Origins of the Neural Crest

IN_PROGRESS BIOLOGY_DOMAIN

Species: XENLA, human, CIOIN

Genes: foxd3-a sox10 snai2 sox9-a twist1 ets1-a myc-a id3-a sox8 pax3-a zic1 MSX1 TFAP2A hes4-a gbx2 pou5f1.1 lin28a snai1 sox2 sox3-a TFAP2B TFAP2C PAX7

Warnings (2)

Project NEURAL_CREST_ORIGINS: the gene regulatory network that made the vertebrate head

Bottom line (scoping): the neural crest is the textbook vertebrate novelty.
It is a migratory, multipotent cell population that builds most of the head
skeleton, the peripheral nervous system and the pigment cells. Comparative work
in lamprey, amphioxus and tunicates now explains much of its origin: an ancient
chordate neural plate border programme was inherited, and new neural crest
specifier circuits were then recruited into it stepwise. A
blastula-stage pluripotency programme was also retained or reactivated. This
project reviews the GO annotations of the genes at each layer of that network.
We start with Xenopus laevis, where most of the functional network was
mapped. The aim is a high-quality annotation set that keeps three roles
apart: the border specifiers (ancestral), the crest specifiers (largely
vertebrate innovations) and the pluripotency factors (repurposed). That
separation is what the evolutionary story depends on, and the current GO
annotations blur it.

Motivation

Gans and Northcutt proposed that the vertebrate "new head" was built from two
embryonic novelties, the neural crest and the cranial placodes
[PMID:17732898 "Neural crest and the origin of vertebrates: a new head"].
Forty years later the question has become one of gene regulation. Which parts
of the neural crest gene regulatory network (GRN) were already present in the
invertebrate chordate ancestor, and which were added at the base of vertebrates?

The evidence so far, from the outgroups:

Lineage Key observation Reference
Lamprey (jawless vertebrate) Most of the gnathostome NC GRN is already deployed in lamprey, so the network is at least as old as crown vertebrates [PMID:17765683 "Ancient evolutionary origin of the neural crest gene regulatory network"], [PMID:19104059]
Lamprey Lamprey cranial NC lacks most of the amniote cranial-specific transcriptional circuit [PMID:27339986] and resembles amniote trunk NC; that circuit was added gradually in gnathostomes, so the ancestral NC was trunk-like [PMID:31645763 "Evolution of the new head by gradual acquisition of neural crest regulatory circuits"]
Lamprey SoxE paralog duplication and subfunctionalisation in the branchial skeleton [PMID:21889937]
Lamprey NC and blastula pluripotency networks share regulatory factors as far back as the vertebrate ancestor. Lamprey pou5 is expressed in animal pole cells but not in NC, yet lamprey and Xenopus pou5 both promote NC formation [PMID:39060477 "Shared features of blastula and neural crest stem cells evolved at the base of vertebrates"]
Amphioxus (cephalochordate) The genome has homologs of most NC GRN genes. Border patterning genes and their BMP responsiveness are conserved, but many NC specifier genes are not expressed at the neural plate border. The AmphiFoxD regulatory region drives mesoderm, not NC, reporter expression in chick [PMID:18562679 "Insights from the amphioxus genome on the origin of vertebrate neural crest"], [PMID:12397104], [PMID:14651928]
Amphioxus / vertebrates Amphioxus FoxD and vertebrate FoxD3 are both repressors with similar DNA binding, but only FoxD3 induces ectopic NC in chick, through a vertebrate-specific N-terminal motif. Protein evolution, not only cis-regulatory change, contributed [PMID:24252777 "A novel N-terminal motif is responsible for the evolution of neural crest-specific gene-regulatory activity in vertebrate FoxD3"]
Ciona (tunicate, sister group of vertebrates) The cephalic melanocyte lineage (a9.49) is a "rudimentary neural crest"; targeted misexpression of Twist (Ci-Twist-like-2) reprograms it into migrating ectomesenchyme [PMID:23135395 "Identification of a rudimentary neural crest in a non-vertebrate chordate"]
Ciona Migratory bipolar tail neurons arise from the neural plate border [PMID:26524532 "Migratory neuronal progenitors arise from the neural plate borders in tunicates"]
Ciona Neural crest and cranial placodes share an evolutionary origin in a border cell population [PMID:30069052 "Shared evolutionary origin of vertebrate neural crest and cranial placodes"], [PMID:26258298]
Lamprey / jawed vertebrates Neofunctionalisation of the endothelin pathway drove NC diversification into skeletal cell types [PMID:32939088 "Evolution of the endothelin pathway drove neural crest cell diversification"]

A second line of work concerns potency. Neural crest cells share a regulatory
programme with pluripotent blastula cells (Pou5f3/Oct25/Oct60, Ventx, Sox2/3,
Snail, Id3, Myc, Sox5, FoxD3 [PMID:25931449]; Lin28 is linked to the crest
separately, from chick and lamprey work [PMID:30520734, PMID:39060477]). This suggests the crest kept, or reacquired, blastula-stage
potential, and that this may be the true vertebrate innovation
[PMID:25931449 "Shared regulatory programs suggest retention of blastula-stage potential in neural crest cells"],
[PMID:33542111 "Reactivation of the pluripotency program precedes formation of the cranial neural crest"],
[PMID:36182685 "Pluripotency factors are repurposed to shape the epigenomic landscape of neural crest cells"],
[PMID:30144418]. Reviews: [PMID:25903629 "Evolution of vertebrates as viewed from the crest"],
[PMID:25564621 "Establishing neural crest identity: a gene regulatory recipe"],
[PMID:31619682 "A genome-wide assessment of the ancestral neural crest gene regulatory network"].

Why this is a GO curation problem

GO has a well-populated neural crest process branch:
GO:0014029 neural crest formation, GO:0014034 neural crest cell fate
commitment, GO:0014036 neural crest cell fate specification, GO:0014033
neural crest cell differentiation, GO:0014032 neural crest cell development,
and GO:0001755 neural crest cell migration. Experimental annotations to it
come mainly from zebrafish, Xenopus and mouse knockdown and loss-of-function
work. Expected problems to test, gene by gene:

  1. Layer conflation. Border specifiers (Pax3, Zic1, Msx1, Tfap2a, Hes4,
    Gbx2) act upstream of crest specification. GO has no "neural plate border
    specification/formation" process term; the nearest are neural plate
    pattern/regionalisation terms (GO:0060896, GO:0060897). So border genes
    may be annotated directly to NC fate specification, and NC specifiers to
    "neural crest formation". Check whether the existing terms are precise
    enough, and whether a border term should be proposed (proposed_new_terms).
  2. Annotate the biology, not the experiment. A GO annotation records our
    synthesised judgement of what the gene product does, reached by biological
    reasoning across several lines of evidence. It is not a transcript of one
    assay. For each gene, the review should bring together:
  3. where and when it is expressed (border, premigratory or migratory crest);
  4. its molecular activity (DNA binding, activation or repression, partners);
  5. its direct targets and enhancer occupancy in the network;
  6. gain-of-function sufficiency (ectopic crest, reprogramming);
  7. loss-of-function phenotypes;
  8. conservation of that role across vertebrates and its state in the
    outgroups.

From that combined picture, place the gene in the network layer where it
does its work. A border specifier, a crest specifier and a competence
factor each warrant a different process term. Where the biology does not
support a process term, it does not stand just because an experiment is
attached to it.
3. Molecular function precision. Many entries are probably still at
generic DNA-binding transcription factor activity or carry protein binding
IPIs. Where the evidence supports it, prefer RNA polymerase II-specific
activator/repressor terms. Examples: Snai2 as a repressor; FoxD3 as a
repressor/pioneer factor; Id3 as a bHLH dimerisation inhibitor, not a DNA
binder.
4. Homeolog and paralog splitting. In X. laevis L/S homeologs
(foxd3-a/-b, pax3-a/-b, sox9-a/-b, id3-a/-b, hes4-a/-b),
experimental annotations may be attached to only one homeolog. Note this in
the reviews; do not "fix" it by assertion.
5. Pluripotency factors and the NC. Can Pou5f3/Oct25, Lin28 or Ventx
properly be annotated to neural crest processes? Or is the role the
maintenance of a competence state that the crest inherits? The second would
argue for a stem cell population maintenance term rather than NC
specification.

The evolutionary claims belong in gene descriptions and core_functions as
biology, and in the project synthesis. GO itself has no "evolutionary novelty"
terms, and we will not invent any.

Scope and species

Accessions below were checked against the UniProt REST API on 2026-10-01.

STATUS

Last updated: 2026-10-01

Tier 1 — Neural crest specifiers (candidates for vertebrate-specific recruitment) — COMPLETE 2026-10-01

Tier 2 — Neural plate border specifiers (ancestral chordate layer) — COMPLETE 2026-10-05

Tier 3 — Blastula pluripotency programme retained in the crest — COMPLETE 2026-10-07

Tier 4 — Missing module members (from module deep research) — COMPLETE 2026-10-08

Human entries are used, as for TFAP2A: the key data are from chick, and no
reviewed chicken or frog entries exist (UniProt checked 2026-10-08).

Tier 5 — Outgroup comparators (optional)

Module

Synthesis deliverables

NOTES

Superseded term choices. Notes dated before 2026-10-09 record decisions
made when border and competence genes were annotated to GO:0014029. That
term is now proposed for obsoletion (see Decision (2026-10-09) below and
NEURAL_CREST_FORMATION_OBSOLETION).
Read GO:0014029 in those notes as neural plate border formation (NTR) for
border specifiers and neural crest progenitor maintenance (NTR) for
competence factors. Where those notes say crest-formation rows were kept or
accepted (e.g. id3-a, hes4-a, zic1), the rows are now MODIFY to the proposed
terms, and the Tier 2 point that TFAP2A sits uneasily at one term predates its
getting both layers (border term + GO:0014036). The STATUS lines and the
synthesis tables use the current terms and actions.

2026-10-01

Tier 1 reviews (in progress)

The TAS GO:0014029 row (source PMID:15242799, an Id2 cardiac-crest paper
whose abstract never mentions Twist) was narrowed to GO:0014036 and its
reference flagged UNVERIFIED. The generic "developmental process" IBA/IEA
rows were first MODIFIED to GO:0014036. On 2026-10-07 they were revised to
KEEP_AS_NON_CORE: the term is correctly inherited at bHLH family level, and
Twist1's specific roles now live in its core functions and in two parts of
the module. Evolution: in Ciona, Twist is
mesoderm-only, and forcing it into a9.49 makes migratory ectomesenchyme
[PMID:23135395], which supports co-option. The early specifier role is
absent in amniotes (Lander), so the conserved vertebrate role is
ectomesenchyme, not specification.

Tier 1 synthesis (2026-10-01)

Gene Layer (review placement) NC process term kept Outgroup / evolution
foxd3-a NC specifier (repressor), with a competence face GO:0014034 fate commitment FoxD repressor activity ancestral; border expression plus N-terminal motif new in vertebrates
sox10 NC specifier GO:0014036 Never in blastula; SoxE recruitment a vertebrate novelty
sox9-a NC specifier, then crest cartilage GO:0014036 Cartilage role ancestral (lamprey Sox9 with Col2a1)
sox8 First-wave NC specifier (frog) GO:0014036 Leading SoxE paralog differs by lineage
snai2 NC specifier, then migration GO:0014036, GO:0001755 Snail border expression ancestral (amphioxus)
twist1 Late, head-only specifier; ectomesenchyme GO:0014036 Ciona Twist mesoderm-only, sufficient for ectomesenchyme: co-option
ets1-a Late cranial-identity specifier, delamination GO:0036032, GO:0001755 (NEW) Absent from lamprey crest, present in skate: a gnathostome addition
id3-a Progenitor maintenance / competence neural crest progenitor maintenance (NTR) (GOA rows MODIFY from GO:0014029) Border expression from lamprey; a change in where it is expressed
myc-a Competence factor from the blastula neural crest progenitor maintenance (NTR) (NEW) Frog c-Myc vs chick N-Myc at the border

Cross-cutting findings:
- Three evolutionary modes appear in Tier 1. (i) Ancestral border
expression (Snail). (ii) Co-option by a change in where a gene is expressed,
with protein activity unchanged (Id, SoxE, FoxD cis-regulation, Twist).
(iii) Protein change (FoxD3 N-terminal motif). Ets1 shows the crest network
kept growing after the vertebrate origin.
- Molecular function. Recurring precise terms: GO:0001227 (FoxD3,
Snai2), GO:0001228 (Sox9, Ets1), and GO:0140416 transcription regulator
inhibitor activity (Id3; Twist on Snail2). Generic protein binding was
replaced or removed everywhere.
- PMID:15242799 (an Id2 cardiac-crest paper) is the TAS source for the
GO:0014029 rows on both twist1 and snai2, and is miscited for both. Report
it to the source curators.
- Homeologs. In every case, experimental rows sit on one homeolog only
(foxd3-b, sox9-b, id3-b, myc-b and snai2.S lack them), while knockdown
reagents typically hit both.
- Flags for curator review: the convention below. (The twist1
"developmental process" rows were resolved on 2026-10-07; see twist1.)

Tier 2 reviews (in progress)

The comparator checks were verified in QuickGO: no Zic in any species
carries GO:0014036, while zebrafish foxi1, gata3 and tfap2a and frog tbx1
carry GO:0060788 by IMP/IGI. Evolution: amphioxus Zic and Pax3/7 already
mark the border without the crest specifiers [PMID:18562679], the same
pattern as pax3-a. Open: PMID:9435279 sources a GO:0014029 IMP row
although its abstract does not mention crest (accepted, deferring to the
curator). Only zic1.S has experimental rows. Zebrafish tfap2a sits at
GO:0014036, which is relevant to the TFAP2A review.

Hearing, retina, iron-response, ROS and retinoblastoma-overexpression rows
were marked over-annotated. Evolution: amphioxus AP-2 is expressed in
non-neural ectoderm only [PMID:12397104]. So, unlike Pax3/7 and Zic, AP-2's
border and crest roles are a vertebrate co-option, probably elaborated by the
TFAP2 paralog expansion. Open: genes that act in both layers, AP-2α the
clearest case, test whether the convention should let them carry
GO:0014036 too.

Tier 2 synthesis (2026-10-05)

Gene Layer NC terms kept Outgroup / evolution
gbx2 Border specifier (posterior; upstream of pax3/msx1) neural plate border formation (NTR) (NEW) Gbx/Otx positioning ancestral (amphioxus); lost in Ciona
MSX1 Border specifier (BMP-responsive; upstream of pax3/zic) neural plate border formation (NTR) (NEW, ISS, flagged) Border expression ancestral (amphioxus)
pax3-a Border specifier; with Zic1, necessary and sufficient for crest neural plate border formation (NTR), GO:0014034 Border expression ancestral (amphioxus)
zic1 Border specifier; also preplacodal neural plate border formation (NTR), GO:0014034 Border expression ancestral (amphioxus)
hes4-a Competence / progenitor maintenance neural crest progenitor maintenance (NTR) (one NOT row UNDECIDED) No outgroup data found
TFAP2A Spans border and crest neural plate border formation (NTR) + GO:0014036 (both NEW, ISS) Amphioxus AP-2 is non-neural only: co-opted

Cross-cutting findings:
- The border layer is ancestral, and the wiring is new. Pax3/7, Zic, Msx
and Gbx already mark the chordate neural plate border (amphioxus). What
vertebrates added is direct activation of the crest specifiers by the
Pax3/Zic1 pair, plus co-option of AP-2 into the border. This matches Tier 1:
most specifiers were co-opted by a change in where they are expressed.
- The convention held up without strain. Every border gene sits naturally
at GO:0014029, with GO:0014034 where it directly induces crest fate
(Pax3, Zic1), and none needed GO:0014036. The one tension is TFAP2A,
which genuinely acts in both layers, and zebrafish tfap2a sits at
GO:0014036.
- Coverage gaps, not conventions. Every border gene's comparator check
(verified in QuickGO) found crest terms on at least one ortholog
(mouse Pax3, mouse Gbx2, mouse/zebrafish Tfap2a) or on same-layer peers
(Msx). Uncurated frog papers account for most of the missing annotations:
gbx2 had no experimental rows at all.
- The case for a "neural plate border formation" term is now made by
six genes. Raise it with GO as a proposed_new_terms / NTR. The
definition should cover the border as a competence territory that gives
rise to crest, placode and dorsal neural tube.

Tier 3 reviews (in progress)

Tier 3 synthesis (2026-10-07)

Gene Layer NC term Module
snai1 Blastula competence, then earliest crest specifier GO:0014036, GO:0001755 Specification part (Zic1 → Snail1 → Snai2)
pou5f1.1 (Oct25) Germ-layer timing (core) + crest competence neural crest progenitor maintenance (NTR) (NEW, flagged) Competence part
lin28a General pluripotency/timing RNA-binding factor (frog) none (chick-only crest evidence) Knowledge gap only
sox2 Blastula pluripotency, neural progenitor none Not a member; SoxB1 → SoxE hand-off gap
sox3-a Maternal germ-layer patterning, genome activation, neural competence none Not a member; SoxB1 → SoxE hand-off gap

Cross-cutting findings:
- "Shared with the blastula" does not mean "participates in the crest."
Of the five pluripotency-programme genes, two keep working inside the
crest-forming cells (Snail1 as a specifier, Oct25 as a competence factor).
One has crest evidence only in chick (Lin28). Two must be switched off
for crest to form (Sox2, Sox3). The retained-potential hypothesis is
really about a subset of the programme, plus a required hand-off.
- Two evolutionary routes into the competence layer. Oct25/pou5 is a
vertebrate-new protein whose crest expression varies by lineage (lamprey
pou5 is absent from crest but still rescues frog crest). Id3 is an old
protein whose crest expression is new. The SoxB1 → SoxE hand-off is
shared with lamprey, so it dates to the vertebrate ancestor.
- Miscitation pattern. PMID:15242799 (an Id2 paper) is the cited source
for TAS crest rows on twist1, snai2 and snai1. Two sox3-a rows cite papers
about other genes. Report both to the source curators.

Tier 4 reviews (in progress)

Tier 4 synthesis (2026-10-08)

Decision (2026-10-09): obsolete GO:0014029

Superseded by NEURAL_CREST_FORMATION_OBSOLETION.
GO:0014029's text describes forming the border region, but its placement in
GO is under epithelial to mesenchymal transition. We therefore propose to
obsolete it, together with its regulation terms, and to add a new term,
neural plate border formation. Replacement rule:
- Border specifiers: the new term. That covers gbx2, MSX1, PAX7, TFAP2A
(border role) and TFAP2C, plus pax3-a and zic1, which also keep GO:0014034.
- Crest specifiers: GO:0014036. That covers the SoxE genes, snai1/2,
twist1, TFAP2B and TFAP2A (crest role).
- Competence factors: a second new term, neural crest progenitor
maintenance (decided 2026-10-09). That covers hes4-a, id3-a, myc-a and
pou5f1.1.
- Overexpression-only evidence (added 2026-10-09): treated as
over-annotation unless the same gene has loss-of-function or other
independent evidence for the role. This affects zic2-a and zic4.

All eleven affected reviews and the module have been updated.

Earlier discussion: GO:0014029 vs GO:0014036

GO:0014029 neural crest formation is defined as forming the region of
ectoderm between the neural plate and non-neural ectoderm. The chain is
GO:0014036 fate specification part_of GO:0014034 fate commitment part_of
GO:0014029. The sox10 review narrowed specifier genes from GO:0014029 to
GO:0014036, leaving GO:0014029 for genes that act on the border region as a
whole. Apply this consistently across Tier 1 (and decide whether border
specifiers in Tier 2 should keep GO:0014029). This needs curator sign-off,
because it modifies IMP rows to a more specific child term. Applied so far:
sox10, sox9-a and sox8 narrowed to GO:0014036; foxd3-a already at GO:0014034
(accepted); id3-a deliberately kept at GO:0014029, as a non-specifier.