Reviewer journal for the GO annotation review. Every assertion below carries provenance.
UniProt entry: A0AAX6R0R7_HETGA, 701 aa, TrEMBL, PE 4: Predicted, derived from RefSeq
model XP_012930388.1. 38 GOA rows, all IEA; zero experimental annotations, in line
with the species-wide picture.
This is the one place where NMR-specific, on-the-protein evidence exists, and it is good.
Tian et al. 2013 blocked CD44 on intact NMR cells with an antibody and got a
proliferation phenotype:
The same blockade removes the transformation barrier:
and the authors place CD44 in a defined axis:
with the cytoplasmic partner named explicitly:
The HA-affinity FACS assay is a whole-cell measurement, not a CD44-specific one, and
I have been careful not to over-read it:
The methods make the assay design explicit — fluorescein-labelled bovine HA on whole
cells read by FACS — so the two-fold difference is attributable to the cell surface as a
whole (CD44 plus RHAMM, LYVE1, layilin and HA synthase-tethered pericellular coat), not
to CD44 alone. It is supporting context for a CD44 hyaluronan-binding annotation, not a
direct measurement of it.
Independent restatement in a 2023 comparative-genomics paper:
and in the "myths" review, which is careful to note the pathway is not classical
cell-contact-dependent inhibition:
Curation consequence. The proliferation phenotype is real and CD44-dependent, but the
mechanism is a soluble/pericellular hyaluronan signal read at the cell surface, not
information transmitted "by direct cell-cell contact" as GO:0060242 contact inhibition
requires. I therefore proposed GO:0008285 negative regulation of cell population
proliferation as the NEW annotation rather than a contact-inhibition term. This is a
small but real ontology-fit issue and I recorded it as a suggested question.
Takasugi et al. 2023 is the most important paper for this review and proposes a CD44
function that is nowhere in the projected GOA set. Crucially, part of the work is done
in naked mole-rat cells with naked mole-rat CD44, so it is not merely a cross-species
inference:
The overall claim, and its explicit hyaluronan-independence:
The sufficiency experiment is done in human CD44-KO cells but is mechanistically
decisive: an ER-retained CD44 ectodomain reproduces the effect, so the function is
exerted from inside the ER rather than from the plasma membrane.
How much of this is NMR-specific? Honest answer: the mechanism is presented as
general mammalian biology (IMR90 human fibroblasts, U2OS, mouse OPCs), and the
NMR-specific claim is one of expression level, not of a different protein activity —
CD44 is unusually highly expressed in NMR OPCs and CD44 expression correlates with maximum
lifespan across mammals:
So the correct framing for the review is: this is a conserved CD44 function that the
naked mole rat uses more of, not a naked-mole-rat-specific invention. That is enough to
justify NEW annotations on the NMR protein, because the loss-of-function experiment was
done on NMR CD44 in NMR cells.
Ontology fit is imperfect. The paper is explicit that the effect is on basal ATF6
tone, not on the stress-induced UPR:
GO:1903893 positive regulation of ATF6-mediated unfolded protein response is the closest
existing term and I used it, but it does not distinguish setting the basal tone of a
sensor from amplifying its stress response. Recorded as a proposed new term and as a
knowledge gap.
The 2020 cytoprotection paper establishes that CD44 is the receptor through which NMR
very-high-molecular-mass HA acts:
Important caveat I kept in front of me throughout: these experiments put NMR
hyaluronan onto human IMR90 cells, so the receptor tested is human CD44. The paper is
evidence about the ligand's species-specific properties, not about the NMR receptor
protein. The key mechanistic result is a negative one for several projected
annotations:
i.e. the very-high-mass hyaluronan that NMR tissue is full of suppresses CD44
clustering and CD44 protein-protein interactions rather than driving them. A review of
NMR hyaluronan restates this:
This bears directly on the receptor-clustering-dependent outputs projected from human —
ERK activation, migration, endocytic uptake — and is the main reason several of those are
marked non-core rather than accepted.
This is abstract-only (full_text_available: false), so I have not quoted any method
detail beyond what the abstract states. Two things it does establish and one it does not:
Curation consequence. This is a positive, NMR-specific argument for downgrading
GO:0005576 extracellular region (which for a type-I membrane protein means the shed
soluble ectodomain) and for treating the migration/lamellipodium cluster as non-core. It
is not enough for REMOVE, because the defect is demonstrably in the cell, not in the
protein, and is rescuable.
A separate, independent line of speculation points the same way — that the material
properties of NMR hyaluronan may themselves protect the receptor from cleavage:
(that paper's cleavage discussion is explicitly speculative and I did not use it as
evidence).
Two independent NMR results:
(the TMEM2 paper is abstract-only; it reports that NMR TMEM2 carries Asn247/Val302 in
place of the catalytic His/Ala and is inactive.)
Curation consequence. GO:0030214 hyaluronan catabolic process was projected from
human, where CD44 is the receptor that routes HA to HYAL1/HYAL2 (PMID:17170110), not an
enzyme. In the naked mole rat, the defining tissue phenotype is hyaluronan accumulation
through reduced degradation. Projecting a catabolic-process role onto NMR CD44 is exactly
the kind of transfer this species falsifies. Marked over-annotated (not removed — nobody
has tested whether NMR CD44 internalises HA).
Note the term actually projected is GO:0051132 NK T cell activation, which by definition
concerns natural killer T cells, a distinct αβ-TCR lineage. I did not conflate the two:
the cached literature says NMRs lack NK cells and says nothing about their NKT compartment.
What tips this row to over-annotated is the combination of (a) no NMR evidence, (b) an
atypical NMR NK/lymphoid compartment, and (c) a term-scoping question at the source
(below).
Written up in Cd44-bioinformatics/RESULTS.md, script
cd44_isoform_architecture.py. Two questions the UniProt record could not answer.
(a) Is the 701-aa entry a particular isoform? The TrEMBL record carries no
ALTERNATIVE PRODUCTS and no VAR_SEQ, so no isoform identity can be read off it — and I
have not asserted one. What the alignment to human P16070-1 does show is architectural:
86.3% (270/313) of the human alternatively spliced insert (residues 223–535, the segment
missing from the short human isoform 11/CD44R2) has an aligned counterpart in the NMR
protein, and the NMR ectodomain is 588 residues versus 629 for the full-length human
canonical form. So the RefSeq gene model chosen as the reference protein encodes a long,
variant-exon-containing (CD44v-like) form, not the short standard CD44s. That is a
statement about the gene model, not about what NMR tissues express — no cached study
reports NMR CD44 splice usage, and this remains a genuine gap.
Does it matter for annotation? Only marginally, and I said so rather than inventing a
consequence:
- The Link/hyaluronan-binding module sits in the invariant N-terminal region present in
every isoform, so GO:0005540 is unaffected by splice form.
- Variant exons are where isoform-specific co-receptor functions live in human CD44
(heparan-sulfate-bearing v3 presenting growth factors; v6 in MET signalling), so a
variant-containing model is at least architecturally compatible with
GO:0044344 cellular response to fibroblast growth factor stimulus. That is an argument
about capability, not evidence, and I did not use it to accept the annotation.
- The variant stem is the least conserved region (74.1% identity vs 92.1% for the Link
domain and 97.2% for the cytoplasmic tail), which is what one expects for a mucin-like
O-glycosylated spacer.
(b) Does the UniProt CAUTION undermine hyaluronan binding? The entry says
[file:HETGA/Cd44/Cd44-uniprot.txt "Lacks conserved residue(s) required for the propagation
of"] feature annotation, cited to PROSITE-ProRule PRU00323. PRU00323 is the Link-domain
rule (trigger PS50963 LINK_2), and its only feature output is two DISULFID bonds
gated on a C-x*-C condition — so the flag concerns automatic feature propagation,
not function. Mapping the human landmarks through the alignment shows the residues that
matter are all present:
| Human | NMR | role (UniProt P16070 feature table) |
|---|---|---|
| R41 | R43 | hyaluronan binding |
| R78 | R80 | hyaluronan binding |
| Y79 | Y81 | hyaluronan binding |
| Y105 | Y107 | hyaluronan binding |
| C28/C129, C53/C118, C77/C97 | C30/C132, C55/C120, C79/C99 | three Link-region disulfides |
Link domain identity to human is 92.1%. So the CAUTION is not evidence against
GO:0005540, and I recorded that explicitly so a later reader does not mistake it for one.
Cd44-deep-research-affinage-human-ortholog.md is an Affinage record for human CD44
(P16070), fetched deliberately as a conserved-mechanism baseline; its frontmatter and
heading both say human. I used it only for mechanism, never as evidence about the naked
mole rat, and I did not import its mechanism_profile GO ids.
What it gave me. A clean mechanistic spine for the conserved protein: HA endocytosis
to lysosomes with no intrinsic catalytic activity (PMID:1370836); glycosylation-gated,
avidity-based HA recognition (PMID:8601595, PMID:10871609); ADAM10 shedding followed by
γ-secretase release of CD44-ICD (PMID:14623895, PMID:15596040); IQGAP1/RhoA/Syk coupling
via the cytoplasmic tail. The ADAM10 entry in particular is what let me read
PMID:36790936 correctly — knowing that ADAM10-dependent shedding is a canonical,
ligation-and-Rac1-augmented CD44 step made the NMR "absence of shedding" result
interpretable rather than isolated.
What it missed — and this is the substantive recall finding. The affinage record does
not contain the ER/ATF6/proteostasis function at all. PMID:37708026 is absent from its
20 citations, and its localization profile lists plasma membrane, nucleus and lysosome —
no ER. That is the single most consequential recent finding about CD44, it is a 2023 Cell
Reports paper with the gene name in the title, and it is the only line of work in which
naked-mole-rat CD44 itself was knocked down and overexpressed. Everything in §1.2, and all
three NEW annotations I proposed, come from outside the provider record.
It also misses, unsurprisingly given its human scope, the entire naked-mole-rat literature:
the CD44-blocking-antibody experiments (PMID:23783513), the vHMM-HA/CD44 interactome
inversion (PMID:32398747), and the absence of ADAM10-mediated CD44 shedding in NMR
fibroblasts (PMID:36790936).
A smaller point of tension worth recording: the affinage entry for PMID:7545465 states
that CD44 engagement inhibits anti-CD3- and dexamethasone-induced apoptosis in T cells
"but not UV-induced (p53-dependent) apoptosis". Two of the Ensembl-projected rows here
(GO:0043518, GO:1902166) assert negative regulation of p53-mediated DNA-damage
signalling. Those come from a different paper and a different cell system, but the
juxtaposition is a reason not to treat the p53 rows as core.
All 38 rows are electronic. There is no curator who read a full text on this protein, so
"do not overrule the curator" is not the binding constraint — but REMOVE still needs a
positive biological argument, and I did not reach that bar for any row. Nothing was
removed.
Provenance of the projections matters for calibration, so I traced the human source
annotations behind the GO_REF:0000107 rows via QuickGO. They are all grounded in human
experimental annotations, and several rows share a source:
| Human source | Evidence | Rows projected to NMR |
|---|---|---|
| PMID:17045821 (MIF–CD74–CD44) | IDA | GO:0004896, GO:0035692, GO:0043518, GO:1902166, GO:0070374 |
| PMID:20962267 (podoplanin) | IDA/IMP | GO:0016324, GO:0031258, GO:0044319, GO:2000392 |
| PMID:16945930 (acylation/rafts) | IDA/IMP | GO:0038024, GO:0045121 |
| PMID:17170110 (HYAL1/2 catabolism) | IDA | GO:0030214, GO:0005540 |
| PMID:15100360 (BMP-7, renal) | IMP | GO:0034116, GO:0070487 |
| PMID:11944887 (chondrogenesis) | IEP | GO:0051216 |
| PMID:19577615 (bFGF, fibrosarcoma) | IDA | GO:0044344 |
| PMID:37006235 (galectin-9, NK cells) | IDA | GO:0038023, GO:0051132 |
| PMID:20522558 (cytokines, endothelium) | IMP | GO:1900625 |
Two observations fell out of that table:
GO:0051216 cartilage development rests on an IEP — an expression-patternA term-scoping issue at the source, flagged rather than resolved: PMID:37006235 is titled
for natural killer cells but grounds GO:0051132 NK T cell activation, which is
defined for natural killer T cells. I have not read that paper's full text and do not
assert the curator erred; I record it as an unresolved concern that adds to the case for
treating the row as an over-annotation in this species.
GO:0034116, GO:0070487 and GO:1900625 all descend from monocyte-adhesion assays in
renal-tubular or endothelial disease models. These are the classic weak transfers the
naked mole rat is most likely to break, and there is no NMR evidence for any of them.
GO_REF:0000108 logical inferencesThese are GO-to-GO inferences with no organism evidence at all, and I treated them
differently from each other:
GO:0016192 vesicle-mediated transport ← GO:0038024 cargo receptor activity. TheGO:0019221 cytokine-mediated signaling pathway ← GO:0004896 cytokine receptor
activity. Worth noting that the source MF row carries contributes_to — CD44 does notinvolved_in, so the inferenceGO:0044344 is marked over-annotatedGO:0030214 andGO:0038024 are marked over-annotated on that reasoning; a direct assay could flipGO:0006954, GO:0042110, GO:2000106 and the NK row rest