UniProt: Q9BXJ8. HGNC: TMEM120A. Paralog: TMEM120B (Q9BXJ9).
Reviewed 2026-09.
TMEM120A carries two mutually incompatible stories in the literature, and GOA carries
both of them at once. The review has to adjudicate which one the annotations should
reflect.
Story A (2020-2021, "TACAN"): a plasma-membrane mechanosensitive ion channel in
nociceptors. PMID:32084332 and PMID:32084332
Story B (2021-2026): an ER membrane protein that binds coenzyme A and works in
glycerolipid synthesis.
Four independent structural/electrophysiology groups published within weeks of each other
in 2021 and none could reproduce the channel activity:
All four instead found a coenzyme A molecule bound in a deep cavity of the
six-transmembrane barrel, and a fold homologous to the fatty-acid elongase ELOVL7:
PMID:34374645, PMID:34409941, PMID:34409941
GO has already ruled on this. In TMEM120A-goa.tsv the QUALIFIER column shows
GO:0005216 monoatomic ion channel activity three times, and every one is NOT|enables:
IMP/PMID:34374645, IDA/PMID:34409941, IDA/PMID:34465718. These are negative assertions —
GO is stating that TMEM120A does not enable ion channel activity.
One row contradicts the three negatives on the same term in the same protein:
UniProtKB Q9BXJ8 TMEM120A enables GO:0005216 ... ECO:0000250 ISS GO_REF:0000024 UniProtKB:Q8C1E7 ... 20200311
Provenance traced: the source is mouse Q8C1E7, whose QuickGO record carries
enables GO:0005216 from PMID:32084332 — i.e. the original Beaulieu-Laroche TACAN
paper. The transfer is dated 2020-03-11, three weeks after that paper and more than a
year before the 2021 refutations. It is a stale orthology projection of the very claim
that the three later human experimental annotations negate. Removing it is not
second-guessing an experimentalist: it is retiring a curator-judgment sequence-similarity
transfer whose donor evidence has been superseded on the same protein. Action: REMOVE.
The three negated rows are ACCEPTed — negated and non-negated annotations of one term
are opposite assertions, so they legitimately take different actions.
The same 2020-03-11 GO_REF:0000024 block also projected GO:0034220 monoatomic ion
transmembrane transport, GO:0050966 detection of mechanical stimulus involved in sensory
perception of pain and GO:0005886 plasma membrane from the same mouse paper, and
GO_REF:0000107 (Ensembl Compara) later duplicated the first two. Those are handled as a
block.
It is an ER protein. PMID:41423633 Independently, with endogenous
tagging in C. elegans: PMID:42098142
It partners acyl-CoA synthetases. PMID:41423633 The interaction depends on the CoA pocket:
PMID:41423633
It activates GPAT4. PMID:42098142 and
PMID:42098142 The
activation is measured directly in vitro: PMID:42098142 and
PMID:42098142 It is not itself the acyltransferase — CHP1
was previously "the only known GPAT4 activator" PMID:42098142 — and the phenotype is conserved to worm
PMID:42098142
It is not an elongase. Despite the ELOVL7-like fold, the pocket cannot hold an acyl
chain: PMID:41423633 So the right MF altitude is not a catalytic term.
TMEM120A modulates mechanotransduction without being a mechanosensor, by changing membrane
lipid composition:
LPA is precisely the product of the GPAT4 reaction that TMEM120A activates
PMID:42098142. So: TMEM120A → GPAT4 activation → LPA/PA →
PIEZO2 inhibition. That is a coherent, lipid-metabolic explanation for why manipulating
TMEM120A in sensory neurons changes mechanical sensitivity, with no channel required.
Caveat recorded honestly: the sign does not obviously line up. Beaulieu-Laroche found KO
reduces mechanical pain; Del Rosario found knockdown increases PIEZO2 currents and
lowers thresholds. Both cannot be the whole story. This goes into suggested_questions.
Neither is grounds to reinstate enables GO:0005216 against three experimental negatives.
GO:0120225 coenzyme A binding is directly demonstrated (cryo-EM density, CoA assay kit,
native MS, W193/W193A loss-of-binding) and is mechanistically load-bearing rather than
incidental — W193A abolishes both the ACSL1/3 interaction and, in worm, function
PMID:42098142. So it is kept as a core MF, not modified away.
But CoA binding alone does not say what the protein does with the CoA. The 2026 result is
an explicit enzyme-activation activity. GO has GO:0008047 enzyme activator activity
(checked via QuickGO: "A molecular function regulator that increases a catalytic activity")
and has precedent for acyltransferase-specific children, e.g.
GO:0060228 phosphatidylcholine-sterol O-acyltransferase activator activity. There is no
glycerol-3-phosphate O-acyltransferase activator activity term (QuickGO search returns
none), so one is proposed, with GO:0008047 used as the core-function MF in the interim.
Deliberately not proposed: a CoA transmembrane transporter term. The authors hedge
their own mechanism PMID:41423633 — "channeling" in their title is metabolic channelling, not demonstrated
transport. Asserting a transporter term would repeat exactly the error that produced the
TACAN channel annotation.
The same-term-consistency rule does not model negation. For GO:0005216 the three
ACCEPTed negated rows are anchored to PMIDs and the single REMOVEd positive row is
anchored to GO_REF:0000024; because the REMOVE reference is disjoint from every kept
reference and all kept rows share one action, the validator's citation-specific-rejection
exemption applies and no inconsistency warning is raised.