Fifth explanation for the Fanconi signal: it runs through menin binding FANCD2, not crosslink repair — tested, refuted, prior 0.07

Five things, in descending order of how easily each could be checked.

1. Menin and FANCD2 associate in a human haematopoietic cell, not only in mouse embryonic fibroblasts. Never tested.
2. That association is functional — losing it changes something FANCD2 does (foci, chromatin loading, monoubiquitination kinetics, checkpoint timing). Never tested in any cell type.
3. The association involves the MBM pocket, or some surface a menin inhibitor perturbs. Otherwise the hypothesis has no pharmacological consequence and cannot explain anything about revumenib. Never tested; the only structural handle is the curated sufficient region menin 214-390 (interface-menin-fancd2).
4. The FA-module dependency in KMT2A-rearranged cells depends on menin, i.e. the FA requirement should collapse in a MEN1-null background or under saturating menin inhibition. Never tested.
5. The effect concentrates on FANCD2 and the ID2 module. It does not: the graph's own measurement puts the signal in the ubiquitin-transfer module (FANCF, FANCL, UBE2T, FANCA, FANCG, FAAP100) with the ID2 substrate weak 4.

Point 5 is already false as stated, which is why the prior was set at 0.12 and not 0.35, before the permutation null took it to 0.07.

MEN1 correlating with FA-module dependency at r = 0.670 [0.490, 0.795], surviving removal of the KMT2A-rearranged lines at 0.486 [0.218, 0.686] and lineage-by-fusion residualisation at partial r = 0.506 [0.274, 0.683], with an ACTB control at 0.096 and -0.034 respectively 1.

That is a well-controlled correlation and it is the only positive evidence for any menin-Fanconi link that was generated in this decade. But it is a correlation of dependency profiles, and there is at least one boring reading that fits it perfectly and requires no menin-FANCD2 contact at all: MEN1 is the deepest selective dependency in this panel (1st of 8,269 on raw effect), and lines that are deeply menin-dependent are KMT2A-rearranged, and KMT2A-rearranged lines are FA-dependent. Residualising on KMT2A status is meant to break that chain, and partially does — but a graded, continuous "how fusion-driven is this line" variable is not captured by a binary label, and the residual correlation is exactly what such a variable would produce.

So: the correlation is real, it is not obviously an artefact, and it does not distinguish menin binds FANCD2 from menin-dependence and FA-dependence are two readouts of the same underlying state. The second is the null and it has not been excluded.

Three things stop it going to zero, and none of them is a dependency number.

The 2008 negative is a fly modifier screen, and its authors said so: with MEN1 loss producing no eye phenotype of its own, they could not evaluate epistasis 3. It is the best test that exists and it is not a definitive one.

The surviving alternative is specific, not a hedge. The same authors proposed that menin and FANCD2 might act together in the IR-induced S-phase checkpoint rather than in crosslink repair 3. FANCD2's checkpoint function runs through ATM phosphorylation of Ser-222, a modification independent of the K561 ubiquitin switch that defines canonical FA repair (gene-fancd2). A checkpoint-arm interaction would be invisible to every assay in the 2008 paper and to every FA-pathway assay in common use. It is a real, narrow, testable version of the hypothesis and nobody has tested it in eighteen years.

A menin-inhibitor-to-DNA-repair link already exists in the right disease. Combining an HDAC inhibitor with the menin-MLL inhibitor MI-3 in MLL-rearranged AML disrupts the DNA damage checkpoint at CHK1 and CHK2 and impairs homologous recombination, selectively in MLL-rearranged cells 6. That is not evidence for the FANCD2 bridge — the paper never looks at FANCD2, and a combination effect is not a menin effect — but it establishes that menin-MLL inhibition and the damage response are not orthogonal in this disease, and it is the nearest thing to a positive control the hypothesis has.

The bridge, as a crosslink-repair mechanism, is dead. It was tested in 2008 and failed, the failing test is now in the graph (mech-men1-and-fancd2-act-in-distinct-crosslink-repair-processes), and the internal DepMap data independently put the selectivity signal in the ubiquitin-transfer module rather than in the ID2 effector menin actually binds.

The third leg has also gone. A dependency triage reached FANCD2 through the MEN1 bait at a difference of -0.153 and looked like weak corroboration. The permutation null that triage had skipped gives p = 0.1241 over 50,000 two-sided label permutations, with nothing in its shortlist surviving Bonferroni across 254 genes. Two independent computations now put FANCD2 outside significance 4. The prior is lowered from 0.12 to 0.07 on that basis, and the hypothesis is recorded as refuted rather than merely improbable.

What is NOT refuted. The complex-level Fanconi result - the FA ubiquitin ligase complex ranking 2nd of 719 Complex Portal complexes, median member rank 125 against a uniform expectation of 8,864 - was measured on a different unit and is unaffected. A set of genes can move together in a way no single member does alone. Open at the complex level, closed at the FANCD2 level is the correct summary, and refusing to let the gene-level negative close the complex-level question is as important as refusing to let the bridge stand.

A narrow residue survives: menin and FANCD2 in the S-phase checkpoint, which the 2008 authors proposed and nobody has examined in eighteen years. It is worth one experiment, not a programme.

The honest summary for the ledger is that the Fanconi signal now has five tested explanations and five failures, and the fifth failed faster and more cheaply than the previous four because the refuting paper already existed and nobody in this project had found it.