ARHGEF19 (WGEF, Ephexin-2) — review notes

Human, UniProt Q8IW93, 802 aa, HGNC:26604, chromosome 1p36.13.
Domain architecture (UniProt): DH 376–560, PH 592–704, SH3 715–776, with a long
disordered N-terminal region (1–~340) that carries the autoinhibitory module.

Headline: this is a coverage problem, not an over-annotation problem

Thirteen GOA rows, of which three are IBA, one IEA, five TAS from Reactome, one
NAS from a review, two IPI protein binding, and one IGI. The measured gap:

• 11 of 14 primary experimental papers on this gene have produced no GO
  annotation of any gene in any species.
• 12 of 14 have produced no GO annotation of ARHGEF19 in any species.
• Exactly one has produced an ARHGEF19 annotation to anything other than
  GO:0005515 protein binding, and that paper (PMID:20643356) is a mouse
  genetics study whose subject is GRHL3.

Method and full table: ARHGEF19-bioinformatics/RESULTS.md §2, reproducible via
go_coverage_by_reference.py (QuickGO by reference, all species; paper list
assembled independently from the Affinage record and a live PubMed sweep).

Two external sources say the same thing in prose. The Ephexin-family review:
"Studies on Ephexin2 have not extensively been done compared with Ephexin1."
PMID:30682817. And the family's structural paper notes that the biological
functions of Ephexin2 and Ephexin3 remain elusive although they are known to
activate RhoA (PMID:33597305, full text via PMC7923574 — the locally cached copy
is abstract-only, so that sentence is not quoted as supporting_text).

Retrieval: what affinage missed, and why

Affinage's trust gates were clear and all eleven of its citations resolve to real
papers about this gene — its precision is fine. Those eleven nevertheless omit
three of the fourteen primary papers plus one commentary.

I first wrote that every miss had the same shape, "the title does not name the
gene". That is false, and the table I wrote it above contained the counterexample.
go_coverage_by_reference.py now computes the answer — it reads each missed
paper's cached title and matches it against the gene's synonyms — so the claim
cannot drift again:

Three of the four have a retrieval story; the fourth does not. That last row is the
sharper finding rather than a blemish on the generalisation: a record whose gates
passed omitted a paper whose title begins with the gene symbol, alongside the
single most annotation-productive paper on the gene. Gates certify the citations
given, not the ones withheld.

What the PAINT table actually says

Resolved from interpro/panther/PTHR12845/PTHR12845-paint.tsv rather than guessed
from the family name:

Two consequences. The GO:0032956 IBA includes the target's own accession among
the seeds — which is the expected shape, not circularity: ARHGEF19's own IGI is
one of the descendant evidences PAINT used to place the IBD, so it reappears as a
source of the IBA the gene later receives. It marks that experimental grounding
exists on the target, and the IBA then adds what the IGI cannot — that actin
regulation is ancestral to PTHR12845 rather than a keratinocyte-specific role
established once in mouse skin. (CLAUDE.md §propagation is explicit on this:
"Never mark such a source CIRCULAR_OR_REDUNDANT or describe it as inflating
support." An earlier draft of this review did exactly that and was corrected.)

And the wound-healing node has a single seed — a legitimate call for a 1:1
orthologue, and CLAUDE.md is equally explicit that a short donor list is not weak
support, so this is recorded as a fact about the breadth of the ancestral call
rather than as a defect.

Molecular function: RhoA, and specifically not Rac1 or Cdc42

The first characterisation reported a broad specificity: "WGEF was shown to
activate RhoA, Cdc42, and Rac1 by pulldown assay, and forced expression of WGEF
resulted in marked rearrangement of the actin cytoskeleton" PMID:15485661.

Four years later the same three GTPases were re-tested side by side, with an
internal positive control, and the answer narrowed:

• RhoA: "Expression of hWGEF and XWGEF increased the level of active RhoA,
  whereas expression of hWGEFΔGEF did not" [PMID:18256687, Fig. 2A].
• Rac1/Cdc42: "we find that hWGEF and XWGEF did not activate Rac or Cdc42 above
  control levels, whereas Ephexin did" [PMID:18256687, Fig. 2B]. Ephexin-1 is the
  positive control in the same panel, so this is a real negative and not a failed
  assay.
• Binding mirrors activity: "We found that hWGEF and XWGEF strongly
  co-precipitated with RhoA at a level comparable to Ephexin, whereas hWGEFΔGEF
  did not" and "hWGEF but not XWGEF showed a very weak interaction with Rac-1,
  and neither WGEF bound to Cdc42" [PMID:18256687, Fig. 2C].

Both papers used human WGEF, and the 2008 experiment is the better controlled
one. The review treats RhoA as the substrate and records the 2004 Cdc42/Rac1
claim as superseded rather than as a second opinion. Note the consequence for
curation: the 2004 paper is the evidence Reactome cites for placing ARHGEF19 in
the RHOA-GEF set ("ARHGEF19 (Wang et al. 2004; Müller et al. 2020)",
R-HSA-8980691), so Reactome's TAS rests on the paper GO never annotated.

Loss of function in human cells: "depletion of WGEF with the aid of RNA
interference (RNAi) attenuates Rho activation in response to Wnt-1"
[PMID:18256687, Fig. 5B; MCF-7 cells].

GO cannot say "RhoA GEF" any more

This matters for how the review is written. Checked across three independent
services (EBI OLS4, the GO API, QuickGO) plus this repo's own cache/go/terms.csv:

QuickGO alone would have been misleading in both directions: it silently resolves
the merges and returns GO:0005085's live record for the obsolete ids, so querying
GO:0005089 there returns a non-obsolete term. The repo's local ontology cache
has no row for any of the merged ids, which is the third, offline confirmation.

So GO:0005085 is already the most specific MF term available and there is
nothing to MODIFY it to. The best-established fact about this protein — that it
is a RhoA GEF and not a Rac1/Cdc42 GEF — is currently inexpressible in a plain
GAF term and would need an annotation extension or a GO-CAM has_input(RHOA).
That is recorded as a knowledge gap, not as a proposed term.

Regulation: a doubly autoinhibited GEF

• N-terminal inhibitory helix, relieved by tyrosine phosphorylation. "Here we
  show that two other Dbl-family proteins, Ngef and Wgef, which like Tim contain a
  C-terminal SH3 domain, are also activated by tyrosine phosphorylation of a
  blocking helix." PMID:18537266. The residue is Y295 in human numbering:
  "This particular tyrosine (Y353xWGEF/295hWGEF) of hWGEF is present in NID and
  locks the GEF by interacting with its DH domain." PMID:38714795. Position 295
  of Q8IW93 is indeed Y (checked in dh_competence_check.py).
• C-terminal SH3 mode. "we show that the C-terminal SH3 domain binding to a
  polyproline region N-terminal to the DH domain of the Tim subgroup of
  Dbl-family proteins provides a unique mechanism of regulated autoinhibition of
  exchange activity" PMID:18537266. The family-wide structural treatment
  (PMID:33597305) calls this the C-terminal (SH3–HC) inhibitory mode and shows it
  is relieved in Ephexin4 and SGEF by PDZ proteins binding a C-terminal PDZ
  motif; ARHGEF19 has no such C-terminal motif, and the paper explicitly leaves
  the activating input for the other Ephexins undefined.
• Dvl2 supplies the missing input, via an internal motif. PMID:38714795 maps
  Dvl2-PDZ binding to an "internal PDZ-binding motif" and reports GEF assays in
  which PDZ binding relieves autoinhibition.

Two published disagreements, both unresolved

1. Where Dvl binds. PMID:18256687 mapped it to the N-terminus: "We found that
   hWGEFΔN and hWGEFDH–PH could not bind Dvl, whereas the other mutants tested,
   in particular the N-terminal domain itself, retained binding activity"
   (deletion constructs; ΔN = residues 1–213 removed). PMID:38714795 says
   otherwise and says so explicitly: "This observation is contrary to the previous
   study in which the Dvl2PDZ binding was mapped towards the N-terminus of hWGEF
   (between amino acid residues 1–213)17, whereas the motif identified by us is
   present between amino acid residues 349–359 of hWGEF."
2. What was actually measured on the human protein in 2024. The exchange
   assays in PMID:38714795 are on Xenopus protein: "Since some of these hWGEF
   constructs were found to be insoluble and/or poorly expressed in E. coli, we
   used WGEF from Xenopus laevis (xWGEF) as a surrogate system for further
   studies." The human contribution is peptide/fragment binding.

UniProt lists PDB 8YR7 under Q8IW93, which reads as "there is a structure of
human ARHGEF19". There is not, in the useful sense: 8YR7 is a Dvl2-PDZ construct
with the WGEF peptide fused to its C-terminus, and the authors report the fusion
did not work — "Even this approach failed to produce the WGEFpep–Dvl2PDZ complex
structure (PDB ID: 8YR7)". The deposited entity carries TFSLWQDIP (unique in
Q8IW93, at 351) but 30 of 109 residues are unmodelled. No ARHGEF19 coordinates
should be relied on.

Is the DH–PH module competent? Both directions

Full analysis in ARHGEF19-bioinformatics/RESULTS.md §1.

Yes, as a module: ARHGEF19 spans 24 of the 25 DH positions that contact RhoA in
both solved RhoA complexes (1X86 LARG:RhoA, 1LB1 Dbs:RhoA), matching the
weakest verified-active comparator in the panel and beating it in one case. Its
PF00621 bitscore (137.0) sits inside the 134.6–166.9 range spanned by five GEFs
that have both a substrate co-crystal and measured exchange activity.

But the same analysis refuses to go further, and this is the important half.
Identity at the exchange surface is 36% for ARHGEF19 against LARG and 36% for
Tiam1, which is a GEF for a different GTPase. A measure that cannot separate a
RhoA GEF from a Rac1 GEF cannot be used to assign substrate. And the truncation
control shows that a Pfam DH hit means very little on its own: ARHGEF19 isoform 2
(Q8IW93-2, which deletes 484–783 and so loses half the DH domain plus all of PH
and SH3) still clears the Pfam gathering threshold at 65.5 bits while spanning
only 5 of the 25 contact positions.

So: no residue-level reason to suspect a pseudo-GEF, and no residue-level licence
to claim exchange activity either. The claim rests on the biochemistry.

Biological process

• Wnt/PCP and convergent extension. The core result, in Xenopus with human
  rescue: WGEF binds Dvl (via the Dvl PDZ domain) and Daam1, overexpression
  activates RhoA and rescues dnWnt-11, depletion blocks CE and is rescued by
  RhoA/Rok PMID:18256687. Human hWGEF mRNA rescues the Xenopus morphant, so the
  human protein is functionally interchangeable in this assay.

• Epidermal wound repair via GRHL3 — and it is human. "we identified RhoGEF19,
  a homolog of a RhoA activator involved in PCP signaling in Xenopus, as a direct
  target of GRHL3" PMID:20643356. This is the source of the human IGI
  (GO:0032956, with GRHL3) and of the mouse Arhgef19 IMP for GO:0042060 that
  the human IBA projects back from — MGI:1925912 is mouse Arhgef19 itself, so
  the wound-healing IBA is a one-to-one ortholog transfer, not a paralogue guess.

Correction, and a lesson. An earlier revision of these notes and of the
review asserted that no human wound-healing experiment on ARHGEF19 had been
published. That was wrong. The cached record was abstract-only, the abstract
says "keratinocytes" with no species, and the paper is framed around mouse
genetics — so I inferred mouse. The full text (re-fetched from PMC2965174, now
cached) says: "we utilized the shRNA lentivirus to target the RhoGEF19
transcript in the human keratinocyte cell line, HaCAT". In those human
cells, RhoGEF19 knockdown leaves the scratch 20% closed "unlike the Scr cells,
which migrated to completely close the scratch within 24 hours", activated RhoA
falls, "RhoGEF19-kd cells formed only rudimentary stress fibers that showed
irregular organization and were not directed towards the scratched area", and
"re-expression of RhoGEF19 substantially rescued the wound healing phenotype
observed with the Grhl3-kd cells (10% closure versus 73%)". The leading-edge
polarity analysis is likewise on "human Grhl3-kd, RhoGEF19-kd, and Scr control
keratinocytes".

This is exactly what CLAUDE.md warns about — reserve confident "this reference
is about organism X" claims for cases where the cached abstract says so. It also
sharpens the coverage finding rather than softening it: a human knockdown-with-
rescue for GO:0042060 exists, in the very paper GOA cites for the mouse
orthologue, and the human gene still carries only an IBA.
- Renal ciliogenesis. "Additionally, the Daam1 partner, ArhGEF19, is also
required for proper cilia formation in MDCKII cells" PMID:31469868. Dog and
mouse lines; the same paper reports that Xenopus kidney knockdown did not
reproduce the cilia loss, so the effect is not obviously conserved.
- Pronephric tubulogenesis. Daam1/WGEF/Rho branch required, in Xenopus and
zebrafish PMID:21804089.
- Adipogenesis. Demethylation of exon 1 CpGs during 3T3-L1 differentiation;
forced WGEF inhibits the adipogenic programme PMID:19503838. Mouse only.
- A published negative. "WGEF shRNA had no apparent effect on Dvl-induced
neurite retraction in N1E-115 cells" and "other Rho-GEFs, including WGEF, are
not related to this process in N1E-115 cells" PMID:20810787. Mouse cells, so
this does not license a human NOT annotation, but it does mean the Wnt→Dvl→RhoA
role is context-dependent rather than universal.

A second, RhoA-independent branch: MAPK/ERK in cancer

Three independent human studies:

• NSCLC: "ARHGEF19 activated the mitogen-activated protein kinase (MAPK) pathway
  in a RhoA-independent manner: ARHGEF19 interacted with BRAF and facilitated the
  phosphorylation of its downstream kinase MEK1/2; both the Dbl homology (DH) and
  Pleckstrin homology (PH) domains of ARHGEF19 were indispensable for the
  phosphorylation of MEK1/2." PMID:29164615
• SCLC: "ARHGEF19-DH and -PD domain interacts with HRAS and activates the
  MAPK/ERK pathway in SCLC cells and SCLC xenografts." PMID:32993957 (the
  abstract's "-PD" is a typographical error for the PH domain)
• Breast: "we found that ARHGEF19 could activate the MAPK pathway in breast cancer
  cells" PMID:34813497

Caveats worth carrying: the NSCLC and SCLC papers share an institution, the
breast paper is in a low-visibility journal, and no structural or direct-binding
work supports a DH–PH:BRAF interface. Two of the three name the same domains as
necessary, which is at least a mechanistic commitment rather than a correlation.
Taken together this is enough to propose GO:0070374 positive regulation of ERK1 and ERK2 cascade as a new IDA-grade annotation, but not enough to call it core.

Partners

Localisation

There is no human localisation experiment worth annotating. The three
GO:0005829 cytosol rows are Reactome compartment assignments attached to GEF
sets, not measurements on this protein. The only positive data are Xenopus —
"XWGEF is associated with the plasma membrane in the Xenopus embryo"
PMID:18256687 — and a qualitative human observation that WGEF shows "a broad
cellular distribution" in proliferating SH-SY5Y and localises to neurites on
differentiation PMID:39409035. GO:0005886 plasma membrane is the biologically
expected site (a GEF must meet membrane-anchored RhoA) but is not supported for
human, so it goes in the gaps, not in core_functions[].locations.

One of the three cytosol rows deserves a flag rather than a shrug.
R-HSA-205039 is "p75NTR indirectly activates RAC and Cdc42 via a
guanyl-nucleotide exchange factor"; its catalyst is a generic
guanyl-nucleotide exchange factor activity of GEFs [cytosol] set and its output
is RAC1:GTP. ARHGEF19 reaches that reaction only by set membership, and human
WGEF does not activate Rac1 [PMID:18256687, Fig. 2B]. The cytosol term is not
false, but the route by which ARHGEF19 acquired it is.

What I deliberately did not do

• Did not propose a Rho-specific MF term: it does not exist any more (three
  services agree), and a failed GO text search would not have been evidence
  either way — the branch was walked.
• Did not propose GO:0005886 plasma membrane: Xenopus only.
• Did not propose a NOT annotation for Rac1/Cdc42 GEF activity: with the
  substrate-specific GEF terms merged away, NOT enables GO:0005085 would read
  as "is not a GEF at all", which is the opposite of the finding.
• Did not propose cilium or ciliogenesis terms: MDCKII (dog) and IMCD3 (mouse),
  and the paper's own Xenopus arm disagreed.
• Did not propose an Eph-receptor term. "Ephexin-2" is a name given by homology;
  the family review is explicit that an Eph receptor interacting with Ephexin2
  has not been identified.