ADNP2 is a nuclear, chromatin-associated protein of 1131 residues built from nine C2H2 zinc fingers arranged in two N-terminal clusters and a single C-terminal homeodomain, followed by a short tail carrying a PxVxL motif. It is the defining subunit of ChAHP2, a three-protein assembly of ADNP2, the HP1 paralogue HP1beta/CBX1 and the CHD4 nucleosome remodeller, which can be reconstituted from purified full-length human ADNP2, HP1beta and CHD4. ADNP2 docks HP1 through its PxVxL motif, and it is HP1beta's recognition of H3K9-trimethylated nucleosomes, rather than any sequence read-out by ADNP2 itself, that determines where the complex sits: ADNP2 occupies several thousand sites that are overwhelmingly heterochromatic repeats - endogenous retroviruses and LINE1 elements - with few peaks at transcription start sites, and no DNA sequence motif accounts for its distribution. Losing ADNP2 de-represses those retrotransposons without lowering H3K9me3, so ADNP2 acts downstream of heterochromatin rather than establishing it. Its paralogue ADNP forms the compositionally similar ChAHP complex but occupies euchromatic SINEs through sequence-specific DNA binding; the two proteins do not co-purify, their target classes barely overlap, and some elements are silenced only when both are removed, so the two complexes divide retrotransposon control between them. Outside chromatin, reducing ADNP2 alters the sensitivity of cultured cells to hydrogen peroxide, and loss of the zebrafish orthologues impairs erythroid maturation and neural induction, the latter through direct repression of BMP-related ventralising homeobox genes.
Definition: A protein-containing complex composed of ADNP2, a heterochromatin protein 1 paralogue (predominantly HP1beta/CBX1) and the CHD4 nucleosome remodeller, recruited to H3K9-trimethylated heterochromatin through the HP1 subunit, where it represses endogenous retroviruses and LINE1 elements. Distinct in composition and chromatin targeting from the ADNP-containing ChAHP complex, with which it does not exchange subunits.
Justification: Directly reconstituted from purified full-length human ADNP2, human HP1beta and human CHD4 and characterised by SEC-MALS, and recovered by affinity purification of endogenously tagged ADNP2 in mouse ES cells. ADNP and ADNP2 do not co-purify, so the two complexes are separate entities rather than variants of one. Should be created alongside the ChAHP complex term proposed by the merged ADNP review.
Parent term: protein-containing complex
Supporting Evidence:
| GO Term | Evidence | Action | Reason |
|---|---|---|---|
| GO:0005634 nucleus | IBA GO_REF:0000033 | ACCEPT | Summary: Accepted as a core localisation. Every seed named in the WITH/FROM is an ADNP-clade gene rather than an ADNP2 orthologue - UniProtKB:Q9H2P0 is human ADNP and ZFIN:ZDB-GENE-061215-112 is zebrafish adnpa - but PANTHER:PTN000405125 is the node ancestral to both paralogue clades, so a paralogue seed is the correct construction here rather than a defect, and both seeds carry their own experimental evidence in the nucleus subtree (human ADNP GO:0090575 IDA PMID:29795351; zebrafish adnpa GO:0005634 IDA PMID:32533114). The term is independently true of ADNP2: ChIP-seq of endogenously tagged ADNP2 recovers thousands of chromatin sites, and full-length human ADNP2 assembles with human HP1beta and human CHD4 into a nuclear chromatin complex. Reason: A nuclear chromatin factor by direct observation. Node PTN000405125 spans 44 ADNP and ADNP2 orthologues and propagates exactly two terms to all of them (this one and GO:0010468), which is an appropriately conservative last-common-ancestor call for a node whose two halves have demonstrably diverged in targeting mechanism. Propagation Review Root cause: NO FAILURE CORE Sources checked: PANTHER:PTN000405125 SUPPORTS TRANSFER PANTHER internal tree node, not a protein. Its reach is 44 gene products spanning both the ADNP and the ADNP2 orthologue clades, and it carries only GO:0005634 and GO:0010468. UniProtKB:Q9H2P0 SUPPORTS TRANSFER Human ADNP, reviewed (Swiss-Prot), 1102 aa. A paralogue rather than an orthologue, which is legitimate for an IBD seed at an ancestral node. Carries GO:0090575 IDA (PMID:29795351) in the nucleus subtree. ZFIN:ZDB-GENE-061215-112 SUPPORTS TRANSFER Zebrafish adnpa, resolves to F1QLG5 ADN1A_DANRE, reviewed (Swiss-Prot), 969 aa. Also an ADNP-clade gene. Carries GO:0005634 IDA (PMID:32533114). Supporting Evidence: PMID:38960717 Peak calling revealed 6315 regions with significant ADNP2 enrichment over input chromatin PMID:38960717 Finally, to directly confirm that ADNP2, HP1Ξ², and CHD4 form a stable complex, we performed in vitro reconstitutions. |
| GO:0010468 regulation of gene expression | IBA GO_REF:0000033 | ACCEPT | Summary: Accepted, but this is the one ADNP2 row whose donor set contains NAP-peptide pharmacology, so the provenance and the soundness have to be stated separately. RGD:71030 is rat Adnp, and its only experimental annotation anywhere in this term's subtree is GO:0010629 IDA from PMID:15314252, a paper titled "NAP mechanisms of neuroprotection" whose subject is the synthetic octapeptide NAPVSIPQ rather than the gene product - the same defect class the merged ADNP review classified SOURCE_BAD plus ROLE_CONFLATION. That is provenance, and on its own it supports no verdict. The substantive question is whether the chain is empty of gene-product evidence, and it is not: the co-seed MGI:MGI:1338758 is mouse Adnp, which carries GO:0010468 IMP from PMID:32533114 and two IDA rows (GO:0000981, GO:0006357) from PMID:17222401, all perturbations of the protein. So one of two seeds contributes peptide pharmacology and the other contributes gene-product experiments, and the row stands on the second. Reason: Independently true of ADNP2 regardless of the donor set: ablating ADNP2 in mouse ES cells raises the RNA of specific retrotransposon families, with MMERVK10C responding to ADNP2 alone. The general term is the right depth for a node that also covers ADNP, whose regulatory targets and recruitment mechanism are different. The row-level root cause is therefore NO_FAILURE_CORE even though one of its three source entities is marked SOURCE_BAD; that combination is deliberate and is the point of the row. The rat seed's peptide-derived contribution is recorded so it is visible if this row is revisited, and it is worth reporting upstream: a seed annotation resting on a synthetic peptide should not be counted toward an IBD. Propagation Review Root cause: NO FAILURE CORE Sources checked: MGI:MGI:1338758 SUPPORTS TRANSFER Mouse Adnp, resolves to Q9Z103 ADNP_MOUSE, reviewed (Swiss-Prot), 1108 aa. Carries its own GO:0010468 IMP (PMID:32533114) plus GO:0000981 and GO:0006357 IDA (PMID:17222401). These are experiments on the protein, and they are what the row rests on. PANTHER:PTN000405125 SUPPORTS TRANSFER PANTHER internal tree node covering both the ADNP and ADNP2 orthologue clades; not a protein. RGD:71030 SOURCE BAD Rat Adnp, resolves to Q9JKL8 ADNP_RAT, reviewed (Swiss-Prot), 1103 aa. Its sole experimental annotation in this subtree, GO:0010629 IDA, cites PMID:15314252, which assays the synthetic NAPVSIPQ octapeptide rather than the Adnp gene product. Marked SOURCE_BAD at the entity level because that is what it is, while the row as a whole is sound because the co-seed supplies gene-product evidence for the same term. Supporting Evidence: PMID:38960717 In contrast, MMERVK10C derepression was exclusively ADNP2-sensitive despite also being weakly bound by ADNP PMID:15314252 An 8-amino-acid peptide, NAPVSIPQ (NAP), was identified as the smallest active element of activity-dependent neuroprotective protein file:human/ADNP2/ADNP2-bioinformatics/RESULTS.md Human and mouse ADNP2 contain **no NAPVSIPQ and not one `NAP` tripeptide** in 1131 / 1165 residues. file:human/ADNP2/ADNP2-bioinformatics/RESULTS.md ADNP's rat-Compara block covers **19 terms**. Terms from that block that also appear on ADNP2: **none**. |
| GO:0003677 DNA binding | IEA GO_REF:0000002 | KEEP AS NON CORE | Summary: Kept, but not as a core function. The signature is genuine and intact: ADNP2 carries a homeodomain at 1043-1102 detected independently by InterPro IPR001356, SMART HOX, PROSITE PRU00108 and SUPFAM Homeodomain-like, alongside nine C2H2 zinc fingers whose zinc-coordinating residues are conserved. GO:0003677 asserts only that the protein interacts selectively with DNA, with no claim of sequence specificity, so nothing in the ADNP2 literature contradicts it. What is missing is any positive measurement: no direct DNA-binding assay on ADNP2 has been published, and the one genome-wide occupancy study attributes ADNP2's chromatin distribution to HP1beta reading H3K9me3 and reports that no sequence motif explains it. Untested rather than refuted. Reason: A fold-derived generic term that survives on its own terms - the fold is present and the definition makes no specificity claim - but it cannot be presented as what ADNP2 does, because the protein's measured route to chromatin runs through a partner rather than through its own DNA contacts. Contrast the paralogue ADNP, where sequence-specific binding is measured and the corresponding term is core. Supporting Evidence: PMID:38960717 The domain architecture of ADNP2 resembles that of ADNP, with nine zinc fingers distributed in two N-terminal clusters and one C-terminal homeodomain. PMID:38960717 In line with this sequence-agnostic recruitment mechanism, we were unable to define a sequence motif that would broadly explain the distribution and specificity of the ADNP2 ChIP-seq signal. |
| GO:0000981 DNA-binding transcription factor activity, RNA polymerase II-specific | ISA GO_REF:0000113 | MODIFY | Summary: Over-specific in a way the definition makes decisive. GO:0000981 inherits from GO:0003700, whose definition requires modulating transcription "via selective and non-covalent binding to a specific double-stranded genomic DNA sequence (sometimes referred to as a motif) within a cis-regulatory region". Three independent observations from the only genome-wide study of ADNP2 occupancy run against each clause: the authors could not define any sequence motif explaining the ChIP-seq signal; the peaks are overwhelmingly in repeats with few at transcription start sites, that is, not at cis-regulatory regions; and recruitment tracks HP1beta binding H3K9me3, since mutating the PxVxL motif nearly abolishes chromatin binding. The zinc fingers reinforce the point - their zinc-coordinating residues are conserved with ADNP but the residues that set sequence specificity are not. The assignment route is a structural classification (TFClass 3.1.8, a homeodomain family), so this is a fold read as an activity, and the homeodomain is exactly the feature the measurement declines to convert into sequence-specific binding. Reason: ADNP2 genuinely controls transcription - deleting it raises retrotransposon RNA - so the right correction is to drop the sequence-specificity claim rather than the regulatory one. GO:0140110 transcription regulator activity is the ancestor that keeps what is measured and drops what is not; it is also the correct term while the mechanism is unresolved, which it is, the authors offering CHD4 remodelling as the likely but untested explanation. Verified against QuickGO that GO:0000981 is a descendant of both GO:0003700 and GO:0140110, so this is a generalisation within one branch and not a move sideways. The node-reach measurement makes this stronger than a one-gene objection: tfclass:3.1.8 reaches exactly 14 human gene products - ADNP, ADNP2, HOMEZ, TSHZ1/2/3, ZEB1/2, ZFHX2/3/4, ZHX1/2/3 - and gives every one of them the identical pair {GO:0000785, GO:0000981}, so the term is a property of being in the class rather than a judgement about ADNP2. Two of those genes cannot both be right here: ADNP's GO:0000981 is independently supported by a measured motif, and ADNP2's is contradicted by the failure to find one. Propagation Review Root cause: SOURCE WEAK OR INFERRED Failure modes: GRANULARITY MISMATCH FUNCTIONAL DIVERGENCE Sources checked: tfclass:3.1.8 SOURCE WEAK OR INFERRED TFClass 3.1.8 is a structural classification of homeodomain-bearing proteins, not an observation about ADNP2. Placing ADNP2 there is defensible as taxonomy; converting it into a sequence-specific RNA polymerase II transcription factor is the step the ChIP-seq data decline to support. The same node also supplies this gene's GO:0000785 row, which is correct - so the classification is not wrong, only over-read in the molecular-function direction. The whole GO_REF:0000113 import is 1436 annotations over 727 entities, 100% ISA, of which 709 receive the identical {GO:0000785, GO:0000981} pair - but 18 receive chromatin ALONE. That exclusion set is the pipeline's own negative control, and it shows the discrimination already exists. Proposed replacements: transcription regulator activity Supporting Evidence: PMID:38960717 In line with this sequence-agnostic recruitment mechanism, we were unable to define a sequence motif that would broadly explain the distribution and specificity of the ADNP2 ChIP-seq signal. PMID:38960717 Intriguingly, the majority of ADNP2 peaks were found in repeat regions, while few overlapped transcription start sites (TSSs) PMID:38960717 the residues that normally determine sequence specificity of zinc fingers vary between ADNP and ADNP2 PMID:38960717 Mutations in the PxVxL motif resulted in a near-complete loss of ChAHP2 binding to H3K9me3-modified target regions, while H3K9me3 levels remained unchanged file:human/ADNP2/ADNP2-bioinformatics/RESULTS.md `tfclass:3.1.8` reaches **14** human gene products (28 annotations). file:human/ADNP2/ADNP2-bioinformatics/RESULTS.md Of those entities, **709** receive `GO:0000785`+`GO:0000981`, **18** receive `GO:0000785` alone, and 0 carry some other signature. file:human/ADNP2/ADNP2-bioinformatics/RESULTS.md ADNP2's own feature is 1043-1102 (Homeobox). So the fold-symmetry precedent is **HMBOX1** β annotated with the identical note β and not HOPX, whose domain UniProt calls **degenerate** |
| GO:0000785 chromatin | ISA GO_REF:0000113 | ACCEPT | Summary: Accepted. This row arrives by the same TFClass classification as the GO:0000981 row above, and the two are being treated differently on purpose: the classification's premise, that ADNP2 is a chromatin-binding homeodomain protein, supports the location and does not support the sequence-specific activity. Chromatin association is directly observed - several thousand ADNP2 ChIP-seq peaks, all of the well-characterised targets being repeat chromatin. Reason: Correct conclusion on independent evidence, whatever the strength of the assignment route. Supporting Evidence: PMID:38960717 Peak calling revealed 6315 regions with significant ADNP2 enrichment over input chromatin PMID:38960717 In contrast to ChAHP, ChAHP2 does not bind euchromatic SINEs but is instead targeted to H3K9me3-modified LTR and LINE1 elements. |
| GO:0005634 nucleus | IEA GO_REF:0000044 | ACCEPT | Summary: Accepted, on the same evidence as the IBA row for this term. The UniProt subcellular-location keyword route is weak in itself - the entry's own SUBCELLULAR LOCATION line is ECO:0000305, a curator inference - but ADNP2 is a chromatin-bound component of a nuclear complex by direct observation, so the conclusion is right. Reason: Same term and same action as the IBA nucleus row, deliberately, since both reach a conclusion that direct measurement supports. Supporting Evidence: PMID:38960717 Peak calling revealed 6315 regions with significant ADNP2 enrichment over input chromatin |
| GO:0006357 regulation of transcription by RNA polymerase II | IEA GO_REF:0000108 | MODIFY | Summary: Directionally under-specified, and resting on an antecedent this review removes. GO_REF:0000108 is logical inference between ontologies, and the WITH/FROM names GO:0000981 - so this row is a mechanical consequence of the TFClass row that is being modified above, and it inherits that row's weakness rather than adding evidence. It nonetheless states something true: the retrotransposon families that ADNP2 silences are transcribed from LTR promoters by RNA polymerase II, and their RNA rises when ADNP2 is removed, so ADNP2 does modulate polymerase II transcription. That independent basis is what the term is kept on - and it is also directional, which the term as written is not. Every functional observation on ADNP2 or its orthologues is a de-repression: ADNP2 loss raises ERV and LINE1 RNA in mouse ES cells, and zebrafish adnp2 directly occupies and suppresses ved/vent/vox. Nothing reports ADNP2 activating a polymerase II gene. So the direction is not merely permitted by the data, it is the only direction the data show. Reason: Replaced by GO:0000122, which is both better supported and independent of the inference chain that produced this row. Two things follow from the change. First, direction: the cited evidence is a de-repression result, so the negative-regulation child states what was measured while the undirected parent discards it. Second, provenance: as written this row's only stated antecedent is GO:0000981, the term this review is modifying, so leaving it in place would leave a row hanging off a premise the same review rejects; GO:0000122 rests on the mouse and zebrafish loss-of-function data instead. Verified against QuickGO that GO:0000122 is a descendant of GO:0006357, so this is a refinement within the branch, and that it neither subsumes nor is subsumed by the NEW GO:0010526 row - the two are non-redundant statements, one about the polymerase II process and one about the transposon outcome. Proposed replacements: negative regulation of transcription by RNA polymerase II Supporting Evidence: PMID:38960717 In contrast, MMERVK10C derepression was exclusively ADNP2-sensitive despite also being weakly bound by ADNP PMID:41822989 Adnp2 functions as a transcriptional repressor, and directly occupies and suppresses the BMP-related ved/vent/vox homeobox genes |
| GO:0005515 protein binding | IPI PMID:21888893 Analysis of the human HP1 interactome reveals novel binding ... | MODIFY | Summary: Partner is P83916 CBX1_HUMAN, HP1beta, reviewed Swiss-Prot, 185 aa, matching the canonical entry. This is the most informative of the nine interaction rows because the study is a focused HP1 interactome rather than a proteome-scale screen: tandem affinity purification of the three human HP1 isotypes followed by mass spectrometry, annotating only nine entities in all of GOA. HP1beta is the physiologically preferred ADNP2 partner, recovered as one of the top two enrichments in ADNP2 immunoprecipitations, and full-length human ADNP2 and human HP1beta form a defined complex when reconstituted from purified protein. Reason: Bare protein binding says nothing. HP1 proteins are engaged through a PxVxL motif docking on the chromo shadow domain, and for ADNP2 this is established by mutation rather than by motif-spotting: PxVxL point mutants fail to bind HP1beta and lose chromatin occupancy. GO:0070087 records the surface actually used, and matches the term the merged ADNP review assigned to the equivalent rows on the paralogue, keeping the family consistent. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:21888893 Our analysis identified with high confidence about 30-40 proteins co-eluted with CBX1 and CBX3 PMID:38960717 As expected, wild-type ADNP2 was able to pull down HP1Ξ² and CHD4, whereas ADNP2PxVxL mutants were unable to bind HP1Ξ² above background levels |
| GO:0005515 protein binding | IPI PMID:21888893 Analysis of the human HP1 interactome reveals novel binding ... | MODIFY | Summary: Partner is Q13185 CBX3_HUMAN, HP1gamma, reviewed Swiss-Prot, 183 aa, matching the canonical entry. Same focused HP1 interactome as the CBX1 row above, which recovered ADNP2 with both CBX1 and CBX3. HP1gamma binding is real but secondary: in ADNP2 immunoprecipitations ADNP2 preferentially binds HP1beta over HP1gamma, the reverse of the preference reported for the ADNP-containing ChAHP complex. Reason: All HP1 paralogues present the same chromo shadow domain to a PxVxL ligand, so the informative molecular function is identical for CBX1 and CBX3; what differs is affinity, which the row cannot express and which is recorded here instead. Recorded separately from the CBX1 row so that the preference is not lost by collapsing partners. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:21888893 Our analysis identified with high confidence about 30-40 proteins co-eluted with CBX1 and CBX3 PMID:38960717 with the notable difference that ADNP2 preferentially binds HP1Ξ² over HP1Ξ³ |
| GO:0005515 protein binding | IPI PMID:24981860 Human-chromatin-related protein interactions identify a deme... | MODIFY | Summary: Partner is P83916 CBX1_HUMAN, HP1beta. This reference is a chromatin-focused affinity-purification survey annotating 314 distinct entities in GOA, so it is a screen rather than a dedicated study of ADNP2 - but it is a chromatin-restricted screen recovering a chromatin partner that four other independent references also recover, and that a purified-protein reconstitution confirms. Not an isolated hit. Reason: Same correction as the other HP1beta rows: replace an uninformative binding term with the domain actually engaged. Judged per partner rather than per reference, which the repo validator permits for GO:0005515. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 The bait protein (ADNP2) was strongly enriched in the ADNP2 IPs compared with control, copurifying CHD4 and HP1Ξ² as the top two significantly enriched interactors |
| GO:0005515 protein binding | IPI PMID:27705803 A High-Density Map for Navigating the Human Polycomb Complex... | MODIFY | Summary: Partner is P83916 CBX1_HUMAN, HP1beta, recovered in a Polycomb-complexome interaction map annotating 135 distinct entities. As with the other screens, the value of this row is corroborative rather than primary: it is one of five independent references placing ADNP2 with HP1beta. Reason: Replace with the domain-level term. The convergence of five references on the same partner, plus an in vitro reconstitution from purified human proteins, is what lifts this above screen noise. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 Finally, to directly confirm that ADNP2, HP1Ξ², and CHD4 form a stable complex, we performed in vitro reconstitutions. |
| GO:0005515 protein binding | IPI PMID:27705803 A High-Density Map for Navigating the Human Polycomb Complex... | MODIFY | Summary: Partner is Q13185 CBX3_HUMAN, HP1gamma, from the same Polycomb-complexome map. One of four independent references recovering the ADNP2-HP1gamma pair. Reason: Same domain-level replacement as the other HP1gamma rows. Kept as a distinct entry from the CBX1 row of the same reference because the GOA table carries them as distinct rows and the two partners are different proteins with different measured affinity for ADNP2. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 with the notable difference that ADNP2 preferentially binds HP1Ξ² over HP1Ξ³ |
| GO:0005515 protein binding | IPI PMID:32296183 A reference map of the human binary protein interactome. | MODIFY | Summary: Partner is Q13185 CBX3_HUMAN, HP1gamma, from HuRI, a systematic yeast two-hybrid reference interactome. This is the weakest single line of evidence on the gene - a binary screen with both partners over-expressed, and one contributing 85343 annotations across GOA, so a reference-projection test on it is uninformative by construction. It is retained because it is a binary assay agreeing with four affinity-purification references on the same partner, which is orthogonal-method agreement rather than replication of one method. Reason: Same replacement. The screen's weakness argues for not resting anything on this row alone, not for discarding a partner that four other references independently support. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 with the notable difference that ADNP2 preferentially binds HP1Ξ² over HP1Ξ³ |
| GO:0005515 protein binding | IPI PMID:33961781 Dual proteome-scale networks reveal cell-specific remodeling... | MODIFY | Summary: Partner is P83916 CBX1_HUMAN, HP1beta, from BioPlex affinity-purification mass spectrometry in two cell lines. BioPlex contributes 9514 annotations to GOA, so an entity count from it is not a meaningful independence test; the row's weight comes from agreeing with the focused HP1 interactome and with the reconstitution. Reason: Same replacement, on the aggregate evidence rather than on this reference. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 The bait protein (ADNP2) was strongly enriched in the ADNP2 IPs compared with control, copurifying CHD4 and HP1Ξ² as the top two significantly enriched interactors |
| GO:0005515 protein binding | IPI PMID:33961781 Dual proteome-scale networks reveal cell-specific remodeling... | MODIFY | Summary: Partner is Q13185 CBX3_HUMAN, HP1gamma, from BioPlex. Fourth independent reference for this pair. Reason: Same replacement as the other HP1gamma rows. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 with the notable difference that ADNP2 preferentially binds HP1Ξ² over HP1Ξ³ |
| GO:0005515 protein binding | IPI PMID:35271311 OpenCell: Endogenous tagging for the cartography of human ce... | MODIFY | Summary: Partner is P83916 CBX1_HUMAN, HP1beta, from OpenCell, which tags proteins at their endogenous loci and so avoids the over-expression objection that applies to the two-hybrid and transfection-based references. Fifth independent reference for the ADNP2-HP1beta pair, and the one closest to native stoichiometry. Reason: Same replacement. Worth noting separately that this is endogenous-level evidence, which is why the HP1beta partnership is treated as established rather than as a screen artefact. Proposed replacements: chromo shadow domain binding Supporting Evidence: PMID:38960717 As expected, wild-type ADNP2 was able to pull down HP1Ξ² and CHD4, whereas ADNP2PxVxL mutants were unable to bind HP1Ξ² above background levels |
| GO:0007399 nervous system development | IEA GO_REF:0000107 | KEEP AS NON CORE | Summary: Kept, but not as a core function, and the route is thinner than it looks. This is the gene's only Ensembl Compara row, and its donor Q8CHC8 is mouse Adnp2 - the true one-to-one orthologue, not the ADNP paralogue - so no ADNP-specific or peptide-derived material enters here. What the donor holds for this term, however, is a single IEP from PMID:18179478, an inference from expression pattern: mouse Adnp2 mRNA is enriched in embryonic brain and rises when P19 cells differentiate. That is an expression correlation transferred across species, the weakest experimental code supporting the whole chain. The conclusion is nonetheless supported from elsewhere: zebrafish adnp2a/adnp2b mutants show reduced pan-neuronal markers, and adnp2 loss impairs neural induction through direct repression of ved/vent/vox. Reason: A real organismal role for the ADNP2 orthologue group, but not what the human protein does at the molecular level, and nothing in the human literature addresses it. Retained rather than removed because independent loss-of-function data in a second vertebrate support the conclusion the IEP only gestures at. Propagation Review Root cause: NO FAILURE NON CORE Sources checked: UniProtKB:Q8CHC8 SUPPORTS TRANSFER Mouse Adnp2, reviewed (Swiss-Prot), 1165 aa. The genuine orthologue. Holds seven annotations in total; for this term it holds GO:0007399 IEP (PMID:18179478) and nothing stronger. ensembl:ENSMUSP00000068560 SUPPORTS TRANSFER The Ensembl protein identifier for the same mouse Adnp2 entry; not an independent source. Supporting Evidence: PMID:18179478 During development, a relatively high level of ADNP2 gene expression was found in the embryonic mouse brain PMID:41822989 we show that ADNP2 orthologs adnp2a and adnp2b are required for proper neural induction in gastrulating embryos |
| GO:0007399 nervous system development | ISS GO_REF:0000024 | KEEP AS NON CORE | Summary: The manually curated counterpart of the Compara row above, from the same mouse Adnp2 donor and resting on the same single IEP. Treated identically: the term is defensible for the orthologue group but is not a molecular function of human ADNP2, and the human protein has never been tested in a neural context. Reason: Same term and same donor as the Ensembl row, so the same action, deliberately. The WITH/FROM here correctly names only the sequence-similar entity, which is what ISS requires. Propagation Review Root cause: NO FAILURE NON CORE Sources checked: UniProtKB:Q8CHC8 SUPPORTS TRANSFER Mouse Adnp2, reviewed (Swiss-Prot), 1165 aa; the one-to-one orthologue of human ADNP2. Its GO:0007399 is IEP from PMID:18179478. Supporting Evidence: PMID:18179478 An increase in the mRNA was detected in differentiated P19 neuronal/glial-like cells as compared with the non-differentiated cells. |
| GO:0010526 transposable element silencing | ISS PMID:38960717 ChAHP2 and ChAHP control diverse retrotransposons by complem... | NEW | Summary: Proposed addition. ADNP2's best-characterised biological role has no representation in its GO record at all: the gene's only biological-process rows are a general regulation-of-gene-expression IBA and a nervous-system-development transfer. Ablating Adnp2 in mouse embryonic stem cells de-represses specific ERV and LINE1 families, with MMERVK10C responding to ADNP2 alone and L1MdA only when ADNP is removed as well, and ADNP2 ChIP-seq places the protein on exactly those elements. Proposed as ISS from mouse Adnp2 rather than as a direct annotation, because the loss-of-function experiments are in mouse; the human protein contributes only the in vitro complex reconstitution. Reason: GO:0010526 and not its child GO:0141005 transposable element silencing by heterochromatin formation, which would be the reflexive choice and is wrong on the definition: GO:0141005 requires the silencing mechanism to involve heterochromatin assembly, and H3K9me3 at ADNP2 sites is unchanged by single or combined removal of ADNP and ADNP2. ChAHP2 reads the heterochromatin mark and does not deposit it, so the parent term is the most specific one the evidence permits. The absence of a term for reader-mediated, assembly-independent silencing is recorded as an ontology gap. Supporting Evidence: PMID:38960717 In contrast, MMERVK10C derepression was exclusively ADNP2-sensitive despite also being weakly bound by ADNP PMID:38960717 H3K9me3 levels at ADNP and ADNP2 peaks were not reduced by either individual or combined removal of these factors PMID:38960717 In contrast to ChAHP, ChAHP2 does not bind euchromatic SINEs but is instead targeted to H3K9me3-modified LTR and LINE1 elements. |
Loading supporting contentβ¦
Download this section (compressed HTML)Q: For PAINT, first of two. PANTHER node PTN000405125 spans 44 gene products covering both the ADNP and ADNP2 orthologue clades and propagates two terms, GO:0005634 and GO:0010468. One of the two IBD seeds named for GO:0010468 is rat Adnp (RGD:71030), whose only experimental annotation in that term's subtree is GO:0010629 IDA citing PMID:15314252, a paper whose subject is the synthetic ADNP-derived octapeptide NAPVSIPQ and not the Adnp gene product. Should a seed annotation derived from a synthetic peptide fragment count toward an IBD? The row survives here only because the co-seed, mouse Adnp, carries a genuine gene-product IMP for the same term.
Suggested experts: GO Central / PAINT curators
Q: For PAINT, second of two, and stated once here for both genes rather than repeated on each. PTN000405125 is the ONLY node in PTHR15740 annotated by PAINT, so human ADNP and human ADNP2 receive byte-identical IBA rows - same two terms, same node, same WITH/FROM strings - and the IBA record cannot distinguish the paralogues at all. Two consequences. (1) The containment of the peptide-derived material to ADNP is not a fine-grained PAINT judgement about the paralogues; it follows from Ensembl Compara projecting along orthologues, while PAINT simply annotated the shared ancestor generically. (2) It leaves a coverage gap in both directions: ADNP's sequence-specific DNA binding and ADNP2's HP1-mediated heterochromatin targeting are each well characterised and clade-restricted, and neither reaches its own clade. The concrete suggestion is that GO:0070087 chromo shadow domain binding is a good candidate for the FAMILY node rather than a clade node - the PxVxL motif is conserved between the two paralogues and across vertebrates, ADNP and ADNP2 both bind HP1, and this review and the merged ADNP review independently assigned that same term - whereas the DNA-sequence-specificity and the euchromatic-SINE versus heterochromatic-ERV target split are exactly the things that should NOT be placed at the shared node, since the two halves demonstrably diverge there.
Suggested experts: GO Central / PAINT curators
Q: For NTNU_SB and TFClass. Would you consider adding ADNP2 to the GO_REF:0000113 exclusion set - the 18 entities that already receive GO:0000785 without GO:0000981? ADNP2's GO:0000981 arrives by ISA from tfclass:3.1.8, whose reach is 14 human zinc-finger/homeodomain proteins that all receive the identical term pair. But the only genome-wide study of ADNP2 occupancy reports that no sequence motif explains its ChIP-seq signal, that its peaks are overwhelmingly repeats with few at transcription start sites, and that chromatin binding is nearly abolished by mutating its HP1-docking motif; GO:0003700 requires binding a specific sequence within a cis-regulatory region. Two precedents already inside the import, and they do different work. HMBOX1 (Q6NT76) carries a DNA_BIND feature annotated "Homeobox" at 267-341 - the identical note to ADNP2's "Homeobox" at 1043-1102, i.e. an intact domain, not a degenerate one - and it receives GO:0000785 without GO:0000981. That is the fold-symmetry precedent. HOPX (Q9BPY8) carries the biological one: UniProt describes it as an "Atypical homeodomain protein which does not bind DNA" and it too is excluded - but its DNA_BIND note reads "Homeobox; degenerate", so it does NOT establish that an intact domain can be excluded, and citing it alone would invite exactly that reply. Every figure in this question is reproducible from the committed ADNP2-bioinformatics/analyze_adnp2_propagation.py, which emits the node's gene list and the chromatin-only exclusion set as sets rather than counts, so the 18 members are diffable rather than a number to be taken on trust; it also asserts that HOPX really is in that set, that both proteins really do carry a DNA_BIND feature, and that at least one excluded entity has a NON-degenerate domain, refusing to report if any of those is false. To be explicit about what carries the ask: the positive argument for ADNP2 is the measured failure to find a motif, not symmetry of fold with any excluded protein - the exclusion set only shows that this import has a mechanism for withholding GO:0000981 and already applies it to an intact-domain homeodomain protein. Worth adding that this repo already adjudicates HMBOX1: projects/TRANSCRIPTION_FACTORS/dbTF-discrepancy-analysis.md lists it among proteins "correctly excluded", with the reason "Telomere binding, not gene regulation", under the heading DBD != dbTF. So HMBOX1 is not excluded for failing to bind DNA - it does bind DNA, at telomeric repeats - but for not being a DNA-binding transcription factor. That is a closer fit to ADNP2 than "does not bind DNA" would be, and it is why GO:0003677 is retained here while GO:0000981 is not: the homeodomain and zinc fingers are intact and unrefuted, and what the ChIP-seq refutes is specifically the sequence-specific polymerase II transcription-factor reading. Finally, the ask needs no new mechanism. The import operates at two granularities - of the 11 nodes reaching an excluded entity, 7 withhold GO:0000981 from a strict subset while the rest of the node keeps it, and 4 withhold it from every member - and the one ADNP2 needs is the first, which class 3.1 already supplies twice: HOPX is excluded alone out of the 47 members of tfclass:3.1.3, and HMBOX1 alone out of the 19 members of tfclass:3.1.10, while ADNP2's own node tfclass:3.1.8 stands at 14/14. So this would take one node from 14/14 to 13/14 and would not touch ADNP, whose sequence-specific binding is measured and whose GO:0000981 should stay. That granularity is measured in the committed script rather than assumed, because had it come out the other way the correct request would have been a different one. The GO:0000785 assignment on ADNP2 is correct and should be kept either way.
Suggested experts: NTNU_SB curators
Q: For UniProt. The Q6IQ32 entry's FUNCTION and SUBUNIT lines are entirely ECO:0000250 by-similarity statements from mouse Adnp2 and predate the ChAHP2 work, so the entry does not mention HP1beta, CHD4 or retrotransposon silencing, and its SIMILARITY line places ADNP2 only in the krueppel C2H2-type zinc-finger protein family. The entry also still carries GO:0003677, GO:0008270 and GO:0006351 as IEA:UniProtKB-KW keyword mappings, a route GOA no longer imports, so the keyword-derived claims are now visible only in UniProt.
Suggested experts: UniProt curators
Q: For GO. Is a term needed for silencing that is targeted by, but does not establish, heterochromatin? GO:0141005 transposable element silencing by heterochromatin formation is the obvious term for ChAHP2 and is wrong for it: H3K9me3 is unchanged when ADNP2 is removed, so the complex reads the mark rather than depositing it. The only correct available term is the bare parent GO:0010526, which loses the mechanism entirely.
Suggested experts: GO ontology editors
Experiment: Deplete ADNP2 in a human cell line and sequence the repeat transcriptome. Every functional claim about ADNP2 currently comes from mouse ES cells or zebrafish; the human protein has only ever been assayed biochemically. Human ERV and LINE1 families differ substantially from mouse, so whether ChAHP2's target set is conserved is an open question, not a formality. Pair with an ADNP double depletion, since the mouse data show that some elements are de-repressed only when both paralogues are lost.
Experiment: Test whether ADNP2 contacts DNA. An in vitro binding assay with purified full-length human ADNP2 - already produced in this system - against naked DNA, and a SELEX or protein-binding-microarray screen on the isolated homeodomain and zinc-finger clusters, would settle whether GO:0003677 is supportable on measurement rather than on the fold. A negative would be as informative as a positive and would justify removing the term.
Experiment: Ask whether ADNP2 is required to bridge HP1beta and CHD4, or merely co-resident with them. The published reconstitution mixed ADNP2/HP1beta with separately expressed CHD4, which does not exclude a direct HP1beta-CHD4 contact. Reconstituting HP1beta with CHD4 in the absence of ADNP2 is the missing control, and it is what would convert the scaffold description into an adaptor activity annotation.
Experiment: Determine whether ADNP2's second PxVxL candidate, PPVLV at 662-666, is functional. The C-terminal PSVLL at 1107-1111 is the one that aligns to ADNP's motif and the one mutated in the published work; the internal candidate has no ADNP counterpart. A double mutant would show whether HP1 binding is fully accounted for by the C-terminal site.
What is not known β curated, literature-grounded statements of the open unknowns (the inverse of core functions).
Gap: What ADNP2 itself contributes catalytically or allosterically inside ChAHP2 is unknown. Its established roles are structural - it holds HP1beta and CHD4 together and is dispensable for the H3K9me3 mark - and the repressive step is attributed to CHD4 remodelling on the strength of elimination rather than of a direct test. ADNP2 may be a pure scaffold.
OPEN BIOLOGY MF_DARK
What is known: Known: ADNP2 binds HP1beta through PxVxL and co-assembles with CHD4 from purified human protein; ADNP2 loss de-represses ERVs and LINE1s. Unknown: whether ADNP2 does anything to chromatin beyond positioning the other two subunits.
Provenance (the field's own admissions):
Gap: Whether ADNP2 binds DNA at all has never been measured. It carries a homeodomain and nine C2H2 zinc fingers with intact zinc-coordinating residues, so GO:0003677 is annotated from the fold, but no binding assay has been reported, the residues that would set zinc-finger specificity differ from ADNP's, and no sequence motif explains its genome-wide occupancy. The generic term is unrefuted; the specific one is contradicted.
OPEN BIOLOGY MF_DARK
What is known: Known: the DNA-binding folds are present and detected by four independent signatures; chromatin occupancy is HP1beta-dependent and motif-free. Unknown: whether the homeodomain or the zinc fingers contact DNA in any context, for instance at the minority of ADNP2 sites that are not repeats.
Provenance (the field's own admissions):
Gap: Two experimentally reported human ADNP2 interactions are absent from GOA. CHD4 co-purifies with ADNP2 as one of the top two enrichments and forms a defined complex with purified full-length human ADNP2 and human HP1beta, yet no GO:0005515 row names it; SMARCA4/BRG1 was reported to interact with ADNP2 and reaches UniProt only as a by-similarity SUBUNIT line. All nine of ADNP2's interaction rows name an HP1 paralogue.
OPEN CURATION
What is known: Known: both interactions are in the primary literature. Unknown to GOA: neither is annotated, so the ChAHP2 complex cannot be reconstructed from ADNP2's GO record.
Provenance (the field's own admissions):
Gap: GO has no term for the ChAHP2 complex, nor for ChAHP, and no term for silencing that is targeted by heterochromatin without establishing it. ADNP2's defining biological identity therefore cannot be recorded as complex membership, and its silencing role can only be recorded at the bare parent GO:0010526, since the obvious child GO:0141005 asserts a heterochromatin-assembly mechanism the data exclude.
OPEN ONTOLOGY
What is known: Known: composition (ADNP2 plus HP1beta plus CHD4), reconstituted from purified human proteins with defined stoichiometry by SEC-MALS; silencing is downstream of an H3K9me3 mark ADNP2 does not place. Missing: a cellular-component term for the complex, and a biological-process term for reader-mediated silencing.
Provenance (the field's own admissions):
Gap: The whole functional description of ADNP2 rests on mouse embryonic stem cells and zebrafish. Human ADNP2 has never been depleted in a human cell line, so whether it silences human retrotransposons, and which ones, is untested; correspondingly the human gene has no experimental annotation to any biological process.
NARROWING BIOLOGY BP_DARK
What is known: Known for human: the protein exists (PE 1), is nuclear, and assembles with HP1beta and CHD4 in vitro from human sequences. Unknown for human: every functional consequence.
Provenance (the field's own admissions):
Loading supporting contentβ¦
Download this section (compressed HTML)Loading supporting contentβ¦
Download this section (compressed HTML)Loading supporting contentβ¦
Download this section (compressed HTML)Loading supporting contentβ¦
Download this section (compressed HTML)