UniProt G5AY81 (HYAS2_HETGA), Swiss-Prot reviewed, 552 aa, NCBI taxon 10181.
GOA: 28 rows collapsing to 24 unique (term, evidence, reference) entries. Every row is
electronic: ISS (GO_REF:0000024, from human Q92819, mouse P70312, rat O35776) or IEA
(GO_REF:0000044 SubCell, GO_REF:0000118 TreeGrafter, GO_REF:0000120 combined). There are
no experimental GO annotations on this protein.
The single most important fact for this review is that GOA's evidence codes understate
what is known. Tian et al. cloned the NMR HAS2 cDNA and expressed it heterologously:
Methods confirm this is a direct assay of the product of the NMR coding sequence in a
heterologous host:
That is IDA-grade evidence for GO:0050501 hyaluronan synthase activity on G5AY81, not
an ortholog projection. Independently, loss of function in NMR cells abolishes the product:
which is IMP-grade evidence for GO:0030213 hyaluronan biosynthetic process. The
functional consequence:
The gain-of-function complement was later done in a whole animal — a transgenic mouse
carrying the NMR gene:
So the core MF and BP are supported by NMR-specific gain-of-function (heterologous cells
and transgenic mice) and loss-of-function (shRNA in NMR fibroblasts). This is unusually
strong for a species with essentially no experimental GOA coverage.
Comparative work refines this: one of the two is not NMR-private, and the gene is otherwise
strongly constrained across mammals:
The purifying-selection result matters for curation: it is a positive argument that the
conserved, ancestral functions of HAS2 (catalysis, ECM output, the developmental roles) are
retained in the NMR, and that the NMR difference is a refinement on top, not a
replacement.
The HA phenotype is a two-sided balance — synthesis up, catabolism down. The catabolic side
is not Has2's doing:
HAS isoenzymes differ systematically in the length of the polymer they extrude, and HAS2 is
the long-polymer synthase. This is generic mammalian biology, not an NMR peculiarity:
The original NMR claim:
This magnitude is genuinely contested. Del Marmol et al. re-measured by size-exclusion
chromatography and gel electrophoresis and could not reproduce it:
The consensus "myths" review adjudicates in favour of the qualitative claim while
rejecting the quantitative one:
It also flags that the mechanistic link from the two substitutions to processivity is not
established:
Direct evidence that HAS2 sequence (not just expression level) sets polymer length does
exist in a sister subterranean species:
Why length matters biologically (and therefore why it deserves an ontology term):
Curation stance taken: I describe the NMR product as unusually long / very-high-molecular-
mass and cite both the 6–12 MDa and the ≤2.5 MDa measurements, rather than asserting either
number as settled. The GO annotations at stake (GO:0050501, GO:0030213) do not depend on the
magnitude, only on the qualitative fact that HAS2 makes HA, which is not in dispute.
These support GO:0085029 extracellular matrix assembly as a real, direct downstream process
for HAS2 output, but they are properties of the polymer, not of the enzyme, so they
inform the BP annotation rather than adding new MF claims.
GO:0035810 positive regulation of urine volume and GO:0070295 renal water absorption are
ISS transfers from rat Has2 (UniProtKB:O35776). Both are organ-level renal-physiology
terms. The rat basis is medullary interstitial HA:
Del Marmol et al. quantified HA in NMR kidney cortex and medulla separately, and the renal
medulla is the single tissue where the NMR runs against the species trend:
and, importantly, the authors say the actual physiological test has never been done in NMR:
There is a countervailing observation from the original paper, based on whole-kidney alcian
blue rather than quantitative cortex/medulla assay:
and the transgenic mouse does raise kidney HA when the NMR gene is expressed
PMID:37612507, but that is a mouse kidney driven by a CAG promoter, not NMR renal
physiology.
Conclusion: the rat organ-physiology terms are transferred on sequence similarity alone,
into a species whose renal medullary HA is the one measurement that does not follow the NMR
pattern and whose urine-concentrating physiology is explicitly described as unremarkable. This
is not enough to call the function contradicted — HA is present in the NMR medulla around the
vasa recta PMID:33846452 —
but it is enough to call the transfer an over-annotation. Both get
MARK_AS_OVER_ANNOTATED, not REMOVE, with the resolving experiment named (a water-loading
protocol with medullary HA quantification and urine osmolality).
GO:0001570 vasculogenesis, GO:0036302 atrioventricular canal development, and
GO:0090500 endocardial cushion to mesenchymal transition come by ISS from mouse Has2
(UniProtKB:P70312), grounded in mouse Has2 knockouts. There is no NMR-specific evidence
for or against them. Two things argue for retaining rather than deleting:
But they are developmental processes of a pleiotropic ECM enzyme, not what HAS2 is for in
the NMR, so they are KEEP_AS_NON_CORE.
HAS enzymes are polytopic plasma-membrane proteins that polymerise HA on the cytoplasmic face
and extrude it directly outward:
So GO:0005886 plasma membrane is the functional site (ACCEPT, core). The remaining CC terms
(GO:0005789 ER membrane, GO:0000139 Golgi membrane, GO:0005794 Golgi apparatus,
GO:0005764 lysosome, GO:0031982 vesicle, GO:1903561 extracellular vesicle) are all
trafficking-itinerary compartments taken from the human SubCell record; UniProt itself
describes them as a route ("Travels from endoplasmic reticulum (ER), Golgi to plasma membrane
and either back to endosomes and lysosomes, or out into extracellular vesicles"). None has
NMR-specific support. They are kept as non-core, with two term-level fixes:
GO:0005794 Golgi apparatus → MODIFY to GO:0000139 Golgi membrane: HAS2 is a seven-passGO:0000271 polysaccharide biosynthetic process → MODIFY to GO:0030213 hyaluronan
biosynthetic process. Verified against QuickGO: GO:0000271 sits underRe-verified independently (QuickGO /ontology/go/terms/<id>/complete plus a text search of
the whole ontology). GO contains exactly eight hyaluronan terms:
| GO id | name |
|---|---|
| GO:0030212 | hyaluronan metabolic process |
| GO:0030213 | hyaluronan biosynthetic process |
| GO:0030214 | hyaluronan catabolic process |
| GO:0050501 | hyaluronan synthase activity |
| GO:0005540 | hyaluronic acid binding |
| GO:1900125 | regulation of hyaluronan biosynthetic process |
| GO:1900126 | negative regulation of hyaluronan biosynthetic process |
| GO:1900127 | positive regulation of hyaluronan biosynthetic process |
GO:0030213 has no is_a children at all (only the three regulates children above),
and none of the eight carries a secondaryIds entry, so no HMM-HA term has been merged away
either. The absence is real.
Since the biologically decisive variable is polymer length — HAS1/HAS3 short vs HAS2 long as
a general mammalian rule, and length-dependent cytoprotection as a measured effect — a child
of GO:0030213 is proposed. Note this is not an NMR-specific request: it would apply to
mammalian HAS2 generally, with the NMR simply being the extreme case.
I considered proposing the distinction at MF level instead (a "high-molecular-mass hyaluronan
synthase activity" child of GO:0050501), since processivity is an intrinsic enzyme property.
I did not, because GO MF terms for glycosyltransferases are defined by the reaction and the
reaction is identical regardless of product length; and because the NMR phenotype demonstrably
depends on the synthesis/degradation balance, which is a process-level property. This is
flagged as an open question for the ontology editors rather than settled here.
Has2-deep-research-affinage-human-ortholog.md is the Affinage record for human HAS2
(Q92819), used here only as a conserved-mechanism baseline. It is good on human mechanism
(AMPK Thr-110 phosphorylation, K190 ubiquitination, Ser-221 O-GlcNAc, HAS1/HAS3 heteromers,
ATG9A-dependent autophagic turnover, the transcriptional inputs), and its 32 citations are
plausible. But measured against what this review actually needed:
mechanism_profile grounds the MF at GO:0016740 transferase activity — three levelsGO:0050501. As instructed, none of its GO ids were imported.This is the expected failure mode for a human-only provider on a species-divergence question:
correct conserved mechanism, zero coverage of what makes the ortholog interesting.
Action counts across the 24 unique GOA entries (28 GOA rows; the seeding key omits WITH/FROM,
so the three GO:0050501 ISS rows, the two GO:0030213 ISS rows and the two GO:0005886 ISS rows
each collapse into one entry — no distinct GO term is lost):
| action | n | terms |
|---|---|---|
| ACCEPT | 8 | GO:0050501 ×2, GO:0030213 ×2, GO:0085029 ×2, GO:0005886 ×2 |
| KEEP_AS_NON_CORE | 12 | GO:0000139 ×2, GO:0005764 ×2, GO:0005789 ×2, GO:0031982 ×2, GO:1903561, GO:0001570, GO:0036302, GO:0090500 |
| MODIFY | 2 | GO:0005794 → GO:0000139; GO:0000271 → GO:0030213 |
| MARK_AS_OVER_ANNOTATED | 2 | GO:0035810, GO:0070295 |
| REMOVE / UNDECIDED | 0 | — |
I initially added two action: NEW entries recording that this protein qualifies for
experimental annotations — GO:0050501 IDA and GO:0030213 IMP, both against
PMID:23783513 — because GOA currently holds no PMID-backed annotation on G5AY81 at all,
despite the heterologous-expression and shRNA-knockdown experiments. The best-practices
validator rejects NEW for any term already present in GOA regardless of evidence code
("Annotation with action=NEW exists in GOA: GO:0030213"), so both were removed and the
recommendation now lives in the review.reason of the corresponding ACCEPT rows.
This is a genuine expressivity gap in the review format, worth flagging: there is currently no
way to say "this term is correctly annotated but the evidence code understates the evidence
available for this species." For a species like the naked mole rat, where the entire GOA is
electronic, that is exactly the recommendation a curator most wants to make.
just validate HETGA Has2 reports ✓ Valid (with 1 warnings); term validation and reference
validation both pass cleanly (all 47 supporting_text quotes verified as verbatim substrings
before writing, and reference titles copied verbatim from the cache).
The remaining warning is no_deep_research_results: "No annotations reference available deep
research files". Satisfying it requires a supported_by entry whose reference_id is the
file: deep-research path — i.e. quoting an Affinage sentence as supporting_text. The brief
forbids that ("never quote an affinage sentence as supporting_text for a mechanistic claim — a
provider sentence is a lead, not evidence"), and the Affinage record here is about the human
ortholog, so any such quote would also be evidence about the wrong species. The record's
provenance is instead recorded properly via additional_reference_ids on the two annotations
whose reasons draw on it (GO:0005789, GO:0005764) and on GO:0050501, plus a full
reference_review. The warning is left standing deliberately.