ACRV1 (SP-10) — review notes

Working journal for the PAINT + affinage review of human ACRV1 / P26436 / ASPX_HUMAN.

1. What the gene is

265-aa precursor with a signal peptide (1–21), testis-restricted, product of a single
gene on chromosome 11. Named SP-10 from the MHS-10 monoclonal antibody that defined it;
ACRV1 ("acrosomal vesicle protein 1") is the HGNC symbol. It was designated a primary vaccine
candidate
PMID:2310816, which is why so
much of the literature is immunolocalisation rather than biochemistry.

UniProt PE 1: Evidence at protein level — the protein is directly sequenced
(residues 78–100, 106–122, 127–151), so this is not a gene lacking protein-level
evidence. What it lacks is biochemical evidence.

2. Localisation — the best-supported part of the record

• Immunocytochemistry on human testis and sperm places SP-10 through the whole acrosome,
  in a bilaminar array against the inner face of the outer acrosomal membrane and the
  outer face of the inner acrosomal membrane; after an ionophore-induced acrosome
  reaction it stays on the sperm head with the inner acrosomal membrane and equatorial
  segment
  PMID:2310816.
• It is not on the surface of acrosome-intact sperm
  PMID:2310816 —
  i.e. it is a lumenal/matrix protein of an intact secretory organelle, and only becomes
  exposed after the acrosome reaction.
• Phase partitioning shows it is peripheral, not integral: both the Triton X-114-released
  and TX-114-resistant pools are hydrophilic, and the resistant pool comes off with a
  chaotrope (150 mM NaSCN) or pH extremes but not with repeated detergent or 1.5 M NaCl
  PMID:1591355.
• EM immunogold on epididymal/ejaculated/capacitated sperm concentrates label in the
  principal segment and posterior bulb of the equatorial segment; after a follicular
  fluid-induced acrosome reaction it is on the inner acrosomal membrane of the equatorial
  segment and on hybrid vesicles
  PMID:7888499.

So GO:0001669 acrosomal vesicle is right, and GO:0043159 acrosomal matrix /
GO:0002079 inner acrosomal membrane are the finer terms the evidence actually supports.
Neither is currently annotated.

3. Processing and isoform heterogeneity

Two independent sources of size heterogeneity, and the literature separates them cleanly.

• Proteolysis. Eight purified SP-10 peptides were Edman-sequenced; their amino termini
  map along the deduced sequence and imply endoproteases cutting after Arg (trypsin-like,
  possibly acrosin) plus after Ser, Pro, Gly and Glu
  PMID:1637938.
  A ~45 kDa full-length precursor is present in testis but absent from epididymal and
  ejaculated sperm; the 25–18 kDa species first appear in caput epididymal sperm and
  nothing further happens during epididymal transit, ejaculation or capacitation
  PMID:7888499.
• Alternative splicing. Eleven authentic spliced mRNAs encoding 81–265 aa, all from
  one or two in-frame deletions in exons 2 and/or 3, with the four largest accounting for

  99% of testis message and isoform 1 alone for 53–72%
  PMID:7619499.

Consequence for curation, computed in ACRV1-bioinformatics/: mapping every UniProt
VAR_SEQ deletion onto the folded domain shows all nine isoforms with measurable
abundance keep the Ly-6/uPAR domain complete; only isoforms 8 and 11 (each <1% of
message, combined with the other rare ones) truncate it by 37 residues. The splice
variation therefore lengthens or shortens the low-complexity spacer, not the folded
module. That is why I set no isoform: field on any GOA row: GOA carries none, and the
biology does not suggest isoform-specific function. alternative_products (seeded from
UniProt) records the 11 forms.

4. Function — what is and is not established

The only functional experiments are antibody blocking.

• A mAb against human SP-10 residues 135–149 inhibited sperm–oolemma binding in the
  zona-free hamster egg penetration test but did not inhibit sperm–zona binding in the
  hemizona assay, and the oolemmal ligand is not a beta-1 integrin
  PMID:10775167,
  PMID:10775167.
  The paper is candid that this was the first functional handle
  PMID:10775167.
• In bovine IVF, mono- and polyclonal anti-human-SP-10 reduced fertilisation rates, acting
  on sperm–zona secondary binding and on the ability of capacitated sperm to complete
  the acrosome reaction
  PMID:8882296.

These two look contradictory on the zona point but are not: the hemizona assay scores
primary binding by acrosome-intact sperm, whereas "secondary binding" is the post-acrosome-reaction
attachment of the inner acrosomal membrane to the zona. SP-10 is inaccessible before the
acrosome reaction PMID:2310816, so a role restricted to the post-AR steps is consistent
with both results. Worth stating explicitly because a curator reading only the two titles
would conclude the papers conflict.

The countervailing evidence, which decides how strongly to annotate. CRISPR knockout
of mouse Acrv1 leaves male fertility intact:
PMID:38697008,
PMID:38697008.
The same paper notes that immunising male rodents against SP-10 did not arrest
spermiogenesis
PMID:38697008.

So: participation in the post-acrosome-reaction gamete-binding step is supported (two
species, antibody blocking), but the gene is dispensable for fertility in mouse. In GO
terms that argues for annotating participation (GO:0007342) while not claiming a
requirement, and against GO:0007283 spermatogenesis, which the KO histology and the
immunisation result both fail to support.

Evidence code. A function-blocking antibody is a specific inhibitor, so the inference
runs from a perturbation phenotype, not from direct observation of the protein acting. That
is IMP territory, not IDA — the native-versus-overexpressed distinction that first
tempted me is not the IDA/IMP discriminator. The two proposed localisation terms stay IDA,
because immunogold EM and phase partitioning are direct observations of where the protein is.

One bovine result deliberately not annotated. The same antibodies also
PMID:8882296,
which would suggest GO:0007340 acrosome reaction. Against taking it: the reagents are
anti-human antibodies applied to bovine sperm; the same abstract reports they
PMID:8882296,
so a general effect on capacitated cells is not excluded as the cause; and a protein that
only becomes accessible after the acrosome reaction is awkward to place as a participant in
triggering it. The sperm–oolemma result avoids the second objection because it used human
sperm, a defined-epitope monoclonal, and an assay scoring binding rather than progression.

5. Molecular function — a tested negative, not an assumption

The 1990 cloning paper concluded SP-10 was unique
PMID:1693291.
That is now superseded, and it matters because it is the reason nobody looked for a
family-based function. ACRV1-bioinformatics/RESULTS.md computes the current picture:

• Residues 188–264 are a canonical Ly-6/uPAR (LU, "three-finger") domain — InterPro
  IPR016054, Pfam PF00021, CDD cd23628 (TFP_LU_ECD_SP10_like) — carrying 10 cysteines,
  the modal count across the 38 LU domains of the 30 reviewed human LU-domain proteins.
  Every cysteine in the protein is inside this domain.
• The 166-residue region between the signal peptide and the domain is 18.7% Ser, 17.5%
  Glu, 13.3% Gly and is MobiDB-lite disordered over 62–181; it is the S-E-H-[GA]-S /
  S-G-E-H / [SV]-G-E-Q-[PSA] repeat block of PMID:1693291.
• ACRV1 is one of 11 of 30 reviewed human LU-domain proteins with no GPI anchor, matching
  the experimental finding that it is a chaotrope-releasable peripheral protein
  PMID:1591355. But anchoring is not the discriminator I first took it for: PATE1,
  PATE4, SLURP1 and SLURP2 are unanchored too, and they are precisely the
  acetylcholine-receptor modulators. The discriminator that works is reachability — all 11
  members carrying an acetylcholine-receptor function are annotated to a secreted or
  cell-membrane location where a surface receptor can be engaged, whereas ACRV1's only
  annotated location is the acrosomal lumen, and it is the sole member of the family with no
  extracellular-accessible location at all. PATE4 is the sharpest test of this and survives
  it: it shares ACRV1's acrosomal annotation but is also annotated as secreted.
• ACRV1 is a subfamily singleton in human: neither PTHR17571 nor cd23628 contains another
  reviewed human LU protein, so no paralogue-based transfer is available — only orthologue
  transfer (mouse Acrv1, baboon ACRV1, red fox FSA-ACR.1).
• Across the family, the most widely shared experimental MF is nicotinic acetylcholine
  receptor modulation (7/30). This is the honest complication: PATE1 and PATE4 carry
  GO:0030548, and the CDD model matching ACRV1 is named for SP-10 and the PATE-like
  proteins. Transfer is still unjustified — on topology (above), on the term not being
  family-wide, and on ACRV1 belonging to neither the Ly-6/LYNX/SLURP nor the PATE PANTHER
  subfamily — but the reason has to be stated rather than assumed, and stated correctly.

Conclusion I acted on: no MF term is currently justifiable for ACRV1. Its only
experimental MF annotation is bare GO:0005515, and neither its own record nor its family
supplies a specific alternative.

6. Resolving the GOA WITH/FROM fields

Rows 1–2, GO:0005737 cytoplasm and GO:0031982 vesicle, IBA, GO_REF:0000033.
WITH/FROM = MGI:MGI:104590 | PANTHER:PTN008565525 | UniProtKB:P26436.

• MGI:MGI:104590 resolves (UniProt xref:mgi-104590, bare number — the doubled-prefix
  form 400s) to P50289 / ASPX_MOUSE, mouse Acrv1, the true orthologue.
• Querying what that donor itself carries (QuickGO geneProductId=UniProtKB:P50289,
  goId=GO:0005737/GO:0031982, goUsage=descendants): 5 hits each, of which two are
  its own IDA annotations to GO:0001669 acrosomal vesicle (PMID:1591350,
  PMID:16093322), plus an ISO and an IEA to the same term, plus the IBA row itself.
• GO:0001669 is a confirmed descendant of both GO:0005737 and GO:0031982
  (QuickGO ancestors of GO:0001669 include both).
• PANTHER:PTN008565525 is an internal tree node, not a protein; PTHR17571 contains no
  other reviewed human member.
• UniProtKB:P26436 is the gene itself — a self-referential token, which in PAN-GO
  records the curator treating the human protein as an annotated member of the node, not a
  circular inference.

So the propagation is mechanically sound and the donor is the correct orthologue, but the
term chosen is three levels less precise than the IDA it was drawn from, and human ACRV1
already has its own IDA to GO:0001669. Hence MODIFY on both rows rather than REMOVE:
the statement is true, just needlessly vague, and the fix is to say what the donor's
evidence says. The qualifier goes with it — the replacement is located_in GO:0001669,
matching the two existing rows for that term, not is_active_in, which presupposes a
molecular activity ACRV1 has not been shown to have. One caveat I record rather than
suppress: PAN-GO annotates from a restricted term set, so the vague terms may be pipeline
policy rather than a curator's misjudgement; TERM_SCOPING_PROBLEM describes where the
annotation sits relative to its evidence either way, and the question is put to GO Central.

Rows 4–8, GO:0005515, IPI, PMID:32814053. IntAct REST
(/intact/ws/interaction/findInteractions/P26436) shows all five GOA partners come from
that one publication and each is recorded three times — two hybrid array,
two hybrid pooling, validated two hybrid — every one with both partners
over-expressed, MI-score 0.56. The UniProt NbExp=3 therefore counts three yeast
two-hybrid variants within a single screen, not three independent methods; the aliases
carry CCSB ORFeome clone ids (ACRV1 = CCSB_1488). IntAct also holds two interactions that
GOA did not import (TRIM68, anti-tag co-IP, PMID:33961781; CFTR, ubiquitin
reconstruction, PMID:35156780), so the five are not even the whole physical-interaction
record.

Partner identities (UniProt REST):

Every one is cytosol-facing. ACRV1 has a signal peptide and lives in the lumen of a
secretory organelle, so in vivo it is on the far side of a membrane from all five, and the
Y2H assay put a signal-peptide-bearing ORF in a yeast nucleus where that peptide is
inert. Add the 166-residue disordered low-complexity spacer, a canonical sticky-prey
feature, and the set reads as screen noise. Hence MARK_AS_OVER_ANNOTATED on all five
rather than REMOVE — a Y2H hit is a real observation, it just does not license a
functional claim, and GO:0005515 conveys nothing anyway.

Row 10, GO:0007283 spermatogenesis, NAS, PMID:21252238. The cited paper is entirely
about TDP-43 as a repressor of the mouse acrv1 promoter
PMID:21252238 —
acrv1 is the transcriptional target, and the paper contains no assay of ACRV1 protein
function. Combined with the fertile mouse null (§4), this is an expression-timing
statement dressed as a process annotation.

7. Ectopic expression in tumours

A 2026 study reports ZNF280A recruiting CUX2 to the ACRV1 promoter in ovarian cancer, with
ACRV1 knockdown attenuating ZNF280A-driven AKT phosphorylation and glycolysis
PMID:41338461,
PMID:41338461.
This is the only loss/gain-of-function
work on human ACRV1, but it is in cell lines aberrantly expressing a testis-restricted
protein, so it describes a consequence of ectopic expression, not a normal function. Kept
out of the annotations; raised as a question instead.

8. Actions taken

Two ACCEPT, two MODIFY, six MARK_AS_OVER_ANNOTATED, three NEW; no REMOVE. Nothing needed
UNDECIDED — every cited paper was retrievable, though only two of the fourteen cited PMIDs
(PMID:38697008, PMID:41338461) have full text cached; the other twelve are abstract-only, which is
why every conclusion here is anchored to a sentence that appears in the abstract rather than to an
inference about what the full text might contain.

9. Process log

• just fetch-gene human ACRV1 — 10 GOA rows, 6 seeded (the five IPI rows collapse to one
  in the stub; expanded back to one entry per GOA row, in GOA order).
• affinage_deep_research.py human ACRV1 --write — trust gates clear at fetch time (no
  self_evaluation_pairwise score), faith_pct: 100,
  13 citations, all numeric PMIDs. Two claims in its narrative needed correction against
  the sources rather than being taken on trust: it presents the sperm–zona results as a
  flat contradiction, and it repeats the 1990 "no homology" conclusion without noting the
  LU domain. Both are handled above.
• Bioinformatics: ACRV1-bioinformatics/fetch_data.py snapshots UniProt / InterPro /
  QuickGO into data/; analyze.py recomputes results.json + RESULTS.md from the
  snapshot with no network access, so a fresh run reproduces the committed report
  byte-for-byte (verified with diff).