The historical notes below are retained as a journal. The 2026-09-26 re-review supersedes the earlier action tally and the categorical peptide, extracellular-localization, unmapped-HP1-interface and disputed-Wnt conclusions in sections 2, 4–7 and 10. Current decisions are in the review YAML.
Working journal for the PAINT + affinage review. Provenance is inline as
[PMID:xxxx "verbatim quote"].
ADNP is a 1102-residue nuclear protein with nine C2H2 zinc fingers (UniProt ZN_FING
74..97, 107..129, 165..188, 221..244, 447..469, 489..510, 512..535, 622..647, 662..686),
a homeodomain (DNA_BIND 754..814, PROSITE PS50071), and a single C-terminal PxVxL motif.
Its defining activity is as the sequence-specific DNA-binding and assembling subunit of
the ChAHP complex:
Repression by ChAHP is not classical HP1 heterochromatin silencing:
Three later papers extend and sharpen this:
The paralogue ADNP2 forms ChAHP2 and covers a different transposon class
PMID:38960717.
26 of ADNP's 53 GOA rows are Ensembl Compara projections (GO_REF:0000107). Nineteen
come from rat Adnp (Q9JKL8) and seven from mouse Adnp (Q9Z103). Resolving each donor's
own primary reference — the campaign's highest-yield check — shows that 13 of the rat-derived
rows trace to experiments on the synthetic ADNP-derived octapeptide NAPVSIPQ ("NAP",
davunetide), applied exogenously, with the ADNP protein never perturbed or measured.
The classification is computed, not hand-assigned: ADNP-bioinformatics/analyze_compara_donors.py
resolves every Compara row to its donor annotation and primary reference, pulls the title and
abstract from PubMed, and reports which peptide markers and which ADNP-scoped protein markers
fired. Output is committed as results.json / RESULTS.md.
The worst cases, in ascending order of how obviously wrong they are:
| human GO row | rat donor reference | what was actually assayed |
|---|---|---|
GO:0005507 copper ion binding, GO:0042277 peptide binding |
PMID:14706557 | NAP peptide vs beta-amyloid aggregation; the only binding reported is biotin-NAP binding Abeta |
GO:0046068 cGMP metabolic, GO:0080164 reg. NO metabolic |
PMID:11438390 | VIP / SNV / NAP dosed onto cortical cultures |
GO:0033484 intracellular NO homeostasis, GO:0043524 neg. reg. neuron apoptosis |
PMID:16938277 | i.p. ADNF-9 + NAP peptides in neonatal rats |
GO:0007614 short-term memory |
PMID:12212775 | inhaled NAP in normal rats, water maze |
GO:0048487 beta-tubulin binding |
PMID:16893427 | NAP peptide on an affinity column |
GO:0010976, GO:0051965 |
PMID:15800376 | femtomolar ADNF-9 + NAP in culture |
GO:0045773 pos. reg. axon extension |
PMID:19047645 | NAP peptide; the only siRNA is against Fyn |
GO:0010629 neg. reg. gene expression |
PMID:15314252 | NAP; new datum is p53 as a NAP target |
GO:0050805 neg. reg. synaptic transmission |
PMID:15963648 | D-NAPVSIPQ, the all-D enantiomer |
The decisive single case is GO:0050805. The molecule is an all-D-amino-acid octapeptide,
which no gene can encode PMID:15963648, and the paper's own control says it did nothing on its own
PMID:15963648.
The suppression required co-application with 50 mM ethanol. The donor evidence code is IMP,
which normally implies a perturbation of the gene; there was none.
Second decisive case: GO:0046068 / GO:0080164. The paper is titled for vasoactive
intestinal peptide, and ADNP is not manipulated anywhere in it
PMID:11438390. The authors themselves decouple the doses
PMID:11438390.
GO:0042277 peptide binding — the argument I first gave was wrong; the verdict is not.
The binding measured is PMID:14706557,
so the entity that binds is the synthetic eight-residue NAP peptide. REMOVE rests on entity
identity alone: no experiment shows that sequence is released from ADNP, so nothing measured
here is a molecular function of the gene product. PMID:14706557 annotates exactly one entity
in all of GOA, so no cross-reference is involved.
Retracted from an earlier draft (kept here because process history belongs in the notes,
not in a row summary). I originally argued the annotation was directionally inverted and that
RGD's own records made the inversion "explicit and checkable", citing the GO:0042277 rows on
rat Tubb3 (Q4QRB4) and Tubb4b (Q6P9T8) that carry WITH/FROM RGD:71030. Both halves were
wrong. Abeta is a peptide, so for the fragment the direction is correct, not inverted. And
those tubulin rows come from PMID:16893427, the donor for GO:0048487 beta-tubulin
binding, not for GO:0042277 — RESULTS.md puts the two references on different rows.
There is no internal contradiction in RGD's records of the kind I described, and the claim
would have misled exactly the RGD/GO curators suggested_questions[0] is addressed to. The
tubulin cross-check is real and is deployed where it belongs, on the GO:0048487 row.
Generalisable lesson: check original_reference_id per row before reasoning across rows.
Guarded by audit_adnp_review.py check J, which selects on the stable tokens
(16893427, RGD:71030, Tubb3, Tubb4b, invert) rather than on the conclusion's wording,
fires on the exact version that shipped (12f374d14), and is clean on the current file. Its
stated limitation: a paraphrase avoiding all five tokens would pass, so the prose still needs
human re-reading.
Two of the peptides are not even ADNP fragments. ADNF-9 is the active site of a different
protein PMID:15800376, yet PMID:16938277 (ADNF-9 + NAP, i.p.) supplies two human IEA rows.
GO:0043524 reaches human twice and both chains end in peptide. The rat donor is
PMID:16938277; the mouse donor is the 1999 cloning paper
PMID:10037502. Checking the
other rodent source is what turned a suspicion into a verdict.
First pass called GO:0045773 MIXED, because PMID:19047645's abstract contains an siRNA. Reading
it: the siRNA is against Fyn kinase, not ADNP
PMID:19047645. Fixed by scoping protein-perturbation markers to
sentences that name ADNP. That changed the call on three references (PMID:15800376,
PMID:15963648, PMID:19047645) and is recorded in RESULTS.md. The self-test that encodes this
shape then caught a second defect in my own patterns: a bare \bdisrupt matched the title
"…by disrupting ADNP signaling", where what is disrupted is downstream NAP signalling, not the
gene product. Narrowed to require ADNP to be named as the thing lost.
positive regulation of
X / negative regulation of X pairs found no opposed pair at all, so no shared-referenceCommentsCorrections on their own PubMed records; none carries RetractionIn,ErratumIn, ExpressionOfConcernIn or a republication link. The sweep is performed by theanalyze_compara_donors.py and its result written into results.json /RESULTS.md, so the claim is reproducible rather than a one-off shell command. Known hole,RESULTS.md: a correction whose own PubMed id is null is invisible to this routerelation/update-to lookup.GO:0030424 axon, GO:0030425 dendrite, GO:0043025GO:0005576 extracellular region rest on genuine immunodetection of theGO:0044849 estrous cycle — decided from the definition, not the labelGO:0044849's definition is "A type of ovulation cycle, which occurs in most mammalian therian
females, where the endometrium is resorbed if pregnancy does not occur." Humans and other
catarrhines menstruate — the endometrium is shed. The formal taxon constraint is
only_in_taxon Theria, which humans satisfy, so the constraint does not block the annotation
but the differentia in the definition does. The rat annotation is separately weak: an IEP
recording mRNA oscillation PMID:16023261. REMOVE for human; the rat row is fine for rat, which is why
source_status is SUPPORTS_SOURCE_BUT_NOT_TARGET rather than SOURCE_BAD.
GO:0009743 response to carbohydrate — an honest non-determinationThe rat IEP cites PMID:19130308, whose abstract describes L-NAME, 7-nitroindazole, ODQ and
kainic acid — no carbohydrate PMID:19130308. The cached record has
full_text_available: false. I am not asserting the rat annotation is wrong — the supporting
experiment may be in the full text. I am declining to propagate an IEP across species, which is
a claim about the human row only. MARK_AS_OVER_ANNOTATED, root_cause: SOURCE_WEAK_OR_INFERRED.
GO:0005576 extracellular region — a real measurement that still should not travelUnlike the rest of the rat block this is a genuine protein measurement
PMID:16845437. Three things
hold it back: human ADNP has no SIGNAL feature (the FT table runs CHAIN 1..1102 with no
cleaved leader) and no non-classical secretion route has been proposed; UniProt lists only
Nucleus and Chromosome; and the antibody was raised against the NAP epitope
PMID:16845437, so it cannot distinguish full-length ADNP from a NAP-containing fragment in
conditioned medium. MARK_AS_OVER_ANNOTATED, not REMOVE.
Housekeeping: this row is the successor of GO:0005615 extracellular space, obsoleted
2026-03-06 and replaced by GO:0005576. The UniProt DR GO; list still shows the old id while
the GOA TSV has migrated — a harmless lag, noted so nobody reads it as a discrepancy.
The campaign's standing lead is "a homeobox does not establish sequence-specific DNA binding".
For ADNP it fails, and that is worth as much as a finding. The ISS/IEA route to GO:0000977 and
GO:0000981 runs through mouse PMID:17222401, a ChIP at promoters
PMID:17222401 — and ChIP occupancy alone genuinely
would not distinguish direct recognition from tethering. Note MGI read that same paper more
conservatively as GO:0003682 chromatin binding while ARUK-UCL read it as GO:0000977 +
GO:0000981. The later ChAHP work vindicates the stronger reading independently of the paper
in question (section 1). ACCEPT, and the directional child GO:0001227 is proposed additively
rather than as a replacement.
GO:0005515 block — 14 rows, one per partnerThe fetch-gene stub had 47 entries against 53 distinct GOA rows. The whole gap is
GO:0005515: 14 GOA rows collapsed to 8 stub entries, one per PMID, dropping the partner
distinction. Restored to one row per (reference, partner). No distinct GO term was missing.
Partners resolve to four reviewed canonical Swiss-Prot entries — CBX5 (P45973, 191 aa), CBX1
(P83916, 185 aa), CBX3 (Q13185, 183 aa), CHD4 (Q14839, 1912 aa). No TrEMBL clones, no ORFeome
fragments. HP1 binding is recovered by seven independent studies, including OpenCell at
endogenous expression (PMID:35271311) and IP-MS in human induced neurons (PMID:36950384), so
it is not screen noise. Per the brief I did not use UniProt's NbExp (5/7/3) as an evidence
proxy; the reference-level count is the meaningful one.
MODIFY targets:
- HP1 rows → GO:0070087 chromo shadow domain binding. Proteins engage HP1 through PxVxL
docking on the chromo shadow domain PMID:20562864, and ADNP's is conserved
PMID:38960717. The scan is committed (analyze_compara_donors.py, pxvxl_scan,
which also asserts Q9H2P0 is still 1102 aa before trusting any position): the single
P.V.[LMIV] match is PGVLL at 820–824, immediately C-terminal to the homeobox — but the
same function reports 0.76 matches expected by chance under the protein's own residue
composition, so one match is what chance predicts and its uniqueness is not evidence. An
earlier draft leaned on "exactly one match in 1102 residues" as though it were enrichment; it
is not. The weight rests on the ADNP↔ADNP2 conservation and on all three HP1 paralogues
binding. Caveat recorded as a knowledge gap: no ADNP PxVxL point mutant has been shown to lose
HP1 binding, so this is motif + paralogue-wide binding, not a mapped surface.
- CHD4 row → GO:0140463 chromatin-protein adaptor activity, whose definition ("brings together
a protein and a region of the chromatin … to establish or maintain the chromatin localization
of the protein, or the complex to which it belongs") is exactly what the experiments assign to
ADNP.
GO:0005654 TAS from Reactome:R-HSA-9940477 ("Formation of ChAHP complex"). Reference-projection
test: 4 entities (ADNP, CBX1, CBX3, CHD4), 1 term. Complex-membership localisation spreading
to exactly the complex's members, with no functional or phenotype term spreading alongside —
the benign shape, not the ACTR8 shape. ACCEPT.
gates_passed: True, 28 citations, faith_pct: 100.0. It is still not the literature search:
complexes: ChAHP complex (ADNP-CHD4-BRG1). ChAHP is ADNP–CHD4–HP1; BRG1/SMARCA4 belongsGO:0090575 row, quoting that exactThe papers it did surface that mattered: PMID:32533114 (Wnt), PMID:25178163 (EB1/EB3 SxIP),
PMID:38479840 (SINE B2 / CTCF in blastocysts), PMID:41174994 (methyltransferase claim).
Where the affinage record is cited. In references only, with a reference_review marking
it MISCITED and recording both defects. It is deliberately not in any supported_by:
supported_by should carry evidence for an annotation, and quoting a statement this review
rejects would misuse the slot. The cost is that the repo's non-blocking
"no annotations reference available deep research files" warning now stands — correctly, since
no annotation here rests on the provider record.
DISPUTED; GOA carries only the positive term.GO:0010976/GO:0045773.LOW_QUALITY for this claim specifically.genes/human/ADNP2/ absent from main), so thePTN000405125 IBA rows could not be run. Recorded as ahuman,Q9H2P0,ADNP verified against UniProt: Q9H2P0 / ADNP_HUMAN / 1102 aa.GO:0005634, GO:0010468). Both adjudicated. GO:0005634's WITH/FROM cites UniProtKB:Q9H2P0GO:0005515cache/go/terms.csv, in two acts. Act 1: just validate added GO:0070087 on a cacheGO:0001675, GO:0009566, 2→1 each). Caught by a multiset Counter comparison; a setorigin/main forward showedsrc/ai_gene_review/tools/cache_lint.py andtests/test_cache_sorted.py, which require every cache/**/*.csv to be sorted by CURIE anduv run python -m ai_gene_review.tools.cache_lint. Asserted afterwards four ways — nothingcheckquotes.py checks 138; the committed audit_adnp_review.py checks 145.knowledge_gaps[].provenance entries thatcheckquotes.py does not walk — derived independently from the parsed document rather thananalyze_compara_donors.py --self-testaudit_adnp_review.py --self-test (baseline + 9 mutations + the SafeLoader dedup baseline).Generated from the review file, not hand-counted. audit_adnp_review.py check I asserts
this table equals the computed counts and fails if they drift.
| action | rows |
|---|---|
| ACCEPT | 18 |
| KEEP_AS_NON_CORE | 7 |
| MODIFY | 14 |
| REMOVE | 1 |
| UNDECIDED | 13 |
| total | 53 |
The first version of this PR's body stated a hand-counted tally that was wrong on three of
six actions (MODIFY, KEEP_AS_NON_CORE and ACCEPT). REMOVE and MARK_AS_OVER_ANNOTATED, the two
that carry the argument, happened to be right — which is exactly why the error survived a
read-through. Check I exists because of it: anything countable should be counted, and then
compared against what was written.
Identity and scope: human ADNP, HGNC:15766, UniProt Q9H2P0; synonyms ADNP1 and KIAA0784. The cached UniProt identity agrees with HGNC-sourced Ensembl and the ENCODE gene record. ADNP2 is a distinct paralog. The parent independently verified all five baseline blobs on main 62134e998e0e4fbda34cc79564fb081e8dbbd951 and found no overlapping open ADNP PR. This session changes only the review, these notes, the rendered page and a newly scaffolded history record. It preserves the 53 original annotation objects, all machine sources and the historical bioinformatics outputs.
The old assertion that ADNP motif mutagenesis had never been performed is false. PMID:21267468, primary PLOS paper Figure 5 and Methods test the ADNP motif and the reciprocal HP1beta interface. Constructs use mouse Adnp cDNA and human HP1 proteins. Results/Figure report V821E, while Methods says V821A; this internal discrepancy and species-specific numbering should remain explicit. A peer independently confirmed these details. The normal PMID cache fetch failed; external reading is documented without creating a substitute cache.
The full cached PMID:38960717 independently contains ADNP experiments, not merely ADNP2 experiments: “To do this, we generated ADNPPxVxL motif point mutants (Supplemental Fig. S6B) and performed ChIP-seq.” ADNP loss from HP1-dependent H3K9me3 repeats and increased SINE binding after interface perturbation demonstrate complementary targeting modes. The human generic HP1-binding rows therefore retain their CSD-specific replacements, supported by additional mechanism rather than by pretending each proteomic screen mapped the interface. Three different HP1 partners across experiments do not imply simultaneous occupancy of one complex by all three.
PMID:29795351 establishes sequence recognition, stable ADNP-CHD4-HP1 assembly and regulation of lineage genes; PMID:31491387 adds CTCF motif competition. PMID:42413491 separates CHD4 remodeling from targeting and finds HP1 dispensable for specified mouse ESC outputs. That is compatible with HP1-dependent targeting in a different chromatin context. PMID:42413492 directly addresses Pol III SINE repression, so its process is assigned to a chromatin-recruitment core rather than coupled to a Pol II-specific molecular function. The three core entries distinguish DNA-binding transcription-factor activity, chromatin-protein recruitment and HP1-domain interaction.
The two pre-existing NEW rows are withdrawn as redundant descendants: GO:0001227 is within the existing Pol II transcription-factor activity branch, and GO:0010526 is a child of the existing GO:0010629 negative regulation of gene expression. Their biology remains in the synthesis. No replacement NEW process is proposed. GO:0070087 was checked against AmiGO; GO:0140463 against the FlyBase ontology report. The stored GO-CAM index had no match for the human accession or the named mouse/rat donor IDs. No claimed missing pathway annotation is inferred from that absence.
The preserved donor table is useful provenance: 26 Compara rows, with mouse Q9Z103 and rat Q9JKL8 and their Ensembl protein identifiers. Its PEPTIDE_ONLY classifier reads abstracts, not complete Methods, and cannot prove that a paper contains no native ADNP experiment. The present review does not label those curated donor records wrong solely from the classifier. Each propagated annotation now records source entities and its own support or unresolved boundary. The two IBA blocks use the ancestral PTN rather than treating its descendants as pairwise donors; target self-inclusion is valid.
Thirteen annotations are now UNDECIDED because full source or native-product linkage remains unresolved. The available assays are still stated explicitly: NAP/amyloid binding, cGMP/nitrite responses, behavioral treatment results, synaptophysin, neurite outgrowth and D-NAP/ethanol effects. This is not a universal assertion that synthetic fragments can never inform a gene function. It distinguishes a positive treatment result from the particular endogenous-product claim under review.
Specific corrections to the historical notes:
Extracellular localization changes to KEEP_AS_NON_CORE. The primary indexed PMC Methods/Results/Discussion for PMID:16845437 provide antibody-competition controls and size-resolved intracellular ADNP detection, alongside positive conditioned-medium immunoreactivity. These support the reported native-protein study; they do not fully resolve release mechanism or every extracellular molecular form. A missing signal peptide or a modest fold increase does not refute location. Direct PMC opening returned a browser challenge, while indexed primary sections were readable; the local cache remains abstract-only. Rat neuronal axon/dendrite/cell-body distributions are retained with source-resolution limits and curator deference.
The estrous-cycle human transfer remains REMOVE on the actual GO:0044849 definition: a cycle involving endometrial resorption does not describe human menstruation. The rat transcript observation is valid in its species, and the human rejection does not depend on declaring IEP invalid.
PMID:32533114 full Results/Methods support endogenous beta-catenin association, domain mapping, in-vitro-translated protein interaction and neural differentiation with rescue. PMID:27903678 supports WNT repression in colorectal-cancer models. These are distinct contexts; neither is marked DISPUTED merely because the directions differ. The developmental interaction/process rows remain non-core, and neuronal differentiation can be a genuine regulatory role rather than merely a phenotype.
PMID:25178163 explicitly reports native ADNP interactions with EB proteins as well as NAP experiments. The source is not exclusively peptide pharmacology. PMID:41174994 reports activity in ADNP-containing immunoprecipitates; intrinsic methyltransferase catalysis remains UNVERIFIED, without asserting contamination or broadly low-quality work. The 2014 disease paper PMID:24531329 is used for human genetic context, not as a direct molecular-function assay.
All original references and their empty findings lists were inspected; manual judgments now match the annotation decisions. Source titles are preserved. full_text_unavailable reflects the local cache, not external reading: it is false for the seven records with extracted full-text sections (PMID:19047645, PMID:24531329, PMID:32533114, PMID:33961781, PMID:35271311, PMID:36950384 and PMID:38960717). The first, BioPlex and OpenCell files are explicitly partial. It is true for abstract-only records and the absent new cache. Full source tables were not inferred from generic study abstracts: the specific human IPI records are retained with curator deference and independently corroborated mechanism.
The genuine Affinage report is unchanged. Its ChAHP composition error is marked in its reference review. The historical bioinformatics report is also unchanged; its claim that no ADNP point mutant was tested is superseded here and in its reference review. Neither is used as a substitute for primary evidence, and no fresh network/retraction sweep is claimed.
Required Falcon research and publication caching were launched concurrently. The normal Falcon wrapper with a 1200-second timeout failed before provider execution: deep-research-client dependency retrieval from PyPI failed DNS after three retries (4.8 seconds). Perplexity-lite fallback encountered the same dependency DNS failure (7.7 seconds). No provider output was created or fabricated. The existing genuine Affinage report was read, and this section records the manual review.
The original 33 PMID records were already cached. A notes-inclusive census found the historical lead PMID:38479840 was missing; the additional mechanism paper PMID:21267468 is also absent. One normal fetch for each failed DNS, with zero cached publications. Keep the publication draft until these two required caches are normally recovered. The older PMID:38479840 lead is not newly used to support a changed biological judgment.
Current tally: 53 original rows retained; 18 ACCEPT, 14 MODIFY, 13 UNDECIDED, 7 KEEP_AS_NON_CORE and 1 REMOVE. Two redundant prior NEW rows were withdrawn. All source assertions, provider files, publication caches, UniProt, GOA and existing bioinformatics files remain unchanged. Validation, history and rendering results are recorded below when complete.
Validation completed: just validate human ADNP passed with two warnings: the missing PMID:21267468 cache and intentional non-use of the faulty provider report as annotation evidence. The history validator and just render human ADNP passed. A separate notes-inclusive census found 35 cited PMIDs, 33 cached and the two missing records listed above. All 53 original source objects and 42 original reference ID/title pairs compare equal to baseline; UniProt and GOA are byte-identical. The YAML has no anchors, aliases or trailing whitespace. The generated HTML was checked for the revised action tally and evidence-boundary text. Exact byte counts, SHA256 hashes and base Git blobs are in /tmp/ADNP-local-manifest.json for parent publication.
Set the YAML status to DRAFT under the literal GeneReviewStatusEnum, which
reserves COMPLETE for reviews without validation warnings. The already documented
source-cache warnings remain unresolved. All biological judgments, source
assertions, reference assessments and core functions are unchanged. This status
label correction does not imply that source retrieval or automated review has
subsequently succeeded.
Both required records now contain normally fetched XML full text.
PMID:21267468 (PMC3022755) is a YAML reference; its local
full_text_unavailable flag is now false. Recovered Methods and Figure 5
agree with the previous external reading: mouse Adnp cDNA NM_009628, human
HP1 constructs, reciprocal HP1beta W170A and ADNP motif perturbation. The
Results/Figure V821E versus Methods V821A discrepancy is retained, without
changing it to a human residue coordinate or claiming direct H3K9me3 binding.
The existing specific domain-binding judgments need no change.
The historical notes-only lead PMID:38479840 (PMC10982698) also has cached
full text. Its mouse preimplantation CUT&RUN and zygotic Adnp knockout
experiments concern SINE B2-derived CTCF-binding sites, with altered
heterochromatin marking and transcription during blastocyst formation. The
recovered Results retain this species and developmental scope. This cache
closure does not convert the historical lead into a new human process assertion
or add it as a supporting reference for an unchanged annotation.
The exact records originate from standard fetch output in Actions run
36286975328, head 5946477c8ac79ade0709264c775ea1262b108438, artifact
10920674630. The transported ZIP SHA-256 is
c0ffe4a66b80278af34b44aab6a3ae354ffd5699236b3a486ca95527be5e9713;
tmp/verified-reference-records/local-import-receipt.json records per-file
hashes. Only these two ADNP-required cache records enter this manifest.
All 53 source assertions and reviews, three cores, 43 reference identities,
machine sources, genuine provider output, historical bioinformatics files and
prior history are preserved. This entry supersedes the missing-cache status
in earlier dated notes. Targeted validation, render and history checks are
recorded in the closure manifest. The existing provider-quotation advisory,
if still emitted, keeps the YAML DRAFT under the zero-warning COMPLETE rule,
separately from the now-closed publication-cache gate.
This entry supersedes historical audit claims and counts in earlier sections;
the generated verdict table above now describes the current 53-row review.
All actions, source assertions and three cores remain unchanged: 18 ACCEPT,
14 MODIFY, 13 UNDECIDED, 7 KEEP_AS_NON_CORE and 1 REMOVE; no NEW rows.
The source-audit harness now reconciles each saved Compara GO/donor pair with
its unchanged source assertion, propagation source and donor PMID(s). It also
recomputes the peptide/protein indexes from the saved per-entry classifications.
It no longer requires a particular action or SOURCE_BAD root cause from an
abstract classifier. Offline mutation checks exercise lost donor references and
index drift. The quote check now preserves case and normalizes whitespace only.
The notes tally includes a checked total. The analyzer's new offline
--render-snapshot mode regenerates RESULTS.md from the unchanged results.json;
it makes no network or fresh retraction-check claim. Its motif interpretation
separates sequence statistics from experimentally established binding. The
historical JSON interpretation is retained as raw provenance, not endorsed as
the current literature conclusion.
The two rat NAP-treatment projections GO:0046068 and GO:0080164 remain
UNDECIDED after explicit reconsideration of their IEA status. The accessible
original PubMed abstract for PMID:11438390
and author institutional record
report positive peptide concentration-response experiments in rat cortical
cultures. They do not establish the endogenous human-protein link; complete
Methods/Results were not recovered. This uncertainty concerns the human transfer,
not an assertion that the rat experiment is wrong. The distinction from
GO:0010629 is positive evidence: independent ChAHP experiments establish
ADNP-dependent repression, without validating that row's original NAP donor
chain. Those experiments do not independently contradict cGMP or NO regulation.
Lack of a corresponding native-protein assay does not itself supply that
contradiction. The peer source consultation recovered no additional full
PMID:11438390 experiments and agreed with this bounded uncertainty.
All chromoshadow-domain refinements now quote the recovered primary
PMID:21267468 Results directly: “Yet, recruitment of ADNP to H3K9me3 was severely
impaired in presence of the mutant HP1β W170A protein compared to the wild type
HP1β protein.” The tested ADNP is mouse and the reciprocal HP1 mutant is human;
the original domain-mapping reference assessment retains the V821E/V821A
reporting discrepancy. These experiments complement the human interaction
records and later ADNP-mutant work; they do not turn each screen into a
contact-mapping assay. The standalone description now states the mouse
embryonic-cell biology directly rather than discussing study provenance.
The IBA propagation source blocks remain PTN-only as required by the annotation
reviewer instructions; descendant evidence remains intact in supporting_entities.
All 35 notes-inclusive PMIDs remain cached; no new reference or cache is added.
The existing provider-quotation advisory is independent of this closed cache
gate, so status remains DRAFT. Validation, rendering and exact file hashes are
recorded in the follow-up manifest.
Follow-up validation completed: just validate human ADNP passed with only the
intentional provider-quotation advisory. The source harness reconciled all 53
GOA rows, 26 donor rows and 130 case-sensitive whitespace-normalized quotes.
Its 14 offline test directions, the analyzer's 10 classifier/motif directions
and snapshot-render equality check passed.
The approved revision at 1c3873ad506c83afc650b15a89840a7b9c697d58
was checked against all 14 local gene/history baseline files before editing.
The latest exact-head formal review (5329130918) approved the source-driven
harness and retained peptide-transfer uncertainty. Its nonblocking suggestions
do not require additional annotation assertions. The historical classifier JSON,
provider report, machine records and all published history remain unchanged.
This entry supersedes the earlier cache census, not the scope of the
previously read experiments. Recursive typed-citation scanning of the YAML,
notes, genuine Affinage report and bioinformatics source/report files finds
57 distinct PMIDs and seven Reactome records, all now present. The
notes-only DOI and PMC links resolve to records already in that set:
10.1371/journal.pone.0015894/PMC3022755 is PMID:21267468,
PMC1502393 is PMID:16845437, and PMC10982698 is PMID:38479840.
The Affinage report has no additional DOI-only or linked raw PDF/HTML
bibliography. Unused UniProt bibliographic entries are not new authored
citations. This census does not recursively turn every bibliography of every
primary paper into a new review dependency.
Source9 supplied the 21 previously absent provider PMIDs and three broad
Reactome records; source8 supplied the other three broad Reactome records.
All 27 exact normal-fetch records are absent from the current PR base and
match the coordinator's immutable canonical-import receipts. Source9 run
36302278898 used head cbe06b74dfe671a5e741f9f9baa04743a9b1f3dd
and artifact ZIP SHA-256
7bca7e6cddc4f37ab6c1b3b6e682c9914a0614dc01f0a6efcec5666c8cba5a65.
Source8 run 36301782511 used head
2be3635f79c080562d2ee255725fc8061d0fed47 and artifact ZIP SHA-256
2ebb896c12cced8cdd2dd2da38a87766e636bc503fcd633967e54d021b0f1fdd.
The earlier ordinary local source9 attempts failed DNS and wrote no records;
the remote standard fetcher subsequently succeeded. No cached text was edited,
manufactured or overwritten, and no redundant local retry was needed.
The new PMID records comprise 13 recovered bodies and eight abstract-only
records. Access labels below reflect the actual local record. Body review
focused on the relevant Methods, Results, figure captions and limitations;
it is not a claim that all supplementary data or every interaction was
independently reproduced. A listed PMCID alone does not change an abstract-only
cache into a full-text record.
| Primary source | Local access and sections inspected | Source-specific judgment |
|---|---|---|
| PMID:16564114 | Abstract only; Cached PubMed abstract | PACAP38 stimulates ADNP mRNA in mouse neuron-glia co-cultures, with receptor/inhibitor effects. ADNP is the regulated gene in this experiment; this does not demonstrate that ADNP itself is the PACAP receptor or performs the upstream signaling step. |
| PMID:18286385 | Abstract only; Cached PubMed abstract | P19 differentiation, ADNP distribution and shRNA phenotypes support developmental and cytoplasmic context. The abstract does not resolve every brain specimen's species; no human neuron-specific localization or direct neurite-building mechanism is inferred from the P19 model alone. |
| PMID:23272107 | Body; PMC3522725 Methods, Results and Figures 1-6 | Administered NAP alters microtubule dynamics in rat PC12, cortical astrocytes and neurons, with NIH3T3 comparisons. The soluble/polymerized assay is nonsignificant at two hours but modestly positive at four hours. These are peptide-treatment experiments, not direct native human full-length ADNP assays. |
| PMID:24489906 | Body; PMC3906161 Methods and Results/Figure 5 | Mouse brain nuclear extracts show ADNP immunoreactivity co-precipitating with BRM and PSF, with antibody-free controls and tissue/age-dependent limitations. Tau-isoform expression correlations and the proposed splicing model do not independently demonstrate direct ADNP splice-site catalysis or a general splicing activity. |
| PMID:25169753 | Body; PMC4195434 clinical presentation and ADNP-in-nBAF sections | The paper combines clinical/genetic analysis, including an additional patient, with discussion of prior BAF interaction work. Its BAF interaction paragraph cites earlier experiments; it is not a newly purified ChAHP composition experiment. The machine PRIMARY_RESEARCH label does not make every mechanistic statement an original assay. |
| PMID:25646590 | Body; PMC4445743 Methods and Results/Figure 3 | Mouse hippocampal ADNP/eIF4E co-immunoprecipitation, predicted binding motifs and sex/genotype-dependent eIF4E expression are positive observations. They do not by themselves measure an ADNP-catalyzed translation-initiation reaction. Human postmortem expression is a separate observational arm. |
| PMID:28115743 | Abstract only; Cached PubMed abstract | NAP changes EB comets, EB3 homodimers and EB-Tau association in cell models; NIH3T3 protection requires introduced Tau. Preserve peptide and model scope. The study's proposed endogenous ADNP mechanism is not equivalent to a direct full-length human-protein assay in the accessible abstract. |
| PMID:28219017 | Abstract only; Cached PubMed abstract | The abstract reports FMDV leader protease interaction with ADNP, ADNP recruitment to IFN-alpha promoters and a BRG1-containing complex. These support a virus-associated transcriptional context. The abstract does not identify all construct/host species, so no human-specific antiviral chemistry is assigned. |
| PMID:29911927 | Abstract only; Cached PubMed abstract | GFP-tagged ADNP variants in HEK293T cells alter nuclear localization, heterochromatin enrichment and stability; MG132 rescues selected truncated forms. Mutant trafficking and degradation are distinguished from wild-type localization and from ADNP being a proteasomal enzyme. The local cache remains abstract-only despite a listed PMCID. |
| PMID:33967268 | Abstract only; Cached PubMed abstract | The source describes ADNP-SIRT1 associations, correlated expression and histone/microtubule contexts. Exact domain contacts and every model's species cannot be reconstructed from the abstract; association does not confer SIRT1 deacetylase chemistry on ADNP. |
| PMID:35538192 | Abstract only; Cached PubMed abstract | ADNP mutant/live-cell microtubule effects and mouse brain actin co-immunoprecipitation are distinct from predicted SH3/actin motifs and NAP rescue in Shank3-mutant mice. The abstract reports motif perturbation, but this does not independently establish an isolated human ADNP actin-binding domain assay. |
| PMID:35775424 | Abstract only; Cached PubMed abstract | Human postmortem ADNP/POGZ expression and viral knockdown in mouse prefrontal cortex are separate arms. Neuroinflammatory, microglial and synaptic outcomes do not demonstrate that ADNP directly transports neurotransmitters or catalyzes inflammatory chemistry. Local cache remains abstract-only. |
| PMID:36230962 | Body; PMC9563912 Methods and Results | CRISPR-edited mouse N1E-115 cells express GFP-Adnp under its endogenous murine promoter, including Pro403 and Tyr718 counterparts of human mutations. Morphology, survival and nuclear/cytoplasmic distribution differ between variants and respond to NAP. Human plasmid results discussed in the introduction are prior studies, not the species of the current edited clones. |
| PMID:37285842 | Body; PMC10501989 Methods, Results and Figures 5-7 | Primary mouse Th2 cells show endogenous ADNP association with CHD4, BRG1 and BATF plus ADNP-dependent chromatin recruitment of remodelers. GATA3/AP-1 binding persists without ADNP; HP1gamma does not colocalize at the activating loci. This is positive recruitment/scaffold evidence, distinct from the HP1-containing repressive ChAHP complex and from a purified direct GATA3-ADNP binding assay. |
| PMID:37365244 | Body; PMC10618100 Methods and Results/Figures 1 and 5 | Adnp exon-5 heterozygous deletion in mice produces age-dependent excitability and synaptic phenotypes. Juvenile excitation/inhibition changes differ from adult basal transmission; adult LTP responds to CaMKII inhibition. The source explicitly leaves the cause of CaMKII hyperphosphorylation unresolved, so it does not establish ADNP kinase activity or a uniform synaptic-transmission direction. |
| PMID:37759476 | Body; PMC10527813 Methods and Results sections 3.1-3.7 | Cy5-NAP enters nuclei of edited mouse neuroblastoma clones and improves mutant-associated distribution defects. Ketamine-ADNP and NAP-zinc-finger contacts are docking predictions, separate from imaging and fractionation. Neither predicted contact is treated as a measured binding activity. |
| PMID:39715923 | Body; PMC12092271 Methods and Results/Figure 4 | Mouse models combine BrdU labeling, hippocampal transcriptomics and NAP treatment. MitoTracker colocalization uses edited mouse GFP-Adnp neuroblastoma cells; human targeting-sequence structure is modeled. Colocalization does not resolve an intramitochondrial subcompartment or prove direct mitochondrial transcription catalysis. |
| PMID:39808658 | Body; PMC11760920 Methods, Results/Figure 7 and Discussion | Mouse telencephalon-specific Adnp/Chd4 knockouts, co-immunoprecipitation and CUT&RUN show ADNP-dependent CHD4 recruitment and cortical-growth gene regulation. Effects involve active regulatory elements and differ from embryonic-stem-cell repression. The authors bound inference from strong conditional knockouts to heterozygous human disease; Chd4's enzymatic activity is not reassigned to ADNP. |
| PMID:40188316 | Body; PMC12339388 Methods and Results/Figures 3-6 | Human iPSC-derived microglia show ADNP-dependent synaptic-material uptake, acidification and motility phenotypes, and ADNP colocalizes with early endocytic markers. No change in bulk dextran endocytic load was observed. A direct trafficking role is suggested, while necessity, localization and proteome changes do not identify a specific catalytic or structural endocytic step for a NEW assertion. |
| PMID:40498903 | Body; PMC12677912 Methods, Results/Figures 6-7 and Discussion; live PubMed identity | In human iPSC-derived NPCs, ADNP is a POU3F2 target and interacting partner. ADNP overexpression leaves baseline Wnt output unchanged and reduces Wnt3a responsiveness; SOX13 is the principal positive effector. This refines the existing context-dependent Wnt explanation without reversing the original positive mouse neural-differentiation result. |
| PMID:40536108 | Body; PMC12257630 Methods and Results/Figures 1-7 | A heterozygous mouse C-terminal frameshift reduces full-length Adnp and excitatory/inhibitory PFC transmission; LSD1 inhibition rescues several outcomes, with sex differences. The study explicitly allows different mechanisms for mutation and inhibitor effects and no baseline difference in four tested histone marks. It does not confer LSD1 histone-demethylase chemistry on ADNP. |
The six recovered provider Reactome citations are broad pathway summaries:
R-HSA-112316 (Neuronal System), R-HSA-1266738 (Developmental Biology),
R-HSA-162582 (Signal Transduction), R-HSA-168256 (Immune System),
R-HSA-4839726 (Chromatin organization) and R-HSA-74160
(Gene expression (Transcription)). Their identities and summaries were read,
but these records do not identify an ADNP-specific catalyst, physical component
or reaction. They are kept as the report's contextual citations, distinct from
the existing specific ChAHP formation event R-HSA-9940477. Their availability
does not validate every provider-predicted pathway assignment.
Two recovered mechanistic distinctions matter to this review. First,
PMID:37285842 demonstrates ADNP-dependent CHD4/BRG1 recruitment in mouse
Th2 cells, with pioneer-factor occupancy maintained in ADNP-deficient cells.
The activating assembly is not evidence that BRG1 replaces HP1 in canonical
ChAHP. The provider's incorrect ChAHP composition remains explicitly flagged.
Second, PMID:40498903 supplies direct human NPC evidence that the direction
of Wnt regulation varies by cellular context. Its Results state:
“ADNP overexpression did not affect baseline Wnt signalling (Fig. 7E) but
significantly reduced Wnt3a responsiveness in both genotypes (Fig. 7F).”
The cached full source and Figure 7 were read, and its title/DOI/PMID were
independently checked at PubMed. This evidence is attached to the existing
Wnt row and existing question. The positive mouse-neural-differentiation
experiment in PMID:32533114 remains valid; neither result is asserted as a
universal direction across cells or species.
All 53 seeded source assertions and actions, all three complete cores,
and all 43 original reference identities are preserved. One primary
reference, PMID:40498903, is added to support the bounded Wnt context;
there are now 44 top-level references. The Wnt reason, its donor explanation
and the existing Wnt question gain that limitation. The provider assessment
records the recovered primary corrections. No NEW row or new core process
is introduced. Human microglial endosomal localization and perturbation
phenotypes in PMID:40188316 are recorded as a scoped lead, without guessing
which endocytic step ADNP performs. NAP/cGMP/NO transfer uncertainties are not
resolved by unrelated chromatin or synaptic studies.
Source closure is separate from the intentional provider-quotation advisory.
YAML remains DRAFT if that advisory is emitted; with no missing required
source, the PR may receive a fresh current-head review. Validation, history,
rendering, exact case-sensitive quote checks, immutable-source comparisons and
final file hashes are recorded in the closure manifest.
Independent read-only peer consultation confirmed the mouse Th2 Methods and
Figures 5-7 in PMID:37285842: endogenous association plus ADNP-dependent
remodeler recruitment supports a distinct activating assembly, without a
purified fixed-stoichiometry bridge or human Th2 assay. No source or annotation
changes were made by that consultation.