007-1093 | 1005-1092; PF00595 1010-1089 | 2-5 aa |
| Afadin domain | — | 1518-1651 (IPR060075) | InterPro only |
The two disagreements worth naming are RA1 and the dilute region. InterPro's afadin-specific RA1 model (IPR061275) starts at 20, 19 residues earlier than UniProt's 39. Because RA1 is the functionally load-bearing module here, that 19-residue difference is the one most likely to matter in a figure, and we record both rather than picking. For the dilute region, PF01843 (785-887) is the tight structural DIL model while IPR037977 (611-932) is a larger afadin cargo-binding region that contains it; we draw the UniProt 668-908 and note the nesting.
The PDZ domain has four independent experimental structures in the 1001-1097 range (PDB 1T2M, 1XZ9, 2AIN, 2EXG, 7QCR) 3, so it is the only domain here whose boundaries are directly observed rather than modelled.
RA1 is the whole oncogenic story, and it is not about Ras
Liedtke and colleagues mapped the minimal AF6 region required for MLL-AF6 activity and landed on RA1 at the amino terminus 4. Three results in that paper together make the point:
- RA1 alone was sufficient for myeloid progenitor immortalisation in vitro and short-latency leukemogenesis in vivo.
- Single substitutions that abolish the AF6-Ras interaction did not abrogate MLL-AF6 oncogenesis — so the Ras-binding function of a Ras-associating domain is not what is being used.
- Fusing MLL to heterologous RA domains from c-Raf1 or RalGDS, or to constitutively active K-RAS, H-RAS or RAP1, was not sufficient. Only AF6's own RA1 worked.
What RA1 does supply is efficient self-association. That places MLL-AF6 in the oligomerisation class of KMT2A fusions rather than the elongation-hijack class, and it is a clean, independent instance of the pattern this graph tracks under hyp-oligomerisation-interfaces-as-mutation-resistant-targets 4.
The negative results in point 3 are the strongest part of the paper, because they exclude the obvious alternative explanations. They also mean the interface is specific to RA1's own surface: a generic dimerisation module will not substitute, which implies a druggable, sequence-specific self-association surface rather than a generic avidity effect.
What the fusion keeps
KMT2A::AFDN is a reciprocal t(6;11)(q27;q23) product. The AFDN contribution begins near the AFDN N-terminus, so the fusion retains RA1 and, depending on the junction, RA2 and downstream sequence; the mapping work shows RA1 is the part that matters and that additional AF6 sequence is not required 4.
We have set fate: "retained" for RA1 and RA2 and left the rest as "variable" rather than asserting a retention status we cannot support at the exon level for every reported junction. The Frankfurt recombinome series reports AFDN among the recurrent partners but the junction distribution for this partner is thin compared with AFF1 5.
A caution for the viewer: AFDN is 1,824 aa, so on a residue-scaled axis the functionally decisive module (RA1, 39-133) occupies about 5% of the lane. The figure will look as though almost nothing matters, which is in this case exactly the correct impression.
| uniprot | P55196 |
|---|---|
| length AA | 1824 |
| gene symbol | AFDN |
| alphafold ID | AF-P55196-F1 |
| isoform note | Coordinates are in UniProt P55196-1 (l-afadin, 1,824 aa) numbering. s-afadin is shorter and its numbering is not interchangeable. |
| assumed fusion for fate | fusion-kmt2a-afdn (KMT2A::AFDN, t(6;11)(q27;q23)), AFDN as the 3' partner |
| domain coordinate caveat | AFDN is well folded and its coordinates are the most reliable in this beat, with two exceptions worth drawing carefully. RA1 — the only domain that matters for leukemogenesis — has a 19-residue disagreement at its N-terminal edge between UniProt (39) and InterPro's afadin-specific model (20). The dilute region is annotated by three nested models spanning 611-932 down to 785-887. Coiled-coil segments and the C-terminal F-actin-binding tail are prediction-derived and soft-edged. Only the PDZ domain has boundaries confirmed by experimental structures. |
PDB IDS
[ "1T2M", "1XZ9", "2AIN", "2EXG", "5A6C", "7QCR" ]
domains — the numbers behind the chart
[
{
"name": "RA1 (Ras-associating domain 1)",
"start": 39,
"end": 133,
"pfam": "PF00788",
"note": "THE oncogenic module. Minimal AF6 region sufficient for MLL-AF6 immortalisation and leukemogenesis; acts by self-association, not by binding Ras. InterPro's afadin-specific model (IPR061275) starts at 20, 19 residues earlier than UniProt's 39 — both recorded.",
"fate": "retained",
"evidence_basis": "curated+functional",
"boundary_precision": "N-terminal edge uncertain by ~19 aa (20 vs 39)"
},
{
"name": "RA2 (Ras-associating domain 2)",
"start": 246,
"end": 348,
"pfam": "PF00788",
"note": "UniProt 246-348 and InterPro 247-347 agree to within a residue. Not required for MLL-AF6 transformation.",
"fate": "retained",
"evidence_basis": "curated",
"boundary_precision": "+/- 1 aa"
},
{
"name": "CC1 (coiled coil)",
"start": 146,
"end": 185,
"note": "First of three coiled-coil segments; sits between RA1 and RA2. Not implicated in the self-association that drives MLL-AF6.",
"fate": "retained",
"evidence_basis": "prediction",
"boundary_precision": "coiled-coil predictions are soft-edged"
},
{
"name": "FHA (forkhead-associated domain)",
"start": 426,
"end": 492,
"pfam": "PF00498",
"note": "UniProt 426-492; Pfam 428-492; the broader InterPro afadin FHA region (IPR061279) runs 384-502.",
"fate": "variable",
"evidence_basis": "curated",
"boundary_precision": "+/- 2 aa core, wider model-dependent region"
},
{
"name": "DIL (dilute / cargo-binding domain)",
"start": 668,
"end": 908,
"pfam": "PF01843",
"note": "UniProt 668-908. Nested models: Pfam DIL is tighter at 785-887; InterPro afadin cargo-binding region is wider at 611-932. The three do not agree and none is obviously right.",
"fate": "variable",
"evidence_basis": "curated",
"boundary_precision": "poor; three models span 611-932 to 785-887"
},
{
"name": "PDZ",
"start": 1007,
"end": 1093,
"pfam": "PF00595",
"note": "The only AFDN domain with direct experimental structures (1T2M, 1XZ9, 2AIN, 2EXG, 7QCR, all in 1001-1097). Binds C-terminal peptide ligands at adherens junctions.",
"fate": "variable",
"evidence_basis": "structure",
"boundary_precision": "+/- 5 aa; several independent structures"
},
{
"name": "Afadin domain",
"start": 1518,
"end": 1651,
"pfam": "PF26705",
"note": "InterPro IPR060075 / Pfam PF26705, an afadin-specific domain UniProt does not annotate. Overlaps the third coiled coil (1523-1667).",
"fate": "variable",
"evidence_basis": "hmm_hit",
"boundary_precision": "HMM only; not curated by UniProt"
},
{
"name": "CTD (C-terminal disordered tail)",
"start": 1569,
"end": 1824,
"note": "UniProt annotates 1569-1824 as a single disordered region — the largest disordered stretch in the protein. l-afadin's F-actin-binding activity is conventionally attributed to this C-terminal tail, but we found no verified residue-level mapping of the actin-binding site, so we have not drawn one.",
"fate": "variable",
"evidence_basis": "prediction",
"boundary_precision": "soft; these are disorder-prediction edges, and no actin-binding site is mapped inside them"
}
]conflicting annotations
[
{
"feature": "RA1 N-terminal boundary",
"claim_a": "UniProt: 39-133",
"claim_b": "InterPro IPR061275 (afadin-specific RA1): 20-131",
"our_reading": "We draw 39-133 to match UniProt, the source the rest of this graph uses, and flag the 19-residue ambiguity. Since RA1 is the functional module, anyone designing a construct should test both boundaries rather than trusting either."
}
]- Wild-type afadin is an essential adherens-junction scaffold; anything that disrupts the RA1 surface risks hitting junction biology in epithelium and endothelium, which is a different toxicity profile from the haematopoietic-restricted targets elsewhere in this graph.
- A self-association interface has no small-molecule precedent as clean as a menin pocket. If RA1 dimerisation is high-avidity and surface-extended, it may simply not be druggable with small molecules, and the honest fallback is degradation of the fusion rather than interface blockade.
- AFDN is an uncommon partner, so even a perfect RA1 inhibitor addresses a small patient population and would struggle to reach a registrational trial on its own.
- Which residues of RA1 form the self-association interface? Liedtke et al. showed RA1 self-associates but did not map the surface, so the most druggable interface in this fusion currently has no coordinates at all — an AlphaFold-Multimer or Boltz co-folding run of RA1 against itself would be a cheap first pass.
- Why will no heterologous RA domain substitute for AF6 RA1? If generic dimerisation were enough, c-Raf1 RA should have worked. The specificity implies RA1 contributes something beyond avidity, and identifying it would sharpen the oligomerisation hypothesis across the whole fusion family.
- MLL-AF6 is a cytoplasmic-partner fusion, yet it drives a nuclear transcriptional program. Where in the cell does the self-association actually happen, and does AFDN's junctional localisation sequester any of the fusion away from chromatin?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
UniProt P55196 (AFAD_HUMAN, afadin) is 1,824 aa with Ras-associating domains at 39-133 and 246-348, an FHA domain at 426-492, a dilute domain at 668-908, a PDZ domain at 1007-1093 and coiled coils at 146-185, 1408-1448 and 1523-1667.
UniProtKB entry P55196 (AFAD_HUMAN), afadin / AFDN · UniProt Consortium · UniProtKB 2026What this evidence cannot show
Retrieved 2026-09-18. AFDN has multiple splice isoforms (l-afadin and the shorter s-afadin); P55196-1 is the long form and all coordinates here are in that numbering. Applying them to s-afadin numbering would be wrong.
Evidence that argues against this (1)
- Two of the three coiled coils sit inside regions that other annotations in the same two databases say are something else. UniProt's own Region 128-194 'Disordered' completely contains the 146-185 coiled coil, so the entry asserts a 40-residue helical bundle inside a stretch it predicts as disordered; and the 1523-1667 coiled coil overlaps InterPro IPR060075 / Pfam PF26705 'Afadin domain' at 1518-1651 by 129 residues, i.e. a folded domain that UniProt does not annotate at all occupies almost the whole of UniProt's longest coiled coil. Which of the two an oligomerisation argument can lean on is not settled by either record. UniProt Consortium 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
InterPro places the afadin-specific RA1 model (IPR061275) at 20-131 and RA2 (IPR061274) at 247-347, an FHA region (IPR061279) at 384-502, an afadin cargo-binding region (IPR037977) at 611-932 containing the tighter Pfam DIL model at 785-887, the PDZ domain at 1005-1092, and an afadin-specific domain (IPR060075) at 1518-1651 not annotated by UniProt.
InterPro entry-to-protein matches for UniProt P55196 · InterPro / EMBL-EBI · InterPro 2026What this evidence cannot show
HMM hit boundaries, not experimental edges. The RA1 discrepancy (20 vs 39) reflects that InterPro's afadin-specific model includes N-terminal flanking sequence that UniProt's generic RA-domain call excludes; neither is 'the' answer. Retrieved 2026-09-18.
Evidence that argues against this (1)
- InterPro's own member databases disagree with that FHA number by more than any of them disagree with UniProt. In the same fetched match set the FHA region is 384-502 under IPR061279 / CDD cd22711, 425-492 under UniProt and InterPro IPR000253, 428-492 under Pfam PF00498, 406-502 under SUPERFAMILY SSF49879 / InterPro IPR008984, and 425-477 under SMART SM00240 - a 44-residue spread at the N-terminus and 25 at the C-terminus, giving domain lengths from 53 to 119 residues, more than double. Quoting the widest member (384-502) as 'the InterPro FHA region' picks one of six without saying so. InterPro Consortium; EMBL-EBI 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
The AFDN PDZ domain has been solved independently several times in the 1001-1097 range, including PDB 1T2M, 1XZ9, 2AIN, 2EXG and 7QCR.
PDB entry 7QCR — MLLT4/Afadin PDZ domain in complex with the C-terminal peptide from protein E of SARS-CoV-2, and the UniProt P55196 PDB cross-reference set · RCSB Protein Data Bank; UniProt Consortium · RCSB Protein Data Bank 2022What this evidence cannot show
Entry title and the 1002-1097 chain range for 7QCR were confirmed against the RCSB data API on 2026-09-18. Crystallised construct boundaries are chosen for expression and crystallisation and are an upper bound on the folded domain, not a definition of it. None of these structures concerns leukemia; the PDZ domain is retained in the fusion only for junctions far downstream and is not the oncogenic module.
Evidence that argues against this (1)
- 1001-1097 is the union of five different constructs, not a range any one of them covers, and four of the five are not independent crystal determinations. SIFTS gives 1T2M 1003-1094, 1XZ9 1001-1096, 2AIN 1003-1094, 2EXG 1001-1096 and 7QCR 1002-1097 - and RCSB records 1T2M, 1XZ9, 2AIN and 2EXG all as SOLUTION NMR with 20 models each, while 1XZ9 and 2EXG come from the same primary citation and 1T2M and 2AIN share construct boundaries. The only crystal structure, 7QCR (2.28 A), has coordinates for author residues 1-22 and 27-91 of chain A, so residues corresponding to UniProt 1025-1027 are unmodelled and nothing beyond about 1092 is observed. PDBe / SIFTS et al. 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The AF6 RA1 domain is the minimal region sufficient for MLL-AF6-mediated myeloid immortalisation and short-latency leukemogenesis; substitutions abolishing AF6-Ras binding do not abrogate oncogenesis, heterologous RA domains and activated RAS do not substitute, and RA1 efficiently mediates self-association.
Self-association mediated by the Ras association 1 domain of AF6 activates the oncogenic potential of MLL-AF6 · Liedtke M, Ayton PM, Somervaille TCP, Smith KS, Cleary ML · Blood 116:63-70 2010What this evidence cannot show
Retroviral overexpression of MLL-AF6 constructs in mouse bone marrow progenitors, not an endogenous or patient-derived model, and immortalisation assays reward self-renewal specifically. The paper does not report residue boundaries for the self-association interface itself, so we cannot draw a sub-domain box inside RA1. The RA1 construct boundaries used are not identical to the UniProt domain call.
Evidence that argues against this (1)
- The Frankfurt recombinome series records AFDN as a recurrent but uncommon partner, so the clinical weight behind this mechanism is much lighter than for AFF1, and the self-association model has not been tested across the range of observed t(6;11) junctions. Meyer C et al. 2023
- E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
AFDN is one of the recurrent KMT2A translocation partner genes catalogued in 3,401 acute leukemia patients analysed between 2003 and 2022, within a set where the seven commonest partners plus partial tandem duplications account for more than 90% of recombinations.
The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, Burmeister T, Gröger D, Sutton R, et al. · Leukemia 37:988-1005 2023What this evidence cannot show
A referral-based diagnostic cohort from a single centre network, so partner frequencies reflect referral patterns rather than a population distribution.
Evidence that argues against this (1)
- A treatment-protocol cohort assembled independently of any referral laboratory gives a different partner mix. The Ponte-di-Legno group pooled 629 non-infant childhood KMT2A-rearranged ALL cases from 17 study groups treated 1995-2010: among the 530 typed cases AFF1 273 (51.5%), MLLT1 106 (20.0%), MLLT3 76 (14.3%), AFDN 20 (3.8%) and MLLT10 14 (2.6%), with a partner gene identifiable in only 84.3% of patients. Against the Frankfurt series' own ALL column (AFF1 56.5%, MLLT1 18.5%, MLLT3 11.8%, MLLT10 3.7%, AFDN 2.1%), AFDN is roughly 1.8-fold commoner here and overtakes MLLT10, inverting the rank order - and 15.7% of cases had no partner assigned at all. Attarbaschi A et al. 2023
- related to →KMT2A::AFDN (MLL-AF6), t(6;11)(q27;q23)
Counter-evidence added cycle 41: a treatment-protocol cohort gives a different partner mix. Among 530 typed non-infant childhood KMT2A-r ALL cases from 17 study groups, AFDN is 20 (3.8%) and overtakes MLLT10 at 14 (2.6%), inverting the rank order against the Frankfurt ALL column (AFDN 2.1%, MLLT10 3.7%), with 15.7% of patients having no partner assigned (Attarbaschi 2023, PMID 36256911).
domain architecture of the 3' partner; RA1 (39-133) is the minimal oncogenic module
AF6 RA1 transforms by self-association, not by Ras binding; heterologous RA domains do not substitute
- fuses with →KMT2A / MLL1 — the gene that breaks
AFDN is the commonest cytoplasmic KMT2A fusion partner, t(6;11)(q27;q23)
the decisive module sits in the first 7% of a 1824-residue partner, so partner-side break position is nearly irrelevant here
two cytoplasmic partners that both transform by self-association rather than by elongation hijack
- ← related toSeparation-of-function test: does KMT2A::AFF1 need its C-terminal homology domain to dimerise?
design follows the AF6 RA1 minimal-region precedent
partner-side domain coordinates for the 3D architecture view
junction falls before RA1 at residue 39, so no annotated domain is discarded