The genetics and embryology of zebrafish metamerism


  • Scott A. Holley

    Corresponding author
    1. Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, Connecticut
    • Department of Molecular, Cellular and Developmental Biology, Yale University, New Haven, CT 06520
    Search for more papers by this author


Somites are the most obvious metameric structures in the vertebrate embryo. They are mesodermal segments that form in bilateral pairs flanking the notochord and are created sequentially in an anterior to posterior sequence concomitant with the posterior growth of the trunk and tail. Zebrafish somitogenesis is regulated by a clock that causes cells in the presomitic mesoderm (PSM) to undergo cyclical activation and repression of several notch pathway genes. Coordinated oscillation among neighboring cells manifests as stripes of gene expression that pass through the cells of the PSM in a posterior to anterior direction. As axial growth continually adds new cells to the posterior tail bud, cells of the PSM become relatively less posterior. This gradual assumption of a more anterior position occurs over developmental time and constitutes part of a maturation process that governs morphological segmentation in conjunction with the clock. Segment morphogenesis involves a mesenchymal to epithelial transition as prospective border cells at the anterior end of the mesenchymal PSM adopt a polarized, columnar morphology and surround a mesenchymal core of cells. The segmental pattern influences the development of the somite derivatives such as the myotome, and the myotome reciprocates to affect the formation of segment boundaries. While somites appear to be serially homologous, there may be variation in the segmentation mechanism along the body axis. Moreover, whereas the genetic architecture of the zebrafish, mouse, and chick segmentation clocks shares many common elements, there is evidence that the gene networks have undergone independent modification during evolution. Developmental Dynamics 236:1422–1449, 2007. © 2007 Wiley-Liss, Inc.


The dorsal neural tube, notochord, pharyngeal pouches, and somites are the defining features of the vertebrate phylotype. The metameric pattern of the somites is inherited by the ribs and vertebral column and influences the development of other somite derivates such and the skeletal muscle and dermis. Somite number varies significantly among the vertebrates, as adult frogs have 6–9 presacral vertebrae, certain amphibians form 15 or fewer while some snakes have several hundred. However, somite number within a species is relatively constant (Richardson et al.,1998). Zebrafish make 30–32 somites at a rate of roughly one bilateral pair every 30 min at 28°C (Hanneman and Westerfield,1989). Broadly speaking, segmentation processes in higher animals can proceed by either of two modes: simultaneous or sequential. As the name suggests, simultaneous segmentation generates all segments at once. Thus, this process only subdivides fully formed fields of cells. Simultaneous segmentation is exemplified by the formation of rhombomeres in the vertebrate hindbrain and segmentation of long germ band insects such as Drosophila. Sequential segmentation is conceptually quite different in that it involves the progressive subdivision of a field of cells that is still growing at one end. Sequential segmentation occurs concomitant with the posterior growth of the embryo in short germ band insects, including the flower beetle tribolium, and during vertebrate somitogenesis. Strikingly, the vertebrate embryo has the capacity to regulate somitogenesis to ensure that the appropriate number of somites is generated even when the number of somite progenitors varies significantly. For example, amphibian blastula, in which up to one third of the blastomeres have been removed, can still give rise to a complete, yet diminutive, embryo. These smaller embryos form the same number of somites, at the same rate, as unmanipulated sibling embryos, but each segment contains fewer cells. Thus, there is not a physical constraint within the segmentation mechanism stipulating that a segment must contain a specific number of cells (Cooke,1975). These observations led to the “clock and wavefront” model of somitogenesis (Cooke and Zeeman,1975; Cooke,1981,1998). In this model, the clock causes oscillations in gene transcription within the cells of the presomitic mesoderm (PSM), and cells are competent to form a somite only at a particular phase of the clock. The wavefront represents the anterior to posterior progression of development of the embryo. Thus, the wavefront sweeps along the body axis once and is correlated with extension of the trunk and tail. The wavefront is often referred to as the maturation front in that it governs the maturation of the PSM. In the clock and wavefront model, the wavefront drives somite morphogenesis, but its activity is gated by the clock: a somite border forms only when the wavefront reaches a group of cells in the appropriate phase of the clock. Thus, in this model, the size of each somite and the rate of somite formation are determined by the speed of the wavefront and the frequency of the oscillator (Supplementary Movie S1, which can be viewed at

Other models have been proposed to explain somitogenesis by invoking Turing equations (Meinhardt,1982,1986), a variation of the progress zone model of limb development (Kerszberg and Wolpert,2000), a “clock and induction” model (Schnell and Maini,2000), and models linking somite periodicity to the cell cycle (Primmett et al.,1989; Collier et al.,2000; Stern and Vasiliauskas,2000; McInerney et al.,2004). More recently, models specific to the zebrafish clock and incorporating increasingly detailed experimental data have been published (Lewis,2003; Horikawa et al.,2006; Cinquin,2007). While it is not yet clear if any of these models precisely describe what happens in vivo, the interaction between theory and experimentation in the field of segmentation has been unusually intense.

Here, a review of our understanding of zebrafish somitogenesis is presented, starting with specification of the somite progenitors in the blastula, creation of the segmental pattern by means of the somite clock, stabilization of the segmental pattern by the wavefront, consummation of segment polarity, and execution of somite morphogenesis. The relationships between segmentation and the differentiation of the myotome and sclerotome are reviewed. Subsequently, differences in somitogenesis along the anterior–posterior (A-P) axis are discussed and then mechanisms that ensure bilateral symmetry in vertebrate segmentation are reviewed. Finally, zebrafish somitogenesis is compared with segmentation of other chordate model organisms.


Progenitors of the somites arise from the ventral and lateral margin of the zebrafish blastula. While fate mapping studies show that the precursors to the trunk and tail somites are intermingled, genetic studies demonstrate that the anlagen of the anterior trunk (somites 1–9), posterior trunk (somites 10–15), and tail (somites (16–30) are specified before gastrulation (Fig. 1A; Kimmel et al.,1990; Warga and Nusslein-Volhard,1999; Szeto and Kimelman,2006). These anlagen are set-aside early in development, and the cells remember when they are allowed to enter the segmentation program several hours later. Direct reception of nodal is required for a cell to be fated as anterior trunk while bmp signaling specifies the tail somites (Szeto and Kimelman,2006). nodal has an indirect role in specifying posterior trunk fates by promoting fgf expression in the margin (Mathieu et al.,2004; Szeto and Kimelman,2006). In turn, fgf appears to promote posterior trunk by repressing bmp expression (Furthauer et al.,2004; Szeto and Kimelman,2006). Later in development, fgf signaling is required to maintain the fates of most of the somite anlagen as fgf8;fgf24 double morphants lack all but the anterior 2–3 somites (Draper et al.,2003). wnt signaling is also needed to maintain posterior trunk and tail somite fates as embryos lacking both wnt3a and wnt8 form only 10–12 somites (Thorpe et al.,2005). These loss-of-function studies are complemented by experiments that have shown that ectopic tails can be induced by injection of bmp4, nodal, and wnt8 mRNA into blastomeres at the animal pole, which would normally give rise to ectoderm (Agathon et al.,2003). Patterning by nodal, fgf, and bmp signaling is mediated largely through the t-box genes: spadetail, no tail, and tbx6. spadetail specifies the anterior and posterior trunk somites while no tail specifies the tail somites (Ho and Kane,1990; Halpern et al.,1993; Amacher and Kimmel,1998; Griffin et al.,1998; Griffin and Kimelman,2002; Goering et al.,2003). tbx6 may function semiredundantly with both spadetail and no tail (Griffin et al.,1998; Goering et al.,2003). Later still, fused somites/tbx24 interacts with the clock and is required for somite maturation while tbx15 and tbx18 are expressed in the somites after segmentation (Holley et al.,2000; Begemann et al.,2002; Nikaido et al.,2002). Thus, nodal, fgf, wnt, and t-box genes act in succession, with some reiteration, to link early patterning and maintenance of the mesoderm to the subsequent segmentation program (reviewed in (Holley,2006a).

Figure 1.

An overview of zebrafish somitogenesis. A: The progenitors for the tail somites (yellow) and trunk somites (teal) are specified in the late blastula 4 hours postfertilization (hpf), dorsal is right and ventral left (adapted with permission from the Company of Biologists (Warga and Nusslein-Volhard,1999). During gastrulation (8 hpf), the anlagen for the anterior trunk somites (green) and some of the posterior trunk somite progenitors (orange) converge toward the dorsal midline, while some of the posterior trunk somite progenitors and the tail progenitors remain ventral. The arrows indicate the direction of epiboly toward the vegetal pole. Somites form sequentially with the anterior somites (green) forming before the posterior trunk somites (orange) and the tail somites (yellow). In the 17 hpf embryo, the unsegmented yellow region in the tail bud is the presomitic mesoderm. The anlagen of the posterior somites form a progenitor zone (red) in the tail bud posterior to the chordal–neural hinge. Somitogenesis is complete at 25 hpf. All views are lateral. B,C: Dorsal views of the tail bud. The progenitors of the posterior somites will exit the progenitor zone laterally (small arrows) to enter the initiation zone as the tail extends posteriorly (large arrow). Cells attenuate their movement and are relatively sessile in the PSM. The PSM is bilaterally symmetric and flanks the notochord and neural tube (not shown). D: A time series representing a single somite cycle showing the posterior displacement of the progenitor zone (red) as the tail extends. Oscillating gene expression (blue) sweeps from posterior to anterior. These oscillations cease in the anterior PSM, and expression of the gene disappears as morphological somite formation occurs. Note that the most-recently formed somite is called the SI while the region in the anterior PSM that will give rise to the next somite is the S0. As the S0 becomes the SI, the old SI becomes the SII (Pourquié and Tam,2001). E: Morphological somite formation occurs as mesenchymal PSM cells (light purple) undergo a mesenchymal to epithelial transition along the somite boundary (yellow). The somite consists of epithelial border cells (purple) and an internal core of mesenchyme (light blue). The adaxial cells (aquamarine) are a morphologically distinct row of medial cells that elongate along the anterior–posterior axis of each segment after somite formation and later give rise to slow muscle fibers.

Fate mapping experiments indicate that in the late blastula, the precursors of trunk somites are located in the dorsal–lateral margin, while the tail somites arise from the ventral margin (Fig. 1A; Kimmel et al.,1990; Warga and Nusslein-Volhard,1999). During gastrulation, the more dorsal and lateral cells converge toward the dorsal midline, while the ventral cells make up a “no convergence zone” that becomes the posterior tail bud as gastrulation is completed (Myers et al.,2002). Fate mapping studies suggest that the anterior 12 somites are derived from the somite anlagen that underwent some dorsal convergence and were already in the segmental plate at the end of gastrulation (Müller et al.,1996; Kanki and Ho,1997; Jülich et al.,2005a). Some of the posterior trunk somite anlagen and all of the tail somite precursors pass through the posterior tail bud or progenitor zone (Fig. 1A–C). During the segmentation period, some somite precursor cells form a dorsal–medial domain above the notochord, continue to migrate posteriorly, and will dive ventrally into the progenitor zone at the posterior tip of the tail (Kanki and Ho,1997). Cells moving into the progenitor zone remain there for variable amounts of time. The progenitors of the posterior trunk somites 13–15 and the tail somites appear to be intermingled in the progenitor zone. Indeed, there is extensive cell mixing in both the progenitor zone and the initiation zone. It is not known how these distinct progenitor populations maintain their positional identities as the two anlagens within the tail bud cannot be distinguished by patterns of gene expression (Kanki and Ho,1997; Mara et al.,2007).


The tail bud can be divided into four regions: progenitor zone, initiation zone, posterior PSM, and anterior PSM (Fig. 1C). The progenitor zone is immediately posterior to the chordal–neural hinge, while the initiation zone is lateral to the progenitor zone and posterior to the tip of the notochord. The progenitor zone contains the precursors for the more posterior somites, and cells in the progenitor zone appear to transcribe the oscillating genes her1, her7, and deltaC in a steady, nonoscillating manner. Cells move laterally from the progenitor zone into the initiation zone where the her1, her7, and deltaC oscillations seem to begin. In 10 somite stage embryos, the posterior PSM extends 25 cells anterior to the posterior tip of the notochord. This anterior boundary corresponds to the point (± two cells) where deltaC expression levels increase in wild-type embryos and the anterior-most extent of her1 expression in fss/tbx24−/− embryos. At the 10 somite stage, the posterior PSM contains the anlagen for approximately five somites that each are roughly five cells in length along the A-P axis (Fig. 1E). The anterior PSM contains all cells lying anterior to the posterior PSM and contains enough cells for approximately two to three somites (Mara et al.,2007). By convention, the most recently formed somite is the SI, while the second-newest somite is the SII. In the PSM, the S0, S-I, and S-II will sequentially give rise to the next three segments (Fig. 1D; Pourquie and Tam,2001).


The notch signaling pathway is one of the most broadly used cell:cell communication mechanisms in metazoan development. Notch is a large transmembrane receptor with multiple extracellular EGF repeats and a cytoplasmic domain with six ankrin repeats and a RAM domain. Notch binds to its transmembrane ligands Delta/Jagged/Serrate on the surface of adjacent cells. Ligand binding causes Notch to be cleaved in its transmembrane domain by γ-Secretase, releasing the cytoplasmic domain in the receiving cell. The cytoplasmic domain translocates to the nucleus where it interacts with the DNA binding protein Su(H)/RBPJκ, converting it from a transcriptional repressor to an activator. notch signaling frequently activates the transcription of members of the hairy/enhancer-of-split family of transcriptional repressors. The ability of Delta to activate Notch requires the ubiquitinylation of the cytoplasmic domain of Delta by an E3 ubiquitin ligase of either the Mindbomb or Neuralized family. This ubiquitinylation causes the sending cells to internalize Delta and the extracellular domain of Notch from the adjacent cell. There is evidence that Delta can also cell-autonomously inhibit the ability of a cell to receive a Delta signal from an adjacent cell (Bray,2006). The mouse Delta-like-3, which is required for murine segmentation, appears to specifically function in this inhibitory manner (Ladi et al.,2005). The fringe genes, such as lunatic fringe, are glycosyltransferases that modify the EGF domain of Notch while the receptor passes through the Golgi. This modification biases Notch's affinity for its different ligands. There are additional genes within the notch pathway, but those mentioned above are the ones that will be discussed here in the context of zebrafish segmentation. For a more detailed review of notch signaling, see Bray (2006).

The first suggestion that the notch pathway may play a role in zebrafish segmentation came from expression analysis. notch1a is expressed in the hypoblast of the germ ring during gastrulation and is later expressed in the progenitor zone, initiation zone, PSM, and posterior of each somite (Fig. 2) (Bierkamp and Campos-Ortega,1993). notch2, formerly notch6, is expressed in the PSM, but not in the posterior tail bud (progenitor zone and initiation zone). In the most recently formed somites, notch2 is transcribed in the anterior half of each segment and along the dorsal and ventral surface of the somite (Fig. 2; Westin and Lardelli,1997; Kortschak et al.,2001). notch3, formerly notch5, is expressed in stripes in the anterior PSM and in the posterior of each somite (Fig. 2; Westin and Lardelli,1997; Kortschak et al.,2001). notch1b is expressed in the posterior of each somite but not the PSM (Fig. 2; Westin and Lardelli,1997). Two notch ligands, deltaD and deltaC, are important for zebrafish somitogenesis. deltaD is expressed in the germ ring during gastrulation and later in the progenitor zone, initiation zone, and throughout the PSM (Dornseifer et al.,1997). In the anterior PSM, deltaD is up-regulated in a broad domain that refines into strong stripes of expression (Dornseifer et al.,1997; Holley et al.,2000,2002). After segmentation, deltaD expression is enriched in the anterior half of each somite (Fig. 2; Dornseifer et al.,1997). deltaC is expressed in the germ ring during gastrulation, in the progenitor zone at the posterior tail bud and in a striped pattern in the PSM (Smithers et al.,2000). In the nascent somite, deltaC is transcribed in the anterior half of the segment, but during somite morphogenesis, deltaC expression switches polarity to be expressed in the posterior of each somite (Fig. 2; Smithers et al.,2000; Jülich et al.,2005b). lunatic fringe, is transcribed in a segmental pattern in the anterior PSM and in the anterior half of the nascent somites (Fig. 2; Prince et al.,2001; Qiu et al.,2004). Zebrafish have two homologues of nrarp, an intracellular protein containing two ankrin repeats, which acts as a negative regulator of notch signaling (Lamar et al.,2001; Topczewska et al.,2003). The zebrafish nrarp a mRNA is seen in the germ ring during gastrulation and throughout the progenitor zone, initiation zone, and posterior PSM. In the anterior PSM, nrarp a is expressed in segmental stripes in the anterior half of the nascent and forming somite. nrarp b mRNA is observed in one segmental stripe in the anterior PSM but not elsewhere in the somitic mesoderm (Fig. 2; Topczewska et al.,2003). mib and the two zebrafish Su(H) homologues, Su(H)1 and 2, are each ubiquitously expressed during gastrulation and segmentation (Fig. 2; Itoh et al.,2003; Sieger et al.,2003; Echeverri and Oates,2007).

Figure 2.

Notch pathway genes and somitogenesis. Whether expression of the gene oscillates, yes (y) or no (n), is indicated. The expression patterns are depicted. The three most recently formed somites are shown with the anterior (a) and posterior (p) halves of each somite labeled. The tail bud is divided into the anterior presomitic mesoderm (PSM; a psm), posterior PSM (p psm), initiation zone (iz), and progenitor zone (pz). The bd indicates that the gene is required for normal oscillating gene expression or for segment polarity but that the phenotype is background dependent. The regions of the body axis that are affected in the mutant or morphant are indicated. Segments in green, yellow, or orange form normally in the mutant or morphant, whereas uncolored segments are missing or abnormal. The segments in the three shades of gray indicate an absence of the relevant loss-of-function data. References: a (Bierkamp and Campos-Ortega,1993), b (van Eeden et al.,1996), c (van Eeden et al.,1998), d (Holley et al.,2002), e (Westin and Lardelli,1997), f (Kortschak et al.,2001), g (Dornseifer et al.,1997), h (Holley et al.,2000), i (Smithers et al.,2000), j (Jiang et al.,2000), k (Jülich et al.,2005b), l (Prince et al.,2001), m (Qiu et al.,2004), n (Topczewska et al.,2003), o (Ishitani et al.,2005), p (Jiang et al.,1996), q (Itoh et al.,2003), r (Sieger et al.,2003), and s (Echeverri and Oates,2007).

The first functional evidence for the role of notch signaling in zebrafish segmentation came from gain-of-function studies. Ectopic expression of an activated form of notch consisting solely of the intracellular domain, called NIC or NICD (2007), perturbs the normal pattern of her1 expression, segment polarity, and morphological somite formation (Takke and Campos-Ortega,1999). Similarly, overexpression of a dominant-negative form of the Xenopus suppressor of hairless homologue that cannot bind DNA, X-Su(H)DBM, perturbs morphological somite formation (Wettstein et al.,1997; Jülich et al.,2005a). It has been reported that ectopic expression of either deltaD or deltaC can perturb zebrafish somitogenesis (Takke and Campos-Ortega,1999). Others have been unable to observe such an effect by overexpressing either wild-type deltaD or deltaC, but have observed segmentation defects when overexpressing a chimeric delta containing the extracellular and transmembrane domain of deltaC and the cytoplasmic domain of deltaD (Mara et al.,2007).


In the large-scale genetic screens performed in Tübingen and Boston, five genes were found that are required for normal segmentation of the mesoderm: fused somites (fss), after eight (aei), deadly seven (des), beamter (bea), and white tail/mindbomb (mib; Jiang et al.,1996; Schier et al.,1996; van Eeden et al.,1996). These mutants displayed defects in morphological segmentation and segment polarity. The segmentation defects subsequently led to mispatterning of primary motoneurons and malformed vertebral bodies (van Eeden et al.,1996). mib embryos were shown to have lost the segmental expression of Myf-5 and notch1a (Jiang et al.,1996). The gene affected in each of these mutants has been identified and four of the five are notch pathway genes. after eight and beamter encode the Notch ligands deltaD and deltaC, respectively (Holley et al.,2000; Jülich et al.,2005b). deadly seven is the receptor notch1a (Holley et al.,2002). mindbomb encodes an E3 ubiquitin ligase required for notch signaling (Itoh et al.,2003). fused somites, the only gene of the five that is not a notch pathway gene, encodes the t-box family transcription factor tbx24 (Nikaido et al.,2002).


The segmentation clock creates oscillations in gene expression that manifest as stripes that traverse the PSM in a posterior to anterior direction (Fig. 1D). The initial genes shown to oscillate were vertebrate orthologues of the Drosophila pair-rule gene hairy, a target of notch signaling. hairy orthologues in zebrafish and mouse are called her (hairy and enhancer of split-related) or hes (hairyenhancer ofsplit) genes, respectively. The existence of a segmentation clock was first demonstrated in the chick by dissecting and culturing the bilateral halves of the PSM. Half was cultured for a longer period of time before fixation and in situ hybridization. These experiments demonstrated that the position of the stripes of c-hairy in the PSM changed over time and that the cyclical characteristic of these changes in gene expression correlated with the length of the somite cycle (Palmeirim et al.,1997). Due to the small size and morphogenesis of the zebrafish tail bud, the bilateral culturing technique has not yet been successfully applied to zebrafish. Instead, her1 expression was shown to oscillate by measuring the distances between the myoD and her1 stripes in double in situ hybridization experiments. The distances between the myoD stripes, representing the segmental pattern of the formed somites, does not vary significantly, while the distances between the her1 stripes showed a large variation. When embryos were staged at the early or late 12 somites stage before fixation, and the distances between the her1 stripes measured, it was found that the stripes moved anteriorly and that distance between the her1 stripes decreased over developmental time (Fig. 3). These changes in gene expression could not be explained by cell migration or cell death (Holley et al.,2000). A similar, but nonquantitative, set of experiments suggested that her1 expression oscillated by analyzing the position of the her1 stripes relative to the segmental stripes of mesp-a in the anterior PSM (Sawada et al.,2000). deltaC expression was shown to oscillate using a different methodology than the “staging-measurement” method applied to her1. Jiang and colleagues used a “heat gradient” to disjoin the rates of development of the bilateral halves of the PSM. Because the rate of zebrafish development is dependent upon temperature, this caused the warmer side of the tail bud to develop faster and the bilateral stripes of deltaC expression to be uncoupled. By incubating embryos in the heat gradient for different lengths of time, the phases of bilateral deltaC expression shifted in a reproducible pattern, indicating that deltaC expression oscillated (Jiang et al.,2000). A subsequent study using the “staging-measurement” method examined deltaC and deltaD expression and showed that expression of the former oscillates while the latter does not (Fig. 3; Holley et al.,2002). Subsequent studies showed that her7, her11, her12, and her15 oscillate by performing double in situ hybridization with her1 and observing cyclical changes in expression in correlation with changes in her1 expression (Fig. 3; Oates and Ho,2002; Gajewski et al.,2003,2006; Sieger et al.,2004; Shankaran et al.,2007).

Figure 3.

hairy/enhancer of split-related genes in somitogenesis. Whether expression of the gene oscillates, yes (y) or no (n), is indicated. The expression patterns are depicted. The three most recently formed somites are shown with the anterior (a) and posterior (p) halves of each somite labeled. The tail bud is divided into the anterior presomitic mesoderm (PSM; a psm), posterior PSM (p psm), initiation zone (iz), and progenitor zone (pz). The bd indicates that the gene is required for normal oscillating gene expression or for segment polarity but that the phenotype is background dependent. The regions of the body axis that are affected in the mutant or morphant are indicated. Segments in green, yellow, or orange form normally in the mutant or morphant, whereas uncolored segments are missing or abnormal. The segments in the three shades of gray indicate an absence of the relevant loss of function data. Note that there are a variety of independently reported morphological defects caused by loss of her1 function. Loss of her1 in combination with loss of other genes can perturb all somites. References: a (Holley et al.,2000), b (Sawada et al.,2000), c (Holley et al.,2002), d (Henry et al.,2002), e (Oates and Ho,2002), f (Gajewski et al.,2003), g (Sieger et al.,2004), h (Sieger et al.,2006), i (Takke et al.,1999), j (Gajewski et al.,2006), k (Pasini et al.,2001), l (Pasini et al.,2004), m (Oates et al.,2005a), n (Shankaran et al.,2007), o (Kawamura et al.,2005b), and p (Winkler et al.,2003).

deltaC, her1, her7, her11, her12 and her15 all generally oscillate in phase, and they are all regulated by notch signaling. her1 and her7 are located head to head in the zebrafish genome separated by 11 kb, and the two genes exhibit similar oscillating expression (Henry et al.,2002; Gajewski et al.,2003). However, there are notable differences in the expression of her11, her12, and her15. her11 appears only to oscillate in the anterior half of the tail bud, while her12 and her15 oscillations are confined to the posterior half of the tail bud (Fig. 3; Gajewski et al.,2006; Sieger et al.,2006; Shankaran et al.,2007). Control of oscillating gene expression differs between the anterior PSM and the rest of the tail bud. This observation was first revealed through study of the fss/tbx24 mutant, which only affects the oscillating expression in the anterior PSM, while the notch pathway mutants exhibit aberrant expression throughout the tail bud (van Eeden et al.,1998; Holley et al.,2000). Furthermore, transgenic analysis of the her1 cis regulatory elements demonstrates that distinct elements are responsible for driving expression in the anterior and posterior PSM (Gajewski et al.,2003).

Fluorescent in situ hybridization using Tyramide Signal Amplification (TSA) allows the identification of different phases of oscillation by means of the subcellular localization of mRNA. Using this method, one can see robust transcription of the chromosomal loci at the anterior of each stripe of her1, her7, and deltaC expression (Jülich et al.,2005b; Horikawa et al.,2006; Mara et al.,2007). More posteriorly, the nuclei fill with transcript and then the whole cell fills with the oscillating mRNA. In the posterior of each stripe, cells with no nuclear and decreasing levels of cytoplasmic staining can be observed as cells have ceased active transcription and the mRNAs are degraded (Jülich et al.,2005b; Mara et al.,2007). The oscillations of her1, her7, or deltaC were characterized in detail by examining the expression of each gene in a large number of cells. In the progenitor zone, her1, her7, and deltaC begin to be expressed, but this expression does not appear to oscillate. Initiation of oscillation seems to be coupled to exit from the progenitor zone laterally into the initiation zone. However, oscillations in the initiation zone do not appear to be as well synchronized among neighboring cells as they are in the posterior PSM. The appearance of synchronized oscillations correlates with an attenuation in cell movement, suggesting that cell movement is a source of noise for the segmentation clock (Mara et al.,2007). Mitosis also produces noise with which the system must contend to establish and maintain synchrony among the clocks of neighboring cells. Visualization of her1 mRNA with TSA concomitantly with a green fluorescent protein (GFP) -tagged histone 2B showed that condensed or segregating chromatin does not exhibit active transcription of the clock genes, while nondividing neighboring cells are transcriptionally active. During every 30-min somite cycle, 10–15% of cells in the PSM undergo mitosis, which minimally takes 15 min. Therefore, a significant portion of cells in the PSM will have their clocks shifted out of phase with their neighboring cells due to mitosis. The segmentation clock must buffer against this noise to create the segmental prepattern (Horikawa et al.,2006).

Notch Function in the Segmentation Clock

The zebrafish notch pathway mutants, bea/deltaC, aei/deltaD, des/notch1a, and mib, each form the anterior three to nine somites before segmentation breaks down (Jiang et al.,1996; van Eeden et al.,1996). Accordingly, oscillating gene expression is normal during anterior somitogenesis but gradually deteriorates before the onset of the morphological defects (van Eeden et al.,1998; Jiang et al.,2000). By the eight somite stage, the expression of her1, her7, or deltaC is no longer in a striped pattern but is in a disorganized “salt and pepper” pattern. Double knockdown of Su(H)1 and 2, disrupted the striped expression of her1, her7, and deltaC, although the overall level of expression was higher than observed in des/notch1a mutants. The elevated expression suggests that, in addition to their positive function in notch signaling, the Su(H) homologues likely function as a repressor in the absence of activated notch signaling (Sieger et al.,2003; Echeverri and Oates,2007). Double knockdown of both nrarp a and nrarp b leads to elevated levels of her1 mRNA but has no apparent affect on the oscillation of her1 expression or on morphological segmentation (Ishitani et al.,2005). Morpholino inhibition of notch3/5 has no significant effect on somitogenesis by itself and can only cause a slight enhancement of the segmentation defect seen in des/notch1a mutants (Susan Truong and Scott Holley, unpublished observations).

In aei/deltaD embryos, the salt and pepper pattern is confined to the anterior PSM, while in bea/deltaC embryos the salt and pepper expression is found throughout the PSM (van Eeden et al.,1998; Holley et al.,2000,2002; Jiang et al.,2000; Oates and Ho,2002). The two delta mutants also differ in the time of onset of the segmentation defect in that bea/deltaC embryos form only three to five normal anterior somites while aei/deltaD embryos generate seven to nine proper somites (van Eeden et al., 1997). These differences suggest that these two deltas have distinct functions in the segmentation clock. Elucidation of these functional differences provides a resolution to the debate about whether notch signaling exclusively synchronizes the oscillations between neighboring cells (Jiang et al.,2000) or if notch also is fundamentally responsible for generating the oscillations (Holley et al.,2000,2002). In short, the answer appears to be that some notch pathway genes drive the clock while others primarily act to synchronize the clocks of adjacent cells (Mara et al.,2007).

Both deltaD and deltaC are homologues of mammalian delta-like-1 (unpublished observations; Ladi et al.,2005). Nonetheless, they have different mutant phenotypes and a series of pharmacogenetic experiments using the deltaD and deltaC mutants indicate that the two deltas represent distinct signals within the segmentation clock. deltaC cannot substitute for deltaD, and chimeric analysis suggests that functional differences between the ligands are conferred by the coding sequence of both the extracellular and intracellular domain (Mara et al.,2007).

Examination of the expression of her1, her7, and deltaC expression using high resolution fluorescent in situ hybridization strongly suggests that these genes do not oscillate in the posterior PSM of aei/deltaD embryos but that asynchronous oscillation persists in bea/deltaC embryos. Moreover, aei/deltaD embryos show a decrease in expression of her1, her7, and deltaC in the progenitor zone. These data suggest that the two deltas perform distinct functions in that deltaD drives the oscillations in the initiation zone and posterior PSM. deltaC appears to primarily function in synchronizing the oscillations among neighboring cells in the PSM (Mara et al.,2007). Given that the expression of deltaC oscillates, it is well suited to couple the oscillations of neighboring cells (Jiang et al.,2000). In contrast, deltaD expression does not oscillate, meaning that it is likely a continuous signal that helps to drive the oscillations while deltaC levels fluctuate on top of the basal level of deltaD. Experiments combining genetic mosaics with high resolution in situ hybridization suggest that a cell needs both deltaD and deltaC to provide a sufficiently strong signal to affect the clocks of neighboring cells. In these experiments, clones of cells lacking her1 and her7, and thus deficient in cyclic gene expression, shifted the phase of oscillation in their neighboring wild-type cells. This experiment demonstrated the coupling of the clocks of adjacent cells, and additional experiments showed that the donor clones required both deltaD and deltaC to influence the oscillations of their neighbors. Furthermore, the phase shift created by the donor clones was successfully recapitulated in a computational model of the clock (Horikawa et al.,2006).

Additional evidence of the coupling of the oscillations among neighboring cells comes from mosaic experiments in which wild-type cells from the posterior PSM of one embryo are transplanted into the PSM of a host embryo. Because the two embryos may not be in the same phase of the somite cycle at the time of transplantation, the donor cells often show different patterns of her1 expression than the host embryo when the mosaics are examined immediately after transplantation. In contrast, mosaics examined three somite cycles after transplantation showed synchronous oscillation of donor and host cells (Horikawa et al.,2006).

In models of the segmentation clock, the instability of both the oscillating mRNAs and proteins is an important feature, an idea supported by computational analysis (Lewis,2003; Cinquin,2007). Interestingly though, embryos homozygous for the tortuga mutation display a posttranscriptional stabilization of the oscillating mRNAs her1, her7, and deltaC as revealed with in situ hybridization with an exon probe, but still have normal transcriptional oscillation of her1 and deltaC as indicated by in situ hybridization with an intron probe. In tortuga embryos, peaks and valleys of expression are still observed, and these differential levels may be sufficient to maintain the oscillating pattern even with persistent low levels of mRNA in the interstripe regions. Consistent with the normal transcriptional oscillations, tortuga mutants have no segmentation defect. her1, deltaC and her7 oscillating mRNAs are differentially affected in the tortuga mutant in that the defects in her1 expression are first seen at the three somite stage, while the elevated levels of deltaC and her7 mRNAs are first observed at the 6 and 16 somite stage, respectively. With each mRNA, the persistent levels of mRNA gradually increase with each successive somite cycle (Dill and Amacher,2005). The increase may reflect diminishing quantities of maternally supplied tortuga, whose molecular identity is currently unknown, or may be indicative of changes in the clock function over time.

Her Function in the Segmentation Clock

her1 and her7 have been the most extensively studied of the oscillating her genes. The functional analysis of her1 has been complicated by differing phenotypes being reported for the her1 morphants. Injection of a translation-blocking morpholino against her1 stabilizes the her1 mRNA in addition to blocking production of protein. The initial report of the her1 morphant phenotype did not recognize the mRNA stabilization (Holley et al.,2002), while subsequent studies did (Henry et al.,2002; Oates and Ho,2002). A later study demonstrated the stabilization definitively by examining her1 morphants with both a probe against the full-length cDNA, which recognizes all transcripts, and with an intron probe, which recognizes only nascent unspliced transcripts. This analysis suggests that oscillations in her1 transcription still occur in the posterior PSM of her1 morphants, but that her1 expression disappears in the anterior PSM (the same effect was seen on her7 expression; Gajewski et al.,2003). Oates and Ho reported very weak effects on both her1, her7, and deltaC expression in the her1 morphants (Oates and Ho,2002). In contrast, two other reports showed abolition of deltaC oscillation in the her1 morphants (Holley et al.,2002; Gajewski et al.,2003). Despite these differences, it is clear that her1 morphants have a weaker phenotype than her7 morphants, which lose the striped pattern of her1, her7, and deltaC (Fig. 3; Oates and Ho,2002; Gajewski et al.,2003). Moreover, concomitantly eliminating both her1 and her7, either by means of morpholinos (Oates and Ho,2002) or chromosome deletion (Henry et al.,2002), leads to a stronger phenotype. This result suggests that her1 and her7 have partially overlapping functions within the zebrafish somite clock. A similar result was attained with the functional analysis of her11. Knockdown of her11 alone had very little effect on segmentation, but in combination with inhibition of either her1 or her7 led to a strong perturbation of the oscillating expression of her1 and her7 (Sieger et al.,2004). Knockdown of her12 perturbs the oscillating expression of her1, her7, and deltaC, indicating a nonredundant function for her12 within the segmentation clock. In contrast, inhibition of her15 alone or in combination with other genes did not lead to a segmentation defect (Shankaran et al.,2007).

Because the Her proteins function as dimers, distinct species of homo- and heterodimer could exist in different regions of the tail bud and these dimers could vary in their binding specificity, repressive activity, and stability (Fig. 4). Based on the patterns of mRNA expression, Her1, Her7, Her12, and Her15 are present throughout the tail bud. Her4.1 expression should overlap with these proteins in the progenitor zone (Fig. 3; Gajewski et al.,2006), while Her13.2 should be coexpressed with them throughout the posterior tail bud (Fig. 3; Kawamura et al.,2005b). Her11 should overlap with the other oscillating Her proteins in the anterior and posterior PSM. Experimentally, Her1 has been shown to interact with Her13.2 in a GST pull-down experiment, and a luciferase reporter assay demonstrated that cotransfection of both her1 and her13.2 had a synergistic effect in repressing reporter gene expression driven by the her1 enhancer. Thus, the Her1:Her13.2 heterodimer may have a stronger repressor activity than the Her1 or Her13.2 homodimer. In vivo, morpholino knockdown of her13.2 leads to a strong perturbation of the oscillating expression of her1, her7, and deltaC, a phenotype similar to that of the her7 morphant (Fig. 3; Kawamura et al.,2005b). It has also been noted that knockdown of her7 leads to a general elevation in transcription of her1, suggesting that her7 is a transcriptional repressor (Oates et al.,2005a). While loss-of-function phenotypes suggest that her7 and her13.2 act as repressors in vivo, the her1 morphant phenotype is more complicated. Given that her1 inhibits its own expression in the luciferase assay, it is odd that knockdown of her1 leads to a loss of her1 and her7 expression in the anterior PSM as one would expect that elimination of a repressor would lead to elevated expression of its target genes (Gajewski et al.,2003; Kawamura et al.,2005b). A resolution to this apparent paradox is suggested by computational modeling of the zebrafish clock. In testing 13 parameters that were independently varied, the her1 morphant phenotype can be accounted for if Her1 and Her7 form a heterodimer and if this heterodimer has a weaker repressive activity than either homodimer or heterodimer with Her13.2. In this model, the Her1/Her7 heterodimer acts as a “protective species” that would bind to cis regulatory elements and prevent binding of the more robust repressors (Cinquin,2007).

Figure 4.

Gene networks governing zebrafish segmentation. A: The oscillator gene network in the posterior tail bud is depicted. Through Notch1a, DeltaD drives expression of the oscillating her genes and deltaC. RPTPψ acts upstream of, or in parallel to, the notch pathway. The Her genes act in a negative feedback loop to block transcription of the oscillating genes. After the Her proteins are degraded, transcription of the oscillating genes is reinitiated. Her proteins can act as homodimers or heterodimers, and possible dimerization partners are shown in brackets. The oscillating her genes are in blue. fgf signaling affects the clock by promoting the expression of her13.2 in the posterior tail bud (red). Other, nonoscillating her genes are in gray. deltaC acts in a feedback loop to coordinate oscillations among neighboring cells. B: In the anterior tail bud, fgf and her13.2 no longer influence the clock circuitry, but the oscillations do become dependent upon fss/tbx24 (yellow). Note that the color-coding corresponds to that of Figure 5. C: The clock, in conjunction with fss/tbx24, establishes segment polarity. mespb appears to promote anterior half segment fates and repress posterior half segment fates. Anterior and posterior half segments may interact to stabilize the segmental pattern. Segment polarity influences somite morphogenesis by means of transcriptional control of ephrin/eph genes and posttranscriptional regulation of Integrins and Cadherins. Genes differentially expressed in the anterior-half somite are notch2/6, deltaD, lfng, nrarp-a, nrarp-b, her6, her11, ripply1, papc, snail2, tbx15, ephA4, ephrinB2b, and fgf8. Genes differentially expressed in the posterior-half somite are notch1a, notch1b, notch3/5, deltaC, her4, hey1, ephrinA1, ephrinB2a, myoD, uncx4, and snail1a.

Figure 5.

The wavefront and somitogenesis. The tail buds are depicted with the oscillating genes on the left half (blue) and wavefront genes on the right half (red and yellow). Note that the color-coding corresponds to that of Figure 4. A: The wavefront consists of a gradient of fgf signaling and her13.2 expression in the posterior tail bud (red). In the anterior presomitic mesoderm (PSM), the wavefront is characterized by a gradual decrease in fgf and her13.2 and the emerging activity of fss/tbx24 (yellow). The fgf + her13.2 and fss/tbx24 domains progress posteriorly with the extension of the body axis. B: In the absence of fss/tbx24, oscillations occur normally in the posterior tail bud but disappear in the anterior PSM. Subsequently, there is no attempt to make segment borders. C: In the absence of normal clock function in the aei/deltaD mutant, oscillations of her1, her7, and deltaC expression do not occur in the posterior PSM. fss/tbx24 promotes the anterior expression of the oscillating genes in a disorganized salt and pepper pattern. The salt and pepper pattern may represent asynchronous oscillations or an attempt to establish segment polarity in the absence of an oscillator-generated prepattern. This salt and pepper pattern arises in an anterior to posterior direction along the receding wavefront. A wave of irregular boundary formation follows the salt and pepper pattern.

The roles of her4 and her6 in somitogenesis are less clear. The expression of neither gene oscillates as her4 is expressed in the posterior of the tail bud in the progenitor zone and in a transient stripe in the nascent somite (Fig. 3; Takke et al.,1999; Gajewski et al.,2006). her6 is expressed in the anterior-most PSM and in the posterior half of each somite (Fig. 3; Pasini et al.,2001). Ectopic expression of either her4 or her6 has been shown to perturb somitogenesis (Takke and Campos-Ortega,1999; Pasini et al.,2004). However, loss-of-function data for these genes are less clear. It has been reported that knockdown of either her6 alone or her6 in combination with her4 leads to a perturbation of the oscillating expression of her1 and deltaC (Pasini et al.,2004). How loss of her6 could have this effect when the gene is not expressed in the posterior PSM is unclear.

hairy is a Drosophila pair-rule gene meaning that it is expressed in every-other segment and is required for the formation of alternating segment boundaries in the Drosophila embryo. her1 was initially reported to be expressed in a pair-rule pattern raising the possibility that Drosophila and vertebrates shared segmentation mechanisms (Müller et al.,1996). However, a more recent study comparing her1 expression to the segmental pattern of myoD showed that her1 is expressed in the anlagen of consecutive somites (Holley et al.,2000). In the initial study, Müller and colleagues labeled cells in the PSM that that were in consecutive her1 stripes and found that these cells were subsequently incorporated into alternating somites. In retrospect, these results are likely due to the fact that, in the PSM, distances larger than a single somite can separate consecutive oscillating her1 stripes. In a genetic screen for mutants with aberrant her1 expression, a mutant with a deletion of her1 and the adjacent her7 gene was isolated. The authors reported that alternating somite borders were affected in the her1;her7 deletion mutant, resurrecting the idea that her1 (and perhaps her7) have pair-rule-like functions (Henry et al.,2002). In contrast, others examining the knockdown phenotypes of her1 and/or her7 have not observed a pair-rule-like phenotype (Holley et al.,2002; Oates and Ho,2002; Gajewski et al.,2003). In theory, the discrepancies in observed phenotypes could be due to differences between the phenotype of the deletion allele and morphants. While many consider this issue resolved in favor of non–pair-rule function, it remains a point of contention. In fact, her15 has recently been reported to be expressed in the anlagen of alternating segments (Fig. 3; Shankaran et al.,2007).

Wnt Signaling and the Clock

In the mouse, the wnt pathway plays a role in segmentation, and the wnt pathway genes axin2 and nkd1 oscillate in the PSM. While axin2 expression does not appear to be under the control of notch, oscillating ndk1 expression is dependent upon hes7 (Aulehla et al.,2003; Ishikawa et al.,2004). Conversely, wnt3a is required for the oscillating expression of lunatic fringe in the mouse (Aulehla et al.,2003). In zebrafish, there is no direct evidence that wnt plays a role in governing segmentation. The strongest indication that wnt signaling may have a function in zebrafish segmentation comes from the analysis of receptor protein tyrosine phosphatase ψ, RPTPψ, whose human and mouse homologues have been shown to bind to and dephosphorylate β-catenin. RPTPψ is broadly expressed during the segmentation period. Knockdown of RPTPψ abolishes the striped expression of her1, her7, and deltaC and subsequently leads to a loss of segmental expression of mesp-a, mesp-b, and papc (Fig. 6; Aerne and Ish-Horowicz,2004). RPTPψ is also required for normal convergence extension, a process controlled by noncanonical wnt signaling, perhaps suggesting that RPTPψ influences both branches of wnt signaling (Topczewski et al.,2001; Jessen et al.,2002; Aerne and Ish-Horowicz,2004).


The wavefront represents the anterior to posterior progression of development during the segmentation period. Mechanistically, it can be thought of as the link between axis elongation and morphological somite formation. In zebrafish, there are three genes that participate as part of the wavefront: fgf8, her13.2, and fss/tbx24 (Figs. 4, 5; Holley et al.,2000; Sawada et al.,2001; Kawamura et al.,2005b). fgf8 and her13.2 act in the posterior tail bud (red in Figs. 4, 5), while fss/tbx24 functions in the anterior PSM (yellow in Figs. 4, 5). A unified model for the action of these genes suggests that fgf signaling couples growth of the tail with the segmentation program. fgf in the posterior keeps the cells in an immature state and when the cells escape influence of fgf in the anterior PSM, the oscillations cease and cells acquire a segmental identity. her13.2 links fgf signaling to the clock in the posterior PSM, while the clock becomes dependent upon fss/tbx24 in the anterior PSM (Fig. 4A,B). fss/tbx24 is required to stabilize the oscillations, establish segment polarity, and initiate somite morphogenesis (Fig. 5B).

As in the mouse and the chick, fgf8 is expressed in a gradient in the posterior tail bud with the highest levels in the posterior (Dubrulle et al.,2001; Sawada et al.,2001; Dubrulle and Pourquié,2004). The low end of the gradient is in the rostral end of the posterior PSM. This pattern of fgf8 results in a corresponding gradient of fgf signaling as indicated by the distribution of phosphorylated ERK in the tail bud. In the model, this gradient keeps cells oscillating and prevents them from fixing their segmental identity in the posterior tail bud. Evidence for this model comes from chemical inhibition of fgf signaling using SU5402 (Dubrulle et al.,2001; Sawada et al.,2001). Transient addition of SU5402 for 8 min at the two somite stages causes the seventh and eighth somite to be larger than usual, because the gradient of fgf signaling would have suddenly shrunk toward the posterior. The reason that this shift in the gradient does not affect the third through sixth somites is likely because (1) there is a time delay before the SU5402 can get into the embryo to efficiently block fgf signaling and (2) the anlagen for the third through sixth somite would have already escaped the fgf gradient by the time the drug took effect. This posterior displacement of the fgf gradient causes the oscillations of her1 to cease earlier/more posteriorly. The expression of anterior-specific genes such as mesp-a is also moved to the posterior. A complementary experiment in which an fibroblast growth factor-8 (FGF8) -soaked bead was implanted next the PSM produced smaller somites because it extended the FGF8 gradient anteriorly (Sawada et al.,2001). In all, the drug and bead experiments provide strong support to the idea that a gradient of fgf signaling controls the maturation of the PSM (Dubrulle et al.,2001; Sawada et al.,2001). However, there is no genetic evidence for this model, as the ace/fgf8 mutant does not have a strong somite defect and neither does a conditional knockout of fgf8 in the mouse (Reifers et al.,1998; Perantoni et al.,2005). There are several possible explanations for the mutant phenotypes. First is genetic redundancy, and there is clear evidence for this in zebrafish, as an fgf8;fgf24 double morphant does not even form a tail (Draper et al.,2003). Because the double morphant lacks posterior mesoderm, the identification of a specific effect on the segmentation program is not possible. However, it is easy to imagine that fgf24 could substitute for loss of fgf8 both in promoting tail formation and in regulating somite maturation. A second explanation for the phenotype of the single mutants is that this patterning system is regulative: a mutation that constantly reduces the fgf gradient can be compensated for, while a sudden perturbation with SU5402 can produce a somite defect before regulative properties of the system offset the disruption. A final explanation for the lack of a segmentation phenotype in the fgf8 mutants is that other factors such as wnt signaling may be the operative molecule in vivo or that wnt may function in parallel with fgf signaling to regulate progression of the wavefront (Aulehla et al.,2003).

In the posterior tail bud, her13.2 connects fgf signaling to the clock (Figs. 4A, 5, red). her13.2 is expressed in virtually the same domain in which phosphorylated ERK is observed: in the progenitor zone, initiation zone, and caudal half of the posterior PSM (Fig. 3). her13.2 expression is reduced in SU5402-treated embryos and expanded by implantation of an FGF8-soaked bead. Unlike her1, her13.2 expression is not effected by perturbation of notch signaling (Kawamura et al.,2005b). However, like the notch pathway mutants, morpholino knockdown of her13.2 perturbs somite formation after the seventh to ninth somite and strongly disrupts the oscillating expression of her1, her7 and deltaC (Fig. 3; Holley et al.,2000; Jiang et al.,2000; Oates and Ho,2002; Kawamura et al.,2005b). Thus, while her13.2 is regulated by fgf signaling, her13.2 functions like the notch pathway components of the clock. Indeed, knockdown of her13.2 has no effect on the expression of no tail or spadetail, both targets of fgf signaling (Kawamura et al.,2005b).

In the anterior PSM as cells escape the influence of fgf8 and down-regulate her13.2, the expression of the oscillating genes becomes dependent upon fss/tbx24 (Figs. 4B, 5, yellow; van Eeden et al.,1998; Holley et al.,2000). fss/tbx24 is broadly expressed in the posterior and anterior PSM and in the anterior half of the two most recently formed somites (Fig. 6; Nikaido et al.,2002). In fss/tbx24 mutants, the oscillations occur normally in the posterior PSM but expression of these genes disappears in the anterior PSM (Fig. 5B; van Eeden et al.,1998; Holley et al.,2000). This loss of expression appears to be due to a failure to reinitiate expression of the oscillating genes after finishing the normal “off” phase of the cycle. These embryos do not establish segment polarity and do not make any somites (van Eeden et al.,1996). In contrast to the notch pathway mutants, which make irregular borders in the posterior trunk and tail, cells in the paraxial mesoderm of fss/tbx24 embryos do not even make irregular borders and in fact fail to undergo a mesenchymal to epithelial transition (Fig. 5B; Durbin et al.,2000; Holley et al.,2000; Barrios et al.,2003).

Figure 6.

Nonnotch pathway genes involved in patterning the presomitic mesoderm (PSM) and/or establishing segment polarity. Whether expression of the gene oscillates, yes (y) or no (n), is indicated. The ∼y indicates that fss/tbx24 is only necessary for normal expression of the oscillating genes in the anterior PSM. The ∼y indicates that loss of fgf8 may have a mild effect on somite pattern. The expression patterns are depicted. The three most recently formed somites are shown with the anterior (a) and posterior (p) halves of each somite labeled. The tail bud is divided into the anterior PSM (a psm), posterior PSM (p psm), initiation zone (iz), and progenitor zone (pz). The regions of the body axis that are affected in the mutant or morphant are indicated. Segments in green, yellow, or orange form normally in the mutant or morphant, whereas uncolored segments are missing or abnormal. The segments in the three shades of gray indicate an absence of the relevant loss of function data. References: a (Aerne and Ish-Horowicz,2004), b (van Eeden et al.,1996), c (van Eeden et al.,1998), d (Durbin et al.,2000), e (Holley et al.,2000), f (Nikaido et al.,2002), g (Reifers et al.,1998), h (Kawamura et al.,2005a), i (Topczewska et al.,2001b), j (Topczewska et al.,2001a), k (Durbin et al.,2000), l (Sawada et al.,2000), and m (Kawahara et al.,2005).

As a cell enters the segmentation program in the initiation zone and becomes relatively less posterior as the tail continues to extend, the architecture of the oscillator circuit changes. Initially, oscillating Her1 will heterodimerize with Her13.2, which is under the control of fgf8. This heterodimer acts in a negative feedback loop to repress the expression of clock genes (Fig. 4A; Kawamura et al.,2005b). As cells enter the anterior PSM, other species of homo- and heterodimer, which may differ in their repressive activity, are predicted to replace Her13.2-containing heterodimers. At the same time, the oscillations become dependent upon fss/tbx24 (Fig. 4B; van Eeden et al.,1998; Holley et al.,2000). The maturation front that controls somite formation can be shifted posteriorly by transiently inhibiting fgf activity while in the absence of fss/tbx24 somite maturation does not occur at all (Holley et al.,2000; Sawada et al.,2001). In the absence of oscillations in the posterior of aei/deltaD embryos, the wavefront creates a salt and pepper pattern of expression of the oscillating genes in the anterior PSM followed by irregular border morphogenesis (Fig. 4C; Holley et al.,2000,2002). If fgf signaling is inhibited in an aei/deltaD embryo, this wavefront of salt and pepper expression is again shifted to the posterior (Sawada et al.,2001). In aei/deltaD;fss/tbx24 embryos, this salt and pepper expression is lost (Holley et al.,2002). Examination of her1, her7, and deltaC expression in the anterior PSM of aei/deltaD embryos suggests that her1 and her7 may oscillate asynchronously but that deltaC expression does not (Mara et al.,2007). (Decoupling of deltaC expression from her1 and her7 is also seen when the retinoic acid catabolizing enzyme, cyp26a1, is inhibited (Echeverri and Oates,2007)). A salt and pepper pattern is not inherently indicative of oscillations, as nonoscillating genes such as the mesps show a salt and pepper pattern in aei/deltaD embryos (Durbin et al.,2000; Sawada et al.,2000).


Within the anterior PSM, segment polarity is established by means of consummation of the pattern created by the clock (Fig. 4C,D). Not surprisingly, mutations that affect clock function also perturb segment polarity, with markers of anterior and posterior-half somites generally expressed in a disorganized manner throughout the somitic mesoderm (van Eeden et al.,1996; Durbin et al.,2000). Three genes, fss/tbx24, ripply1, and foxc1a, play early roles in establishing segment polarity, with fss/tbx24 and ripply1 also having effects on oscillating gene expression in the anterior PSM (Fig. 4C; van Eeden et al.,1998; Holley et al.,2000; Topczewska et al.,2001a; Kawamura et al.,2005a). Downstream of fss/tbx24, the pathway to segment polarity branches with fss/tbx24, ripply1, and foxc1a upstream of mesp-b, which appears to promote anterior half-somite identity. mesp-a expression overlaps that of mesp-b and is downstream of fss/tbx24 but not ripply1 or foxc1a (Fig. 4C; Durbin et al.,2000; Sawada et al.,2000; Topczewska et al.,2001a; Kawamura et al.,2005a; Oates et al.,2005b).

fss/tbx24, which is broadly expressed in the PSM, is required for the expression of ripply1, foxc1a, her4, mesp-b, and mesp-a (Fig. 6; unpublished observations; Durbin et al.,2000; Sawada et al.,2000; Topczewska et al.,2001a; Kawamura et al.,2005a; Oates et al.,2005b). Furthermore, some markers of anterior half-somites such as ephA4, ephrinB2b, lfng, fgf8, and papc and a few markers of the posterior half-somite such as ephrinA1 and notch3/5 are not expressed in fss/tbx24 mutants. Other genes that are normally segmentally expressed in the posterior half-somite, including myoD, ephrinB2a, deltaC, notch1a, and notch1b, and anterior-half somites such as notch2 and deltaD are expressed throughout the somitic mesoderm (unpublished observations; van Eeden et al.,1996; Durbin et al.,2000; Holley et al.,2000; Sawada et al.,2000; Topczewska et al.,2001a; Barrios et al.,2003; Kawamura et al.,2005a; Oates et al.,2005b).

Genetic mosaic experiments indicate that the dependence of her1 expression on fss/tbx24 is cell autonomous as is the expression of mesp-b, papc, and fgf8 (Holley et al.,2000; Oates et al.,2005b). However, wild-type cells transplanted into an fss/tbx24−/− embryo can induce surrounding host cells to express notch3/5. mesp-b, papc, and fgf8 are expressed in the anterior half-somite, while notch3/5 is a posterior marker. Thus, fss/tbx24 may function primarily cell-autonomously to promote expression of anterior half-somite fates which then induce or promote the adoption of posterior fates in neighboring cells (Oates et al.,2005b). Therefore, there is cell–cell signaling in the S-II, S-I, and S0 that helps convert the oscillator-generated pattern into segment polarity. This fss/tbx24-dependent, anterior to posterior wave of cell–cell communication may contribute to the salt and pepper pattern of gene expression in the anterior PSM of the notch pathway mutants such as aei/deltaD−/−, as this salt and pepper pattern is missing in fss/tbx24;aei/deltaD double mutants (Holley et al.,2000).

Ripply1 is a nuclear protein with a WRPW domain, which allows it to physically interact with Groucho, a corepressor that also binds to the WRPW domain of Hairy-related proteins (Paroush et al.,1994; Kawamura et al.,2005a). ripply1 is expressed in a segmental pattern in the anterior PSM and in the anterior half the most recently formed somites. A related gene, ripply2 shows only the striped PSM expression (Fig. 6). ripply1 mRNA is largely absent in fss/tbx24 mutants and is not segmented in aei/delta−/−, mib−/−, or mesp-b morpholino-injected embryos. Ectopic expression of ripply1 repressed mesp-b expression, but not mesp-a. Inhibition of ripply1 does not eliminate the striped pattern of mesp-b, but does cause mesp-b expression to persist throughout the somitic mesoderm. These results suggest that Ripply1 acts as a transcriptional repressor and that ripply1 and mesp-b regulate each other's expression. Genetic mosaics indicate that the repression of mesp-b by ripply1 is cell autonomous. By contrast, overexpression or inhibition of ripply1 has little effect on mesp-a, again underscoring the difference in regulation of the two mesp homologues (Fig. 4C). ripply1 does not affect the oscillations but does cause expression of her1 to persist within the somitic mesoderm. deltaC, deltaD, fgf8, and myoD are also expressed throughout the somitic tissue. ripply1 appears to function downstream of fss/tbx24 in promoting the maturation of PSM to somitic mesoderm (Fig. 4C; Kawamura et al.,2005a).

foxc1, a forkhead/winged helix class transcription factor, is broadly transcribed throughout the PSM with particularly strong, somewhat segmental expression in the S0 and S-I (Fig. 6; Topczewska et al.,2001b). Foxc1a protein is evenly distributed in the nuclei of the PSM and nascent somites. The transcription of foxc1a in the anterior PSM is dependent upon fss/tbx24. Like fss/tbx24 and ripply1, inhibition of foxc1a perturbs formation of all somites. Knockdown of foxc1a leads to a loss mesp-b expression but has no affect on mesp-a. In the anterior PSM and nascent somites of the morphants, stripes of ephrinB2a were fused but papc expression was normal. notch3/5 and notch2/6 expression was strongly reduced. Oscillating expression of her1 and deltaC was unaffected, but the expression of deltaC in the somite was missing (Topczewska et al.,2001a). foxc1a, therefore, specifically affects the transition in deltaC expression from the anterior half of the S0 to the posterior half of each somite (Jülich et al.,2005b). In summary, foxc1a is a broadly expressed gene that acts downstream of fss/tbx24 to help establish segment polarity (Fig. 4C; Topczewska et al.,2001a).

mesp-a and mesp-b are basic helix–loop–helix (bHLH) genes, homologous to the thylacine genes of Xenopus and mesp genes in the mouse, the latter of which have been shown to play a major role in establishing segment polarity during mouse somitogenesis (Saga et al.,1997; Sparrow et al.,1998; Durbin et al.,2000; Sawada et al.,2000; Takahashi et al.,2000,2003). mesp-a and b are expressed in a segmental pattern in the S0, S-I, and S-II. mesp-b stripes are in the future anterior half segment while mesp-a expression initially encompasses all of the S-II and refines to only the anterior half of the S-I (Fig. 6; Durbin et al.,2000; Sawada et al.,2000). mesp-a is expressed in a faint, disorganized manner in aei/deltaD−/−, bea/deltaC−/−, des/notch1a−/−, and mib−/− embryos and exhibits a stronger salt and pepper pattern in her7 morpholino-injected embryos (Durbin et al.,2000; Sawada et al.,2000; Oates et al.,2005b). mesp-b mRNA is in a salt and pepper pattern in bea/deltaC−/− and mib−/− (Sawada et al.,2000). Expression of both genes is absent in fss/tbx24−/− (Durbin et al.,2000; Sawada et al.,2000). Ectopic expression of mesp-a causes a defect in gastrulation precluding an analysis of its effect on segmentation. Ectopic expression of mesp-b blocks somite formation, inhibits the expression of myoD and notch3/5, which are normally expressed in the posterior of each somite, and expands the expression of fgfr1, notch2/6, and papc, which are normally expressed in the anterior half of each somite (Sawada et al.,2000). Overexpression of mesp-b did not affect either her1 or mesp-a expression. mesp-b appears to act downstream of fss/tbx24 and foxc1a to promote anterior-half somite and repress posterior-half somite fates (Fig. 4C).

Two zebrafish gadd45β (Growth Arrest and DNA Damage) homologues are expressed in a single stripe in the anterior PSM (Fig. 6). Mammalian gadd45β has been implicated in cell cycle control. Simultaneous knockdown of both homologues blocks formation of all somites and perturbs the segmental expression of mesp-a and myoD. her1 expression is also affected in the morphants, but still appears to oscillate. Interestingly, the posterior domain of fgf8 expression is expanded anteriorly but the expression of fss/tbx24 and the posteriorly expressed genes such as wnt3a and tbx6 are unaffected. Overexpression of gadd45β represses mesp-a and myoD transcription (Kawahara et al.,2005). The exact role of gadd45β in somitogenesis is currently unclear. It could be involved in either establishing segment polarity or regulating the maturation program.

There are indications that specific delta/notch signals may help establish anterior and posterior half-somite identities by means of cell:cell communication. For example, deltaD and deltaC are expressed in the anterior and posterior half-somite, respectively (Dornseifer et al.,1997; Smithers et al.,2000). notch3/5 expression in the posterior half-somite is lost in embryos lacking deltaC but not in aei/deltaD−/− embryos (Oates et al.,2005a). Similarly, the hairy/enhancer of spilt orthologue, hey1 is differentially affected in these mutants. hey1 is expressed in the posterior half of each somite and in the anterior PSM. In wild-type embryos, the level of expression in the most recently formed one to two somites is always weaker than the other expression. This weak expression domain, albeit disorganized, is seen in aei/deltaD−/− embryos but not in bea/deltaC or des/notch1a mutants (Winkler et al.,2003; Sieger et al.,2004). In contrast, her11, which is expressed in an oscillating pattern in the PSM and in the anterior of the nascent somites, displays a similar expression pattern in the these three notch pathway mutants. her11 expression is sometimes missing in the region of the nascent somites in aei/deltaD−/−, bea/deltaC−/−, and des/notch1a−/− embryos (Sieger et al.,2004). This variable expression domain is too far anterior to likely be due to oscillations and probably results from an attempt to generate some segment polarity in these mutant embryos. Note that expression of hey1 and notch3/5 are differentially affected in the mutants while her11 is not. This difference is perhaps due to hey1 and notch3/5 being normally expressed in the posterior half-somite, while hey1 is expressed in the anterior half of each somite. Together, these segment polarity phenotypes suggest that deltaD and deltaC represent distinct signals within the somites that help establish segment polarity by means of cell:cell communication.


As somite polarity is established, morphological segmentation commences. Somite morphogenesis involves a mesenchymal to epithelial transition (MET) of the boundary cells. The typical trunk somite averages around five cells in length and consists of epithelial anterior and posterior boundary cells separated by internal mesenchymal cells (Fig. 1E). The internal mesenchymal cells are not needed for segmentation to occur as knypek;trilobite double mutants form somites consisting of only a row of anterior boundary cells and a row of posterior boundary cells. In wild-type embryos, there is little cell movement in the anterior PSM and only local cell jostling among nascent boundary cells (Henry et al.,2000). The initiation of border morphogenesis is marked by the clustering Integrinα5, along the basal side of the prospective boundary cells (Fig. 8A1) and alignment of the nuclei (not shown; Jülich et al.,2005a). Shortly thereafter, a fibronectin matrix begins to form between adjacent segments (not shown; Crawford et al.,2003; Jülich et al.,2005a). Cells then begin to adopt a columnar morphology with basally aligned nuclei, a process that continues as the somite matures to be the SIII (Fig. 8A2; Henry et al.,2000). Filamentous actin becomes enriched at the boundary (Fig. 8A2; Barrios et al.,2003). Paxillin and activated, and phosphorylated Focal Adhesion Kinase (Fak) also accumulates along the basal side of the boundary cells (Fig. 8A3; Henry et al.,2001; Crawford et al.,2003). The centrosomes localize apically (Fig. 8A2), and β-catenin is enriched along the apical cortex of the boundary cells (Fig. 8A3; Barrios et al.,2003). Several hours after the somite initially forms, a Laminin matrix is assembled along the somite boundary (Crawford et al.,2003).

Figure 8.

Somite and myotome morphogenesis. A: A rendering of a dorsal view of one bilateral half of the paraxial mesoderm. Medial is down, and lateral is up. Note that the boundary cells become gradually more columnar as the somite matures. Inset 1, depicts clustering of Integrinα5 (green) along the nascent somite boundary at the basal surface of the boundary cells. Inset 2, as the boundary matures, filamentous actin (yellow) concentrates along the boundary and the nuclei (red) localize basally. The centrosomes (green) adopt an apical position. Inset 3, Integrin-associated proteins such as Focal Adhesion Kinase and Paxillin (red) are localized to each somite boundary, while adherins junctions proteins such as β-catenin (green) are enriched at the apical side of the boundary cells. B: A schematic time series of myotome morphogenesis adapted with permission from the Company of Biologists (Stellabotte et al.,2007). In each diagram, anterior is up and medial is to the right. Slow muscle fibers (teal) originate medially and are the first to elongate and migrate to the lateral edge of the myotome. This lateral migration induces the differentiation of fast fibers (purple). Myogenic precursor cells (orange) responsible for later growth of the myotome, derive from the anterior somite. These cells rotate to the lateral surface, divide, intercalate into the myotome, and give rise to additional fast muscle fibers in the lateral myotome.

Somite morphogenesis appears to depend upon the coordinated action of eph/ephrin, integrin, cadherin, and notch signaling. Eph proteins are receptor tyrosine kinases that bind to their ligands, Ephrins on the surface of adjacent cells. Ephrins exist in two forms, a six glycosylphosphatidylinositol (GPI)-linked Ephrin-A class, in which the GPI tethers the protein to the plasma membrane, or the transmembrane Ephrin-B class ligands. The Ephrin-B class can activate receptors and also transduce a signal cell-autonomously by means of its cytoplasmic domain. There are two classes of Eph receptors, EphA and EphB, that bind to either the Ephrin-A or Ephrin-B ligands, respectively, although there are examples of promiscuous interactions between the two classes (Pasquale,2005). In the zebrafish, ephA4 and ephrinB2b are expressed in the PSM, and in the anterior of the S-I, S0, and nascent somites. ephrinA1 mRNA is seen in the posterior tail bud, throughout the S-I, and in the posterior of the S0 and each somite, while ephrinB2a is expressed in the posterior of S-I, S0 and each segment (Fig. 7; Durbin et al.,1998; Barrios et al.,2003). EphA4 can bind to EphrinB2a and EphrinA1, while an EphB can only interact with EphrinB2a. Ectopic expression of ephrinB2a, but not ephA4 or ephB, causes a morphological segmentation defect along with aberrant segmental expression of myoD, fgf8, deltaD, paraxis, and her1. Secreted forms of the Ephrins or variants of the receptors lacking the cytoplasmic/kinase domain have a dominant-negative activity when ectopically expressed and produced somite defects. Together, these experiments suggest that eph/ephrin signaling is involved in regulating somite morphogenesis and somite polarity (Durbin et al.,1998).

Figure 7.

Segmentally expressed genes involved in somite morphogenesis and/or commonly used as markers for somite pattern. Whether expression of the gene oscillates, yes (y) or no (n), is indicated. The expression patterns are depicted. The three most recently formed somites are shown with the anterior (a) and posterior (p) halves of each somite labeled. The tail bud is divided into the anterior presomitic mesoderm (PSM; a psm), posterior PSM (p psm), initiation zone (iz), and progenitor zone (pz). References: a (Begemann et al.,2002), b (Durbin et al.,1998), c (Barrios et al.,2003), d (Yamamoto et al.,1998), e (Sawada et al.,2000), f (Weinberg et al.,1996), g (Kawakami et al.,2005b), h (Hammerschmidt and Nüsslein-Volhard,1993), i (Thisse et al.,1993), j (Nisha Tamhankar and Scott Holley, unpublished observations), k (Kudoh et al.,2001), and l (Clements and Kimelman,2005).

The paraxial mesoderm cells in fss/tbx24 embryos do not attempt somite morphogenesis and do not undergo the mesenchymal to epithelial transition (Durbin et al.,2000; Holley et al.,2000; Barrios et al.,2003). Expression of zebrafish paraxis, the homologue of a mouse gene required for somite epithelialization, is maintained in fss/tbx24−/− embryos (Burgess et al.,1996; Shanmugalingam and Wilson,1998; Topczewska et al.,2001a). fss/tbx24 mutant embryos do lack expression of ephA4, and genetic mosaic experiments show that clones of ephA4-expressing cells in fss/tbx24−/− embryos form borders at the interface of the donor and host cells (Durbin et al.,2000; Barrios et al.,2003). ephA4 causes the expressing cells to adopt a columnar morphology and to apically localize β-catenin. Additionally, ephA4 induces the nuclei of adjacent host cells to align basally and centrosomes to reside apically (Barrios et al.,2003). Other aspects of normal border formation are absent from the ephA4-expressing clones, suggesting that additional factors downstream of fss/tbx24 are necessary for a complete MET.

Cadherins are located around the cell cortex of cells in the PSM and in the mature somites (Crawford et al.,2003). Paraxial protocadherin, papc, mRNA is seen throughout the tail bud and S-II but refines to the anterior half of S-I, S0, SI, and SII (Fig. 7; Yamamoto et al.,1998; Sawada et al.,2000). Ectopic expression of a full-length papc does not affect segmentation, but injection of a secreted form encoding three extracellular cadherin domains perturbs somite morphogenesis and myoD expression (Yamamoto et al.,1998). These experiments suggest that, while regulation of papc expression is not needed for normal somitogenesis, homophilic interactions among cadherins are necessary.

Integrins are heterodimeric transmembrane proteins, consisting of an α and a β subunit, that link the extracellular matrix to the actin cytoskeleton. Integrins can signal bidirectionally to modify the extracellular matrix or to alter the polarity of the cystoskeleton and affect gene expression (Hynes,2002). Intracellular effectors of Integrins include the adapter protein Paxillin and Fak, both of which are localized to the basal side of the somite boundary cells (Fig. 8A; Hynes,2002; Crawford et al.,2003). Integrin α5β1 is the primary receptor for fibronectin, and Integrin α6β4 is the Laminin receptor (Hynes,2002).

mRNAs for the integrin-associated genes are generally not observed in a segmental pattern, suggesting that regulation of these genes is largely posttranscriptional. The exceptions to this are fak1a and fibronectin 1b (fn1b), formerly fibronectin 3. fak1a is transcribed in the anterior PSM and in a segmental pattern in the posterior of each somite, and this pattern is perturbed in fss/tbx24−/− embryos and in the notch pathway mutants (Henry et al.,2001). fak1b, in contrast, is ubiquitously expressed during somitogenesis (Crawford et al.,2003). Indeed while Fak protein is seen in all cells in the presomitic and somitic mesoderm, Fak and phosphorylated/activated Fak is enriched along the somite boundary (Henry et al.,2001; Crawford et al.,2003). fn1b is expressed in the PSM and in the somites, while fn1a is expressed in the posterior tail bud and ubiquitously during gastrulation. integrinα5 is transcribed ubiquitously at the shield stage, but is restricted to the posterior tail bud and adaxial cells during somitogenesis (Jülich et al.,2005a; Koshida et al.,2005). integrinβ1 mRNA is seen in the posterior tail bud and throughout the somites (Jülich et al.,2005a).

Mutation of either integrinα5 or fn1a leads to a failure to maintain somite boundaries in the anterior seven to nine somites (Jülich et al.,2005a; Koshida et al.,2005). Knockdown of fn1b leads to a mild extension defect and a failure to maintain all somite boundaries, while elimination of both fn1a and fn1b leads to a severe truncation of the body axis, suggesting that the two genes are partially redundant (Jülich et al.,2005a). Loss of integrinα5 does not affect expression of the oscillating genes or mesp-b but does perturb the segmental expression of myoD. In integrinα5−/− embryos, the boundary cells do not show a polarized distribution of the centrosomes or phosphorylated Fak and the Fn matrix is disjointed. These observations indicate that integrinα5/fn function is required for the completion and/or maintenance of MET (Jülich et al.,2005a; Koshida et al.,2005). Integrin signaling may function with Eph/Ephrin signaling during somite morphogenesis as morpholino knockdown of ephrinB2a only leads to a short delay in boundary formation, while inhibition of both ephrinB2a and integrinα5 leads to a broad failure to maintain the borders of all somites (Koshida et al.,2005).

Integrinα5-GFP clusters along the basal side of the somite boundary cell in wild-type embryos and in the irregular boundaries that form in the posterior of the notch pathway mutants (Fig. 8A). These irregular boundaries require integrinα5, as they do not form in the double mutants between integrinα5 and either aei/deltaD, bea/deltaC, or des/notch1a. In fact, in the double mutants, cells of the paraxial mesoderm fail to undergo any MET. Because MET occurs in the posterior of the integrinα5 mutant and along the irregular boundaries in the posterior of the notch pathway mutants, the absence of MET in the double mutants is a synergistic genetic interaction between the mutations. This synergy suggests that notch and integrinα5 signaling may work in concert to cause MET during somite morphogenesis. The exact relationship between delta/notch, ephrin/eph, and integrinα5 signaling in somite border morphogenesis are not understood. The parsimonious explanation is that notch defines discrete domains of ephrin/eph expression, which then leads to local activation of integrinα5 in the prospective somite boundary cells (Jülich et al.,2005a).



After segmentation, the zebrafish somite is subdivided into myotome, sclerotome, and dermomyotome, which give rise to the musculature, vertebrae, and myogenic progenitors that facilitate subsequent growth of the skeletal muscle (Morin-Kensicki and Eisen,1997; Stickney et al.,2000; Morin-Kensicki et al.,2002; Devoto et al.,2006; Hammond et al., 2006; Hollway et al.,2007; Stellabotte et al.,2007). Here, aspects of the somite derivatives that specifically relate to segmentation are reviewed in detail.

The zebrafish myotome differentiates into the muscle pioneers, slow muscle and fast muscle, with the bulk of the myotome becoming fast muscle. The muscle pioneers are early-developing, Engrailed-expressing slow muscle fibers that are located at the level of the horizontal myoseptum that divides the myotome into dorsal and ventral halves (Waterman,1969; van Raamsdonk et al.,1978,1982; Felsenfeld et al.,1991; Hatta et al.,1991). The muscle pioneers, slow muscle precursors, and horizontal myoseptum are missing in embryos in which transduction of hedgehog signaling to the somites is perturbed, and the somites in these embryos are u-shaped, rather than the chevron shape seen in wild-type embryos (Currie and Ingham,1996; Devoto et al.,1996; van Eeden et al.,1996; Blagden et al.,1997; Schauerte et al.,1998; Karlstrom et al.,1999; Lewis et al.,1999; Barresi et al.,2000; Chen et al.,2001; Coutelle et al.,2001; Du and Dienhart,2001; Roy et al.,2001; Baxendale et al.,2004; Hirsinger et al.,2004; Sekimizu et al.,2004; Wilbanks et al.,2004; Wolff et al.,2004; Kawakami et al.,2005a; Woods and Talbot,2005; Feng et al.,2006; van der Meer et al.,2006).

The slow muscle fibers, including the muscle pioneers, exhibit an interesting developmental progression in that they are derived from the adaxial cells in the medial somites but migrate through the somite to the lateral surface (Figs. 1E, 8). After segmentation, the 15–20 columnar adaxial cells in a somite are arranged roughly 4 × 4 along the anterior–posterior and dorsal–ventral axes. In the SI and SII, the adaxial cells can be seen rearranging such that they stack along the dorsal–ventral axis and each cell extends along the entire anterior–posterior length of the somite, a process that takes approximately 2 hr. Over the next 3–4 hr, the slow muscle fibers migrate laterally through the somite to the lateral surface (Fig. 8B; Devoto et al.,1996). As the slow muscle fibers migrate, they induce a medial to lateral wave of fast muscle fiber differentiation (Blagden et al.,1997; Henry and Amacher,2004). The migration of the slow muscle cells to the lateral surface requires the function of two cadherins, n-cadherin and m-cadherin, which display complementary changes in expression during the slow muscle migration period. n-cadherin is initially expressed only in the medial somite, but the expression expands laterally to encompass the entire somite. n-cadherin, in contrast, is initially expressed throughout the somite but disappears in a medial to lateral wave and is ultimately only expressed in the postmigratory slow fibers on the lateral surface of the myotome. In embryos lacking either of the two cadherins, the lateral migration is perturbed (Cortes et al.,2003). The period of migration of the slow fibers and elongation of the fast fibers has been termed the transitory phase in myotome border formation. The initial phase is the formation of the epithelial somite with a Fibronectin matrix at the border. The third phase of myotome border morphogenesis is marked by the completion of myofiber development and the production of a Laminin matrix. The 10th somite reaches this third phase around 24 hours postfertilization (hpf; Henry et al.,2005).

Each of the Tübingen segmentation mutants are viable to some extent and actually display segmented myotomes, although the borders are somewhat irregular (van Eeden et al.,1996). These boundaries are hedgehog-dependent: they are missing when hedgehog signaling is inhibited in the segmentation mutants (van Eeden et al.,1998; Henry et al.,2005). Inhibition of hedgehog signaling does not affect the initial irregular boundaries that form in the posterior of the notch pathway mutants (Henry et al.,2005). The earliest evidence of segmentation recovery in fss/tbx24 mutant embryos can be seen in the segmental alignment of the nuclei of the slow muscle cells before their lateral migration (van Eeden et al.,1996). Given that the border rescue is seen after the slow muscle migration occurs at 24 hpf, it is likely that the migrating slow muscle fibers are responsible for the boundary recovery in these mutants. The length of slow muscle fibers appears to help define the length of the fast fibers and collectively they may create a myotome boundary. These boundaries, in turn, help to define or stabilize the length of the myofibers (Henry et al.,2005).

The rescue of the myotome boundaries in fss/tbx24 and the notch pathway mutants also requires integrinα5. integrinα5 mutants have u-shaped somites, but in contrast to the hedgehog pathway mutants, the horizontal myoseptum is formed and the specification and lateral migration of the slow muscle fibers is normal. integrinα5 expression is elevated in the adaxial cells, and this Integrinα5, in conjunction with the Integrinα5 expressed in the fast fibers, is likely required for the formation of the extracellular matrix that constitutes the myotome boundary (Jülich et al.,2005a). These Integrins would normally help to anchor the myofibers to the myotome boundary, which acts as a tendon to transfer the force generated by the contracting fibers. If the migrating slow muscle fibers can cause the fast fibers to coordinately differentiate and elongate, collectively these fibers may be able to generate a myotome border de novo, by means of their Integrins, in the segmentation mutants.

The first fast fibers to differentiate in each somite derive from the cells of the posterior somite expressing myoD. Anterior–posterior patterning of the somite also affects later myogenesis in that myogenic progenitors that contribute to the later growth of the myotome are derived from the anterior of each somite. These fast-fiber myogenic progenitors ultimately express pax3/pax7 and are located at the lateral surface of the mature myotome. These anterior cells rotate laterally after somite formation and before the migration of the slow muscle fibers (Fig. 8B). During later growth of the myotome, these lateral progenitors appear to intercalate into the myotome, suggesting that the later growth of the myotome proceeds by lateral addition of fast fibers (Fig. 8B; Hollway et al.,2007; Stellabotte et al.,2007).

Sclerotome and Vertebral Column

While most of the zebrafish somite forms myotome, a small portion is fated to be sclerotome. In contrast, the amniote somite is mostly sclerotome. The zebrafish sclerotome is located in the ventral–medial portion of the somite and contributes to the axial skeleton. Fate mapping has shown that the posterior ventral–medial somite gives rise to both sclerotome and myotome, while the anterior ventral–medial somite only becomes sclerotome (Morin-Kensicki and Eisen,1997). After somite formation, the sclerotome cells undergo an epithelial to mesenchymal transition and migrate dorsally to surround the notochord. These cells express the bHLH gene twist, and the pattern of migration can be seen in transverse sections of progressively more mature somites. As the adaxial cells rearrange and begin to migrate laterally as slow muscle fibers, the twist-expressing cells begin their dorsal migration. As the slow muscle fibers reach the lateral myotome, the twist-expressing cells reach the dorsal side of the notochord (Stickney et al.,2000). The migration path of the anterior sclerotome is less variant than that of the posterior sclerotome and colocalizes with ventrally migrating neural crest cells and specific motor axons within each somite (Morin-Kensicki and Eisen,1997).

The segmental relationship between the somites and the vertebrae is thought to be offset by one half segment, meaning that one somite contributes to the posterior portion of one vertebra and the anterior portion of another vertebra (Fig. 9A; Remak,1855). While the idea of resegmentation has its detractors, it is generally accepted that it does occur in amniotes (Verbout,1976; Stern and Vasiliauskas,2000). Fate mapping experiments suggest that, while resegmentation may occur in the zebrafish, it appears to be leaky in that cells from the anterior or posterior half of a somite may contribute to two consecutive vertebrae rather than strictly giving rise to a specific portion of a single vertebra (Morin-Kensicki et al.,2002).

Figure 9.

The relationship between zebrafish somites and vertebrae. A: The anterior limit of expression of hox genes in the zebrafish trunk is indicated. References: a (Prince et al.,1998), b (Sordino et al.,1996), and c (van der Hoeven et al.,1996). Zebrafish have 30–32 somites, which align to the posterior three quarters of one vertebral body and the anterior one quarter of the next posterior vertebra. The total number of vertebrae range from 29–33, with the variability arising within the transition from rib-bearing to hemal-arch bearing vertebrae (Morin-Kensicki et al.,2002). The different vertebral subtypes are indicated. The two cervical and the first two rib-bearing vertebrae contribute to the Weberian apparatus (Morin-Kensicki et al.,2002; Bird and Mabee,2003; Fleming et al.,2004). Ambiguity regarding the alignment and total number of somites and vertebrae is indicated by the hatched posterior two somites. Adapted with permission from the Company of Biologists from (Morin-Kensicki et al.,2002). B: Ichthyological convention classifies the first four vertebrae as “Weberian,” the rib-bearing vertebrae as “precaudal,” the rib- and hemal arch-bearing as “transitional,” and the hemal arch-bearing as “caudal.” Adapted with permission from (Bird and Mabee,2003).

Fate mapping studies indicate that sclerotome may contribute to the vertebral arches and the centra (vertebral bodies; Morin-Kensicki et al.,2002). However, the centra may derive largely from the notochord and not the sclerotome. Notochords excised before any evidence of vertebra formation and cultured in isolation can form a segmented bone matrix. Moreover, histological analysis suggests that the centra do not contain osteoblasts and may not form by means of a cartilage intermediate. The centra appear to form from bone matrix secreted by the notochord (Fleming et al.,2004). In fss/tbx24−/− embryos, somite borders do not form and the neural and hemal arches are malformed, but the vertebral bodies are not fused (van Eeden et al.,1996; Fleming et al.,2004). Thus, disruption of somitic metamerism does not eliminate segmentation of the vertebral bodies. In contrast, laser ablation of notochord cells does lead to fusion of the centra (Fleming et al.,2004). In counterpoint to the fss/tbx24 mutant, mouse mesp2 knockouts exhibit extensive vertebral fusion (Saga et al.,1997). The difference in severity could be due to the fact that the mouse somite is mostly sclerotome and loss of somitic segmentation in this larger population of cells may not be rescued by any pattern present within the notochord.

Zebrafish usually form 32 vertebrae, but the number may vary from 29–33 (Fig. 9; Morin-Kensicki et al.,2002). There are generally 2 cervical vertebrae, 10 rib bearing vertebrae, 2 rib and hemal arch-bearing vertebrae, 14 hemal arch-bearing vertebrae, and 4 tail fin vertebrae (Morin-Kensicki et al.,2002; Bird and Mabee,2003). The variability in total vertebrae occurs in the number of posterior rib-bearing and anterior hemal arch-bearing vertebrae (Fig. 9A; Morin-Kensicki et al.,2002). The first four vertebrae, encompassing the two cervical vertebrae and the first two rib-bearing vertebrae, contribute to the Weberian apparatus, which is thought to transmit vibrations from the swim bladder to the otic vesicle (Fig. 9; Morin-Kensicki et al.,2002; Bird and Mabee,2003; Fleming et al.,2004). In the ichthyological literature, the term “precaudal vertebrae” is used to describe the rib-bearing vertebrae posterior to the “Weberian vertebrae,” the rib- and hemal arch-bearing vertebrae are called “transitional,” and “caudal vertebrae” are the hemal arch-bearing vertebrae (Fig. 9B; Bird and Mabee,2003). The developing centra can be seen at day 9 by Alizeran red staining for bone, and most structures of the adult axial skeleton are completed by day 21 (Du et al.,2001; Morin-Kensicki et al.,2002; Bird and Mabee,2003).

Alignment of the myotome with the vertebrae indicates that each myotome spans the posterior three quarters of one vertebral body and one quarter of the next posterior vertebra. The first two to three somites do not align next to vertebrae, suggesting that they may contribute to the posterior skull by analogy to the occipital somites of amniotes. Myotome 5, which is derived from somite 5, aligns with the second and third vertebra (Fig. 9A; Morin-Kensicki et al.,2002). Thus, as in the mouse and chick, the anterior limit of hox6 paralogue group expression coincides with the transition from cervical vertebrae to rib-bearing/thoracic vertebrae (Fig. 9A). By contrast, the anterior limit of hox9 paralogue group expression is somite 7 or 8, within the precaudal/rib-bearing/thoracic region, while in mouse and chick the hox9 boundary coincides with the thoracic/lumbar transition (Fig. 9A; Burke et al.,1995; Prince et al.,1998; Morin-Kensicki et al.,2002). Somite 16 should give rise to the last rib-bearing vertebra, and somite 17 should contribute to the first caudal vertebrae. This transition corresponds to the anterior limit of hoxd12a expression and is also the first of the nodal-independent somites specified during the late blastula (Fig. 9A; van der Hoeven et al.,1996; Morin-Kensicki et al.,2002; Szeto and Kimelman,2006).


Distinct anlagen for the anterior trunk, posterior trunk, and tail somites are specified in the late blastula before gastrulation (Szeto and Kimelman,2006). These subdivisions in the somite primordia are reflected in the loss of portions of the paraxial mesoderm when various t-box genes, nodals, wnts, or fgfs are perturbed (reviewed in Holley,2006a). The transitions from the anterior trunk (somites 1–9), posterior trunk (somites 10–15), and tail (somites 16–30) are also reflected in differences in the development and segmentation of these regions of the body axis, with the most prominent differences occurring at the progression from anterior to posterior trunk.

Three genes are known to be exclusively expressed in the anterior somites, tbx15, a nanos-related gene, and soxlla (de Martino et al.,2000; Begemann et al.,2002; Jülich et al.,2005a). Notably, there is no known function for any of these genes in somitogenesis, and none of the genes that show anterior- or posterior-specific somite defects are differentially expressed along the body axis.

In several respects, the development of the anterior trunk somites occurs more synchronously than the more posterior segments. The segmental expression of myoD, snail1a, and engrailed arises simultaneously in the first roughly six somites at the six to seven somite stage, while expression of these genes in the remaining somites arises sequentially with the formation of each segment (Hatta et al.,1991; Ekker et al.,1992; Hammerschmidt and Nüsslein-Volhard,1993; Thisse et al.,1993; Weinberg et al.,1996; Jülich et al.,2005a). These differences in gene expression mirror the morphogenesis of the adaxial cells. The adaxial cells begin to rearrange simultaneously in the anterior ∼6 somites at the 7–10 somite stage, but rearrange sequentially in the posterior somites after segment morphogenesis is initiated (Felsenfeld et al.,1991; van Eeden et al.,1996). As discussed above, the adaxial cells are capable of generating a segmental pattern in the absence of fss or notch pathway signaling (van Eeden et al.,1998; Henry et al.,2005). Therefore, there is a greater temporal separation between this secondary, adaxial cell-mediated segmentation and initial border morphogenesis during anterior somitogenesis, relative to posterior somitogenesis, where adaxial cell rearrangement appears more as a continuation of somite border formation and epithelialization.

fss/tbx24, foxc1a, and ripply1 are required for formation of somites along the entire body axis, but the formation of the anterior five to seven somites is generally resistant to perturbation (van Eeden et al.,1996; Topczewska et al.,2001a; Kawamura et al.,2005a). Mutations in the notch pathway genes bea/deltaC, aei/deltaD, des/notch1a, and mib do not affect the anterior somites (van Eeden et al.,1996; Holley et al.,2000,2002; Itoh et al.,2003; Jülich et al.,2005b). The formation of the anterior somites in these mutants is not due to functional redundancy between these genes or to maternal contribution (van Eeden et al.,1996; Jülich et al.,2005b). The anterior somites are also resistant to dominant perturbation of notch signaling by means of injection of low levels of a dominant negative Su(H) or a constitutively active form of notch (Jülich et al.,2005a). Moreover, treatment of embryos with a chemical inhibitor of notch signaling, DAPT, does not affect the anterior somites (Geling et al.,2002; Mara et al.,2007). Inhibition of non-notch pathway genes also tends to affect all but the anterior trunk somites. Morpholino knockdown of RPTPψ does not affect the formation of the first seven somites. The anterior trunk somites are also normal when RPTPψ function is inhibited in aei/deltaD embryos (Aerne and Ish-Horowicz,2004). Additionally, while fss/tbx24−/− embryos typically lack all somites, there are alleles of fss/tbx24 that form one or two somites (unpublished observations). Collectively, these phenotypes suggest that the genetic control of anterior somitogenesis differs in some way from the control of posterior somitogenesis. fss/tbx24−/+;aei/deltaD−/+;des/notch1a−/+ embryos display defects centered around somites 7–9, suggesting that the transition from anterior trunk to posterior trunk somitogenesis is particularly sensitive to the dosage of these genes (Jülich et al.,2005a). Only two zebrafish mutants are known that specifically affect the anterior somites. integrinα5−/− and fibronectin1a−/− embryos fail in the morphogenesis of the anterior trunk somites (Jülich et al.,2005a; Koshida et al.,2005). Double mutants between integrinα5 and the notch pathway mutants display defects along the entire body axis (Jülich et al.,2005a). One possible explanation for the difference in anterior somite formation is that, at the time that the first few somites are forming, the paraxial mesoderm is still undergoing dorsal convergence, and the tissue is relatively shallow along the dorsal–ventral axis and broad along the medial–lateral axis. Thus, aspects of somite formation may be delayed or regulated differently to allow for the continuation of dorsal convergence.

Although not affected by inhibiting notch signaling, anterior trunk somitogenesis can be perturbed by blocking the function her genes, either alone or in concert with other genes. Morpholino knockdown of her1 has been reported to cause, at most, a very mild perturbation of the anterior trunk somites, and there is debate about the effect, if any, inhibition of her1 has on the posterior trunk and tail somites (Holley et al.,2002; Oates and Ho,2002; Sieger et al.,2006). Inhibition of her7 affects somites posterior to somites 7–10, but knockdown of both her7 and deltaC affects all somites (Oates and Ho,2002; Oates et al.,2005a). Elimination of her1 and her7 also affects all somites (Henry et al.,2002; Oates and Ho,2002). In other genetic combinations, her1 and her7 do not behave equivalently. Knockdown of her13.2 affects the posterior somites, but knockdown of both her1 and her13.2 affects all somites and abolishes oscillating expression of her1, her7, and deltaC by the tail bud stage. In contrast, knockdown of her7 or both Su(H) paralogues in conjunction with her13.2 or knockdown of her13.2 in bea/deltaC mutants does not affect formation of the anterior somites (Kawamura et al.,2005b; Sieger et al.,2006). Knockdown of her1 with both Su(H) genes affects oscillating gene expression from the beginning of somitogenesis (Sieger et al.,2006). Why does elimination of notch target genes in combination with other genes perturb anterior somitogenesis while double mutants among notch pathway genes themselves, or chemical inhibition of notch signaling, do not affect the first few somites? Perhaps these results suggest that there are other signals that activate the expression of the her genes during the first rounds of segmentation.

There are conflicting reports regarding the phenotype of the embryos lacking both Su(H)1 and 2. The initial study focused on Su(H)1, but the morpholino used in this study also effectively inhibits Su(H)2 due to high sequence homology around the translational start site. The authors found that injection of this morpholino abolished oscillating expression of her1, her7, and deltaC and that, like the notch pathway mutants, the segmentation defect started around the 7th—9th somite (Sieger et al.,2003,2006). A second study used two morpholinos, one for each Su(H) paralogue. They observed a similar effect on gene expression but found that the anterior somites were perturbed (Echeverri and Oates,2007). This difference may be due to different efficacies of the morpholinos or to the genetic background in which the experiments were performed. However, it would be curious if knockdown of the two Su(H) paralogues affects the anterior somites, while otherwise inhibiting notch signaling does not. The answer to this puzzle may be that the perturbation of the anterior somites is due to elimination of the repressive function of Su(H), which operates in the absence of notch signaling. That basal levels of her1 and her7 expression are higher in the Su(H) morphant than in the des/notch1a embryos, indicates that the Su(H) paralogues do function as default repressors in the PSM (Sieger et al.,2003).

That the anterior somites form in the notch pathway mutants has been explained as a gradual desynchronization of the clock over the first few somite cycles (Jiang et al.,2000). This model correlates well with the gradual breakdown in expression of her1, her7, and deltaC, but as discussed above, is inconsistent with many other aspects of the notch mutant phenotypes. Implicit in this model is the idea that all of the somite precursors begin synchronized oscillations at the same time, early in development. However, a detailed examination of her1, her7, and deltaC expression in the posterior somite precursors strongly suggests that the expression of these genes do not oscillate in the progenitor zone. The precursors to the posterior ∼18 somites appear to initiate oscillations at different times as the cells leave the medial progenitor zone into the lateral initiation zone (Mara et al.,2007).

As somitogenesis proceeds, the tail bud shrinks in size such that two to three her1 stripes are typically seen in a 7–8 somite stage embryo, while most 15 somite stage embryos have only one to two stripes (Müller et al.,1996; Holley et al.,2000). At the 7–8 somite stage, the anterior and posterior PSM combined have the anlagen for roughly 10 future somites. By the 12 somite stage, the PSM has the precursors for ∼8 somites and by the 15 somite stage the PSM has shrunk to less than 7 future somites. The decease in the number of cells in the PSM may make the system more sensitive to noise generated by mitosis, cell movement, and stochastic gene expression. Thus, any perturbation of the genetic circuitry of the clock may have a proclivity to cause segmentation defects during posterior trunk and tail somitogenesis.

Differences between the anterior approximately seven somites and the more posterior segments are also observed in amphioxus, Xenopus, mouse, and human. Genetic perturbation of notch signaling in mice and humans (Conlon et al.,1995; Oka et al.,1995; Hrabé Angelis et al.,1997; Wong et al.,1997; Evrard et al.,1998; Kusumi et al.,1998; Zhang and Gridley,1998; Bulman et al.,2000; Bessho et al.,2001; Dunwoodie et al.,2002) or wnt3a (Takada et al.,1994; Aulehla et al.,2003) and mesp2 (Saga et al.,1997) in mice leads to a somite defect in the posterior but not the anterior somites. Mice mutant for either of the transcription factors mesogenin or tbx6 form only the anterior paraxial mesoderm, revealing genetic differences in the specification of anlagen of the anterior somites (Chapman and Papaioannou,1998; Yoon and Wold,2000). Anterior-specific defects are observed in mice mutant for PDGFRα, which is believed to mediate signaling between the myotome and sclerotome. Mice mutant for this receptor show extensive fusion of the cervical vertebrae but more subtle defects in thoracic and lumbar vertebrae (Soriano,1997; Tallquist et al.,2000). This phenotype is similar to congenital human defects known as Klippel–Feil syndrome, in which the cervical vertebrae are fused but the rib cage is only moderately affected, if at all (Clarke et al.,1998; Pourquié and Kusumi,2001). Anterior somitogenesis in the mouse and the cephalochordate amphioxus occurs more rapidly than posterior somitogenesis (Tam,1981; Schubert et al.,2001). Conversely, the rotation of the anterior seven to nine somites during Xenopus segmentation is slower than the rotation in the posterior somites (Afonin et al.,2006).


The left–right patterning of the vertebrate trunk establishes the asymmetric position/morphology of the heart, lung, liver, and other viscera (Raya and Belmonte,2006). Signals that establish or maintain left–right asymmetry also act on the paraxial mesoderm, and if these signals are not buffered in some way, somitogenesis will occur asymmetrically, with one side of the embryo forming somites more rapidly than the other. This asymmetry only effects somites ∼8–15 and is transient, as the bilateral alignment of the somites recovers later during the segmentation period (Kawakami et al.,2005b; Vermot et al.,2005). Neither the mechanisms of this recovery nor the exact cause of the initial asymmetry are fully understood, although it is clear that retinoic acid has a roll in balancing the bilateral rate of somite formation (Diez del Corral et al.,2003; Moreno and Kintner,2004; Kawakami et al.,2005b; Vermot et al.,2005; Vermot and Pourquie,2005; Echeverri and Oates,2007; Sirbu and Duester,2006).

raldh2 is the enzyme that catalyzes the last step in the biosynthesis of retinoic acid. During the segmentation period, raldh2 is expressed in the anterior PSM and somites in the zebrafish, Xenopus, mouse, and chick (Begemann et al.,2001; Diez del Corral et al.,2003; Moreno and Kintner,2004; Sirbu and Duester,2006). It has been proposed that retinoic acid promotes the transcription of bHLH genes in anterior PSM of the Xenopus embryo (Moreno and Kintner,2004). In the chick, somite explants treated with a retinoic acid agonist cease to express fgf8 and reciprocally, FGF8 represses the expression raldh2 in explants. These results led to the hypothesis that reciprocal gradients of retinoic acid and fgf exist in the PSM (Diez del Corral et al.,2003). In raldh2-deficient mid-somitogenesis stage embryos, fgf8 expression in the PSM is expanded anteriorly, particularly on the right side. The timing of the asymmetric fgf8 expression correlates with the appearance of asymmetric somite formation (Kawakami et al.,2005b; Vermot et al.,2005; Vermot and Pourquie,2005; Sirbu and Duester,2006). During the early somitogenesis stages in raldh2−/− mouse embryos, fgf8 expression is normal in the PSM but expanded anteriorly in the overlying neural ectoderm (Sirbu and Duester,2006). Moreover, segmentation in raldh2−/− mouse embryos can be rescued if retinoic acid is provided through the mother's diet up until the beginning of somitogenesis (day E8.25). Thus, retinoic acid is not required throughout the segmentation period for normal somitogenesis to occur, arguing that retinoic acid is not needed to continually regulate the fgf gradient or to induce gene expression in the anterior PSM. Moreover, using a retinoic acid responsive transgene, the raldh2−/− embryos rescued by dietary supplement, only showed evidence of retinoic acid signaling in the neural plate and cranial mesoderm not in the somites or PSM. These results suggest that retinoic acid produced in the paraxial mesoderm primarily acts to regulate fgf8 expression in the neural ectoderm and that deregulation of the neural expression of fgf8 may later affect the mesodermal expression of fgf8 in some way that is not understood (Sirbu and Duester,2006). This explanation contrasts with the model that opposing gradients of retinoic acid and fgf8 regulate somite maturation throughout somitogenesis. Obviously, more experiments are required to better understand the relationship between retinoic acid, left–right patterning and somitogenesis.

Perturbation of early establishment of left–right asymmetry by inhibiting H+/K+-ATPase activity, left–right dynein, notch signaling, or the transcription factor terra, leads to asymmetric somite formation (Kawakami et al.,2005b; Saude et al.,2005; Vermot and Pourquie,2005; Echeverri and Oates,2007). Similarly, inhibition of raldh2, by means of mutation, antisense, or pharmacology, results in asymmetric somitogenesis in the zebrafish, mouse, and chick (Kawakami et al.,2005b; Vermot et al.,2005; Vermot and Pourquie,2005; Sirbu and Duester,2006). In each case, asymmetric somite formation is preceded by a loss of bilaterally symmetric activity of the somite clock (Kawakami et al.,2005b; Vermot et al.,2005; Vermot and Pourquie,2005; Echeverri and Oates, 2007). Antisense inhibition of zebrafish cyp26a1, a retinoic acid catabolizing enzyme expressed in the posterior tail, also leads to asymmetric oscillation of the somite clock (Echeverri and Oates, 2006). When early left–right patterning is perturbed, the handedness of the asymmetry is random, meaning that half of the embryos show more somites on the right side and half show more somites on the left side (Kawakami et al.,2005b; Vermot and Pourquie,2005). In contrast, when retinoic acid synthesis is inhibited, the left side typically exhibits more somites (Kawakami et al.,2005b; Vermot et al.,2005; Vermot and Pourquie,2005). When both early left–right patterning and retinoic acid synthesis is perturbed, the handedness of the somite defects is randomized, indicating that retinoic acid acts to buffer somitogenesis against signals that establish left–right asymmetry (Kawakami et al.,2005b; Vermot and Pourquie,2005).


Among multicellular organisms, perhaps the best understood developmental process is subdivision of the Drosophila embryo along the anterior–posterior axis. However, unlike vertebrates where segmentation and axis elongation occur as the embryo grows posteriorly, the entire field of cells or body axis of Drosophila is laid down before segmentation. Long germ band insects like Drosophila do not have a posterior growth zone. Thus, it is perhaps not surprising that the genetic control of Drosophila segmentation has no apparent similarity to the control of somitogenesis. However, short germ band insects like Tribolium, which have a posterior growth zone, share extensive homology with Drosophila segmentation and have no known similarity to vertebrate somitogenesis (Sommer and Tautz,1993; Maderspacher et al.,1998; Choe and Brown,2007; Choe et al.,2006). As a counterpoint, in the spider, Cupiennius salei, which has a posterior growth zone, notch signaling is involved in segmentation, a mechanistic similarity to vertebrate somitogenesis but not Drosophila segmentation (Stollewerk et al.,2003; Schoppmeier and Damen,2005). It is not clear, however, if spider segmentation is regulated by a clock (Damen et al.,2000). There currently is not enough data to conclude if some arthropods use a segmentation mechanism homologous to that used in somitogenesis. Thus, we cannot conclude whether urbilateria, the last common ancestor of the protostomes (arthropods) and deuterostomes (vertebrates), was a segmented organism or if segmentation arose multiple times in animal evolution.

Amphioxus, a cephalochordate, has somites that express notch, wnt, hairy, and tbx orthologues (Holland et al.,2001; Schubert et al.,2001; Minguillon et al.,2003; Beaster-Jones et al.,2006). As yet, there is no evidence of a somite clock in amphioxus, and lunatic fringe is not even expressed in the mesoderm (Mazet and Shimeld,2003; Minguillon et al.,2003). Engrailed is only expressed in the anterior approximately eight somites of amphioxus (Holland et al.,2001). These anterior somites form by pinching off from the archenteron, a process called enterocoely, whereas the posterior segments directly bud off from a proliferative zone without going through a mesenchymal PSM intermediate stage. The anterior somites form much more rapidly than the posterior somites: one somite per hour for the former and one somite per 18 hr for the latter (Holland et al.,2001; Schubert et al.,2001). The first eight somites of amphioxus may be homologous to the anterior somites in vertebrates or may have given rise to the head mesoderm of higher vertebrates, an issue that remains unresolved (Kuratani et al.,1999; Beaster-Jones et al.,2006). The posterior somites of amphioxus form asymmetrically, with the left side preceding the right side (Schubert et al.,2001). Interestingly, the somites in the lamprey, Lampetra japonica, are not bilaterally symmetric and have irregular borders when they initially form, but this pattern refines over time such that bilateral symmetry is achieved (Kuratani et al.,1999).

Medaka, Oryzias latipes, is a small (2.5–3 cm), freshwater fish more closely related to Fugu (pufferfish) than to zebrafish. In Japan, there is an extensive history of research on medaka, and more recently, this organism has been used to study vertebrate development (Shima and Mitani,2004). Medaka make approximately 35 somites at a rate of 1 per hr at 26°C (Elmasri et al.,2004b). In a large-scale genetic screen, mutations in nine genes were found that disrupt somite formation. samidare mutants form the first six somites but not the posterior somites, while planlos and schnelles ende form only the first approximately two somites. bremser mutants do not form any somites. These mutants affect PSM patterning and segment polarity. In the other mutants, kurzer, orgelpfeifen, fusel, and zahnleucke, irregular somites form despite the presence of normal pattern in the PSM. These four mutants may have defects in the establishment or maintenance of segment polarity or in somite morphogenesis. The genes affected in the medaka mutants are not yet known (Elmasri et al.,2004b). Transgenic analysis of the medaka mesp-b enhancer identified putative Tbx and Su(H) binding sites that are required for appropriate expression in the PSM (Terasaki et al.,2006). An interesting observation about medaka somitogenesis is that the expression of the oscillating genes, her1/11, her7, and her5, resemble tetrapod not zebrafish oscillations (Elmasri et al.,2004a; Gajewski et al.,2006). The expression of these genes in the PSM of medaka, chick, and mouse typically consists of a single broad stripe or one broad posterior stripe and a one narrow anterior stripe. In the zebrafish, the oscillating genes are usually expressed in two to three stripes at the seven to eight somite stage, when the length of the PSM would be similar to that of the chick or mouse. This difference in pattern may be due to the rapid progression of zebrafish somitogenesis, 30 min per cycle, while medaka, chick, and mouse somite cycles are 60, 90, and 90–120 min, respectively (Tam,1981; Hanneman and Westerfield,1989; Palmeirim et al.,1997; Elmasri et al.,2004b). The existence of the extra stripe of expression in the zebrafish PSM is likely indicative of an oscillator running at higher frequency, which would facilitate the more rapid progression of zebrafish somitogenesis.

Xenopus laevis forms roughly 25 somites at a rate of one bilateral pair every 50 min. Unlike the other vertebrate models systems, Xenopus do not form epithelial somites. Rather, the PSM consists of elongated myofibers that are aligned perpendicular to the notochord. During segmentation, these myofibers rotate 90 degrees so that they align parallel to the notochord (Hamilton,1969; Afonin et al.,2006). This rotation is slower in the anterior seven to nine somites than in the more posterior somites (Afonin et al.,2006). The posterior, but not anterior, somites also include cells that ingress from the gastrocoele (Shook et al.,2004). Gain-of-function experiments indicate that alternating activation and repression of notch signaling is needed for Xenopus segmentation (Jen et al.,1997,1999). However, oscillations in gene expression have not been clearly demonstrated in the Xenopus PSM (Li et al.,2003).

Genetic studies in mice, chick, and humans demonstrate that the notch signaling pathway is required for somitogenesis and that oscillations in gene expression occur in mouse and chick PSM. While the details of somite formation cannot be studied in the human embryo, identification of the genes responsible for a heterogeneous class of congenital vertebral defects in humans called spondylocostal dysostosis, reveals similarities between human, mouse, and zebrafish segmentation (Fig. 2). There are 24 distinct mutations in the human notch ligand homolog dll3 responsible for ∼20–25% of cases of spondylocostal dysostosis (Bulman et al.,2000; Sparrow et al.,2002; Turnpenny et al.,2003; Whittock et al.,2004). Mutations in the human homolog of lunatic fringe (lfng), a glycosyltransferase that modifies Notch, cause some forms of spondylocostal dysostosis (Bruckner et al.,2000; Moloney et al.,2000; Sparrow et al.,2006). Mesp genes, bHLH transcription factors, are necessary for mouse somitogenesis and function along with the notch pathway to establish segment polarity (Saga et al.,1997; Takahashi et al.,2000,2003). Recently, mutations in the human mesp homologue were shown to cause some forms of spondylocostal dysostosis (Sparrow et al.,2006).

The function of snail genes in zebrafish segmentation likely differs from the role that snail homologues play in tetrapod somitogenesis. Snail proteins are zinc-finger transcription factors and have a well-characterized role in regulating cell morphology. For example, Snail promotes mesenchymal versus epithelial cell morphology by repressing the expression of E-cadherin (reviewed in Barrallo-Gimeno and Nieto,2005; Holley,2006b). snail2 and snail1 expression oscillates in the PSM of the chick and mouse, respectively. However, expression of the snail genes disappears before somite epithelialization, suggesting that snail may act to keep the cells of the PSM in a mesenchymal state. In support of this idea, overexpression of snail2 in the chick prevents cells from adopting an epithelial morphology and thus disrupts somitogenesis. These data suggest that snail expression promotes a mesenchymal morphology and that its disappearance in the S0 is a prerequisite for somite epithelialization. Because snail oscillates, these genes also may link the mouse and chick segmentation clock with control of somite morphogenesis (Dale et al.,2006). There are three snail homologues expressed in the PSM of the zebrafish embryo, but our analysis of the expression of these genes using exon probes suggests that neither snail1a, snail1b, nor snail2 expression oscillates (Fig. 7; Nisha Tamhankar and Scott Holley, unpublished observations; Hammerschmidt and Nüsslein-Volhard,1993; Thisse et al.,1993; Kudoh et al.,2001; Clements and Kimelman,2005). Moreover, snail1a and snail2 expression persists in the posterior and anterior half of each somite, respectively, indicating that expression of these snail genes in zebrafish is not sufficient to prevent PSM cells from undergoing a mesenchymal to epithelial transition. It therefore appears that the regulation and function of snail genes in zebrafish somitogenesis differs from the role that snail homologues play in chick and mouse segmentation.

Another interesting difference in segmentation mechanisms among the model systems is the function of lunatic fringe (lfng). lfng expression oscillates in the mouse and chick (Forsberg et al.,1998; McGrew et al.,1998). lfng mRNA is seen in the S0 and the anterior of each somite in zebrafish and medaka, but expression does not oscillate in the PSM (Prince et al.,2001; Elmasri et al.,2004a; Qiu et al.,2004). Moreover, lfng is not expressed in the mesoderm of amphioxus (Mazet and Shimeld,2003). We have not observed a consistent segmentation defect in zebrafish embryos injected with lfng mRNA or a morpholino targeting the gene (unpublished observations). However, lfng is required for segmentation in the mouse and leads to constitutive activation of notch in the PSM (Evrard et al.,1998; Zhang and Gridley,1998; Morimoto et al.,2005). lfng gain of function studies in the mouse and chick perturb somite formation, but the results of specific experiments differ (Sato et al.,2002; Dale et al.,2003; Serth et al.,2003). The variation in results may be due to technical differences or to multiple or distinct functions of lfng in tetrapod somitogenesis. That lfng is not expressed in the mesoderm in amphioxus and does not oscillate in either the zebrafish or medaka, certainly suggests that the role of this gene in chordate segmentation has continued to evolve. These data suggest that lfng was co-opted by the segmentation program in early vertebrates as a segment polarity gene and that expression later came under the control of the clock in tetrapods.


The past decade has seen astounding progress in our understanding of vertebrate segmentation. During this time period, there has also been a marked shift in our perception of evolutionary conservation of developmental mechanisms. The homologies between invertebrates and vertebrates are so striking that it is now expected that developmental mechanisms among the vertebrate model systems will be virtually identical. Our current understanding of vertebrate segmentation certainly underscores the conserved roles of notch signaling and the clock in governing vertebrate segmentation. Thus, generally speaking, the mechanisms appear to be homologous. However, the data regarding lunatic fringe suggests that, while the general gene network governing segmentation may be conserved, the wiring of the network and the functions of orthologous genes within the network may vary. For other developmental processes for which we have a large number of studies in the different vertebrate model species, such as early induction and patterning of the germ layers, discordance in the data could be ascribed to significant differences in the morphology of the blastula or gastrula of the different species. In contrast, during somitogenesis in the zebrafish, mouse, and chick, the morphology of the tissue is very similar: a mesenchymal field of cells is converted into a series of epithelial spheres with a core of mesenchyme. This common tissue organization means that somitogenesis is well suited as a subject for understanding the plasticity of the genetic mechanisms that govern vertebrate development. Because understanding the evolution of gene networks will require more than compiling coding sequences, expression patterns, or expression profiles, this endeavor necessitates a mechanistic understanding of segmentation in multiple species. While the phylogenetic distances between the vertebrate models species may be greater than one would prefer for an evolutionary comparison, the ability to generate detailed mechanistic data in each system compensates for this shortfall. It seems likely that the combined efforts of the research community focused on vertebrate segmentation will lead to insights into the evolvability of fundamental developmental mechanisms.

Vertebrate segmentation is also of great general interest because it is a process governed by a clock. Biological clocks are central to governing the cell cycle, circadian rhythms, and neuronal differentiation. However, the segmentation clock is unique in the intertwining of spatial and temporal information. It is a temporal mechanism involving cell:cell signaling that creates a spatial pattern within a tissue. Vertebrate segmentation is, thus, a rich subject for computational modeling. As experimental methods continue to improve the spatial and temporal resolution of the data, more empirically substantiated models can be generated. Such advances will allow the study of complex and emergent phenomena in biology, that is, how the whole may be greater than the sum of its parts. A central question within systems biology is how does the combined function of many genes within a network lead to higher levels of organization? Does the interaction among specific genes have unexpected consequences that may influence, or perhaps, determine the major characteristics of the gene network or biological system? The dynamic, regulative capacity of cells in the PSM to organize themselves in space and time is a great subject for such analysis.

Somitogenesis is also a particularly good context to study the mechanisms of morphogenesis. Because somitogenesis is reiterative, one can observe multiple cycles of somite formation in a single individual, a distinct advantage when performing live embryo microscopy. From the movement of cells into the growing tail bud, to the mesenchymal to epithelial transition during somite morphogenesis and later differentiation of the somite derivatives, zebrafish embryos are excellent subjects for understanding cell biology in intact tissues. This area is the particular strength of the zebrafish with its optical transparency and single cell resolution. These embryological features combined with powerful genetics will likely lead to the zebrafish being the first vertebrate in which we have a complete understanding of its development from fertilization to establishment of the basic vertebrate bauplan.


I thank Stephen Devoto, Judith Eisen, and Paula Mabee for sharing published artwork and Tim Brend, Dörthe Jülich, and Andrew Mara for critical comments on the manuscript.