Of 262 curated interactors this graph never wrote down, five are selectively required in KMT2A-rearranged lines - and one of them is FANCD2, which binds menin and sits inside a selectivity signal this project has failed four times to explain

This graph has carried an unresolved Fanconi result for many cycles. The Fanconi complex ranks 2nd of 719 complexes for KMT2A-rearranged selectivity, median gene rank 125 against a uniform expectation of 8,864. Four explanations were tested and all four failed, and the ledger records that the signal survives with none standing. Lineage was refuted in the surprising direction: both halves clear the null and the lymphoid half is stronger.

Separately, and never connected to it, PMID 12874027 is titled Menin associates with FANCD2, a protein involved in repair of DNA damage (Jin, Mao, ... D'Andrea, Hua; Cancer Res 2003; fetched and read back). It is cited nowhere in this graph. And the graph has written gene-fanca, gene-fancf, gene-fancg, gene-fanci, gene-fancl, gene-brip1, gene-ube2t and gene-faap100 - but not gene-fancd2, which is the central effector of the pathway and the only FA protein with a published direct menin interaction.

Now this analysis, which did not set out to look for it: FANCD2 comes back selective, via the MEN1 bait, and it is more required in the lymphoid half (-0.70) than the myeloid (-0.48) - the same direction as the Fanconi complex result.

The candidate fifth explanation is that the FA selectivity signal runs through menin binding FANCD2, not through crosslink repair. That would explain why pathway-level reasoning kept failing: the unit would be one protein with a second job, not a pathway.

It is a candidate and nothing more. A 2003 co-immunoprecipitation predates most of what is known about the ID2 complex; an association is not a function; a -0.153 difference on 11 lines is small; and three numbers agreeing is the situation in which a story becomes attractive before it becomes true. This project has already adopted and then killed four explanations for this signal, and the right posture toward a fifth found by its own instruments is suspicion. Q36 was dispatched to read the 2003 paper properly, with instructions that a clean negative is worth more than a stretched positive.

HDGFL2 is UniProt Q7Z4V5, synonyms HDGF2, HDGFRP2 and HRP2 - fetched, with the primary name read back. It is LEDGF's paralogue, and it appears here against the PSIP1 bait.

That is not a coincidence in this graph. The Q33 LEDGF beat corrected a premise of mine using PMID 22396646: LEDGINs remain active in cells with the PSIP1 IBD exons deleted, by acting through HRP-2's IBD. HRP-2 is the documented functional substitute for LEDGF in the one system where LEDGF removal was pushed to completion.

So HRP-2 is a paralogue that (a) carries the same IBD module, (b) has been shown to substitute when LEDGF is gone, and (c) is now selectively required in KMT2A-rearranged lines with no entry in this graph. The ledger's standing LEDGF hole - no published experiment removes LEDGF from an established KMT2A-rearranged leukaemia in any animal - acquires a second half: any such experiment that does not also remove HRP-2 may be measuring paralogue buffering rather than LEDGF dispensability.

The numbers are weak and should be said so: pan-essentiality -0.119 and an index mean of -0.239 mean this is a small effect on a gene that is barely required anywhere, and the myeloid/lymphoid split (-0.35 / -0.05) runs opposite to FANCD2's.

DAXX (Q9UER7) has the largest selective difference in the whole unwritten set, -0.305, appears against four baits (AFF4, DOT1L, MLLT10, MLLT3), is not pan-essential (-0.374), and is most required in the lymphoid half at -0.89 on four lines.

DAXX is the H3.3 histone chaperone partner of ATRX and a component of PML nuclear bodies - both of which this graph already touches, at protein-pml-domains and in the subnuclear beat. It has no entry.

Its database evidence is one laboratory, so on the previous ranking it sat mid-queue. That is the point of this analysis: the two rankings disagree, and where they disagree the dependency ranking is the one tied to something the cells do.

TRAF2 (Q12933) and QKI (Q96PU8) complete the five. QKI's index mean is -0.122 against +0.035 in other leukaemia lines - it is close to not required at all in the comparator, so the difference is real and the effect is tiny.

64 of the 254 are pan-essential at mean gene effect at or below -0.5 across all lines: the ribosome (RPL13, RPL13A, RPLP0, RPL10, RPS11), the spliceosome and RNA machinery (SRSF6, SRSF7, NCBP2, CNOT1, RAN), the chaperonin CCT6A, MCM6, HUWE1, OGT, CDK7 - and MYC, MED19 and CDK6.

Eighteen of the 23 largest selective differences are in this group, and the single most extreme difference in the whole set (-0.527) belongs to a pan-essential gene. A difference between two lethal effects is dominated by where each arm sits against the Chronos floor, which this project has corrected three times and did not have to rediscover here because the filter was built in.

CDK6 deserves a note of its own. The ledger carries a standing priority - CDK6, not CDK4/6 - requiring any entry recommending the class to say which gene carries the rationale. CDK6 has no entry of its own, and it comes back pan-essential, which is an argument about the therapeutic window that the priority as written does not make.