Oncofusion

start / graph / protein-aff1-domains

Proteinmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q2b-atlas2

AFF1 / AF4 — domain architecture and residue coordinates

also called AF4 · MLLT2 · ALF transcription elongation factor 1 · P51825

AFF1 is the partner protein most often stuck onto the broken KMT2A gene in infant leukemia, and this entry pins down where each of its working parts sits along the protein chain. Most of AFF1 is a floppy, shapeless string, so several of its "domains" are conventions drawn on a diagram rather than things anyone has measured.

The same thing, for a scientist

AFF1 (UniProt P51825, 1,210 aa canonical) is the scaffold subunit of the super elongation complex and the commonest KMT2A fusion partner in infant B-ALL. UniProt annotates **no** named domains for it: 87% of the chain falls inside three annotated disordered regions (1-45, 73-314, 366-957). The only residue-level features that come from measurement rather than convention are the CATH-Gene3D structured N-terminal element (5-78), the Pfam/InterPro AF4 interaction motif that binds the AF9/ENL ANC1 homology domain (761-775, NMR construct 738-779, PDB 2LM0), the Pfam AFF C-terminal homology domain (949-1209), and a GAL4-fusion transactivation domain mapped to residues 480-560. The nuclear localisation signal, which every review draws as a discrete box, has no published residue assignment we could verify.

What it looks like
AFF11,210 aa
ALF/ALF/NHD (N-terminal homology domain) · residues 5–78 · lost in the fusion · Only structured element in the N-terminal half; CATH-Gene3D G3DSA:6.10.250.2670 hit. Reviews conventionally extend the ALF domain to ~aa 100. The paralogous AFF4 N-terminal region (aa 2-73) is the Cyclin T1/P-TEFb-binding element resolved in PDB 4IMY, so this is the likely P-TEFb contact. Carries the SIAH1/2-sensitive N-terminal portion (not residue-mapped).pSERpSER (serine-rich low-complexity core) · residues 423–497 · retained in the fusion · Contiguous UniProt compositional-bias features: acidic 423-439, Pro-rich 442-453, low-complexity 454-472, acidic 473-497. A 50-residue sliding scan of P51825 puts peak serine density at 441-510 (Ser fraction 0.36-0.44). This is a composition boundary, not a fold boundary — it has no edges in any physical sense.TAD TAD (transactivation domain) · residues 480–560 · retained in the fusion · Smallest AF4 fragment reported as a strong activator in a GAL4-fusion reporter assay (Prasad 1995). Overlaps the pSER core. An upper bound on extent, not a mapped boundary.NLS (nuclear localisation signal, predicted) · residues 603–614 · retained in the fusion · PREDICTION ONLY. No AFF-family member carries an NLS feature in UniProt and we found no verifiable primary mapping study. Sequence KQPKKPVKASAR, the strongest basic cluster in the region where reviews place the NLS. Draw this differently from the measured features, or omit it.AF9/ENL-binding motif · residues 761–775 · retained in the fusion · Pfam 'AF4 interaction motif' / InterPro IPR043639. This — NOT the CHD — is the element that binds the AF9 and ENL ANC1 homology domains. The NMR construct in PDB 2LM0 spans 738-779; the conserved core is 761-775.CHD (C-terminal homology domain)CHD (C-terminal homology domain) · residues 949–1209 · retained in the fusion · Pfam PF18876 / InterPro IPR043640. An AFF-family DIMERISATION domain: the AFF4 equivalent (899-1163) is a novel eight-helix fold, solved at 2.2 A as PDB 6R80, that mediates AFF4 homodimerisation and AFF1-AFF4 heterodimerisation, binds RNA and DNA, and carries a CDK9-substrate surface loop. Widely but, on the structural evidence, incorrectly described as the AF9/ENL-binding domain; see the AF9/ENL-binding motif entry above. Fully retained in every observed KMT2A::AFF1 break, so the fusion carries an oligomerisation interface.05001,000
loading AF-P51825-F1…
AlphaFold AF-P51825-F1 ↗
colourshape

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.

very high (90+)confident (70–90)low (50–70)very low (<50)
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.

Go deeper
Why this entry exists, and what it is honest about6 deeper

The dashboard's 3D domain viewer draws residue position to scale. AFF1 is the single most important partner in this project and it had no data.domains array at all, so its lane fell back to nominal ordered blocks.

Filling that gap turned out to be less straightforward than for a folded protein. UniProt P51825 carries zero Domain features. It carries three Region features of type Disordered covering residues 1-45, 73-314 and 366-957, plus one functional region (601-900, required for recruitment to DNA-damage sites and for PARP1 interaction) 1. That is the whole structured annotation.

So every boundary below is labelled with how it was obtained:

FeatureBoundaryWhere the number comes from
ALF/NHD5-78CATH-Gene3D HMM hit G3DSA:6.10.250.2670 on P51825 2
pSER low-complexity core423-497UniProt compositional-bias features, contiguous 1
TAD480-560GAL4-fusion reporter assay, Prasad 1995 3
NLS603-614 (predicted)Sequence prediction made for this graph. Not a measurement. 6
AF9/ENL-binding motif761-775Pfam PF18875 / InterPro IPR043639 2; NMR construct 738-779, PDB 2LM0 4
CHD949-1209Pfam PF18876 / InterPro IPR043640 2

Treat the first, fourth and fifth of those as soft-edged. Treat the CHD and the AF9-binding motif as hard, because they are HMM-defined on a family alignment and, in the motif's case, backed by a solved complex.

A correction the brief for this beat needs

The standing description in this project — and in much of the review literature — says that the C-terminal homology domain binds AF9 and ENL. The structural evidence says otherwise.

The solved AF9-binding element of AFF1 is a short motif in the middle of the protein, not the CHD. PDB 2LM0 is the solution NMR structure of the AF9 ANC1 homology domain bound to an AFF1 peptide, and its chain A spans AFF1 residues 738-779 45. Pfam and InterPro have since promoted the conserved core of that element to its own entry, AF4 interaction motif (PF18875 / IPR043639), mapped to P51825 residues 761-775 2. The CHD (949-1209) is a separate, C-terminal element.

This matters for the viewer and for target selection. If a figure labels the CHD as the AF9/ENL interface, it puts a drug-relevant surface roughly 200 residues away from where the structure says it is. It also matters for breakpoint reasoning: every common KMT2A::AFF1 junction retains both 761-775 and 949-1209, so no clinically observed break separates them — which is precisely why the mislabelling has survived without being caught by a phenotype.

We have recorded both the conventional claim and the structural one in data.domains[].note rather than silently overwriting the conventional one.

What the CHD actually is: a dimerisation domain the fusion keeps

Having established that the CHD is not the AF9/ENL interface, the obvious next question is what it is. There is an answer, and it is more interesting than the thing it was mistaken for.

Chen and Cramer solved the AFF4 CHD at 2.2 Å (PDB 6R80, chain A spanning AFF4 899-1163 — the positional equivalent of AFF1 949-1209) and characterised it biochemically 9. It folds into a novel eight-helix domain distantly related to tetratricopeptide-repeat motifs, and it mediates AFF4 homodimerisation or AFF1-AFF4 heterodimerisation. Fluorescence anisotropy indicates it binds both RNA and DNA, and a surface loop in it is a substrate for CDK9.

Three consequences for this project:

  1. KMT2A::AFF1 retains an oligomerisation interface. The dimerisation module is at the far C-terminus and every observed break is far upstream of it, so the fusion keeps it in full. Oligomerisation has been treated in this graph as the mechanism of the rare cytoplasmic partners — AFDN's RA1, EPS15's coiled coil. It is also, apparently, available to the commonest partner of all.
  2. The fusion can heterodimerise with wild-type AFF4. If KMT2A::AFF1 dimerises through its CHD with endogenous AFF4, the fusion is not a free-standing unit but is physically coupled to the intact super elongation complex scaffold. That is a different picture from 'the fusion recruits SEC' and it predicts a stoichiometric dependence on AFF4 levels.
  3. It is a candidate interface. An eight-helix folded domain with a defined dimerisation surface is a far more tractable target than a coiled coil or a disordered motif, and it is retained in essentially 100% of KMT2A::AFF1 patients.

The caveat is that all of this is AFF4 structural work. The heterodimerisation result explicitly involves AFF1, and the CHD is conserved across AFF1-4, but no AFF1 CHD structure exists and the fusion protein's CHD behaviour has not been tested.

The nuclear localisation signal is the weakest number here

Every schematic of AF4 shows an NLS box between the serine-rich region and the C-terminus. We could not find a primary paper that assigns it residue numbers, and no member of the AFF family — AFF1 P51825, AFF2 P51816, AFF3 P51826, AFF4 Q9UHB7 — carries an NLS motif feature in UniProt 16.

What we can say from the sequence itself: scanning P51825 in 50-residue windows, the only basic-residue density peak in the region where schematics place the NLS is at 601-650 (Lys+Arg fraction 0.28), and within it the sequence K603-QPK606-K607-PVK610-ASAR614 has the shape of a classic bipartite signal 6.

We have entered that as start: 603, end: 614 with evidence_basis: "prediction" and confidence weak. Anyone using this coordinate in a figure should draw it differently from the others, or not at all. This is the single boundary in this entry we would most expect to be revised.

Which numbering, and a trap worth naming

All coordinates here are in UniProt P51825-1 numbering, 1,210 aa.

Ensembl's canonical transcript for AFF1, ENST00000395146 (ENSP00000378578), encodes 1,218 aa — eight residues longer 7. Any coordinate quoted against the Ensembl canonical protein and pasted into a UniProt-numbered figure will be displaced. Eight residues is small enough to look like rounding and large enough to move a 15-residue motif off its own footprint.

UniProt also describes two further isoforms, P51825-2 (differing at residues 1-5 and at 1096) and P51825-3 (differing at 1096) 1. Neither is the displayed sequence and neither is used here.

A 300-residue ambiguity at the AFF1-side junction that this graph currently papers over

This graph's fusion-kmt2a-aff1 node lists two breakpoints, bp-kmt2a-intron-11-aff1-e4 and bp-kmt2a-intron-9-10-aff1-e4, both joining to AFF1 exon 4, and both list the ALF/N-terminal homology domain as retained. This entry lists it as lost. We looked at why, and the disagreement turns out to be real and consequential rather than a slip.

We mapped the coding exons of the Ensembl canonical AFF1 transcript ENST00000395146 to codon ranges 10:

ExonCoding ntCodons
1—non-coding
2381-13
312113-53
490054-353
545354-368

So which exon 4 is meant decides the answer, and the two readings differ by three hundred residues:

  • If "AFF1 exon 4" means Ensembl exon 4, the AFF1 contribution starts at codon 54 — inside the ALF/NHD (5-78). The fusion would keep roughly the last 25 residues of that domain and lose the first 53. Neither "retained" nor "lost": a fragment, and a fragment of a 74-residue structured element is not a functional domain.
  • If the numbering counts from the first coding exon — a common convention, and Ensembl exon 1 here is entirely non-coding — then "exon 4" is Ensembl exon 5 and the AFF1 contribution starts at codon 354, cleanly downstream of the ALF/NHD, which is then wholly lost.

We cannot resolve this from the sources we verified. Meyer and colleagues characterise the KMT2A side of the junction in detail and the partner side barely at all, and the per-gene exon tables that would settle it sit in a supplementary Excel file we could not read.

We keep fate: "lost" because the functional literature supports it — Bursen and colleagues place the SIAH-recognised element in "the N-terminal portion of AF4", and that literature describes the fusion throughout as lacking the AF4 N-terminus 8 — but fate_caveat records the ambiguity. The honest position is that the AFF1-side junction residue is not known to us to better than ±300 residues. For the most important fusion in this project, that is the largest single coordinate gap this beat found.

Note also that both listed breakpoints join the same AFF1 exon while differing on the KMT2A side. If that is accurate, partner-side domain content is invariant across KMT2A break position — consistent with the observation elsewhere in this entry that no observed break separates the 761-775 motif from the CHD.

What the KMT2A::AFF1 fusion keeps and loses on the AFF1 side

Subject to the junction ambiguity above, the break in AFF1 falls in the 5' part of the gene, so the fusion protein keeps the C-terminal majority of AFF1 and discards its N-terminus.

Lost: the ALF/N-terminal homology domain (5-78) — wholly on one reading of the junction, all but its last ~25 residues on the other. This is not a cosmetic loss. The N-terminal portion of AF4 carries the element recognised by the E3 ubiquitin ligases SIAH1 and SIAH2, and Bursen and colleagues showed that removing it protects the fusion protein from proteasomal degradation 8. A fusion that cannot be turned over is a fusion whose dose is set by transcription alone.

Retained: the serine-rich low-complexity core, the transactivation domain (480-560), the predicted NLS, the AF9/ENL-binding motif (761-775) and the CHD (949-1209).

One caution about 8. The protein Bursen et al. showed to be stabilised is AF4·MLL, the reciprocal der(4) product, not MLL·AF4. The directionality of the argument as applied to MLL·AF4 — which retains the same AFF1 C-terminal portion and also lacks the AF4 N-terminus — is an extrapolation, and we mark it as such.

Facts and figures
uniprotP51825
length AA1210
gene symbolAFF1
ensembl geneENSG00000172493
cytoband4q21.3-q22.1
alphafold IDAF-P51825-F1
alphafold noteAF-P51825-F1 resolves and covers residues 1-1210, but the model is overwhelmingly very-low-confidence ribbon: UniProt annotates 87% of the chain as disordered, so the AlphaFold model is a disorder cartoon, not a structure to measure domains from.
isoform noteAll coordinates are in UniProt P51825-1 numbering (1,210 aa). Ensembl canonical ENST00000395146/ENSP00000378578 encodes 1,218 aa — an eight-residue offset. UniProt also describes P51825-2 (variant at 1-5 and 1096) and P51825-3 (variant at 1096).
assumed fusion for fatefusion-kmt2a-aff1 (KMT2A::AFF1, t(4;11)), AFF1 as the 3' partner
domain coordinate caveatAFF1 is an intrinsically disordered protein and most of these 'domains' are conventions. Only the AF9/ENL-binding motif (761-775, backed by PDB 2LM0) and the CHD (949-1209, a Pfam HMM on a family alignment) should be treated as firm. The ALF/NHD C-terminal edge floats by ~20 residues between the CATH hit (78) and the conventional review boundary (~100). The pSER boundary is a composition threshold with no physical edge. The TAD (480-560) is the smallest tested active fragment, an upper bound. The NLS is a prediction we made and has no published residue assignment. Everything is in UniProt P51825-1 numbering; Ensembl canonical is 1,218 aa, eight residues longer.
domains
ALF/NHD (N-terminal homol…5.000ALF/NHD (N-terminal homology domain): 5pSER (serine-rich low-com…423.0pSER (serine-rich low-complexity core): 423TAD (transactivation doma…480.0TAD (transactivation domain): 480NLS (nuclear localisation…603.0NLS (nuclear localisation signal, predicted): 603AF9/ENL-binding motif761.0AF9/ENL-binding motif: 761CHD (C-terminal homology …949.0CHD (C-terminal homology domain): 949
PDB IDS
[
  "2LM0"
]
paralogue structures used
[
  {
    "pdb": "6R80",
    "protein": "AFF4",
    "range": "899-1163",
    "relevance": "CHD fold and dimerisation; positionally equivalent to AFF1 949-1209"
  },
  {
    "pdb": "4IMY",
    "protein": "AFF4",
    "range": "2-73",
    "relevance": "N-terminal element bound to P-TEFb; positionally equivalent to the AFF1 ALF/NHD at 5-78"
  },
  {
    "pdb": "5JW9",
    "protein": "AFF4",
    "range": "301-351",
    "relevance": "the 'ELLBow' element bound to the ELL2 OCEL domain; the AFF1 equivalent is unmapped"
  }
]
domains — the numbers behind the chart
[
  {
    "name": "ALF/NHD (N-terminal homology domain)",
    "start": 5,
    "end": 78,
    "pfam": "PF05110",
    "note": "Only structured element in the N-terminal half; CATH-Gene3D G3DSA:6.10.250.2670 hit. Reviews conventionally extend the ALF domain to ~aa 100. The paralogous AFF4 N-terminal region (aa 2-73) is the Cyclin T1/P-TEFb-binding element resolved in PDB 4IMY, so this is the likely P-TEFb contact. Carries the SIAH1/2-sensitive N-terminal portion (not residue-mapped).",
    "fate": "lost",
    "fate_caveat": "DISPUTED. This graph's fusion-kmt2a-aff1 node lists the ALF/NHD as RETAINED at both of its breakpoints, which both join 'AFF1 exon 4'. Ensembl exon 4 spans codons 54-353, so on that numbering the fusion would keep only the last ~25 residues of the 5-78 domain; on a first-coding-exon numbering, 'exon 4' is codons 354-368 and the domain is wholly lost. We use 'lost' because the functional literature describes the fusion as lacking the AF4 N-terminus, but the AFF1-side junction residue is not known to us to better than +/-300 aa.",
    "evidence_basis": "hmm_hit",
    "boundary_precision": "+/- ~20 aa at the C-terminal edge (5-78 measured vs ~1-100 conventional)"
  },
  {
    "name": "pSER (serine-rich low-complexity core)",
    "start": 423,
    "end": 497,
    "note": "Contiguous UniProt compositional-bias features: acidic 423-439, Pro-rich 442-453, low-complexity 454-472, acidic 473-497. A 50-residue sliding scan of P51825 puts peak serine density at 441-510 (Ser fraction 0.36-0.44). This is a composition boundary, not a fold boundary — it has no edges in any physical sense.",
    "fate": "retained",
    "evidence_basis": "composition",
    "boundary_precision": "soft; a composition threshold, not a measurement"
  },
  {
    "name": "TAD (transactivation domain)",
    "start": 480,
    "end": 560,
    "note": "Smallest AF4 fragment reported as a strong activator in a GAL4-fusion reporter assay (Prasad 1995). Overlaps the pSER core. An upper bound on extent, not a mapped boundary.",
    "fate": "retained",
    "evidence_basis": "functional_assay",
    "boundary_precision": "upper bound; no deletion titration was done"
  },
  {
    "name": "NLS (nuclear localisation signal, predicted)",
    "start": 603,
    "end": 614,
    "note": "PREDICTION ONLY. No AFF-family member carries an NLS feature in UniProt and we found no verifiable primary mapping study. Sequence KQPKKPVKASAR, the strongest basic cluster in the region where reviews place the NLS. Draw this differently from the measured features, or omit it.",
    "fate": "retained",
    "evidence_basis": "prediction",
    "boundary_precision": "unverified; this is the weakest coordinate in the entry"
  },
  {
    "name": "AF9/ENL-binding motif",
    "start": 761,
    "end": 775,
    "pfam": "PF18875",
    "note": "Pfam 'AF4 interaction motif' / InterPro IPR043639. This — NOT the CHD — is the element that binds the AF9 and ENL ANC1 homology domains. The NMR construct in PDB 2LM0 spans 738-779; the conserved core is 761-775.",
    "fate": "retained",
    "evidence_basis": "structure",
    "boundary_precision": "core hard (HMM + solved complex); the folded-on-binding segment may extend into 738-760"
  },
  {
    "name": "CHD (C-terminal homology domain)",
    "start": 949,
    "end": 1209,
    "pfam": "PF18876",
    "note": "Pfam PF18876 / InterPro IPR043640. An AFF-family DIMERISATION domain: the AFF4 equivalent (899-1163) is a novel eight-helix fold, solved at 2.2 A as PDB 6R80, that mediates AFF4 homodimerisation and AFF1-AFF4 heterodimerisation, binds RNA and DNA, and carries a CDK9-substrate surface loop. Widely but, on the structural evidence, incorrectly described as the AF9/ENL-binding domain; see the AF9/ENL-binding motif entry above. Fully retained in every observed KMT2A::AFF1 break, so the fusion carries an oligomerisation interface.",
    "fate": "retained",
    "evidence_basis": "hmm_hit+structure(paralogue)",
    "boundary_precision": "+/- a few residues between InterPro releases; the AFF4 paralogue's crystallised construct 899-1163 is positionally equivalent"
  }
]
conflicting annotations
[
  {
    "feature": "AF9/ENL-binding element",
    "claim_a": "The C-terminal homology domain (949-1209) binds AF9 and ENL — the conventional statement in reviews and in this project's own brief.",
    "claim_b": "A short motif at 761-775 binds the AF9/ENL ANC1 homology domain; PDB 2LM0 resolves AFF1 738-779 bound to the AF9 AHD, and Pfam/InterPro define 761-775 as a distinct 'AF4 interaction motif'.",
    "our_reading": "Claim B is the one with a solved structure behind it. Both are recorded. No clinically observed KMT2A::AFF1 break separates 761-775 from 949-1209, which is why the mislabelling has no phenotypic consequence and has gone uncorrected."
  },
  {
    "feature": "Is the ALF/NHD retained in KMT2A::AFF1?",
    "claim_a": "This graph's fusion-kmt2a-aff1 node: RETAINED at both listed breakpoints, each joining AFF1 exon 4.",
    "claim_b": "This entry: LOST. The functional literature describes the fusion as lacking the AF4 N-terminus, which carries the SIAH-recognised degron element.",
    "our_reading": "The disagreement is a numbering question, not a biology question. Ensembl AFF1 exon 4 = codons 54-353, so on that convention the fusion keeps only the last ~25 residues of the 5-78 domain (a fragment, not a domain); on a first-coding-exon convention, 'exon 4' starts at codon 354 and the domain is wholly absent. Neither reading supports 'retained' as a functional statement. RESOLVED 2026-09-24 (Q191): the AFF1-side junction is residue 347. QTYSNEVH occurs exactly once in UniProt P51825, and the siMA6 reconstruction shows both KMT2A::AFF1 chimeras resuming at QTYSNEVHCVEEILK, so the AFF1 acceptor is constant across the two modelled breakpoints. This agrees with the codons 347-361 derived above for exon 4 under ENST00000307808. CITATION CORRECTED 2026-09-24 (Q194): this field previously said to settle it from 'the Meyer 2024 per-gene supplementary tables'. That is the wrong paper - Meyer 2024 (PMID 38678092) is a four-page Letter on the KMT2A/MLL consensus GENE STRUCTURE and carries no partner-gene tables. The partner-side breakpoints were determined in Meyer 2023 (PMID 37019990, n=3,401), whose abstract states that breakpoints within KMT2A and the involved translocation partner genes were determined. Meyer 2023 remains the right source for the one part still open: whether rarer KMT2A::AFF1 breakpoints use an AFF1 acceptor other than exon 4."
  },
  {
    "feature": "ALF/NHD extent",
    "claim_a": "CATH-Gene3D structured element at 5-78.",
    "claim_b": "Reviews conventionally draw the ALF domain as ~aa 1-100.",
    "our_reading": "We use 5-78 because it is the only one derived from structural modelling; the ~22-residue discrepancy is recorded rather than averaged away."
  }
]
essentiality
{
  "note": "Not assessed in this entry; this is a coordinates entry. See gene-aff1 for essentiality."
}
How this could bite us back
  • The AF9/ENL-binding motif at 761-775 is short, disordered and folds only on binding. A peptidomimetic against it inherits the usual problems of that class — poor cell permeability, proteolytic lability — and the interface is shared with wild-type AFF1 and AFF4 in the normal super elongation complex, so selectivity for the fusion is not obviously available.
  • If the fusion's advantage partly comes from escaping SIAH-mediated turnover [E8], then any therapy that lowers fusion transcription without restoring degradation is fighting a protein with no off-switch, and dose-response will be shallow.
  • Drawing the NLS as a hard box in a figure, when it is a prediction, is exactly the kind of error this beat exists to prevent. If it propagates into a grant figure it will be cited back as if measured.
Still open
  1. Does AFF1 have its own nuclear localisation signal at all, or does it reach the nucleus only as cargo of the super elongation complex? A GFP-fusion import assay on the 590-650 fragment would settle in a week a question that every review currently answers with a drawn box.
  2. Does KMT2A::AFF1 dimerise through its retained CHD, and does it heterodimerise with endogenous AFF4? The AFF4 CHD does both, and the fusion keeps the domain intact, but nobody has tested it in the fusion. If it does, the commonest KMT2A fusion of all belongs in the oligomerisation class alongside AFDN and EPS15, and it has a folded, tractable interface that a coiled coil does not offer.
  3. If the fusion's CHD heterodimerises with AFF4, is the fusion's activity stoichiometrically dependent on AFF4 levels? That is a straightforward AFF4 knockdown-titration experiment and it would distinguish 'the fusion recruits the SEC' from 'the fusion is a subunit of the SEC'.
  4. Does the 761-775 motif behave differently in the fusion than in wild-type AFF1 — for instance, is its AF9/ENL occupancy higher because the fusion is not turned over? That would make a peptide-competitive inhibitor of this motif fusion-selective by dose rather than by sequence.
  5. Is the eight-residue Ensembl/UniProt discrepancy an annotation artefact or a real alternative N-terminus, and has it silently displaced coordinates in any published AFF1 figure?
  6. RESOLVED 2026-09-24 (Q187 follow-up): the fusion junction begins at AFF1 residue 347. QTYSNEVH occurs exactly once in UniProt P51825, at 347, and the siMA6 reconstruction shows both KMT2A::AFF1 chimeras resuming at QTYSNEVHCVEEILK, so the AFF1 acceptor is constant across breakpoints. This matches the codons 347-361 that this entry derived for 'exon 4' under ENST00000307808 - exon structure and protein sequence agree. The codon-54 reading (Ensembl ENST00000395146 exon 4) is refuted. Consequence: the ALF/NHD is wholly lost, as this entry already recorded, and so is the SIAH degron at human 282-304.
Every source behind this entry (10)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt P51825 (AFF1_HUMAN) is 1,210 aa and annotates no named domain features; FOUR Region features of type Disordered cover 1-45, 73-314, 366-957 and 1098-1119, and a functional region 601-900 is required for DNA-damage-site recruitment and PARP1 interaction.

    What this evidence cannot show

    Retrieved 2026-09-18 via the UniProt REST API. Absence of a domain annotation is weak evidence of absence of a domain: UniProt annotates folded units, and an intrinsically disordered protein can carry functional elements that UniProt's curation model has no feature type for. The 'Disordered' regions are themselves predictions (MobiDB-lite consensus), not experimental disorder measurements. CORRECTION 2026-09-19: this item previously listed THREE disordered regions and omitted 1098-1119, which is the one that matters, because it falls inside the Pfam/InterPro AF4/FMR2 C-terminal homology domain at 949-1209.

    Evidence that argues against this (1)
    • There are four disordered regions, not three, and a structural database places a fold inside one of them. The fetched P51825 record carries Disordered regions at 1-45, 73-314, 366-957 AND 1098-1119 - the fourth sitting inside the region Pfam/InterPro call the AF4/FMR2 C-terminal homology domain (PF18876 / IPR043640, 949-1209), so UniProt predicts a 22-residue disordered stretch inside a domain other databases model as folded. In the other direction, CATH-Gene3D G3DSA:6.10.250.2670 assigns a homologous superfamily at residues 5-78, straddling UniProt's disordered 1-45 and 73-314. 'No named domain features' is a property of UniProt's curation model, not a finding about the protein. UniProt Consortium 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    InterPro maps four signatures onto P51825: CATH-Gene3D G3DSA:6.10.250.2670 at 5-78; Pfam PF05110 (AF-4 proto-oncoprotein N-terminal region) at 8-520; Pfam PF18875 / InterPro IPR043639 (AF4 interaction motif) at 761-775; and Pfam PF18876 / InterPro IPR043640 (AF4/FMR2 C-terminal homology domain) at 949-1209.

    InterPro entry-to-protein matches for UniProt P51825 · InterPro / EMBL-EBI · InterPro 2026
    What this evidence cannot show

    These are HMM hit boundaries on a family alignment, not experimentally determined domain edges; hit ends typically float by a few residues between database releases. PF05110 at 8-520 is a family-level model spanning the whole N-terminal half and should not be drawn as a single folded domain. Retrieved 2026-09-18.

    Evidence that argues against this (1)
    • The fetched InterPro record returns eight matches, not four, and the two largest swallow the whole protein. Alongside the four named, the set contains PANTHER PTHR10528 at 6-1209 and InterPro IPR007797 'AF4/FMR2 family' at 6-1209 - two family-level calls spanning 1,204 of 1,210 residues - plus the InterPro parents IPR043639 (761-775) and IPR043640 (949-1209). Counting four also mis-types two of them: PF18875 / IPR043639 is registered as a conserved_site, not a domain, so drawing it as a 15-residue bar alongside PF18876 puts two different annotation classes on the same axis. InterPro Consortium; EMBL-EBI 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    An AF4 polypeptide containing residues 480-560 showed strong transcriptional activation potential when fused to the GAL4 DNA-binding domain.

    Domains with transcriptional regulatory activity within the ALL1 and AF4 proteins involved in acute leukemia · Prasad R, Yano T, Sorio C, Nakamura T, Rallapalli R, Gu Y, Leshkowitz D, Croce CM, Canaani E · Proc Natl Acad Sci U S A 92:12160-12164 1995
    PMID 8618864DOI
    What this evidence cannot show

    A heterologous GAL4 reporter assay from 1995 that tested a set of fragments, not a boundary-mapping titration; 480-560 is the smallest fragment they reported as a strong activator, so it is an upper bound on the domain's extent and says nothing about whether 479 or 561 are required. The assay also does not establish that this region acts as a transactivation domain in the context of full-length AFF1 or of the fusion.

    Evidence that argues against this (1)
    • The same paper's headline finding is that KMT2A's own strong activation domain sits at 2829-2883, a region lost in every fusion. That the fusion's activation output nonetheless depends on the AFF1 side is an inference from other work, not from this assay. Prasad R et al. 1995
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PDB 2LM0 is the solution NMR structure of the AF9 ANC1 homology domain in complex with an AFF1 peptide; its AFF1 chain spans residues 738-779.

    PDB entry 2LM0 — Solution structure of the AF4-AF9 complex · Leach BI, Kuntimaddi A, Schmidt CR, Cierpicki T, Johnson SA, Bushweller JH · RCSB Protein Data Bank 2013
    What this evidence cannot show

    The 738-779 range is the construct the authors used, cross-referenced from UniProt P51825, not a statement that all 42 residues contact AF9. Solution NMR of a short peptide bound to an intrinsically disordered partner defines the folded-on-binding segment but leaves the flanks poorly determined. Entry title and the 738-779 chain range were confirmed against the RCSB data API on 2026-09-18. CORRECTION 2026-09-19: 2LM0 is a SINGLE tethered polypeptide, not a two-chain complex. SIFTS maps both P42568 490-568 and P51825 738-779 to the same chain A of one polymer entity, and RCSB gives polymer_entity_count = 1, so the entry measures no bimolecular affinity.

    Evidence that argues against this (1)
    • 2LM0 is not a two-chain complex - AF9 and AFF1 are a single tethered polypeptide, so nothing in it measures a bimolecular interaction. The fetched SIFTS record maps BOTH P42568 (AF9 490-568, PDB residues 47-125) and P51825 (AFF1 738-779, PDB residues 2-43) to the same chain A of the same single polymer entity, and RCSB reports polymer_entity_count = 1. Only 15 of the 42 AFF1 residues in that chain (761-775) correspond to the conserved AF4 interaction motif Pfam PF18875 and InterPro IPR043639 recognise, so the 738-779 bar is a construct length on a construct that could not dissociate. PDBe / SIFTS et al. 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The AF9 ANC1 homology domain is intrinsically disordered and recruits AF4/AFF1 and other partners by coupled folding and binding, with conserved aliphatic residues forming the hydrophobic core of the complex.

    Leukemia fusion target AF9 is an intrinsically disordered transcriptional regulator that recruits multiple partners via coupled folding and binding · Leach BI, Kuntimaddi A, Schmidt CR, Cierpicki T, Johnson SA, Bushweller JH · Structure 21:176-183 2013
    PMID 23260655DOI
    What this evidence cannot show

    PubMed records the electronic publication date as 2012-12-20; the journal issue (Structure volume 21, issue 1) is dated 2013 and we use 2013 throughout this graph for consistency. The work is on isolated peptides in solution, so the affinities and the folding transition are properties of the fragments, not demonstrated in the context of the assembled super elongation complex on chromatin.

    Evidence that argues against this (1)
    • The same laboratory's follow-up shows AF9's ANC1 homology domain is a single binding slot, not a recruiter of several partners at once. The NMR structure of the DOT1L(877-900)-AF9(499-568) complex (PDB 2MV7) superimposes on the AF4-AF9 complex (2LM0) with a backbone RMSD of 1.4 A, and the authors state that AF4 and DOT1L bind in the same site on AF9, consistent with mutually exclusive binding; near-identical NMR spectra of the AF9 complexes with BCoR and CBX8 make it likely those partners also bind the same site, and an AF9 D546R mutant that abolishes DOT1L binding INCREASES CBX8 binding. 'Recruits multiple partners' therefore means sequential competition for one surface. The paper also flags a methodological limit on this class of measurement: its Site 2 Kd differed significantly from a previously published value, attributed to the AF9 AHD's propensity to aggregate. Kuntimaddi A et al. 2015
  6. E6weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.database

    No AFF-family member (AFF1 P51825, AFF2 P51816, AFF3 P51826, AFF4 Q9UHB7) carries a nuclear-localisation-signal motif feature in UniProt; the strongest basic-residue cluster in AFF1 in the region where reviews place the NLS is residues 603-614 (KQPKKPVKASAR).

    What this evidence cannot show

    This is a sequence-composition prediction we made, not a published or experimental NLS assignment, and it is the least reliable coordinate in this entry. A basic patch is necessary but nowhere near sufficient for importin binding; the residues could equally be a DNA- or RNA-contacting surface, and AFF1 could reach the nucleus passively as part of the super elongation complex without its own signal. We searched PubMed for a primary AF4 NLS mapping study and found none we could verify.

    Evidence that argues against this (1)
    • A published experimental NLS mapping of an AFF-family protein shows this exact kind of basic-cluster prediction both over-calls and under-calls. PSORT II predicted seven nuclear localisation signals in the 1,163-aa AFF4/MCEF protein; testing twenty EGFP/DsRed deletion constructs in HeLa cells showed only two were functional (the bipartite NLS 1 in exon XI and NLS 5 in exon XII), while predicted NLS 6 and NLS 7 gave no nuclear localisation when tested directly. Critically, a non-NLS sequence - AFF4 residues 804-845 in exon XIII, with no basic-cluster signature - was by itself sufficient to drive nuclear localisation, with partner-mediated co-transport via P-TEFb proposed. Five of seven predicted sites were false positives and one real targeting element carried no basic cluster at all. BOUNDED ABSENCE for AFF1 specifically: PubMed `(AF4[tiab] OR AFF1[tiab] OR MLLT2[tiab]) AND (nuclear localization[tiab] OR nuclear localisation[tiab])` returns 5 records (PMIDs 32518061, 17389929, 9892693, 9349501, 8358709) and none maps an AFF1 NLS; the only AFF1-adjacent assertion (PMID 8358709) says the retained AF-4 portion begins at codon 491 containing nuclear localization and GTP-binding motifs, with no residue mapping, placing it downstream of 491 rather than at 603-614. Niedzielski MF et al. 2007
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Ensembl's canonical AFF1 transcript ENST00000395146 (ENSP00000378578) encodes a 1,218 aa protein, eight residues longer than the 1,210 aa UniProt canonical P51825-1, so coordinates are not interchangeable between the two references.

    What this evidence cannot show

    Retrieved 2026-09-18. We confirmed the gene identity (ENSG00000172493 returns display_name AFF1, chr4:86,934,203-87,141,073) rather than assuming it. We did not establish which residues account for the eight-residue difference, so we cannot say whether the offset applies uniformly to all downstream coordinates or only past a specific exon.

    Evidence that argues against this (1)
    • Ensembl also contains a transcript encoding exactly 1,210 aa, so this is a disagreement about which transcript to call canonical, not a disagreement about the protein. The fetched gene record for ENSG00000172493 lists 34 protein-coding AFF1 transcripts, among them ENST00000307808 (AFF1-201) at 1,210 aa - the UniProt length - plus ENST00000544085 at 1,211, ENST00000863512 at 1,202 and four separate transcripts at 1,218. Framing it as an eight-residue offset between two databases hides that the real spread across Ensembl's own AFF1 translations is 1,202-1,218, and that a zero-offset transcript is available in the same record. Ensembl / EMBL-EBI 2026
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    SIAH1 and SIAH2 E3 ubiquitin ligases act on the N-terminal portion of AF4, and the AF4·MLL fusion protein is protected against proteasomal degradation.

    Interaction of AF4 wild-type and AF4.MLL fusion protein with SIAH proteins: indication for t(4;11) pathobiology? · Bursen A, Moritz S, Gaussmann A, Moritz S, Dingermann T, Marschalek R · Oncogene 23:6237-6249 2004
    PMID 15221006DOI
    What this evidence cannot show

    The stabilised protein in this study is AF4·MLL, the reciprocal der(4) product, not MLL·AF4. Applying the argument to MLL::AFF1 — which also lacks the AF4 N-terminus — is an extrapolation we have not found directly tested. The work is yeast two-hybrid plus overexpression in cell lines, and the paper does not map the SIAH-recognised element to residue numbers, so we have not drawn a degron box in the domain array.

    Evidence that argues against this (1)
    • Whether the reciprocal AF4·MLL product contributes to t(4;11) leukemogenesis at all is contested; a substantial fraction of t(4;11) patients lack a detectable reciprocal transcript, which weakens any model that requires it. Meyer C et al. 2023
  9. E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The AFF4 C-terminal homology domain, solved at 2.2 Å (PDB 6R80, chain A covering AFF4 899-1163), folds into a novel eight-helix domain distantly related to tetratricopeptide-repeat motifs; it mediates AFF4 homodimerisation or AFF1-AFF4 heterodimerisation, binds RNA and DNA by fluorescence anisotropy, and contains a surface loop that is a CDK9 substrate. The CHD is conserved across AFF1, AFF2, AFF3 and AFF4.

    PMID 31147444DOI
    What this evidence cannot show

    The structure is of AFF4, not AFF1; no AFF1 CHD structure exists, and the AFF1 involvement is in the heterodimerisation biochemistry rather than the crystallography. The nucleic-acid binding is fluorescence anisotropy with purified domain and naked RNA/DNA, which does not establish sequence specificity or chromatin relevance. Whether the CHD dimerises in the context of the KMT2A::AFF1 fusion protein, and whether the fusion heterodimerises with endogenous AFF4 in cells, has not been tested. PDB 6R80's title, journal and the 899-1163 chain range were confirmed against the RCSB data API on 2026-09-18.

    Evidence that argues against this (1)
    • An independent crystal structure of the same region bounds three parts of this at once. Human AFF4 902-1163 solved by Se-SAD at 2.4 A (PDB 6K7P) has only ONE molecule in the asymmetric unit, localises the interface to alpha4/alpha5/alpha6 burying 3,219 square angstroms, and renames the fold THD because the base is a three-TPR-like solenoid. Two findings cut against the dimeric-scaffold reading: single mutants F1014A or Y1096A convert AFF4-THD to a monomer (analytical ultracentrifugation 3.43 S / about 61.2 kDa wild type against 2.22 S / about 28.9 kDa mutant) yet in-vivo pull-downs show those monomers bind other super-elongation-complex subunits as strongly as wild type; and although AFF1-THD is 44.3% identical with the key residues conserved, the equivalent AFF1 mutants F1063A and Y1145A FAILED to disrupt AFF1-THD dimerisation. The paper also discounts the nucleic-acid result directly, saying functional assays are required to support that conclusion. Fetched PDBe coverage for 6R80 adds that residues 1048-1080 - 33 residues including the surface loop - have no coordinates. Tang D et al. 2020
  10. E10strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Mapping the coding exons of Ensembl canonical AFF1 transcript ENST00000395146 to codon ranges gives: exon 1 non-coding; exon 2 codons 1-13; exon 3 codons 13-53; exon 4 codons 54-353 (a single 900-nt exon); exon 5 codons 354-368. A junction at 'AFF1 exon 4' therefore begins at codon 54 or, if exons are numbered from the first coding exon, at codon 354 — a 300-residue difference that determines whether the ALF/N-terminal homology domain is truncated or wholly absent from the fusion.

    What this evidence cannot show

    The exon-to-codon mapping is our own computation from Ensembl coordinates, retrieved 2026-09-18, and is arithmetic rather than a literature claim. It is on the 1,218 aa Ensembl protein, not the 1,210 aa UniProt canonical, so codon numbers carry the same eight-residue offset described in E7. Critically, we could NOT establish which exon-numbering convention the KMT2A fusion literature uses when it says 'AFF1 exon 4' — that is the unresolved part, and this computation only quantifies how much the answer depends on it.

    Evidence that argues against this (1)
    • There is a third answer, because Ensembl's 1,210-aa AFF1 transcript uses a different first exon entirely and shifts the whole numbering by one. In the fetched ENST00000307808 record the first exon is chr4:87,006,988-87,007,424 with the CDS starting at 87,007,408 - so exon 1 IS coding, in its last 17 nt - and the 900-nt exon that is exon 4 in ENST00000395146 is exon 3 here. Working the coding lengths through (17 + 121 + 900 = 1,038 nt before it), 'exon 4' in this transcript is the 45-nt exon at 87,084,120-87,084,164, i.e. codons 347-361. The two transcripts' first exons are 72 kb apart, so the ambiguity is not two-way between codon 54 and codon 354 but at least three-way, and it turns on transcript choice before it turns on numbering convention. Ensembl / EMBL-EBI 2026
Connected to