Protists: Gene Function Outside Animals, Fungi and Plants

Bottom line: most of the diversity of eukaryotes is protist, but very little
of this repository is. Of 7,176 gene folders, about 104 (1.4%) are protists, and
62 of those are Dictyostelium. Before this project there were no reviews at
all for the parasites that carry most protist GO curation (Plasmodium,
Toxoplasma, trypanosomes, Leishmania), nor for the ciliate and diatom
models. This project sets out to fill that gap, lineage by lineage, starting
with a seed list of 27 well-studied genes whose UniProt accessions we have
checked. Tracks A (Apicomplexa), B (Kinetoplastida) and C (ciliates) are done:
23 genes (11 P. falciparum, 3 T. gondii, 4 T. brucei, 2 L. major,
2 T. thermophila, 1 P. tetraurelia) have COMPLETE reviews covering 361
existing annotations plus 34 new ones. The most common corrections were removing
terms carried over from animal homologs and replacing generic membrane and
protein binding rows with the compartment or binding partner each paper shows.

"Protist" is not a clade. Here it means any eukaryote that is not an animal, a
fungus or a land plant. Green algae and the unicellular relatives of animals
fall inside this definition, but other projects already cover them (see
Related projects), so this project only tracks them.

Why protists need their own project

1. Curation is concentrated in a few databases, and GO coverage is uneven.
   Experimental annotation of protists comes mainly from dictyBase,
   VEuPathDB/GeneDB (apicomplexans and kinetoplastids) and a few UniProt
   curators. Outside these, organisms with tens of thousands of IBA rows have
   almost no experimental rows (table below).
2. Function is inferred mostly from animals, fungi and plants. IBA and IEA
   transfer to divergent lineages is where over-annotation is most likely:
   animal-specific processes, ancestral activities that a lineage has lost, and
   localisations to organelles a protist does not have (or has in a modified
   form, such as the apicoplast, kinetoplast, hydrogenosome or mitosome).
3. Many protist functions are lineage-specific and have no animal homolog.
   Examples are antigenic variation, RNA editing, programmed genome
   rearrangement, apical-complex invasion and silica biomineralisation. GO
   terms for these may be missing or awkward, which makes them a good source of
   proposed_new_terms.
4. They matter for health and the environment. Malaria, toxoplasmosis,
   sleeping sickness, leishmaniasis, amoebiasis, giardiasis and trichomoniasis
   are protist diseases. Diatoms carry out roughly a fifth of global
   photosynthesis.

GOA census (2026-10-14)

Annotation counts from the QuickGO API, for the exact taxon of each row
(descendant taxa are not included), which is a strain except where the row says
otherwise. Toxoplasma appears twice because its curation really is split that
way: the species-level taxon 5811 carries SAG1's Swiss-Prot entry, while ME49
(508771) carries the TrEMBL entries, and neither row's count includes the
other's. "Experimental" means evidence codes
descending from ECO:0000269 (EXP, IDA, IPI, IMP, IGI, IEP). The review counts
are current as of 2026-10-15; T. brucei brucei (TRYBB, taxon 5702) holds 3
more and is not a row here because its annotations sit under the TREU927 strain.

Several counts need checking before we rely on them. The Leishmania major
Friedlin taxon has only 529 annotations in total, so its curation is probably
filed under a different taxon id. Toxoplasma annotations may be split across
strains (ME49, GT1, RH). And the high T. brucei experimental count probably
comes largely from a genome-wide localisation screen, which would make it mostly
cellular-component rows. A first task is to redo the census by species
(including descendants) and by GO aspect.

Scope

In scope: gene products of eukaryotes outside Metazoa, Fungi and Embryophyta,
reviewed with the standard gene-review workflow. The project is organised into
tracks by lineage, roughly following the eukaryote supergroups:

Tracks F to H are counted here so the project can report protist coverage as a
whole, but new work on them belongs to their own projects. Lineages with no
model organism in the repository yet (Rhizaria, haptophytes, cryptophytes,
euglenids, red algae) are out of scope for the first phase.

What to look for in each review

These are the review questions most specific to protists. They add to the
normal guidelines in CLAUDE.md; they do not replace them.

• Taxon-inappropriate terms. Flag electronic terms that describe animal-only
  biology (for example immune or nervous-system processes, or
  extracellular matrix in organisms without one) or plant-only biology.
• Lost or modified organelles. Check mitochondrion annotations on
  Giardia, Entamoeba and Trichomonas (they have mitosomes or
  hydrogenosomes instead), and check plastid annotations on apicomplexans (the
  apicoplast is non-photosynthetic).
• Paralog expansions. Protist genomes often carry large lineage-specific
  families (PfEMP1/var, VSGs, Giardia VSPs, Trichomonas BspA-like
  proteins). IBA and IEA rows may be assigned to the wrong subfamily.
• Host–pathogen terms. Use the GO "symbiont" and "host" process terms
  correctly. A parasite protein acts on the host; it is not a participant in
  the host's own processes.
• Ontology gaps. Record missing terms for lineage-specific processes as
  proposed_new_terms and in the gene notes. They are not knowledge gaps.

Related projects

• DICTYOSTELIUM_DEVELOPMENT: 62 Dictyostelium
  reviews in 14 development modules.
• ORIGINS_OF_MULTICELLULARITY: choanoflagellate
  and Capsaspora genes.
• CO2_CONCENTRATING_MECHANISMS and
  PHOTOSYNTHESIS: Chlamydomonas and algal photosynthesis.
• IBA_REVIEW and PAINT: the propagation patterns
  that most protist annotations depend on.

STATUS

Last updated: 2026-10-20

Seed gene list

Species codes are UniProt mnemonics and match the genes/<CODE>/ folders.
Accessions were checked against UniProt on 2026-10-14. Entries marked
unreviewed are TrEMBL entries that carry no GN Name= in UniProt. These were
fetched by accession with --alias, so the folder uses the standard
nomenclature name; gene_symbol is set to that same standard symbol and the
accession stays in id. Expect this for any TrEMBL-only apicomplexan gene.

Track A: Apicomplexa

• [x] PLAF7 CSP (Q7K740): circumsporozoite protein, sporozoite surface coat
• [x] PLAF7 CRT (Q8IBZ9): chloroquine resistance transporter, digestive vacuole
• [x] PLAF7 MDR1 (Q7K6A5): ABC transporter, drug resistance
• [x] PLAF7 PMII (Q8I6V3): plasmepsin II, haemoglobin digestion
• [x] PLAF7 CDPK1 (P62344): calcium-dependent protein kinase 1
• [x] PLAF7 ATP4 (A0A143ZZK9): P-type Na+ ATPase, antimalarial drug target
• [x] PLAF7 MSP1 (Q8I0U8): merozoite surface protein 1
• [x] PLAF7 RH5 (Q8IFM5): erythrocyte invasion ligand (binds basigin)
• [x] PLAF7 K13 (Q8IDQ2, unreviewed): kelch13, artemisinin resistance
• [x] PLAF7 AMA1 (Q7KQK5, unreviewed): apical membrane antigen 1, moving junction
• [x] PLAF7 HAP2 (Q8IJQ3, unreviewed): gamete fusion factor
• [x] TOXGO SAG1 (P13664): major surface antigen; Swiss-Prot entry is at species level (taxon 5811)
• [x] TOXGM MIC2 (S8F9G8, unreviewed): microneme adhesin, gliding motility
• [x] TOXGM ROP18 (A0A125YQL1, unreviewed): rhoptry kinase, virulence factor

Toxoplasma folders follow the UniProt mnemonic of each entry, so they split
between TOXGO (species level, taxon 5811; SAG1's Swiss-Prot entry) and
TOXGM (strain ME49, taxon 508771; the TrEMBL entries MIC2 and ROP18). Put a
new Toxoplasma gene wherever its own UniProt entry sits rather than picking
one folder for the species.

Track B: Kinetoplastida

• [x] TRYB2 REL1 (P86927): RNA-editing ligase 1, mitochondrial RNA editing
• [x] TRYBB TPR (P39051): trypanothione reductase; Swiss-Prot entry at subspecies level (taxon 5702)
• [x] TRYBB PFRA (P22225): 69 kDa paraflagellar rod protein (PFR-A/PFR-B)
• [x] TRYBB PFRC (Q26789): 73 kDa paraflagellar rod protein (PFR-C/PFR-D)
• [x] LEIMA gp63 (P08148): leishmanolysin, surface metalloprotease
• [x] LEIMA PTR1 (Q01782): pteridine reductase 1, antifolate resistance

Track C: Ciliates

• [x] TETTS TERT (O77448): telomerase reverse transcriptase
• [x] TETTS TAP65 (W7X6T2): p65 telomerase holoenzyme protein (LARP7 family)
• [x] PARTE A0BFB4: existing draft, finished

Track D: Metamonads and Entamoeba

• [ ] ENTH1 hgl1 (P32022): Gal/GalNAc lectin heavy subunit, adherence
• [ ] ENTH1 APA (P34095): amoebapore A, pore-forming peptide
• [ ] TRIV3 A2FPI7: existing draft, to finish

Track E: Stramenopiles

• [ ] PHATC Lhcx1 (B7FYL0, unreviewed): LHCX photoprotection protein
• [ ] PHATC Sit1 (B7GBI4, unreviewed): silicic acid transporter
• [ ] THAPS Sil1 (Q5Y2C2, unreviewed): silaffin, silica biomineralisation
• [ ] PHATC B7FXQ8: existing draft, to finish

Most species codes in this list (all except PARTE, PHATC and TRIV3) are new
to genes/. Before fetching, check that just fetch-gene
resolves them, and decide whether entries filed at species or subspecies level
should stay there or be moved to the strain proteome entry.

Candidates not yet identified in UniProt

Lookups on 2026-10-14 found no entry by gene name for these; find the right
accession (often a systematic locus id) before adding them:
T. brucei VEX1 and RBP10; Tetrahymena PDD1, TWI1 and TPB2; Paramecium
PiggyMac; Phaeodactylum aureochrome 1a; Giardia (strain WB) alpha-giardin.

Tasks

• [ ] Redo the GOA census per species (with descendant taxa) and per GO aspect;
  resolve the Leishmania and Toxoplasma taxon questions
• [x] Confirm just fetch-gene works for the new species codes (it does; entries
  with no UniProt gene symbol need just fetch-gene <ORG> <ACC> --alias <NAME>)
• [x] Review Track A (Apicomplexa)
• [x] Review Track B (Kinetoplastida)
• [x] Review Track C (ciliates): finish the PARTE draft, add Tetrahymena TERT and TAP65
• [ ] Review Track D (metamonads and Entamoeba)
• [ ] Decide whether to submit "food vacuole membrane" and "moving junction" as
  GO term requests (see 2026-10-15 notes)
• [ ] Collect protist-specific over-annotation patterns in a
  PROTISTS/patterns.md sub-page as reviews accumulate
• [ ] Collect proposed GO terms for lineage-specific processes

NOTES

2026-10-20 (Track C)

Track C reviewed: 2 new Tetrahymena genes plus the long-standing Paramecium
draft, all status: COMPLETE and passing just validate. 52 annotations newly
reviewed with 1 NEW row; A0BFB4's 14 rows were reviewed in earlier rounds and this
batch only closed the file out.

"Other" is TOO_GENERAL, UNDECIDED or MARK_AS_OVER_ANNOTATED. TERT's eight are all
TOO_GENERAL; A0BFB4's five are all MARK_AS_OVER_ANNOTATED. A0BFB4's counts are those of
the 2026-10-21 re-review merged to main separately from this track, which this branch
takes as-is; Track C's own contribution to that gene is the close-out to COMPLETE.

Patterns, continuing the Track A and B lists.

• The expected import did not happen. Tracks A and B were dominated by animal
  terms transferred onto divergent lineages, and both ciliate genes were set up for
  it — TERT by the enormous human-telomerase literature, TAP65 by the LARP7 family's
  7SK/P-TEFb role. Neither GOA contained any of it: no shelterin, TPP1/POT1, Cajal
  body, dyskerin or senescence terms on TERT, and no 7SK or transcription-elongation
  terms on TAP65. Worth recording as a negative result, because it suggests the
  over-annotation risk tracks how much experimental curation a protist gene has
  rather than how divergent the lineage is.
• Generic ancestors, judged rather than removed. TERT's eight TOO_GENERAL rows are
  the polymerase and telomere-maintenance ancestors of terms accepted elsewhere in
  the same review. Five are EXP rows, which is exactly the case the enum covers:
  the term is uninformative, the evidence is not in doubt, and REMOVE would be wrong.
  A0BFB4 had the same pattern and resolved it the other sanctioned way: its four broad
  catalytic rows are MODIFY to the specific descendant, per the correct-but-general
  rule, in the re-review merged to main separately. Both actions are available when the
  gene already carries the specific term — TOO_GENERAL says the row adds nothing,
  MODIFY names what should replace it — and which reads better is worth settling as a
  convention rather than per gene.
• Refinement where the assay is sharper than the term. Both telomeric repeat DNA binding rows became single-stranded telomeric DNA binding (telomerase engages
  the 3' overhang), and TAP65's generic nucleic acid binding / RNA binding became
  telomerase RNA binding — a specificity that was tested and held in Tetrahymena,
  where p65 knockdown collapses TER while leaving 5S, SRP, U2, U3 and U6 untouched.
• The non-catalytic subunit stayed non-catalytic. No telomerase activity term was
  transferred onto p65, and the explicit negative result is now recorded on the
  reference rather than merely absent: coassembly of p65 did not affect TERT's
  primer-extension activity in reconstitution.
• Two nuclei, one term. Tetrahymena has a macronucleus and a micronucleus and GO
  has terms for both, but no cached source says which compartment was assayed for
  TERT, so nucleus stays generic and the refinement is a question rather than an
  assertion.

One asymmetry left standing deliberately. TERT carries GO:0000333 telomerase catalytic core complex and p65 does not, although p65 is a subunit of that very
core. The reason is procedural, not biological: TERT's GOA already contains the term,
so reviewing it is an ACCEPT, while adding it to p65 would be a NEW row — and
NEW is governed by the rule against proposing a term that is a descendant of one
the gene already carries, which p65 satisfies eight times over with GO:0005697.
Both reviews are correct under the rule and reach opposite content. This is the same
shape as the MODIFY-to-a-non-descendant gap found in Track B: a rule that behaves
differently depending on which side of the fence an annotation already sits. Raised
as a question on TAP65 rather than resolved by either reviewer.

Caveat, as in Tracks A and B. Automated deep research was unavailable (perplexity
is not a registered provider, falcon fails authentication, the OpenAI key is
rejected), so each -notes.md is a manual literature review and says so. All 11
cited PMIDs were already cached, 8 with full text.

2026-10-15 (Track B)

Track B reviewed: 6 kinetoplastid genes, one review agent each, all
status: COMPLETE and passing just validate. 54 existing annotations
reviewed, 7 NEW rows added.

"Other" is TOO_GENERAL, UNDECIDED or MARK_AS_OVER_ANNOTATED.

Patterns, continuing Track A's list.

• Family-level transfer onto a diverged enzyme. TPR's glutathione-disulfide reductase activity is a TreeGrafter transfer from the wider flavoprotein
  family. The purified enzyme strongly prefers trypanothione disulfide, and
  kinetoplastids have no glutathione reductase, so the row was removed and the
  generic oxidoreductase terms were narrowed to GO:0015042. The same logic
  moved glutathione metabolic process to trypanothione metabolic process.
• Lineage-specific compartments beat generic ones. cytoplasm became
  cytosol (TPR), cytoplasm became mitochondrion (REL1, where a TrypTag
  cytosolic call conflicts with a transit peptide), and cytoskeleton became
  paraflagellar rod (both PFR subunits).
• Paralog pairs must be decided together. PFRA (69 kDa, called PFR2 in the
  literature) and PFRC (73 kDa, PFR1) are partner subunits whose reviews first
  diverged: one proposed structural constituent of cytoskeleton plus
  paraflagellar rod assembly, the other structural molecule activity plus
  cilium-dependent cell motility. OLS settles it: GO:0097740 paraflagellar rod is a supramolecular complex and is not under cytoskeleton, so
  GO:0005200 does not fit. Both now use GO:0005198 and GO:0060285, with
  rod assembly left as a question, since a subunit incorporated into the rod is
  arguably what is assembled rather than an agent of assembly. The naming
  inversion is a live trap: UniProt's mnemonic for the 73 kDa entry is
  PFR2_TRYBB.
• No GO term for RNA editing. REL1's editing role had to go under
  mitochondrial mRNA modification; GO has no term for guide-RNA-directed
  U-insertion/deletion editing. Raised as a question rather than a proposed term.
• Obsolete anchored-component terms. gp63's GPI anchoring could not use
  GO:0031362/GO:0046658 (both obsolete); GO:0009897 external side of plasma membrane carries it instead.

One row left UNDECIDED: TPR mitochondrion, from a 1991 fractionation
paper with no cached abstract or full text, against a protein with no
mitochondrial targeting sequence.

2026-10-15

Track A reviewed: 14 genes, one review agent each, all status: COMPLETE
and passing just validate.

"Other" is TOO_GENERAL or MARK_AS_OVER_ANNOTATED.

No deep research. Falcon authentication failed and the OpenAI key was
rejected in this environment, so no -deep-research- files exist. Each gene's
-notes.md is a manual literature review and says so at the top. Re-running
deep research later would be a useful cross-check.

Recurring patterns (candidates for PROTISTS/patterns.md):

• Animal or plant terms carried over electronically. ROP18 nucleus and mitotic DNA damage checkpoint signaling (TreeGrafter, from an animal kinase family); ATP4 potassium and
  calcium transport (inherited from ENA and SERCA nodes; the activity is
  insensitive to both ions); CDPK1 calmodulin binding (CDPKs carry their own
  calmodulin-like domain). All removed. A related case went the other way: RH5's
  bicellular tight junction is equally an animal epithelial term imported on a
  word match, but the parasite structure it describes is real, so that row is
  MODIFY to the proposed moving junction term rather than a removal.
• Glutathione transport on CRT. Comes from Arabidopsis CLT donors and from
  mutant alleles only; marked over-annotated rather than removed.
• Drug-response terms. response to xenobiotic stimulus (CDPK1, K13, MDR1)
  describes inhibitor screens or resistance alleles, not a process the protein
  carries out. It was removed, marked over-annotated, or changed to
  xenobiotic transmembrane transport depending on the case.
• Sparse GOA. MIC2, SAG1, HAP2 and K13 had 1 to 3 annotations each, with no
  molecular function. Most of their core content came from NEW rows. MIC2
  (7 NEW rows) is the heaviest and is a good candidate for a second look.
• Non-symbol entries. MDR1, K13, AMA1 and HAP2 have no GN Name= in UniProt,
  so just fetch-gene seeded gene_symbol with the accession. All four were
  set to the standard P. falciparum symbol, matching the folder, with the
  accession kept in id. A convention note in CLAUDE.md would stop this
  recurring; left for a maintainer.

Proposed terms to consolidate.

• food vacuole membrane (a child of GO:0005774): proposed independently by the
  CRT and MDR1 reviews, which worded their definitions differently. Canonical
  text for the eventual term request: "The lipid bilayer surrounding the food
  vacuole (GO:0020020), the acidic digestive compartment of intraerythrocytic
  Plasmodium and related apicomplexans in which host haemoglobin is degraded."
  Parent: GO:0005774 vacuolar membrane.
• moving junction (AMA1 and RH5, parent GO:0030054 cell junction): the
  merozoite-erythrocyte ring that RH5's bicellular tight junction row was
  reaching for. Both reviews carry identical definition text.
• RH5-CyRPA-Ripr invasion complex (RH5).
• Kelch13 compartment and cytostome-mediated endocytosis of host cell cytosol (K13).

Calls to check. AMA1 replaces its RON2 and RON4 protein binding rows with
host cell surface binding, on the grounds that RON2 sits in the host membrane
when AMA1 binds it. That stretches the term, since RON2 is a parasite protein.
The open question is whether a parasite-ligand-in-host-membrane case needs
its own term.

2026-10-14

Project created. Inventory of genes/: eight protist species folders, holding
62 Dictyostelium, 19 Chlamydomonas, 11 Salpingoeca, 6 Capsaspora, 2
Monosiga and one each of Paramecium, Phaeodactylum and Trichomonas
reviews. Apart from Dictyostelium (all COMPLETE) and Chlamydomonas (18 of
19 COMPLETE), every protist review is DRAFT. There are no apicomplexan, kinetoplastid,
Tetrahymena, Giardia or Entamoeba folders. The census table above was
built from QuickGO API queries on the same day; seed accessions were checked
with the UniProt REST API.