Eight claims this corpus carries from PMID 23995757, checked one at a time against the full text: five confirmed, one scope-limited, one materially different, one not in the paper at all
also called PMC3813012 verification table · q166 claim audit
We re-read the 2013 AKAP95 paper line by line to check what our own notes say it contains. Most of it was right. Two things were not: a binding experiment was described more precisely than it was actually done, and a localisation fact came from a database, not from that paper.
The same thing, for a scientist
Full text of PMC3813012 obtained from the NCBI PMC OAI service (49,400 characters, Methods and figure legends included; the Europe PMC `fullTextXML` endpoint returned **HTTP 500** on four attempts across two sessions). Each claim this corpus attributes to the paper was scored **CONFIRMED / SCOPE-LIMITED / DIFFERENT / NOT-FOUND** against the located text, with the figure or section named. **Result: 5 CONFIRMED, 1 CONFIRMED-but-scope-limited, 1 DIFFERENT, 1 NOT-FOUND.** The two that moved are: the **direct-binding experiment**, which tested full-length AKAP95 against AKAP95(101-692) and never made a 1-100 fragment at all; and the **nucleolus**, which appears **zero times** in the paper [E1]. **A third finding is arguably the most important and was not on the checklist**: the enzymatic enhancement of H3K4 methylation is measured only on the **MLL2/KMT2B** complex, and the paper's own Discussion says the other MLL complexes were not tested [E1].
The table
| # | claim as carried | verdict | located at |
|---|---|---|---|
| 1 | AKAP95 is a major FLAG-HA-DPY30-associated protein | CONFIRMED | Fig 3A, MALDI-MS of the FH-DPY30 Flp-In-293 line; confirmed by anti-HA IP, Supp Fig 2A |
| 2 | endogenous DPY30 co-IP from HeLa nuclear extract | CONFIRMED | Fig 3B, anti-DPY30 vs normal rabbit IgG |
| 3 | WDR5 and ASH2L IPs recover AKAP95 reciprocally | CONFIRMED | Fig 3C (FLAG-HA-WDR5), Supp Fig 2B (ASH2L), Supp Fig 2C (tagged AKAP95 pulls both) |
| 4 | purified AKAP95 residues 1-100 bind purified DPY30 directly | DIFFERENT | Fig 5F compares GST-AKAP95 full length vs GST-AKAP95(101-692) against bacterially purified M2-captured FH-DPY30. No 1-100 fragment exists in the paper. 101-692 binds 'much weaker', not zero. The 1-210 sufficiency is Fig 5E, a Myc-DPY30 co-IP from transfected G-293T cells, not purified protein. |
| 5 | recombinant AKAP95 enhances H3K4 me1/me2/me3 on recombinant chromatin | CONFIRMED, SCOPE-LIMITED | Fig 6B (3H-SAM), Fig 6C (state-specific immunoblot). Enzyme is MLL2C = KMT2B only. Discussion: 'It will also be interesting to examine whether AKAP95 directly enhances H3K4 methylation mediated by other MLL complexes.' No MLL1/KMT2A HMT assay exists. |
| 6 | no methyltransferase activity of its own, no free-H3 binding | CONFIRMED | Fig 6D lanes 5-6 and 8-9; Supp Fig 4B lanes 3-4 |
| 7 | localisation to nuclear matrix and nucleolus | NOT-FOUND | nucleol occurs 0 times in the full text. nuclear matrix occurs twice: the Introduction, citing prior work on p68 RNA helicase, and the NMTS acronym expansion. The paper performs no localisation experiment. Both annotations are UniProt's, from PMID 10601332 and PMID 26683827. |
| 8 | the NMTS is the one region NOT required for the function | CONFIRMED | Fig 5D + Supp Fig 3C: NLS and both zinc fingers required; NMTS and RII-binding not; the RII point mutant I582P has 'only a little or marginal effect' |
Where the error in claim 7 came from, since it matters for how the corpus handles this class. The existing entry interface-kmt2a-akap8-a-prey-only-grade-on-a-direct-mll-coactivator sources both localisations correctly, to UniProt, in its evidence item E3 with support_location: "database". The entry is right. The dispatch brief for this beat re-described E3's content as a full-text claim of the paper. An entry that correctly separates a database annotation from a paper's experiments can still be summarised into one that does not, and the summary is what the next agent reads.
Residue coordinates recovered, and the internal check that validates them
The paper never draws a domain map with numbers. Its deletion constructs are given only as joined coordinate pairs, buried in one Methods sentence:
dNMTS: 1-110 + 141-692 -> NMTS = 111-140
dNLS : 1-287 + 306-692 -> NLS = 288-305
dZF : 1-389 + 504-692 -> ZF region = 390-503
dRII : 1-571 + 590-692 -> RII-bind = 572-589
Reconstructed coordinates need a control, and there is one. The RII interval reconstructed this way, 572-589, is character-for-character the UniProt RII-binding feature; the zinc-finger interval 390-503 brackets UniProt's two C2H2 fingers at 392-414 and 481-504; the NLS 288-305 is one residue inside UniProt's bipartite NLS motif at 289-306; and the paper's own I582P point mutant, said to disrupt RII binding, falls inside 572-589, with residue 582 of O43823 confirmed as isoleucine by reading the downloaded sequence. Four independent agreements. The one interval with no external check is the NMTS, 111-140, which has no UniProt feature - it should be quoted as the region deleted in the published dNMTS construct, not as a database-backed domain.
| script | analyses/q166-the-akap95-dpy30-axis/pmc3813012-per-claim-verification.md |
|---|---|
| status | run |
| result summary | 8 claims scored: 5 CONFIRMED, 1 CONFIRMED-but-scope-limited (the HMT enhancement is KMT2B-only), 1 DIFFERENT (no AKAP95(1-100) fragment was ever purified or tested; the direct assay is full-length vs 101-692 and 101-692 binds weakly rather than not at all), 1 NOT-FOUND (the word nucleolus does not occur in the paper; that annotation is UniProt's, from PMID 26683827). Four reconstructed residue intervals corroborated against UniProt. |
inputs
[ "NCBI PMC OAI full text of PMC3813012", "UniProtKB O43823" ]
measurements — the numbers behind the chart
[
{
"quantity": "count",
"subject": "claims from PMID 23995757 scored CONFIRMED",
"value": 5,
"units": "claims",
"n": 8,
"universe": "the 8 claims this corpus and its dispatch brief attribute to PMID 23995757",
"null": null,
"note": "plus 1 confirmed-but-scope-limited, 1 DIFFERENT, 1 NOT-FOUND"
},
{
"quantity": "count",
"subject": "occurrences of 'nucleol' in the full text of PMC3813012",
"value": 0,
"units": "occurrences",
"n": 1,
"universe": "49,400-character NCBI PMC OAI text, case-insensitive literal grep, Methods and legends included",
"null": null,
"note": "the absence is the support for the NOT-FOUND verdict on claim 7"
},
{
"quantity": "count",
"subject": "AKAP95 constructs tested in the purified DPY30 binding assay",
"value": 2,
"units": "constructs",
"n": 2,
"universe": "PMID 23995757 Fig 5F and Methods: GST-AKAP95 full length and GST-AKAP95(101-692)",
"null": null,
"note": "neither is a 1-100 fragment; the corpus claim named a construct that does not exist"
},
{
"quantity": "count",
"subject": "reconstructed deletion intervals independently corroborated against UniProt features",
"value": 3,
"units": "intervals",
"n": 4,
"universe": "NMTS 111-140, NLS 288-305, ZF 390-503, RII 572-589 reconstructed from joined coordinate pairs in Methods",
"null": null,
"note": "RII exact, ZF brackets both fingers, NLS within one residue. NMTS has no UniProt feature and is uncorroborated."
},
{
"quantity": "count",
"subject": "HTTP 500 responses from the Europe PMC fullTextXML endpoint for PMC3813012",
"value": 4,
"units": "requests",
"n": 4,
"universe": "four attempts across two sessions on 2026-09-24; the NCBI PMC OAI route succeeded on the first try",
"null": null,
"note": "instrument note: when Europe PMC fullTextXML 500s, fall through to NCBI rather than concluding the text is unavailable"
}
]- A verification analysis that reports mostly CONFIRMED invites the reader to treat the source as fully checked. It is not: figure panels were not inspected and supplementary figures were unavailable, so this audit verifies that the paper SAYS what the corpus says it says - not that the figures support the paper's text.
- Would reading the supplementary figures change any verdict? Supp Fig 2B/2C, 3C, 4A/4B and 5A-D carry the reciprocal IPs, the NMTS dissection half and every ESC steady-state negative. They were not obtainable here, and three of the eight rows rest partly on the main text's description of them.
- How often does this corpus carry a database annotation as though it were a paper's finding? The AKAP8 case was caught only because a beat re-read the paper. A corpus-wide check would compare each evidence item's support_location against the claim's actual source and is scriptable.
- Has anyone ever made AKAP95(1-100)? A 100-residue fragment against purified DPY30 would convert a necessity claim into a sufficiency claim and is a one-week experiment that has been available for thirteen years.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Full text of PMC3813012 (PMID 23995757), 49,400 characters including Methods, Acknowledgments and all figure legends, was obtained and searched literally. The in vitro binding assay (Fig 5F, Methods 'Recombinant proteins and in vitro binding assay') uses GST-AKAP95 or GST-AKAP95(101-692) expressed from pGEX in bacteria against M2 bead-captured FLAG-HA-DPY30 purified from bacteria; no AKAP95(1-100) construct appears anywhere. The string 'nucleol' occurs zero times; 'nuclear matrix' occurs twice, neither an experiment of this paper. The HMT enhancement (Fig 6B-D) is measured on MLL2C only, and the Discussion states 'It will also be interesting to examine whether AKAP95 directly enhances H3K4 methylation mediated by other MLL complexes' and 'it is possible, but has not been rigorously demonstrated, that AKAP95 may also be commonly associated with all MLL complexes'. Deletion constructs are defined in Methods as dNMTS 1-110+141-692, dNLS 1-287+306-692, dZF 1-389+504-692, dRII 1-571+590-692. Microarray data are GEO GSE48128.
Regulation of transcription by the MLL2 complex and MLL complex-associated AKAP95. · Jiang H, Lu X, Shimada M, Dou Y, Tang Z, Roeder RG · Nat Struct Mol Biol 2013What this evidence cannot show
SUPPLEMENTARY FIGURES WERE NOT OBTAINED. Claims located in Supp Fig 1-8 are scored against the main text's own description of them, which is weaker than reading the panel. This affects claims 3 (partly), 5 (partly) and 8 (partly), and it affects every steady-state ESC negative. FIGURE PANELS WERE NOT INSPECTED for any claim: 'much weaker' and 'significantly enhanced' are the authors' words with no densitometry available. The expected_absent_terms here are positive findings, verified by literal grep over the complete downloaded text - the absence IS the support for rows 4 and 7 of the table.
Evidence that argues against this (1)
- A 'DIFFERENT' verdict on claim 4 is a verdict about precision, not about truth. The contact IS direct - both partners were bacterially purified - and the N-terminal region IS required. What is not supported is that residues 1-100 constitute a purified, sufficient binding module. Anyone reading this audit as 'the direct binding did not happen' would be over-correcting. Jiang H et al. 2013
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
UniProt O43823's RII-binding feature is 572-589 and its two AKAP95-type C2H2 zinc fingers are 392-414 and 481-504; its bipartite NLS motif is 289-306; residue 582 of the canonical 692-aa sequence is isoleucine. These independently corroborate the intervals reconstructed from the paper's joined deletion coordinates.
UniProtKB entry O43823 (AKAP8_HUMAN) feature table and sequence, retrieved live 2026-09-24 by q166 · UniProt Consortium 2026What this evidence cannot show
The UniProt features and the paper are not fully independent: the 'Interaction with DPY30' feature is explicitly sourced to this very paper (ECO:0000269 PubMed 23995757). The RII-binding, zinc-finger and NLS features predate it and come from other work, so those three corroborations are genuine. Residue 582 was read from the downloaded sequence string directly.
Evidence that argues against this (1)
- Corroborating a reconstruction against a database does not validate the reconstruction method, only its output on the intervals the database happens to annotate. The NMTS at 111-140 has no UniProt feature and is therefore the one interval with no check at all - and it is the interval the corpus most wants a number for, because the NMTS dissociation is the interesting functional result. It must be quoted as 'the region deleted in the published dNMTS construct', never as a domain boundary. Jiang H et al. 2013
claim 4 scored DIFFERENT - no AKAP95(1-100) fragment was ever purified or tested
claim 7 scored NOT-FOUND - 'nucleol' occurs zero times in PMC3813012
claim 5 scored CONFIRMED but scope-limited to KMT2B by the authors' own Discussion
the four reconstructed residue intervals and their corroboration against UniProt