The menin/Polycomb split between NLRC5 and CIITA does not extend to their pathways, and the MHC locus is structurally invisible to the method that found it
also called NLRC5 CIITA split · antigen presentation enrichment · MHC alt haplotype blind spot · class I class II menin
Two genes switching on the two halves of the immune-visibility system sit in opposite categories in a published dataset, which is a tidy story that does not hold for the rest of the machinery - the other genes are distributed exactly as chance would put them. The genes that matter most cannot be examined by this method at all, because that part of the genome exists in many alternative versions.
The same thing, for a scientist
The NLRC5 (Menin-solo) versus CIITA (H2AK119ub-solo) split in Zhou's locus set is real at the level of those two genes and was found independently by two beats and one desk analysis this cycle. **It is not a pathway property.** Testing the GO class I (GO:0002474) and class II (GO:0002495) sets against Zhou's own called set with a 20,000-permutation label shuffle: | set | n tested | Menin solo | shared | H2AK119ub solo | |---|---|---|---|---| | **class I** | 16 | 5 (exp 5.9, p=0.86) | 3 (exp 3.0, p=1.00) | 8 (exp 7.1, p=0.84) | | **class II** | 9 | 3 (exp 3.3, p=1.00) | 1 (exp 1.7, p=0.94) | 5 (exp 4.0, p=0.73) | **Observed equals expected everywhere.** No skew in either direction. **And the negative is bounded far more tightly than it looks**, because **27 of 83 annotated symbols were dropped for mapping to multiple Ensembl gene ids** - and they are not a random 27.
What was dropped, and why it is the whole point
The genes that could not be joined one-to-one:
B2M,HLA-A,HLA-B,HLA-C,HLA-DMA,HLA-DMB,HLA-DOA,HLA-DOB,HLA-DPA1,HLA-DPB1,HLA-DQA1,HLA-DQA2,HLA-DQB1,HLA-DQB2,HLA-DRA,HLA-DRB1,HLA-DRB3,HLA-DRB4,HLA-E,HLA-F,HLA-G,HLA-H,MARCHF8,TAP1,TAP2,TAPBP
That is every classical heavy chain, every class II alpha and beta chain, B2M, and TAP1, TAP2 and TAPBP. In other words, the presenting molecules themselves and the peptide-loading transporters - the objects the pathway is named after.
The reason is structural, not accidental. The MHC on chromosome 6p21 is the most alternative-haplotype-dense region of the human genome; Ensembl represents those haplotypes as additional gene ids, so a symbol there resolves to several ids and any instrument requiring a 1:1 symbol-to-id join must either guess or refuse. This one refuses, which is correct and is also why it cannot see.
So the clean null above is a null about the accessory machinery - cathepsins, ERAP1/2, PDIA3, AP3B1, PIKFYVE, the ULBPs - and says nothing whatever about HLA itself. A reader who takes 'no enrichment in antigen presentation' from this entry has taken the opposite of what it measured.
What survives about NLRC5 and CIITA
The two-gene observation stands and the generalisation does not. NLRC5 is in the Menin-solo group and CIITA is in the H2AK119ub-solo group; both resolve 1:1, both are outside the MHC proper, and the assignment is read directly from the published table.
What it licenses: a specific, cheap, unrun experiment - measure NLRC5 transcript under a menin inhibitor. If the class I transactivator is menin-occupied, that is a candidate direct route to class I which no published model includes, and one qPCR settles whether occupancy translates into regulation.
What it does not license: the sentence 'the class I arm is menin-controlled and the class II arm is Polycomb-controlled'. Two transactivators are two genes. The pathways behind them show nothing, and the molecules in front of them were not measurable.
A further caution on direction. Occupancy is not dependence. Sparbier reports that VTP50469 globally reduced menin occupancy while producing comparatively few expression changes, so a menin-bound promoter is not a menin-regulated gene until someone measures the transcript.
CORRECTED the same day: the MHC was joinable, and the rescued test is not flat
The claim that a symbol-to-Ensembl join cannot reach the MHC is wrong. MyGene returns all ids including alt-haplotype scaffolds, so an exact-match resolver refuses - but HGNC assigns one primary ensembl_gene_id per symbol, which is what Zhou's table uses. Resolving that way recovers B2M, HLA-A, HLA-B, HLA-C, HLA-E, TAP1, TAP2 and TAPBP.
And the null above does not survive the rescue. With 24 class I genes instead of 16, the bivalent group holds 9 against 4.5 expected, nominal p = 0.047 - though six tests were run (Bonferroni p = 0.284) and five of the nine members lie within 6p21-6p22, so the honest claim is domain-level and moderate. Class II remains flat.
What survives from this entry unchanged: the NLRC5/CIITA split, the warning that two transactivators are not a pathway, and the general lesson about counting what a join dropped. See ana-the-mhc-was-not-invisible-and-the-rescued-test-is-not-flat.
| n dropped | 26 |
|---|---|
| null | 20,000-permutation label shuffle within the called set, two-sided |
class I
{
"go": "GO:0002474",
"n_tested": 16,
"observed": {
"H2AK119ub solo": 8,
"Menin solo": 5,
"Menin/H2AK119ub shared": 3
},
"p": {
"Menin solo": 0.8578,
"Menin/H2AK119ub shared": 1,
"H2AK119ub solo": 0.8368
}
}class II
{
"go": "GO:0002495",
"n_tested": 9,
"observed": {
"Menin solo": 3,
"H2AK119ub solo": 5,
"Menin/H2AK119ub shared": 1
},
"p": {
"Menin solo": 1,
"Menin/H2AK119ub shared": 0.9394,
"H2AK119ub solo": 0.7301
}
}multi ensg dropped
[ "B2M", "HLA-A", "HLA-B", "HLA-C", "HLA-DMA", "HLA-DMB", "HLA-DOA", "HLA-DOB", "HLA-DPA1", "HLA-DPB1", "HLA-DQA1", "HLA-DQA2", "HLA-DQB1", "HLA-DQB2", "HLA-DRA", "HLA-DRB1", "HLA-DRB3", "HLA-DRB4", "HLA-E", "HLA-F", "HLA-G", "HLA-H", "MARCHF8", "TAP1", "TAP2", "TAPBP" ]
- Do not cite this entry as evidence that antigen presentation is unaffected by menin status. It measured the accessory machinery and could not see a single HLA chain, B2M, or the TAP transporters. The null is about what remained after the interesting genes were dropped.
- And do not take the NLRC5/CIITA split to the clinic or to a trial rationale. It is a promoter-occupancy assignment in one adult myeloid cell line, with no transcript measurement behind it, in a table with no condition column. The experiment it suggests is cheap; the claim it suggests is not yet supported.
- And the entry was wrong about its own central lesson. It said a 1:1 join CANNOT see the MHC; the truth is that one resolver cannot and another can. 'The instrument is blind here' is itself a claim that needs checking before it becomes a rule.
- What is NLRC5 transcript doing under a menin inhibitor? One qPCR. It is the only cheap test that would turn the occupancy observation into a regulatory claim.
- Can the MHC be joined at all? An allele-aware or coordinate-interval join over chr6:29-33 Mb would reach the classical chains that a symbol-to-Ensembl join cannot.
- How much else in this graph's gene layer is silently unjoinable for the same reason? The MHC is the worst case but it is not the only multi-haplotype region.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
In OCI-AML2, an adult KMT2A::AFDN AML line, the Gene Ontology MHC class I (GO:0002474) and class II (GO:0002495) gene sets show no enrichment in any of the three menin/H2AK119ub groups of Zhou's called locus set: 16 class I genes distribute 5/3/8 against expectations of 5.9/3.0/7.1 and 9 class II genes distribute 3/1/5 against 3.3/1.7/4.0, with all six two-sided permutation p-values between 0.73 and 1.00 over 20,000 label shuffles.
antigen-presentation-menin-polycomb: GO class I and class II gene sets from the EBI QuickGO annotation API tested against Zhou supplemental Table 4 by permutation · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Run 2026-09-20. **The universe is Zhou's CALLED set**, so every statement is conditional on a gene being called at all; genes absent from the table were not tested. **27 of 83 annotated symbols were dropped** for multi-Ensembl mapping, and they comprise every classical HLA chain, B2M, TAP1, TAP2 and TAPBP - so this tests accessory machinery and NOT the presenting molecules. Sets are small (16 and 9), so power against a modest effect is poor and this is an absence of evidence at that size rather than evidence of absence. GO annotation is itself an instrument and most rows carry IEA electronic evidence. **Strong rather than established** for that reason. System is adult myeloid KMT2A::AFDN, not infant B-lineage KMT2A::AFF1 (rule 73).
Evidence that argues against this (2)
- A null on OCCUPANCY does not contradict the published FUNCTIONAL results. Menin inhibition demonstrably raises cell-surface MHC class I dose-dependently and enhances T-cell killing, and the same paper reports that reduced menin occupancy genome-wide produced comparatively few expression changes. So occupancy and regulation are loosely coupled here in both directions, and an occupancy-based test is simply the wrong instrument for deciding whether a pathway is drug-responsive. Sparbier CE et al. 2023
- SUPERSEDED IN PART, same day. The null reported here was measured after dropping every classical HLA chain, B2M and the TAP transporters for multi-mapping. Those genes ARE joinable through the HGNC primary ensembl_gene_id, and with them included the class I set is no longer flat - 9 bivalent against 4.5 expected, nominal p = 0.047, not surviving correction. The class II null stands. Zhou X et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The master transactivators of the two arms fall in opposite groups of the same published table: NLRC5, the class I transactivator, is menin-bound and H2AK119ub-negative, while CIITA, the class II transactivator, is H2AK119ub-marked and menin-unbound.
Epigenetic regulation of noncanonical menin targets modulates menin inhibitor response in acute myeloid leukemia · Zhou X, Zhang L, Aryal S, et al. · Blood 2024What this evidence cannot show
Read directly from supplemental Table 4 and independently confirmed by a separate beat. **Established for the group assignment in that dataset and nothing more.** Both genes resolve 1:1 and lie outside the MHC proper, so neither is affected by the alt-haplotype problem. The table carries no condition column, the coordinates are 2 bp TSS anchors, and **occupancy is not dependence** - no transcript measurement under a menin inhibitor exists for NLRC5.
Evidence that argues against this (1)
- The published mechanism runs the other way for class II: menin inhibition derepresses endogenous retroviruses, which raises interferon-stimulated genes, which raises CIITA and MHC class II. That is an INDIRECT route and is consistent with CIITA being menin-unbound. If the analogous class I route through NLRC5 is direct, the two arms differ in kind as well as in target - but no experiment has tested the NLRC5 limb, so the asymmetry is a hypothesis carried by an occupancy table, not a result. Fetsch V et al. 2026
THE MENIN-BOUND HALF OF THE SPLIT, and the entry that should carry the qPCR experiment: occupancy without any transcript measurement.
THE POLYCOMB-MARKED HALF: menin-unbound, consistent with the published indirect route.
EXTENDS THAT JOIN AND BOUNDS IT: the same table, now with a null, and with the reason the MHC layer never appeared in it.
CONSTRAINS THE MECHANISM WITHOUT DECIDING IT: no occupancy skew across the class I set, so the route is not a broad menin footprint over the pathway.
CONSISTENT: CIITA being menin-unbound is what an indirect HERV/interferon route predicts.
DROPPED BY THE METHOD: B2M maps to several Ensembl ids and is invisible to a 1:1 join, in both GO sets.
DROPPED BY THE METHOD: every classical heavy chain is multi-mapped in the alt-haplotype-dense MHC and cannot be joined one-to-one.
RULE 73 AT WRITING TIME: the system - adult KMT2A::AFDN myeloid OCI-AML2 - is in the first clause of both claims.
scored this entry at 0.000, the lowest of 141 pairs, and adjudicated it as correctly attached support in supplemental Table 4
this entry's NLRC5/CIITA assignments come from the supplemental table that cannot be re-fetched
CORRECTS ITS METHOD AND ITS RESULT: the MHC is joinable through HGNC primary ids, and the class I null does not survive the rescue.
CONFIRMS ITS CENTRAL CAUTION from the other direction: two transactivators are not a pathway, and the bivalent group is the least functionally coherent of the three.