PROM1/CD133 tracks the HOXA arm and ANTI-tracks the IRX1 arm in 48 KMT2A::AFF1 infant ALL cases - so the one surface target with a published human kill is depleted in the iALL-IRX1 subgroup
also called q149 PROM1 vs IRX1 · CD133 is a HOXA-arm marker
CD133 is the most promising surface handle for treating this leukaemia. In 48 babies with the t(4;11) fusion, CD133 is switched on almost every time HOXA is on - and it is much less often on in the babies whose leukaemia runs on IRX1 instead. A CD133-directed therapy would under-treat that second group.
The same thing, for a scientist
**PROM1 is not a surrogate for IRX1 status. It is a surrogate for HOXA status, and therefore an anti-surrogate for IRX1 status.** Measured on the 48 KMT2A::AFF1 infant ALL cases of GSE68720, all three genes scored by the identical per-sample floor rule [E1]: | within KMT2A::AFF1, n=48 | PROM1 ON | PROM1 OFF | Fisher two-sided | |---|---:|---:|---| | **HOXA10 ON** | **24** | **1** | **0.00021** | | HOXA10 OFF | 11 | 12 | | | **IRX1 ON** | **10** | **11** | **0.00079** | | IRX1 OFF | 25 | 2 | | **PROM1 is ON in 24 of 25 HOXA-positive cases and in 11 of 23 HOXA-negative ones.** Because HOXA and IRX1 are mutually exclusive in this group (`ana-q149-the-hoxa-irx1-mutual-exclusivity-is-aff1-specific-and-absent-in-mllt1`), PROM1 inherits the opposite relation to IRX1: **PROM1-ON cases are IRX1-negative 25 of 35 times, and PROM1-OFF cases are IRX1-positive 11 of 13 times.** **And in KMT2A::MLLT1 the relation is not testable and looks trivial:** 13 of 16 HOXA-positive, **PROM1 ON in 16 of 16**. PROM1 is simply on in every ENL case. So PROM1 carries no subgroup information outside AFF1. **Why this matters more than a correlation.** `gene-prom1` records PROM1/CD133 as *the one surface target in this disease with a published human-relevant kill*, activated by direct MLL-AF4 binding through an intragenic H3K79me2/3 enhancer and repressed by PRC2 when off. **This analysis says that activation is concentrated in the HOXA arm.** The 19 of 48 AFF1 infants in the IRX1 arm are the ones a CD133-directed agent is least likely to reach - and they are also the arm whose biology this corpus understands least. **It also retires PROM1 as a proxy.** `exp-q147-swap-the-aff1-c-terminus-onto-kmt2a-af9-and-score-irx1-occupancy` and the q147 beat validated a chromatin proxy against PROM1. At the transcript level PROM1 and IRX1 move in **opposite** directions inside the index fusion, so a PROM1-validated proxy is not validated for IRX1. **The cell lines agree and make it vivid.** NALM-19 carries PROM1 at **160.7 RPKM** and HOXA9 at exactly **0.00** and IRX1 at **0.0000**; SEM carries PROM1 **105.9** with IRX1 **0.0000** [E2]. So a PROM1-high KMT2A::AFF1 model can be both HOXA-negative and IRX1-negative - the third state, which the infant ALL 2x2 puts at 4 of 48.
The direction, stated carefully, because it is easy to invert
PROM1 is not the HOXA arm's marker in the sense of being exclusive to it. It is ON in 24 of 25 HOXA-positive cases and in 11 of 23 HOXA-negative cases. So:
- PROM1-OFF is highly informative: 12 of 13 PROM1-negative cases are HOXA-negative, and 11 of 13 are IRX1-positive. A CD133-negative KMT2A::AFF1 blast is very likely an iALL-IRX1 blast.
- PROM1-ON is weakly informative: 24 of 35 are HOXA-positive, 10 of 35 are IRX1-positive. CD133 positivity does not tell you which arm you are in.
That asymmetry is the practically useful part and it is the opposite of how a positive surface marker is normally used. The negative call carries the information.
In the KMT2A-germline infant ALL controls the relation is different again: HOXA+PROM1+ 4, HOXA+PROM1- 3, HOXA-PROM1+ 0, HOXA-PROM1- 10, Fisher two-sided p = 0.015. PROM1-ON requires HOXA-ON there (0 of 10 HOXA-negative germline cases are PROM1-positive), which is a tighter dependence than in AFF1. So the 11 HOXA-negative PROM1-positive AFF1 cases are a state that does not occur in the germline controls at all - a fusion-associated state, and nobody has named it.
What could make this wrong
- HOXA10 is standing in for HOXA9. GPL570 has no single-gene HOXA9 probe. If PROM1 tracks HOXA9 and not HOXA10, the 24-of-25 becomes uncertain. The two are adjacent and co-regulated in this disease, which is why the substitution is defensible, but it is a substitution.
- A shared per-sample floor induces positive dependence between any two genes. A sample with a low floor calls more genes ON. That mechanism would push the HOXA/PROM1 and IRX1/PROM1 tables in the SAME direction, and they go in opposite directions - which is the argument that the floor is not driving it.
- PROM1 has exactly one probe on GPL570 -
204304_s_at, annotatedprominin 1, fetched from the platform annotation file. No second measurement, so a probe-level artefact cannot be detected. - n = 13 in the PROM1-OFF cell. The clinically interesting claim - that a CD133-negative AFF1 blast is probably iALL-IRX1 - rests on 11 of 13. Wilson 95% on 11/13 is roughly 57-96%.
| cohort | GSE68720 infant ALL, GPL570; all three genes scored by max-of-ten-silent-genes per sample |
|---|---|
| hoxa gene used | HOXA10 - GPL570 has no single-gene HOXA9 probe |
| absence class | measured-not-analysed |
aff1 hoxa prom1 2x2 — the numbers behind the chart
{
"HOXA_on_PROM1_on": 24,
"HOXA_on_PROM1_off": 1,
"HOXA_off_PROM1_on": 11,
"HOXA_off_PROM1_off": 12,
"n": 48,
"fisher_two_sided": 0.00021
}aff1 irx1 prom1 2x2 — the numbers behind the chart
{
"IRX1_on_PROM1_on": 10,
"IRX1_on_PROM1_off": 11,
"IRX1_off_PROM1_on": 25,
"IRX1_off_PROM1_off": 2,
"n": 48,
"fisher_two_sided": 0.00079
}mllt1 hoxa prom1 2x2
{
"HOXA_on_PROM1_on": 13,
"HOXA_on_PROM1_off": 0,
"HOXA_off_PROM1_on": 3,
"HOXA_off_PROM1_off": 0,
"n": 16,
"fisher_two_sided": 1,
"note": "PROM1 is ON in 16 of 16 - no information"
}germline hoxa prom1 2x2
{
"HOXA_on_PROM1_on": 4,
"HOXA_on_PROM1_off": 3,
"HOXA_off_PROM1_on": 0,
"HOXA_off_PROM1_off": 10,
"n": 17,
"fisher_two_sided": 0.01471,
"note": "0 of 10 HOXA-negative germline cases are PROM1-positive; the 11 HOXA-negative PROM1-positive AFF1 cases are a state absent from the controls"
}aml irx1 high prom1 range
{
"median": 6.37,
"min": -4.17,
"max": 9.79,
"n": 20,
"note": "in paediatric AML the IRX1-high non-AFF1 samples span the whole PROM1 range - no relation there"
}measurements — the numbers behind the chart
[
{
"quantity": "fisher_exact_p_two_sided",
"subject": "HOXA10-ON x PROM1-ON within KMT2A::AFF1 infant ALL",
"value": 0.00021,
"units": null,
"n": 48,
"universe": "GSE68720 GPL570, all genes scored against the same per-sample max-of-10-silent-genes floor",
"null": 1,
"note": "24 of 25 HOXA-positive cases are PROM1-positive; the null cited is the same 2x2 in the 16 KMT2A::MLLT1 cases where PROM1 is ON in 16 of 16"
},
{
"quantity": "fisher_exact_p_two_sided",
"subject": "IRX1-ON x PROM1-ON within KMT2A::AFF1 infant ALL, ANTI-association",
"value": 0.00079,
"units": null,
"n": 48,
"universe": "GSE68720 GPL570, same floor",
"null": 0.54118,
"note": "11 of 13 PROM1-negative cases are IRX1-positive; the null cited is the same 2x2 in the 17 KMT2A-germline controls"
},
{
"quantity": "proportion_IRX1_ON_among_PROM1_OFF",
"subject": "KMT2A::AFF1 infant ALL, PROM1-negative cases",
"value": 0.846,
"units": null,
"n": 13,
"universe": "GSE68720; 11 of 13. Wilson 95% roughly 0.57-0.96",
"null": 0.4375,
"note": "the null is the AFF1 group's overall IRX1-ON rate of 21 of 48. A CD133-negative AFF1 blast is roughly twice as likely to be iALL-IRX1 as an unselected one"
}
]- A CD133-directed CAR or antibody in KMT2A::AFF1 infant ALL would be aimed at a marker concentrated in the HOXA arm. 11 of the 13 PROM1-negative infants in this cohort are IRX1-positive - so the subgroup most likely to escape a CD133 therapy is the subgroup whose biology is least understood, and relapse after CD133 therapy would be expected to be IRX1-enriched.
- This also predicts a specific resistance route: selection for the IRX1 arm under CD133 pressure. Because the two arms are mutually exclusive rather than continuous, that selection would look like a lineage switch rather than a gradual loss of antigen.
- The mirror risk: CD133 is on normal haematopoietic stem and progenitor cells and on several epithelia, and PROM1's OFF state is PRC2-dependent per gene-prom1 - so a PRC2 or DOT1L combination intended to sensitise could move CD133 on normal cells too.
- Does surface CD133 protein, measured by flow with a stated clone, stratify the same way as PROM1 transcript? The AC133 epitope is glycosylation-dependent and this is a transcript result being read as a surface-marker result.
- What are the 11 HOXA-negative PROM1-positive KMT2A::AFF1 cases? That state does not occur in any of the 17 KMT2A-germline controls, so it is fusion-associated, and it is not the iALL-IRX1 state either - only 10 of those 11 overlap with IRX1-positive.
- Is NALM-19 really HOXA9-zero? A KMT2A::AFF1 line with no HOXA9 and PROM1 at 160.7 RPKM would be the only HOXA-negative AFF1 model in existence and nothing in this corpus mentions it.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In 48 KMT2A::AFF1 infant ALL cases, PROM1 transcript calls co-occur with HOXA10 calls (24 of 25 HOXA-positive cases are PROM1-positive, Fisher exact two-sided p = 0.00021) and anti-co-occur with IRX1 calls (25 of 35 PROM1-positive cases are IRX1-negative and 11 of 13 PROM1-negative cases are IRX1-positive, Fisher exact two-sided p = 0.00079), all three genes scored against the same per-sample silent-tissue floor.
Gene expression profiles predictive of outcome and age in infant acute lymphoblastic leukemia: a Children's Oncology Group study · Kang H, Wilson CS, Harvey RC, et al. · Blood 119:1872-1881 2012What this evidence cannot show
My analysis of their deposited data, not their result. HOXA10 substitutes for HOXA9 because GPL570 carries no single-gene HOXA9 probe. Transcript, not surface protein - CD133 is a glycosylated pentaspan membrane protein and the AC133 epitope is glycosylation-dependent, so a transcript call is two steps from what a CD133-directed agent sees. Thresholded calls with a floor this beat has shown to be threshold-sensitive. The practically interesting cell (PROM1-OFF, 13 cases) is small.
Evidence that argues against this (1)
- The AC133 epitope is glycosylation-dependent and surface CD133 protein dissociates from PROM1 transcript in several systems, so a transcript-level association with the HOXA arm could vanish at the protein level - which is the level a CAR or antibody sees. No surface CD133 measurement stratified by IRX1 status exists anywhere, so the clinical inference in this entry is a transcript inference wearing a surface-marker claim and should be read as a hypothesis for flow cytometry. Kang H et al. 2012
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In CCLE RNA-seq, the KMT2A::AFF1 lines NALM-19 and SEM carry PROM1 at 160.7 and 105.9 RPKM with IRX1 at exactly 0.0000, and NALM-19 additionally carries HOXA9 at exactly 0.00 - so a PROM1-high KMT2A::AFF1 model can be negative for both HOXA and IRX1.
Next-generation characterization of the Cancer Cell Line Encyclopedia · Ghandi M, Huang FW, Jane-Valbuena J, et al. · Nature 569:503-508 2019What this evidence cannot show
Two cell lines, RPKM, 2019-era. Cell lines are not patients and this corpus has just recorded that no KMT2A-r CCLE line expresses IRX1 at all, so the IRX1 zeros here carry no information about the PROM1 relation - only the PROM1-high-with-HOXA9-zero observation in NALM-19 does. NALM-19's HOXA9 at exactly 0.00 in a KMT2A::AFF1 line is itself surprising enough to deserve its own check.
Evidence that argues against this (1)
- NALM-19's HOXA9 of exactly 0.00 may be a mapping or annotation artefact rather than a biological zero - HOXA9 is a short, GC-rich, cluster-embedded gene and exact zeros in RPKM tables are worth suspecting. If NALM-19 is in fact HOXA9-positive, the 'neither arm' state this evidence item describes disappears from the cell-line panel. Ghandi M et al. 2019
the safety half of that finding: the agent it implies would act on kidney at 5.27x marrow
PROM1 is ON in 24 of 25 HOXA-positive and 10 of 21 IRX1-positive KMT2A::AFF1 infant ALL cases
PROM1 inherits its anti-relation to IRX1 from the HOXA/IRX1 exclusivity
the iALL-IRX1 arm is PROM1-depleted: 11 of 13 PROM1-negative AFF1 cases are IRX1-positive
- related to →iALL-HOXA9 — the HOXA-high state of infant KMT2A::AFF1 ALL, and the one everything is modelled on
the iALL-HOXA9 arm is PROM1-enriched: 24 of 25 are PROM1-positive
the cohort this 2x2 is computed on
PROM1 105.9 RPKM with IRX1 at 0.0000