At gastrulation, epiblastic cells from the embryonic plate caudal to the head process invaginate through the primitive streak to form mesoderm on each side of the neural plate, while cells from both sides of the dorsal lip of Hensen’s node migrate through the primitive pit to integrate into the midline notochord. The embryonic plate thus elongates by new additions to its rear (caudal) aspect [24].

The anterior-posterior (rostrocaudal) polarity of the embryo is determined very early during gastrulation. The prechordal mesoderm rostral to the otic vesicle (and notochord) forms most of the bones and muscles of the head and face without ever developing somites. Caudal to this tissue is the somitic region which extends along the body axis down to the tip of the tail (Fig. 1). The anterior somitic region from the otic vesicle to the blastopore (future anus) corresponds to the future body axis from the occiput to the anus. Here, the epiblastic cells condense to form the parachordal mesoderm on each side of the notochord following systematic ingression movements through the primitive streak. This initially homogenous column of cells, also called pre-somitic mesoderm (PSM) or segmental plate, subsequently segregates into segmental clusters called somites, which will eventually give rise to the smooth muscle of the dermis, the axial musculature, the vertebral column, and support structures of the peripheral nervous system. After blastopore closure and complete regression of the primitive streak, the gastrulating region for the tail is restricted to a small region called the tail bud, located at the caudal tip of the primitive streak. The tail bud functions as a blastema of undifferentiated cells [20, 21]. The pre-somitic mesoderm and later somites of the caudal and tail regions are therefore not formed by epiblastic ingression but by progenitor condensation in situ.

Prior to the appearance of somites, the PSM remains a loose mesenchyme without specific cellular polarity or stratification pattern alongside the lengthening notochord and neural plate [75]. During somitogenesis, the loose mesenchymal cells of the PSM undergo transformation into tightly-apposed epithelial cells with definite polarity and orientation. A newly formed somite is a compact epithelial sphere (“somitomere”) composed of a single layer of radially arranged cells with apices pointing towards a central lumen (somitocoele), which contains a few mesenchymal cells [45] (Fig. 2). The epithelial cells have polarised Golgi zones, basally aligned nuclei, and actin molecules near the luminal border, the entire sphere being enveloped by a collagen-containing basal lamina [4]. This epithelial conversion of the PSM is aided by an increase in cell-cell adhesion mediated by a sharp but transient rise in the levels of the calcium-dependent adhesion molecule N-cadherin, of fibronectin and possibly cytotactin [88].

Somitogenesis begins soon after internalisation of the prochordal (head) mesoderm and continues through subsequent production of the body axis. The first somite forms immediately caudal to the otic vesicle [38], followed by sequential transformation such that a new pair of somites is regularly added in a rostrocaudal direction until a fixed species-specific number of somites is reached (Fig. 3). Thus, depending on the stage of gastrulation, the growing column of body mesoderm consists of a rostral section of mature somites already undergoing differentiation into sclerotomes and dermomyotomes, a middle section of new pre-differentiated epithelial somites, and a caudal section of pre-somitic mesoderm just rostral to the remaining primitive streak, the whole enterprise necessarily evolving in parallel with the neurulating neural plates (Fig. 4).

Current thinking regarding the mechanism of metameric transformation of the PSM follows what is called the “Clock and Wavefront” model, first proposed by Cooke and Zeeman in 1976. In this model, PSM cells oscillate between a permissive and a nonpermissive state for somite formation. These oscillations are phase-linked and controlled cell-autonomously by a “Segmentation Clock”. Somitic formation is triggered when cells of the rostral PSM, while in the permissive phase of the Clock, are hit by a wavefront of maturation (or determination) that slowly moves caudally along the embryonic axis (Fig. 5). At this exact site, a transverse column of “boundary cells” forms across the width of the PSM, such that at the next periodic activation of the “Segmentation Clock”, a consecutive column of boundary cells, together with the previous boundary, now segregates out a segment of PSM that will ultimately evolve into a somite. Thus, the Clock generates a temporal periodicity that is translated spatially into the periodic boundaries of the somites [75] (Fig. 6).

At the molecular level, the phasic oscillation of the Segmentation Clock is reflected by rhythmic expressions of a class of genes called cycling genes [74]. These genes include the c–hairy1 family in fish, chick, frog, and mouse, which encode transcription factors such as the split (HES) family and the glycosyltransferase Lunatic Fringe [28, 56], all tightly involved in the Notch signalling pathway and its ligands, suggesting that periodic Notch activation plays a critical role in the oscillator (Fig. 7). Also, these genes oscillate largely in synchrony in the PSM, suggesting they are downstream of a common cycling activator.

In chick and mouse embryos, the maturation or determination wave is generated by the expression of genes which include the fibroblast growth factor gene fgf8 [82]. It is known that the caudal domain of the PSM is very high in the factor FGF8, which seems to actively maintain the mesenchymal identity of caudal PSM cells. The FGF8 gradient decreases towards the rostral axis, so that at the rostral domain of the PSM where somitogenesis is taking place, FGF8 level is very low. Also, overexpression of fgf8 in this region results in inhibition of somitogenesis. Thus, the activation of epithelisation of PSM cells is negatively regulated by FGF8, whose absence allows the PSM cells to become competent to respond to the clock signal and initiate somitic boundary formation [26]. The limit between the two domains of high (in the caudal PSM) and low (in the rostral PSM) fgf8 expression therefore marks the determination wavefront of somitogenesis [75] (Fig. 8).

Due to the constant caudal elongation of the body axis during gastrulation and the addition of new PSM cells with strong fgf8 expression to the rear, the FGF8 gradient is continuously displaced caudally. Accordingly, the determination wavefront also moves slowly in a caudal direction along the body axis. This ensures that the old and new metameric boundaries, at both ends of a new somite, are separated by a distance corresponding to the caudal displacement of the determination wavefront during one period of oscillation of the Segmentation Clock [74] (Fig. 8). The speed of new somite production is thus linked to the Clock period, which is species specific: 30 min for zebra fish, 90 min for chick, and 120 min for mouse [70].

Within hours after its formation, each somite starts differentiating along its dorsoventral axis. Cells from its ventromedial part lose their epithelial arrangement and migrate towards the notochord to form, with the luminal cells, the mesenchymal sclerotomes [12, 45]. Together with the notochord, the sclerotomes provide exclusive material for the vertebral column. Cells from the dorsolateral part of the somite retain their epithelial arrangement to produce the dermomyotomes (Fig. 9a, b). The dermomyotomes later subdivides into the dorsal dermatome immediately beneath the ectoderm to give rise to the dermis and its smooth muscles, while the disaggregated cells between the dermatome and the sclerotomes remain closely packed as the myotome, forming ultimately the axial skeletal muscles [10, 12, 24] (Fig. 9c, d).

While specification of the anterior-posterior pattern of the somite appears to be determined very early [1, 39, 46], the dorsoventral values are not intrinsic to the somites. Thus, if the somite were to be surgically rotated dorsoventrally by 180º, sclerotomes still develop in the ventromedial position while the dermomyotomes remain dorsolateral. Moreover, a dorsally implanted notochord represses dermatomal differentiation but encourages sclerotomal formation dorsally, while notochord ablation completely prevents sclerotomal appearance [24] (Fig. 10). These findings suggest that the dorsoventral differentiations of the somite depend on appositional instructions from the notochord, very possibly through sonic hedgehog (Shh) expression [48] (Fig. 11).

The term “resegmentation” was originated by Remak in 1855 (“Neugliederung”) [81] and remains still a subject of controversies [16, 24]. It refers to the fact that the early metameric boundaries between the somites are once again changed and “reshuffled” during the development of the sclerotome, so that the later boundaries between the vertebral bodies do not match up with the original intersomitic clefts [12].

During somitic differentiation, the ventromedial cells destined to become sclerotome subdivide into paired lateral clusters flanking the ventral aspects of the neural tube, and an unpaired median (axial) cluster surrounding the midline notochord (Fig. 12). The formation of the vertebral column takes place first in the lateral sclerotome, where a conspicuous subdivision into a densely packed caudal half and a more loosely cellular cranial half soon appear, separated by a fissure (of von Ebner [27]). The loose cranial half attracts, promotes, and supports the growth and expansion of peripheral nervous tissue from the neural tube and neural crests, but itself never chondrifies into vertebral parts (Fig. 13, middle). In contrast, the caudal densely-packed mesenchyme of the lateral scleroderm soon takes on a triangular shape. The side of the triangle facing the axial perichordal sclerotome, which will become the future vertebral body, gives rise to the pedicle. The side facing dorsolaterally away from the axial sclerotome forms rudiments of the neural arch, and the side of the triangle facing ventrolaterally becomes the costal process (Fig. 9d).

Shortly after the specifications of the lateral sclerotome into its arcual, costal, and pedicular components, the initially loosely-meshed mesenchyme within the perichordal axial sclerotome also begins to show compartmentalisation. A cell-dense zone develops at the same level as the dense caudal half of each lateral sclerotome, partly due to medial expansion of the lateral band of condensed tissues from the lateral sclerotome. The axial sclerotome in between these median dense zones remains loosely cellular. Later, the cranial-most layer of the axial dense zone, in line with von Ebner’s fissure in the lateral sclerotome, becomes even more tightly packed and forms the intervertebral boundary zone (IBZ) (Fig. 13, middle). This intervertebral boundary mesenchyme (IBM) ultimately forms the ringlike annulus fibrosus of the intervertebral disc, permanently enclosing a looser central core of nucleus pulposus made partly of notochordal remnants. The vertebral body itself is mainly made from chondrogenesis in the loose-cell zone of the perichordal sclerotome, now called the prevertebra [11, 12, 24, 67], although contributions from the condensed tissue adjacent to the IBZ have been observed [16, 23]. As a final step, the pedicular anlagen from the lateral sclerotomes fuse with the chondrifying prevertebra of the axial sclerotome, while the neural arches surround the neural tube to complete the vertebral ring (Fig. 13, right). The costal component becomes the future transverse process but only in the thoracic region is it predestined to form ribs.

Thus, the neural arch of the vertebra is derived from the caudal-lateral part of a single somite [16]. However, labelling and transplantation experiments of half-somites have repeatedly demonstrated that each vertebral body is made up of cells from the axial zones of two adjacent somites [2, 3]. Although the exact boundary of individual somite participation is not known [12], the juxtapositioning of axial and lateral sclerotomal components suggests that each vertebra probably comes from the caudal half of one somite and the rostral half of the somite below [67]. This will explain the slightly off-step registration between the levels of the somite and the resegmented sclerotome [12, 17, 67] on the embryonic axis such that the middle of the resegmented sclerotome lines up with the original intersomitic cleft (Fig. 13, left and middle). Also, since the dense portion of the lateral sclerotome is in line with the dense zone of the axial sclerotome adjacent to the IBZ, it makes perfect sense that the mature pedicle is joined to the cranial and not the caudal half of the vertebral body [12]. It also follows that the spinal nerve, ganglion, and blood vessel from the corresponding somitic segment, being associated with the loose cranial half of the lateral sclerotome, must cross above its own neural arch and that the corresponding segment of the spinal cord is always slightly more rostral to its companion vertebral body (Fig. 13, right). Given there are eight cervical somites but only seven resegmented axial sclerotomes and consequently seven cervical vertebrae, the C₁ nerve root emerges above the C₁ neural arch, and the C₈ nerve root comes through below the C₇ neural arch and above the T₁ neural arch, which in fact is derived from the C₈ somite. Finally, resegmentation of the sclerotome explains why the original “pre-resegmented” intersomitic vessel ultimately enters the midpoint of the vertebral body as the segmental nutrient artery.

The CVJ is the product of the occipital somites and the first three cervical somites. There is controversy regarding the proper number of occipital somites in vertebrates. Wilting et al. and others [14, 80, 99] thought there were five occipital somites in the chick and mouse but conceded that the first somite either disappeared early or was an insignificant clan of cells that lacked sclerotomal lineage. Müller and O’Rahilly [67] studied staged human embryos and concluded that, in humans, there are only four occipital somites that participate in formation of the skull base. The transitional zone between the skull and the cervical spine is thus taken to be between the 4th and 5th somites. During the fourth week of gestation, there are consequently 4 occipital, 8 cervical, 12 thoracic, 5 lumbar, 5 sacral, and 8–10 coccygeal somites, 42 pairs all told.

Following primary segmentation, the determination of the positional identity of the prevertebral segments along the embryonic axis, which in turn ordains the regional developmental specifications of the vertebral phenotypes, is controlled by Hox genes. The mammalian Hox genes encode transcription factors used in regulating the establishment of the body plan. They contain the phylogenetically highly conserved homeobox domain [24, 43, 44]. In mouse and humans, there are 39 Hox genes distributed in four linkage clusters, Hox A, B, C and D, on four different chromosomes (chromosomes 6, 11, 15 and 2) (Fig. 19). The members of each cluster, designated by Arabic numerals, are also grouped vertically along the clusters because analogous members of each group are linked by common origin from a single ancestral gene. For example, Hox a– ₄ , b– ₄ , c– ₄ , and d– ₄ of the mouse, called paralogues, are held kin by all being the phylogenetic descendants of the much older Dfd gene of the Drosophila, and a–6, b–6, and c–6 are similarly related to the Antp gene of the Drosophila’s antennapedia complex (Fig. 19). Topographically, the lower numbered paralogues are located on the anterior 3′ axis of the chromosome, and the higher numbered ones are on its posterior 5′ locations (Fig. 19).

Hox genes are expressed in mesodermal and ectodermal cells along the body axis. Each gene has a characteristic and distinct anterior boundary of expression. A temporal and structural colinearity exists between the position of a gene in a cluster and its expression pattern. Thus, genes from the more anterior 3′ locations in the Hox clusters are expressed earlier and always occupy more anterior (cranial) expression domains than genes closer to the posterior 5′ location in the clusters (Fig. 19). For example, Hox a–3 has a more anterior expression domain than Hox a–9 and similarly between Hox d–4 and Hox d–10.

Functionally, only the anterior boundary of the Hox expression is important. Since multiple genes have the same anterior expression boundary along the prevertebral axis, each metameric segment can be identified by its own characteristic combination of Hox gene expression domains, i.e., its own Hox code (Fig. 20). A specific Hox code acts as a master switch in determining the exact positional identity of a mesodermal segment along the body axis, and via regional idiosyncrasies in development, the Hox code thus becomes translated into a specific vertebral anatomy [24, 34, 44].

The importance of the Hox code is illustrated by the severe abnormality in numbers and structures of vertebral segments when the code is altered by Hox gene mutations and teratogenic disturbance of Hox gene expressions. One example at the CVJ is the inactivation in mice of Hox d–3, an important component of the Hox code for the first cervical prevertebra. This results in a more caudal anterior boundary of Hox d–3 expression, shifting the transitional sclerotomal properties from C₁ to C₂ prevertebra and transforming the C₁ prevertebra to a more anterior (i.e., occipital) identity, producing mutant mice with atlas assimilation to the basioccipital [13]. This is called anterior homeotic transformation (Fig. 21). Conversely, extension of an expression domain rostrally can transform prevertebral segments into a more caudal (posterior) identity, a phenomenon known as posterior homeotic transformation. This is exemplified in “gain-of-function” transgenic mutation of the murine Hox d–4 gene, also a component of the Hox code for the C₁ prevertebra, such that its expression domain is forced to extend towards the occipital somites. The exoccipital region of the transgenic mutant shows no occipital condyles but instead ectopic neural arches resembling cervical neural arches, and the basioccipital is fused to the apical dens, “imitating” the behaviour of the atlas centrum [53]. In the human analogue of posterior homeotic transformation, the basion, from the O₄ somite, “imitates” the C₁ prevertebra by being detached from the upper clivus, and the anterior C₁ arch, from the hypochordal bow of the C₁ sclerotome, is fused to the apical dens as if C₁ is trying to become C₂ by acquiring a “true centrum” (Fig. 22). Thus, an altered interpretation of axial position cues in the developing occipital somites leads to an imposition of cervical-vertebral phenotype upon the occipital bones. These various observations suggest that Hox gene abnormality may underlie many malformations in the CVJ [17].

The Pax family of regulatory genes is implicated in sclerotomal resegmentation. Pax genes in vertebrates all contain the highly conserved DNA sequence called “paired-box”, in essence a paired domain with a partial or complete homeodomain. There are nine Pax genes in the mammalian family; all except Pax–1 and Pax–9 are involved in development of the central neuraxis. These two exceptions, especially Pax–1 (PAX–1 in the human homologue) [85, 97], control boundary formation between tissues by keeping two cell populations separate, presumably because the transcription factor encoded by Pax–1 differentially regulates cell surface molecules expressed by these two cell populations, thereby divergently influences their respective fates. This scenario of cellular partitioning is a necessary condition for resegmentation, and Pax–1 action at the future IBZ thus helps to demarcate the site and extent of sclerotomal segregation. The downstream target gene for Pax–1 is unknown but may involve cell adhesion molecules such as NCAM or cytotactin or molecules that promote cell-cell communication such as connexins [41, 42, 68, 85].

Pax–1 expression is detected very early in the pre-differentiated somites before their dorsoventral specification. Signals from the notochord and ventral floor plate of the neural tube, mediated by the SHH protein, induce the somite to divide into dermomyotome and the ventromedial sclerotome. This coincides with intense expression of Pax–1 ventrally within the sclerotomal field, suggesting that the earliest role of Pax–1 is mediating the dorsoventral specification of somites [48, 99].

After somitic differentiation, Pax–1 expression plays a more clearly segregating role within both the lateral and axial sclerotomes, where its timing and fluctuating levels coincide with crucial events of resegmentation; i.e., the opposite functions of partitioning and aggregation of adjacent sclerotomal clusters are matched to high and low Pax–1 expression, respectively. For example, during condensation of the axial sclerotome into the loose and dense halves, Pax–1 expression is weak within the loosely cellular prevertebrae but intense within the dense IBZ [85, 99]. Later, with chondrification of the prevertebra to form the homogeneous vertebral body, Pax–1 is further actively repressed in this location, but persists in high levels at the IBZ where partition of centra takes place, until formation of the intervertebral disc is well underway [99]. Pax–1 level is also enhanced during condensation of the lateral sclerotome to form the neural arch [23, 25, 98].

Conversely, normal fusion of certain adjacent sclerotomes takes place only when Pax–1 expression is turned off. At the CVJ of chick embryos, Pax–1 repression is timed exactly when the occipital sclerotomes fuse to form the basiocciput. The fusion of the two dens primordia with the axis body also coincides with the downregulation of the Pax–1 gene [99]. Ectopic Pax–1 expression disrupts normal assemblage of the dens-axis and basiocciput, producing mutants with mobile dental fragments and loosely beaded clivus [98]. Pax–1 is also highly expressed within the transitional zone between the proatlas and the first cervical sclerotome; it may thus also play a role in the separation of the head from the trunk.

Murine Pax–1 mutants undulated show multiple fusion of vertebral bodies and fusion of the dens with the anterior atlantal arch [98]. Extreme cases of complete fusion of the basiocciput-prevertebral column, literally from the clivus to the thoracolumbar spine, have been seen in human spondylocostal dysostosis (Fig. 23). These and the milder form of Klippel-Feil syndrome have been postulated to reflect severe PAX–1 loss-of-function mutations [85, 86]. It is therefore conceivable that hyper- and hyposegmentation defects in human may be explained by over- and under-expression of PAX–1 during vertebral development.

The making of any part of the vertebral column requires the successful completion of three developmental phases: First, the mesodermal primordium has to be properly formed and, in some cases, assembled during the membranous phase; second, the mesodermal primordium undergoes chondrification in the cartilaginous phase; and finally, in the osseous phase, ossification takes place within the cartilaginous mould to complete the end product. In the case of the dens-axis, there is a fourth phase which involves bony fusion of the upper and lower dental synchondroses (Fig. 16).