ADAMTSL5 (Q6ZMM2) — review notes

PAINT + affinage campaign. Branch paint/ADAMTSL5.

What the gene is

ADAMTSL5 (ADAMTS-like protein 5; synonym THSD6) is a small secreted, N-glycosylated
extracellular-matrix glycoprotein of the ADAMTS superfamily. At 481 aa it is the
smallest member of the family by a wide margin (ADAMTSL1 1762, ADAMTSL2 951, ADAMTSL3
1691, ADAMTSL4 1074, THSD4 1018, PAPLN 1278 aa). Domain architecture from the UniProt
feature table: signal peptide 1–42, mature chain 43–481, a single TSP type-1 domain
(45–97), and a C-terminal NTR module (360–479), joined by a proline-rich,
disordered segment (331–361).

The catalytic question — established, not assumed

The campaign brief flagged "the ADAMTSL proteins lack the catalytic metalloprotease
domain" as a lead to establish. It holds for ADAMTSL5, on three independent lines:

• UniProt CAUTION: [file:human/ADAMTSL5/ADAMTSL5-uniprot.txt "lacks the metalloprotease and disintegrin-like domains which are typical of that family"]
• The gene's own primary paper: PMID:23010571
• Sequence scan (this review): the zinc-binding signature HExxHxxGxxHD is absent, and
  there is no HExxH substring at all in the 481-residue sequence. Its InterPro
  match set contains no peptidase/reprolysin (M12B) signature.

But the predicted annotation error did not occur. ADAMTSL5's GOA contains no
peptidase term of any kind — no GO:0004222, no GO:0008237, no GO:0006508. I
recorded this as a hypothesis that was not confirmed, not as a finding. PAINT
actually got this right: the catalytic terms at node PTN000347317 reached all four
catalytic ADAMTS controls and none of the seven non-catalytic members, and ADAMTSL2
even carries an explicit NOT|enables GO:0004222 (IKR at PTN002673039).

The shape of the error does appear one aspect over, in BP — see GO:0030198 below.

Annotation-by-annotation reasoning

12 GOA rows. The fetch-gene stub seeded only 9 — it collapsed the three
GO:0005515 partner rows into one and the two GO:0031012 IDA rows (MGI- and
UniProt-assigned) into one. Restored to 12 so every row gets its own verdict.

ACCEPT — the experimental core (5 rows, all PMID:23010571)

GO:0008201 heparin binding IDA, GO:0050436 microfibril binding IDA,
GO:0031012 extracellular matrix IDA ×2, GO:0005576 extracellular region IDA.

Heparin binding is well documented and localised to the NTR module:
PMID:23010571,
with the interaction shown to be ionic
PMID:23010571.

Microfibril binding rests on direct affinity co-isolation with both fibrillins plus
colocalisation:
PMID:23010571
GO:0050436 is_a GO:0050840 extracellular matrix binding, and there is no
"fibrillin binding" term in GO (OLS returns nothing), so GO:0050436 is the most
specific molecular function available.

Caveat noted but not acted on: these assays used recombinant/exogenous ADAMTSL5
added to fibroblast cultures rather than native protein, which by the campaign's rule
of thumb leans toward IMP. I did not overturn the curators' IDA calls — for a secreted
protein this is standard practice, and the authors ran a non-permeabilised staining
control confirming the signal was extracellular
PMID:23010571.

ACCEPT — GO:0031012 IBA, and it is unusually well founded

WITH/FROM has 17 tokens = 16 protein donors + the node PANTHER:PTN000347317. This
matches the cached PAINT seed list exactly (asserted in code). All 16 protein
donors carry their own experimental IDA/HDA for the term or a descendant, so
SOURCE_WEAK_OR_INFERRED would be contradicted by my own measurement. One token is
self-referential (UniProtKB:Q6ZMM2) — a PAINT curator judging the function core,
which is valid, hence NO_FAILURE_CORE.

The donor set spans four distinct locations — extracellular matrix (14), basement
membrane (4), interstitial matrix (1), microfibril (1). GRANULARITY_MISMATCH requires
donors to agree; they do not, so GO:0031012 is the LCA and refining it would mean
arbitrarily preferring one donor's compartment. ACCEPT with no specificity upgrade.

Also checked and negative (the ACRV1 shape): the IBA does not land above its donors —
14 of 16 hold the term itself — so no downward MODIFY is warranted.

ACCEPT — GO:0005576 IEA from UniProtKB-SubCell:SL-0243

Straightforward: a signal peptide (1–42) plus experimentally demonstrated secretion.
The SubCell→GO mapping is doing exactly what it should.

MARK_AS_OVER_ANNOTATED — GO:0030198 extracellular matrix organization IEA

Two independent legs:

1. The source signature bundles proteases with non-proteases.
   InterPro:IPR013273 is named, literally, "ADAMTS/ADAMTS-like" (26,580 proteins) and
   carries exactly one GO mapping — GO:0030198. For the catalytic ADAMTS members, ECM
   organization follows from ECM proteolysis. ADAMTSL5 has no such activity. This is the
   familiar "a domain's name is not an activity" failure moved one aspect over: a
   family signature too broad to carry a process.
2. The only direct test of it was negative.
   PMID:23010571
   The abstract puts it as colocalisation "but without discernible effect on microfibril
   assembly". Direct binding to fibronectin was also not supported.

Why not REMOVE. The negative is a "data not shown" result from a single
exogenous-protein assay. It shows no role has been demonstrated; it does not refute
one. UniProt itself hedges — "May play a role in modulation of fibrillin microfibrils".
MARK_AS_OVER_ANNOTATED is the honest ceiling.

A third leg I deliberately did NOT use. ADAMTSL5 has no GO:0030198 IBA, while
ADAMTSL2/ADAMTSL4/THSD4 got one from the same node. That looks like PAINT declining to
propagate the process to ADAMTSL5 — but the propagation is incoherent family-wide (see
below), and ADAMTSL1 and PAPLN received nothing at all. So the absence is more likely
a propagation gap than a judgement, and leaning on it would be rationalising a number I
could not explain.

MARK_AS_OVER_ANNOTATED — the three GO:0005515 rows

All three from PMID:32296183 (HuRI). IntAct shows each logged under three
sub-methods of one experiment — two hybrid array + two hybrid prey pooling approach + validated two hybrid, MI-score 0.56. UniProt's NbExp=3 is therefore
one screen counted three ways — the ACRV1 finding, replicated here on a second gene.

Distinct partner counts (derived as entity sets; IntAct records are not partners):

Every partner is topologically incompatible with a signal-peptide secreted ECM protein.
Decided per partner as the brief requires; all three independently come out the
same. All three resolve to reviewed canonical Swiss-Prot entries at canonical lengths —
no ORFeome/TrEMBL substitution of the ACRV1 kind (negative result, recorded).

I differ here from the merged ADAMTSL4 review, which used REMOVE on four
comparable Y2H GO:0005515 rows. Per this campaign's convention an unreplicated screen
hit is MARK_AS_OVER_ANNOTATED, and REMOVE is reserved for demonstrably wrong
inferences. Flagged as a cross-family inconsistency rather than silently diverging.

MODIFY — GO:0071953 elastic fiber TAS → GO:0001527 microfibril

The TAS source PMID:23962539 is a review (full_text_available: false) used as
the reference for 62 distinct entities, assigning GO:0071953 to 41 of them.
Its abstract never mentions ADAMTSL5. The same curation from the same review assigned
the more specific GO:0001527 microfibril to 15 other proteins including
THSD4/ADAMTSL6 — so the specific term was available and simply not chosen here.

Meanwhile the gene's own primary data supports microfibril association specifically,
and UniProt already records GO:0001527; C:microfibril; IDA:UniProtKB — a term
GOA does not carry (verified: QuickGO returns exactly 12 annotations for Q6ZMM2 and
GO:0001527 is not among them). GO:0001527 is current and is a part_of child of
GO:0071953, so the replacement retains the parent by closure while gaining precision.
The merged ADAMTSL4 review independently proposed GO:0001527 as a NEW term, which is
useful convergent support.

The main PAINT finding: one node, incoherent propagation

All eight human ADAMTSL/papilin proteins sit in PTHR13723. Node PTN000347317
carries four IBD terms. Who received them:

From a single node, GO:0031012 reached 5 of 7 and GO:0030198 reached 3 of 7, in no
biologically coherent pattern. The sharpest case is PAPLN, which received nothing
although three of its own orthologs are seeds for GO:0031012 at that very node —
fly Ppn (FB:FBgn0003137), mouse Papln (MGI:MGI:2386139) and worm mig-6/ppn-1
(WB:WBGene00003242). Human PAPLN and human ADAMTSL1 both have no IBA at all.

This is not the "right term, wrong node" defect (AADACL) nor the "mis-placed member"
defect (ACTL8). It is a propagation-coverage defect: the node and its term
assignments look correct, and the term simply failed to reach some descendants.

Reported once in suggested_questions, naming all affected genes, rather than repeated
per gene.

Cross-check with the sibling agents (ADAMTSL1, ADAMTSL3)

Derived independently from QuickGO rather than read off their branches:

• ADAMTSL3 carries an GO:0031012 IBA identical to mine — same node
  PTN000347317, same 17 WITH/FROM tokens. Our two derivations should agree; if that
  review reaches a different verdict on the same row, the disagreement is the finding.
  ADAMTSL3 also shares the GO:0071953 TAS from PMID:23962539 and the InterPro
  GO:0030198 IEA — so all three of my non-ACCEPT verdicts have a direct counterpart
  there.
• ADAMTSL1 has only 4 annotations and no IBA at all despite being in
  PTHR13723 (SF157). Its only BP row is the same InterPro GO:0030198 IEA. If that
  review reports ADAMTSL1 as having IBA support, that is a discrepancy worth resolving.

An ontology issue this surfaced

GO:0001527 microfibril is part_of GO:0071953 elastic fiber, which asserts every
microfibril is part of an elastic fiber. That contradicts GO:0001527's own
definition — "Extracellular matrix components occurring independently or along
with elastin" — and contradicts the ciliary zonule, a fibrillin-microfibril structure
essentially devoid of elastin (the reason FBN1 mutations cause ectopia lentis).
ADAMTSL5's own expression pattern is a further mismatch: cartilage and bone are not
elastic-fibre tissues
PMID:23010571

Non-GO biology (informs description, not annotations)

• Psoriasis autoantigen. ADAMTSL5 is an HLA-C*06:02-presented melanocyte
  autoantigen PMID:26621454.
  The recognised peptide VRSRRCLRL maps to residues 67–75, inside the TSP type-1
  domain (verified against the UniProt sequence). This is a property of the peptide in
  an MHC groove, not a molecular function of the secreted protein, so it generates no
  GO annotation.
• Hepatocellular carcinoma. PMID:33197513
  Knockdown/overexpression in a cancer context; I did not propose a signalling GO term
  from it, since the RTK effects are downstream and the mechanism connecting a secreted
  ECM protein to receptor levels is not established.

Provider record (affinage)

gates_passed: False. The tripped gate is specifically the self-evaluation pairwise
tie (self_evaluation_pairwise: tie), not a faithfulness failure — faith_pct is
100.0. All 7 citations are numeric PMIDs; no PMID:bio_* bioRxiv ids. I verified
every claim I used against the cited PMID directly and quoted the PMIDs, never the
provider's prose. The narrative is broadly accurate on this gene; its GO grounding
however proposes GO:0008289 lipid binding and GO:0098772 molecular function regulator activity, neither of which has any support — the ligand demonstrated is
heparin, a glycosaminoglycan (GO:0008201), not a lipid. Not imported.

Checks run, including the ones that came back negative

• Retraction / erratum / expression of concern on all 9 PMIDs relied on, via PubMed
  publication types and each record's own CommentsCorrections (a Publisher
  Correction is invisible to a pubtype query): none flagged. PMID:26621454 has two
  Comment in entries — commentaries, not corrections.
• Reference projection test: PMID:23010571 annotates only 2 entities
  (ADAMTSL5 + mouse ortholog) — gene-specific curation, not a projection. Negative.
• Partner accession integrity: all three IPI partners reviewed, canonical, correct
  length. Negative.
• IBA-above-donor precision defect: absent. Negative.
• Dead-accession check: every WITH/FROM UniProt lookup returns a named entry; the
  resolver raises rather than reporting a silent zero.

Process notes worth carrying to the next gene

• QuickGO's geneProductId rejects MOD ids (MGI:, FB:, WB:, RGD:) with
  HTTP 400 in every form. Donor evidence can only be queried through the resolved
  UniProt accession. My first script printed "NONE" for all 13 MOD tokens — a silent
  degradation reading as a null result, which would have inverted the conclusion from
  "16/16 donors have their own experimental evidence" to "almost none do".
• UniProt does not index WBGene ids (xref:wormbase-WBGene00003242 → []); it keys
  WormBase xrefs on the CDS name. Resolve WBGene → symbol via the WormBase REST first.
• That symbol search must use gene_exact:. Fuzzy gene:mig-6 also returns
  mig-10/mig-5/mig-14/mig-18 and puts the wrong entry first. The 16/16 count would
  otherwise have been right for the wrong reason.
• QuickGO strips the DB prefix from WITH/FROM tokens. withFrom[].connectedXrefs[]
  returns {"db": "FB", "id": "FBgn0003137"} as two fields, so the id alone is
  FBgn0003137, not FB:FBgn0003137. Comparing that set naively against a GOA TSV
  WITH/FROM string reports 17 vs 17 tokens that are "not identical" for data that is
  in fact the same — the token must be reassembled as db + ":" + id first. I hit this
  while verifying the ADAMTSL3 cross-check; the sibling comparison itself was sound
  (both sides came from QuickGO, so both were in the stripped serialisation), but the
  same-size-yet-different result is exactly the kind of discrepancy that must be
  investigated rather than waved through.