Endogenous aldehydes and silenced ALDH2 — the rival explanation for an FA dependency in acute leukaemia, and the only one with an experiment behind it

Two parallel defences stand between a cell and its own aldehyde output. ALDH2 destroys acetaldehyde before it reaches DNA. The FA pathway repairs the interstrand cross-links that get through. Knock out one and the cell is fine; knock out both and it is not. That epistasis is established in mouse haematopoiesis and is the basis of the leukaemia result here 1.

What makes it the serious rival rather than a footnote:

• It nominates UBE2T and FANCL specifically — the E2 and the E3 — which is exactly the sub-module that mech-fa-core-complex-heterogeneity-kmt2a identifies as carrying this project's signal. Two independent routes converging on the same two enzymes is not nothing.
• It is a synthetic lethality with a defined, measurable marker: ALDH2 promoter methylation. That is a testable reassignment, not a vague alternative.
• It explains the dependency without invoking transcription at all, so it needs none of the unmeasured links in mech-transcription-replication-conflict-kmt2a-r.
• It is in acute myeloid leukaemia, the right disease, with human cells.

Against the ALDH2 model explaining the KMT2A finding. If the FA dependency in acute leukaemia were simply AML-with-silenced-ALDH2, then an AML-versus-other-leukaemia contrast should show it. It shows the reverse: the eight headline genes average +0.095 in non-KMT2A AML lines against the rest of the leukaemia panel, and the FA set's permutation p is 0.999 — enriched away from dependency 2. And the KMT2A increment survives when lineage is held exactly constant, in both directions: -0.334 for KMT2A-rearranged AML against non-KMT2A AML, -0.308 for KMT2A-rearranged B-ALL against non-KMT2A B-ALL 2.

Why that is not a refutation. The published screen compared AML lines against other cancer lineages, not against other leukaemias. If every leukaemia line in the comparator arm also silences ALDH2, an AML-versus-other-leukaemia contrast cancels a real shared dependency exactly to zero, and a small excess in one subtype is all that remains visible. The two observations are compatible; they are measured against different baselines.

The experiment that separates them is cheap and has not been done here. ALDH2 expression and promoter methylation are available for every DepMap line. If ALDH2 expression predicts FA dependency across the 54-line leukaemia panel better than KMT2A rearrangement status does, this entry replaces the transcription-replication-conflict entry as the explanation. If KMT2A status still adds after conditioning on ALDH2, it does not. That single regression would resolve the most important open question in this beat.

Speculative and flagged as such, but worth writing down because it is testable.

KMT2A-rearranged infant leukaemia arises in utero, and the fetal haematopoietic compartment is where the aldehyde-FA axis has its sharpest known phenotype — Fanconi anaemia's haematopoietic failure is a fetal and early-postnatal stem-cell attrition, not an adult one. A fusion that expands a fetal-liver-like progenitor pool could be placing cells in a compartment where aldehyde defence is already marginal, which would make an FA requirement a property of the cell of origin rather than of the fusion's transcriptional output.

That predicts the dependency should be strongest in the most fetal-like models — infant KMT2A::AFF1 lines and patient-derived xenografts — and weakest in adult-derived AML lines. The four B-lineage lines in this project's index arm are the closest thing available to that test, and they are indeed the stronger half for seven of the eight genes 2. That is a suggestive coincidence and nothing more: four lines, and B-lineage is confounded with everything else about them.