GPR37 (O15354, GPR37_HUMAN / Pael-R / ETBR-LP-1) — curation notes

The problem

GPR37 is one of the most over-deorphanized receptors in the literature. Four
different endogenous ligands have been proposed, from three different chemical
classes, and the 2025-2026 literature asserts several of them simultaneously
without the proposing groups engaging one another. Separately, GPR37 has a
completely non-ligand body of work — it is the parkin substrate "Pael-R" — that
is arguably its most reproducible molecular phenotype.

1. Head activator — refuted, and the GOA rows should go

GO:0008188 neuropeptide receptor activity (IDA), GO:0007218 neuropeptide signaling pathway (IDA) and GO:0042923 neuropeptide binding (IPI) all trace to
PMID:16443751. The IPI WITH/FROM field is the giveaway: UniProtKB:P69251,
which is MORN_HYDVU, "Morphogenetic neuropeptide / Head activator" of
Hydra vulgaris. So a human GPCR carries an experimental binding annotation
whose partner is a cnidarian peptide.

The claim did not survive:

• [PMID:26635605 "These studies contrast with that of Dunham et al. (2009), who attempted to replicate the finding that HA was a ligand for GPR37 but found no evidence of HA-mediated internalization, ERK1/2 phosphorylation or cAMP stimulation. HA was also included in a larger screen of all remaining orphan GPCRs but did not register as a "hit" for any GPCR tested (Southern et al., 2013). Finally, perhaps the most damning evidence against HA as the endogenous ligand for GPR37 is the fact that it has not been found in the human genome (Davenport et al., 2013). Thus, it appears that HA is unlikely to be an agonist at GPR37 and is certainly not its endogenous ligand."]
• Even the prosaposin paper, which is friendly to the HA result, concedes HA's
  standing is analogical rather than orthologous:
  PMID:23690594

This is not "second-guessing a curator I disagree with": it is a published,
direct failure to replicate by an independent laboratory, plus the non-existence
of the ligand in the human genome. That is exactly the "genuinely contradicted
function" case for which REMOVE exists. No information is lost by removing them,
because the peptide-receptor content survives in GO:0008528 and GO:0036505,
which rest on the prosaposin work.

The GO:0005737 cytoplasm IDA from the same paper reflects
PMID:16443751
— i.e. ER retention of a misfolding-prone multipass membrane protein, not a
cytosolic pool. MODIFY to GO:0005789 endoplasmic reticulum membrane.

2. Prosaposin — the pairing GOA already carries, and the one I keep

Meyer et al. screened orphan neuropeptides and got exactly one hit
PMID:23690594
with Gi-type signalling and ERK phosphorylation
PMID:23690594

The 2015 authoritative review was sceptical, for reasons worth recording:
PMID:26635605
and PMID:26635605

But the 2015 objection has largely been answered. The Ji laboratory (Duke),
independent of the Hall laboratory that made the original pairing, reproduced
TX14 binding and GPR37-dependent signalling
PMID:30010619
and in 2026 showed that the in vivo actions of intrathecal TX14A require GPR37 in
a defined neuronal population
PMID:42144155

So: two laboratories, cell-free binding + heterologous signalling + cell-type-specific
conditional knockout in vivo. I ACCEPT GO:0036505 prosaposin receptor activity as
core, and record the residual reservations (potency, IUPHAR non-ratification, the
possibility that prosaposin also acts as a folding chaperone for GPR37) in the
reason field rather than pretending they do not exist.

3. Protectin D1 — a real second ligand, of a different chemical class

The brief asks whether "a peptide receptor and a lipid receptor are not easily the
same receptor". They are not — and yet the same experiments, in the same paper,
report both. PMID:30010619 tested NPD1 (= protectin D1) and TX14 side by side and
found both bind GPR37 and both raise Ca2+ in a GPR37-dependent way. The lipid arm
has since been replicated across models: sepsis/infection PMID:33731716,
chemotherapy-induced neuropathy PMID:40287118, spinal nociceptive sensitization
PMID:42144155, and a 2026 rat pituitary study that treats both as established
PMID:42649016

Two honest readings:
1. GPR37 is genuinely polypharmacological — peptide at one site, lipid at another
(plausible for a class A GPCR with a very long N-terminus that is constitutively
shed, PMID:26869225).
2. The two "ligands" are converging on a shared downstream Ca2+/phagocytosis
readout that is not a clean measure of orthosteric occupancy, and one of them is
not a direct agonist.

Nobody has done the experiment that distinguishes these: competition binding of
PD1 against labelled TX14A on the same receptor preparation. Most of the
protectin D1 work shares a senior author (Ji). So the chemical-class objection
stands, and the corroboration is less independent than the paper count suggests.

There is no GO term for protectin D1 receptor activity, and GOA carries no lipid
MF for GPR37. I therefore propose a new term rather than shoehorning it into
GO:0045125 bioactive lipid receptor activity, and flag the unresolved
competition-binding question.

4. Osteocalcin — reproduced, but not independently

Original pairing: PMID:34678058
2026 extension: PMID:41679312

Two caveats that matter for curation:
- These are not independent groups. Zhengjiang Qian, Xiang Li and Liping Wang
are authors on both the 2021 Sci Adv paper and the 2026 Neuron paper, from the
same Shenzhen institute. A second paper from the same lab is corroboration, not
replication.
- The Neuron paper carries a published erratum spanning four journal pages
(PMID:42001851, Neuron 2026;114:1695-1698), which is substantial for an erratum.
- OCN already has two other claimed receptors, GPRC6A (periphery) and GPR158
(brain); a third is not impossible but raises the bar.

No GO annotation for osteocalcin exists on GPR37 and no osteocalcin receptor GO
term exists at all. I propose the term and leave the pairing out of
core_functions.

5. Pael-R / parkin — the best-replicated molecular fact about GPR37

PMID:11439185, with
PMID:12150907 and a second E3
PMID:17059562

Is this better supported than any ligand claim? In reproducibility, yes — it
comes from at least three laboratories, was found by unbiased interaction
screening, is echoed in the UniProt PTM block, and is corroborated by the
independent observation that GPR37 aggregates and is ER-retained on overexpression
(PMID:16443751, PMID:25977097 for the ASD R558Q mutant). But it is not the same
kind of claim. Being a parkin/HRD1 substrate describes ER quality control of an
aggregation-prone GPCR; it is a property the protein has because it folds badly,
not a function it evolved to perform. I therefore keep these annotations as
KEEP_AS_NON_CORE rather than promoting them over the receptor function, and I
mark GO:0000151 ubiquitin ligase complex as over-annotated, because a substrate
held by an E3 is not a component of the ligase.

Actions summary

• REMOVE (3): GO:0008188, GO:0007218, GO:0042923 — all head-activator-derived.
• MODIFY (1): GO:0005737 cytoplasm -> GO:0005789 endoplasmic reticulum membrane.
• MARK_AS_OVER_ANNOTATED: GO:0005515 protein binding (x4 references),
  GO:0000151 ubiquitin ligase complex.
• KEEP_AS_NON_CORE: GO:0031625, GO:0030544, GO:0031072, GO:0042277,
  GO:0034614, GO:0016358, GO:0043235.
• ACCEPT: the prosaposin-receptor / Gi / MAPK / localisation set.
• NEW: nothing added to existing_annotations for the lipid or osteocalcin
  pairings; both are recorded as proposed_new_terms plus suggested_questions,
  because in my judgement neither has the independent replication that a
  ligand-specific MF assertion requires.

No NOT/negated qualifiers are present in the GOA file.