Propagation Audit: Animal Terms on Unicellular Holozoans

Propagation Audit: Animal Terms on Unicellular Holozoans

← back to Origins of Animal Multicellularity

Bottom line: almost every GO annotation on choanoflagellate, Capsaspora
and sponge proteins is propagated from animals, so we checked where that
propagation goes wrong. The 16 Track A reviews, chosen because the genes have
experimental literature, carry 53 propagated rows from TreeGrafter
(GO_REF:0000118) and the UniProt multi-method merge (GO_REF:0000120).

Reviewer action Rows
Accepted 33
Kept as non-core 6
Undecided 3
Down-graded 11

All 11 down-grades are TreeGrafter rows.

Five further reviews were added because their proteins carry the terms found
below: M. brevicollis LATS (MONBRDRAFT_1233) and four choanoflagellate
cadherins. They add 48 propagated rows, 38 of them down-graded, including IBA
rows on the M. brevicollis reference genome. They were chosen for their
errors, so they are excluded from the rates above. All 21 reviews are in
propagation_audit_rows.tsv (101 rows, 49 down-graded).

The down-grades come from four mechanisms:

The S. rosetta protein knocked out as yorkie also landed with the MAGI
scaffolds (Case 3), but its inherited terms happened to be harmless.
- An animal-tissue term placed on a node that includes choanoflagellates.
PAINT put four process IBDs on the LATS node PTN002390470. One of them,
regulation of organ growth, is meaningless in a unicellular organism
(Case 1).
- A lineage-specific term on a broad node. A mannan IBD seeded only by
Candida genes reaches a choanoflagellate mannosyltransferase, oomycetes
and green algae (Case 4).
- A family-level node term. Talin "cell-cell adhesion" (Case 5).

Most propagated molecular-function and signalling terms were sound, for
example hippo signaling on the Capsaspora kinases and coactivator. The
problems are specific and fixable at named nodes. Each down-graded row
carries a propagation_review (root cause, failure modes, source node) in its
gene review YAML; propagation_audit_rows.tsv lists them. The proteins in Cases
1 and 6 that reach these terms without literature have reviews too:
M. brevicollis LATS (MONBRDRAFT_1233) and the four choanoflagellate
cadherins (PTSG_05882, PTSG_06458, PTSG_11235, MBCDH12). Their PANTHER node
analysis is in genes/MONBE/MBCDH12/MBCDH12-bioinformatics/.

Data: propagation_audit.py regenerates
propagation_audit_rows.tsv (every propagated
row with the reviewer's action) and
propagation_audit_spread.tsv (how often each
down-graded term occurs on unicellular holozoan proteins in GOA). Counts are
from 2026-10-01.

Scope

Organism Reviews Propagated rows Down-graded
Salpingoeca rosetta (choanoflagellate) rosetteless, jumble, couscous, hippo, warts, yorkie, SrSeptin2, SrSeptin6 19 4
Capsaspora owczarzaki (filasterean) coHpo, coWts, coYki, CoBra, coITGB2, coVIN 28 6
Oscarella pearsei (sponge) VIN1, TLN 6 1
S. rosetta and M. brevicollis, chosen for their propagated terms PTSG_05882, PTSG_06458, PTSG_11235, MBCDH12, MONBRDRAFT_1233 48 38

None of these organisms is a PANTHER reference genome, so none of the
proteins gets IBA rows. Their tree-based annotations all come from
TreeGrafter, which grafts the protein onto a reference tree and copies the
terms on the graft node. The one reference genome among unicellular
holozoans is Monosiga brevicollis, which does get IBA rows. It appears below
because it inherits the same LATS-node terms.

Case 1: organ growth, apoptosis and G1/S on choanoflagellate Warts

What happened. The S. rosetta Warts kinase (F2U943) is correctly placed in
the LATS family (PTHR24356:SF418). It received seven TreeGrafter rows from the
M. brevicollis Warts leaf PTN001220369, whose parent PTN002390470 carries
four PAINT IBDs (from interpro/panther/PTHR24356/PTHR24356-paint.tsv):

IBD term Seeds Date Review action on S. rosetta warts
GO:0035329 hippo signaling fly wts, mouse Lats1/Lats2, human LATS1/LATS2 2020-08-09 ACCEPT
GO:0046620 regulation of organ growth mouse Lats2 only 2022-04-15 REMOVE
GO:0043065 positive regulation of apoptotic process fly wts only 2017-02-28 MARK_AS_OVER_ANNOTATED
GO:0000082 G1/S transition of mitotic cell cycle mouse Lats2, human LATS2 2017-02-28 MARK_AS_OVER_ANNOTATED

The same four terms reach M. brevicollis LATS (A9UVF9) as IBA rows. They
also reach the reference-tree leaf for Drosophila wts, where they are fine.

Why it is a problem.
- Organ growth. No choanoflagellate has organs. GO's taxon constraints on
GO:0046620 do not exclude choanoflagellates, so nothing downstream catches
this.
- Apoptosis. This rests on one fly gene. We found no evidence of apoptosis
in choanoflagellates.
- G1/S transition. This rests on mammalian LATS2. In S. rosetta,
knocking out warts makes rosettes larger and slows growth, which is not a
G1/S-checkpoint result. Choanoflagellates do have a cell cycle, so the term
is not impossible; it is unsupported for this protein.
- Hippo signaling fits the Capsaspora evidence and was accepted.

Spread. Regulation of organ growth occurs on exactly two unicellular
holozoan proteins, the two choanoflagellate Warts kinases, and on none in
Filasterea or Ichthyosporea. The other two terms are more widespread in
unicellular holozoans, but on different nodes. Unicellular organisms divide
and some may die by regulated pathways, so those other rows are not taxon
errors as such.

Recommendation for PAINT (PTHR24356).
- Move the GO:0046620 IBD from PTN002390470 to the metazoan descendant node,
or add an exception there for the choanoflagellate leaves.
- Consider doing the same for GO:0043065, which has one seed.
- Separately, GO could consider a taxon constraint excluding unicellular
lineages from organ-level growth terms.

Case 2: Capsaspora Warts grafted into the citron/ROCK family

What happened. Gene targeting identifies coWts (A0A0D2VGR4, CAOG_00619) as
Capsaspora Warts, and knocking it out sends coYki into the nucleus
(PMID:38517944). Its sequence has a LATS-type MOB-binding region and an AGC
kinase domain. TreeGrafter nevertheless placed it in PTHR22988:SF71
"CITRON RHO-INTERACTING KINASE", part of the "MYOTONIC DYSTROPHY S/T
KINASE-RELATED" family that contains ROCK, MRCK and citron. Human LATS1/2 are
in PTHR24356. The graft node PTN001122925 supplied cytoskeleton and
actomyosin structure organization. The review marked cytoskeleton
over-annotated and removed actomyosin structure organization. It also added
hippo signaling by IMP, which coWts never received because it missed the
LATS node.

Why the Drosophila ortholog escapes. UniProt's PANTHER cross-reference
also classes Drosophila wts (Q9VA38) in PTHR22988 (SF76). But fly wts is a
leaf of the PANTHER reference tree and gets its IBA rows from the LATS node
PTN002390470 in PTHR24356, including hippo signaling, with no
cytoskeleton/actomyosin rows. Our reading is that the family HMMs and the
reference trees disagree about Warts kinases. Reference proteomes follow the
tree, while non-reference proteomes such as Capsaspora follow the HMM call
and graft into the wrong family. This is an inference from the classifications
and annotation rows, not from re-running TreeGrafter.

Spread. GO:0031032 actomyosin structure organization is on 9
Capsaspora proteins, 4 of them from PTN001122925, and on 7 ichthyosporean
proteins. Several are probably genuine ROCK/MRCK/citron kinases, for which the
term is reasonable; we have not reviewed them.

Recommendation for PANTHER. Check whether the PTHR22988 subfamily HMMs
(SF71, SF76) out-score the PTHR24356 LATS subfamilies for Warts/LATS
sequences. If so, the LATS HMM or the family boundary needs attention. A quick
test: rescore coWts, fly wts and S. rosetta Warts (F2U943, which classifies
correctly) against both families.

What happened. The 2025 knockout study names PTSG_06057 (F2UDK1) as the
S. rosetta Yorkie homolog. This is stated in version 2 of the preprint
behind PMID:41037400. TreeGrafter placed F2UDK1 in PTHR10316:SF68
"MEMBRANE ASSOCIATED GUANYLATE KINASE-RELATED", on graft node PTN002569196,
whose members are ecdysozoan genes. That gave it cytoplasm and
signal transduction. Both terms are plausible for a Yorkie protein, so the
review kept them as non-core, while flagging that they were reached by the
wrong route.

PANTHER places a different S. rosetta WW protein, F2U5K0 (PTSG_03848), in the
YAP1 family subfamily PTHR17616:SF8, alongside Drosophila yki (Q45VV3) and
Capsaspora coYki (A0A0D2WY30). That conflicts with the experimental
literature. Our motif scan
(yorkie-bioinformatics)
finds:
- F2UDK1 has the four Warts phosphorylation motifs the literature reports for
srYki, and no recognizable TEAD-binding motif.
- Full-length alignment scores put F2UDK1 closer to YAP1, Yki and coYki than
F2U5K0 is. Much of that similarity is in the WW domains.

Status: unresolved. Either PANTHER's family call or the literature's
ortholog call is wrong. A phylogeny of the holozoan WW-domain proteins would
settle it. Until then, neither protein should receive YAP1-family process
terms on the strength of this audit.

Case 4: fungal mannan biosynthesis on a choanoflagellate mannosyltransferase

What happened. Couscous (F2UJ78) is a predicted MNN2-type
alpha-1,2-mannosyltransferase needed for rosette development (PMID:30556809).
Its TreeGrafter node PTN001270341 gave it the following rows:

Term Review action Reason
alpha-1,2-mannosyltransferase activity ACCEPT Plausible family-level activity
Golgi apparatus UNDECIDED Tagged Couscous was "clearly not localized to the Golgi"
GO:0046354 mannan biosynthetic process REMOVE Defined in GO as the main hemicellulose of softwood; in fungi, Mnn2 builds cell-wall mannan. Nothing suggests choanoflagellates make mannan.

Spread. GO:0046354 is on 4 choanoflagellate proteins and 1
ichthyosporean protein. From this node it also reaches green algae and oomycetes,
and all 56 of the node's IBA rows list only Candida seed genes. Recommendation: restrict the mannan IBD in the
MNN2 family to the fungal node.

Case 5: talin cell-cell adhesion (minor)

The talin family node PTN001072690 gave sponge TLN cell-cell adhesion. Talin
works at integrin-matrix adhesions, and nothing links the sponge protein to
cell-cell contacts, so the review changed it to cell-matrix adhesion. This is
a family-level node term, not a taxon problem; GO:0098609 occurs widely in
unicellular holozoans through other families.

Extension: IBA nodes of the human toolkit genes (Track B)

For each existing human review re-checked in Track B (CDH1, ABL1, MYC,
NOTCH1, TP53), we took the PANTHER nodes behind its IBA rows and asked QuickGO
whether those nodes' terms reach proteins of Choanoflagellata, Filasterea or
Ichthyosporea, by IBA or by TreeGrafter IEA. Details and the query results are
in each gene's notes (genes/human/<GENE>/<GENE>-notes.md, section
"Premetazoan origin").

Gene Node(s) Reaches unicellular holozoans? Assessment
CDH1 PTN000616280 (PAINT taxon Bilateria) Yes: 30 TreeGrafter IEA rows. Ten terms each on three S. rosetta cadherins: F2UD23, F2UFV3 and F2USU1 Case 6, below
CDH1 PTN008601603 Yes: 5 IBA rows on M. brevicollis A9V8Y4 Case 6, below
TP53 PTN000893833 Yes: 7 IBA terms each on M. brevicollis A9UZX3 and A9V4M3 Mostly DNA-binding and transcription terms; regulation of apoptotic process is an untested extrapolation
TP53 PTN000154790 3 TreeGrafter rows on Capsaspora A0A0D2X0F0 (DNA binding, TF activity, regulation of transcription); the apoptosis term itself is not transferred Plausible from domain conservation (PMID:31861340)
MYC PTN001691821 Yes: 4 IBA rows on M. brevicollis Myc (A9V5B4) Supported: the transcription-factor and nuclear terms match the experiments on choanoflagellate Myc (PMID:21571926)
ABL1 PTN002521457 Yes: tyrosine kinase activity and plasma membrane on 59 M. brevicollis kinases The MF is fine; plasma membrane is a localization default on cytoplasmic kinases
NOTCH1 PTN001933897 (Eumetazoa), PTN002911625 No Expected for the animal-innovation control

Case 6: junction and catenin terms on choanoflagellate cadherins

What happened. PAINT records node PTN000616280 in the cadherin family
PTHR24027 at taxon:33213 (Bilateria) in
interpro/panther/PTHR24027/PTHR24027-paint.tsv. Its IBDs, seeded by mouse,
rat, human and zebrafish classical cadherins, include adherens junction,
catenin complex, cell-cell junction assembly, calcium-dependent cell-cell
adhesion and cell morphogenesis. TreeGrafter nonetheless grafts three
S. rosetta cadherins (subfamily PTHR24027:SF422) onto that node. Each of
the three receives 10 IEA rows, 30 rows in all:

Protein Domains (Pfam)
F2UD23 cadherin repeats, SH2
F2UFV3 cadherin repeats, fibronectin type II
F2USU1 cadherin repeats, tyrosine phosphatase

Separately, M. brevicollis A9V8Y4 (MBCDH12) receives five IBA rows: beta-catenin
binding, catenin complex, cadherin binding, cell migration and cell-cell
adhesion. They come from the cadherin family root PTN008601603, labelled
"Metazoa-Choanoflagellida" in the PANTHER tree. A9V8Y4 is the only
choanoflagellate leaf in PTHR24027 and hangs directly from that root, so its
rows are a PAINT node-placement problem (IBDs placed too deep), not a
TreeGrafter graft. Four of the ten TreeGrafter rows on the S. rosetta
proteins are these same root IBDs; the other six are PTN000616280's own.

Why it is a problem. None of the four proteins has PF01049, the
cytoplasmic domain through which classical cadherins bind beta-catenin. A
UniProt census finds 0 PF01049 proteins in choanoflagellates, filastereans
and ichthyosporeans, against 23,575 in animals (CDH1-bioinformatics/).
Choanoflagellates lack classical cadherins (PMID:22837400, PMID:27189570), and
S. rosetta colonies have no adherens-junction-like structures
(PMID:22837400). So the catenin and junction terms describe a protein
complex these proteins cannot form.

Why this matters for TreeGrafter. This is the clearest case so far of
TreeGrafter grafting onto a node that PAINT itself restricts to a clade the
query is not in. Either the graft ignores the node's taxon, or the node label
is wrong. Both readings point to a fix in the PTHR24027 tree.

Recommendation for PANTHER. Check how the S. rosetta SF422 cadherins
come to graft onto a Bilateria node. Graft them onto a pre-bilaterian
cadherin node, or block terms from PTN000616280 for non-bilaterian queries.

Case 7: Capsaspora integrin betas grafted onto the vertebrate ITGBL1 node

What happened. The 9 TreeGrafter rows on Capsaspora integrin beta 2
(coITGB2, A0A0D2WRB3) all come from node PTN002560695 in PTHR10082. The
PANTHER tree API (2026-10-01) gives this node as subfamily SF3, "INTEGRIN
BETA-LIKE PROTEIN 1", with species Euteleostomi. Every leaf is a vertebrate
ITGBL1, a secreted integrin-beta-like protein without the transmembrane
receptor architecture. coITGB2 is a full transmembrane integrin beta:
- a MIDAS-type motif;
- NPxY-type tail motifs;
- shown to mediate adhesion to fibronectin-coated surfaces (antibody
blocking; DOI:10.1101/2020.02.27.967653).

Spread. In QuickGO, the node's 9 terms reach six Capsaspora integrin-beta
entries, 54 rows in all:
- A0A0D2U6H2, A0A0D2VIQ2, A0A0D2WRB3;
- the cDNA entries D7PE18, D7PE19 and D7PE20.

They also reach one ichthyosporean protein (A0A0L0G967, 9 rows). The terms
include focal adhesion and cell-cell adhesion. The coITGB2 review marked both
over-annotated, and modified cell-matrix adhesion to cell-substrate adhesion,
since Capsaspora has no fibronectin ortholog and no known matrix ligand.

Recommendation for PANTHER. Graft the unicellular holozoan integrin betas
onto a pre-metazoan integrin-beta node, not the vertebrate ITGBL1
subfamily.

Case 8: "cell fate specification" from an all-animal T-box node

What happened. TreeGrafter grafts Capsaspora Brachyury (CoBra,
A0A0D2VUC6) onto PTN000137774 in the T-box family PTHR11267. In the PANTHER
tree API this is a duplication node whose leaves come from 34 organisms, all
animals, from Trichoplax and Nematostella to human. Among its terms is
GO:0001708 cell fate specification, an embryonic-development process seeded
by animal T-box genes. The CoBra review removed it:
- nothing links CoBra to cell fate in Capsaspora;
- even in Xenopus, CoBra activates mesendodermal genes without the
target selectivity of animal Brachyury (PMID:24043797).

Spread. The same term from this node reaches 5 Capsaspora proteins
(all T-box) and 3 ichthyosporean proteins in QuickGO.

Note. PANTHER classifies CoBra in subfamily PTHR11267:SF181,
"OPTOMOTOR-BLIND PROTEIN", a Tbx2-class name. That conflicts with the
paper's phylogeny placing CoBra in the Brachyury class.

What held up (negative controls)

So signalling and molecular-function terms on conserved premetazoan pathways
transfer well. The failures are at the edges:
- tissue- and organ-level process terms placed too deep;
- family boundaries for divergent paralogs (Warts vs ROCK, Yorkie vs MAGI);
- lineage-specific metabolic terms.

Caveats