ABRACL (Q9P1F3) — review notes

Human ABRACL, "ABRA C-terminal like", formerly C6orf115 / HSPC280 / PRO2013. 81 aa, ~9 kDa,
usually one copy per genome, but not universally: PANTHER PTHR46334 resolves into two subfamilies,
SF1 (human, Dictyostelium and most plants) and a plant-restricted SF3, and Arabidopsis has 2 members
while maize has 3.

Starting position: three annotations, no molecular function

The whole GOA record is three rows:

Nothing is experimental, and there is no molecular function annotation at all — despite a
2021 paper that reports a direct in-vitro binding assay and a direct in-vitro activity assay
on purified recombinant protein.

What the primary literature actually shows

Dictyostelium: the origin of the phylogenetic annotation

cosA was found in a chemotaxis screen. cosA-null cells move slower but steer normally
PMID:20940261, and their actin cytoskeleton
is disturbed PMID:20940261.

The decisive point for the IBA is functional complementation: PMID:20940261 The human
protein does the Dictyostelium protein's job in Dictyostelium cells. That is a stronger basis
for an ortholog transfer than sequence identity alone.

Human: cofilin, F-actin, and the leading edge

All from PMID:33670794 (Hsiao et al. 2021, full text available):

• Direct, weak F-actin binding by purified untagged recombinant protein in a co-sedimentation
  assay: PMID:33670794
• Colocalisation of endogenous protein with F-actin at the leading edge:
  PMID:33670794
• Cofilin interaction in cells, three ways — colocalisation especially at lamellipodia,
  proximity ligation that disappears on knockdown of either partner, and co-IP:
  PMID:33670794 and PMID:33670794
• The functional interaction, on purified proteins: PMID:33670794, and the necessary
  control ruling out simple displacement: PMID:33670794
• Loss-of-function: PMID:33670794

The sign problem — why the vague GO term is the right one

Three lines of evidence disagree about which way ABRACL pushes the F-/G-actin balance.

1. Losing the Dictyostelium ortholog raises cytoskeleton-associated actin
   PMID:20940261.
2. Losing human ABRACL lowers it PMID:33670794.
3. Purified ABRACL on its own inhibits polymerisation PMID:33670794, which the authors
   themselves flag as inconsistent with (2) PMID:33670794.

Only in the presence of cofilin does ABRACL clearly favour F-actin. So GO:0030835 negative regulation of actin filament depolymerization and GO:0030837 negative regulation of actin filament polymerization are each supported by part of the evidence and contradicted by another
part. The unsigned parent GO:0032970 is not a lazy annotation here — it is the only claim all
three datasets support. Left as ACCEPT for exactly that reason.

The cilium annotation: a four-step chain with nothing at the end

This is the main finding of the review.

GO:0005929 cilium is IEA from GO_REF:0000044, i.e. a mechanical mapping of the UniProt
subcellular-location line Cell projection, cilium {ECO:0000305|PubMed:37759737}. Tracing back:

1. GOA transfers UniProtKB-SubCell:SL-0066 without judgement — as designed.
2. UniProt flags its own statement ECO:0000305 (curator inference, not observation).
3. PMID:37759737 is an expression study of mouse and cat embryonic telencephalon. It
   reports no ciliary localisation of its own. It says only:
   PMID:37759737 — a citation of somebody else's work. In the
   same paragraph it reports data pointing the other way:
   PMID:37759737 — the ventricular zone being where the ciliated progenitors are.
4. That "high-throughput proteomics study" is PMID:26638075, the Gupta et al. centrosome-cilium
   BioID interactome. It is the only source of physical-interaction data for ABRACL in IntAct:
   five proximity-labelling hits, one publication, one detection method, with baits SASS6,
   CNTRL, DCTN1 and RPGRIP1L (see ABRACL-bioinformatics/RESULTS.md).

So a proximity-labelling neighbourhood became, three citations later, a subcellular location on
the human protein. Nobody along the chain did anything unreasonable; the failure is that no step
re-examined the one before it.

The phyletic check in ABRACL-bioinformatics/RESULTS.md adds a family-level argument: Costars is
retained in Arabidopsis, rice, maize and Dictyostelium, none of
which builds a cilium, basal body or centriole, while IFT88, IFT52, BBS1 and ARL13B are absent
from all four. Whatever the family is conserved for, it is not ciliary. That does not exclude a
human-specific ciliary role, and the notes say so — but there is no evidence for one either.

The lamellipodium annotation is the opposite problem

GO:0030027 lamellipodium is also IEA-from-SubCell, but here the underlying UniProt statement is
ECO:0000269 (experimental) and the human experiment exists: endogenous protein, endogenous
antibody, colocalised with F-actin at the leading edge, in the paper's Figure 4D. This
annotation deserves to be IDA. Kept as ACCEPT with a recommendation to upgrade the evidence code
rather than the term.

What is missing from GO itself

ABRACL binds cofilin and reduces cofilin-stimulated filament disassembly without displacing
cofilin from the filament. GO has a molecular function for the opposite sign — GO:0000513 actin severing activator activity, "Binds to and increases the activity of a actin severing protein" —
and no inhibitor counterpart. The absence of that sibling term is a large part of why this gene
has no MF annotation. Proposed as a new term in the review.

GO:0051015 actin filament binding is proposed as a NEW annotation in the meantime: it is
directly demonstrated (purified protein, co-sedimentation, quantified against a no-actin
control), and it is honest about being weak.

Loose ends not annotated

• Nuclear localisation. PMID:26537243 reports HSPC280/Abracl in the subventricular zone and
  that overexpression in Neuro2a inhibits neuronal differentiation. Nuclear localisation there is
  hard to reconcile with the lamellipodial pool; the protein is 9 kDa and would diffuse through
  nuclear pores regardless, so this needs a targeted experiment, not an annotation.
• Winged-helix fold without DNA binding. PMID:21082705 solved the NMR structure (PDB 2L2O).
  The fold is winged-helix-like but the groove is hydrophobic and the recognition helix negatively
  charged, so the natural reading is protein interaction, not DNA. No DNA-binding annotation
  exists and none should be added.
• Cancer biology. Multiple knockdown papers in oesophageal, breast, gastric and glioma lines,
  plus upstream regulators MYBL2, CBX4 and miR-145-5p. These are disease-context studies of a
  normal actin regulator; they do not add a distinct GO process for the gene and are not
  annotated here.

Correction: the family is not single-copy, and the subfamily structure matters

A first draft of this review called PTHR46334 a single-copy family, in five places. The PR's own
fetched data contradicts that — RESULTS.md shows Arabidopsis = 2 and Zea mays = 3, and the
PANTHER metadata records subfamilies: 2.

Resolving the entries file:

This does not weaken the GO:0032970 ACCEPT — it strengthens the provenance argument. The IBA
donor (Dictyostelium cosA) and the human target are both SF1, so the transfer is
within-subfamily and is not crossing the one boundary the family contains. The propagation_review
now names that boundary rather than denying one exists.

The lesson: the copy-number claim was a summary written alongside the analysis rather than read out
of it, and the analysis script printed it as hardcoded prose regardless of the counts fetched. Any
claim a script asserts should be derived from what the script fetched.

Two verifications added after review

GO:0000513 is real. It is absent from the local ontology cache, so the reviewer could not
check it. QuickGO confirms: GO:0000513 actin severing activator activity, molecular_function,
not obsolete, defined as "Binds to and increases the activity of a actin severing protein." That
is exactly the positive counterpart of the inhibitor activity proposed under proposed_new_terms,
so the asymmetry the proposal rests on is genuine.

RPGRIP1L is genuinely ciliary. Conceded in the review rather than glossed: of the four BioID
baits that labelled ABRACL, three (SASS6, CNTRL, DCTN1) are centriolar or dynactin proteins, but
RPGRIP1L is a transition-zone protein. The bait panel is therefore not uniformly non-ciliary. It
does not rescue the annotation — proximity to one ciliary bait over hours of labelling is a
neighbourhood observation, not a localisation — but the argument is stronger for stating the
inconvenient part.