Oncofusion

start / graph / localization-at-hooks-targeting-determinant

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q4b-localization

The three AT-hooks (aa 169-180, 217-227, 301-309) as a localisation determinant — and the correction that MLL's hooks read DNA *structure*, not AT-richness

also called MLL AT hook · AT-hook cruciform · A.T hook 1 2 3 KMT2A · minor groove tether

KMT2A has three short hook-shaped motifs near its front end that grip DNA and are kept in every leukemia fusion. The textbook says they grab AT-rich DNA through its narrow groove — but the one paper that actually tested the MLL hooks found they recognise a DNA *shape*, the four-way cruciform, and said so explicitly: structure rather than sequence.

The same thing, for a scientist

UniProt Q03164 (human KMT2A, 3,969 aa) annotates three A.T hooks at residues **169-180, 217-227 and 301-309** [E1]. All three lie inside the N-terminal fragment retained by every leukemogenic fusion, and deletion of the AT-hook motifs substantially impairs HRX-ENL myeloid transformation in primary murine cells [E2]. The generic AT-hook motif, defined by Reeves & Nissen from HMG-I/Y, is a minor-groove binder of A.T-rich DNA whose consensus core is the palindromic T-P-K-R-P-R-G-R-P-K-K peptide, competed by netropsin, distamycin and Hoechst 33258 [E3]; Aravind & Landsman later classified the motif across DNA-binding proteins, naming HRX/ALL-1 explicitly, and described it as a versatile minor-groove tether [E4]. **But the MLL-specific experiment says something different.** Zeleznik-Le, Harden & Rowley assayed the bacterially expressed MLL AT-hook region by EMSA and reported binding to **cruciform DNA**, recognising *structure rather than sequence of the target DNA* [E5]. That is a real distinction with consequences: a structure-selective tether predicts enrichment at bent, supercoiled, cruciform-prone and scaffold/matrix-attached DNA rather than at AT-rich sequence per se — and the KMT2A breakpoint cluster region is itself flanked by scaffold attachment regions and topoisomerase II consensus sites [E6]. The dispatch premise that this beat should describe 'the AT-hooks' minor-groove binding' is therefore only half right, and the half that is wrong is the half specific to this protein.

What it looks like
The three AT-hooks (aa 169-180, 217-227, 301-309) as a localisation determinant — and the correction that MLL's hooks read DNA *structure*, not AT-richness3,969 aa
AT-hook 1 · residues 169–180AT-hook 2 · residues 217–227AT-hook 3 · residues 301–309CXXC (CpG-binding zinc finger) · residues 1147–1195SNL-1 (subnuclear localisation signal 1) · residues 388–43201,0002,0003,000

No three-dimensional model is available for this protein, so the map above is the whole picture. That is itself informative: AlphaFold does not model chains this long, and the parts that have been crystallised are held under the entries for the individual domains.

Go deeper
The residue numbers, from the database rather than from memory

From UniProt Q03164 (KMT2A_HUMAN, Homo sapiens, 3,969 aa, 431,764 Da), fetched directly 1:

elementresidues
A.T hook 1169-180
A.T hook 2217-227
A.T hook 3301-309
CXXC-type zinc finger1147-1195
PHD-1 / PHD-2 / PHD-31431-1482 / 1479-1533 / 1566-1627
Bromodomain1635-1765
SET3829-3945

Two housekeeping notes. First, UniProt annotates three AT-hooks; papers that describe 'two AT hooks' are describing a truncation construct (for example the 35-kDa MLL N-terminal fragment that drives cell-cycle arrest and monocytic differentiation in U937 7) rather than the full protein. Second, all three hooks sit well 5' of every breakpoint in the major or minor breakpoint cluster region, so they are retained in KMT2A::AFF1, KMT2A::MLLT3, KMT2A::MLLT1 and the rest without exception.

What the hooks actually bind — the premise that does not check out1 deeper

The generic motif is unambiguous. Reeves & Nissen synthesised the consensus oligopeptide, showed it binds the minor groove of A.T-rich DNA stretches, and showed competition by the minor-groove ligands netropsin, distamycin and Hoechst 33258 3. Aravind & Landsman's computational survey then found the motif across chromatin proteins including HRX/ALL-1 and called it 'a versatile minor groove tether' 4. If one reasons only from the motif, 'MLL AT-hooks bind AT-rich minor groove' follows.

The experiment on MLL's own hooks reports something narrower and stranger. Zeleznik-Le et al. expressed the MLL AT-hook domain in bacteria and ran gel-shift assays; the binding they characterise is to cruciform DNA, with the explicit statement that the domain recognises structure rather than sequence of the target DNA 5. Cruciform extrusion happens at inverted repeats under negative supercoiling — a topological state, not a base composition.

The graph should hold both facts and not collapse them. The honest formulation: the motif class is a minor-groove AT tether; the only direct assay of MLL's copies of it reports structure-selective cruciform binding. Nobody has since resolved the discrepancy for MLL, which is itself a publishable gap.

Why structure-selectivity would be the more interesting answer

If KMT2A's hooks are topology readers rather than sequence readers, three things in this graph connect that currently do not.

  1. The breakpoint region. Broeker et al. mapped 11q23 breakpoints in 31 de novo acute leukemia and 8 t-AML samples and correlated them with scaffold attachment regions and topoisomerase II consensus binding sites flanking the bcr 6. SARs are AT-rich, base-unpairing, cruciform-prone DNA. A protein whose own N-terminus prefers that DNA topology, and whose gene's fragile region is made of it, is a coincidence worth testing.
  2. The nuclear matrix. The graph already records that the MLL N-terminus targets the protein to the interphase nuclear matrix and to mitotic chromosome scaffolds alongside topoisomerase II 8. Matrix attachment is operationally defined by SAR/MAR DNA. A cruciform-preferring tether is a mechanism for that observation; an AT-sequence tether is a weaker one.
  3. Chromocentres. AT-hook proteins that are sequence readers (HMGA1, Drosophila D1) bundle AT-rich pericentromeric satellite DNA into chromocentres. KMT2A is not reported at chromocentres. If MLL's hooks are structure-selective rather than AT-selective, that absence is explained rather than merely unexplained.
Is the AT-hook a localisation determinant or a chromatin-affinity determinant?

These are different claims and the literature supports only the weaker one cleanly. AT-hook deletion impairs transformation 2 and the AT-hook-containing N-terminus is sufficient to confer the punctate nucleoplasmic/nucleolar-peripheral/nuclear-matrix distribution and mitotic scaffold association 8. But sufficiency of a fragment is not the same as necessity of the hooks within the full protein: the fragment used in the targeting study spans the AT-hook region and also the SNL-1 element at aa 388-432 recorded elsewhere in this graph, so the targeting could be SNL-driven with the hooks contributing chromatin affinity rather than address.

The clean experiment is a three-hook alanine substitution (the GRP cores at 169-180, 217-227, 301-309) in an otherwise intact endogenously tagged fusion, scoring subnuclear distribution and chromatin occupancy separately. It has never been done, and it is the same experimental design that settled the equivalent question for CXXC.

Facts and figures
uniprotQ03164
length AA3969
binding preference generic motifminor groove of A.T-rich DNA
binding preference measured for mllcruciform DNA; structure rather than sequence
retained in all leukemogenic fusionstrue
premise correctedthe dispatch brief's 'AT-hooks' minor-groove binding' is the motif-class property, not what was measured for MLL's own hooks
domains
AT-hook 1169.0AT-hook 1: 169AT-hook 2217.0AT-hook 2: 217AT-hook 3301.0AT-hook 3: 301SNL-1 (subnuclear localis…388.0SNL-1 (subnuclear localisation signal 1): 388CXXC (CpG-binding zinc fi…1,147CXXC (CpG-binding zinc finger): 1147
domains — the numbers behind the chart
[
  {
    "name": "AT-hook 1",
    "start": 169,
    "end": 180
  },
  {
    "name": "AT-hook 2",
    "start": 217,
    "end": 227
  },
  {
    "name": "AT-hook 3",
    "start": 301,
    "end": 309
  },
  {
    "name": "CXXC (CpG-binding zinc finger)",
    "start": 1147,
    "end": 1195
  },
  {
    "name": "SNL-1 (subnuclear localisation signal 1)",
    "start": 388,
    "end": 432
  }
]
How this could bite us back
  • Targeting the AT-hook-DNA interface with a minor-groove ligand (the netropsin/distamycin class competes the generic motif) would hit every AT-hook protein in the cell, including HMGA1 which organises pericentromeric chromatin — a predictable genome-instability toxicity in a paediatric population that will live for decades.
  • If the hooks are structure readers, an inhibitor designed against an AT-rich sequence complex would miss the physiological interaction entirely, and a screen built on the wrong probe DNA would return only artefacts.
  • Three redundant hooks is a resistance architecture: loss of any one is unlikely to be limiting, so a competitive inhibitor faces avidity from a multivalent tether rather than a single site.
Still open
  1. Do MLL's AT-hooks prefer AT-rich sequence or cruciform/supercoiled structure? A modern binding survey — SELEX plus a topology panel with relaxed, supercoiled and inverted-repeat substrates — would resolve a 32-year-old ambiguity in one experiment and is cheap.
  2. Does a triple AT-hook alanine mutant in an endogenously tagged KMT2A::AFF1 change subnuclear distribution, chromatin occupancy, or both, and are the two separable? CXXC is already known to be separable from localisation; the hooks have never been tested this way.
  3. If the hooks are structure-selective, do they enrich the fusion at scaffold attachment regions genome-wide? A SAR/MAR-annotated overlay on existing KMT2A::AFF1 CUT&RUN would be a same-week reanalysis with no new data generation.
  4. Does the structure-selective binding preference explain why KMT2A is not found at AT-rich pericentromeric satellite, where other multi-AT-hook proteins concentrate?
Every source behind this entry (8)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    Human KMT2A (UniProt Q03164) is 3,969 amino acids and carries three annotated A.T hook DNA-binding motifs at residues 169-180, 217-227 and 301-309, with the CXXC-type zinc finger at 1147-1195.

    UniProtKB Q03164 (KMT2A_HUMAN) — Histone-lysine N-methyltransferase 2A · UniProt Consortium · UniProt Knowledgebase 2026
    What this evidence cannot show

    Database feature annotations are curated predictions anchored on sequence motifs and literature, not experimentally determined boundaries for every feature; the AT-hook boundaries in particular are motif-matched. Isoform numbering follows the canonical sequence.

    Evidence that argues against this (1)
    • The annotation gives three equivalent-looking motifs, and the structural literature on this motif family says copies are not equivalent. Huth et al. determined the solution structure of an HMG-I(Y) fragment containing the second and third DNA-binding domains bound to a dodecamer, found one protein bound to two DNA molecules, identified a central Arg-Gly-Arg motif plus two accessory contact modules, and showed that the third domain lacks one of those modules and binds roughly 100-fold more weakly than the second. A UniProt feature table listing 169-180, 217-227 and 301-309 carries no such gradation, so any argument in this entry that treats 'the AT-hook region' as a single functional unit is assuming uniformity that the reference protein does not have. Huth JR et al. 1997
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Deletion of the AT-hook motifs, or of the methyltransferase-homology/CXXC domain, substantially impairs HRX-ENL myeloid transformation, establishing that the DNA-binding motifs of HRX are required for the oncogenic capacity of the fusion.

    PMID 9418860DOI
    What this evidence cannot show

    Retroviral overexpression of a deletion series in primary murine myeloid cells with an in-vitro immortalisation readout; the fusion is HRX-ENL (KMT2A::MLLT1), not KMT2A::AFF1, and deletion of a DNA-binding element can destabilise a protein or disrupt neighbouring structure, so 'required for transformation' does not by itself localise the defect to DNA contact.

    Evidence that argues against this (1)
    • Deleting the AT-hooks may disable the fusion without the AT-hooks being the address, because the retained N-terminus works as a composite. Milne et al. mapped an essential MLL1 recruitment domain spanning the CXXC domain and the PHD fingers, controlled by direct interaction with the PAF elongation complex and by H3K4me2/3, and showed that fusion proteins - which lack the PHD fingers - require prebinding of a wild-type MLL1 complex PLUS CXXC recognition of DNA for stable HoxA9 association. In that architecture, removing a basic DNA-contacting element upstream could destabilise a multivalent assembly whose specificity is supplied elsewhere. A 1998 deletion series scoring immortalisation cannot separate 'the hooks are the address' from 'the hooks are one of several contacts a composite needs', and the entry should not read it as the former. Milne TA et al. 2010
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The A.T-hook is a peptide motif, consensus T-P-K-R-P-R-G-R-P-K-K, that binds the minor groove of A.T-rich DNA stretches and is competed by the minor-groove ligands netropsin, distamycin and Hoechst 33258.

    PMID 1692833
    What this evidence cannot show

    Synthetic peptide and purified HMG-I in vitro, with molecular modelling; the motif definition is from HMG-I/Y, and transfer of its binding preference to any other protein's copies of the motif is an inference. Note the paper's own title says 'recognizing DNA structure'.

    Evidence that argues against this (1)
    • The motif's own structure describes an architectural cofactor rather than a targeting address, which cuts against using it as the model for KMT2A targeting. Huth et al. solved HMG-I(Y) bound to the interferon-beta PRDII site and reported a one-protein-to-two-DNA stoichiometry and a new architectural minor-groove binding motif whose function is to STABILISE B-DNA so that other transcription factors can bind in the opposing major groove, with binding mediated by a central Arg-Gly-Arg plus two accessory modules of unequal contribution. Netropsin competition establishes minor-groove occupancy; it does not establish that occupancy specifies location. A protein whose AT-hooks work like HMG-I(Y)'s would be recruited WITH other factors, not by its own hooks. Huth JR et al. 1997
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    AT-hook motifs fall into three sequence types, are prevalent in nuclear and chromatin proteins including HRX/ALL-1, and function as a versatile minor-groove tether.

    AT-hook motifs identified in a wide variety of DNA-binding proteins · Aravind L, Landsman D · Nucleic Acids Research 1998
    PMID 9742243DOI
    What this evidence cannot show

    Purely computational: pattern searches and position-dependent matrices over a sequence database, with no experimental binding data. 'Versatile minor groove tether' is the authors' functional interpretation of sequence conservation.

    Evidence that argues against this (1)
    • A computational classification cannot be used to predict binding behaviour for a specific protein's copies, and the size of the gap is known. In the protein the motif was defined in, Huth et al. measured roughly 100-fold lower affinity for the third DNA-binding domain than for the second, tracing it to the absence of one of two accessory contact modules that flank the central Arg-Gly-Arg. Sequence-type assignment does not capture that difference. So 'HRX/ALL-1 carries AT-hooks of type X, which are versatile minor-groove tethers' predicts nothing quantitative about KMT2A's hooks 1, 2 and 3, and this entry's targeting argument rests on an unmeasured assumption of equivalence. Huth JR et al. 1997
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The MLL AT-hook region binds cruciform DNA, recognising the structure rather than the sequence of the target DNA, and 11q23 translocations split this AT-hook cruciform DNA-binding region and the transcriptional repression domain away from the activation domain.

    PMID 7938000DOI
    What this evidence cannot show

    Bacterially expressed recombinant AT-hook domain assayed by gel mobility shift, plus GAL4-fusion reporter assays in transfected cells. A 1994 EMSA with a limited probe panel; the cruciform preference has not been re-tested with modern methods, and no structure of an MLL AT-hook bound to DNA exists. It is nonetheless the only direct assay of MLL's own AT-hooks.

    Evidence that argues against this (1)
    • The motif class to which these hooks belong is defined as a minor-groove A.T-rich binder, so the structure-selective result may be a property of the probe set rather than of the protein. Reeves R et al. 1990
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    11q23 breakpoints within the MLL breakpoint cluster region correlate with scaffold attachment regions and topoisomerase II consensus binding sites.

    PMID 8634439
    What this evidence cannot show

    Breakpoint mapping in 31 de novo acute leukemia and 8 treatment-related AML samples with in-vitro SAR-binding assays and predicted topo II consensus sites; correlation, not causation, and SAR mapping methods of the era are known to be sensitive to extraction conditions.

    Evidence that argues against this (1)
    • The SAR half of this correlation is an operational artefact risk, not a stable annotation. Pederson's review sets out the long-running case against the nuclear matrix fraction: it is generated by harsh extraction under precipitation-prone conditions, the in vivo biophysical properties of the nucleoplasm are difficult to reconcile with the inferred filament network, and scepticism about whether the fraction represents the living nucleus has persisted for decades. Scaffold attachment regions are defined by binding to that fraction. A 1996 correlation between 11q23 breakpoints and SAR assignments is therefore a correlation with an extraction outcome, and because both SAR capture and AT-hook binding favour AT-rich sequence, the correlation may be about base composition rather than architecture. Pederson T 2000
  7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A 35-kDa MLL amino-terminal fragment spanning two AT hooks is sufficient to cause cell-cycle arrest, p21 and p27 upregulation and partial monocytic differentiation when conditionally expressed in U937 cells.

    The amino terminus of the mixed lineage leukemia protein (MLL) promotes cell cycle arrest and monocytic differentiation · Caslini C, Shilatifard A, Yang L, Hess JL · Proceedings of the National Academy of Sciences USA 2000
    PMID 10688900DOI
    What this evidence cannot show

    Tetracycline-inducible overexpression in the U937 monoblastic line, a system where supraphysiological levels of a basic DNA-binding fragment could arrest cells non-specifically. Two of the three annotated AT hooks are present in the construct.

    Evidence that argues against this (1)
    • Tetracycline-induced overexpression of a small basic DNA-binding fragment is the condition under which concentration-dependent artefacts are strongest. Chong et al. used live-cell single-molecule imaging to show that intrinsically disordered low-complexity domains form local high-concentration interaction hubs at genomic loci, that those hubs STABILISE DNA binding and recruit RNA Pol II, and that the behaviour is concentration- and partner-dependent, with rapid reversible interactions at physiological levels. An untethered 35-kDa fragment expressed above endogenous levels can therefore occupy and stabilise binding across the genome in a way the intact protein does not, and cell-cycle arrest with p21/p27 induction is a plausible downstream consequence of that rather than evidence that the fragment carries a specific address or function. Chong S et al. 2018
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The MLL amino terminus, spanning the AT-hook region, is sufficient to confer punctate nucleoplasmic distribution including nuclear-envelope-attached regions and the nucleolar periphery, retention in the interphase nuclear matrix, and colocalisation with topoisomerase II on mitotic chromosomal scaffolds, with the same pattern for MLL-AF9 and MLL-AF4 chimeras.

    PMID 11069025DOI
    What this evidence cannot show

    Tetracycline-inducible overexpressed constructs in U937 and HeLa, neither KMT2A-rearranged; the targeting fragment contains the AT-hooks and also the SNL-1 element at aa 388-432, so the experiment cannot attribute the address to the hooks specifically.

    Evidence that argues against this (1)
    • Two of the three compartments in this claim are extraction-defined, and the review that catalogues the objection is unusually direct about it. Pederson describes the nuclear matrix as a fraction whose functional interpretation has never been convincingly established, produced under harsh extraction and precipitation-prone conditions, and argues that the biophysical properties of the living nucleoplasm sit poorly with an extensive internal filament network; he notes the scepticism that continues to surround the fraction as a representation of the in vivo state. 'Retention in the interphase nuclear matrix' and 'colocalisation with topoisomerase II on mitotic chromosomal scaffolds' are both readouts of that preparation. What survives without it is the punctate nucleoplasmic pattern in fixed, overexpressing cells, which is a much smaller claim. Pederson T 2000
Connected to