Where does the human ABRA GO record come from, and does its one plasma-membrane call have any support?

Reproduce with uv run --no-project --with requests python audit_abra_record.py.
Q1, Q2, Q4 and Q5 are fetched at run time from the UniProt, QuickGO and InterPro REST APIs,
plus this gene's own ABRA-goa.tsv. Q3 instead reads the repository's committed PANTHER
family export, interpro/panther/PTHR22739/PTHR22739-entries.csv (refresh it with
just fetch-panther-family PTHR22739); if that file is absent the script aborts rather
than dropping the section. Nothing is hardcoded, and a failed fetch is reported rather
than counted as an absence. Machine-readable output is written to results.json.

Run date: 2026-07-25.

Q1. The human record is almost entirely rodent

The 16 GOA rows break down as IBA 3, IEA 6, ISS 5, IPI 1, IDA 1 — two experimental rows
out of sixteen, and neither of them is about what ABRA does. Resolving every WITH/FROM
accession to a species gives:

Every WITH/FROM protein accession is a genuine one-to-one ortholog of the gene under review
(mouse or rat Abra), not a paralog, so the transfers are legitimate in kind. The point is
the concentration: fourteen of the sixteen rows trace to two rodent proteins, and the mouse
entry supplies eight of them under four different identifier styles, which makes the record
look better-sourced than it is. No human experiment has established any molecular function,
process or native localisation for ABRA.

Q2. The plasma-membrane annotation is unsupported, and it has already spread

GO:0005886 plasma membrane is the only cellular-component annotation on human ABRA not
derived from a rodent ortholog. It is an IDA under GO_REF:0000054 — the LIFEdb survey of
GFP-tagged proteins expressed in cultured cells. Three independent checks:

• UniProt does not place ABRA there. Its subcellular location is
  Cytoplasm, myofibril, sarcomere and Cytoplasm, cytoskeleton. No plasma membrane.
• Nothing in the sequence targets a membrane. UniProt was queried for all four
  membrane-targeting feature classes — Transmembrane, Intramembrane, Signal, Lipidation —
  and ABRA carries none of them. A peripheral protein can of course reach a membrane through
  a partner, but no such anchor is reported for ABRA, whereas its F-actin binding gives a
  ready alternative reading of a GFP rim signal as cortical actin.
• The call has been projected back into the mouse. Mouse Abra now carries GO:0005886
  twice — an Ensembl Compara IEA (GO_REF:0000107) and an ISO (GO_REF:0000119) — and both
  cite UniProtKB:Q8N0Z2, i.e. the human protein. The single human GFP-overexpression image
  is the sole origin of a plasma-membrane annotation in two species.

Q3. The PANTHER family carries no paralog hazard

All 4 reviewed members of PTHR22739 are named ABRA (human, mouse, rat, pig) and all fall in
the single subfamily PTHR22739:SF20. There is no second subfamily and no paralog in the
reviewed set, so the IBA annotations from node PTN001100454 and the ISS transfers are
operating inside a one-to-one ortholog group. Whatever else is wrong with this record, the
propagation topology is not.

Q4. "Costars" is a domain shared with a different gene

Exactly two reviewed human proteins carry the Costars domain PF14705:

ABRACL is essentially the domain alone; ABRA is a 381-residue protein in which the domain
occupies the C-terminus. Work reported on "Costars" — the 82-residue Dictyostelium protein
of PMID:20940261 and its small mammalian counterpart — is therefore a statement about the
ABRACL-type protein, not about ABRA. Cytoskeletal phenotypes from that line of work should
not be carried over to ABRA on the strength of the shared domain.

Q5. The mouse ortholog carries the import term, but not the regulation term

The review proposes GO:0042307 positive regulation of protein import into nucleus as a
NEW annotation for human ABRA. That proposal previously rested on reasoning about the mouse
record and the MRTF literature without being checkable here, unlike the Q2 back-propagation
finding. Projecting both import terms onto mouse Q8BUZ1:

Two things follow, and they point the same way.

The mouse annotation is a direct assay with an empty WITH/FROM, so it is not an ortholog
transfer from human — it is primary evidence, and PMID:15798203 is the same paper the review
already cites for the MRTF import mechanism. The NEW proposal is therefore anchored to a
real experiment rather than to an inference chain.

But the mouse record uses the unqualified process term, and that is very likely the less
accurate of the two. GO:0006606 reads as the annotated protein being part of the import
event; ABRA is not the cargo. What PMID:15798203 shows is ABRA promoting the import of
MRTF-A/B, which is what GO:0042307 says. So the human proposal is not merely a copy of the
mouse annotation at a different granularity — it is the term the mouse row arguably should
have used, and the gap in the mouse record is itself the reason no transfer supplied it to
human.