The target is unambiguous. UniProt entry CRTB_PANAN / P21683, 309 aa,
reviewed (Swiss-Prot), evidence level PE 1: Evidence at protein level
[file:PANAN/crtB/crtB-uniprot.txt "Evidence at protein level"]. The recommended
name is [file:PANAN/crtB/crtB-uniprot.txt "Full=15-cis-phytoene synthase"] with
[file:PANAN/crtB/crtB-uniprot.txt "EC=2.5.1.32 {ECO:0000269|PubMed:9593819};"].
Nomenclature caution that matters for reading the evidence: the organism is
recorded by UniProt as [file:PANAN/crtB/crtB-uniprot.txt "Pantoea ananas (Erwinia
uredovora)."], NCBI taxon 553. Both cached primary papers say Erwinia uredovora
in their titles. That is the same organism under its historical name, not a
different species — so neither paper is an off-target citation. The GOA rows
likewise carry TAXON NAME: Pantoea ananas. I note this explicitly because the
CLAUDE.md rule about not dismissing a citation whose title names something else
applies in reverse here: the title looks wrong but is right.
Both cached publications are abstract-only (full_text_available: false).
Neither is a case where I need the full text, because in each the abstract itself
states the result being annotated.
Abstract-only, but unusually informative. It is a direct in-vitro biochemical
characterisation of this protein:
The 15-cis/all-trans mixture is not an ambiguity about the enzymatic product.
UniProt resolves it: [file:PANAN/crtB/crtB-uniprot.txt "The enzyme produces
15-cis-phytoene."], with the all-trans form arising non-enzymatically by
photoisomerisation. So GO:0046905 (15-cis-phytoene synthase activity) is the
correct specific term, not a generic phytoene synthase term.
A detail from this abstract does a lot of curation work below. Expressing
CrtB alone poisoned the host: PMID:9593819,
and the fix was to supply more GGPP from a different enzyme: PMID:9593819 CrtB therefore consumes GGPP, and the GGPP
synthase of this organism is a different gene, crtE. This is direct
experimental refutation of the GO:0004311 IEA row (section 4.1).
Abstract-only. Cloned the six-gene cluster from this organism: PMID:2254247, from PMID:2254247, and established
the pathway by PMID:2254247. This is not a
GOA reference row for crtB; I use it only for pathway position and to establish
that crtE and crtB are separate genes with separate activities. Fetched with
uv run ai-gene-review fetch-pmid 2254247.
The abstract's inline pathway string is mangled by the text extraction (the
arrows and gene labels ran together), so I have deliberately not quoted it and
have taken no ordering claim from that fragment. The gene-cluster and
transformant-analysis sentences quoted above are clean.
UniProt gives three experimentally-evidenced (ECO:0000269|PubMed:9593819)
reaction records, all one overall transformation split into its two half-steps:
the overall RHEA:34475 (2 GGPP = 15-cis-phytoene + 2 diphosphate), the first
half RHEA:22296 (2 GGPP = prephytoene diphosphate + diphosphate), and the second
half RHEA:34479 (prephytoene diphosphate = 15-cis-phytoene + diphosphate). The
FUNCTION line describes the same two-stage mechanism. Pathway:
[file:PANAN/crtB/crtB-uniprot.txt "Carotenoid biosynthesis; phytoene
biosynthesis."]. Family: [file:PANAN/crtB/crtB-uniprot.txt "Belongs to the
phytoene/squalene synthase family."].
modules/carotene_backbone_biosynthesis.yaml lists this exact protein
(UniProtKB:P21683) as a representative member of the phytoene-synthesis step,
with required function GO:0046905 and RHEA:34475 as evidence, in PANTHER family
PTHR31480. The UniProt record agrees: [file:PANAN/crtB/crtB-uniprot.txt "PANTHER;
PTHR31480; BIFUNCTIONAL LYCOPENE CYCLASE/PHYTOENE SYNTHASE; 1."] (the family name
reflects the fungal bifunctional members; bacterial CrtB is single-domain). The
review below is consistent with the module and asserts nothing beyond it.
Eight GOA rows. No CC rows at all, no IBA rows, no protein binding row, no
negated rows, no isoform rows.
Verified definition (QuickGO): "Catalysis of the reaction: (2E,6E)-farnesyl
diphosphate + isopentenyl diphosphate = (2E,6E,10E)-geranylgeranyl diphosphate +
diphosphate." That is head-to-tail chain elongation producing GGPP.
CrtB does the opposite: GGPP is its substrate. Three independent lines:
The source signature is IPR044843, [file:PANAN/crtB/crtB-uniprot.txt "InterPro;
IPR044843; Trans_IPPS_bact-type."] — a fold/superfamily-level trans-isoprenyl
diphosphate synthase entry (cf. Gene3D 1.10.600.10 "Farnesyl Diphosphate Synthase"
and SSF48576 "Terpenoid synthases" on the same record). Mapping a shared all-alpha
terpenoid-synthase fold to one specific chain-elongation activity assigns the
paralog's activity (CrtE's) to CrtB. This is exactly the InterPro2GO
paralog-overannotation case the brief describes. REMOVE.
Verified definition (QuickGO): "Catalysis of the reaction: 2 (2E,6E)-farnesyl
diphosphate + H+ + NAD(P)H = 2 diphosphate + NAD(P)+ + squalene." Every component
is wrong for CrtB: the substrate is FPP (C15) not GGPP (C20), the product is
squalene (C30) not phytoene (C40), and the reaction consumes NAD(P)H.
The NAD(P)H point is the decisive one and is mechanistically informative rather
than merely bookkeeping. Squalene synthase and phytoene synthase are genuine
homologues and both form a cyclopropylcarbinyl diphosphate intermediate by
head-to-head condensation (presqualene-PP and prephytoene-PP respectively — the
latter is explicit in UniProt's RHEA:22296 record for this protein). The
families then diverge at the second half-step: squalene synthase reductively
opens the intermediate using NADPH, whereas phytoene synthase instead eliminates a
proton, installing the central double bond. That elimination is precisely why the
product is 15-cis-phytoene and why no nicotinamide cofactor is needed. The
experimentally determined cofactor requirement for this protein contains no
NAD(P)H at all: PMID:9593819 So the one feature that names
GO:0051996 is the step CrtB demonstrably does not perform.
The source signature IPR033904, [file:PANAN/crtB/crtB-uniprot.txt "InterPro;
IPR033904; Trans_IPPS_HH."], is the shared head-to-head trans-IPPS domain
(same record carries CDD; cd00683; Trans_IPPS_HH), common to both families by
construction, so it cannot discriminate between them. REMOVE.
Counter-argument I considered and rejected. CrtB is inhibited by squalestatin
PMID:9593819, and squalestatin is a squalene synthase inhibitor. Does that
support GO:0051996? No. Inhibitor cross-reactivity reports active-site similarity,
which is already explained by the shared fold and is the reason the InterPro
signature matches in the first place. Being inhibited by a drug aimed at a
homologue is not evidence of performing the homologue's reaction. Noted here so
the removal is made with the contrary evidence on the table, not around it.
Not wrong, just uninformative. I confirmed by QuickGO ancestor query that
GO:0046905 is a descendant of GO:0016765 (path: GO:0046905 -> GO:0004659
prenyltransferase activity -> GO:0016765 -> GO:0016740 transferase activity), so
the row is a true grandparent-level assertion. Its supporting signatures are
ARBA:ARBA00026321 and IPR019845, the latter being
[file:PANAN/crtB/crtB-uniprot.txt "InterPro; IPR019845;
Squalene/phytoene_synthase_CS."] — a conserved-site pattern shared across the
whole phytoene/squalene synthase family. A family-wide site pattern can only
license a family-wide statement, so the generic term is a reasonable automatic
output. But this protein has a directly demonstrated, EC- and Rhea-matched
activity, so retaining a term that says only "transfers some non-methyl alkyl
group to something" adds nothing. MODIFY to GO:0046905.
I weighed the annotation-reviewer guidance that broader IEAs may simply be
ACCEPTed. I judged this one too general to leave: it is two levels above even
prenyltransferase activity and conveys no substrate, no product, and no pathway.
The IDA is as clean as this evidence type gets: purified-enzyme in-vitro assay,
product confirmed against authentic standards, on the protein from this organism.
The IEA row is grounded in RHEA:34475 and EC:2.5.1.32, which are the same
reaction and match UniProt's experimentally-evidenced catalytic-activity record
exactly. Duplicate GO IDs across evidence codes are expected and fine; I accept
both rather than collapsing them.
Participation test: CrtB catalyses a step of this process — the committed
head-to-head condensation that makes the first C40 carotenoid. It is not merely
required for, consumed by, or acted on by the pathway. Passes cleanly; this is the
straightforward catalyst case, no scaffold/cofactor reasoning needed.
Inter-ontology logical inference from GO:0046905, which the gene holds with IDA.
Sound on the chemistry: GO:0016120 is defined (QuickGO) as "The chemical reactions
and pathways resulting in the formation of carotenes, hydrocarbon carotenoids",
and 15-cis-phytoene is exactly that — the C40 hydrocarbon precursor of every
downstream carotene.
Ontology-structure note, recorded but deliberately not acted on. I verified via
QuickGO that GO:0016120 is not currently a descendant of GO:0016117 (its
ancestors run through GO:0016119 carotene metabolic process and GO:0008299
isoprenoid biosynthetic process). The gene therefore legitimately holds both terms
and neither is redundant with the other. Per the brief I make no claim that one
subsumes the other and propose no ontology fix.
Three candidates were considered and all three were declined.
Prephytoene diphosphate synthase activity. UniProt records the first half-step
(RHEA:22296) as a separately evidenced catalytic activity, so a term for it
would be defensible in principle. There is no such term to use: GO:0016767
"geranylgeranyl-diphosphate geranylgeranyltransferase activity" was exactly this
reaction and I confirmed against QuickGO that it is obsolete (isObsolete:
True). GO has therefore already decided this half-step is not separately
annotated, and the overall GO:0046905 covers the transformation. Nothing to add.
Metal ion binding (GO:0000287 / GO:0030145 / GO:0046872). The cofactor
requirement is experimentally established for this protein PMID:9593819
and UniProt carries the Metal-binding, Magnesium and Manganese keywords. I
still declined. A cofactor requirement in a bulk assay does not by itself localise
the metal to the protein: divalent cations also chelate the diphosphate leaving
groups of the substrate. Decisively, the UniProt record contains no FT BINDING
features at all — only FT CHAIN and a disordered FT REGION — so UniProt's own
curators, with the full text, did not place a metal site on this sequence. The deep
research report reaches the same limit, noting that catalytic-residue assignments
here "should remain family-based predictions, not protein-specific experimental
facts". Asserting a binding term would be filling a slot the primary curators left
empty. Raised as a suggested experiment instead.
A stress/photoprotection process term. The literature associates P. ananatis
carotenoid production with oxidative-stress and toxoflavin tolerance, and a crtB
lesion would abolish all downstream carotenoids. This is necessity, not
participation: the protective chemistry is done by the coloured downstream
products (lycopene, beta-carotene, zeaxanthin), not by CrtB and not by its own
colourless product phytoene. Under the CLAUDE.md participation rule, knockout
abolishing an outcome establishes that the gene is required, which is precisely
what being the pathway's entry enzyme means, and cannot substitute for asking
which entity performs the step. CrtB performs the condensation step and nothing
else. Declined.
GOA has no cellular-component row for this protein, and I add none. There is no
localization experiment for P21683: the deep research report states plainly that
cytosolic or cytoplasmic-face placement [file:PANAN/crtB/crtB-deep-research-falcon.md
"remains inferred rather than experimentally demonstrated for P21683"]. The
sequence offers no positive signal either way — no signal peptide, no transmembrane
segment, the only feature besides the chain being a disordered C-terminal region.
Cytoplasm would be a low-risk guess for a bacterial soluble enzyme, and the module
asserts GO:0016020 membrane at its taxon-neutral root on substrate-hydrophobicity
grounds. Those two plausible answers point in different directions, which is itself
the reason not to pick one. I therefore left locations out of the core function
rather than manufacturing a component assertion, and recorded it as a CC_DARK
knowledge gap with a concrete resolution. Substrate-specificity breadth is recorded
as a second gap on the same basis ([file:PANAN/crtB/crtB-deep-research-falcon.md
"No rigorous alternative-substrate panel for P21683 was recovered."]).
Action tally over 8 rows: ACCEPT 5, MODIFY 1, REMOVE 2. No UNDECIDED (both
cached papers are abstract-only, but each abstract states the result being
annotated, so nothing was unadjudicable) and no NEW. One core function.