magi3b (LOC564220) / wu:fi36a10 (magi3a)

Back to pairs

Bottom line: UNRESOLVED, leaning toward an asymmetric pair in which both copies are still
needed. The two MAGI3 scaffolds keep the same domain set. In the only functional test, F0 crispants
of either copy impaired brain vascularization equally, and the double crispant was no worse. That
argues against simple backup in endothelium. The copies differ in public expression: wu:fi36a10 is
called in 22 tissues, a pattern like that of the single gar magi3, while magi3b is called in 9 of
those, with retina the highest. The data cannot separate a dosage requirement, a partition of
other tissues, or a degenerating magi3b. The TGD origin rests on the PANTHER tree. Only wu:fi36a10
keeps gar-like synteny.

Sample record: fate=UNRESOLVED; level=expression; evidence=experimental_both; identity=58.2%

	magi3b (directory LOC564220)	wu:fi36a10
UniProt	A0A8M6YYZ5 (TrEMBL, 1406 aa)	A0A0R4IGT8 (TrEMBL, 1355 aa)
Current names	ZFIN magi3b; NCBI magi3b (GeneID 564220; PANTHER label LOC564220)	ZFIN wu:fi36a10; NCBI and RefSeq magi3a (GeneID 561689)
Human ortholog	MAGI3	MAGI3
Chromosome	6	23
ZFIN	ZDB-GENE-060503-301	ZDB-GENE-030131-6139
Ensembl	ENSDARG00000025974	ENSDARG00000101869
GOA rows	12 (3 experimental)	10 (IEA/IBA only)
Review	genes/DANRE/LOC564220	genes/DANRE/wu_fi36a10

Drawn at random (draw 8, seed 20260928) from the 778 clean 1:1 TGD_tree pairs
(batch3_sample.tsv). The page uses the ZFIN symbols in prose. The literature uses the NCBI names
(magi3a = wu:fi36a10, magi3b = magi3b). I checked this mapping by matching the paper's probe
primers to each gene (section 3).

1. Evidence the pair comes from the TGD

PANTHER. The row in panther_tgd_pairs.tsv is:

tgd_call	branch	class	family	human orthologs (a / b)	gar co-orthologs	medaka co-orthologs	medaka parallel dup
TGD_tree	Neopterygii|Teleostei	1:1	PTHR10316 (MEMBRANE ASSOCIATED GUANYLATE KINASE-RELATED)	MAGI3(O) / MAGI3(LDO)	1 (same gar gene for both)	2	(blank)

Independent checks (RESULTS.md,
synteny_output.md):

• Synteny with gar. Within +/- 1.5 Mb, the wu:fi36a10 region (chr 23) shares ptpn22 and RSBN1 with the
  gar magi3 region (LG3). The magi3b region (chr 6) shares none. No gene gives double-conserved
  synteny. wu:fi36a10 keeps the ancestral neighbourhood. The magi3b region is not recognizably its TGD
  duplicate at this scale, but duplicated blocks often lose most second copies, so this does not rule
  out a TGD origin.
• Ensembl Compara places the duplication at Euteleostomi rather than at the teleost root. PANTHER's
  placement (with one shared gar co-ortholog) is the one consistent with a single gar magi3.
• Literature. No paper discusses the origin of the pair.

Status. TGD origin is supported by the PANTHER tree only. Synteny neither confirms nor contradicts
it, and Ensembl disagrees on timing. Treat it as probable but not established.

2. Protein-level comparison

Source: RESULTS.md and
annotation-comparison.md.

• Identity. 58.2% identity and 68.9% similarity over 1480 alignment columns. wu:fi36a10 is the
  more conserved copy: 63.5% identical to gar magi3, compared with 55.1% for magi3b, and 54.5% vs 50.4%
  to human MAGI3. Each medaka co-ortholog is closest to its PANTHER-assigned zebrafish partner.
• Domains. Both copies keep the complete MAGI architecture: PDZ0, the GK-like domain, WW1, WW2 and
  PDZ1-5. Per-domain identity to human MAGI3 is 69-96% (wu:fi36a10) and 69-88% (magi3b). The two
  zebrafish copies are 79-96% identical domain by domain. Only the disordered C-terminal tail has
  diverged, as it has in all fish MAGI3 proteins (about 20% identity to human).
• GK domain. Neither copy, nor gar or human MAGI3, has a Walker A/P-loop motif at the site UniProt
  labels "ATP binding". This fits the structural finding that MAGI GK domains are truncated protein-binding
  modules:
  [PMID:37163606 "Unlike other MAGUK GKs, which are composed of three subdomains including guanosine monophosphate (GMP)–binding, LID, and CORE subdomains, MAGI GKs lack the last two subdomains"]
• Biochemistry. Only wu:fi36a10 has been tested. As "Magi3a" it bound the Gpr124 C-terminal motif in
  a yeast two-hybrid screen:
  [PMID:35649360 "the PDZ proteins membrane-associated guanylate kinase, WW, and PDZ domain containing 3 (Magi3a) and discs large MAGUK scaffold protein 4 (Dlg4a) bound to Gpr124 via its C-terminal ETTV motif."]
  [PMID:35649360 "Not unexpectedly, the ETTV motif of mouse and human GPR124 also interacted with Dlg4a and Magi3a in a yeast two-hybrid assay (Figure 5D)."]
  magi3b was not tested for binding. A forced-recruitment assay used "MAGI3" without naming the copy:
  [PMID:35649360 "Similarly, the recruitment of the ETTV motif, DLG4, and less potently MAGI3 could restore brain angiogenesis."]

Does each copy keep the ancestral molecular function? Probably yes: the scaffold, with all
protein-binding domains intact. For magi3b this is inferred from sequence, not tested. Neither copy has
kinase activity, and the ancestral MAGI3 has none either.

3. Expression

• Which gene the paper means. The in situ primers of PMID:35649360 map uniquely. The magi3a primers
  match wu:fi36a10 with 0 mismatches and magi3b with 7-8; the magi3b primers match magi3b with 0
  mismatches (RESULTS.md).
• Whole-mount in situ was uninformative for both:
  [PMID:35649360 "As the diffuse whole-mount in situ hybridization expression patterns of dlg4a, dlg4b, magi3a, and magi3b did not permit to establish their vascular expression unambiguously (Figure S2A), we isolated 48 hpf kdrl:GFP+ ECs by fluorescence-activated cell sorting (FACS) and determined their expression levels by qRT-PCR analysis."]
• Endothelium: both copies are expressed.
  [PMID:35649360 "dlg4b, magi3a, and magi3b transcripts could readily be amplified, while dlg4a transcripts were not (Figure S2B)."]
• Public RNA-Seq (Bgee). wu:fi36a10 has calls in 22 anatomical entities, including gill, blastula,
  skin, intestine, ovary, muscle, bone, liver, heart, brain, kidney, spleen and retina. magi3b has calls
  in 9: retina (the highest), blastula, ovary, early embryo, brain, gill, gastrula, tail and larva. Every
  magi3b call is shared with wu:fi36a10. The single gar magi3 has calls in eye, ovary, heart, embryo,
  brain, testis, skin, kidney, bone, intestine, liver, gill and muscle. That is a broad pattern resembling
  wu:fi36a10. Mammalian MAGI3 is also broadly expressed:
  [PMID:37163606 "Among the three paralogs of MAGIs, MAGI1 and MAGI3 are ubiquitously expressed, whereas MAGI2 is mainly expressed in neuronal synapses and kidney podocytes (10)."]
• Caveat. Bgee lists present calls only. magi3b may be present at lower levels in tissues without
  a call. No tissue-matched quantitative comparison or in situ side-by-side exists. ZFIN has one RT-PCR
  record (magi3b, whole embryo) and none for wu:fi36a10.

Summary. The copies overlap in endothelium, early embryo, ovary, brain, retina and gill. wu:fi36a10
alone appears to keep the broad, gar-like profile. magi3b has no expression domain of its own in the
available data.

4. Experimental evidence of function

Both copies (PMID:35649360, F0 CRISPR "crispants", brain angiogenesis):

• [PMID:35649360 "Targeting magi3a, magi3b, or dlg4b, but not dlg4a, partially impaired brain vascularization (Figures 4H and 4I), in agreement with the expression data."]
• [PMID:35649360 "This lack of additive effect suggests that Dlg4b, Magi3a, and Magi3b act within the same pathway, in which they are individually required."]
• [PMID:35649360 "Strikingly, magi3a, magi3b, or dlg4b crispant phenotypes were indistinguishable from the ΔETTV phenotypes (Figures 4H and 4I), with an identical reduction in number of cerebral vessels and a similar defect in vessel lumenization."]
• [PMID:35649360 "Accordingly, injections of magi3a, magi3b, dlg4a, or dlg4b sgRNAs did not affect the gpr124ΔETTV mutant vasculature (Figure 4K), suggestive of an interaction between the ETTV motif of Gpr124 and these PDZ proteins."]

Allele type: F0 mosaic crispants. The paper checked them by amplicon sequencing. No stable lines and no
RNA-less alleles were used, and transcriptional adaptation was not assessed.

magi3b only (PMID:25139857, morpholino screen): a "magi3" morphant, which ZFIN assigns to magi3b, had
shorter kidney motile cilia:
[PMID:25139857 "We found that the loss of six genes – BX470211.1, Dr.81747 (GLB1L2 in human), ect2l, magi3, si:ch211-71m22.1 and si:dkey-26i13.8 (KIF18B in human) – showed a significant (P<5.0×10−7, Student's t-test, two-tailed) reduction in the length of motile cilia in the kidney tubules"]
This is single-morpholino screen evidence. The paralog was not tested, and no mutant confirmation exists.

wu:fi36a10 only: in an adult vivo-morpholino screen, magi3a knockdown did not change factor VIIa levels:
[PMID:33134781 "The knockdown of zgc:123295, zgc:136461, acrosin like, magi3a, serine protease 56, si:dkey‐32n7.7, si:dkey‐21e2.12, serine protease 23, and tmprss12 did not affect Fviia levels."]
This result says nothing about fate.

No cross-rescue, stable mutant, double mutant or compensation data exist for either copy.

5. Fate classification

UNRESOLVED. Confidence that it is not pure BACKUP: moderate. Confidence in any positive fate: low.

What the data show

• Protein: conserved in both copies (sequence only). No domain loss, no gain.
• In the one tissue where both copies were tested (endothelium, during brain angiogenesis), each is
  individually required, and loss of both is no worse than loss of one. That is not what redundancy
  predicts. Under backup, the single crispants would be mild and the double crispant severe. The
  result fits a dosage or stoichiometric requirement in a shared Gpr124/Wnt7 pathway, with a
  phenotype ceiling set by the ETTV mutant. It also fits the two proteins acting together.
• Expression: asymmetric. wu:fi36a10 keeps a broad gar-like pattern; magi3b has a narrower
  profile with no unique domain.

Why no positive call

• DOSAGE would need matched expression and evidence that the total dose matters, such as
  heterozygote or double-heterozygote effects, or rescue by extra copies of either gene. F0 crispants
  are mosaic and cannot be titrated.
• PARTITION would need a magi3b-only or wu:fi36a10-only domain shown by in situ, together with a
  copy-specific phenotype. The only copy-specific phenotype (cilia) comes from a morpholino screen.
• INNOVATION: nothing suggests a new function.
• A degenerating magi3b (asymmetric loss, a nonfunctionalization trajectory) is compatible with the
  lower conservation and narrower expression, but it is contradicted by its requirement in
  angiogenesis.

What would change the call: stable (ideally RNA-less) single and double mutants with
central-artery counts; endothelial cross-rescue; quantitative expression of both copies in sorted
endothelium, kidney and retina alongside gar magi3.

6. GO annotation consistency across the pair

Source: annotation-comparison.md, after both reviews.

Shared and correctly so. Every IBA row (cytoplasm, cell-cell junction, signal transduction) is on
both copies and was accepted on both. So were the UniProt location rows (plasma membrane, bicellular
tight junction). Same-node IBA propagation is appropriate: both copies keep the full scaffold. For
cytoplasm, the magi3b IDA (PMID:25139857) is among the PAINT descendant evidence (ZFIN:ZDB-GENE-060503-301
in WITH/FROM), which is expected and not circular.

Removed on both, identically. The UniRule rows for kinase activity, transferase activity, ATP binding
and nucleotide binding. MAGI GK domains lack the catalytic subdomains, and neither copy has a P-loop.
This is a family-wide IEA error, not a pair asymmetry.

Asymmetries

• magi3b alone has the cilium assembly and cilium organization IMPs and the cytoplasm IDA. These reflect
  which copy the Choksi screen targeted, not a demonstrated biological difference. I kept the IMPs as
  non-core and did not propagate them.
• wu:fi36a10 alone has nucleus (IEA). This reflects different UniProt rule coverage, not biology. Kept
  as non-core.

Missing from both. Neither copy is annotated to the Gpr124-dependent brain angiogenesis role that
PMID:35649360 shows for both. ZFIN annotated only gpr124 (adgra2) from that paper. Because the magi3 data
are F0 crispants plus yeast two-hybrid, I did not add NEW process terms. Instead I raised the question
in both reviews. If added, the term should go on both copies together.

7. Open questions

• Do stable magi3b and wu:fi36a10 mutants, including an RNA-less allele to avoid transcriptional
  adaptation, reproduce the crispant angiogenesis phenotype? Are double heterozygotes affected (dosage)?
• Does Magi3b bind the Gpr124 ETTV motif like Magi3a?
• Is magi3b expressed at low levels in the tissues where only wu:fi36a10 has a call?
• Is the magi3 morphant kidney cilia phenotype real, and which copy was targeted?
• ZFIN still uses wu:fi36a10; NCBI uses magi3a.

References

PMID:10748157, PMID:25139857, PMID:33134781, PMID:35649360, PMID:37163606. Also the files
panther_tgd_pairs.tsv, annotation-comparison.md and
wu_fi36a10-bioinformatics/RESULTS.md.