q172's four-line index arm is three KMT2A::AFF1 lines from TWO donors plus an ENL line; dropping the ENL line moves MEN1 7.7x more than DAXX, and menin's lost significance is variance, not effect
also called q176 · the index arm is not all AFF1 · menin fails on variance
A dependency comparison last cycle used four leukaemia cell lines as the 'fusion' group. One of the four carries a different fusion from the one this project is about, and two of the other three come from the same child - so the real evidence is three lines from two people. Removing the odd line barely changes DAXX but changes menin a lot. And menin's apparently weaker result is not menin becoming less needed: the comparison group becomes more menin-dependent, while menin stays exactly as needed as before.
The same thing, for a scientist
q172's numbers reproduce exactly; this is about what the **index arm** is made of and what its p-values mean. Both findings bear on the menin-interface hypothesis. 🔴 **1. One of the four 'index' lines is not the index fusion.** `HB11;19` (CVCL_8227) is **KMT2A-MLLT1 — ENL**, not AFF1 [E1]. The arm is correctly labelled *KMT2A-rearranged*, but this corpus's menin-interface hypothesis is about **KMT2A::AFF1**. And `SEMK2` (CVCL_S906) is **derived from `SEM`** (CVCL_0095), the same 5-year-old. **For the actual index fusion the evidence is 3 lines from 2 donors.** 🟠 **2. The odd line is load-bearing for menin and not for DAXX.** Dropping HB11;19 moves the index mean by **MEN1 +0.2383 vs DAXX +0.0308 — a 7.7× difference** [E2]. HB11;19 is the most menin-dependent line in the arm at **−1.9831**, so menin's index mean was substantially propped up by a fusion that is not the subject. 🔴 **3. 'MEN1 stops discriminating' is true of the p-value and invites the wrong inference.** Decomposed [E3]: - **MEN1's index arm is CONSTANT across all three strata at −1.2682** — it is the same four lines every time. - **MEN1's comparator strengthens: −0.435 → −0.594 → −0.639** as the comparator narrows to fusion-driven B-ALL. - So **menin does not become less required. Other fusion-driven B-ALL becomes more menin-dependent.** - **MEN1's effect in the tightest stratum (−0.6292) is still LARGER than DAXX's (−0.4545).** It loses significance on **variance** (index sd **0.594**, range **1.520**) against DAXX's (sd **0.078**, range **0.181**), at n=4. 🔵 **4. At the level the hypothesis is actually about, the two genes are indistinguishable.** Index fusion only, donor-collapsed: **MEN1 −0.8882, DAXX −0.8471, a gap of 0.0411 on n=2 donors** [E2]. Every claim that DAXX outperforms menin in the B-lineage stratum rests on the ENL line and on treating two donors as four samples. 🟠 **5. The genuinely interesting observation, and a real challenge to the hypothesis:** **MEN1's two AFF1 donors differ 2.84× in magnitude — SEM −1.3132 versus RS4;11 −0.4632** — while DAXX's differ by **1.06×** [E4]. **RS4;11 is a canonical KMT2A::AFF1 line and is only weakly menin-dependent.** If menin dependence were a direct consequence of the fusion's menin interface, two lines carrying the same fusion should not differ this much. **This is the sharpest quantitative challenge to the menin-interface hypothesis currently in the corpus — and it is n=2.**
This corpus already knew the arm is mixed — and the corpus-wide number is worse
Novelty check, run after the analysis and before publishing it. analyses/depmap-scan-line-identities/results.json, generated 2026-09-21, already resolved the identities of the whole DepMap KMT2A arm and records the fusion of every line. So "the index arm is a mixture of fusions" is not a discovery of this analysis — it is a fact this repository established three days earlier and then did not apply.
What that record shows, corpus-wide, is stronger than what q176 found in one lymphoid subset. Of 16 resolved KMT2A-rearranged DepMap lines:
| fusion | lines |
|---|---|
| KMT2A-AFF1 (the index fusion) | 5 |
| KMT2A-MLLT3 (AF9) | 5 |
| KMT2A-AFDN (AF6) | 3 |
| KMT2A-MLLT1 (ENL) | 2 |
| KMT2A-FOXO4 | 1 |
Only 5 of 16 — 31% — carry the index fusion, and they are SEM, SEMK2, RS4;11, MV4-11 and ALL-PO. SEMK2 derives from SEM, so that is 5 lines from 4 donors. Every analysis in this corpus that compares a "KMT2A-rearranged arm" against other leukaemia is, for roughly two-thirds of its index arm, measuring a fusion that is not the subject of the menin-interface hypothesis.
The same record also carries a separate, still-open defect in the source scan: arm membership is fixed from the first gene with data rather than per gene, so 5 of the 16 eligible lines are excluded from every comparison. That is recorded and unfixed.
So q176's contribution is narrower than it first looked: not that the arm is mixed, but (a) that q172's lymphoid arm inherits the problem, (b) that the sensitivity is 7.7× larger for MEN1 than for DAXX, and (c) the required/discriminating decomposition below. The composition point belongs to the 2026-09-21 record.
Why 'required' and 'discriminating' must not be collapsed
A dependency difference is index_mean − comparator_mean. It can shrink for two opposite reasons, and they mean opposite things:
- the index arm weakens — the gene matters less to the fusion. This would be evidence against the hypothesis.
- the comparator strengthens — the gene matters to the comparators too. This is evidence the gene is not selective, and says nothing about whether the fusion needs it.
For MEN1 it is entirely the second: the index arm is a fixed set of four lines and its mean does not move at all between strata. Menin remains the most strongly required of the four genes examined (index mean −1.2682, more negative than DAXX's −0.8864).
That distinction matters because the corpus's hypothesis is about a mechanism — a retained interface the fusion uses — not about selectivity relative to other B-ALL. Menin being needed by ETV6::RUNX1 and TCF3::PBX1 lines too is expected: menin is a broad cofactor. A non-significant difference against fusion-driven B-ALL is not a refutation of the interface; it is a statement about the comparator.
I stated the collapsed version in a prior summary — 'MEN1 and DOT1L both stop discriminating in the B-lineage stratum where DAXX does not' — which is literally true and reads as though menin had weakened. It had not.
What this does NOT rescue, and what it genuinely threatens
Does not rescue DAXX. DAXX's own result survives the restriction almost unchanged (index mean moves 0.0308 on dropping the ENL line, 0.0085 more on donor collapse), which is a point in its favour and is why its consistency, not its size, is the interesting thing about it. But it is still n=2 donors for the index fusion, still fails Bonferroni over the 254-gene selection set, and still has no physical contact in the only MLL-fusion AP-MS interactome.
Does genuinely threaten the interface hypothesis: the 2.84× spread between SEM and RS4;11 on MEN1. Two lines, same fusion, same nominal mechanism, and one needs menin roughly a third as much as the other. Candidate explanations that must be excluded before this means anything — none of which was tested here: RS4;11 is from a 32-year-old and SEM from a 5-year-old, so age/biology differ; the two may differ in MEN1 copy number or expression; Chronos scores are normalised per screen and a single line's value carries real technical variance; and n=2 cannot separate any of these from the fusion.
The experiment this implies is cheap and already half-built in this corpus: menin CUT&RUN or menin-inhibitor dose-response across several KMT2A::AFF1 lines, asking whether menin dependence varies with something measurable — breakpoint exon, HOXA status, IRX1 status — rather than being uniform. A uniform interface predicts uniform dependence.
| script | analyses/q176-the-index-arm-contains-a-non-index-fusion/run.py |
|---|---|
| status | executed |
| corrects framing of | ana-daxx-lymphoid-dependency-under-three-nulls |
| n aff1 lines | 3 |
| n aff1 donors | 2 |
| men1 donor spread fold | 2.84 |
| daxx donor spread fold | 1.06 |
| verdict | menin is the most REQUIRED of the four genes and loses significance only because the comparator strengthens; the index arm is 3 AFF1 lines from 2 donors plus an ENL line |
| prior record | analyses/depmap-scan-line-identities/results.json (2026-09-21) |
index arm
{
"SEM": "KMT2A-AFF1",
"RS4;11": "KMT2A-AFF1",
"SEMK2": "KMT2A-AFF1 (derived from SEM)",
"HB11;19": "KMT2A-MLLT1 (ENL) - NOT the index fusion"
}index mean four line — the numbers behind the chart
{
"DAXX": -0.8864,
"MEN1": -1.2682,
"ATRX": -0.6285,
"DOT1L": -0.683
}index mean aff1 only — the numbers behind the chart
{
"DAXX": -0.8556,
"MEN1": -1.0299,
"ATRX": -0.6399,
"DOT1L": -0.5103
}index mean aff1 two donor — the numbers behind the chart
{
"DAXX": -0.8471,
"MEN1": -0.8882,
"ATRX": -0.6236,
"DOT1L": -0.4746
}men1 comparator by stratum
[ -0.435, -0.5939, -0.639 ]
daxx comparator by stratum
[ -0.2856, -0.3484, -0.432 ]
measurements — the numbers behind the chart
[
{
"quantity": "chronos_mean",
"subject": "MEN1, index fusion only, donor-collapsed",
"universe": "2 KMT2A::AFF1 donors in the DepMap lymphoid set",
"value": -0.8882
},
{
"quantity": "chronos_mean",
"subject": "DAXX, index fusion only, donor-collapsed",
"universe": "2 KMT2A::AFF1 donors in the DepMap lymphoid set",
"value": -0.8471
},
{
"quantity": "delta",
"subject": "shift in MEN1 index mean on dropping the ENL line",
"universe": "4-line index arm",
"value": 0.2383
},
{
"quantity": "delta",
"subject": "shift in DAXX index mean on dropping the ENL line",
"universe": "4-line index arm",
"value": 0.0308
},
{
"quantity": "fold",
"subject": "MEN1 dependency spread between the two AFF1 donors",
"universe": "SEM(+SEMK2) vs RS4;11",
"value": 2.84
},
{
"quantity": "fold",
"subject": "DAXX dependency spread between the two AFF1 donors",
"universe": "SEM(+SEMK2) vs RS4;11",
"value": 1.06
},
{
"quantity": "share",
"subject": "DepMap KMT2A-arm lines carrying the INDEX fusion (AFF1)",
"universe": "16 resolved KMT2A-rearranged DepMap lines",
"value": 0.3125
}
]corpus wide kmt2a arm — the numbers behind the chart
{
"KMT2A-AFF1": 5,
"KMT2A-MLLT3": 5,
"KMT2A-AFDN": 3,
"KMT2A-MLLT1": 2,
"KMT2A-FOXO4": 1,
"total": 16,
"index_fusion_share": 0.3125
}aff1 lines corpus wide
[ "SEM", "SEMK2", "RS4;11", "MV4-11", "ALL-PO" ]
- Recompute the MEN1 donor spread across all FIVE KMT2A::AFF1 DepMap lines - SEM, SEMK2, RS4;11, MV4-11, ALL-PO, which is 4 donors - not the 2 available in the lymphoid subset. The line identities are already resolved in analyses/depmap-scan-line-identities/results.json, so this needs only the Chronos values. If the 2.84-fold spread holds at n=4 donors it is the sharpest quantitative challenge to the menin-interface hypothesis in the corpus; if it collapses, it was two cell lines.
- Re-fetch per-line comparator values and recompute the permutation null for the AFF1-only and donor-collapsed arms. It is possible DAXX's result does not survive the restriction, and this analysis deliberately does not claim that it does.
- Is RS4;11's weak menin dependence explained by MEN1 copy number, MEN1 expression, HOXA status or differentiation state? All are retrievable and none was checked. A uniform interface predicts uniform dependence; anything that explains the difference rescues the hypothesis.
- Should this corpus's DepMap comparisons be routinely restricted to KMT2A::AFF1 rather than KMT2A-rearranged? Four of this project's DepMap scripts use the broader definition, and this is the first time the difference has been shown to matter.
- Fix the arm-membership defect recorded in analyses/depmap-scan-line-identities: membership is fixed from the first gene with data rather than derived per gene, so 5 of 16 eligible lines are excluded from every comparison in the full selective scan. Recorded 2026-09-21, still unfixed.
- Should the corpus's DepMap comparisons default to KMT2A::AFF1 (5 lines, 4 donors) rather than KMT2A-rearranged (16 lines, 31% index fusion)? The AFF1-only arm is small enough that the answer may be that neither is adequate and the honest output is an effect size with no p-value.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Of the four DepMap lines q172 treats as the index arm, three carry KMT2A-AFF1 (SEM CVCL_0095, RS4;11 CVCL_0093, SEMK2 CVCL_S906) and one carries KMT2A-MLLT1 / MLL-ENL (HB11;19 CVCL_8227). SEMK2 is recorded in the same source as derived_from SEM, and both are annotated age 5Y, so the three AFF1 lines represent two donors. RS4;11 is annotated age 32Y.
What this evidence cannot show
The fusion and derived_from annotations are DepMap/Cellosaurus metadata carried through q172's fetch, not independently re-fetched from Cellosaurus by this analysis. The 'Age unspecified' value for HB11;19 means age could not be used as a covariate for that line.
Evidence that argues against this (2)
- Including an ENL line is defensible and may even be deliberate: q172's beat was framed as KMT2A-rearranged, not KMT2A::AFF1, and for a chromatin-chaperone question the shared retained N-terminus is arguably the relevant biology rather than the 3' partner. The criticism only bites for claims about the INDEX FUSION specifically, which is what this corpus's menin hypothesis is about.
- Treating SEM and SEMK2 as one donor is conservative but not obviously correct for a dependency screen: SEMK2 is a distinct subline with its own passage history and possibly divergent genotype, and DepMap screens them separately. If they have genuinely diverged, collapsing them discards real information and n=2 understates the evidence.
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Recomputing the index mean three ways from q172's per-line values: 4-line, AFF1-only (3 lines), and AFF1 donor-collapsed (2 donors). DAXX: -0.8864, -0.8556, -0.8471. MEN1: -1.2682, -1.0299, -0.8882. ATRX: -0.6285, -0.6399, -0.6236. DOT1L: -0.6830, -0.5103, -0.4746. Dropping the ENL line moves MEN1 by +0.2383 and DAXX by +0.0308, a 7.7-fold difference; HB11;19's MEN1 score is -1.9831, the most negative in the arm. Donor-collapsed on the index fusion, MEN1 is -0.8882 and DAXX -0.8471, a gap of 0.0411.
What this evidence cannot show
These are index-arm means only. No permutation p-value is recomputed for the AFF1-only or donor-collapsed arms, because q172 cached comparator MEANS but not per-line comparator values, and a null would require re-fetching the comparator lines from the live API. The effect sizes stand; their significance under the restricted arm is unmeasured.
Evidence that argues against this (2)
- With n=2 donors, a difference of 0.0411 between MEN1 and DAXX is far below what this design can resolve, so 'indistinguishable' is a statement about power, not about the biology. Equally, the 7.7x sensitivity to one line is what any statistic on n=4 does when one member is extreme - it is a demonstration of fragility rather than evidence that the ENL line is wrong to include.
- Not recomputing the null is a real gap, not a formality: it is possible that DAXX AFF1-only fails significance where the 4-line version passed, which would change the conclusion of the parent analysis. This entry should not be cited as confirming q172's DAXX result under restriction - only as showing the index MEAN barely moves.
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Across q172's three comparator strata the MEN1 index-arm mean is constant at -1.2682 while the MEN1 comparator mean moves -0.4350 (18 lymphoid) -> -0.5939 (9 B-ALL) -> -0.6390 (6 fusion-driven B-ALL); the corresponding DAXX comparator means are -0.2856 -> -0.3484 -> -0.4320. In the tightest stratum MEN1's difference is -0.6292 against DAXX's -0.4545, while MEN1's p rises to 0.1032 and DAXX's is 0.0141. MEN1's index-arm sd is 0.594 (range 1.520) against DAXX's 0.078 (range 0.181).
What this evidence cannot show
The comparator strata are nested subsets chosen by q172, not independent samples, so the three p-values are not independent tests. The n for the tightest stratum is 6 comparators against 4 index lines.
Evidence that argues against this (2)
- That the comparator strengthens does not fully exonerate menin: if fusion-driven B-ALL generally needs menin as much as KMT2A::AFF1 does, then menin dependence carries little information ABOUT the fusion, and a therapeutic rationale built on fusion-specific menin addiction is weakened even though the fusion genuinely needs menin. 'Required but not selective' is a real problem for a targeting argument, just not for a mechanistic one.
- Attributing the p-value change to variance rather than effect relies on comparing an sd computed on 4 points. An sd from n=4 is itself extremely unstable, so 'MEN1 fails on variance' is a plausible reading of the arithmetic rather than a demonstrated cause.
- E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Within the index fusion, the two donors' MEN1 Chronos scores are SEM -1.3132 (averaged over SEM and SEMK2) and RS4;11 -0.4632, a 2.84-fold difference in magnitude. The same two donors' DAXX scores are -0.8727 and -0.8214, a 1.06-fold difference. ATRX gives 1.17-fold and DOT1L 1.58-fold.
What this evidence cannot show
n=2 donors. RS4;11 is from a 32-year-old and SEM from a 5-year-old, so age, lineage maturity and any number of other differences are fully confounded with donor. MEN1 copy number and expression in the two lines were not retrieved. Chronos is normalised per screen and single-line values carry technical variance that was not bounded here.
Evidence that argues against this (2)
- A 2.84-fold spread on n=2 is not a finding, it is an observation awaiting a third line. DepMap contains further KMT2A::AFF1 lines outside this lymphoid universe (the parent analysis counted 11 index lines across lineages, of which only 4 are lymphoid), and the spread should be recomputed across all of them before any weight is put on it.
- Heterogeneous menin dependence is fully compatible with the menin-interface hypothesis if the lines differ in something downstream - HOXA status, differentiation state, cooperating lesions. The interface can be uniform while the consequence of blocking it is not. This observation therefore constrains a STRONG version of the hypothesis ('the interface is the whole story') and leaves a weaker version untouched.
reproduces its figures exactly, then shows the index arm is 3 AFF1 lines from 2 donors plus an ENL line
DAXX's index mean barely moves under restriction to the actual index fusion (0.0308), unlike MEN1's
menin remains the most REQUIRED of the four genes; it loses significance because the comparator strengthens
answers q176's first open question: the 2.835x spread survives but RS4;11 is last on KMT2A too, so the challenge is largely inherited from fusion dependence
q176 said menin fails on variance not effect; this measures that variance against 358 genes and finds it near the top of the distribution