ACTRT2 (Q8TDY3, ARP-T2, ARPM2) — review notes

Human actin-related protein T2. 377 aa, Swiss-Prot, PE 1: Evidence at protein level,
HPA "Tissue enriched (testis)", Pharos Tdark. UniProt carries no FUNCTION comment, and
the only subcellular-location line is Cytoplasm, cytoskeleton {ECO:0000250} — a
by-similarity statement with no source entry cited.

Identity and naming

UniProt's own entry settles the synonym question, which matters because two papers name the
protein differently:

So hArpM2 = ACTRT2 and hArpM1 = ACTRT3 (ARPM1). Harata et al. 2001 is therefore a paper
about this gene and its close paralogue, even though it never uses the symbol ACTRT2. (The
task brief glossed ACTRT2 as "ARPM1"; UniProt says ARPM1 is ACTRT3, and the affinage narrative
agrees, writing "ACTRT3 (ARPM1/ACTRT3)".)

What is actually known

1. It is a bulk structural component of the sperm cytoskeletal calyx (founding paper)

Heid et al. 2002 purified the calyx from bull sperm and found ARP-T1/ARP-T2 as major
components — not trace constituents:

Two things to be careful about here. The calyx preparations were from bull; the antibodies
were raised against the human proteins, but the abstract does not state which species was used
for immunofluorescence, so the human localisation claim rests on ortholog transfer rather than
on a stated human observation in this paper. And the calyx's stated composition is
"basic proteins calicin, cylicin I and II, and two major actin-capping proteins"
PMID:12243744
— the Arp-Ts being the new additions to that inventory.

The authors' own reading of their result points away from actin-like behaviour, and this is
worth quoting because it is the opposite of what a fold-based annotation would assume:

2. It is a member of a multimeric perinuclear-theca ARP complex (mouse)

ACTRT3 was added to the complex four years later, again by co-immunoprecipitation:

Note the subacrosomal location here, versus the calyx (posterior) location in Heid 2002.
Both are the perinuclear theca; they are different subdomains of it. Calicin behaves the same
way — first around the acrosome, later postacrosomal
PMID:35793634
— so redistribution during spermiogenesis is the norm for this protein set, not a discrepancy.
This is the reason GO:0033011 perinuclear theca is the correct term and not an
under-specification: it is the least common ancestor of ACTRT2's two reported subdomains.

3. It interacts with calicin, the PT organising centre (mouse)

Mouse Actrt2's own GO:0033011 IDA and its GO:0005515 IPI both come from the calicin paper
(QuickGO, UniProtKB:Q9D9L5; the IPI partner resolves to Q8CDE2 = mouse Calicin/Ccin):

4. Human sperm localisation and an abundance change in asthenozoospermia

This is the only human protein-level localisation for ACTRT2. Note that "middle piece" is
not a perinuclear-theca compartment, so the human immunofluorescence reports a PT pool and
a flagellar-midpiece pool. The paper is a differential proteomics study, so the abundance
finding is correlative.

5. A mouse knockout with a spermatogenesis phenotype, and a ferroptosis claim

Treat the ferroptosis mechanism cautiously. It is a single paper; the in vivo busulfan
comparison uses heterozygotes, not nulls; and the mechanism is read out as expression
changes in ferroptosis regulators (ACSL4, ALOX15 up; SLC7A11, GPX4 down) rather than as a
direct activity of ACTRT2. It is also biologically odd for a perinuclear-theca protein: the
PT forms in spermatids, whereas the ferroptosis phenotype is in spermatogonia, several
stages earlier. The shrunken-tubule phenotype of the null is the solid part.

6. The 2001 cloning paper, and a claim worth testing rather than repeating

Two notes. First, the apparent conflict with testis specificity is not a conflict: Harata
measured mRNA ("Their mRNAs are expressed in all tested human tissues, but in smaller
amounts"), Heid measured protein synthesis, and HPA's call is tissue-enriched rather than
tissue-exclusive. Low ubiquitous transcript with testis-restricted protein is consistent with
all three. Second, the closing sentence is an explicit speculation from sequence
inspection, made before any localisation was known, and it is the intellectual ancestor of the
GO:0007010 cytoskeleton organization row. The ATP-binding claim, by contrast, is testable —
see below, where it is confirmed.

Bioinformatics (see ACTRT2-bioinformatics/RESULTS.md)

All numbers are computed by analyze_actrt2.py from live UniProt/RCSB/QuickGO/IntAct plus the
repository's cached PANTHER PAINT table; a fresh run reproduces RESULTS.md byte-identically.

The nucleotide pocket is largely retained — Harata 2001 was right where it matters

Contacts computed from PDB 2BTF (profilin:beta-actin, ATP + Sr in the Mg site), 19 contact
positions. Whole contact set first, so no sub-selection can flatter the result: ACTRT2 has
13 identical, 4 conservative, 2 non-conservative, 0 gaps; unchanged under a second
substitution matrix and gap model.

Split by role, which is where the interesting asymmetry is:

group positions id cons non-cons substitutions
phosphate loops, cation site, sensor D11, S14, G15, K18, Q137, D154, G156, D157, V159, R183 10 0 0 none
adenine/ribose region E214, Y306, K336 0 1 2 E214→K, Y306→F, K336→W

A correction to an earlier draft of this review, caught by the reviewer. That draft reported
only the first row and called the pocket "fully retained". The number was true but selectively
bounded: the ten-position set omitted E214, Y306 and K336 — three of the script's own computed ATP
contacts, and exactly the three where ACTRT2 differs — while including D11, D154 and R183, which
the same computation places outside the 4 Å contact set. The set was, in effect, selected so that
"all ten identical" would follow. Both groups and their contact-set membership are now printed by
the script, so the sub-selection cannot recur silently. This is the brief's "a verbatim quote can
be TRUE and selectively bounded" failure committed in a number rather than a quote.

The honest statement loses nothing: every phosphate-binding-loop residue, the cation ligand D154,
the catalytic-water glutamine Q137 and the sensor R183 are identical to actin, while the
adenine/ribose region has diverged. Harata 2001's ATP-binding-motif claim is confirmed where it
matters. But a retained pocket means the question is untested, not answered: no structure, no
ligand and no binding measurement exists for ACTRT2, so no nucleotide-binding term is proposed. Worth noting
that this gap is family-wide rather than gene-specific — QuickGO returns zero GO:0005524
annotations for beta-actin itself (P60709), which is the same observation the merged ACTR10
review made.

The ATP-hydrolysis trigger His161 is lost

Actin's ATPase is coupled to filament incorporation, and the coupling runs through His161:

ACTRT2 has H161→C — the trigger itself. It has also lost both Pro-rich loop residues
(A108→P, P109→S), but a second correction from the reviewer applies here, and reading the cached
full text confirms it: the actin mutants A108G and P109A
PMID:37009486
and
PMID:37009486
— so substituting those residues does not abolish polymerisation or hydrolysis. The Pro-rich
loop modulates the His161 rotamer; it does not gate hydrolysis. The earlier framing, which
grouped all three residues together as jointly coupling filament incorporation to hydrolysis,
overstated two of them. His161 carries the argument alone, and that is how it is now stated. Same failure shape as the ACBD3
proline lesson: a substitution whose actin counterpart has no effect cannot be cited as evidence
of lost function.

The census across the panel is the interesting part, and it is computed rather than asserted:

The merged ACTL7A review found His161→Tyr in that one gene and noted ACTL7B retains it; the
contribution here is turning that into a clade-level convergent loss with a control.

The filament interface is not intact either — and this is the same finding, not a second one

38 protomer-protomer contact positions computed from PDB 6DJO (cryo-EM ADP-F-actin). ACTRT2:
14 identical, 6 conservative, 18 non-conservative, 0 gaps; D-loop 2/10 identical. Stable under
both alignment schemes.

The D-loop motif is where the comparison discriminates (actin P38 R39 H40 Q41 G42 V43 M44 V45
M47 Q49):

protein motif id/10
ACTB / ACTG1 / ACTA1 / ACTC1 PRHQGVMVMQ 10/10
Arp53D (divergent, does polymerise) PRHLNVLLSI 4/10
ACTR1A (builds the dynactin minifilament) PKHVRVMAAE 4/10
ACTRT2 LKFQAPSAAQ 2/10

Both polymerisation-competent divergent controls keep Pro38 and His40 and a hydrophobic
residue at 43–44. ACTRT2 loses Pro38 and His40, replaces the M44/V45/M47 hydrophobic docking
triad with Ser/Ala/Ala, and introduces a proline at position 43, inside the loop that must
dock into the neighbouring protomer.

Honest bounding, and a mistake caught in this review's own first draft: ACTRT2's 14/38 is below
every filament builder in the panel (lowest is ACTR1A at 20/38) but it is not below Arp2
(15/38) or Arp3 (5/38), which nucleate a filament without extending one. The first version of
the summary put Arp2/Arp3 into a single "known polymerisers" set and consequently asserted that
14/38 was "below" 5/38, which is false. So the correct statement is: the measurement argues
against ACTRT2 extending an F-actin filament, and does not exclude an Arp2/3-like role —
for which, separately, nothing has ever been proposed.

Also note these two results are one coupled observation: hydrolysis happens in the F-form,
so a protein that cannot make the F-form contacts has no route to the hydrolysis step
regardless of His161. Counting them as two independent lines would inflate the case — as would
counting the Pro-rich loop substitutions as a third.

A truncated reference sequence in the panel — flagged, and load-bearing nowhere

ACTL10's Swiss-Prot entry (Q5JWF8) is 245 aa, against 366–435 aa for every other panel member
and 374 observed residues in the structure. It is ~130 residues short of the actin fold, so its
tallies contain gaps and apparent substitutions that reflect absent residues rather than
divergence — the same artefact that has already propagated into a merged review elsewhere in this
campaign.

The script now audits panel lengths before any tally is presented, with the cut derived from the
panel's own distribution rather than hand-assigned: 0.75 × the structure's observed chain, i.e. 280.5
aa, which lies inside a real observed gap (shortest unflagged member 366 aa, longest flagged 245 aa).
ACTL10 is the only flag.

Checked which conclusions could be affected: none. The three reference sets that carry arguments
are enumerated explicitly in synthesis — filament_builders (conventional actins, Arp53D, ACTR1A),
nucleators_not_polymerisers (Arp2, Arp3) and the PT-complex ARPs — and ACTL10 is in none of them.
It appears in this review only in the relatives census, where the figure is a count of IBA rows and
is independent of sequence length, and as one of the five un-adjudicated PAINT clade members. Its
truncation was already visible as gap calls (-!) in the named-site table; what was missing was the
statement of why.

IBA donor quality — the objection cannot be about the donors

Every WITH/FROM token was resolved programmatically and each source was then asked, through
QuickGO, what evidence it itself carries for the term it donated:

So SOURCE_WEAK_OR_INFERRED / SOURCE_EVIDENCE_WEAK would be factually contradicted by this
review's own measurements. Any objection to these rows has to be about propagation.

A defect found and fixed while doing this: the first run reported 18/24 and 8/10 because
multi-hit tokens (every MGI/RGD identifier resolves to one Swiss-Prot entry plus several TrEMBL
isoform entries) were excluded from the "carries experimental evidence" filter, silently
dropping the IDA-carrying canonical actins.

The rows are literally shared with siblings — so consistency is mandatory

Mechanical comparison of WITH/FROM fields against the merged sibling reviews:

sibling shared IBA row WITH/FROM byte-identical
ACTL8 GO:0015629 (IBA) yes, 25/25 tokens
ACTR10 GO:0005200 (IBA) yes, 11/11 tokens
ACTL7A / ACTL7B none —

This resolves the apparent GO:0005198-versus-GO:0005200 divergence in the cluster. ACTL7A and
ACTL7B do not share ACTRT2's molecular-function row: they carry GO:0005198 from node
PTN008986528, and their removed GO:0005200 rows were legacy TAS rows citing
PMID:10373328. So there is no single row with two verdicts — there are two different PAINT node
decisions, and the ACTRT clade has not been given one.

Where PAINT has and has not negated GO:0005200

From the cached PTHR11937-paint.tsv, GO:0005200 is propagated at exactly one node
(PTN000940351, IBD from ten experimentally annotated seeds) and explicitly negated as an IRD
with negated=true at eight:

node clade (from that node's other PAINT annotations) date
PTN000233596 Arp2 (GO:0005885 Arp2/3 complex, GO:0051015) 20260416
PTN000233796 Arp3 (GO:0005885, GO:0051015) 20260416
PTN000233752 nuclear ARP, Ino80 complex (GO:0031011, GO:0030234) 20250805
PTN000233887 nuclear ARP, Swr1 complex (GO:0000812, GO:0031491) 20250805
PTN000234048 nuclear ARP, Ino80 complex (GO:0031011, GO:0003729) 20250805
PTN001732543 NuA4 / SWI-SNF (GO:0035267, GO:0016514) 20250805
PTN007551901 (GO:0106006) 20260416
PTN008986528 ACTL7A / ACTL7B — and the only node given the parent GO:0005198 as replacement 20250805

GO:0005198 appears at exactly one node in the whole family. So PAINT has worked through the
divergent-ARP clades one at a time and, at the ACTL7A/7B node, replaced the child with the
parent. The clade carrying ACTRT1, ACTRT2, ACTRT3, ACTL9, ACTL10 (and, on the same
un-negated path, ACTR10) has not been reached, and those genes still inherit GO:0005200
straight from PTN000940351. Live QuickGO confirms the outcome: the human IBA recipients of
GO:0005200 include ACTL9, ACTL10, ACTR10, ACTRT1, ACTRT2, ACTRT3, while ACTL7A and ACTL7B
appear under GO:0005198.

The recommendation is deliberately not "extend the negation". The answer is not uniform
across that clade: ACTR10 is a cryo-EM-verified structural subunit of the dynactin minifilament
and the ACTRT proteins are bulk components of the perinuclear theca on their own evidence,
whereas ACTL10 has no evidence at all. What is needed is for PAINT to address the clade
explicitly, and — if the IRD is applied — for ACTRT1/2/3's structural-constituent term to be
re-asserted from their perinuclear-theca evidence rather than dropped.

The beta-actin subfamily mis-placement does NOT apply to ACTRT2

Verified live from QuickGO rather than read off the sibling review. ACTRT2 draws on
PTN000940351 and PTN002631484 only; the narrow beta-actin-subfamily nodes PTN002631586 and
PTN007551913 donate to ACTL8 alone among the eight divergent human actin-like/actin-related-T
proteins. IBA row counts: ACTL7A 3, ACTL7B 3, ACTL8 11, ACTL9 2, ACTL10 2, ACTRT1 5,
ACTRT2 2, ACTRT3 2 — so ACTRT2 sits at the modal value (2, in 4 of 8 genes). Medians, since
they depend on membership: 2.5 over all eight, 2 excluding ACTL8 (the figure the ACTL8 review
reports for its seven relatives), 3 excluding ACTRT2. No inflation here.

PTN002631484 is a genuinely deep node — it donates GO:0015629 to 18 human genes spanning
33.7–100% identity to beta-actin — so the generic term it carries is the true LCA of a
heterogeneous donor set, exactly as the ACTL8 review argued.

The protein-binding row is the CCT actin-folding pathway, not screen noise

GOA records one GO:0005515 IPI, partner Q9H2J4 = PDCL3 (phosducin-like protein 3 /
PhLP2A, Swiss-Prot, 239 aa), from BioPlex 3.0 (PMID:33961781). IntAct returns 10 records for
ACTRT2, all from that one publication and all anti tag coip; nine are spoke-expanded at
MI-score 0.35 and the PDCL3 pair also has a non-expanded record at 0.50.

The eight partners besides PDCL3 are not a random set: TCP1(CCT1), CCT2, CCT3, CCT6A, CCT6B,
CCT7 plus SLC25A19 and ACSL4 — six CCT/TRiC chaperonin subunits alongside the chaperonin's
co-chaperone. (Ten records, nine distinct partners, so eight besides PDCL3; PDCL3 appears
twice, once spoke-expanded and once not, both at MI 0.50, and the other eight spoke rows are at
0.35. An earlier draft put the non-PDCL3 partner count at nine rather than eight, and the count at
MI 0.35 at nine rather than eight — both off by one, caught by the reviewer.)
Querying PDCL3 itself confirms the reading — across IntAct it has 91 partners including
12 actin-superfamily proteins (ACTA2, ACTB, ACTBL2, ACTG1, ACTR1A, ACTR1B, ACTR2, ACTRT1,
ACTRT2, ACTRT3, POTEF, POTEI), all nine CCT subunits, and three tubulins.

So this is a chaperonin-client contact shared with beta-actin itself and with both ACTRT
paralogues in the same experiment. It is real and mechanistically interpretable — corroborating
that ACTRT2 folds by the actin route — but being a CCT substrate is not a molecular function of
ACTRT2, and there is no informative GO MF term for it. Notably ACTRT1 and ACTRT3 hit PDCL3 in
the same experiment, so any verdict here should hold for all three.

One lead, explicitly not a finding: ACSL4 appears both as an ACTRT2 co-purifying protein
here and as the pro-ferroptotic enzyme upregulated in ACTRT2-deficient testis in PMID:40811009.
That is a single spoke-expanded AP-MS association at MI 0.35 in one experiment, so it is a
hypothesis to test, not evidence.

Reference scope: does each supporting reference observe this protein, or project onto it?

Querying QuickGO by reference rather than by gene. A reference that annotates many entities
to the same term with identical evidence is one projection, not N independent findings.

Two units must not be conflated, and the second column is where it would happen: QuickGO's total is
an annotation count, not an entity count — one reference can annotate several terms per entity
(the calicin paper is 35 annotations over 19 entities). And large result sets are paginated, so
a page total is not the whole; where the walk is capped the entity count is reported as not
counted rather than replaced by the sample size.

reference GOA annotations entities distinct terms assigned by
PMID:12243744 (calyx, founding) 0 0 — —
PMID:11750065 (cloning) 0 0 — —
PMID:35616329 (theca ARP complex) 0 0 — —
PMID:41668650 (Actrt3 KO) 0 0 — —
PMID:40811009 (Actrt2 KO) 0 0 — —
PMID:25293813 (human sperm IF) 0 0 — —
PMID:33961781 (BioPlex 3.0) 9,514 not counted (330+ in a partial walk) GO:0005515 only IntAct only
PMID:35793634 (calicin) 35 19 (walked exhaustively) GO:0005515, GO:0007286, GO:0033011 UniProt

An earlier draft of this table wrote "thousands" in the entities column for BioPlex — an inference
standing in for a measurement, which is the same error as reading a page total as a whole. The
script now returns None with an explanatory note whenever the walk was capped.

Two conclusions, opposite in direction.

BioPlex. 9,514 annotations (not entities), every one GO:0005515, every one from IntAct —
checked on pages 1, 20 and 40, so the uniformity is not a first-page artefact, though sampling three
pages of a paginated set is evidence of uniformity rather than proof of it. That is the strongest possible statement
that the term carries no gene-specific information, and it is part of why the row moved to
MARK_AS_OVER_ANNOTATED. It is not the ACTR8 projection failure, though: ACTRT2 was
individually assayed in that experiment, so this is a real if uninformative observation, not a
complex-level annotation copied onto subunits that were never perturbed.

Calicin — and this one had to be tested, not assumed. The mouse GO:0033011 IDA that both
of ACTRT2's theca rows descend from comes from a paper that gave that same term to 12 entities. If
the curator had projected "these are theca proteins" onto every protein named, that IDA would be
one observation counted twelve times, and the ortholog transfer would be correspondingly weaker.
The discriminating test is whether the term went to a subset:

So the curator discriminated per protein rather than projecting — is_subset_not_blanket = True.
The IDA stands. Note also the curation pattern: localisation went to 12 proteins, but the only
biological-process term (GO:0007286, IMP) went to Ccin alone, the gene actually knocked out.
That is correct practice, and it explains ACTRT2's process gap precisely — the gap is not an
oversight in this paper's curation, it is that no Actrt2 perturbation has been curated at all
(PMID:40811009 exists and has zero annotations).

Both NEW rows are ISS onto donors that hold nothing — stated as a precondition

The reviewer's sharpest catch, and this review's own reference-scope table is what proves it:
PMID:12243744 and PMID:40811009 have each produced zero GOA annotations anywhere, so
bovine Q2TA43 cannot donate GO:0033150 and mouse Q9D9L5 cannot donate GO:0007283
(confirmed directly: 0 QuickGO hits for each accession/term pair). GO's ISS convention wants the
WITH/FROM entry to carry an experimental annotation to the same term, so as first written both
rows were dangling — a curator importing them would find the source empty.

The biology is unaffected and the terms and evidence codes are unchanged. What changed is that
each row now states its precondition explicitly: GO:0033150 is contingent on a prior bovine
IDA from PMID:12243744, and GO:0007283 on a prior mouse IMP from PMID:40811009. For
GO:0033150 the alternative is also named — a human IDA on PMID:25293813, which the reviewer
rightly notes is a shorter journey than the original reason allowed, since GO places the calyx "at
the posterior end of the perinuclear theca" and that paper localises ACTRT2 to the post-acrosomal
region.

The annotation gap, measured

For each reported PT-complex member and its mouse ortholog, all GO annotations were pulled from
QuickGO. All six carry GO:0033011 in both species. Experimental biological-process terms:

gene human experimental BP mouse experimental BP
ACTL7A none GO:0001675, GO:0007286, GO:0009566
ACTL9 GO:0001675, GO:0009566 GO:0001675, GO:0009566
CCIN GO:0007283 GO:0007283, GO:0007286
ACTRT1 GO:0008589, GO:0045892 none
ACTRT2 none (1 BP row total) none (2 BP rows)
ACTRT3 none (1 BP row) none (1 BP row)

ACTRT2 and ACTRT3 are the only two members with no experimental BP annotation in either
species. The cause is traceable: PMID:35616329 names ACTRT2 explicitly as a member of the
complex that "mediates the acrosome-nucleus connection" and has produced BP annotations for
ACTL7A and ACTL9 but none for Actrt1 or Actrt2; PMID:35793634 gave mouse Actrt2 a
cellular-component IDA and a protein-interaction row but no process term; and PMID:40811009's
knockout phenotype has not been curated at all. Because the human record is fed by ortholog
transfer, the human gap is downstream of the mouse gap.

GO:0033150 cytoskeletal calyx exists, and nothing in this family uses it

GO has a term for the exact structure ACTRT2 was purified from: GO:0033150 cytoskeletal calyx,
part_of GO:0033011, defined as "A large cytoskeletal structure located at the posterior end of
the perinuclear theca of a mammalian sperm head. The nucleus is tightly associated with the
calyx, which contains calicin and basic cylicin proteins."

In human, that term is annotated to exactly three genes — CYLC1, CYLC2 and CCIN — i.e.
precisely the three proteins its own definition names, mostly by the UniProt subcellular-location
keyword SL-0032. Meanwhile the paper that named ARP-T1/ARP-T2 identified them as major acidic
components of that same purified structure, and listed two actin-capping proteins there too
(CAPZA3 and CAPZB carry GO:0033011 but not GO:0033150). The term's definition and its
annotation set have both stayed at the three basic proteins.

This is additive, not corrective: GO:0033011 must stay, because ACTRT2's subacrosomal pool
(PMID:35616329) has no GO term at all — GO has no term for the subacrosomal layer, the
postacrosomal sheath or the acroplaxome (OLS and QuickGO searches return nothing), which is
exactly the gap the merged ACTL7A review filed as a proposed acroplaxome term, naming ACTRT2
among the genes that need it. That proposal is supported here rather than duplicated.

Verdicts and why

# term ev action basis
1 GO:0015629 actin cytoskeleton IBA KEEP_AS_NON_CORE Byte-identical row to ACTL8's, same deep LCA node; matches that merged verdict. Kept because nothing contradicts the compartment; non-core because the evidenced compartment is the PT, and GO:0033011 is verified not to be a GO:0015629 descendant, so this row is an independent unevidenced claim rather than a redundant ancestor.
2 GO:0005200 structural constituent of cytoskeleton IBA MODIFY -> GO:0005198 Reversed after the merged ACTRT3 review, which holds the byte-identical row and modified it. See the reversal section below.
3 GO:0005856 cytoskeleton IEA KEEP_AS_NON_CORE Strict ancestor of GO:0033011 (verified in the QuickGO ancestor list), which is on the gene; true but redundant. Matches ACTL7A, whose situation is identical.
4 GO:0007010 cytoskeleton organization IEA KEEP_AS_NON_CORE Inter-ontology inference whose sole input is row 2, so its machine provenance lapses once row 2 is generalised (no such link runs from GO:0005198). Survives on the compartment's own classification: ACTRT2 is a bulk theca component and GO places the theca under GO:0005856. Same action as ACTR10 and ACTRT3; propagation classification follows ACTRT3's.
5 GO:0005515 protein binding IPI MARK_AS_OVER_ANNOTATED Reversed after the merged ACTRT3 review, which holds this byte-identical row (same publication, same partner). Real partner, but PDCL3 is the CCT co-chaperone and the row explains itself away as a folding-pathway artefact of an over-expressed actin fold. Follows ACTL8/ACTR10/ACTRT3, not ACTR1B.
6 GO:0033011 perinuclear theca IEA ACCEPT Core location. Donor mouse Actrt2 holds the same term by IDA, so no downward MODIFY is warranted — the ACRV1-style check run and reported negative.
7 GO:0033011 perinuclear theca ISS ACCEPT Same, by the curated route.
8 GO:0033150 cytoskeletal calyx ISS NEW The structure the founding paper purified ACTRT2 from; term exists and is unused outside the three basic proteins. ISS because the fractionation was bovine and the paper does not state the immunofluorescence species.
9 GO:0007283 spermatogenesis ISS NEW Actrt2-/- mice have significantly shrunken seminiferous tubules and reduced spermatogenesis. Restores the only BP claim ACTRT2 has any experimental basis for.

Reversal: GO:0005200 ACCEPT -> MODIFY, after the merged ACTRT3 review

An earlier round of this review ACCEPTed GO:0005200. That is withdrawn. ACTRT3's review
merged while this one was in flight, it holds the byte-identical row (11/11 tokens, node
PTN000940351 — verified mechanically, not by eye), and it modified the row to GO:0005198. It
also pre-empted the exact argument I had used, and its refutation is a checkable fact, which I
checked before conceding:

the merged ACTR10 review's ACCEPT of this same row … ACTR10 has an ortholog-strength donor in
the seed set and ACTRT3 has none.

Resolving all ten seed donors confirms it: mouse Actg1, rat Actg1, human ACTB, yeast ACT1, yeast
ARP1, yeast ARP10, human ARP2, human ARP3 and two Dictyostelium actins — no ARP-T of any
kind. ACTR10's ACCEPT rests on yeast ARP10 being its own ortholog in that set. ACTRT2 has no
such donor, so the precedent I leaned on does not transfer, and my invocation of it was wrong on a
fact rather than on a judgement.

Three further reasons the reversal is right, not merely conceded:

  1. The interface measurements do not distinguish the two genes — ACTRT2 20/38 chemically
    compatible protomer contacts, ACTRT3 18/38, both far below the 28/38 floor set by a dynactin
    Arp1 paralogue that does build a filament. Whatever verdict fits one fits the other.
  2. Two more independent sources prefer the parent: PAINT substituted GO:0005198 at the adjacent
    node PTN008986528 on the same day it rejected the child, and the affinage record's own GO
    grounding also lands on GO:0005198
    [file:human/ACTRT2/ACTRT2-deep-research-affinage.md "**molecular_activity:** GO:0005198 structural molecule activity, GO:0008092 cytoskeletal protein binding"]. (The second half of
    that line is a fold-to-activity leap and is not adopted; the ACTL7B review flagged the identical
    proposal for its gene.)
  3. Nothing real is lost. GO:0005198 plus GO:0033011 — a GO:0005856 descendant — already
    convey "structural molecule in a cytoskeletal structure". The only thing the child adds is the
    filament reading, which is the false part.

The evidence behind the withdrawn argument is not withdrawn. ACTRT2 is one of Heid's two
major acidic components of the purified calyx, and that is bulk structural constituency of a
structure GO classifies as cytoskeletal — evidence neither ACTRT3 (not one of the two) nor ACTL7A
(residence only, and its review withdrew the residence argument) possesses. ACTRT3's review filed
"can the theca proteins re-earn GO:0005200 on evidence of their own?" as an open question; this
review's contribution is the strongest instance of exactly that evidence, and it is carried into
suggested_questions rather than used to publish a filament inference the structure audit excludes.

GO:0005515 also reversed, and strengthened in the process

KEEP_AS_NON_CORE -> MARK_AS_OVER_ANNOTATED, again aligning with ACTRT3's byte-identical row
(same publication, same partner Q9H2J4). ACTRT3's argument is that the partner explains the
observation away: PDCL3/PhLP2A is a CCT/TRiC co-chaperone that modulates cytoskeletal actin
folding, actin is an obligate CCT client, and BioPlex was run in HEK293T and HCT116 while these
proteins are testis-restricted.

My evidence strengthens that argument rather than opposing it, and this is the one place the
two genes' data differ materially on a shared row: ACTRT3 had to infer the chaperonin from
PDCL3's identity, having only two IntAct records. ACTRT2 has ten records over nine
partners, and the eight besides PDCL3 are TCP1, CCT2, CCT3, CCT6A, CCT6B, CCT7, SLC25A19 and
ACSL4 — six of them CCT subunits — so for ACTRT2 the holo-chaperonin
is directly observed and the inference is confirmed. Querying by reference rather than by gene
adds the scale: PMID:33961781 is the source of 9,514 GOA annotations, every one GO:0005515,
every one assigned by IntAct.

So the correct reading is ACTRT3's: the row shows the ACTRT2 polypeptide is recognised as a
foldable actin-fold client, which is mildly informative about the fold and says nothing about the
gene's function. ACTR1B kept its row from this same publication because its partner set was
assembled dynactin, its own complex; here the partner set is the folding machinery.

Retraction and correction checks

Two separate checks, because they catch different things.

Process notes