Non-elongation KMT2A partners considered and not written: what the searches actually returned, partner by partner

All searches were raw E-utilities esearch against PubMed with [Title/Abstract] scoping, not the PubMed MCP search_articles tool, which ANDs every keyword and returns zero on topics with large literatures. Both halves of each intersection were counted separately so that a zero could be interpreted: 'menin'[Title/Abstract] returns 1264 records and 'septin OR septins'[Title/Abstract] returns 1936, so their intersection of 0 is informative rather than an artefact of a dead search 2.

Septin nomenclature is a trap in its own right. HGNC renamed the family (SEPT9 to SEPTIN9 and so on) and the historical literature uses three further names for SEPTIN9 alone. Intersecting (MLL OR KMT2A) with each symbol in turn returns: SEPT9 = 20, MSF = 14, SEPTIN9 = 1, AF17q25 = 3; SEPT6 = 19, SEPTIN6 = 5; SEPT5 = 9, SEPTIN5 = 1, CDCREL = 1, hCDCrel = 6; SEPT2 = 10, SEPTIN2 = 0; SEPT11 = 5, SEPTIN11 = 1 2. Every septin's literature is retrieved mainly by its retired symbol. An enumeration keyed to the current HGNC symbol would have concluded that KMT2A-septin fusions are a curiosity with one or two papers each. This is stated in full because it is the single most reusable finding in this analysis.

One further caution recorded for the next agent: the wildcard SEPT* in a PubMed query matches 'September' and produces large amounts of unrelated noise (dairy science, lupus meta-analyses). It was checked by reading titles and then abandoned in favour of explicit symbol lists.

All six clear the bar of at least two independent published observations. SEPTIN2 is included on literature count (4 independent case reports plus an expression study) despite a recombinome count of 1, and the discrepancy is itself recorded as a finding about referral ascertainment 1.

• SEPTIN3 — 1 patient in the recombinome (adult AML) 1; (MLL OR KMT2A) AND (SEPT3 OR SEPTIN3)[Title/Abstract] AND (fusion OR translocation OR leukemia OR leukaemia) returns 0 records 2. A partner known only from a genomic-breakpoint table with no publication describing it. Rejected.
• SEPTIN4 — not a KMT2A partner at all. The single record intersecting MLL/KMT2A with SEPT4 is the Sept6-knockout paper, in which Sept4 appears only as a genetic background 23. Rejected, and recorded because SEPT4 is easy to mistake for a partner when scanning septin-family reviews.
• MYO1F — 5 patients in the recombinome; 1 PubMed record on the intersection 12. Deferred rather than rejected: the count is real but there is nothing to read.
• SNX9 — 1 patient, 1 record 12. Rejected.
• ITSN1 — 1 patient in the recombinome, 0 records on the intersection 12. Rejected.
• ABI2 — 2 patients, 1 record 12. Rejected; noted because ABI2 is the paralogue of a partner that was written, and the pair would be the natural comparison if anyone ever studies ABI1.

• GAS7 — 3 patients in the recombinome but 5 PubMed records including the foundational dimerisation work, where the GAS7 coiled coil was shown necessary and sufficient 12. GAS7 is mechanistically load-bearing for the whole oligomerisation argument and is cited throughout this beat, but it is a cytoplasmic non-septin partner whose own entry belongs with EPS15/GAS7 rather than here. Flagged for a future beat.
• GPHN (gephyrin) — 3 PubMed records; not in the recombinome top list 2. Also mechanistically load-bearing (the 15-aa minimal oligomerisation domain) and cited here, but clinically negligible. Flagged.
• MLLT11/AF1q — 28 patients, mostly AML, 14 PubMed records 12. This is a substantial partner and the graph holds only four passing mentions of it. It is 90 aa with no chromatin domain and no annotated oligomerisation module, which makes it the sharpest remaining challenge to every model in this beat. Deliberately deferred rather than written badly; see the open question.
• EPS15 — 55 patients, 13 records 12. The graph already holds protein-eps15-domains and subtraction-eps15-eh-domains but no fusion entry. Flagged to the coordinator as a real gap in an adjacent beat.
• MLLT6/AF17 — 27 patients, 13 records, an existing gene-mllt6 entry 12. MLLT6 has a leucine-zipper dimerisation motif and is arguably a bridge case between the classes; not written here because an entry exists.
• NEBL — 3 patients, 5 records 12. Relevant mainly because it shares the 10p11-12 ambiguity with ABI1 and MLLT10.

This enumeration covers partners that supply no elongation or chromatin module. It does not cover the 63 KMT2A partners seen only once in the recombinome as a group 1; each would need its own nomenclature-aware search, and the expected yield is near zero by construction. It also does not cover reciprocal fusion proteins. A reader should treat the counts here as a floor for the non-elongation class and as no statement at all about the long tail.

Three record counts in this enumeration were wrong, all in the direction of UNDER-counting, and all found when the partners that had been deferred here were actually written. The counting method is the common cause: intersections run on one symbol spelling, against a recombinome table that is itself inconsistent.

GPHN is two index cases, not one therapy-related patient. PMID 11289145 reports t(11;14)(q23;q24) generating an MLL-gephyrin fusion in de novo acute monoblastic leukaemia; PMID 11579461 reports GPHN as a novel MLL partner in a therapy-related acute undifferentiated leukaemia. Describing GPHN as a single therapy-related patient understated both its recurrence and its de novo occurrence — which matters, because gephyrin is the load-bearing counterexample in this project's arguments about whether SEC recruitment is the mechanism.

MYO1F has two records, not one, and the second is the one that counts: PMID 21620472, The t(11;19)(q23;p13) fusing MLL with MYO1F is recurrent in infant acute myeloid leukemia. A partner dismissed as a singleton is recurrent in the exact population this project is about. (That paper has no abstract in PubMed, so the claim here is limited to its title.)

MLLT11 counts understate by roughly four-fold — the controlled intersection is 20 records, not 14, and the gene-level count is 88.

A fourth alias failure mode was found alongside them and belongs with the others recorded in the brief: the recombinome prints MYO1F as "MYOF1", a typographic variant that collides with the real and unrelated gene MYOF. And the recombinome is internally inconsistent on EPS15, with Table 1 giving 55 patients while the prose says "12 out of 51".