ACAP2 (Q15057) — review notes

Journal for the PAINT + affinage review of human ACAP2 / centaurin-β2 / CENTB2 / KIAA0041.

1. What the protein is

778 aa, five-module architecture, all confirmed in the UniProt feature table
[file:human/ACAP2/ACAP2-uniprot.txt "FT DOMAIN 399..520"]:

PE 1: Evidence at protein level — the protein is detected, so nothing here should be
described as "no experimental data"; there is in fact good biochemistry.

2. The two established activities

(a) ARF6-preferring GTPase-activating protein. The founding paper identified ACAP1 and
ACAP2 as Arf6 GAPs
PMID:11062263
with genuine substrate preference
PMID:11062263.
Catalysis needs the conserved arginine
PMID:11062263
and the activity is lipid-gated
PMID:11062263,
which UniProt records as
[file:human/ACAP2/ACAP2-uniprot.txt "CC -!- ACTIVITY REGULATION: GAP activity stimulated by phosphatidylinositol"].
Specificity is real in the negative direction too: ACAP2 is not a GAP for the Rab it binds
[file:human/ACAP2/ACAP2-uniprot.txt "CC (ARF6). Doesn't show GAP activity for RAB35 (PubMed:30905672)."].
Compartment specificity: the effect is peripheral, not Golgi
PMID:11062263.

(b) RAB35 effector. ACAP2 binds GTP-loaded RAB35 through its terminal ankyrin repeat
PMID:30905672,
which UniProt localises to the ANK repeats
[file:human/ACAP2/ACAP2-uniprot.txt "CC -!- DOMAIN: The ANK domains are required for interaction with RAB35."]
— note UniProt's SUBUNIT line writes "KANK domains", which looks like a typo for ANK; the
DOMAIN line and the crystal structure both say ankyrin repeats. The specificity is extreme
PMID:25694427.

The two modules compose into one mechanism: RAB35 recruits ACAP2, which then switches ARF6 off
PMID:24600047.

Caution on residue numbering: PMID:25694427 names Asn-610/Asn-691 in the minimal RAB35-binding
site, while PMID:30905672's human mutagenesis is at Asp-721/Arg-727/Met-731/Asp-756. The two
papers are not on the same numbering, so the 610/691 pair must not be quoted as human positions.
The affinage record does exactly that, which is why it was not used as the source for any
residue-level claim here.

3. Cellular readouts

• Overexpression suppresses PDGF-induced dorsal ruffles PMID:11062263,
  yet the protein still goes to the ruffles that form PMID:11062263.
• In human HeLa cells PMID:11062263
  ACAP2 blocks ARF6-driven protrusions PMID:11062263
  by half PMID:11062263, GAP-dependently
  PMID:11062263.
  These are overexpression/point-mutant experiments, i.e. IMP-grade, not IDA-grade, evidence for the process.
• Steady state it is largely soluble but visits ARF6 tubules
  PMID:11062263,
  and those tubules are the recycling route
  PMID:11062263.
• Non-human deployments of the same axis: NGF-triggered neurite outgrowth in rat PC12
  PMID:22344257,
  FcγR phagocytosis in macrophages PMID:22045739,
  and oligodendrocyte differentiation PMID:24600047.
• One human loss-of-function phenotype outside membrane traffic
  PMID:25853217,
  in a paper that also reports direct lipid binding
  PMID:25853217.
  Abstract-only in the cache, so the lipid species and the assay cannot be read off; unreplicated.

4. WITH/FROM resolution (the highest-yield step)

Every accession in column 11 of ACAP2-goa.tsv, resolved:

All of this is now reproducible: ACAP2-bioinformatics/resolve_withfrom.py regenerates the
table from primary APIs and writes RESULTS.md. It was added in response to PR review,
which correctly pointed out that the blow and MGI:MGI:2153589 claims were load-bearing
but had nothing behind them in-repo.

Does each source hold its own evidence for what it donates?

An IBA/ISS WITH/FROM list is supposed to name experimentally annotated family members, so it
is worth asking rather than assuming. Querying QuickGO for each source's own annotations to
the donated term (descendants included):

Every single source carries wet-lab evidence for the term it donates. So none of these
transfers is an inference recycling another inference, and no verdict here should say
otherwise. That sharpens the blow objection rather than softening it: blow's actin
annotation is a real experimental result, and the problem is purely that a protein with no
ArfGAP domain cannot be producing that phenotype the way ACAP2 would. It also softens the
mouse-Acap3 note — the paralog is a competent donor for GAP activity (its own IMP), so the
only observation is that the ortholog is absent from that support set.

Two things fall out of this table.

(i) blow is a family-boundary artifact. PANTHER places blow in PTHR23180 (CENTAURIN/ARF)
as subfamily SF399, and blow carries experimental GO:0030036 actin cytoskeleton organization
from myoblast fusion. But A1Z714/P91678 has one annotated domain, a PH domain at 204–307 —
no BAR, no ArfGAP catalytic domain, no ankyrin repeats, and UniProt records no PANTHER
cross-reference for it at all. Whatever blow does to actin during myoblast fusion, it cannot be
doing it as an ARF GAP. Its inclusion among the four supporting entities of ACAP2's
GO:0030036 IBA is therefore not real support.

(ii) PTN001142372 is pan-eukaryotic. All four IBA rows come from this single node. Querying
QuickGO for annotations carrying PANTHER:PTN001142372 returns the identical four-term set
(GO:0005096, GO:0005886, GO:0010008, GO:0030036) on Brassica napus, Cucumis sativus,
Medicago truncatula, Theobroma cacao, Zostera marina and Mimulus guttatus genes, on
yeast AGE1, and on Xenopus acap2.L. So the node ancestral to these annotations predates the
plant/fungal/animal split; it is not the ACAP1/2/3 node. GAP activity survives that depth
(everything under it except blow has an ArfGAP domain). A specific process term does not.

5. The ontology problem

The best-characterised fact about ACAP2 — that it is an ARF6-preferring GAP — cannot be
expressed in current GO. GO:0008060 ARF GTPase activator activity has been merged into
GO:0005096 GTPase activator activity, together with GO:0005097 (Rab GAP), GO:0005098 (Ran
GAP), GO:0005099 (Ras GAP), GO:0005100 (Rho GAP), GO:0005101, GO:0017123, GO:0030675 and
GO:0046582 — all nine appear as secondaryIds of GO:0005096, which now has no substrate-specific
children (its only child is GO:1902773 GTPase activator complex).

The same has happened on the binding side: GO:0017137 Rab GTPase binding is now a secondary id
of GO:0031267 small GTPase binding, which has no children at all.

So GOA's GO:0005096 and GO:0031267 rows for ACAP2 are already maximally specific. The
ARF6-versus-ARF1/ARF5 preference, and the exclusive recognition of RAB35 among 60 mammalian
Rabs, are recoverable only through has_input annotation extensions or a GO-CAM — not through
term choice. This is a deliberate GO design decision, so the right recommendation is extensions,
not resurrecting the obsoleted terms.

By contrast the process side has a real, unused, non-obsolete term:
GO:0032013 negative regulation of ARF protein signal transduction. GOA gives ACAP2 four BP
annotations (actin cytoskeleton organization IBA, actin filament-based process IDA, endocytic
recycling ISS, cellular response to NGF ISS) and none of them says that ACAP2 turns ARF6 off,
which is the whole of its characterised biology. That is the main gap this review closes.

5b. A curation asymmetry with ACAP1

PMID:11062263 characterised ACAP1 and ACAP2 together, in the same figures, and ACAP1 was the
more potent of the two in the protrusion assay
PMID:11062263.
Yet QuickGO returns three annotations from that reference for ACAP2 (GO:0005096, GO:0030029,
GO:0001726, all IDA) and zero for ACAP1 (UniProtKB:Q15027) — verified directly:

.../annotation/search?geneProductId=UniProtKB:Q15027&reference=PMID:11062263  -> numberOfHits 0
.../annotation/search?geneProductId=UniProtKB:Q15057&reference=PMID:11062263  -> numberOfHits 3

So the asymmetry is a curation artifact, not biology. Two consequences for this review: ACAP2's
IDAs from this paper are not over-calls (they are the recorded half of a symmetric experiment),
and ACAP1 is under-annotated from the same source. Cross-checked against the parallel ACAP1
review (PR #2251), which reached the same conclusion independently.

6. Interactome triage

IntAct has 33 records for Q15057. Almost all are single-shot high-throughput: a PI(3)P affinity
proteomics screen that pulled 681 proteins PMID:23416715, socioaffinity inferences, and
one-off tag co-IPs (MAPT, DOK2, ELK4, RASA1, RAB5A, RAB8A, CAPZB, MED4, Tgs1, Naa10, PDK3,
HSPB1). Two entries are worth separating out:

• ACAP3 (Q96P50) is recovered by two orthogonal proteome-scale datasets — BioPlex 3.0
  PMID:33961781 and OpenCell PMID:35271311 — and UniProt promotes it to an INTERACTION line
  with NbExp=4. The pair is probably real, and hetero-oligomerisation through the shared
  N-terminal BAR domain would be the obvious mechanism, but nothing has been tested; AP-MS
  cannot license a dimerisation MF. Both GOA rows stay as uninformative protein binding.
• Vaccinia K1L binds ACAP2 PMID:16806385, but the paper's point is that the host-range
  function of K1L is separable from ACAP2 binding, so it says nothing about ACAP2's own function.

Not in GOA and not used: RNF126-driven degradation of ACAP2 in ovarian cancer PMID:40251363 —
a single 2026 paper, ACAP2 as substrate rather than agent.

7. Decisions taken

• GO:0005096 (IBA, IEA, IDA) — ACCEPT ×3. Maximally specific term available; substrate
  specificity noted for extension/GO-CAM capture.
• GO:0031267 (IEA from mouse ortholog) — ACCEPT. GO:0005515 with RAB35 — MODIFY to
  GO:0031267 + GO:0030742 (GTP-dependent protein binding), since ACAP2 recognises only the
  GTP form.
• GO:0005515 with ACAP3 ×2 — MARK_AS_OVER_ANNOTATED. Real but functionally silent.
• GO:0030029 (IDA) and GO:0030036 (IBA) — MODIFY both to GO:0032956 regulation of actin cytoskeleton organization. ACAP2 is upstream signalling, not cytoskeletal machinery; and the
  IBA's supporting set is the pan-eukaryotic node plus blow.
• GO:0016020 (HDA) — MARK_AS_OVER_ANNOTATED. Bulk NK-cell membrane fraction; the study's own
  method removed peripheral membrane proteins and it concedes ~40% plausible membrane proteins,
  while ACAP2 is peripheral. Subsumed by the specific membrane terms anyway.
• GO:0005737 (IEA from the BAR signature) — KEEP_AS_NON_CORE. True (largely cytosolic at
  steady state) but uninformative, and BAR→cytoplasm is a poor inference for a curvature-binding
  module.
• NEW: GO:0032013 (IMP, PMID:11062263) and GO:0035091 phosphatidylinositol binding
  (IDA, PMID:25853217).
• Not proposed: GO:0043065 positive regulation of apoptotic process. Only PMID:25853217
  supports it, abstract-only, unreplicated, and the paper itself notes ACAP2 is not caspase-cleaved
  so the CNT-1 mechanism does not transfer. Filed as a knowledge gap and an experiment instead.
• Not proposed: phagocytosis. PMID:22045739 says only "in macrophages" and the cache is
  abstract-only, so the species is not stated and an ISS cannot be justified from it.