PROM1 (CD133) occupancy tracks HOXA occupancy at Spearman +0.952 — a second, independent marker of the same split
A completely separate marker that doctors have proposed as a CAR-T target lines up almost perfectly with the HOXA half of this leukaemia. That makes it much less likely the split we measured is an accident.
The same thing, for a scientist
🟠 **SCOPED 2026-09-24 (cycle 131): this co-occupancy is specific to KMT2A::AFF1 and does not generalise across fusion partners.** Measured across the other five partners in the same deposit: PROM1 is **essentially unoccupied in KMT2A-AF9 (0.069-0.430x) and KMT2A-AFDN (0.090-0.308x) while HOXA in those same samples is 5.7-29.1x** - the extreme case is `TB040576`, with HOXA at **29.1x** and PROM1 at **0.069x**. In KMT2A-ENL, the only other group with enough samples to carry a correlation, PROM1-vs-HOXA is **+0.086 (n = 6, exact two-sided p = 0.919)**, i.e. flat. **So PROM1 is not a general HOXA co-target.** The +0.952 below stands for KMT2A::AFF1 and must not be quoted as a property of KMT2A-rearranged leukaemia. Note the partner groups are confounded with lineage (the AF9 and AFDN samples are all AML). See `ana-q147-prom1-occupancy-does-not-track-hoxa-outside-kmt2a-aff1-and-is-absent-in-af9-and-afdn`. --- Symeonidou reports **PROM1 (CD133) upregulated in iALL-HOXA9 and lower in iALL-IRX1**, and proposes it as a subgroup-discriminating therapeutic target [E1]. Measured as oncoprotein occupancy across the seven KMT2A::AFF1 patients, PROM1 tracks HOXA9/HOXA10 at **Spearman +0.952** — numerically identical to HOXA9's own self-correlation, and the **99.90th percentile** of all 19,404 promoters [E2]. The pre-declared flat control **CSPG4 (NG2)**, which Symeonidou reports as similar in both subgroups, sits at or below **0.09x** mean promoter in every patient [E2]. Two independently published markers of the same subgroup therefore agree in chromatin, which is the strongest available evidence that the occupancy partition measured here corresponds to the published expression partition.
The numbers
Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean promoter coverage, so 1.0 = an average promoter in that library. GSE252378 has no spike-in and no drug arm, so no between-sample absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
| patient | HOXA9/HOXA10 20 kb | PROM1 8 kb | PROM1 promoter N1 | CSPG4 N1 |
|---|---|---|---|---|
| p179 | 11.78 | 7.35 | 13.24 | 0.00 |
| p247 | 10.65 | 7.72 | 12.19 | 0.09 |
| TB051939 | 9.18 | 3.26 | 6.05 | 0.00 |
| p279 | 2.92 | 3.41 | 5.53 | 0.77 |
| TB920488 | 0.97 | 3.07 | 1.59 | 0.14 |
| p454 | 0.33 | 1.15 | 2.65 | 0.29 |
| p318 | 0.23 | 1.48 | 2.57 | 0.38 |
| TB986292 | 0.11 | 1.02 | 2.47 | 0.45 |
It is a gradient, not a switch — and that is the honest reading
PROM1 is fusion-occupied in all seven patients (1.02-7.72x at 8 kb); it is merely higher in the HOXA-high ones. So PROM1 does not partition the cohort; it co-varies with HOXA. The pre-registered outcome C1 ("PROM1 high exactly in the HOXA-high patients") therefore fires in its correlational form and not in its categorical form, and the categorical wording is withdrawn.
A limitation of my own instrument, found by the pre-declared floor check
CSPG4's own Spearman against HOXA is -0.619, at the 2.95th percentile — more anti-correlated than IRX1's -0.571. Taken at face value that would make my supposed flat control the strongest hit in the beat. It is not: every CSPG4 value is at or below 0.09x, so the rank statistic is being computed over eight essentially-zero numbers and their ordering is noise.
The lesson generalises and is recorded here because it nearly bit: a rank correlation is scale-free, which means it cannot tell you whether anything was there at all. Any locus reported on a Spearman must also clear an absolute floor. IRX1 does (up to 6.35x); CSPG4 does not.
| spearman prom1 vs hoxa promoter | 0.952 |
|---|---|
| spearman prom1 vs hoxa 8kb | 0.929 |
| percentile | 99.9 |
| spearman hoxa9 self | 0.952 |
| cspg4 max share | 0.77 |
| cspg4 spearman | -0.619 |
| scope | KMT2A::AFF1 only; does not generalise to AF9, AFDN, ENL, AF10 or EPS15 |
| rho prom1 vs hoxa enl n6 | 0.0857 |
prom1 range 8kb
[ 1.019, 7.718 ]
prom1 range af9
[ 0.069, 0.43 ]
prom1 range afdn
[ 0.09, 0.308 ]
- If CD133-directed CAR-T is developed against the HOXA subgroup, the IRX1 subgroup - the one with the worse prognosis - has the lower target density and is the one that relapses.
- PROM1 is expressed on normal haematopoietic and epithelial stem cells, so on-target off-tumour toxicity is the governing constraint regardless of subgroup.
- Does PROM1 surface protein, as opposed to promoter occupancy, track the split? A CAR-T decision depends on surface CD133, which is two measurements away from what was scored here.
- Is PROM1 a direct oncoprotein target or is its occupancy a consequence of the HOXA9/MEIS1 program it co-varies with?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
PROM1 (CD133) is specifically upregulated in the iALL-HOXA9 subgroup with lower expression in iALL-IRX1, while CSPG4 (NG2) is expressed at similar levels in both.
HOXA9/IRX1 expression pattern defines two subgroups of infant MLL-AF4-driven acute lymphoblastic leukemia · Symeonidou V, Ottersbach K · Exp Hematol 2021What this evidence cannot show
Transcript abundance in two reanalysed RNA-seq cohorts. The paper renders the gene as 'GSPG4' at one point, evidently a typographical error for CSPG4/NG2; I use CSPG4 and flag the discrepancy rather than silently correcting the source.
Evidence that argues against this (1)
- If PROM1's subgroup association does not replicate in a prospective cohort, the CD133 CAR-T concern built on it - and this entry's use of PROM1 as an independent marker - both fall. Symeonidou V et al. 2021
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Across the seven patients, PROM1 oncoprotein occupancy tracks HOXA9/HOXA10 occupancy at Spearman +0.952 (99.90th percentile of 19,404 promoters), identical to HOXA9's own value, while CSPG4 never exceeds 0.09x mean promoter.
q145 re-analysis of GSE252378: IRX1 and HOXA oncoprotein occupancy across seven KMT2A::AFF1 patients (analyses/q145-irx1-in-hoxa-negative-infants) · Iqbal Lab oncofusion knowledge graph, beat q145 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
n = 7 patients / 8 sample records. PROM1 is occupied in every patient, so this is a correlation of degree and not a partition. The +0.952 being numerically identical to HOXA9's own value reflects the coarse resolution of a Spearman over 8 points, where many distinct orderings collapse to the same value - it is not evidence of an unusually tight relationship. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean promoter coverage, so 1.0 = an average promoter in that library. **GSE252378 has no spike-in and no drug arm**, so no between-sample absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
Evidence that argues against this (1)
- One deposit, one lab, one assay, one genome build; seven patients. A second cohort profiled by a different group could reverse any of this. Janssens DH et al. 2024
PROM1 occupancy tracks HOXA at Spearman +0.952 across seven patients
- supports →iALL-HOXA9 — the HOXA-high state of infant KMT2A::AFF1 ALL, and the one everything is modelled on
a second, independently published marker of the same subgroup
validates that the chromatin partition corresponds to the published expression partition