Neural Crest Formation (GO:0014029) — Obsoletion Proposal

Bottom line: we propose to obsolete GO:0014029 neural crest formation
and its three regulation terms. We also propose two new terms, neural plate
border formation and neural crest progenitor maintenance. GO:0014029 contradicts itself: its text defines forming a region of
ectoderm (the neural plate border), while its place in GO is under epithelial
to mesenchymal transition. The 1,454 annotations to it therefore have no single
meaning. They mix border specifiers, crest specifiers, competence factors,
signalling modulators and ribosome-biogenesis genes. 1,322 of them (91%) are
electronic, and one machine-learned UniProt rule (ARBA00027170) alone produces
665. Only 84 rows are experimental (48 gene products, 36 papers). Each of these
can be remapped to a precise existing term, or to one of the two new terms. This
project holds the case, the replacement rules, the remapping of every
experimental annotation, and the draft GO request. It originated in the
Origins of the Neural Crest project. Nothing has
been submitted to GO yet.

The defect

GO:0014029 neural crest formation (checked against QuickGO and OLS, 2026-10-09):

• Text definition: "The formation of the specialized region of ectoderm
  between the neural ectoderm (neural plate) and non-neural ectoderm. The neural
  crest gives rise to the neural crest cells that migrate away from this region
  as neural tube formation proceeds." This describes forming the neural plate
  border region.
• Placement: is_a GO:0001837 epithelial to mesenchymal transition;
  ancestors include GO:0048762 mesenchymal cell differentiation and GO:0060485
  mesenchyme development. This describes making migratory crest cells.
• Children: GO:0014034 neural crest cell fate commitment is part_of
  both GO:0014029 and GO:0014033 neural crest cell differentiation, and
  GO:0014036 fate specification is part_of GO:0014034.

The two readings pick out different genes. A border specifier such as Gbx2,
Msx1 or Pax7 builds the border territory but does no EMT. A competence factor
such as Id3 or Myc keeps crest progenitors undifferentiated and does neither.
Both are annotated to the same term as Sox10. The evolutionary data make the
split unavoidable. Amphioxus and tunicates have a neural plate border, with
Pax3/7, Msx and Zic expressed there [PMID:18562679], but no neural crest. Border
formation cannot be part of neural crest formation.

How the term is used

Inventory from QuickGO, exact annotations only (inventory.py; full rows in
NEURAL_CREST_FORMATION_OBSOLETION/annotations.tsv, counts in
NEURAL_CREST_FORMATION_OBSOLETION/summary.md):

Where the electronic rows come from:

• ARBA:ARBA00027170 (GO_REF:0000117), 665 rows. One UniProt rule with six
  unrelated condition sets: Zic-type C2H2 zinc fingers, HMG box (Sox),
  winged-helix (Fox/Ets), BTB-kelch and others, each restricted to Amphibia,
  Anura, Pipoidea, Pipidae or Craniata. It learns "is a crest gene" across all
  network layers and writes one term for all of them.
• Ensembl Compara orthology (GO_REF:0000107), 570 rows, and the combined
  automatic pipeline (GO_REF:0000120), 87 rows, propagating from the
  experimental rows below.

No production GO-CAM in the local cache (gocams/) uses any of the four terms.

Proposed changes

1. New term: neural plate border formation. "The formation of the neural
   plate border, the region of ectoderm at the boundary between the neural plate
   and the non-neural ectoderm that is competent to give rise to neural crest,
   preplacodal ectoderm, dorsal neural tube and epidermis." Proposed part_of
   GO:0007398 ectoderm development. It should not sit under any crest or
   mesenchyme term.
2. New term: neural crest progenitor maintenance. "The process by which
   neural plate border and premigratory neural crest progenitor cells are kept in
   an undifferentiated, proliferative and multipotent state until neural crest
   specification." Proposed is_a GO:0019827 stem cell population
   maintenance. It is for competence factors such as Myc, Id3, Hairy2 and
   Pou5f3/Oct25. These are required for crest to form, but their loss causes
   progenitor arrest or death rather than a change of fate, and sustained Id3
   blocks differentiation.

Why a stem-cell parent. Premigratory crest cells are transient, multipotent
embryonic progenitors rather than lifelong stem cells, so the parent needs
arguing. GO already treats an equivalent pool this way: nephron progenitors
(cap mesenchyme) are covered by GO:0072038 mesenchymal stem cell maintenance
involved in nephron morphogenesis, which is_a GO:0035019 somatic stem cell
population maintenance (checked in QuickGO 2026-10-12). By that precedent the
tighter parent would be GO:0035019; the request should offer it and let GO
choose. The new term is needed because GO:0019827 on its own does not say
which population is maintained (option (c) below).
3. Obsolete GO:0014029. Replacements to suggest ("consider"): neural plate
border formation (new), neural crest progenitor maintenance (new), GO:0014033 neural crest cell differentiation,
GO:0014034 fate commitment, GO:0014036 fate specification, GO:0036032
delamination. GO:0014034 keeps its existing part_of GO:0014033.
4. Obsolete the regulation terms, replaced by the existing regulation
terms for crest cell differentiation:

Gaps considered but not proposed

All three were checked as absent from GO in OLS on 2026-10-09. None has an
annotation that clears our own evidence bar, so none goes into the request.

• (Negative) regulation of neural crest cell delamination. This would
  negatively_regulate GO:0036032. It is a standard regulation term of the kind
  GO creates on request. Its only candidate is chick TSPAN18 [PMID:23418345],
  and that case fails the overexpression rule. Knockdown removes Cad6B protein,
  yet "the timing of neural crest emigration, basal lamina breakdown and Cad7
  upregulation proceed normally in Tspan18-deficient cells". That cells are
  held back is shown only by extending Tspan18 expression. GO:0090301 is
  currently is_a GO:0010719, so its EMT sense would naturally land on this
  term. We will propose it once a loss-of-function case exists.

• Regulation of neural crest cell migration (general). GO has only the
  cardiac outflow-tract version (GO:1905310 and children), not a general
  parent. The one candidate in scope was zebrafish hdac4 [PMID:22676467], but
  its authors cannot tell specification from migration, so it stays on
  GO:1905292. We will propose the term when an annotation needs it.

• Neural crest cell apoptotic process / survival. polr1b knockdown kills
  crest precursors through p53, but that is a ribosome-supply requirement
  shared with any fast-growing tissue (→ GO:0006360). No gene in the remapping
  does crest-specific survival work, so a term made now would only collect
  necessity evidence.

Replacement rule for existing annotations

Decide each annotation by what the gene product does, using the layers worked
out in NEURAL_CREST_ORIGINS and the
neural crest GRN module:

Competence factors (decided 2026-10-09). Three options were considered:
(a) GO:0014033 neural crest cell differentiation; (b) a new term; (c)
GO:0019827 stem cell population maintenance. Option (b) was chosen. (a) is
awkward for factors that hold off differentiation. (c), used on its own, is too general
to say which population is maintained, so the new term sits beneath it as a
child. (The lin28a review marks a transferred GO:0019827 row as over-annotated
because there is no frog evidence for any such role; it does not argue against
the parent.)

Remapping of every experimental annotation

NEURAL_CREST_FORMATION_OBSOLETION/remapping.tsv has one row per distinct GOA
annotation: all 84 experimental GO:0014029 rows plus the 17 regulation-term
rows (12 distinct). That is 94 rows and 62 distinct gene–paper decisions. Each
row gives an action, replacement ids and labels, a rationale and a verbatim
quote. Every quote and every replacement id passes check_remap.py (live GO
terms with matching labels, quotes found in the cached papers). The table is
assembled by merge_remap.py from four review batches (remap/*.tsv, each with
a .notes.md) and from the gene reviews already in this repo.

Decisions by gene (species; paper):

What the remapping shows:
- Only 14 of 94 rows land on a precise crest-cell term (fate
specification or delamination): the SoxE genes, the ubiquitin group and SOX9.
The rest move to the border term (11), the progenitor term (11), a regulation
term (14), the general differentiation term (32, of which 28 are zeb2
accessions from one paper) or a non-crest term (polr1b), or are removed or left
undecided. The obsoleted term was carrying all of these meanings.
- Ribosome biogenesis is split on purpose. polr1b knockdown makes crest
precursors die from a ribosome shortage, which is a requirement, not a crest
step. TCOF1 and NOLC1 knockdown switches cells from crest to CNS identity
without affecting rRNA output or survival, which is participation in
specification [PMID:26399832].
- GO lacks regulation terms for crest migration and delamination. TSPAN18
therefore goes to a generic EMT term. That row needs a second look under
the overexpression rule: knockdown shows Tspan18 keeps Cad6B protein levels
up, but not that it delays emigration (see Gaps considered but not
proposed).
- Overexpression alone is not participation (rule added 2026-10-09). A
factor that induces crest markers when overexpressed shows it can drive the
program, not that it does. Rows resting only on such data are marked
over-annotation: zic2-a and zic4. zic1 and sox10 also have overexpression-only
rows (PMID:9435279, PMID:9655809, PMID:9739105; PMID:12812785), but these are
carried forward because each gene has its own knockdown evidence for the same
role (PMID:15843410; PMID:12885557). zic5 is kept because its paper includes a
dominant-negative construct, and bmper because its paper shows it is required.
These calls rest on abstracts; a loss-of-function experiment in a full text
would change them.
- Weaker calls flagged in the notes: zeb2 and cnbpa rest on abstracts; rgs2 rests on its abstract
only; for KBTBD8, GO:1905297 (regulation) is a defensible alternative to
GO:0014036.

Impact on this repo

18 gene reviews mention one of the four terms (2026-10-09):

• Border specifiers, done. gbx2, MSX1, PAX7, TFAP2A and TFAP2C: NEW rows
  now point to the proposed term (id: NTR). pax3-a and zic1: their GOA
  GO:0014029 rows are MODIFY → NTR (+ GO:0014034). All seven carry a
  proposed_new_terms entry. Applied with
  NEURAL_CREST_FORMATION_OBSOLETION/apply_border_ntr.py.
• Crest specifiers, already consistent. sox8, sox9-a, sox10, snai1, snai2
  and twist1 MODIFY their GO:0014029 rows to GO:0014036.
• Competence factors, done. hes4-a and id3-a: their GOA GO:0014029 rows
  are MODIFY → NTR neural crest progenitor maintenance. myc-a and pou5f1.1:
  NEW rows re-pointed to it. All four carry a proposed_new_terms entry.
  Applied with apply_border_ntr.py --competence.
• Outside the crest project. human LRP6 keeps GO:0014029 (IDA,
  PMID:11029007) as non-core. Revisit it with the signalling-modulator group.
• lin28a, sox2, sox3-a and TFAP2B mention the term only in prose.

Draft request to GO (not submitted)

Obsolete GO:0014029 neural crest formation (and GO:0090299/0090300/0090301); add "neural plate border formation" and "neural crest progenitor maintenance"

GO:0014029 is defined as "the formation of the specialized region of ectoderm
between the neural ectoderm (neural plate) and non-neural ectoderm", but it is
is_a GO:0001837 epithelial to mesenchymal transition. The definition
describes the neural plate border, a territory that also gives rise to
preplacodal ectoderm, dorsal neural tube and epidermis, and that exists in
chordates without neural crest (amphioxus, tunicates). The placement describes
the generation of migratory crest cells. Current annotations (1,454; 84
experimental) mix border specifiers (Pax3, Zic), crest specifiers (SoxE),
competence factors (Id3, Hairy2) and others; 665 come from a single ARBA rule
spanning unrelated families.

Proposal: (1) new BP "neural plate border formation", part_of GO:0007398
ectoderm development; (2) new BP "neural crest progenitor maintenance",
is_a GO:0019827 stem cell population maintenance (or its child GO:0035019
somatic stem cell population maintenance, following the GO:0072038 nephron
progenitor precedent), for factors that keep
border/crest progenitors undifferentiated and competent (Myc, Id3, Hairy2,
Pou5f3); (3) obsolete GO:0014029 with consider: the two new terms, GO:0014033,
GO:0014034, GO:0014036, GO:0036032; (4) obsolete GO:0090299, GO:0090300 and
GO:0090301, replaced by GO:1905292, GO:1905294 and GO:1905293.
A per-annotation remapping of the 84 experimental annotations is attached.

STATUS

Last updated: 2026-10-09

• [x] Defect documented (text vs placement), checked in QuickGO and OLS
• [x] Inventory of all annotations to the four terms (inventory.py, annotations.tsv, summary.md)
• [x] Source of the electronic bulk identified (ARBA00027170; Ensembl orthology)
• [x] Replacement rule drafted; regulation-term mappings identified
• [x] Border-specifier reviews in this repo moved to the proposed term (7 reviews)
• [x] Competence-factor replacement decided: new term neural crest progenitor maintenance (2026-10-09)
• [x] Applied to hes4-a, id3-a, myc-a, pou5f1.1; module competence part updated
• [x] Review the remaining experimental annotations (four batches; 2026-10-09)
• [x] Review the 17 regulation-term annotations (12 distinct)
• [x] Per-annotation remapping table for GO (remapping.tsv, 94 rows, checker clean)
• [ ] Resolve the 5 UNDECIDED rows (needs PMID:27716508 full text; Chrd full text; hes4-a NOT row)
• [x] Gap terms checked (2026-10-09): delamination regulation, migration regulation and crest apoptosis are absent from GO; none proposed yet (no annotation clears the bar)
• [ ] Re-decide TSPAN18 (GO:0010719 rests on overexpression; knockdown supports Cad6B stabilisation only)
• [x] Overexpression-only rule applied: zic2-a and zic4 marked over-annotated; zic1 and sox10 kept because they have loss-of-function evidence
• [ ] Report ARBA00027170 to UniProt (see the rule-reviewer workflow)
• [ ] Submit the GO request (needs sign-off)

NOTES

2026-10-09 (OpenScientist checks)

Three OpenScientist jobs were run on the new terms and the module's
regulatory links. None ran the requested data analysis; all three are
literature reviews (issue #4469). Two of them argued the new terms are
redundant with GO:0014029, but they did not know it is proposed for
obsoletion. One suggested GO:0019827 with an extension instead of
neural crest progenitor maintenance. Decision: keep the new term. Its
definition ("kept in an undifferentiated, proliferative and multipotent
state until neural crest specification") already says nothing about whether
the blastula program is retained or re-activated, which is still open
[PMID:35534333 "This suggests that neural crest cells either retain or
re-activate a pluripotency GRN from the blastula"]. New evidence for the
border/specifier split: pax3 and zic1 "initially promote neural crest gene
activation but are downregulated as neural crest identity emerges"
(Montequin & LaBonne 2026, PMID:41718037; PubMed-verified and cached 2026-10-12).

2026-10-09 (overexpression rule, gap terms)

The project lead's call: overexpression evidence alone is over-annotation. It
is applied per gene. A row is carried forward only if the same gene has
loss-of-function or other independent evidence for the role. zic2-a and zic4
fail this test. zic1 (PMID:15843410) and sox10 (PMID:12885557) pass. The
remapping now has 52 replace, 3 over-annotated, 2 remove and 5 undecided
decisions. Gap terms were checked in OLS. The delamination regulation term looked
needed for TSPAN18, but on rereading the paper the delay of emigration is
shown only by overexpression, so it was not proposed and TSPAN18 is flagged
for re-decision.

2026-10-09 (remapping)

Four review batches covered the remaining 72 distinct rows, and 22 more rows
were taken from the existing gene reviews. Result: 94 rows, 62 distinct
gene–paper decisions (52 replace, 3 over-annotated, 2 remove, 5 undecided). All quotes and ids
pass check_remap.py. Spot checks by the coordinator: GO:0044335 (Wnt in crest
differentiation) is indeed obsolete, which is why LRP6 maps to GO:1905294; and
the chicken SOX9 paper's title and text support delamination. Inventory
coverage was checked: every experimental and regulation row is in the table.

2026-10-09 (later)

Decided option (b) for competence factors: a second new term, neural crest
progenitor maintenance (is_a GO:0019827). Applied to the four competence
reviews and to the module. No GO:0014029 term IDs remain in the module or in
the project's core functions. The 4 competence reviews and the 7 border
reviews still have GOA rows on GO:0014029, now MODIFY to an NTR, which is the
intended state until GO acts.

2026-10-09

Project created from the NEURAL_CREST_ORIGINS convention discussion. The
earlier convention (border and competence genes at GO:0014029, specifiers at
GO:0014036) used the term's text definition. Checking its placement showed the
text and placement disagree, so the term itself was judged unfit and an
obsoletion was proposed rather than a definition fix. Inventory run: 1,471 rows
across the four terms. All 665 ARBA rows come from ARBA00027170, whose six
condition sets span Zic, SoxE, Fox/Ets and kelch families.