OTX2 curation notes

UniProt: P32243 (OTX2_HUMAN). HGNC: OTX2. Locus 14q22.3.
PANTHER family: PTHR45793 (Paired homeobox transcription factors); bicoid subfamily
of the paired-like homeodomain class.

Journal-style working notes. Provenance is recorded inline as
[PMID:xxxxx "<verbatim supporting text>"]. Anything without a citation is my own
synthesis and is flagged as such.

2026-09-20 — initial pass

Seeded with ai-gene-review fetch-gene human OTX2 (36 GOA annotations, 15 seeded
references). Deep research run with the falcon provider →
OTX2-deep-research-falcon.md. That report is good on the disease/cancer side and on
the hypothalamic–pituitary axis, but it under-covers three things I had to research
separately: (1) the classic mouse null phenotypes, (2) the OTX2/GBX2 boundary that
positions the isthmic organizer, and (3) the non-cell-autonomous homeoprotein transfer
biology. Those are filled in below.

The immediate trigger for this review is
PMID:42760324(https://pubmed.ncbi.nlm.nih.gov/42760324/), Two parallel neural
ectoderm progenitors contribute to the developing brain (Jokhai, Dundes et al., Nat
Neurosci 2026), which makes OTX2 the marker of one of the two parallel brain
progenitors. That paper is the reason the neural_ectoderm_anteroposterior_specification
module needs OTX2 grounded, and this review is the prerequisite.

1. What the protein is

A 289/297-aa bicoid-class homeodomain transcription factor; the homeodomain spans
residues ~38–97 and is the DNA-binding module. Recognizes bicoid-type TAATCC sites;
binds as monomer or dimer. DNA binding alone does not determine output — activation
vs. repression depends on cofactors and promoter context.

Worth noting for the molecular-function call: OTX2 belongs to the extended homeodomain
family that prefers CpG-methylated sites, established by methyl-SELEX across 542 human
TFs PMID:28473536. So the
GO:1990837 sequence-specific double-stranded DNA binding IDA from that paper is a
high-throughput-but-careful in-vitro binding measurement, not a functional assay.

2. Core developmental role: anterior neural ectoderm

This is the function the gene is named for and the one the 2026 paper turns on.

Mouse Otx2 nulls delete the entire forebrain and midbrain, and the defect is traced
to a specification failure during gastrulation, not to later degeneration
PMID:7588062, with the authors concluding
PMID:7588062. Independently replicated with
different alleles by Matsuo et al. PMID:7590242 and Ang et al. PMID:8565836, the
latter also showing gastrulation and axial-mesoderm defects — i.e. OTX2 acts earlier
than the neural plate proper.

The posterior limit of the OTX2 domain is the midbrain–hindbrain organizer. GBX2
represses OTX2 and positions that border
PMID:10490024, and the relationship is instructive, not
merely correlative — ectopic GBX2 shifts the OTX2 border and a new organizer forms at
the new border
PMID:10490024. Their conclusion
PMID:10490024.

This mutual-repression pair is exactly the aNE/pNE axis of
PMID:42760324. The new contribution is commitment: challenged with
hindbrain-inducing signals, human OTX2+ aNE does not convert
PMID:42760324, with a chromatin correlate
PMID:42760324. Conservation extends to hemichordates
PMID:42760324.

Curation consequence. GOA has GO:0030900 forebrain development and
GO:0030901 midbrain development only as TAS from PMID:15705863, a medulloblastoma
digital-karyotyping paper. The terms are right; the citation is thin (see §6). The much
better support is the mouse null series above, and there is no human-specific
experimental annotation for the anterior-patterning role at all — the strongest,
best-replicated OTX2 function is the one with the weakest evidence code in GOA. That is
worth saying out loud in the review.

3. Retina, RPE and pineal

Second well-established program, and the one behind the human ocular phenotypes.

Photoreceptor fate determination and pineal development, by conditional ablation
PMID:14625556,
acting through CRX PMID:14625556,
and sufficient as well as necessary PMID:14625556.

In human RPE specifically, OTX2 drives RPE-restricted target genes. Direct promoter
occupancy and functional requirement for DCT
PMID:12559959 and
PMID:12559959. This is the
experiment behind the human GO:0000978 / GO:0001228 / GO:0045944 IDAs — a genuine,
human, promoter-level activator assay.

The second human RPE target is BEST1, where OTX2 works combinatorially rather than alone
PMID:20530484. Note this is the sole basis for the
GO:0005515 protein binding, GO:0032991 protein-containing complex and
GO:0065003 protein-containing complex assembly annotations — the interaction is real,
but "protein binding" is uninformative per project guidance and the complex terms are
generic restatements of the same co-activation result.

Human fetal expression is consistent and RPE-restricted at these stages
PMID:19414065 — this
is the GO:0005634 nucleus EXP.

4. Hypothalamic → pituitary axis (non-cell-autonomous, via FGF10)

The mechanism is worth stating precisely because it is a case where OTX2's effect on
another tissue is mediated entirely by transcriptional control of a secreted ligand —
OTX2 itself is not secreted here. Patient-derived organoid work with R127W: reduced
hypothalamic FGF10 → reduced LHX3 in oral ectoderm → pituitary progenitor apoptosis,
rescued by CRISPR correction PMID:31845906. Human embryonic expression supports a
hypothalamic origin for the endocrine phenotype rather than an autonomous anterior-
pituitary defect PMID:33950863.

This is the mechanistic basis on which a regulation of fibroblast growth factor receptor signaling pathway annotation could be defended — but note GOA's actual citation for
GO:0040036 is again PMID:15705863, not this work (§6).

5. Non-nuclear / moonlighting biology (homeoprotein transfer, eIF4E)

OTX2 is one of the homeoproteins that transfers between cells and engages the
translation machinery. Two strands:

eIF4E binding. Direct, demonstrated by GST pull-down with human OTX2 protein
PMID:15247416. Construct provenance confirms the human
protein was used PMID:15247416. This is the real evidence for GO:0008190, and it is not the
reference GOA cites. See §6.

Growth cone localization and axon turning. In the Brunet et al. Engrailed-2 paper,
OTX2 was the specificity control and was assayed — so this is not a paralog
mis-attribution despite the title
PMID:16267555. Internalization into
growth cones is positive PMID:16267555, supporting GO:0030426 growth cone. But the guidance result is
explicitly negative for the property the assay was measuring
PMID:16267555 — OTX2 only "weakly attracted both the nasal and temporal
populations". Annotating human OTX2 to GO:0007411 axon guidance from the experiment
whose published conclusion is that OTX2 lacks the guidance behaviour is an
over-annotation. Flagging rather than removing: a weak bidirectional attraction was
measured, so the term is not flatly false.

Critical-period plasticity. The most important instance of transfer, though not
currently in GOA: visual experience drives non-cell-autonomous OTX2 accumulation in
cortical PV interneurons and this times the critical period
PMID:18692473, and it is required
PMID:18692473. I considered proposing a NEW term here and decided against it —
see §7.

6. Reference problems found (the main QC finding of this review)

Five annotations hang off one reference that probably does not support most of them.
GOA has a block of five TAS annotations, all UniProt, all dated 20050607, all citing
PMID:15705863 (Boon, Eberhart & Riggins, Genomic amplification of orthodenticle
homologue 2 in medulloblastomas, Cancer Res 2005):

The paper is a digital-karyotyping/copy-number study; its abstract's only functional
statement is "OTX2 functions to specify the fate of neuroectoderm in various regions of
the developing brain". It is not open access (Europe PMC: isOpenAccess: N,
inEPMC: N), so I cannot read the intro/discussion to check whether the FGFR and
smoothened statements are traceable there. Per project policy I do not overrule a curator
on evidence I cannot see, so those two are kept and flagged rather than removed.

For GO:0008190 the situation is different: I found affirmative evidence for the
claim in a different paper (PMID:15247416, human OTX2 GST pull-down), so the annotation
is scientifically sound and only its citation is wrong. Recorded as
reference_review.replacement on PMID:15705863 rather than by editing the GOA-supplied
original_reference_id.

GO:0071542 dopaminergic neuron differentiation (TAS, PMID:24431302) cites a Wnt
review (Arenas 2014, J Mol Cell Biol) — review-derived TAS, weaker than the companion
IGI from PMID:19951692, where OTX2 is genuinely part of the tested combination
(Lmx1a + Otx2 + FoxA2 synergy).

GO:0005634 nucleus (IDA, PMID:24399192) cites a hEPI-NCSC→dopaminergic-neuron
differentiation protocol paper. Nucleus is certainly right for OTX2, so this is
harmless, but the citation is incidental (OTX2 used as a marker), not a localization
study.

ISS block (GO_REF:0000024, WITH UniProtKB:P80206 = mouse Otx2). GO:0090009 primitive streak formation, GO:2000543 positive regulation of gastrulation,
GO:0040019 positive regulation of embryonic development,
GO:0045893 positive regulation of DNA-templated transcription. The mouse source is
sound (Ang et al. gastrulation defects, PMID:8565836) and these are legitimate transfers,
but they sit at a much lower resolution than what is actually known — "positive regulation
of embryonic development" says almost nothing for a gene whose null deletes the forebrain.
Keeping as non-core; the informative content lives in the anterior-patterning terms.

7. NEW terms considered and rejected

Running the project's NEW discipline over the candidates:

• Critical-period / PV-interneuron plasticity (PMID:18692473). Tempting — the
  evidence is strong and the biology is distinctive. Rejected for this review because the
  work is mouse visual cortex with no human experimental counterpart, and because the
  cleanest candidate terms describe the plasticity process, which the PV interneuron
  network performs; OTX2's role is as the transferred signal that times it. Better raised
  as a suggested_questions item about whether a "homeoprotein transfer" representation
  exists, than asserted as human involvement.
• GO:0021999 neural plate anterior/posterior regionalization. This is the term the
  new paper is nominally about, and it is not on OTX2. Comparator check settles it
  against proposing: in mammals the term is annotated to CELSR2, NOG and SSBP3 (all IEA),
  and its experimental annotations are zebrafish axin1, bptf, smad2, sfpq,
  sbno1 and gsc — morphogen and chromatin machinery, not regional selector
  transcription factors. A systematic absence across species and MODs is a convention, not
  a gap. The claim belongs in the module, where a developmental program is the right unit.
• GO:0009952 anterior/posterior pattern specification. Proposed, then withdrawn. The
  comparator evidence was good — mouse Otx2 carries it by both IMP and IGI, as does Hoxa1
  (IMP, IGI), and Gbx2 (GO:0030917) and Six3 (GO:0021797) carry terms that entail it,
  while the human orthologues GBX2, SIX3, EN1 and HOXA1 have none, which reads as a species
  artefact rather than a convention. The redundancy check is what killed it: human OTX2
  already carries GO:0090009 primitive streak formation by ISS, and an annotation to
  GO:0090009 already entails GO:0009952, so proposing the latter would have added an
  ancestor of an existing annotation — which the project rules reject outright.
• The GO:0090009 → GO:0009952 edge, stated precisely. An earlier draft of the bullet
  above (and of the YAML reason, and of the 2026-09-20T163726Z history record) called
  GO:0090009 a descendant of GO:0009952 without qualifying the relation. That wording
  was retracted on 2026-10-03, because the whole withdrawal rests on this single edge and
  the unqualified word is wrong for it. Queried against the local oaklib sqlite:obo:go
entailed_edge closure — the same artefact the module validator resolves labels through,
  GO release 2026-07-26:
• GO:0090009 has exactly one is_a parent, GO:0048646 anatomical structure formation involved in morphogenesis, and its is_a closure contains no term in the
  regionalization branch. There is no is_a path to GO:0009952.
• GO:0009952 is reached from GO:0090009 only through BFO:0000050 part_of (and
  RO:0002131 overlaps).
• The same holds for GO:0030917, the term that did survive: it reaches GO:0009952
  through BFO:0000050, not is_a. GO:0021797 is the one genuine rdfs:subClassOf
  descendant of the four checked.
  The conclusion is unchanged, because GO's true-path rule propagates annotations over
  part_of as well as is_a: an annotation to GO:0090009 entails GO:0009952 either
  way. Recorded because an is_a-only reading of the closure would make GO:0009952 look
  proposable, and that reading was argued four times across review passes before the
  closure was actually queried.
• GO:0021978 telencephalon regionalization. Mouse Otx2 carries it by IMP. Rejected on
  the same ground: it reaches GO:0030900 forebrain development, which human OTX2 already
  has, by BFO:0000050 (it is not under GO:0009952 at all).
• GO:0030917 midbrain-hindbrain boundary development — this one survived, and is
  proposed. The claim is narrower and better grounded than the AP-patterning one: the
  organizer forms at the OTX2 caudal border, and moving that border with ectopic GBX2
  moves the organizer PMID:10490024. Comparator check is favourable
  rather than merely neutral — queried against QuickGO, GO:0030917 has 8 experimental
  annotations in mouse (Gbx2, Fgf8, Wnt1, En1, Lrp6, all IMP) and 138 in zebrafish, so GO
  does not withhold it from this class of gene. Redundancy check passes: checked against all
  37 OTX2 annotations in the part_of-inclusive closure, only the self-match comes back;
  GO:0030900 and GO:0030901 are siblings of GO:0030917 under GO:0007420 brain development, which OTX2 does not itself carry.
• The GO:0030917 ISS donor was wrong, and the qualifier was too strong. Both found on
  2026-10-04 after a reviewer asked whether the ISS WITH actually pointed at a holder of
  the term. It did not.
• Donor. The row cited UniProtKB:P80206 (mouse Otx2). QuickGO returns zero
  GO:0030917 annotations for P80206 — and zero for human P32243 — so the WITH named a
  gene that does not carry the term, which is not a transfer at all. The actual holders
  among the orthologues are zebrafish: otx2 (UniProtKB:Q91981, Swiss-Prot), otx2a
  and otx2b, all IGI from PMID:16611693. The donor is now Q91981, the reviewed entry.
  My earlier note said "zebrafish otx2a, by IGI" — the right species and evidence code,
  but the paralog I named is a TrEMBL entry when a reviewed one carrying the same
  annotation exists.
• Qualifier. The row said involved_in. All 146 experimental GO:0030917 rows in
  mouse and zebrafish use acts_upstream_of_or_within; every involved_in row on the term
  is IEA/ARBA. An ISS must not strengthen what it transfers, and this one was doing exactly
  that. Now acts_upstream_of_or_within, which is also what the project's own NEW rule
  requires: the boundary structure is built by the isthmic organizer genes, and what OTX2
  contributes is where it forms. The zebrafish paper puts it well — loss of Otx function
  turns the presumptive midbrain into extended r1 PMID:16611693, and Fgf8 only maintains the boundary that the Otx domain defines
  PMID:16611693.
• Consequence for core_functions: GO:0030917 is deliberately not in
  directly_involved_in, since listing it there would re-assert the participation the
  annotation now declines to claim. The border-positioning role stays in the prose.
• Worth recording because the error was invisible to every check that ran. The term id and
  label were right, the comparator argument was right, the mechanism was right, and the
  donor accession was a real mouse Otx2 entry — so schema validation, label validation and
  the redundancy check all passed. Only asking "does the WITH gene actually carry this
  term?" catches it, and nothing in the pipeline asks that.
• Anything medulloblastoma. Oncogenic re-expression (PMID:21964830, PMID:28213356,
  PMID:32686664, PMID:39025928) is disease biology, not normal gene function. The
  chromatin/enhancer work (PMID:28213356) and the LASR/splicing work (PMID:39025928) are
  mechanistically interesting and I cite them in references, but they establish what
  OTX2 can do when aberrantly retained in a tumour, not a normal-physiology process term.

8. Open questions for experts

Carried into suggested_questions / suggested_experiments:

• Is the eIF4E-binding activity used physiologically by OTX2, or is it a shared
  homeodomain-family property with no OTX2-specific readout?
• Should the GOA TAS block from PMID:15705863 be re-refereed by UniProt? Three of the
  five look unsupported by the cited paper.
• Does human OTX2 have a demonstrable role at the aNE stage that could be annotated
  directly, now that PMID:42760324 provides a human in-vitro system?

9. A deep-research summary that compressed its source wrongly

Recorded here because it is a reusable warning about the -deep-research-*.md files, not
an OTX2 fact. Added 2026-10-04.

The companion module asserted that in chick the Otx2 and Gbx2 domains "are separate at
HH4–HH8, overlap slightly around HH9, and are contiguous and mutually exclusive by HH10".
That came from the falcon deep-research file, which says "separated at HH4–8". Fetching the
cited primary (Hidalgo-Sánchez et al. 2022, PMID:35401126, full text in PMC) shows the
HH4 half is wrong:

• At HH4 the genes are simply expressed at opposite ends — "At stage HH4, Otx2 and Gbx2
  expressions are observed in the rostral and caudal portion of the chick embryo,
  respectively". Nothing about a gap.
• The non-contiguity is specific to HH8 — PMID:35401126.
• PMID:35401126.

So "HH4–HH8" silently extended a stage-specific observation backwards across four stages.
The lesson: a deep-research file's citations are reliable enough to resolve and fetch —
every primary it named for the module's four comparative claims turned out to be real and
on point — but its compressions of those primaries are not quotable and should not be
restated as fact without reading the source. This one survived because no validator can
check a claim in a prose description; it was caught only when a reviewer asked for
source_id on exactly those claims, which forced the primaries to be read.