KMT2A::GPHN (MLL-gephyrin), t(11;14)(q23;q24) — the 15-residue counterexample

Case 1 (Kuwada et al., Cancer Res 2001). De novo acute monoblastic leukaemia with t(11;14)(q23;q24). A human homologue of gephyrin — then known mainly as a rat glycine-receptor-associated protein that anchors glycine and GABA-A receptors to microtubules — was found fused to MLL. Alternative splicing of the human gephyrin gene produced two different fusion cDNAs: in one, GPHN fused in frame to MLL exon 9 and the chimeric product carried the gephyrin C terminus including the tubulin-binding site; in the other the reading frame terminated shortly after the fusion point, attaching only seven amino acids of no known function to the MLL N terminus — a de facto truncated MLL whose functional significance the authors explicitly left unresolved 1.

Case 2 (Eguchi et al., Genes Chromosomes Cancer 2001). A child with acute undifferentiated leukaemia three years after intensive chemotherapy including the topoisomerase-II inhibitor VP16. Screening a cDNA library from the leukaemic cells found a novel MLL-GPHN transcript. The fusion encodes MLL AT-hook motifs and the DNA-methyltransferase-homology (CXXC) domain fused to the C-terminal half of gephyrin, including a presumed tubulin-binding site and a MoeA-homologous domain. Genomic breakpoint analysis found potential in vitro topoisomerase-II DNA-binding sites spanning the breakpoints in both genes, and no flanking sequences that might mediate homologous recombination — the authors' inference being VP16/topoisomerase-II-induced double-strand breaks followed by error-prone NHEJ 2.

The enumeration this beat inherited described GPHN as coming "from a single therapy-related patient". That is not right: there are two independently published index cases, one de novo monoblastic and one therapy-related undifferentiated, and only the second is therapy-related. The correction is recorded in ana-kmt2a-deferred-partner-enumeration.

The dominant model of KMT2A leukaemia is that the fusion partner supplies transcription-elongation machinery — AFF1, AFF4, MLLT1, MLLT3, ELL, and through them P-TEFb, the super elongation complex and DOT1L. KMT2A::GPHN cannot work that way. Gephyrin is a molybdenum-cofactor-biosynthetic enzyme and a synaptic receptor scaffold; UniProt annotates no chromatin-binding, transactivation or elongation-complex function for it 4, and Eguchi et al. state plainly that MLL-GPHN transforms "despite lack of transcriptional activity" 3.

What remains is proximity. The 15-amino-acid tubulin-binding domain is necessary and sufficient, and it confers oligomerisation capacity on MLL 3. Fifteen residues is small enough that it is hard to argue the module smuggles in a hidden effector complex: whatever it recruits, it recruits by making two or more MLL N-termini adjacent.

This is the reason the fusion earns an entry at a patient count of two. Almost every mechanistic statement in this graph about KMT2A leukaemia is a statement about what the partner brings. KMT2A::GPHN is the case where the partner brings, as far as anyone can tell, nothing but a dimerisation surface — and it still makes leukaemia in mice, in two lineages.

Three limits, all of which matter for how hard this counterexample can be pushed.

1. Sufficiency is not necessity for the common fusions. Martin et al. showed that induced dimerisation of truncated MLL immortalises bone marrow and upregulates Hoxa7, Hoxa9 and Meis1, and in the same paper stated that MLL-AF9 is not dimerised in vivo 6. Forced oligomerisation is a second, sufficient route into the same disease, not a re-description of the first one. An argument that runs "GPHN shows SEC recruitment is not the mechanism" overreaches; what GPHN shows is that SEC recruitment is not the only mechanism.
2. Single laboratory, abstract-level reading. The 15-residue result is one paper from one group in 2004, and this entry's reading of it is from the abstract: the full text is subscription-only, with no PMC record and HTTP 403 from two publisher mirrors. The residue numbers, the oligomerisation assay, the mouse strain, the latencies and any Hox data are not verified here and are asserted nowhere.
3. Everything downstream is unmeasured. No menin dependence, no H3K79me2, no H3K4 methylation, no HOXA ChIP, no transcriptome, no localisation imaging, no structure 5. The fusion is mechanistically decisive and biochemically almost blank.

Cell type of origin / target. Mouse haematopoietic progenitor cells transduced with MLL-GPHN produced both myeloid and lymphoid lineage leukaemias, which the authors read as MLL-GPHN targeting multipotent progenitor cells 3. No marker-defined compartment (LMPP, GMP, CMP) was resolved. In patients, one case was acute monoblastic (M5-type, myeloid) 1 and the other acute undifferentiated leukaemia 2 — the human phenotypes are consistent with a multipotent target, and the undifferentiated case is about as close to a lineage-ambiguous presentation as the classification allows.

Cell state. Not characterised. No differentiation-block phenotype has been described for this fusion beyond the leukaemias themselves, and no serial-replating differentiation readout is reported in a source this entry could verify.

Subcellular localization. Not reported for the fusion. Wild-type gephyrin is submembranous and tubulin-associated 4. The obvious experiment — does MLL-GPHN relocalise to the nucleus, as MLL-SEPTIN6 does to nuclear speckles — has not been done.

Subnuclear localization. Not reported. Stated explicitly rather than inferred from other cytoplasmic-partner fusions.

Epigenetics. No epigenetic readout of any kind exists for KMT2A::GPHN 5. This entry declines to import the MLL-AF9 H3K79me2 picture onto it.

There is no clinical literature. Two patients, no cohort, no survival estimate, no trial, no cell line, no PDX. Nothing in this entry should be read as prognostic.

The menin question is the actionable one and it is open in an interesting way. Menin binding is a property of the retained KMT2A N-terminus (MBM1/MBM2), which both reported KMT2A::GPHN transcripts keep 12. If menin inhibition kills a fusion whose partner supplies nothing but an oligomerisation surface, that is strong evidence that menin dependence is a property of the KMT2A moiety rather than of partner-recruited elongation machinery — which would strengthen the menin-interface hypothesis and broaden it to every partner class. If it does not, the hypothesis narrows sharply. The experiment is cheap: the retroviral MLL-GPHN construct is published, and revumenib and ziftomenib are available compounds.