MAKING REVIEWS
THE OF THE SOMITE: MOLECULAR EVENTS IN VERTEBRATE SEGMENTATION
Yumiko Saga* and Hiroyuki Takeda
The reiterated structures of the vertebrate axial skeleton, spinal nervous system and body muscle are based on the metameric structure of somites, which are formed in a dynamic morphogenetic process. Somite segmentation requires the activity of a biochemical oscillator known as the somite-segmentation clock. Although the molecular identity of the clock remains unknown, genetic and experimental evidence has accumulated that indicates how the periodicity of somite formation is generated, how the positions of segment borders are determined, and how the rostrocaudal polarity within somite primordia is generated.
PARAXIAL MESODERM
A subpopulation of mesoderm that lies on both sides of the neural tube, which gives rise to somites.
CEPHALOCHORDATE
A subphylum of chordates that includes the amphioxus Branchiostoma, which has a notochord, dorsal nervous system and segmented trunk, but lacks characters such as complex paired sensory organs and a true brain.
*Division of Mammalian Development and Division of Early Embryogenesis, National Institute of Genetics, Yata 1111, Mishima 411-8540, Japan. Department of Biological Sciences, Graduate School of Science, University of Tokyo, Hongo 7-3-1, Bunkyo-ku, Tokyo 1130033, Japan. Correspondence to Y.S. e-mail: [e-mail]
Somites are transient segments of the PARAXIAL MESODERM that are present in developing CEPHALOCHORDATES and vertebrates. In many vertebrate species, such as the frog, chick, mouse and zebrafish, somites form as blocks of cells, which bud off in a highly coordinated fashion from the anterior end of the unsegmented PRESOMITIC MESODERM (PSM)1 (FIG. 1). The strict temporal and spatial regulation of somitogenesis is of crucial developmental importance, because segmental structures, such as the peripheral spinal nerves, vertebrae, axial muscles and early blood vessels, develop according to the periodicity of somite segmentation. Until recently, the mechanisms that create the periodicity of somite development were largely unknown, although theoretical models predicted the existence of a clock mechanism that was responsible for generating this periodicity. However, the identification of CYCLING GENES, the expression of which turns on and off, has provided evidence for an intrinsic oscillatory mechanism in PSM cells2 (BOX 1). Furthermore, it has recently been found that the Notch/Delta signalling pathway has a crucial role in establishing both temporal periodicity in the PSM and ROSTROCAUDAL POLARITY in somite primordia1. This was an exciting finding because segmentation in Drosophila melanogaster probably does not use the Notch/Delta signalling pathway3. Indeed, in addition to the morphological differences between the segmentation process of vertebrates and flies, vertebrate
homologues of fly SEGMENT-POLARITY GENES, such as engrailed (en)4, sonic hedgehog (shh)5 and wingless-related (Wnt)6 genes, do not have segmental expression patterns during embryonic development, indicating that flies and vertebrates have different segmentation mechanisms. After somites separate from the PSM (in a process called `initial segmentation'), somitic cells begin to differentiate into axial structures in response to signals derived from the surrounding tissues. The SCLEROTOME of each somite is subdivided into rostral and caudal compartments, and each compartment then re-fuses with its neighbour to form a vertebra. This `resegmentation' process recreates segment borders within the sclerotome to generate parts of two neighbouring vertebrae; the division reflects the pre-existing rostrocaudal polarity of the somite (BOX 2). Experimental and gene-expression data strongly indicate that the generation of somite periodicity and the establishment of rostrocaudal polarity takes place before segment-border formation. In this review, we focus on the molecular events in the PSM that control how the segmental property of the PSM is established and how the positions of somite borders are determined.
Vertebrate segmentation
Since the first description of somite development at the beginning of the nineteenth century, much of our
NATURE REVIEWS | GENETICS
VOLUME 2 | NOVEMBER 2001 | 8 3 5
2001 Macmillan Magazines Ltd
REVIEWS
understanding of this process has come from morphological observations and experimental manipulations of avian embryos7. Paraxial mesodermal cells, which derive initially from the PRIMITIVE STREAK and later in development from the TAILBUD, acquire a PSM fate as they are separately laid down on both sides of the neural tube. The PSM is then segmented in the anteroposterior direction during the process of initial segmentation. Soon after their formation, EPITHELIAL SOMITES become patterned in response to local signals that are derived from the surrounding tissues. The dorsal part of a somite differentiates into the dermomyotome, which maintains epithelial morphology, and the ventral part differentiates into the sclerotome. Newly formed somites are thought to have rostral and caudal compartments, which subsequently give rise to a sclerotome subdivision that provides the basis for a later resegmentation process (BOX 2). Graft-reversal experiments in the chick have shown that the rostrocaudal polarity of the somite is established in the PSM and is maintained independently of its orientation with respect to the environment 8. Further grafting experiments have indicated that somite borders form only when anterior and posterior somite compartments are juxtaposed with each other9. However, until recently, no molecular mechanism had been proposed to explain such experimental results.
Key molecules in vertebrate segmentation
Mouse embryo Rostral
Zebrafish embryo Rostral
Heart
Head
9 dpc
Tailbud
PSM Caudal
Caudal
*
PSM
14 hpf
Figure 1 | Somitogenesis in mouse and zebrafish embryos. Epithelial somites bud off sequentially from the rostral end of the presomitic mesoderm (PSM; marked by a red asterisk in the zebrafish embryo), while more PSM cells are supplied from the paraxial mesoderm in the caudal region of the tailbud (shown in the mouse embryo). The speed of somite formation is coordinated with the speed of elongation of the tailbud region. Somites are generated every 90120 min in the mouse, and every 30 min in zebrafish. Arrows show previously formed somite segment borders. dpc, days post-coitum; hpf, hours post-fertilization.
Box 1 | The presomitic mesoderm oscillator The initial finding by Palmeirim and colleagues that the expression of c-hairy1 (chick hairy homologue 1) in chick presomitic mesoderm (PSM) is periodic, is widely accepted as the mechanism by which the PSM generates somites of a similar size at regular intervals2 . Chick PSM cells oscillate in synchrony with their neighbours in terms of c-hairy1 RNA expression. These oscillations are bilaterally synchronous, and appear as anteroposterior waves of expression that sweep across the PSM approximately every 90 min (see figure, in which c-hairy1 expression is shown in purple), which equals one cycle of somite formation in the chick embryo. The expression of c-hairy1 is an intrinsic cell-autonomous property of the PSM and is independent of cell movement. However, c-hairy1 is unlikely to regulate its own transcription in this oscillatory cycle because inhibiting protein synthesis does not arrest cyclic c-hairy1 expression2. So, c-hairy1 production is an output rather than a key regulator of the oscillator, and the regulatory mechanism remains unknown. The other cycling gene, the mouse lunatic fringe (Lfng) gene, might act even further downstream than c-hairy1, as de novo protein synthesis is required for the oscillations in its expression30. S, somite.
Anterior SI S0 SI S0 SII SI S0 SII SI S0 Formed somites (S) Newly forming somite (S0) Neural tube
Tailbud
Posterior
90 min
Studies of knockout mice have contributed greatly to our understanding of somitogenesis. One important finding has been that the Notch signalling pathway is involved in vertebrate segmentation (FIG. 2). So far, somite defects have been reported in mutant mice in which important components of the Notch pathway have been disrupted, such as Notch1 (REF. 10), the two Notch ligands in mice -- delta-like 1 (Dll1)11 and deltalike 3 (Dll3)12 -- presenilin 1 (Psen1)13, lunatic fringe (Lfng)14,15 and Rbpsuh (recombining binding protein suppressor of hairless, also known as Rbp-j)16 (TABLE 1). In addition, mesoderm posterior 2 (Mesp2), a basic helixloophelix (bHLH) transcription factor, has been found to have a key role in vertebrate segmentation, and has been shown to interact with the Notch pathway in the PSM17. Forward-genetic analyses in zebrafish have also made an important contribution to this field. A largescale mutagenesis screen that was carried out in the mid-1990s successfully isolated several somite mutants -- so-called `fused somites (fss)'-type mutants (TABLE 2) -- that are defective in both segmentation and rostrocaudal somite patterning. The fss mutant itself seems to be different from the other members of this group, and is crucially important18, because in fss embryos, essentially no somites form and the defect is somitogenesis specific. By contrast, in other fss-type mutants, several anterior somites are formed, and extra defects are observed. For example, the mind bomb (mib) mutation makes cells unresponsive to Notch signalling in many other tissues19, and deadly seven (des) mutants show neurogenic abnormalities in the neural
PSM
836
| NOVEMBER 2001 | VOLUME 2
www.nature.com/reviews/genetics
2001 Macmillan Magazines Ltd
REVIEWS
Box 2 | Segmentation and resegmentation in amniotes In AMNIOTES, before the formation of metameric structures, mesenchymal cells have to experience two types of segmentation: one is initial segmentation, in which segment-border formation produces the epithelial somites, and the other is resegmentation, a process by which sclerotomal cells are further segregated into rostral or caudal compartments. Segment-border formation occurs at fixed intervals and continues until the end of the supply of PSM cells. Once somites are formed, somitic cells start to differentiate, depending on their position within the somite (see figure). Cells facing the SURFACE ECTODERM differentiate into the DERMOMYOTOME, which gives rise to the dermatome and myotome. Conversely, the medial cells differentiate into the sclerotome under the influence of the notochord. The second segmentation, called resegmentation, takes place only in the sclerotome. The rostral and caudal halves of somites segregate, and re-fuse with their neighbouring halves to form vertebrae36,48,49, owing to underlying gene-expression differences between rostral and caudal compartments. Embryonic manipulation and gene-expression studies have shown that rostrocaudal polarity is established in the anterior PSM before the initial segmentation. Somites of fish50,51 and amphibians52 consist mainly of the myotomal component, and sclerotomal cells differentiate at later stages of development, after segmentation.
Neural tube Dermomyotome Dermatome
SEGMENT-POLARITY GENE
PRESOMITIC MESODERM
Precursor unsegmented mesoderm, which generates somites on segmentation.
CYCLING GENE
A gene, the expression of which oscillates rostrocaudally in the presomitic mesoderm.
ROSTROCAUDAL POLARITY
The difference between the rostral (anterior) and caudal (posterior) halves of a somite that underlies a difference in future developmental fates.
Somite Notochord
A gene originally identified in Drosophila early development, the expression of which divides the embryo into units that are 14 segments wide.
SCLEROTOME
Sclerotome
Myotome Initial segmentation (border formation)
Mesenchymal cell mass located in the medial region of a somite, from which the axial skeleton derives
PRIMITIVE STREAK
Anterior Dermomyotome Sclerotome
C
R
C
R
C
R
C
R
C
R
C
R
C
Posterior
A longitudinal cleft formed on the surface of the amniote early embryo by a convergence of cells. At the onset of gastrulation, epiblast cells migrate towards and into the streak, and in so doing acquire mesodermal cell fate.
TAILBUD
Resegmentation Lamina
Rostrocaudal polarity
Tailbud region
Dorsal
Pedicle
Neural arch
The caudal end of the tail region at which gastrulation continues to generate precursors for the paraxial mesoderm and the neural tube.
EPITHELIAL SOMITE
Ventral
Vertebral body
Spherical epithelial structure made up of epithelial cells that differentiate from mesenchymal cells on segmentation.
AMNIOTE
Animal, such as reptile, bird or mammal, whose eggs contain an amnion a membrane that surrounds the embryo and helps retain fluids.
SURFACE ECTODERM
plate20, indicating that the responsible genes are likely to encode components of the Notch signalling pathway. Indeed, the after eight (aei) mutant, which has NEURONAL HYPERPLASIA in addition to the somite phenotype, was found to be mutant for deltaD (a zebrafish homologue of Delta)21. These results confirm that the Notch pathway has a central role in vertebrate somitogenesis, as predicted by analyses of knockout mice.
Subdivision of the presomitic mesoderm
The outermost germ layer of the embryo that develops during gastrulation. Also the cell layer that covers the paraxial mesoderm, from which several diffusible factors are secreted that induce somitic cells to take on the dermomyotome fate.
Expression analyses of segmentation genes in normal and mutant embryos of both mice and zebrafish show that the PSM can be divided into at least two distinct regions, region I and region II, which correspond to two distinct cellular states -- state I and state II (FIG. 3). Although the size of each region varies between animals, region I corresponds approximately to the caudal twothirds of the PSM, including the tailbud, whereas region
II covers the anterior PSM, which is the width of several somites and lies posterior to the last somite formed. In a wild-type mouse embryo, Dll1 is expressed uniformly and at a high level in region I. By contrast, Dll1 expression is downregulated in region II and becomes restricted to the presumptive caudal half of the somite, showing a segmental expression pattern22. Interestingly, in Psen1 homozygous knockout mice, the stripe of Dll1 expression in region II disappears, leaving expression in region I intact23. This indicates that Dll1 expression is dependent on Psen1 in region II, but not in region I. The expression of cycling genes, such as mouse Lfng24 and zebrafish hairy-related 1 (her1; REFS 21,25), is also differentially regulated in the PSM. The expression domain of these cycling genes appears first in the tailbud, and is subsequently propagated through the posterior PSM. When it reaches the anterior PSM, it becomes stabilized, and is localized to either the rostral or caudal
NATURE REVIEWS | GENETICS
VOLUME 2 | NOVEMBER 2001 | 8 3 7
2001 Macmillan Magazines Ltd
REVIEWS
segments when cells enter the transient zone. In the somitomeric region, Notch signalling seems to promote the expression of Thylacine1 (a putative Xenopus homologue of Mesp2), and Delta2 in a positive feedback loop that specifies the rostral half of the segment. In addition to differentially regulated genes, there are several other genes, the expression of which is specific to each region. For example, Mesp genes are expressed only in region II17, whereas mesogenin, which belongs to a bHLH family that is similar but distinct from the Mesp family, is expressed only in region I27, such that Mesp and mesogenin expression domains are mutually exclusive. Furthermore, several observations indicate that anteroposterior differences in cell states exist within the PSM. For example, cellular morphology is different between anterior and posterior PSM regions: dorsal and ventral epithelialization of the PSM is evident in the anterior PSM, whereas posterior PSM cells show mesenchymal cellular morphology 28. Recently, Dubrulle and colleagues 29 reported that anterior but not posterior PSM cells become determined with respect to their future segmental identity29. To define the point at which rostrocaudal polarity is irreversibly determined, Dubrulle et al. carried out rostrocaudal inversions of regions of the PSM in chick embryos that are one somite in length. They found that the transition from the undetermined to the determined state takes place at a particular position in the PSM, about halfway from the end of the tailbud. This transition is accompanied by changes in gene expression, such as by the downregulation of expression of fibroblast growth factor 8 (FGF8). Although the transition zone could be included in region I, PSM cells in region I are largely maintained in an undetermined and immature state. So, a segmentation programme that leads to border formation is suspended in region I, and initiates when PSM cells enter the anterior PSM (region II).
Events in region I
Cytoplasm DII3 Furin Nucleus Notch1 Presenilin
Rbp-j ICN Rbp-j DII1 Kuzbanian Golgi Lfng bHLH genes
Figure 2 | Components of the Notch signalling pathway that might be involved in somitogenesis. Extracellular regions of the Notch receptor (Notch1) interact with Delta ligands (Dll1 or Dll3) to activate the Notch cascade. On activation, the extracellular domain of Notch1 is released by the protease Furin61, and the membrane-bound intracellular domain of Notch1 (ICN) is further processed by presenilin 1 (Psen1)62,63. The ICN interacts with Rbp-j and enters the nucleus to activate downstream target genes, such as the bHLH family of transcription targets64. Delta ligands might be processed by the metalloprotease Kuzbanian65. Lunatic fringe (Lfng) is known to act in the Golgi as a glycosyltransferase enzyme that modifies the extracellular domain of Notch1 (REFS 66, 67). (bHLH, basic helixloophelix; Dll1, Delta-like 1; Dll3, Delta-like 3; Rbp-j, recombining binding protein suppressor of hairless.) (Modified with permission from REF. 3.)
part of the future somite. Intriguingly, only the stabilized band of her1 expression is affected in zebrafish fss mutants, whereas the posterior two bands show normal cyclical patterns, indicating that the regulatory mechanisms that maintain her1 expression differ in regions I (fss independent) and II (fss dependent)20,21. Expression of her1 is propagated in a manner that is dependent on Notch signalling21. It has been shown in Xenopus laevis that the PSM can be divided into three regions -- the tailbud domain, transient zone and somitomeric region -- and different factors might regulate the Notch pathway in each region. According to Jen and colleagues26, Notch signalling is active when PSM cells are in the tailbud domain, but is repressed in the posterior-half
Table 1 | Knockout mice with somite defects
Gene
DERMOMYOTOME
No signs of segment specification can be detected at a molecular or cellular level when PSM cells are in region I. But cyclical gene expression that reflects a segmentation clock occurs in this region (BOX 1).
Protein function Receptor Ligand Ligand Glucosyl transferase Nuclear protein Transcription factor -secretase? Transcription factor Irregular No segment border No segment border No segment border Irregular No segment border No segment border No somite posterior to the cervical region
Knockout phenotype Segment border formation Rostrocaudal polarity Not clear No polarity Random Random Not clear Caudalized Rostralized No polarity
References 10 11 12 14,15 6 17 13 68
Epithelial cell layer in the dorsolateral region of the somite that faces the ectoderm and further differentiates into the most dorsal dermatome, which later differentiates into dermis and myotome future skeletal muscles.
NEURONAL HYPERPLASIA
Notch1 Dll1 Dll3 Lfng Rbp-j Mesp2 Psen1 Msgn1
Excessive formation of neuronal tissues due to transdifferentiation as a result of defects in Notch signalling .
Dll1, Delta-like 1; Dll3, Delta-like 3; Lfng, lunatic fringe; Mesp2, mesoderm posterior 2; Msgn, mesogenin; Psen1, presenilin 1; Rbp-j, recombining binding protein suppressor of hairless.
838
| NOVEMBER 2001 | VOLUME 2
www.nature.com/reviews/genetics
2001 Macmillan Magazines Ltd
REVIEWS
Table 2 | Zebrafish mutants with somite defects
Gene fss bea des mib (wit) aei Mutant name fused somites beamter deadly seven mind bomb (white tail) after eight Phenotype No segmentation First 34 somites formed, thereafter no somites First 79 somites formed, thereafter no somites First 79 somites formed, thereafter no somites First 79 somites formed, thereafter no somites Responsible gene ND ND ND ND deltaD References 21,25 21 20,21 20,21 21
ND, not determined.
Rostral Segmented Region II Region I
Caudal
Dll1 (mouse, chick)
Dll3 (mouse, chick)
Cycling genes
Stabilized
Cycling her1 (zebrafish)
Cycling c-hairy1 (chick) deltaC (zebrafish) Cycling Lfng (mouse, chick) deltaD (zebrafish) Stage-specific genes Mesp2 (mouse) MESO1 (chick)
mesogenin (mouse)
Figure 3 | Expression pattern of segmentation genes in the presomitic mesoderm. Two distinct regions, regions I and II, are shown by changes in the expression pattern of the following genes: Dll1 and Dll3; the cycling genes, her1, c-hairy1, deltaC, deltaD and Lfng; and the stage-specific genes, Mesp2, MESO1 and mesogenin. The expression of the cycling genes oscillates in region I, but becomes stabilized in either the caudal (c-hairy1 and deltaC) or rostral (Lfng and deltaD) compartment in somite primordia in region II. Because the size of the presomitic mesoderm varies between species and with developmental stages, this figure is a simplification based on an 11.5-dpc mouse embryo. Generally, mouse embryos have only two presumptive somites in region II, whereas 34 presumptive somites exist in chick, zebrafish and Xenopus embryos. Colour gradients indicate changes in the level of gene expression. (Dll1, Delta-like 1; Dll3, Delta-like 3; dpc, days post-coitum; her1, hairy-related 1; Lfng, lunatic fringe; Mesp2, mesoderm posterior 2.)
Over the past three decades, several models have been proposed to explain how somites are formed at regular intervals (BOX 3). Most of these models were proposed before molecular evidence for the existence of a segmentation clock and the involvement of Notch signalling in somitogenesis was obtained. It is now known that the molecules that have this cyclic RNA expression are components of the Notch signalling pathway: c-hairy1 2 and LFNG 2,30 in chick; Hes1 (hairy and enhancer of split 1; REFS 2, 31), Hey32/Hesr33 (hairy and enhancer of split related protein) and Lfng 24 in mouse; and Her1, DeltaC and DeltaD in zebrafish20,21,25. This indicates that Notch activity oscillates and sweeps up the PSM once per somite formation. One of the proposed functions of Notch signalling is to keep the oscillations of neighbouring PSM cells synchronized20. This idea comes from observations of gene expression in mutants in which Notch signalling is defective. In Dll1 mouse mutants, Lfng expression is very low and no propagation wave of gene activity is observed31. Similarly, in zebrafish aei/deltaD mutants, the expression of deltaC in PSM cells seems to be randomized, resulting in no cycling pattern20. It is possible that, through the Notch signalling pathway, oscillations in each cell can be communicated to neighbouring cells. However, the molecular nature of the oscillator and its regulation remain unknown. On the basis of the experimental results described above, Schnell and Maini developed a new theoretical model to explain how the oscillations are stabilized34. In this `clock-and-induction' model, Lfng has an inductive role. In Drosophila, Fringe acts to potentiate Notch activation by Delta and to inhibit Notch activation by the alternative ligand Serrate, therefore controlling the formation of the wing margin35. In Lfng mutant mice, somite formation and rostrocaudal patterning are disrupted14,15. So, Schnell and Maini speculated that the accumulated Lfng protein stabilizes Notch signalling and therefore arrests the segmentation clock, leading to segment-border formation in region II. However, this model cannot explain how the first few anterior somites are generated without Lfng in Lfng knockout embryos14,15. Formation of anterior somites is a general feature of embryos with defects in the Notch pathway20; a possible explanation of this phenomenon has been
NATURE REVIEWS | GENETICS
VOLUME 2 | NOVEMBER 2001 | 8 3 9
2001 Macmillan Magazines Ltd
REVIEWS
Box 3 | Previous models for somite segmentation In 1976, Cooke and Zeeman were the first to propose the existence of a cellular clock that might interact with a slowly progressing wave (wavefront) that moves along the presomitic mesoderm from anterior to posterior at a constant velocity53 (clock-and-wavefront model). The interaction between the wave and the oscillator was proposed to allocate cells to individual somites in a regular fashion along the anteroposterior axis. Although this model explains the control of the periodicity and cell-autonomous nature of c-hairy1 oscillation, it does not explain the formation of the anterior and posterior halves of a somite. Meinhardt proposed a reactiondiffusion model that involves two autocatalytic substances that behave as shortrange activators and long-range inhibitors54. The model assumes that the autocatalytic substances generate a spatial pattern, which results in a spatially homogeneous arrangement of cells that oscillate between two states that correspond to the anterior and posterior halves of a somite. Meinhardt's model is in agreement with two observations of Palmeirim and co-workers2: first, one full cycle of c-hairy1 oscillation corresponds to the formation of one somite, and second, c-hairy1 expression resembles the spatiotemporal dynamics of one of the autocatalytic substances (its propagating expression stops and is maintained in the posterior half of the somites). However, the model cannot explain the periodic abnormalities that are observed after chick embryos are exposed to a single heat shock55,56. In heat-shocked embryos, anomalies that are several somites wide seem to be separated by relatively constant distances of six to seven normal somites. The repeated anomalies indicated that heat shock might affect an oscillatory process within the somite precursors. The heat-shock phenomenon can be explained by the cell-cycle model57. This model relies on an intracellular oscillator, which controls cell division and interacts with a kinematic wave that produces a signal that recruits other cells in the vicinity shortly before segmentation. However, the length of the cell cycle, which is 9 h in chick, does not fit the 90-min cycling time of c-hairy1 expression, which is coincident with the formation of a single somite. Several other models have been proposed; the `wave gradient' model58, the `wave and cell polarization' model59 and a `clock and trail' model60. Some of these successfully account for several aspects of somitogenesis, but either fail to explain, or even contradict, other observations.
suggested by Jiang and colleagues20. They speculate that the PSM oscillator that controls somite patterning is initially synchronous in the PSM precursor cells, and if the cellcell communication through Notch signalling is defective, the cells will gradually drift out of synchrony until the lack of coordination causes somitogenesis to fail. Although this hypothesis can explain the normal formation of anterior somites, accumulating experimental evidence, as we describe below, strongly indicates that at least two distinct mechanisms might be involved in somite formation: one is a clock mechanism that operates in region I, and the other is a mechanism that arrests the clock and initiates border formation in region II. In the following two sections, we address how gene interactions are established in region II, and how the transition from region I to II is regulated.
The maturation processes in region II
is crucial for the maturation of the PSM. Unfortunately, the nature and function of Fss are at present unknown. Consistent with the phenotype of Mesp2 knockout mice, the expression of caudal-specific genes is expanded in fss mutants because no zebrafish mesp-b expression is induced25. Rostrocaudal polarity and resegmentation. At least in amniotes, the establishment of rostrocaudal polarity in region II is an essential step for later resegmentation. The rostral compartment of the somite gives rise to the caudal half of the vertebral body and intervertebral disc, whereas the caudal compartment generates the rostral half of the vertebral body and the pedicle of the neural arch36. It is known that, in mice, the level and pattern of Dll1 expression in region II prefigure the segmental features of vertebrae 23. As shown in FIG. 4a, Dll1 expression in region II in wildtype embryos is restricted to the caudal half of somite primordia, and this caudal expression is maintained after somite borders have formed. By contrast, in the embryos of Psen1 knockout mice, which show no caudal expression of Dll1 in region II but maintain normal Dll1 expression in region I, the vertebrae are rostralized. In the embryos of Mesp2 knockout mice, in which Dll1 expression in region II is expanded (FIG. 4a) , vertebrae are caudalized. In addition, in the embryos of Dll3 or Lfng knockout mice, Dll1 expression is randomized14,15 and the resulting vertebrae show mixed and randomized patterns with respect to their rostrocaudal identity. Consistently, inactivation of Uncx4.1 (a paired-type homeobox transcription factor, which seems to be a downstream target of
Having experienced a wave of Notch signalling in region I, PSM cells enter region II, where they acquire rostrocaudal polarity and become competent to segment (state II). State II is characterized by a change in the expression and regulation of key genes, such as an induction of members of the Mesp gene family. A principal role of Mesp in this region might be to downregulate Dll1 expression, because in Mesp2 knockout mice, rostral Dll1 expression fails to be suppressed17,23 (FIG. 4a). As a result, no Dll1 stripe is generated, the intense expression of Dll1 in region I continues and, because the rostral identity of the presumptive somite is lost, no segment border is formed. In zebrafish fss mutants, expression of nearly all key genes in region II is lost, which results in no segmentation21,25, indicating that fss
840
| NOVEMBER 2001 | VOLUME 2
www.nature.com/reviews/genetics
2001 Macmillan Magazines Ltd
REVIEWS
Dll1), results in the loss of the caudal component of vertebrae37,38, strengthening the idea that the expression of Dll1 in region II determines rostrocaudal patterning of the vertebrae. Although at the moment we cannot directly analyse the rostrocaudal polarity of vertebrae in Dll1-null mice because Dll1-null embryos die around 10.5 days post-coitum11, determining the regulation of Dll1 expression in region II is important for understanding how rostrocaudal polarity is established.
DII1 Uncx4.1 Vertebrae lm Wild type pd vb
HOMOTYPIC INTERACTION
Protein between interaction molecules of the same type.
a
tr tr
Mesp2 null
tr
Psen1 null
Pudgy (DII3 null)
b
SI
S0
State II Mesp2 Notch Notch DII1 Psen1 DII1 Notch Mesp2 Psen1 DII1
State I
Rostral
Caudal
Figure 4 | Genetic evidence for the requirement of Mesp2 in establishing rostrocaudal polarity. a | Expression of Dll1 and Uncx4.1 and vertebral morphology in wild-type and three mutant mice. Dll1 expression in region II and subsequent Uncx4.1 expression in the segmented region reflects the rostrocaudal properties of vertebrae. The pedicle (pd) of the neural arch is derived from the caudal half of the somite, which is defined by the expression of the caudal markers Dll1 and Uncx4.1. Mesp2-null embryos, which show expanded expression of caudal markers, generate completely caudalized vertebrae. By contrast, a rostralized phenotype is observed in Psen1-null embryos, which is characterized by the lack of both the caudal markers and the pedicles. Pudgy mice, which carry a point mutation in the open reading frame of Dll3, have a randomized phenotype with respect to the rostrocaudal polarity of somites and vertebrae. Arrowheads in Dll1 expression indicate the boundary between region I and II. b | Establishing rostrocaudal polarity in somite primordia (S). Dll1 expression is regulated by Mesp2 and Psen1 through two Notch signalling pathways. When Mesp2 is initially expressed in both prospective rostral and caudal regions, Mesp2 suppresses Dll1 in the entire somite primordium by suppressing the Psen1-dependent Notch pathway and by activating the Psen1-independent Notch signalling pathway. When Mesp2 expression becomes localized to the presumptive rostral half of the somite, Dll1 expression is induced in the caudal half by means of Psen1. Blue shading indicates Dll1 expression. The vertical arrow indicates the position of the next segmental border. SI, the last completely segmented somite; S0, the next somite to be segmented. (Dll1, Delta-like 1; Dll3, Delta-like 3; lm, lamina; Mesp2, mesoderm posterior 2; Psen1, presenilin 1; tr, transverse process; vb, vertebral body.)
Mesp2 and the Notch signalling pathway. The expression pattern of Dll1 in region II is established by Mesp2 through the Notch signalling pathway23. Dll1 expression is stronger in region I, but it is quickly downregulated by Mesp2 when cells enter region II. It seems that this effect of Mesp2 is mediated by the Notch signalling pathway, because Mesp2 knockout embryos show reduced levels of Notch1 expression17. In addition, when a dominant-active form of Notch1 -- Notch-IC, which lacks the extracellular domain -- is introduced into the Mesp2 locus, Dll1 expression is reduced in region II, even in the absence of Mesp2 (REF. 23) . It is thought that Dll1 induction itself depends on Notch signalling, because Psen1 knockout embryos completely lack Dll1 expression in region II23. It has been proposed that there are two Notch signalling pathways that regulate Dll1 expression in region II23: one is Psen1 dependent and is involved in inducing Dll1 expression, and the other is Psen1 independent and is involved in inhibiting Dll1 expression. Mesp2 might stimulate the inhibitory pathway and suppress the induction pathway (FIG. 4b). In Notch1 knockout mice10, both pathways might be defective, and the phenotype (uncoordinated segmentation with rostrocaudal polarity39) is, therefore, difficult to understand. It is noteworthy that the rostral restriction of Mesp2 expression is crucial for generating the striped pattern of Dll1 expression in region II. Initially, Mesp2 expression is induced in both the prospective rostral and the prospective caudal halves of somite primordia, suppressing Dll1 expression through the Psen1-independent Notch signalling pathway. Subsequently, Mesp2 expression becomes restricted to the rostral half of the somite primordia and Dll1 suppression is maintained only here. In the caudal half, however, Dll1 expression is induced through the Psen1-dependent Notch signalling pathway. Therefore, Mesp2 and Psen1 knockout mice show contrasting phenotypes, and double-knockout Mesp2/Psen1 mutants phenotypically resemble the Mesp2 knockout mouse23; lack of Mesp2 results in an initial failure to suppress Dll1, irrespective of the presence or absence of Psen1. These results show that the rostral restriction of Mesp2 expression is crucial for creating rostrocaudal polarity in somite primordia. The mechanism that restricts Mesp2 expression to the rostral half is not known, but Psen1-dependent Notch signalling must be involved, as no rostral restriction of the Mesp2 expression domain is observed in Psen1 knockout mice 40. As shown in FIG. 4b, the complicated gene network that is centred on Mesp2 is established in region II. The downstream target of Mesp2 has not yet been identified, but many genes expressed in the presumptive rostral compartment are affected in Mesp2-null embryos, including Notch2, cerberus 1 and the Fgf receptor 1 gene (Fgfr1). In addition, the paraxial protocadherin gene (PAPC) has been implicated as a downstream gene of Thylacine1 (frog homologue of Mesp2)41 and mespb25 (zebrafish homologue of
NATURE REVIEWS | GENETICS
VOLUME 2 | NOVEMBER 2001 | 8 4 1
2001 Macmillan Magazines Ltd
REVIEWS
Mesp2)25. The disruption of PAPC activity in frog embryos leads to alterations in somite morphology, a lack of maintenance of rostrocaudal polarity and a lack of segment-border formation41. PAPC might function to prevent cell mixing across the rostrocaudal boundary, because PAPC is a very potent HOMOTYPIC cell-adhesion molecule. Therefore, in addition to establishing rostrocaudal polarity, Mesp2 might function as a regulator of segment-border formation by regulating a cell-adhesion molecule.
Transition from an immature to a mature state
a Activation of MAPK
Expression of fgf8
b
Rostral Region II
Control + SU5402
Control + FGF bead
Region II
Region II
Region II Small somite
Large somite
Fgf8 bead
Region I
Region I
Caudal
Figure 5 | Manipulation of an Fgf signal alters somite size in zebrafish. a | The pattern of MAPK activation (left) closely resembles that of fgf8 expression (right) in 10somite zebrafish embryos. The embryos were immunostained with an antibody against an activated (di-phosphorylated) form of ERK (a vertebrate MAPK). Lateral views are shown. The strong activation is detected in the posterior two-thirds of the PSM (arrow). The activation domain roughly corresponds to region I. Weak activation is detected in newly formed somites (arrowhead). (Modified from REF. 42.) b | Schematic representations of the effect of Fgf signalling on somite-border formation. The expression of her1, which is cyclic (grey) in the presence of higher Fgf signalling, becomes stabilized (purple) as Fgf levels decline in region I. The stabilized her1 marks a presumptive future somite border (dashed line). When an Fgf signal is transiently compromised by SU5402, a kinase inhibitor that might be specific to all types of Fgf receptors, waves of her1 expression prematurely terminate in the intermediate PSM instead of in the anterior PSM, due to acceleration of PSM maturation. The posterior shift in her1 termination is accompanied by a posterior shift in the expression of segmentation genes, such as a zebrafish mesp gene42. As a result, the point of segment formation moves towards the tailbud, giving rise to a larger somite (left panel). By contrast, an exogenous Fgf signal that is provided by Fgf-soaked beads maintains the PSM in an immature state for a longer period and, therefore, causes anterior expansion of region I. Eventually, her1 expression reaches further to the anterior than it does on the control side, and a somite border forms more anteriorly, resulting in a smaller somite (right panel). (fgf8, fibroblast growth factor 8; her1, hairy-related 1; MAPK, mitogen-activated protein kinase; mesp, mesoderm posterior; PSM, presomitic mesoderm.)
What determined the point of the transition from state I to state II in the PSM remained unknown until recently. However, two papers29,42 have recently shown that Fgf signalling -- especially when mediated by FGF8 -- functions as a crucial positional cue that determines the position of the state transition. Signalling mediated by Fgfr had previously been implicated in somitogenesis, as Fgfr1 is expressed in the PSM and in the anterior part of segmented somites in mice43 and zebrafish25. However, the embryos of Fgfr1 knockout mice have disturbed segment borders44, whereas Fgf8 mutant mouse embryos do not proceed through gastrulation45, and zebrafish fgf8 mutant embryos do not show severe somitogenesis phenotypes46. So, the precise role of Fgf signalling in segmentation had been unclear. A breakthrough came when Dubrulle and colleagues29 reported that chick FGF8 is expressed in the caudal PSM in a graded fashion that extends to region I. Similarly, in zebrafish, Sawada and colleagues42 showed that there is a rough correlation between region I and the activated domain of Fgf signalling, as indicated by the phosphorylation of mitogen-activated protein kinase (MAPK) -- one of the main downstream targets of Fgf signalling47. The activation pattern of MAPK closely resembles the expression pattern of zebrafish fgf8 (FIG. 5a). The pattern of Fgf8 expression and MAPK activation indicates a possible role for Fgf signalling in the state transition of the PSM. Indeed, when FGF8 is misexpressed in the entire PSM of chick embryos, the caudal PSM marker brachyury is upregulated in the rostral PSM, and segmentation is suppressed, indicating that FGF8 maintains the caudal identity of the PSM29. Furthermore, a transient manipulation of FGF signalling in chick and zebrafish embryos alters the size of the somites, because it has been found that the transient inhibition of FGF signalling with SU5402, a general inhibitor of FGF signalling, results in the formation of larger somites, whereas transiently activating FGF signalling by implanting FGF8-soaked beads into the PSM causes smaller somites to form (FIG. 5b). Detailed analyses of FGF-manipulated chick embryos have shown that FGF levels do not affect the oscillation frequency but regulate the position of the state transition29. The altered pattern of her1 waves of expression in manipulated zebrafish embryos further strengthens this idea42. These results indicate that PSM cells are maintained in an immature state when they are in the posterior PSM because of high levels of Fgf signalling. Once they reach region II, which is devoid of Fgf signalling, a maturation programme is initiated, which includes the induction of Mesp genes. Importantly, as development proceeds, the Fgf activation domain gradually moves back at a constant speed, indicating that Fgf signalling might function as a positional cue within the PSM. The analysis of zebrafish mutants has shown that Fgf signalling functions independently of Notch and Fss activity42. Therefore, Fgf signalling might directly control the transition from state I to state II in the PSM.
842
| NOVEMBER 2001 | VOLUME 2
www.nature.com/reviews/genetics
2001 Macmillan Magazines Ltd
REVIEWS
front' model (BOX 3), in which the transition point between region I and region II could function as a wavefront. As PSM cells pass this front, they acquire the ability to arrest the oscillation and to induce the expression of segmentation genes. This transition from state I to state II is controlled by the level of Fgf signalling. The high level of Fgf activation in the posterior PSM maintains PSM cells in an immature state, whereas low Fgf levels accelerate the maturation process of PSM cells in the anterior PSM. The oscillation wave travels through the posterior to the intermediate PSM and, as it enters the anterior PSM, the wave becomes Fss-dependent and stabilized (FIG. 6). It is probable that only a group of cells that are in a certain phase of the oscillation might be able to initiate a segmentation programme, which assures spatial periodicity of boundary formation. Because the Fgf activation domain retreats as somitogenesis proceeds, the wavefront gradually moves back. So, a periodic interaction between the wavefront and the oscillation wave could create a regularly spaced somite border.
Conclusion and future questions
Level of Fgf signalling
A model for vertebrate segmentation
The events that lead to somite-border formation can be subdivided as follows. First, PSM cells, which have been maintained in an immature state in the posterior PSM, sequentially enter state II in the anterior PSM, where they mature and become competent to segment. Second, the oscillation wave that appears in the tailbud travels through the posterior to the intermediate PSM, and this process depends on Notch signalling. Third, the oscillation wave is stabilized in the anterior PSM (region II), and the segmentation programme initiates. In zebrafish, these processes depend on Fss activity. Through their interaction with the Notch pathway, Mesp genes have a crucial role in the initiation of the segmentation programme and in the establishment of rostrocaudal polarity. Fourth, the level of Fgf signalling determines the point of transition from state I to state II. All these events can be integrated into a model (FIG. 6) that is largely consistent with a `clock-and-wave-
Wavefront
Region II
Region I
Maturation
Immature state Oscillator Fss (zebrafish)
Induction of segmentation genes (e.g. Mesp)
Periodic activation of Notch signalling
Establishment of rostrocaudal polarity
Stabilization of oscillation
Establishment of periodicity
Segment-border formation
Figure 6 | A model of somite segmentation. Already-segmented somites and determined somites that have yet to be segmented are indicated by open rectangles and dotted rectangles, respectively. Anterior character within segmented and presumptive somites is shown in dark blue. In region I, a high level of Fgf signal maintains PSM cells in an immature state and, through Notch signalling, the oscillation in each cell might be translated into periodic expression of cycling genes. Reducing Fgf signalling is one of the key events for PSM cells in their transition from an immature to a mature state. The position of the state transition could function as the wavefront. PSM cells initiate a segmentation programme and an induction of segmentation genes when cells in a certain phase of the oscillation encounter the wavefront. As development proceeds, the activation domain of Fgf signalling retreats and the wavefront moves back through the PSM. As somitogenesis proceeds, the wavefront interacts with the oscillation wave at regular intervals, resulting in the periodic induction of segmentation genes. (Fgf, fibroblast growth factor; PSM, presomitic mesoderm.)
Accumulating experimental and genetic data lead us to conclude that a molecular clock and wavefront activity establishes the periodic pattern of the PSM. However, there are many questions still to be answered. We still do not know the molecular identity of the clock. An endogenous oscillator might exist within cells and might function as, or be translated into, a segmentation clock. Alternatively, there could be no oscillator in the cell, and a cyclic wave could be generated by interactions between two kinds of molecule -- activators and inhibitors -- as proposed by Alan Turing's `reactiondiffusion' model (BOX 3). In this model, the oscillation wave travels when the inhibitor is increased and is stabilized when the activator increases. Either way, Notch signalling must be involved in the generation, propagation and stabilization of the cyclic wave. At present, we cannot conclude which mechanism is used in vivo, because no experimental data strongly support either model. Further analyses of the regulatory mechanisms of cycling genes will shed light on the molecular identity of the clock. The interaction between the oscillation wave and wavefront in region II is still awaiting molecular analysis. The formation of the wavefront might depend on Fss, and one of the key molecules in region II is likely to be Mesp2. Zebrafish fss mutants do not form a single somite, whereas the embryos of Mesp2 knockout mice form at least several anterior somites, like other zebrafish fss-group mutants (TABLE 1). These observations indicate that Mesp2 is only one of the targets of Fss. Therefore, a promising study will be to clone fss and identify the downstream targets of Fss and Mesp. Another aspect that deserves to be addressed is the relationship between the establishment of rostrocaudal polarity and segment-border formation. In fact, it
NATURE REVIEWS | GENETICS
VOLUME 2 | NOVEMBER 2001 | 8 4 3
2001 Macmillan Magazines Ltd
REVIEWS
is very difficult to uncouple these two events experimentally because most knockout mice simultaneously show both defects. Normally, the establishment of rostrocaudal polarity precedes border formation, and the former might therefore be required for the latter. We need more mutants to discriminate between these two events. Notch signalling is involved in several steps in somitogenesis: synchroniziation of oscillation in region I; suppression and activation of Dll1 expression in the presumptive rostral half and caudal half of the somite primordia in region II; and re-shaping the Mesp2 expression domain. There must be many unidentified molecules that modulate the Notch signalling pathway to differentiate between these Notch activities. Most analyses, however, are done to investigate the dynamic change in gene expression but not the functional proteins or their activities. To carry out more sophisticated analyses, we need good antibodies to visualize the active molecules, or animals that are transgenic for green fluorescent protein-tagged cycling genes to monitor the dynamic changes in situ and in vivo. At present, there is no information on how Fgf signalling is regulated in somitogenesis. A full understanding of the mechanisms of segmentation needs further studies on the function of modified Notch signalling and on the possible interplay with other signalling and adhesion molecules.
1.
2.
3.
4.
5.
6.
7. 8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
Gossler, A. & Hrabe de Angelis, M. Somitogenesis. Curr. Top. Dev. Biol. 38, 225287 (1998). An excellent overview of vertebrate somitogenesis. Palmeirim, I., Henrique, D., Ish-Horowicz, D. & Pourquie, O. Avian hairy gene expression identifies a molecular clock linked to vertebrate segmentation and somitogenesis. Cell 91, 639648 (1997). This article provided the first molecular evidence that presomitic mesoderm cells show oscillatory behaviour by the expression of c-hairy1. Artavanis-Tsakonas,...