The check I called decisive was not: the haem signal is still open (run)

ana-depmap-kmt2a-selective-scan-full recorded its own first open question in strong terms: "checking their lineage subtype is the first thing to do — it is free, and if it explains the signal the rest is moot." That was right to prioritise and wrong to call decisive.

The erythroid version is genuinely excluded. Cellosaurus disease annotation for all 11 rearranged lines shows four childhood/adult B-ALL and seven myeloid, none erythroid or megakaryocytic 1.

And the comparison group works against the signal, not for it. Eight erythroid or megakaryocytic lines — HEL, HEL 92.1.7, SET-2, TF-1, F-36P, M-07e, CMK-11-5 and OCI-M2 — sit among the 43 non-rearranged leukemia lines. Those cells have high haem demand by their nature, so their presence makes the comparison group's mean haem dependency more negative, which shrinks the difference attributed to the fusion. The haem cluster was detected despite them 1.

But checking the annotation surfaced a different confound, and it is the more plausible one: six of the seven rearranged AML lines are monocytic or monoblastic — MV4-11, THP-1, NOMO-1, SHI-1, MOLM-13, MOLM-14. Monocytes and macrophages handle iron and haem as ordinary physiology.

The design was clean: exclude the rearranged lines entirely, split the remaining leukemia panel by monocytic annotation, and ask whether the haem genes are still selectively required. If they are, the signal is lineage and the fusion story collapses.

It failed on sample size. Only 4 monocytic non-rearranged leukemia lines exist in the panel — OCI-AML-3, U-937, Mono-Mac-1 and MUTZ-8 — against 35 non-monocytic, with 4 unresolved.

Two things make this uninterpretable rather than merely negative.

First, GAPDH moves at p = 0.031 while ACTB does not. A housekeeping gene separating the groups means there is a global sensitivity offset between monocytic and non-monocytic lines, so every difference in the table is partly that offset. This is the same failure the line-normalisation control was built for elsewhere in this project, and at n = 4 there is not enough data to correct it.

Second, SPI1 does not reach significance (p = 0.126) despite a large difference. SPI1/PU.1 is a monocyte-identity gene and should separate cleanly on a split that captures monocytic identity. That it does not means the split is underpowered to detect even what it was designed around, so its silence on the haem genes carries almost no information.

This entry and the one before it rest on four genes picked because they were already extreme - ALAD 14, IREB2 15, UROS 41, ALAS1 47 in a scan of 17,728. That is the same after-the-fact selection this project has had to correct twice elsewhere, most recently when three hand-picked pathway controls made the Fanconi result look isolated and a systematic sweep of fourteen found a second pathway a tenth away.

So: all eight canonical haem biosynthesis enzymes, fixed by the pathway's chemistry rather than by their ranks.

All eight are in the top 3% of the genome. Median rank 115 against a uniform expectation of 8,864, and 0 of 20,000 random eight-gene draws reach it. The signal is not four cherry-picked genes; it is the pathway.

Five of the eight are not broadly essential (ALAS1, ALAD, HMBS, UROS, PPOX), so this is not simply a rediscovery of genes every cell needs - though UROD, CPOX and FECH are broadly essential and do inflate the pathway median.

The internal control is the part worth keeping. ALAS1 and ALAS2 catalyse the same first committed step; ALAS1 is ubiquitous, ALAS2 erythroid-specific. ALAS1 ranks 47 and ALAS2 ranks 9,992 - a separation of 9,945 ranks between two isoforms of one reaction. If the pathway were scoring for some artefactual reason there would be no reason for the isoforms to separate at all.

But that control also cuts the other way, and it is the sharpest caveat in this entry. ALAS2 scoring low is exactly what "ALAS2 is not expressed in these lines" predicts, and expression has not been checked. It demonstrates that the measurement discriminates; it does not show the discrimination is about the fusion. The monocytic-lineage confound this entry was written about is untouched by the pathway test - a broader haem signal is, if anything, easier to explain by myeloid identity than a narrow one.

The section above offered ALAS1 at rank 47 against ALAS2 at 9,992 as an internal specificity control, while flagging that "ALAS2 scoring low is exactly what 'ALAS2 is not expressed in these lines' predicts, and expression has not been checked".

It has now been checked. ALAS2 is at 0.10 nTPM in all seven panel lines held by the Human Protein Atlas - at or below detection. The flagged explanation is the correct one and the control is withdrawn.

What survives: all eight canonical enzymes are expressed, at 7.9 to 67.0 nTPM, so the pathway-level result - median rank 115 of 17,728, 0 of 20,000 random draws - is not an expression artefact. The monocytic-lineage confound this entry exists to discuss is untouched, and remains the open question.