Human PAINT/affinage campaign. Sources: UniProt P01019 (ANGT_HUMAN, 476 aa), the
GOA TSV (114 rows), AGT-deep-research-affinage.md, the PANTHER PAINT slice for
PTHR11461, the cached publications, and three analyses in AGT-bioinformatics/.
AGT is a heavily annotated gene whose annotation set is not really about AGT.
Sorting the 114 GOA rows by what molecule was actually assayed gives three layers:
| layer | what was assayed | rows |
|---|---|---|
| 1. the precursor itself | full-length secreted AGT protein | ~6 |
| 2. the released peptides | synthetic angiotensin I/II/(1-7) added to cells, or Ang II binding a receptor | ~45 |
| 3. neither | Reactome reaction membership, plasma proteomics, editorial statements, family inference | ~63 |
Layer 1 — the biology that is uniquely angiotensinogen's, and the only layer where
the 476-residue protein is the experimental subject — is almost entirely missing
from GOA. The single row that touches it is a bare GO:0005515 protein binding
against renin.
That framing decided most of the actions below. It did not lead me to strip
the layer-2 peptide rows: GOA's annotation unit is the UniProt accession, and
UniProt itself hangs the angiotensin functions on P01019 (as FUNCTION:
[Angiotensin-2], FUNCTION: [Angiotensin-3], FUNCTION: [Angiotensin 1-7]
sub-entries and as PRO_ chain-level IntAct entries). There is no separate gene
for angiotensin II. Peptide pharmacology belongs on AGT. What does not belong is
(a) a serpin activity AGT provably lacks, and (b) treating every cellular
response to added Ang II as a process the gene product is involved in.
Three rows give AGT GO:0004867 serine-type endopeptidase inhibitor activity:
an IBA, an IEA from InterPro, and a TAS. The GO definition is "Binds to and
stops, prevents or reduces the activity of a serine-type endopeptidase", which
is falsifiable, and it is false.
Four independent lines say so.
Explicit statement in the primary structural literature.
PMID:20927107
Structural demonstration that the inhibitory mechanism is absent. Serpins
inhibit by cleaving at the reactive centre loop and undergoing a stressed-to-relaxed
transition that traps the protease. Yan et al. crystallised loop-cleaved AGT
specifically to test this:
PMID:30563843
and, in the discussion,
PMID:30563843
Sequence features (AGT-bioinformatics/serpin_inhibitory.py, 19-serpin panel).
UniProt annotates a SITE "Reactive bond" for 10 of the 12 inhibitors in the
panel and for AGT it annotates none. At the hinge position equivalent to
SERPINA1 Ala371 (P12), AGT carries Pro430 — a proline where loop insertion
into β-sheet A requires a small residue. AGT's whole P17–P9 hinge reads
ADEREPTES (P01019 residues 425–433) against SERPINA1's EKGTEAAGA, 2/4 versus
4/4 small residues at P12–P9.
MEROPS. AGT is I04.953. MEROPS reserves the .9xx range of family I04 for
non-inhibitor homologues. Every one of the 7 resolvable seed proteins of the IBD
node that donated the term is in the inhibitor range (I04.001, I04.004, I04.027,
I04.082, …). 7/7.
And the physiology: AGT's protease partner is renin, an aspartyl protease, and
AGT is renin's substrate, not its inhibitor. The GO term is doubly wrong —
wrong catalytic class and wrong role.
The PAINT slice (interpro/panther/PTHR11461/PTHR11461-paint.tsv) shows the
family carries GO:0004867 at three separate IBD nodes. AGT inherits from
PTN008970140 (taxon:7711, Chordata), the SERPINA-clade node, seeded by
SERPINA1, SERPINA5, SERPINA9, bovine SERPINA3-1/A3-3, rat Serpinc1/Serpina1/
Serpina5 and mouse Serpina1b. All inhibitors. AGT is the one non-inhibitory
member of that clade and the inference reaches it mechanically.
This is PROPAGATION_BAD, not SOURCE_BAD: no seed annotation is wrong.
Worth recording, because it shows the fix already exists in this family: at node
PTN002606963 PAINT curators entered an IRD (negated=true) against
GO:0005576 to stop the extracellular-region inference reaching the
ER-resident serpin subclade. The machinery to block a bad propagation is in use
in PTHR11461; it simply has not been pointed at angiotensinogen for
GO:0004867. An IRD at the angiotensinogen node would fix all IBA-derived
copies at once.
The IEA copy has a different route but the same shape: InterPro:IPR000215 is
Serpin_fam, the whole family. Note the contrast that makes this diagnosable —
AGT also matches two angiotensinogen-specific signatures, IPR000227
(Angiotensinogen) and IPR033834 (Angiotensinogen_serpin_dom), and those map to
GO:0003081 regulation of systemic arterial blood pressure by renin-angiotensin,
which is exactly right. The family-level signature produces the wrong molecular
function; the gene-level signatures produce the right process.
The TAS copy is the origin story. It cites a 1988 mouse gene-cloning paper whose
own words are careful and hedged:
PMID:3397061
— homologous, putative reactive center, an alignment. The same abstract then
says the human site differs from the rodent one. A sequence-similarity
observation became a molecular function annotation.
The layer-1 biology is genuinely interesting and entirely absent from GOA.
AGT is not a passive substrate. The renin cleavage site is buried and has to
be released by a conformational change:
PMID:20927107
The renin complex is not a simple active-site encounter:
PMID:20927107
Yan et al. worked out the mechanism and showed the specificity determinants sit
outside renin's catalytic cleft:
PMID:30563843
PMID:30563843
That is a real molecular function of the precursor and it is currently recorded
only as bare protein binding. GO:0002020 protease binding is the
informative replacement, and it is what the GO:0005515 IPI against renin
(PMID:20927107) has been MODIFIED to.
Why angiotensinogen is absent from GO:0002003 — and why that absence is
correct. GO:0002003 angiotensin maturation is defined as "The process
leading to the attainment of the full functional capacity of angiotensin by
conversion of angiotensinogen into mature angiotensin in the blood." Querying
QuickGO for it across human/mouse/rat returns 110 annotations covering
essentially every other participant in the cascade — REN, ACE, ACE2, ENPEP,
ANPEP, MME, PREP, PRCP, CTSG, LVRN, Ace3, and even ATP6AP2, the (pro)renin
receptor — and angiotensinogen is not among them. The same is true of
GO:0002002 regulation of angiotensin levels in blood (117 annotations, no
AGT).
An earlier draft of this review read that as a pathway-completeness gap and
proposed both terms as NEW. That was wrong. The absence is deliberate GO
practice, on four counts.
The term is a proteolysis term. GO:0002003 is_a GO:0016486 peptide hormone
processing, which sits under GO:0016485 protein processing, GO:0051604
protein maturation and GO:0006508 proteolysis. What the term describes is the
cleaving. Renin and ACE do the cleaving; angiotensinogen is what is cleaved, and
contributes no chemistry to its own conversion.
GO applies the same rule to every comparable prohormone. Checked live in
QuickGO: insulin (P01308) is absent from GO:0030070 insulin processing, whose
annotations are PCSK1, PCSK2, P4HB, ERO1B, SLC30A8 and YIPF5. POMC (P01189),
pro-glucagon (P01275), proenkephalin-A (P01210) and pro-ANP (P01160) are all absent
from GO:0016486 peptide hormone processing, whose 132 human annotations are
convertases and peptidases — CORIN, which cleaves pro-ANP, is on it; pro-ANP is
not. APP (P05067) is absent from GO:0034205 amyloid-beta formation, whose 182
annotations include BACE1, PSEN1/2, NCSTN and APH1A/B but not the substrate.
Angiotensinogen's absence is the same absence, six times over.
GO's own causal model of this pathway makes the distinction explicitly. GO-CAM
6246724f00000549 "Renin-angiotensin pathway (Human)" (cached at
gocams/6246724f00000549/) types REN as GO:0004190 aspartic-type endopeptidase
activity part_of GO:0002003, and ACE as GO:0008241 peptidyl-dipeptidase
activity part_of GO:0002003. It contains AGT too — as GO:0005179 hormone
activity part_of GO:0002034 maintenance of blood vessel diameter homeostasis by
renin-angiotensin, and separately as three molecules: instances of
UniProtKB:P01019, i.e. as the proteases' input and output. A curator built that
model, put angiotensinogen in it, and chose not to give it GO:0002003. So the
claim that "the protein they act on is invisible" is false: AGT is in the
pathway model, and in GOA it already carries GO:0003081 regulation of systemic
arterial blood pressure by renin-angiotensin (IEA) and GO:0002034 (IDA, TAS).
GO:0003081 is an ancestor of both GO:0002003 and GO:0002002, so AGT is
already inside this branch of GO at the level that describes what it does. The
proposal would have moved it further down a branch whose added specificity is
exactly the part it does not do.
The two proposed rows were parent and child. GO:0002003 part_of GO:0002002,
so proposing both would have been redundant even on its own terms.
The perturbation data cited in support — knockout removing the product,
PMID:7989296
and AAV re-expression restoring it,
PMID:25691624
— is real, but it establishes that angiotensinogen is necessary for angiotensin
production, which is what being the substrate means. Necessity is not
participation; every one of the prohormones above would pass the same test.
There are partial counterexamples worth recording rather than explaining away:
thyroglobulin (P01266) is annotated to GO:0006590 thyroid hormone generation
by IDA, IBA and ISS, fibrinogen to GO:0042730 fibrinolysis, and C3 to
GO:0006956 complement activation. None of them is a passive substrate, but they
are not passive in three different ways, and it is worth being precise about
which — an earlier draft of this note implied thyroglobulin does its own
chemistry, and it does not.
Thyroglobulin is a scaffold. Thyroid peroxidase (TPO, P07202) does the
chemistry — UniProt's function line for it is "Iodination and coupling of the
hormonogenic tyrosines in thyroglobulin to yield the thyroid hormones T(3) and
T(4)", and every one of its curated catalytic-activity reactions is written on a
[thyroglobulin]-bound residue, e.g. [thyroglobulin]-L-tyrosine + iodide + H2O2
+ H(+) = [thyroglobulin]-3-iodo-L-tyrosine + 2 H2O. Thyroglobulin supplies those
tyrosines and holds the donor and acceptor in position; it catalyses nothing.
Both TG and TPO are annotated to GO:0006590 by IDA (checked live in QuickGO,
among 49 human annotations that also include DUOX1/2, SLC5A5 and DIO1/2), which
is the clearest available signal that GO puts the catalyst and the scaffold of a
reaction on the same process term.
Fibrin is a cofactor: it polymerises, then accelerates its own lysis by
serving as the template for tPA-mediated plasminogen activation. C3 does its own
chemistry: an internal thioester, buried in native C3 and exposed by
cleavage, is what C3b uses to bond covalently to the target surface.
Angiotensinogen is none of the three. It contributes no residue to the product
beyond the bond that is cut, positions nothing, and catalyses nothing. So the
distinction is not substrate versus non-substrate but "does this gene product do
any of the work", and the three exceptions above span what that can mean.
If AGT's substrate role should be machine-readable at all, the GO mechanism for
it already exists and belongs on the enzymes: has_input on REN's and ACE's
GO:0002003 annotations, or the molecule instances in the GO-CAM above. Nothing
needs to be added to AGT.
No NEW rows are proposed in this review.
The redox switch — and the negative that keeps it out of the review's claims.
UniProt records DISULFID 42..162 (mature Cys18–Cys138) and Zhou et al. built a
model on it: the bridge is labile, plasma holds a ~40:60 reduced:oxidised
mixture, and the oxidised form binds receptor-bound renin better
PMID:20927107.
It is a beautiful mechanism and I nearly proposed a GO term for it. Then I found
the in vivo test, which affinage did not surface:
PMID:25691624
Cys→Ser mutants delivered by AAV into hepatocyte-specific Agt-null mice gave
equivalent plasma Ang II, systolic blood pressure and atherosclerotic lesion
size. So the structural observation stands but the physiological claim is
contested, and no GO term is proposed for it. It is recorded as a knowledge
gap instead.
The M235T variant is worth one line because it is so often mis-told. The
structure places it away from the functional sites:
PMID:20927107
Concentration, not function — which is also why GO:0007267 cell-cell signaling
TAS from the preeclampsia association paper (PMID:8513325) is not a real
functional annotation.
Four of these, all checkable, none requiring me to second-guess a full text I
could not read.
GO:1903598 positive regulation of gap junction assembly (IGI, PMID:17416596)
has the sign backwards. The model is dTGR — rats carrying human renin and
human angiotensinogen, which is why the row is correctly coded IGI with
UniProtKB:P00797. The result:
PMID:17416596
Connexin 43 goes down when the human RAS transgenes are active. MODIFIED to
GO:1903597 negative regulation of gap junction assembly.
GO:0008083 growth factor activity (IDA, PMID:10406457) is contradicted by its
own paper. The definition is "The function that stimulates a cell to grow or
proliferate". The cited study is an AT2-receptor intracellular-loop mutagenesis
in PC12 cells, and what it measures is
PMID:10406457
— apoptosis and ERK inhibition. A second AGT-cited paper independently reports
no proliferative effect in its own system:
PMID:15652490
Marked over-annotated rather than removed, because Ang II is genuinely mitogenic
for vascular smooth muscle via AT1 — the term is not absurd, this citation just
does not support it.
GO:0003014 renal system process (IDA, PMID:21183621) rests on an
autoantibody biomarker study. The full text is cached and complete. It reports
that anti-angiotensinogen autoantibody titres are raised in chronic kidney
disease, and it explicitly cannot say what the antibodies even bind:
PMID:21183621
There is no renal-process experiment in it. AGT's genuine renal biology is
already carried by GO:0002019 and GO:0035813.
GO:0005515 protein binding with the HCV F protein (IPI, PMID:16237761,
assigned by AgBase) could not be verified. The cached full text runs from
abstract through the complete discussion, discusses each named hit in turn, and
never mentions angiotensinogen — the word appears zero times. The serpin it
does report is C1 inhibitor (SERPING1). The limit of that argument is worth
stating, because I first overstated it: the Results identify only 25 of the 36
positive colonies, and the Discussion names a handful more, so roughly a third of
the colonies are never identified anywhere in the text. Angiotensinogen cannot be
found in the paper, but neither can it be positively excluded. That is precisely
why this is marked over-annotated rather than removed, and raised as a question
for the assigning group.
27 rows are one fact. All 27 Reactome TAS rows say located_in GO:0005576
extracellular region. They differ only in which reaction identifier is cited —
R-HSA-2022403 (renin:prorenin receptor cleaving AGT), R-HSA-2022368 (neprilysin
making Ang-(1-7)), and so on. That is 24% of AGT's GOA record expressing a single
correct claim once per reaction the protein or one of its peptides appears in.
The term is right, so each is accepted, but no one should read 27 as 27 findings.
11 rows rest on a two-page editorial. PMID:17159080 is typed
Comment / Editorial / Review in PubMed, is a comment on another paper, and the
cached record contains no abstract text at all — only the citation line. It is
the sole source for 11 GOA rows, three of them coded TAS (traceable author
statement) and eight NAS. Several of the terms are perfectly correct RAS
physiology (GO:0002016, GO:0002018, GO:0002019, GO:0019229,
GO:0035813) and those are accepted on the biology while the citation is flagged
in reference_review. The generic ones (GO:0001558 regulation of cell growth,
GO:0042127 regulation of cell population proliferation) are marked
over-annotated: uninformative terms with an unreadable source.
10 rows are one large-scale yeast two-hybrid screen. PMID:32814053 is an
interactome of neurodegenerative disease proteins. AGT is not one. Its ten
partners are, per UniProt: mitochondrial intermembrane space and matrix (NME4),
cytosol (EIF2B4), nucleus/cytoplasm (PRMT5, SLFN12), ER membrane (VKORC1L1),
trans-Golgi network (TGOLN2), dendrite (TMEM185A), membrane (SNX12), and two
with no annotated location (NPHP1, PRRG2). 0 of 10 share a compartment with
a secreted plasma protein (AGT-bioinformatics/y2h_partners.py). Y2H
reconstitutes a transcription factor in the yeast nucleus, so a signal-peptide-
cleaved, disulfide-bonded, four-site N-glycosylated protein is being tested
somewhere it never is. UniProt's NbExp=3 on each is replicates within this one
study. None has functional follow-up. All ten marked over-annotated.
5 HDA proteomics rows. GO:0005576 from colostrum and venous-hypertension
ECM proteomes is correct and accepted. GO:0031012 extracellular matrix (colon
cancer ECM), GO:0070062 extracellular exosome (prostatic-secretion exosomes)
and GO:0072562 blood microparticle (plasma microvesicles) are marked
over-annotated: an abundant plasma protein appears in every such preparation,
and detection is not localisation of function.
All four resolved through resolve_withfrom.py; 80 distinct WITH/FROM tokens
across the GOA file, 0 unresolved.
GO:0005576 / PTN000156123 / 51 donors — accepted, NO_FAILURE_CORE. TheGO:0038166 angiotensin-activated signaling pathway / PTN008518321 /MGI:MGI:87963 + UniProtKB:P01019 — accepted, NO_FAILURE_CORE. The nodeSUPPORTS_TRANSFER, neverCIRCULAR_OR_REDUNDANT. A three-token donor list is not weak support here; itGO:0004867 / PTN008970140 / 9 inhibitor seeds — removed,PROPAGATION_BAD. See §1.GO:0042981 regulation of apoptotic process / PTN008518321 / mouse Agt +NO_FAILURE_NON_CORE, no failureGRANULARITY_MISMATCH, butGO:2001238) and anti-apoptotic/proliferative through AT1, soThe six ISS rows (GO_REF:0000024) transfer from rat Agt (P01015) and mouse Agt
(P11859) — true 1:1 orthologues, not paralogues. I checked each donor actually
still carries the term via QuickGO, and all six do, with experimental evidence:
GO:0005179 IDA PMID:8348686, GO:0005576 IDA ×2, GO:0010613 IDA
PMID:8348686, GO:0014873 IMP PMID:8252633, GO:0048146 IDA PMID:8348686,
GO:0010718 IMP PMID:34615811. So SOURCE_WEAK_OR_INFERRED would be wrong for
all of them; they are SUPPORTS_TRANSFER.
The two IC rows infer GO:0005179 and GO:0005576 from GO:0031702. Both
inferences are sound — a molecule that binds a cell-surface angiotensin receptor
is extracellular and is acting as a hormone. They are kept as non-core and
classified EVIDENCE_CIRCULAR_OR_REDUNDANT, in the specific sense the enum
defines ("the target already has stronger direct evidence"): AGT already holds
both terms by IDA and ISS. The inference is not wrong, it just adds nothing.
The trust gate tripped (self_evaluation_pairwise: tie, not win), and the
record earns LOW_QUALITY on its own merits independently of that.
It confused AGT with AGXT. The narrative asserts: "Independently of its
angiotensin-precursor role, AGT possesses peroxisomal alanine:glyoxylate
aminotransferase enzymatic activity required for glyoxylate-to-glycine
metabolism", citing PMID:40203111, a primary hyperoxaluria type 1 study in
AgxtQ84-/- rats. "AGT" is also the common abbreviation for alanine-glyoxylate
aminotransferase, whose HGNC symbol is AGXT (P21549) — a different gene, a
different protein, a different compartment. The error then propagates into
affinage's own GO grounding, which lists GO:0016740 transferase activity and
GO:0005777 peroxisome for a secreted plasma serpin. This is exactly why the
brief forbids importing mechanism_profile GO ids, and none were imported.
Its recall was poor on the gene's own literature. Of the 11 papers it cites,
none appears in AGT's GOA record, and it missed every paper that matters for
the molecular biology: PMID:20927107 (the Nature structure of AGT and of the
AGT–renin complex, and the redox switch), PMID:30563843 (the renin-specificity
structures and the loop-cleaved AGT that settles the serpin question),
PMID:7989296 (the angiotensinogen knockout mouse), PMID:25691624 (the in vivo
negative on the disulfide), PMID:12045255 (the prorenin receptor). I found these
by searching Europe PMC on the mechanism rather than the symbol — "angiotensinogen
Cys18 Cys138", "angiotensinogen AND renin AND crystal structure",
TITLE:"angiotensinogen" sorted by citation count. The affinage report's centre
of gravity is transcriptional regulation and cancer associations, which is where
the recent literature is, not where the gene's function is.
Not everything in it is wrong — the hepatic-knockdown and M235T-secretion items
are real — but nothing from it is cited as supporting_text for a mechanistic
claim in this review.
PRO_ chains of the same accession.y2h_partners.py is notRESULTS.md says so.AGT-bioinformatics/check_goa_reconciliation.py maps every GOA row to exactly
one review entry on (GO id, evidence, reference, qualifier, normalised
WITH/FROM), verifies supporting_entities against the GOA field verbatim, and
requires propagation_review on every IBA/ISS/IEA/IC row that carries a
WITH/FROM (15 rows). 114 GOA rows map to 113 entries: the one collapse is a
genuine exact duplicate in GOA, GO:0005179 IDA PMID:1567413 enables, which
appears twice with identical fields.