The paper that named the gene, reporting hArpM1 (ACTRT3) and hArpM2 (ACTRT2) together. Any ARPM-keyed search therefore returns both genes and the M1/M2 suffixes must be resolved before a claim is attributed to either.

A sequence-level prediction that the nucleotide site is preserved, made in 2001 and confirmed by measurement in this review: 17 of 19 computed ATP/cation contacts of beta-actin are chemically compatible in ACTRT3, the joint-best score among the eight divergent human actin-like and actin-related-T proteins.

The sole evidence behind UniProt's "Ubiquitously expressed" tissue-specificity line. It is an mRNA survey with an explicit abundance caveat, and the same UniProt entry's own cross-references record HPA "Tissue enriched (testis)".

Establishes the mouse orthologue as testis-exclusive and confined to haploid germ cells, which is the protein-level counterpart to the transcript survey in PMID:11750065 and the main reason to read UniProt's "ubiquitously expressed" line narrowly.

The source of the male-germ-cell-nucleus IDA that reaches human ACTRT3 by Compara transfer.

The experiment behind the ARP-T3-profilin interaction. GOA records it on the mouse entry as a bare GO:0005515 protein binding IPI with WITH/FROM UniProtKB:Q9DAD6, which resolves to PROF3_MOUSE (Pfn3, 137 aa, Swiss-Prot), so the partner is a profilin and the specific term GO:0005522 profilin binding is available on both the mouse and the human side.

The authors' own framing of the complex's role, stated as a suggestion rather than a demonstration.

Reciprocal genetic evidence for the PFN3-ARPM1 partnership: in the Pfn3 knockout, ARPM1 is undetectable in the nuclear fraction. This is a stability dependency, not merely co-purification, and it is the second independent strand supporting profilin binding.

The same result confirmed by a second method, immunohistochemistry.

The complex is compartment-restricted, which matters for the nuclear-localisation rows: it is detected in the sperm nucleus and not in the cytoplasm.

The result that argues against refining the cytoskeleton-organization row to an actin-specific child. In the one in vivo setting where ARPM1 is depleted from the nuclear fraction, F-actin organisation was not altered. This does not contradict the Actrt3 knockout (different compartment, indirect depletion, and cytoplasmic ARPM1 persists) but it does mean the actin-remodelling link is not robust enough to license a specificity upgrade.

The qualification that keeps the preceding result from being over-read: ARPM1 is not absent from Pfn3-null sperm, only from the nuclear fraction.

The reference cited by the mouse Actrt3 GO:0033011 perinuclear-theca IDA, and therefore the original basis for the two perinuclear-theca rows on human ACTRT3. The paper is titled for CCIN/Calicin, a different gene, and reports interactions among many perinuclear-theca components; the Actrt3 localisation call was made by a curator from its full text, which is not in the cache.

The dedicated knockout study. Confirms the perinuclear-theca localisation independently of PMID:35793634.

The loss-of-function phenotype, which is entirely absent from the current GO record for this gene and is the basis for the proposed GO:0001675 acrosome assembly row.

The mechanism behind the acrosome defect, and the reason GO:0001675 (defined as formation of the acrosome from the spermatid Golgi) fits the phenotype's mechanism and not only its outcome.

Places ACTRT3 in a perinuclear-theca protein assembly with ACTRT1, ACTRT2, ACTL7A and SPEM2 - i.e. its own clade plus another divergent actin-like theca protein - rather than with conventional actin.

The authors' scaffold interpretation, which is what a structural-molecule-activity term expresses.

The actin-remodelling evidence, from over-expression in a heterologous cell line and hedged by the authors. Weighed against the contrary in vivo observation in PMID:34869336, it is not sufficient to refine the cytoskeleton-organization row to an actin-specific child.

Co-enrichment of two actin regulators, a further but indirect strand of the actin link.

The authors' overall conclusion, which includes a contribution to perinuclear-theca structure - the fact that makes GO:0005200 arguable on grounds unrelated to the phylogenetic transfer that actually carries it.

BioPlex 3.0, the source of the single GO:0005515 row. IntAct returns exactly two records for Q9BYD9 and both are this one anti-tag co-immunoprecipitation with PDCL3 (MI score 0.5), one of them a spoke expansion, so UniProt's NbExp=2 is one experiment counted twice.

The human PFN3 interaction is recorded only by similarity and names no source entry, which is why the proposed profilin-binding row uses ISS with the mouse orthologue as its supporting entity rather than treating this line as evidence in itself.

The tissue-specificity line that a merged sibling review quotes as evidence the divergent actin clade is not uniformly testis-restricted.

The same entry's own cross-reference, which says the opposite. HPA's API gives a single tissue above its specificity threshold, testis at 54.1 nTPM.

A family-specific CDD signature for the ARP-T1/2/3 clade rather than a bare actin fold, which is a materially better-founded starting point for a domain-derived inference than a generic fold hit.

The protein is detected, so any statement of absent data here must be scoped to functional and biochemical data rather than to evidence of existence.

GO:0005200 is asserted at exactly one node in this family, PTN000940351, the conventional-actin node, from ten seeds. Every other GO:0005200 row in the family is a negation of this one.

The nearest neighbour clade to ACTRT3's, PTN008986528, which covers ACTL7A and ACTL7B, carries the negation of GO:0005200 against that same node. This is the row the MODIFY on this gene follows.

And on the same day PAINT asserted the general parent GO:0005198 at that node in the specific term's place, which is why ACTL7A and ACTL7B hold structural molecule activity by IBA today. Substituting the parent rather than leaving the clade with no molecular function is the precedent this review applies to ACTRT3.

A second batch of rejections landed on 2026-04-16, reaching the Arp2, Arp3 and dynactin Arp1 clades, so the sweep is ongoing rather than finished. The ARP-T branch has not been reached.

The provider's own summary of the gene, cited only because every clause of it was independently checked against the abstract of PMID:41668650 and against HPA. It is used as corroboration of location and process, never for a mechanistic detail.

The nucleotide site is essentially intact, so a "fold without function" reading is refuted for nucleotide binding. Controls in the same run: beta-actin and Drosophila Arp53D 19/19 compatible, Arp1/ACTR1A 18/19, Arp11/ACTR10 11/19.

The filament protomer interface is not intact. Every panel member known to assemble into a two-stranded actin-like filament scores far higher - conventional actins 38/38, the divergent but polymerising Arp53D 33/38, and the dynactin Arp1 paralogues 28/38 - so ACTRT3 sits ten below the weakest true polymeriser.

The profilin-binding surface of beta-actin is largely retained, a directed structural prediction for the one interaction the gene has ortholog-level experimental support for. Comparators in the same run: beta-actin 21/21, Arp53D and Arp1 20/21, ACTL8 10/21, ACTR10 8/21.

Within the eight divergent human actin-like and actin-related-T proteins, all scored in one run, ACTRT3 has the joint best-preserved nucleotide site. The ARP-T trio leads the ACTL set.

On the filament interface the same clade is uniformly poor, and ACTRT3 sits above ACTL8's 11, the score that with three additional deletions earned ACTL8's REMOVE. Retaining the nucleotide core while losing the protomer interface is the signature of a monomeric nucleotide-binding scaffold rather than a polymer subunit.

The profilin surface is retained across the ARP-T trio specifically, which is consistent with the profilin interaction being a clade property rather than a one-off.

GO:0005200 is asserted once in PTHR11937 and then explicitly negated at eight descendant nodes covering thirteen human genes, in two dated batches (2025-08-05 and 2026-04-16). ACTRT3's own path carries no such negation, so its row is a survivor of an ongoing sweep rather than an adjudicated decision.

At the nearest neighbour clade PAINT did not merely reject the specific term, it substituted the general parent GO:0005198 on the same day. That is the in-resource precedent for the MODIFY proposed on this gene, and it is why ACTL7A and ACTL7B hold GO:0005198 by IBA while ACTRT3 still holds GO:0005200.

The mouse source of every experimental location on this gene is a genuine 1:1 orthologue, not a paralogue: 79.4 per cent identical to human ACTRT3 against 47.0 per cent to beta-actin.

The mouse orthologue carries a protein-binding IPI whose partner resolves to profilin-3, and the human GOA has no counterpart row. This is the informative molecular function missing from the human record.

ACTL8's published census is reproduced independently here, including the figure it assigned ACTRT3. ACTRT3 sits under neither narrow beta/gamma-actin node, so the roughly five-fold record inflation ACTL8 documented does not affect this gene.

The perinuclear-theca reference annotates 19 entities across 3 terms, and every one of its rows was returned - 35 examined against 35 total. Row completeness, not term-list closure, is the precondition for asking the projection question, because the per-term entity counts are collected from the returned rows: a term list can close while the rows are still a sample, and a distinct-entity count taken from a sample is a lower bound rather than a total.

Within that reference, GO:0033011 reaches 12 distinct mouse proteins while the only functional row, GO:0007286, reaches exactly one - the gene that was knocked out. A complex-to-subunit projection spreads the phenotype across the whole set; this one does not, so the twelve localisation calls are twelve observations rather than one written twelve times.

The profilin reference covers only 2 entities in total, ACTRT3's orthologue and its partner, and is likewise complete. That is how the apparent duplicate IPI resolves into one co-immunoprecipitation logged by two databases and recorded reciprocally on both partners - note the assigning databases are UniProt and IntAct, not one source twice.

The BioPlex reference behind the single GO:0005515 row accounts for 9514 annotations, of which 9508 are protein binding itself. It is also the one reference here that a single page cannot cover, and the summary says so: 200 rows examined, TRUNCATED, 1 annotation still unaccounted. Its assigning databases are ComplexPortal 5 and IntAct 9509 - so this reference does contain a small complex-to-subunit projection, invisible in the sampled page, which is why the counts quoted in this review come from filtered count queries rather than from that page.

The perinuclear-theca rows trace to mouse Actrt3, ACTT3_MOUSE, which the table shows carries the term itself. Built from the GOA WITH/FROM field with an assertion on the token count, so the source lists in this review cannot drift from GOA.