Natural Science Forum / Biology / Biology / November 2007

Patterning the vertebrate neuraxis. (Developmental Neurobiology)(Cover
Story).

Andrew Lumsden and Robb Krumlaf.

Abstract:
Neuraxial patterning is a continuous process that extends over a
protracted period of development. During gastrulation a crude
anteroposterior pattern, detectable by molecular markers, is conferred
on the neuroectoderm by signals from the endomesoderm that are largely
inseparable from those of neural induction itself. This coarse-grained
pattern is subsequently reinforced and refined by diverse, locally
acting mechanisms. Segmentation and long-range signaling from
organizing centers are prominent among the emerging principles
governing regional pattern.

Full Text :COPYRIGHT 1996 American Association for the Advancement of
Science. Due to publisher request, Science cannot be reproduced until
360 days after the original publication date.

The central nervous system (CNS) arises from the neural plate, a
cytologically homogeneous sheet of epithelial cells that forms the
dorsal surface of the gastrula-stage embryo. The peripheral nervous
system arises from ectodermal placodes and neural crest cells that
form at the lateral fringes of the plate. The neural plate
subsequently rolls up on its anteroposterior (AP) axis to form a tube,
the expanded anterior end of which then partitions into a series of
vesicles representing the anlagen of fore-, mid-, and hindbrain.
Posteriorly, the long, uniformly narrow tube forms the spinal cord.
These early morphological features of the neuraxis, accompanied by
position-specific expression of developmental control genes, dictate
the overall plan of the CNS and predict its regional specializations.
Within each region, a large diversity of neuronal cell types is then
generated, each with distinct identities in terms of morphology,
axonal trajectory, synaptic specificities, neurotransmitters and so
on. The intricate spatial order of differentiated neurons, essential
to the subsequent formation of functional circuits, is crucially
dependent on correct regional specification.

Signals from adjacent tissues are involved at all stages of neuraxial
patterning. Neural-inducing factors and modifiers produced during
gastrulation by the (endo) mesoderm establish an initial crude AP
pattern in the overlying neural plate. Although the precise nature of
this early patterning information remains unclear, the inductive
signals that confer forebrain identity appear to differ qualitatively
from those that operate more posteriorly. The coarse-grained pattern
that emerges at the end of gastrulation is progressively refined,
resulting in a precise regional variation in cell identity.

Patterning of cell types appears to be organized on a Cartesian grid
of positional information, the coordinates of which correspond with
the AP and dorsoventral (DV) axes of the neural tube: analyses of cell
fate after experimental rotation of the neural plate (1,2) have
indicated that regional fate is determined along the AP axis before
and independently of fate restriction on the DV axis. We will confine
our discussion to the assignment of AP regional identity, as
patterning on the DV axis has been comprehensively reviewed elsewhere
(3,4). We will focus on two well-studied regions, the hindbrain and
the midbrain, to illustrate distinct but not mutually exclusive modes
of local patterning. segmentation plays a prominent role in the
hindbrain, whereas a discrete signaling region at the isthmus sets up
the AP polarity of the midbrain. The forebrain has been less
intensively studied, but interesting parallels with hindbrain
development are considered.

Early Role of the Mesoderm in

Regionalization

Investigation of the earliest developmental events has focused on the
amphibian embryo in which it was first seen that neural fate is
imparted to competent ectoderm by signals emanating from the dorsal
blastopore lip. Spemann noted that the early dorsal lip, grafted
heterotopically, could induce an entire neuraxis but the later dorsal
lip could induce only posterior CNS (5). Although the details of how
the dorsal lip (Spemann's organizer) confers polarity on the neuraxis
are still unknown, emphasis has been placed on the posterior-to-
anterior (P-to-A) progression of mesoderm involution during
gastrulation and a two-step, activation transformation action of its
signals on the dorsal ectoderm. Activating signals from early-
involuting mesoderm are thought to induce a default state of anterior
neural differentiation, which is then modified to a more posterior
character by transforming signals from later-involuting mesoderm (6).
The level of transforming, signal impinging on any one level of the
neuraxis might confer its local identity (Fig. 1). Inducing signals
could pass to the ectoderm either vertically, from underlying cells,
or tangentially, from organizer cells still in the plane of the
ectoderm (7,8).

Whereas the molecular identity of activating signals remains
uncertain, three secreted proteins expressed by the organizer-
follistatin, noggin, and chordin - are capable of inducing the
expression of anterior neural plate markers in naive ectodermal cells
(9). These candidate activators share no obvious structural features,
but each can bind directly to and antagonize the actions of members of
the BMP family of signaling molecules (10,11). Induction of anterior
neural plate may thus involve inhibition of the neural inhibitors BMP2
and BMP4, which are present in the presumptive neurectoderm.

Candidate transforming signals include basic fibroblast growth factor
(bFGF) (12) and retinoic acid (RA) (13), both of which can
posteriorize anterior neural tissue but have little neuralizing
capacity on their own. However, signaling through the bFGF receptor is
not necessary for this process; when blocked in transgenic Xenopus
embryos by the expression of a truncated, dominant-negative form of
FGFR1, the posterior CNS forms normally (14). Strongly indicating a
posteriorizing role for RA, however, the expression of a
constitutively active retinoid receptor results in a posteriorized
axis, whereas dominant-negative retinoid receptor expression results
in an anteriorized axis (15). The posteriorizing role of RA will be
considered again in the context of local patterning in the hindbrain.

In addition to signals that may influence pattern along the entire
neuraxis, there is also evidence for head-specific induction pathways.
The Lim-1 and Otx-2 homeobox genes are expressed in the organizer
region (Hensen's node of amniote embryos), and their inactivation in
mice deletes all head structures, including prosencephalon,
mesencephalon, and anterior hindbrain, whereas the posterior hindbrain
and spinal cord are unaltered (16-18). The requirement for these genes
appears to distinguish early organizer functions from later ones. Both
genes are later expressed in the prechordal plate (16, 19),
emphasizing the importance of this structure in endowing the overlying
forebrain with unique characteristics. A recently identified secreted
protein, cerberus, has potent forebrain-inducing activity (20).
Cerberus is abundantly expressed in the deep-layer cells of the
organizer that constitute the leading edge of the gastrulating
endomesoderm. Both the maintained expression and inducing activity of
cerberus would seem to depend on coexpression of other organizer
factors. It seems likely that in vivo forebrain-inducing activity lies
in the prechordal plate and the endomesoderm immediately anterior to
it, where expression of cerberus and chordin overlap (20). The
different inductive activities of early and later dorsal lips, first
recognized by Spemann, are now being dissected at molecular and
genetic levels; it appears that the acquisition of coarse-grained
pattern along the neuraxis is controlled by mechanisms that differ
between the anterior-most (prechordal) and the more posterior
(epichordal) regions of the neuraxis.

Hindbrain-a Segmented

Region of the Neuraxis

Pronounced axial variation involving a comparatively small repertoire
of cell types makes the hindbrain an attractive and accessible system
for the study of local CNS pattern. Furthermore, early development of
the hindbrain is characterized by metamerism, suggesting the early
allocation of defined sets of precursor cells and the existence of
precise boundaries to both cellular assemblies and realms of gene
action. In the chick embryo, the segmented pattern of the hindbrain
emerges upon neural tube closure as a series of bulges-rhombomeres-and
is virtually complete at the onset of neurogenesis. Segmentation of
the vertebrate hindbrain bears a superficial resemblance to
segmentation of the Drosophila embryo:rhombomeres form by internal
subdivision rather than by budding from a growth zone, and they have a
pair-wise organization (21).

Neuronal pattern. Two patterns of metameric cellular organization can
be distinguished in the hindbrain, one involving neurons of the
reticular formation and the other involving motor neurons (Fig. 2).
Eight identified types of reticular neuron are repeated through
sequential rhombomeres such that each contains a more or less complete
set (22) Motor neurons also develop in each rhombomere, but have
rhombomere-specific identities (23,24). The segmental disposition of
branchiomotor nuclei in the early hindbrain has a close anatomical
(23) and functional to correspondence with target structures
associated with the segmented series of branchial arches that lie
beneath it. Later in development, the segmental origins of these cells
become obscured as certain reticular cells become more numerous in
particular rhombomeres (26) and the motor nuclei condense and migrate
to new positions. Fate-mapping studies have also revealed metameric
origins for the adult sensory nuclei (27). In the hindbrain,
segmentation is involved in specifying the pattern of developing
structure, but not in deploying them in the adult.

Compartment-like properties of rhombomeres. Developmental compartments
provide a way of allocating blocks of cells with distinct properties
(28). The containment of polyclonal assemblages of neuroepithelial
cells within rhombomeres has been shown by lineage-tracing studies in
chick (29). Compartmental restriction of cell mingling persists while
the epithelium is predominantly germinative (30); later, young neurons
may escape the restriction once they have acquired their ultimate
positional specification. Rhombomeric domains of the germinative
(ventricular) zone remain lineage-restricted up to late stage, when
neurogenesis is nearing completion (31).

Rhombomeres partition from one another according to an adhesion
differential that displays a two-segment repeat (Fig. 3) Consistent
with an expected tendency of neighboring cell groups to separate,
enlarged intercellular space is the earliest specialization of
rhombomere boundaries (33). Complementing this adhesive differential
is an alternating periodicity to the expression domains of the Eph-
like receptor tyrosine kinases and their ligands (Fig. 4) (34-36).
Perturbing Rtk-1 (Sek-1) function in zebrafish and Xenopus embryos by
expression of a dominant-negative form of the mouse Sek-1 receptor
results in failure to establish sharp inter-rhombomere boundaries
(37). These ligand-receptor partners may thus mediate repulsive
interactions that serve to sharpen rhombomere borders. They may also
provide a potential mechanism whereby cells in adjacent rhombomeres
interact with each other to establish additional cell states at the
inter-rhombomere boundaries (21,33,38-41). Finally, the inter-
rhombomere boundaries become colonized by axons, perhaps on account of
both the local expression of growth-promoting molecules (23) and the
availability of extracellular space (33).

Hox genes encode positional value along the AP axis. Prime candidates
for conferring rhombomere identity are the clustered homeobox-
containing genes of the Hox family (42), homologs of the HOM-C genes
that encode parasegment identity in Drosophila. Expression of genes at
the 3' ends of the Hox clusters precedes rhombomere formation and
becomes progressively restricted (43) such that expression boundaries
coincide with the interfaces between rhombomeres (44). Their
expression patterns form an ordered and nested set of domains along
the neuraxis, with a two-rhombomere periodicity. Superimposed on this
pattern are rhombomere-specific variations in expression levels (Fig.
4).

Considering the distribution of transcripts and the general synergy
among Hox genes detected in mouse null mutants, it is possible that
the identity of individual rhombomeres could be defined by the co-
operative action of Hox proteins (42). They may also have singular
effects: ectopic expression of Hoxa-1, for example, results in the
transformation of r2 to an r4 identity (45,46). However, loss of
Hoxa-1 function results in the deletion of r5, reduction of r4, and
loss of specific neuronal nuclei (42), abnormalities that are not
obviously consistent with conferring specific identity on an existing
repetitive ground plan; but it remains possible that Hox genes could
have dual roles, both in segmentation and segment identification.

Positional values appear to be conferred on rhombomeres by Hox
expression, but it is unclear how the Hox genes become activated at
appropriate levels of the neuraxis. Candidates for this role include
kreisler, a b-Zip member of the c-maf proto-oncogene family (47)
expressed in r5 and r6, and Krox-20, a zinc finger gene that is
expressed in two stripes in the neural plate that become r3 and r5
(48). In [kr.sup.-/-] mouse embryos, the neural tube posterior to the
r3/r4 boundary appears unsegmented, a defect that is attributable to
the loss of r5 and r6 as identifiable territories (49,50). Targeted
disruption of Krox-20 results in the elimination nation of r3 and r5
and the formation of a partially fused r2/r4/r6 territory (51). This
phenotype suggests that Krox-20 may be responsible for generating
single-compartment periodicity from cues established by upstream
genes. Absence of the r5 stripe of Krox-20 expression and the more
anterior expression of group 4 Hox genes in [kr.sup.-/-] mice is
consistent with their regulation by kreisler, although direct
interaction has yet to be shown. In contrast, Krox-20 is a direct
modulator of the r3/r5 activity of both Hoxa-2 (52) and Hoxb-2 (53)

A major gap in our understanding of hindbrain segmentation is the lack
of candidate segmentation genes. Despite the conserved role of Hox/HOM
genes in specifying segmental identity, the upstream pathway appears
not to be conserved. However, segmentation is a generic property of
metazoan organization that has evolved many times (54), making it
likely that Hox/ HOM genes have been coupled independently to
segmentation.

Retinoid signaling and AP position. In addition to the putative role
of kreisler and Krox-20 in locally regulating Hox expression, RA has
strong candidacy as an overall mediator of nested Hox expression,
consistent with its posteriorizing effect on CNS regionalization.
Excess RA causes both an anterior shift of Hox gene expression and an
A-to-P transformation of regional fate (55) that includes the ordered
transformation of anterior rhombomeres to a more posterior type (56,
57). Conversely, RA-deficient quail embryos have a small hindbrain,
lacking posterior rhombomeres (58). In addition, the suppression of RA
signaling by expression of a dominant-negative retinoid receptor also
results in anteriorization (15). Furthermore, Hox genes have the
molecular machinery for responding directly to retinoid signaling
(55).

Consistent with a direct role for the organizer, Hensen's node is a
rich source of RA and produces increasing amounts during regression
(59); thus, the nested expression of Hox genes could be controlled
either by a P-to-A gradient of RA diffusing directly from the node, or
by an increasing exposure to RA of cells that pass through the node,
in A-to-P succession (60). However, it has yet to be shown that RA
normally forms a graded signal of either kind or, indeed, that a
gradient is necessary; the activity of retinoids could be locally
modified. by coactivators and corepressors of retinoid signaling
(61).

Rhombomere autonomy and plasticity. Transplantation experiments in
avian embryos reveal a direct correlation between commitment to
rhombomere-specific fate and Hox expression: Grafts of neural plate-
stage tissue acquire the complement of Hox transcripts and
neuroanatomical features of their new location (62), whereas grafts of
emerging rhombomeres maintain both their identity and specific Hox
expression (2,63). By contrast, transplantation of rhombomeres into
the post-otic region results in the activation of posterior Hox gene
expression (64), suggesting that their commitment is not irreversible.
However, they cannot not easily be shifted from an even- to an odd-
numbered fate, suggesting that commitment to "odd" or "even" may be an
early step in segmentation.

In addition, the even-numbered rhombomeres appear to influence the
fate of odd-numbered rhombomeres, thus providing a secondary mechanism
for establishing positional differences. Inter-rhombomere interactions
control cell survival in the neural crest of r3 and r5, the
maintenance of Krox-20 expression in r3, and the repression of
follistatin in r3 (65)

Midbrain - the Role of the

Isthmic Signaling Region

In the midbrain, beyond the anterior limit of Hox gene expression,
local AP pattern is generated within an unsegmented field through the
activity of a long-range signaling region, the isthmic constriction at
the junction of mesencephalic and rhombencephalic vesicles (Fig. 5)

Establishment of midbrain polarity by Engrailed Signals from the
isthmus regulate expression of two Engrailed genes (66) in a gradient
that decreases both anteriorly, through the mesencephalic vesicle, and
posteriorly, through r1 (Fig. 5). Knockout experiments have shown that
En-1 has a critical role in the early specification of the entire
region of its expression, whereas En-2 function is restricted to
cerebellar morphogenesis (66). However, the En-1 mutant phenotype,
agenesis of the tectum (dorsal midbrain) and cerebellum (anterior
hindbrain), is completely rescued by insertion of the En-2
complementary DNA into the En-1 locus (66,67), demonstrating that the
contrasting phenotypes of En-1 and En-2 mutations reflect differences
in the temporal and spatial expression of the respective proteins and
not a divergence in their biochemical activity.

En expression is the earliest known marker for mesencephalic polarity,
later manifested in a pronounced variation in cytoarchitecture and the
acquisition of different sets of afferent inputs from the retina: the
posterior tectum receives axons from the nasal retina, whereas the
anterior tectum becomes innervated by temporal retina. The molecular
basis of this discrimination may involve ligands for Eph-related
receptor tyrosine kinases, RAGS (68) and ELF-1 (69), that are
expressed in decreasing P-to-A gradients - reflecting the earlier
pattern of En - and that may function as growth inhibitors of Mek-4
receptor-bearing temporal axons.

Transplantation studies in avian embryos have shown that En expression
correlates with later morphology. Thus, when the mesencephalic vesicle
is reversed on the AP axis at E2, the En gradient readjusts to its
original polarity, and both the graded architecture and pattern of
retinotectal projections develop normally (66). When reversed at E3,
however, the En gradient does not adjust, and both cytoarchitecture
and retinotectal projection are subsequently inverted. This
association has been strengthened by experiments in which En is
misexpressed in the anterior tectum through use of a retroviral
vector. nasal axons arborize ectopically in the anterior tectum,
whereas temporal retinal axons fail even to enter the midbrain (70).
Furthermore, the altered retinotectal specificity after En
misexpression in the anterior midbrain is associated with ectopic up-
regulation of RAGS and ELF-1, defeating their normal P-t-A expression
gradient and effectively converting temporal axon - specific anterior
tectum into nasal axon - specific posterior tectum (71). Expression of
these effector genes, downstream of En, suggests that the normal
graded expression of En may polarize the dorsal mesencephalon.

Regulation of Engrailed expression. Graded mesencephalic expression of
En appears to be regulated by signaling from the posterior border of
the mesencephalic field. When grafted to the caudal forebrain, the
posterior border (isthmus) induces En expression and the formation of
a complete optic tectum from the surrounding tissue (72). Two secreted
signal molecules, Wnt-1 and FGF8, have been implicated in the isthmic
control of En expression. Wnt-1, a homolog of the segment polarity
gene wingless (a regulator of Engrailed in Drosophila), is expressed
in the midbrain region of the neural plate and later in a ring of
cells that lies just anterior to the isthmus. As for their cognates in
flies, Wnt-1 and En expression appears to be mutually interdependent:
in [Wnt-1.sup-/-] mice, En is expressed normally at first but is then
progressively lost (73) along with the dorsal midbrain. Thus, although
Wnt-1 is critically involved in the maintenance of En expression, it
is not a candidate for inducing En expression or for directly setting
up midbrain polarity. However, another secreted factor, FGF8,
expressed in a circumferential ring immediately posterior to that of
Wnt-1, has midbrain-inducing and -polarizing abilities (74). When a
bead coated with recombinant FGF8 is implanted in the posterior
diencephalon of chick embryos, expression of Fgf8, Wnt-1, and En.2 is
induced in the surrounding cells. These cells later display the
character of a complete ectopic midbrain, whose AP polarity is
reversed with respect to that of the "host" midbrain. Thus,
neuroectodermal Fgf8 expression may be sufficient to establish both
midbrain pattern and polarity. Fgf8 is expressed earlier in axial
mesoderm cells that lie beneath the presumptive isthmic region of the
neural plate (75,76) and that have the capacity to induce En
expression (77, 78); mesodermal FGF8 is thus implicated as a
homeogenetic inducer in this local control of neural pattern.

Other likely targets of FGF8 are the paired box genes, Pax-2, Pax-5,
and Pax-8, which may be required, singly or together, for
specification of the isthmus. In [Pax-5.sup.-/-], mice (79) and
zebrafish treated with function-blocking antibodies to pax(zf[b]), a
presumed homolog of Pax-5 (80), the isthmus is deleted. In the
zebrafish experiments, the expression of both Wnt-1 and En-2 was also
repressed, suggesting their direct positive regulation by pax(zf[b])
Indeed, consensus Pax-binding sites have been identified within an
enhancer region of the En-2 gene: when these sites are mutated, the
midbrain/hindbrain domain of reporter expression is lost (81).

Whereas isthmic grafts induce tectal development in the caudal
diencephalon, the same grafts to the dorsal hindbrain induce
cerebellar development (82), demonstrating that the competence of
rhombencephalic tissue to respond to isthmic signals differs from that
of mesencephalic and caudal diencephalic regions. FGF8 alone appears
to be insufficient for inducing ectopic En-2 expression or cerebellar
development in the hindbrain (74), implicating additional signaling
molecules at the isthmus.

Although signals from a the isthmus are involved in patterning both
the dorsal mesencephalon and the, dorsal anterior rhombencephalon, the
constriction does not correspond precisely with the midbrain/hindbrain
junction. Separating structurally and functionally distinct tectal and
isthmocerebellar regions of the brain, this junction forms some
distance anterior to the constriction and registers with the posterior
limit of Otx.2 expression in the early mesencephalic vesicle (83). The
posteriormost, Otx-2-negative region of the vesicle is fated to join
rl and r2 in the formation of the cerebellum (84). Thus, it cannot be
assumed that obvious morphological features of the neural tube, such
as the constrictions between vesicles, necessarily correspond in a
predictable way to future subdivisions of the brain.

Forebrain - Is Segmentation

Involved?

In contrast to hindbrain and midbrain patterning, where restricted
patterns of gene expression have been tightly linked either to
segmentation or to the activity of a signaling region, our
understanding of early forebrain patterning is virtually limited to
the gene expression patterns. Most notable among these (85) are the
Emx, Dlx, and Nkx homeobox genes, the paired box gene Pax-6, the
winged helix genes BF-1 and BF-2, the Brachyury homolog Tbr-1, and the
secreted factor-encoding gene Wnt-3. Some of these genes are expressed
in the ventricular zone, suggesting a role in regional specification,
whereas the expression of others (Dlx and Tbr-1) is restricted to the
mantle zone, suggesting a role in the control of differentiation. In
the former category, Emx and Otx genes are expressed in the forebrain
and midbrain in a nested array reminiscent of that of the Hox gene
more posteriorly, although with reversed AP symmetry (86).

Largely on the basis of descriptive molecular studies, it has been
proposed that the forebrain is built piecemeal, like the hindbrain,
from a series of metameric units or prosomeres (87). Experimental
evidence for compartmentation is limited to the diencephalon where
cell lineage restriction boundaries, aligned with prominent axon
tracts, define four neuromeres (88). However, the significance of
diencephalic neuromeres is brought into question by an analysis of
retrovirally marked clones (89), which has shown that sibling cells
can occupy multiple nuclei throughout the AP extent of the
diencephalon. Further anteriorly in the telencephalon, the patchwork
expression of putative developmental control genes displays no
evidence of repetition, the essence of metamerism. Nor does the early
cellular organization of the telencephalon support the notion of a
segmental origin; rather, this region appears to be subdivided
longitudinally into two subregions, the anlagen of cortex (pallium)
and striatum (90). These subregions express different regulatory genes
(Emx-1/2, Pax-6,, and Tbr-1 dorsally; Dlx-1/2 ventrally) and appear,to
be segregated by differential adhesion (91). Within the dorsal
(cortical) subregion, Cells migrate extensively in the AP direction,
so that clones cross functional boundaries and sibling cells
contribute to widely separated structures (92). Supporting the view
that the telencephalon is a single field, which becomes subdivided
longitudinally, BF-1 is expressed in the prospective telencephalic
domain before the telencephalic/diencephalic boundary appears. In [BF.
1.sup.-/-] mice, the cerebral hemispheres are severely diminished and
ventral telencephalic markers are not expressed (93.)

Cell marking and transplantation experiments are required to test the
postulate that segmentation is involved in forebrain regionalization.
Alternatively, or additionally, forebrain pattern could depend on an
as yet undiscovered signaling region. Whether or not a segmented
transverse organization exists, a major constraint on understanding
forebrain pattern has been our uncertainty regarding its topological
coordinally particularly with respect to the trajectory of the
longitudinal axis. Here, however, detailed descriptions of DV-
restricted gene expression patterns (94) have more precisely defined
the topology of the domains whose mechanism of formation we are
seeking. The rapid accumulation of molecular data has provoked excited
speculation: with the combined application of experimental
embryological and genetic methods, we can expect this excitement soon
to be relieved by enlightenment.

Spinal Cord - Late to Role of the

Mesoderm in Regionalization

Although superficially uniform, there are subtle variations in
cellular composition along the AP axis of the spinal cord. Motor
neurons are arranged in discontinuous longitudinal columns that occupy
different DV and mediolateral positions at different, plurisegmental
levels of the neuraxis. Thus, the neurons that form the lateral motor
columns at limb (brachial and lumbar) levels are distinct from those
that form at cervical and thoracic levels, not only in the identity of
their peripheral targets but also in the expression of different
combinations of LIM-homeobox genes that are thought to confer
targeting specificity (3,95). Recent studies of the spinal cord have
focused on the control of its pronounced DV pattern (3), whereas
classical studies of AP regionalization and the influence of paraxial
mesoderm (96) still need to be put into a molecular context. However,
genes that lie 5' in the Hox clusters have sharp boundaries of
expression along the spinal neural tube, suggesting, by analogy with
the hindbrain, that they might underlie this regional diversity.
Transposition of prospective brachial and thoracic regions leads to
the respecification of Hox and Lim gene expression, and motor neuron
subtypes develop according to their new positions (3). The most likely
source of signals that effect the acquisition of this regional
identity is the paraxial mesoderm (64). Mesodermal control has also
been implicated in the specification of primary motor neurons of the
zebrafish: transplantation of single cells to new AP positions with
respect to the adjacent somite results in respecification of both the
Lim gene code of the motor neuron and its subsequent axon trajectory
and target specificity (3).

Many aspects of cell pattern are conserved between the hindbrain and
the spinal cord, particularly with regard to the DV axis, where
ventral (Sonic Hedgehog) and dorsal (BMP) signaling systems appear to
be identical in the two regions (3). It is also apparent that these DV
signals act on cells that have already acquired a stable and heritable
indication of their position on the AP axis, and thereby an AP
position-specific competence to respond (2). The time at which AP fate
becomes restricted differs, however, between spinal cord and
hindbrain. Whereas the regional AP identity of the spinal neural tube
appears to be uncommitted for some time after closure (3,97),
hindbrain pattern is fixed and independent of position relative to the
cranial mesoderm from as soon as the rhombomeres become defined (2).
The difference may stem from a phylogenetically ancient distinction
between head and body with respect to patterning strategy. In the
head, the paraxial mesoderm is patently unsegmented and may contribute
little to patterning (98), whereas the neural crest predominates,
furnishing the ectomesenchymal cells that construct the segmented
branchial skeleton and pattern the cranial nerves and muscles. Thus,
in the hindbrain/branchial region, segmentally restricted positional
information originates in the neural tube and is imposed on the
surrounding mesoderm. A relatively rigid set of positional values
within the hindbrain region may have evolved both for correct
deployment of its emigrant neural crest cells and in compensation for
the lack of patterning information in the mesoderm. In the body, by
contrast, the mesoderm imposes its AP positional information on the
neural tube (99). This is seen both for cell pattern within the tube
(3, 100) and for the pattern of motor roots and dorsal root ganglia,
whose overtly segmented disposition is controlled, apparently
exclusively, by the AP polarity of the somitic sclerotome (101).

Conclusions

Considerable advances have recently been made toward understanding the
mechanisms involved in neuraxial regionalization, particularly with
respect to the earliest events, during gastrulation, when the
molecular identity of activating and transforming signals is being
revealed. Especially promising is the evidence, from dominant gain-
and loss-of-function experiments with retinoid receptors (15), that RA
acts as a concentration-dependent posteriorizing signal in vivo and is
required for the correct spatial restriction of anterior markers. The
action of RA in regionalizing the entire posterior CNS, as studied in
early Xenopus embryos, is mediated, at least in part, by its direct
action on the spatial regulation of Hox genes, best known from the
amniote hindbrain. A synthesis of data from these diverse experimental
systems is needed to advance our understanding of this crucial
molecule, as are experiments directed at elucidating its regulation
and mode of action, whether as a gradient from a single source, the
organizer, or as discrete local signals from axial or paraxial
mesoderm (or both).

On a wider level, expression studies and functional analyses of
developmental control genes in different vertebrate systems have
revealed the existence of a neuraxial ground pattern that is highly
conserved. It will be important to discover the genetic and cellular
mechanisms involved in the subsequent elaboration of this ground
pattern that produce the very different brains of fish and mammals.
Further understanding of CNS regionalization will depend on
discovering how region-specifying genes confer a particular potential,
or set of potentials, with respect to the ultimate selection of
regionally appropriate cell identity.