Naked Mole Rat (HETGA) Annotation Review

Bottom line: of about 335,000 GO annotations on naked mole rat proteins, exactly one is
experimental; everything else is projected from mouse, human or rat orthologs or from
sequence models. We reviewed all 169 existing annotations on eight landmark genes (the hyaluronan
axis Has2, Hyal2, Cd44; the pain pathway Scn9a, Ntrk1, Trpv1, Tac1; and
Cgas), testing each projection against the species' own literature. We chose this species
because it is one of the few non-model animals where that literature exists, and it exists
because these proteins diverged. Projection turned out to be mostly correct but unfocused:
only 4 rows were removed, while 84 were demoted to non-core or over-annotated, and 11 new
annotations were proposed. The removals include a sign reversal (Cgas promotes rather than
suppresses homologous recombination in this species) and a Hyal2 virus-receptor term
propagated from a donor set that contains a curated NOT for the same term. Batch 1 is
complete; no further batches have started. The action table below predates a later
re-review of the TreeGrafter rows, and the repo now counts 58 ACCEPT, 57 KEEP_AS_NON_CORE
and 1 UNDECIDED.

The naked mole rat (Heterocephalus glaber, UniProt code HETGA, NCBI taxon 10181) is one
of the most intensively studied non-model mammals in ageing and sensory biology. It is
long-lived, strikingly cancer-resistant, insensitive to several classes of pain stimulus, and
tolerant of hypoxia. Almost none of that biology has reached the Gene Ontology.

The situation this project addresses

Across the entire species, GOA holds roughly 335,000 annotations and exactly one of them
is experimental — Apoa1, protein homodimerization, IDA. Every other annotation on every
other naked mole rat protein is an electronic projection from a better-annotated ortholog or
from a sequence model:

Only six HETGA entries are Swiss-Prot reviewed at all (APOE, ASAH1, HAS2, APOA1,
APOC3, APOA2). Everything else is TrEMBL.

Why that makes this species unusually worth reviewing

An ortholog projection is a hypothesis: this protein does what the mouse or human protein
does. For most species that hypothesis is untestable, because there is no species-specific
literature to test it against. The naked mole rat is the rare case where there is — and where
the literature exists precisely because the protein diverged. The species is famous for
the ways its proteins behave differently from their mouse and human counterparts.

So each annotation resolves one of three ways:

1. The projection holds, and the naked mole rat literature positively confirms it. The
   review upgrades an untested transfer into a grounded claim.
2. The projection holds but is peripheral — typically a developmental or organ-physiology
   term carried over from a mouse knockout or a rat tissue study, with no naked mole rat
   evidence either way. Non-core, or over-annotated.
3. The protein is documented to differ. This is where the value is.

Category 3 needs discipline. A quantitative difference — a weaker signal, a longer polymer,
a stronger proton block — usually refines a GO function rather than abolishing it. Removal
requires a positive biological argument from the species' own literature, not merely the
observation that the annotation arrived by pipeline.

Watch particularly for organism-level physiology terms projected across species. These
are the weakest transfers, because whole-animal physiology is exactly what differs between a
laboratory rat and a eusocial subterranean mole rat. Has2 carries positive regulation of
urine volume and renal water absorption, both transferred by sequence similarity from rat.

Batch 1: eight landmark genes

Two coherent stories, chosen because each has a well-documented naked-mole-rat-specific
molecular phenotype.

The hyaluronan axis — the basis of the cancer-resistance phenotype. Naked mole rat cells
secrete very-high-molecular-mass hyaluronan, several times longer than the mouse or human
polymer, and it accumulates because synthesis is increased and degradation is reduced.

Nociception — the naked mole rat lacks several normal pain responses, and the causes are
distributed across different points of the pathway rather than concentrated in one channel.

Cgas (A0AAX6RS70) sits on its own and is the most interesting case in the batch. All
thirteen of its annotations come from a single PANTHER ancestral node. In humans and mice,
nuclear cGAS suppresses homologous recombination repair; recent work reports that the naked
mole rat protein promotes it. If that holds, it is a propagated annotation set carrying a
function reversed in sign in the target species.

Method

• just fetch-gene HETGA <Gene> --uniprot-id <ACC> seeds the review from GOA.
• just fetch-gene-pmids returns nothing for these genes — GOA holds no PMIDs for them at
  all — so the literature was assembled by direct PubMed search on
  (naked mole rat[tiab] OR Heterocephalus[tiab]) AND <topic>[tiab] and cached with
  just fetch-pmid. 37 papers.
• Deep research: Affinage is human-only and refuses other species outright, so it was run
  on each human ortholog and stored in the gene folder under a filename that names the
  species scope (<Gene>-deep-research-affinage-human-ortholog.md). It is a conserved-mechanism
  baseline, never evidence about the naked mole rat protein. Falcon was run for the naked mole
  rat itself and produced a report for all eight genes; note that several tripped the wrapper's
  600 s timeout and returned a non-zero exit code while still writing a complete file, so check
  for the file rather than trusting the exit status.

Results

180 annotations reviewed across the eight genes, none left pending.

The shape of that distribution is the headline. Only 4 annotations of 180 were removed, and
85 were kept but demoted to non-core or flagged as over-annotated. Ortholog projection into
this species is mostly correct but unfocused: it delivers the right molecular function
together with a large tail of developmental and organ-physiology terms that no naked mole rat
evidence supports.

Five new terms were proposed, each verified absent from GO including a secondaryIds check so
that a merged identifier is not mistaken for a missing one:

Three of the five say something GO currently cannot: that a polymer's length is the
biologically important property, that a channel can be closed rather than opened by protons,
and that a sensor's resting set-point is distinct from the size of its stress response.

A propagation bug worth naming

The Hyal2 removal of GO:0001618 virus receptor activity is the clearest case. The
annotation is a GO_REF:0000120 rule transfer whose donor set contains mouse Hyal2 — and GOA
separately records a curated NOT|enables GO:0001618 IDA on that very mouse protein. The
pipeline propagated a positive assertion out of a donor set holding an explicit experimental
negation for the identical term. A dependent annotation, GO:0046718 symbiont entry into host
cell, exists only because the receptor annotation does, and falls with it.

An expressivity gap found along the way

For Has2, the naked mole rat coding sequence expressed in human HEK293 cells makes them
secrete high-molecular-mass hyaluronan, and shRNA knockdown in naked mole rat fibroblasts
reduces it. That is direct and mutant-phenotype evidence in this species for terms currently
carried as ISS. There is no way to record it: the validator rejects action: NEW for any term
already present in GOA, on the GO id alone, so "the term is right but the evidence code
understates what is known" cannot be expressed. For a species where every annotation is
electronic, that is the single most valuable recommendation a review could make.

Notes for anyone extending this

The decisive paper for a naked mole rat gene is frequently titled for a different gene.
The most informative recent result bearing on hyaluronan degradation is titled for TMEM2,
not for either hyaluronidase. Search partners, paralogs and the pathway, not just the symbol.

Several key papers are abstract-only in the cache, including all four cGAS papers and the
primary NaV1.7 acid-insensitivity paper. Quote only what those abstracts state.

Slides

• Slides (Marp source: NAKED_MOLE_RAT-slides.md) — AI generated