RGPD8 is the top-ranked MLL-AF4-specific prey on 3 spectra, and 96.5% of its tryptic peptides are shared with seven near-identical paralogues — only 4 peptides in the whole protein could ever have discriminated it — declined
also called MLL-AF4-RGPD8 · KMT2A-RGPD8 · RANBP2L3 · RanBP2-like and GRIP domain-containing protein 8
RGPD8 is the strongest of the proteins said to bind only the MLL-AF4 fusion. It belongs to a family of eight almost-identical human proteins created by a duplicated stretch of chromosome 2, and almost every fragment a mass spectrometer would see from it is also a fragment of one of its siblings. With only three detections, we cannot say it was RGPD8 at all.
The same thing, for a scientist
**Verdict: DECLINED.** RGPD8 (O14715, 1765 aa) is the lowest-p MLL-AF4-specific prey in Skucha 2018 at p = 0.0147 — and the evidence is **2|1 = 3 spectral counts** [E1]. RGPD1-8 are a chr2 segmental-duplication family derived from RANBP2/Nup358, spanning 1748-1765 aa, and UniProt gives **RGPD5 and RGPD6 a single shared entry (Q99666) because they cannot be distinguished** [E2]. An in-silico fully-tryptic digest of O14715 against RANBP2 and the five other reviewed RGPD entries finds **110 of 114 distinct peptides (96.5%) present verbatim in at least one paralogue, 67 of them in RANBP2 itself; exactly 4 peptides are unique to RGPD8** [E3]. The study's GLM used only peptides unique to the protein group, so the call is not a naive paralogue mis-assignment — but the discriminating space is four sequences, one of which is 33 aa and unlikely to be observed under the top-15 CID method used, and the evidence is three spectra. **Which peptides those were is recoverable from PXD009338 and has never been looked at.**
The family
Fetched live from UniProt on 2026-09-24 2:
| accession | gene | length |
|---|---|---|
| P49792 | RANBP2 (Nup358) | 3224 |
| O14715 | RGPD8 | 1765 |
| Q99666 | RGPD5; RGPD6 (one entry for two genes) | 1765 |
| A6NKT7 | RGPD3 | 1758 |
| Q7Z3J3 | RGPD4 | 1758 |
| P0DJD1 | RGPD2 | 1756 |
| P0DJD0 | RGPD1 | 1748 |
Seven reviewed entries for eight genes, all within 17 residues of each other, all derived from the RANBP2 locus by segmental duplication. The length check that verify-accessions.py relies on provides no protection whatever in this family — it is the RAS case of rule 68 again, one step worse, because here even the database declines to separate two of the members.
In-silico fully-tryptic digest of RGPD8, no missed cleavages, K/R not before P, minimum 6 aa 3:
- 114 distinct peptides
- 110 (96.5%) occur verbatim in at least one paralogue; 67 in RANBP2
- 4 (3.5%) are unique to RGPD8: a 33-mer, a 16-mer, a 14-mer and an 11-mer
A 33-residue tryptic peptide is at the far edge of what a top-15 CID method on an LTQ Orbitrap Velos will identify, so the practical discriminating space is nearer three peptides than four.
What the study's method does and does not rescue
The PRIDE data-processing record states the GLM was fitted to spectral counts after Mascot+Phenyx search against human SwissProt at <1% protein FDR, and the paper's Methods add "Only peptides unique to the protein groups were used" 4. That matters and it should be stated in RGPD8's favour: the call is not simply shared peptides being assigned to the first family member alphabetically.
What it does not rescue:
- The unique-peptide rule constrains the count to the four discriminating sequences, so three spectra across at most four candidate peptides is the entire dataset, and at a per-protein FDR of 1% over ~960 reported proteins the expected number of wrong protein calls is of that order.
- The search database version is disputed between sources — PRIDE says human SwissProt v2012.05, the paper's Methods say v2013.01 4. Which paralogues were in the search space at that date determines what "unique to the protein group" meant. RGPD1 and RGPD2 (P0DJD0, P0DJD1) are comparatively late SwissProt creations; if either was absent from the search database, peptides shared with it would have been scored as RGPD8-unique. This is checkable against a SwissProt release archive and was not checked here.
- Nothing about RGPD8 is nuclear or chromatin-associated. UniProt gives O14715 no subcellular-location comment at all, and its parent RANBP2 is a nuclear-pore component — a compartment that co-purifies with nuclear extracts routinely and is represented in this very table by TPR and NUP50.
RGPD8[tiab] returns 2 records in all of PubMed 5. There is no literature to weigh.
| partner a | gene-kmt2a |
|---|---|
| partner a uniprot | Q03164 |
| partner a length AA | 3969 |
| partner b | RGPD8 |
| partner b uniprot | O14715 |
| partner b length AA | 1765 |
| verdict | DECLINED |
| verdict scope | Declined because the IDENTIFICATION cannot be resolved to RGPD8 rather than to a paralogue on the evidence available, not because an RGPD8-KMT2A contact is implausible. Reopenable from PXD009338 at peptide level. |
| adjudicated by | Q180-the-af4-specific-interactors, cycle 159, 2026-09-24 |
| source publication pmid | 29777171 |
| source dataset | PXD009338, Skucha 2018 seven-bait MLL-fusion AP-MS |
| detection method | single-step StrepTactin affinity purification from Jurkat nuclear extract, LC-MS/MS spectral counting with a GLM background model |
| experimental role partner | prey |
| experimental role kmt2a | bait (as the MLL-AF4 fusion) |
| bait exists in dataset | true |
| n publications | 1 |
| n methods | 1 |
| orthogonal method exists | false |
| intact curated | false |
| af4 p | 0.0147 |
| af4 spectral | 2|1 |
| af4 spectral total | 3 |
| neg ctrl | 0|0 |
| rank in af4 by p | 50 |
| n fusions | 1 |
| druggability | Not assessable and not worth assessing. No compartment annotation, no literature, no interface, and an identification that cannot be resolved to the gene. |
| absence class | measured-not-searched |
| absence class note | The peptide sequences and PSM scores behind these three spectra are in PXD009338, public since 2018-05-25. Reading them would settle this outright and needs no wet work. |
| shown for fusion | MLL-AF4 only, as one prey row in one AP-MS experiment in a Jurkat T-ALL host with a tagged transgene. Nobody has looked in any other fusion, lineage or system. |
| epigenetic readout | none - no histone mark, occupancy or chromatin measurement exists for RGPD8 in any context. |
host cell lines
[ "Jurkat (human T-ALL)" ]
paralogue family
{
"parent": "RANBP2 (P49792, 3224 aa, Nup358)",
"members": [
"RGPD1 P0DJD0 1748",
"RGPD2 P0DJD1 1756",
"RGPD3 A6NKT7 1758",
"RGPD4 Q7Z3J3 1758",
"RGPD5;RGPD6 Q99666 1765 (one entry for two genes)",
"RGPD8 O14715 1765"
],
"note": "chr2 segmental duplication; length spans 17 aa across the family, so verify-accessions.py's length check gives no protection here - the rule 68 RAS case, one step worse"
}measurements
[
{
"quantity": "fraction_of_tryptic_peptides_shared_with_a_paralogue",
"subject": "RGPD8 (O14715)",
"value": 0.965,
"units": null,
"n": 114,
"universe": "fully-tryptic peptides >=6 aa, no missed cleavages, K/R not before P, against RANBP2 and five reviewed RGPD entries fetched from UniProt 2026-09-24",
"null": null,
"note": "110 of 114 distinct peptides shared; 67 shared with RANBP2 itself; 4 unique to RGPD8 at 33, 16, 14 and 11 aa"
},
{
"quantity": "spectral_counts",
"subject": "RGPD8|MLL-AF4",
"value": 3,
"units": "spectra (2|1 over two replicates)",
"n": 2,
"universe": "Skucha 2018 Suppl Table 5",
"null": 0,
"note": "null column is the neg_ctrl (0|0). Three spectra over a discriminating space of at most four peptide sequences."
}
]interface residues a
[]
structures
[]
ligands
[]
- Nothing here touches an intervention. The transferable risk is procedural: an entry written from this row would have carried an accession that verify-accessions.py validates perfectly, because O14715 is a real reviewed human accession of the stated length. A length check cannot detect a paralogue family whose members differ by 17 residues.
- Which peptide sequences produced the three RGPD8 spectra in PXD009338, and were any of them among the four RGPD8-unique tryptic peptides? This settles the entry and is a deposit read.
- Was RGPD1 (P0DJD0) or RGPD2 (P0DJD1) present in the human SwissProt release used for the search? If either was absent, peptides shared with it were scored as RGPD8-unique and the identification is weaker still.
- Should data/symbol-collisions.json gain a row for the RGPD family? It has 216 rows and none for RGPD1-8, and this is a cleaner instance of the rule-68 length-check failure than RAS: eight genes, seven entries, 17 aa of total spread, and two genes sharing one accession.
RGPD8 is the strongest member of the retained 110 at p 0.0147 on 3 spectra, the comparison that entry uses to calibrate the class
- related to →KMT2A / MLL1 — the gene that breaks
declined KMT2A::AFF1-RGPD8 pair: the identification cannot be resolved to the gene
whether any of RGPD8's three spectra maps to one of its four discriminating tryptic peptides; a negative turns that decline into a refutation