Focused Hypothesis Report — Horse OLFML2A (A0A9L0SKW1) localizes to the extracellular matrix (GO:0031012) OpenScientist openscientist-autonomous 2 artifacts 2026-09-08T16:12:22.029227

OpenScientist final report (text/html)

OpenScientist final report (application/pdf)

Focused Hypothesis Report — Horse OLFML2A (A0A9L0SKW1) localizes to the extracellular matrix (GO:0031012)

Focus type: function_assignment (CC localization) · Term: GO:0031012 extracellular matrix
Target: Equus caballus (NCBITaxon:9796), UniProt A0A9L0SKW1, 599 aa, PE=4 (Predicted), Ensembl gene model ENSECAG00000023908 / protein ENSECAP00000076704.
Human comparison lead: Q68BL7 (OLFML2A / Photomedin-1), 652 aa, PE=1.

Executive Judgment

Partially supported (support via justified mammalian orthology transfer; one important sequence caveat).

Identity confirmed. The supplied 599-aa sequence is byte-identical to the current UniProt A0A9L0SKW1 record and is an unambiguous OLFML2A/photomedin-1 ortholog: 87.2% identical to human OLFML2A over the mature chain, retaining the diagnostic domain architecture (N-terminal tandem CXCXCX9C cysteine motifs, mid-molecule coiled-coil, C-terminal olfactomedin domain).
The ortholog is genuinely an ECM protein. Direct experimental work on the mouse ortholog (photomedin-1) shows it is secreted, disulfide-bonded, O-glycosylated, and binds the ECM glycosaminoglycans chondroitin sulfate-E and heparin (PMID 15836428). Human OLFML2A is PE1 with UniProt CC "Secreted." This justifies ECM localization by mammalian transfer.
Decisive caveat on the exact sequence. The supplied horse ORF begins exactly at the mature-chain boundary and lacks the entire N-terminal signal peptide. Horse residue 2 aligns to human residue 28 (the first mature residue, immediately after human SIGNAL 1..27), and the horse N-terminus has no hydrophobic h-region (max Kyte-Doolittle window-7 = 0.83 vs 3.09 in the human signal).
The locus DOES encode secretion-competent isoforms (isoform resolution). Ensembl gene ENSECAG00000023908 has 6 protein-coding transcripts; the canonical is 662 aa (ENSECAP00000071146, N-term MAPGTSGNLLLAVCPW…, hydrophobic, KD max 2.61) and isoform ENSECAP00000056820 (618 aa) starts MAAAALPPRP…, matching the human OLFML2A signal peptide (KD max 3.03). NCBI RefSeq likewise lists full-length models (XP_023484611.2, 650 aa; XP_023484612.1, 621 aa). The supplied A0A9L0SKW1 = ENSECAP00000076704 (599 aa) is a minor, non-canonical, N-terminally-truncated isoform — so the missing signal peptide is an isoform/gene-model choice, not loss of secretion by the gene.
Functional ECM-binding domain is intact. The C-terminal olfactomedin (OLF) domain in the supplied 599-aa model (residues 341–599) is 99.6% identical to the human OLF domain (258/259) and ends at the natural C-terminus (…HFVV). The GAG/proteoglycan-binding module that mediates ECM association is therefore fully preserved; only the N-terminal signal peptide is missing.
Paralog confusion ruled out. Smith-Waterman: horse vs human OLFML2A (Q68BL7) = 983 vs human OLFML2B/photomedin-2 (Q68BL8) = 294 — unambiguously OLFML2A.
The horse UniProt "Secreted" tag is ARBA rule-based (automated), not experimental, and must not be treated as biological validation.

Bottom line for the curator: ECM/secreted localization for horse OLFML2A is well supported by orthology to an experimentally characterized ECM protein, but is not directly observed in horse and is not encoded by the supplied truncated ORF. GO:0031012 is defensible as an ISS/ISO annotation with the signal-peptide-truncation caveat noted; it should not be asserted as directly-observed horse evidence.

Evidence Matrix

#	Citation	Evidence type	Stance	Claim tested	Key finding	Context	Confidence / limitations
1	This report (computed)	Structural/evolutionary (Smith-Waterman, own run)	Supports (identity)	Is A0A9L0SKW1 truly OLFML2A?	Horse 2–599 ↔ human 28–652, score 983, 87.2% identity; retains OLF domain (UniProt 341–599) + N-terminal CxCxCx9C motifs + coiled-coil	Horse vs human protein sequence	High; identity/architecture unambiguous
2	This report (computed)	Computational (hydrophobicity + alignment)	Qualifies / competing	Does the exact sequence encode a signal peptide?	Horse res.2 = human res.28 (mature-chain start); no signal peptide, no hydrophobic N-term (KD max 0.83 vs human 3.09); UniProt lists no SIGNAL/TM	Exact 599-aa ORF	High for the ORF; interpreted as 5′-truncated gene model
3	PMID 15836428 (Furutani et al., 2005, Biochem J)	Direct assay + localization	Supports (ortholog)	Is OLFML2A/photomedin-1 an ECM protein?	Identified in an ECM-protein screen; secreted disulfide-bonded dimer, O-glycosylated; binds chondroitin sulfate-E and heparin; retinal photoreceptor outer segment	Mouse cDNA / recombinant protein; retina	High for ortholog; species = mouse, not horse
4	UniProt Q68BL7 (human)	Review/database	Supports (ortholog)	Human OLFML2A localization/features	PE1; SIGNAL 1..27, CHAIN 28..652, OLF 394..652, COILED 157..183; KW Signal/Glycoprotein/Disulfide/Secreted; CC "Secreted"	Human	High; "Secreted" is broader than ECM
5	UniProt A0A9L0SKW1 (horse)	Database (automated)	Qualifies	Does the record itself assert ECM?	CC "Secreted" is ARBA automated; no SIGNAL feature; no explicit GO:0031012; PE4 Predicted	Horse	Automated only — not evidence
6	Ensembl ENSECAG00000023908 + NCBI RefSeq (own queries)	Computational (isoform resolution)	Qualifies (resolves caveat)	Does the locus encode signal-peptide isoforms?	Canonical 662 aa (`MAPGTSGNLLLAVCPW`, KD 2.61) and isoform 618 aa (`MAAAALPPRP`= human signal peptide, KD 3.03) have signal peptides; supplied 599-aa model is a minor truncated isoform (KD 0.83)	Horse locus, chr25	High; shows truncation is isoform-specific, not gene-level
7	UniProt Q68BL8 vs Q68BL7 (own SW)	Structural/evolutionary	Supports (identity)	Is it OLFML2A not the OLFML2B paralog?	SW horse↔OLFML2A=983 vs ↔OLFML2B=294	Horse vs human	High; paralog confusion excluded

Provenance: sequence identity check, Smith-Waterman alignment, N-terminal Kyte-Doolittle hydrophobicity scan, and glyc/cysteine mapping were executed in-session (code + numeric outputs retained in the iteration transcript). SHA-256 could not be recomputed in-sandbox (hashlib/struct blocked), but the provided sequence was verified byte-identical to the live UniProt A0A9L0SKW1 FASTA and length 599 matches the frozen spec.

GO Curation Implications (leads — require curator verification)

GO:0031012 (extracellular matrix, CC): Biologically defensible for the ortholog — photomedin-1 was identified as an ECM protein and binds ECM proteoglycan GAG chains (PMID 15836428). Recommend annotating by ISO/ISS from mouse/human, with evidence code reflecting orthology (not IDA), and a NOT-fully-encoded-signal-peptide caveat on the horse model.
Alternative/parent term consideration: UniProt curators annotate the human protein only as "Secreted" (≈ GO:0005576 extracellular region). GO:0031012 is more specific than the minimal database claim; it is supported by the CS-E/heparin-binding assay but a conservative curator could prefer GO:0005576 extracellular region if requiring the more specific ECM residence to be demonstrated for a mammalian ortholog. Either is more informative than "protein binding."
Supportable companion terms (leads): MF — glycosaminoglycan binding / heparin binding (GO:0005539 / GO:0008201) from PMID 15836428; CC — GO:0005576 extracellular region (parent, safest).
Do not cite the horse record's ARBA "Secreted" as independent support; it is a prediction, not validation.

Mechanistic Scope

Immediate molecular/cellular function being tested is localization/residence, not catalysis. The gene product is a secreted olfactomedin-domain glycoprotein whose extracellular-matrix association is a direct binding property (olfactomedin/coiled-coil scaffold binding chondroitin sulfate-E– and heparin-type GAGs on proteoglycans). ECM residence is thus a primary property of the mature protein, not a downstream phenotype. The N-terminal CXCXCX9C motifs mediate disulfide-linked dimerization; the signal peptide (present in orthologs) drives secretory-pathway entry.

Conflicts and Alternatives

Isoform-specific truncation (principal conflict, now resolved): the supplied ORF lacks the signal peptide, so a naïve SignalP/secretion check on the exact 599-aa sequence would be negative. Isoform resolution shows this is one alternative product of the locus: the canonical 662-aa protein and a 618-aa isoform (MAAAALPPRP… = human signal peptide) are secretion-competent, so the truncation is an isoform/gene-model choice, not loss of secretion by the gene.
Term-scope mismatch: databases annotate the human protein as "Secreted" (extracellular region), which is broader than "extracellular matrix"; ECM specificity rests on the mouse binding assay.
Paralog note — resolved: OLFML2B (photomedin-2) is a close paralog with the same architecture and ECM binding. SW scoring (horse↔OLFML2A 983 vs ↔OLFML2B 294) unambiguously assigns the horse locus to OLFML2A.
Species evidence gap: no horse-specific localization data exist; all functional evidence is mouse/human.

Knowledge Gaps

Does the true horse transcript encode a signal peptide? — largely RESOLVED. Checked Ensembl (6 transcripts) and NCBI RefSeq: canonical 662-aa and 618-aa (MAAAALPPRP) isoforms carry hydrophobic signal peptides; the supplied 599-aa model does not. Remaining gap: which isoform(s) predominate in horse tissue (RNA-seq/proteomics) and whether the 599-aa ORF is a real product or a mis-prediction/NMD substrate.
Is ECM (vs generic extracellular) the right granularity for horse? Checked: orthologs bind ECM GAGs (mouse). Resolve with horse tissue localization or GAG-binding assay.
Locus disambiguation OLFML2A vs OLFML2B. Resolve via synteny/phylogenetic tree of equine olfactomedin genes.

Discriminating Tests

Signal-peptide resolution: SignalP-6.0 / DeepTMHMM on (a) the supplied 599-aa ORF and (b) a reconstructed full-length horse ORF from genomic + 5′-UTR; compare.
Transcript completeness: map equine RNA-seq (e.g., FAANG) to ENSECAG00000023908 to confirm the true start codon and signal-peptide exon.
Orthology tree: ML phylogeny of OLFML2A/OLFML2B across mammals to lock identity.
Localization/binding (if wet-lab): express horse OLFML2A, test secretion and CS-E/heparin binding as in PMID 15836428.

Curation Leads (verify before applying)

Reference to add: PMID 15836428 — snippets to verify: "characterized by the presence of tandem CXCXCX9C motifs in the N-terminal region, a coiled-coil domain and an olfactomedin domain"; "secreted as disulphide-bonded dimers (photomedin-1) … with O-linked carbohydrate chains"; "photomedins preferentially bound to chondroitin sulphate-E and heparin."
Candidate GO action: retain GO:0031012 as an ISO/ISS (orthology) annotation with a curator note that the horse Ensembl model is signal-peptide-truncated; optionally back off to GO:0005576 if ECM specificity for a mammalian ortholog is deemed insufficiently transferable. Consider adding GO:0008201/GO:0005539 (heparin/GAG binding, MF) as leads.
Suggested question: Is the horse OLFML2A gene model 5′-complete? If not, flag the "Secreted"/ECM annotation as inference pending a corrected model.
Do NOT rely on the horse record's ARBA "Secreted" or agreement with predictions as validation.

Provenance Tables (computed this run)

Signal-peptide analysis across horse OLFML2A isoforms (Ensembl ENSECAG00000023908; Kyte-Doolittle window-7):

Ensembl protein	length (aa)	status	N-terminus	KD max	KD windows >1.6	interpretation
ENSECAP00000071146	662	canonical	MAPGTSGNLLLAVCPW	2.61	4	signal-peptide-like (secretion-competent)
ENSECAP00000056820	618	non-canonical	MAAAALPPRPRPRP	3.03	9	= human signal peptide (secretion-competent)
ENSECAP00000036732	621	non-canonical	MDSQVFGDMDQVRM	0.83	0	NO signal peptide (truncated)
ENSECAP00000076704	599	TARGET = A0A9L0SKW1	MDSQVFGDMDQVRM	0.83	0	NO signal peptide (truncated)

Paralog disambiguation (Smith-Waterman, match +2/mismatch −1/gap −2):

Comparison	SW score	Call
horse A0A9L0SKW1 vs human OLFML2A (Q68BL7)	983	assigned
horse A0A9L0SKW1 vs human OLFML2B / photomedin-2 (Q68BL8)	294	excluded

OLF domain integrity: horse 341–599 vs human 394–652 → 258/259 = 99.6% identity; both start SCEGTLRAVDPPVRHHSYGR and end …HFVV.

GO decision table (leads — verify):

GO id	label	aspect	recommended action	basis
GO:0031012	extracellular matrix	CC	Retain as ISO/ISS orthology annotation + caveat	Ortholog is ECM protein (PMID 15836428); locus has signal-peptide isoforms; supplied 599-aa model lacks signal peptide but has intact OLF domain (99.6% id)
GO:0005576	extracellular region	CC	Safer parent alternative	Matches UniProt "Secreted"; use if ECM specificity deemed insufficiently transferable
GO:0008201 / GO:0005539	heparin / GAG binding	MF	Candidate lead	Photomedin-1 binds CS-E and heparin (PMID 15836428)

(Provenance note: CSV export to disk was blocked by the read-only execution sandbox; tables above are the verbatim computed outputs retained from the executed code.)

Limitations: No horse-specific experimental data; functional evidence is mouse/human orthologs. Signal-peptide absence is inferred from the current Ensembl ORF and may be corrected by future gene models. SHA-256 not recomputed in-sandbox (equivalent byte-identity verified against live UniProt FASTA).