TreeGrafter on Unicellular Holozoans: Hippo Pathway Case Study

TreeGrafter on Unicellular Holozoans: Hippo Pathway Case Study

← back to TreeGrafter Inference Evaluation

In the main-page tables since the 2026-10-03 refresh. These rows come
from reviews added on 2026-09-30 and 2026-10-01 for the
Origins of Animal Multicellularity
project. They were not in the 2026-10-01 tables; at the 2026-10-03 refresh
(f81b9f300) the sidecars on the main page include 65 TreeGrafter rows from
S. rosetta, Capsaspora and Oscarella review files, 41 of them
down-graded, and the failure-mode table classifies each down-graded row. The full audit,
with a script that regenerates its tables, is the
propagation audit.

Bottom line: choanoflagellates (Salpingoeca rosetta), Capsaspora and
sponges are not PANTHER reference genomes, so every tree-based GO annotation
they carry is a TreeGrafter IEA (GO_REF:0000118). The main TreeGrafter
sidecars now include the 65 live GO_REF:0000118 rows from reviewed
S. rosetta, Capsaspora and Oscarella proteins, 41 of them down-graded.
The broader Origins propagation audit also scored 101 propagated rows: 53 in
literature-selected reviews and 48 in targeted follow-ups, the latter including
M. brevicollis IBA controls outside the TreeGrafter-only denominator. Together
they show the main page's failure modes in a new setting: genes from lineages
that the reference trees sample thinly.

Case Protein Graft Failure mode (main page numbering) Rows down-graded
Warts grafted with citron/ROCK kinases Capsaspora coWts, A0A0D2VGR4 PTHR22988:SF71, node PTN001122925 4, mis-placement (across families, not just within a superfamily) cytoskeleton (over-annotated); actomyosin structure organization (removed). Also missed hippo signaling from the LATS node
Yorkie grafted with MAGI-related scaffolds S. rosetta yorkie, F2UDK1 PTHR10316:SF68, node PTN002569196 (ecdysozoan) 4, mis-placement; terms happened to be harmless none (cytoplasm and signal transduction kept as non-core)
Animal-tissue IBDs inherited at a correct graft S. rosetta warts, F2U943 PTHR24356:SF418, leaf PTN001220369 under PTN002390470 inherited PAINT over-placement, like the FliI/SctN case regulation of organ growth (removed); positive regulation of apoptotic process and G1/S transition (over-annotated)
Fungal pathway term on a choanoflagellate S. rosetta couscous, F2UJ78 node PTN001270341 (MNN2 family) 3, out-of-context process (pathway absent in host) mannan biosynthetic process (removed)
Family node term sponge TLN, A0A3G2LGI8 node PTN001072690 1, granularity / sibling term cell-cell adhesion → cell-matrix adhesion
Graft onto a node PAINT restricts to Bilateria three S. rosetta cadherins: F2UD23, F2UFV3, F2USU1 PTHR24027:SF422, node PTN000616280 (taxon:33213) 4-like, taxon-blind graft 30 current rows, ten on each of three S. rosetta cadherin gene reviews; none of the proteins has the beta-catenin-binding domain (PF01049). See the audit for the node-level spread check
Graft onto the vertebrate ITGBL1 node Capsaspora integrin beta 2 (coITGB2, A0A0D2WRB3) and five other Capsaspora integrin-beta entries PTHR10082:SF3, node PTN002560695 (Euteleostomi) 4, mis-placement onto an animal-only node focal adhesion and cell-cell adhesion (over-annotated); cell-matrix adhesion → cell-substrate adhesion
Developmental term from an all-animal T-box node Capsaspora Brachyury (CoBra, A0A0D2VUC6) PTHR11267, node PTN000137774 (duplication node; 34 leaf organisms, all animals) 3, out-of-context process cell fate specification (removed)

What is new relative to the main evaluation

  1. Placement errors here cross family boundaries. The main page's mode-4
    cases are within-superfamily mis-placements, such as MDH grafted onto the
    L-LDH subfamily. Here, a LATS/Warts kinase lands in the citron/ROCK family,
    and a Yorkie candidate lands in the MAGI-related family.

The Warts case points to a specific cause. UniProt's PANTHER classification
also puts Drosophila wts (Q9VA38) in PTHR22988. But fly wts is a leaf of the
reference tree, so its IBA rows come from the LATS node PTN002390470 in
PTHR24356, and they are correct. Our working hypothesis is that the family
HMMs prefer PTHR22988 for some Warts sequences. Reference proteomes are
protected by their tree position; non-reference proteomes are grafted by
the HMM call. S. rosetta Warts (F2U943) does classify into PTHR24356, so
the effect is sequence-dependent. A rescoring test is listed below.

  1. A correct graft can still carry wrong terms when an IBD sits above a
    lineage split.     The four IBDs on PTN002390470 (hippo signaling, organ
    growth, apoptosis, G1/S) apply to a node that includes the
    choanoflagellate leaves. M. brevicollis (a reference genome) receives
    them as IBA; S. rosetta receives them through TreeGrafter.
  2. Regulation of organ growth is the clearest error: choanoflagellates have
    no organs, and GO's taxon constraints do not catch it.
  3. Hippo signaling at the same node is supported by the Capsaspora data.

So the node is right for the pathway term and too deep for the tissue
terms. This is the same shape as the FliI/SctN case, where an IBD on a
duplication node reaches paralogs that do not share the function.

  1. Conserved signalling terms transfer well. The three TreeGrafter rows on
    Capsaspora Yorkie (coYki) were all accepted, and the knockout and
    heterologous data support each one: coactivator activity, hippo signaling,
    and positive regulation of transcription by RNA polymerase II.

  2. TreeGrafter grafts onto taxon-restricted nodes, repeatedly. PAINT records cadherin
    node PTN000616280 at Bilateria (taxon:33213), yet three choanoflagellate
    cadherins graft onto it and inherit its adherens-junction and catenin
    terms (audit, Case 6).
    The main evaluation never tested this, because its corpus is
    bacterial-heavy. Two more cases followed. Capsaspora integrin betas
    graft onto the Euteleostomi ITGBL1 node, and Capsaspora T-box factors
    graft onto an all-animal T-box node that carries "cell fate specification"
    (audit, Cases 7 and 8).
    With three of five graft errors in this corpus of the same kind, a cheap QC
    rule would pay off: suppress terms from a graft node whose PAINT taxon does
    not include the query organism.

Where the findings are recorded

Each down-graded row carries a propagation_review in its gene review YAML,
naming the PANTHER source node with a source_status:

The choanoflagellate cadherin graft and the M. brevicollis LATS IBAs are
recorded in reviews of those proteins (under genes/SALRS/ and
genes/MONBE/).