Control of dorsoventral pattern in the chick paraxial mesoderm

During vertebrate gastrulation, the paraxial mesoderm is laid down as two strips of mesenchyme laterally flanking the neural tube and the axial mesoderm, the notochord. While the primitive streak continuously adds new material to the posterior ends of these segmental plates, the cells at their anterior ends condense and epithelialise to form discrete somites, enclosing some mesenchymal remnants in the somitocoel (reviewed in Christ and Ordahl, 1995). After somitogenesis, the paraxial mesoderm differentiates dorsoventrally. Ventrally, the somite de-epithelialises and, together with the somitocoel cells, forms the mesenchymal sclerotome, which gives rise to the vertebral column and ribs. Medially, cells leave the edge of the dorsal, still epithelial dermomyotome and spread ventrally beneath it to form the myotome, the source of the epaxial musculature (Denetclaw et al. 1997). Cells at the lateral edge of the dermomyotome give rise to the hypaxial musculature, migrating laterally to form the limb muscles or spreading ventrally to form the body wall. Last, mesenchymal cells leave the dermomyotome dorsally to form the dorsal dermis.

While segmentation appears to be an intrinsic property of the paraxial mesoderm, its dorsoventral pattern is regulated by extrinsic cues. Induction and maintenance of the sclerotome requires either the notochord or the floor plate of the neural tube as shown by ablation and heterotopic grafting experiments in chick embryos (Brand-Saberi et al., 1993; Pourquié et al., 1993; Goulding et al., 1994), by mouse notochord mutants (Dietrich et al., 1993; Koseki et al., 1993) and by coculture experiments (Fan and Tessier-Lavigne, 1994).

The development of the myotome in vivo similarly depends on extrinsic signals (reviewed in Molkentin and Olson, 1996). However, the source of these signals is controversial: myotome formation in the avian embryo may depend on signals from neural tube and notochord (Teillet and Le Douarin, 1983; Rong et al., 1992; Pownall et al., 1996) or from the dorsal neural tube alone (Spence et al., 1996). The whole neural tube has been suggested to be essential (Xue and Xue, 1996), or dispensable for myotome formation (Spence et al., 1996), or to play only a role in its long-term maintenance (Bober et al., 1994). The notochord has been suggested either to have no effect (Teillet and Le Douarin, 1983; Rong et al., 1992), to induce the myotome (Bober et al., 1994; Pownall et al., 1996; Xue and Xue, 1996) or to repress it (Brand-Saberi et al., 1993; Pourquié et al., 1993; Goulding et al., 1994; Xue and Xue, 1996), and an inhibitory function has been ascribed to the lateral plate mesoderm (Pourquié et al., 1996).

In vitro studies on myotome formation are no more conclusive (reviewed in Molkentin and Olson, 1996): autonomous expression of myotomal markers has been reported for high density cultures of segmental plate or somite cells, and for cultured half-somites (George-Weinstein et al., 1994; Gamel et al., 1995). In other studies, the presence of both neural tube and notochord was essential to allow the expression of any myotome-specific gene from segmental plate explants (Mün-sterberg and Lassar, 1995; Stern et al., 1995). The ventral neural tube/notochord has been implicated as a source of inductive factors (Buffinger and Stockdale, 1995) as has the dorsal neural tube alone (Stern et al., 1995; Stern and Hauschka, 1995; Spence et al., 1996). A myotome-inducing capacity for the surface ectoderm has also been both suggested (Kenny-Mobbs and Thorogood, 1987; Cossu et al., 1996; Spence et al., 1996; Maroto et al., 1997) and denied (Buffinger and Stockdale, 1995; Stern and Hauschka, 1995), whereas the dorsal neural tube (Buffinger and Stockdale, 1995), the lateral plate mesoderm (Cossu et al., 1996) and the intermediate mesoderm (Gamel et al., 1995) have been suggested to delay or inhibit the myogenic response to inducers.

Finally, whether extrinsic influences are required for the formation of the dorsal dermomyotome has not been clarified. Some authors have suggested that the paraxial mesoderm is dorsally predisposed (Dietrich et al., 1993; Williams and Ordahl, 1994), while others found that dermomyotomal features develop in response to influences from surface ectoderm or dorsal neural tube (Fan and Tessier-Lavigne, 1994; Kuratani et al., 1994; Brill et al., 1995; Maroto et al., 1997).

The confusion with respect to tissue influences on formation of the myotome may be diagnostic of the complexity of the system. In vitro, the myogenic response of mesodermal cells may be stimulated by culture conditions promoting withdrawal from the cell cycle (reviewed in Molkentin and Olson, 1996) or by interruption of intercellular communication (George-Weinstein et al., 1994; Gamel et al., 1995), suggesting that the requirement of inducers may be bypassed by artificial culture conditions. On the contrary, in vivo manipulation of a single signalling structure may cause various responses as signals patterning the paraxial mesoderm may influence the environment, may posses multiple functions or may be redundant. Thus, we undertook a systematic study to identify the structures controlling dorsoventral pattern in the paraxial mesoderm.

We gradually ablated and heterotopically transplanted all the structures suggested to pattern the paraxial mesoderm. The simultaneous detection of dorsoventrally restricted markers for the paraxial mesoderm and the axial structures in triple- and quadruple-labelling in situ hybridisations allowed us to identify complex responses to the microsurgical manipulations. Our observations suggest that the newly formed paraxial mesoderm is a naïve tissue, as only the competence to respond to external signals is regulated intrinsically. Both the dorsal neural tube and the surface ectoderm are sources of dorsalising factors which are essential to induce the dermomyotome. Both the notochord and the floor plate of the neural tube are sources of ventralising factors which, at high concentrations, induce the sclerotome and repress the dermomyotome. Where the ventralising and dorsalising factors are present at appropriate relative levels, as occurs in the intermediate region of the somite, they act together to induce the myotome.

Fertilised hens’ eggs were obtained from a mixed flock (Poyndon Farm, Enfield) and incubated in a humidified atmosphere at 39°C. The embryos were staged according to Hamburger and Hamilton (1951). For removal of the newly formed and nascent notochord, Hamburger-Hamilton stage 10 (HH10) embryos were used. For all other types of operation, HH10-13 embryos were used.

For the chick somites, we used the nomenclature of Christ and Ordahl (1995). Somites present at the time of operation were assigned positive roman numerals, beginning with the most recently formed somite. Somites that emerged from the segmental plate postoperatively were assigned negative roman numerals in order of their appearance.

After windowing the eggs and ink injection into the yolk beneath the embryos, the vitelline membrane was opened. In the region to be operated, a slit was cut into the ectoderm between neural tube and paraxial mesoderm or lateral to the paraxial mesoderm using flame-sharpened 100 μm tungsten wire. A tiny amount of dispase (1 mg/ml, Boehringer) was applied with a mouth pipette to loosen the contact between the paraxial mesoderm and its environment. The dispase was washed away with Ringer and the operations performed as described in the results section. For grafting operations, donor embryos were transferred into Ringer solution, the ectoderm or endoderm opened and 50 μl dispase applied until the structures to be grafted came loose. The tissues were excised with tungsten needles, washed in Ringer, transferred into the host with a serum-coated pipette and manoeuvred into position with tungsten needles.

To test for induction of mesodermal markers, most operated embryos were reincubated for 6G or 9G hours, allowing the formation of 4 and 6 somites, respectively. To assay for marker maintenance, the embryos were harvested after 24 hours, so that 13-17 somites had been formed. Embryos in which the posteriormost notochord had been ablated were re-incubated for 24 hours to test for induction of the markers, and for 36 hours or 48 hours to analyse their maintenance. To check that the procedure did not harm the paraxial mesoderm, the neural tube/notochord complex was ablated between somites XV and −VII, and the embryos harvested immediately (n=7) or after reincubation for 1 hour (n=2). In addition, neural tube/notochord were excised and immediately thereafter re-inserted (Fig. 1A), and the embryos cultivated for 24 hours (Fig. 1B). All control embryos displayed the wild-type expression pattern of the dorsoventrally restricted markers MyoD, Pax1 and Pax3 (Fig. 1B-D).

Entire chick embryos were fixed in 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) overnight, and subsequently hybridised simultaneously with two digoxygenin (DIG, Boehringer)-labelled and one or two fluorescein isothiocyanate (FITC, Boehringer)-labelled antisense RNA probes according to Henrique et al. (1995). After hybridisation, the DIG-labelled molecules were detected using NBT/BCIP (Boehringer) as substrate for the anti-DIG-antibody-coupled alkaline phosphatase. After overnight treatment with 4% PFA and extensive washes with PBS, the alkaline phosphatase-coupled anti-FITC-antibody was applied in the same way, and the staining reaction performed with Fast Red (Sigma). The staining reaction was optimised for the detection of the mesodermal markers, leading to some variation in the signal intensity of the notochord or neural tube markers.

The Bmp4 probe described by Francis et al. (1994) labels the roof plate of the neural tube and the overlying ectoderm, the lateral mesoderm and the endoderm (Liem et al., 1995; Pourquié et al., 1996; Fig. 1D).

The Wnt4 probe represents a PCR product encompassing the 3′ end of the coding region (L. Tumiotto, A. Graham, A. Lumsden, unpublished results). Expression is found in dorsal neural tube and the intermediate mesoderm, in line with Hollyday et al. (1995).

The Pax3 probe has been described by Goulding et al. (1993). The mRNA is found in the dorsal neural tube, the anterior segmental plate and the dermomyotome, and at high levels in progenitors of hypaxial musculature at the lateral edge of the dermomyotome (Goulding et al., 1994; Williams and Ordahl, 1994; Fig. 1B,C).

The Pax6 probe of Goulding et al. (1993) detects messages in the intermediate neural tube, dorsally overlapping with Pax3.

The Shh probe and the expression pattern of the gene in notochord, floor plate, endoderm, branchial arches and limbs has been described by Johnson et al. (1994) (Fig. 1B-D).

MyoD (Fig. 1B,C) represents the earliest expressed myotomal marker in the avian somite. Expression starts in the medial halves of somites I or II along the ventral border of the Pax3 expression domain (Williams and Ordahl, 1994; reviewed in Molkentin and Olson, 1996). Details of the probe are described in Bober et al. (1994).

The Pax1 probe is described in Ebensperger et al. (1995). Expression starts simultaneously with or one somite behind MyoD in the ventromedial region of the epithelial somite. After somite deepithelialisation, the transcription of Pax1 is confined to the medial sclerotome (Johnson et al., 1994; Ebensperger et al., 1995; Fig. 1D).

The homeobox of the chick Lbx1 gene was isolated by RT-PCR. The antisense probe derived from this clone labels the lateral edge of the dermomyotome (Fig. 1D) and the migrating progenitors of the limb musculature, coincident with the lateral expression domains of Pax3 (S. Dietrich, F. Schubert, S. C. Chapman and A. Lumsden, unpublished observations), similar to its murine homologue (Jagla et al., 1995).

Embryos were embedded in 20% gelatine in PBS and stored at 4°C overnight. Blocks were trimmed, fixed in 4% PFA overnight, washed in PBS and sectioned to 40 μm on a Pelco 1000 Vibratome. The sections were collected on gelatinised slides and covered with Moviol (Hoechst).

After in situ hybridisation, the embryos were cleared in 80% glycerol in PBS. Whole embryos and sections were photographed on a Zeiss Axiophot, using Nomarski optics.

We first studied the requirement of possible signalling tissues for the dorsoventral pattern of the somite by systematically ablating dorsal and ventral structures from the chick embryo. We then confirmed our results by transplanting these structures to ectopic locations.

Dorsal structures in proximity to the paraxial mesoderm are the dorsal neural tube and the surface ectoderm. To determine their influences on dorsoventral somite pattern, we ablated the dorsal neural tube, the entire neural tube and the surface ectoderm, either alone or in combination, and analysed the operated embryos for the expression patterns of dorsoventrally restricted genes.

Following a vertical incision between the axial structures and the paraxial mesoderm at the level of somites VII to −V, the dorsal half to two thirds of the neural tube was ablated with a horizontal cut (Fig. 1E). As a consequence, the expression domains of Bmp4, Wnt4, Pax3 and occasionally Pax6 were lost, but Shh persisted in floor plate and notochord. After 6G-9G hours, the ectoderm had re-established contact to the cut edge of the neural tube rudiment, which failed to re-close dorsally and to reinstate expression of the dorsal markers (Fig. 1F-H). Possibly as a result of the dorsally flared neural tube, the medial lips of the dermomyotomes appeared less strongly curved than in unoperated embryos (Fig. 1G,H). However, the time course and the dorsoventral distribution of Pax3, MyoD and Pax1 in the paraxial mesoderm remained unaltered (Fig. 1F-H; compare to controls in Fig. 1B-D).

After separating neural tube and paraxial mesoderm with a vertical incision, the neural tube was removed (Fig. 1I), leaving the notochord as the only source of Shh signalling (Fig. 1J-L). In the operated area, extending from somites XIII to −IX, the left and right segmental plates and somites approached each other, fusing above the notochord. Similarly, the edges of the cut ectoderm fused, so that the gap was closed after 24 hours. In all cases, Pax3, MyoD and Pax1 were expressed normally (Fig. 1J-L).

To verify that the paraxial mesoderm can be fully dorsoventrally patterned in the absence of the neural tube, we removed the neural tube between somites XIII and −VIII and inserted one or two segmental plates of a stage-matched donor embryo into the slit on top of the notochord, maintaining anteroposterior orientation (Fig. 1M,N). After 6G hours, the grafts had segmented in the same way as the host paraxial mesoderm and the newly formed somites had activated Pax1 in a normal fashion in ventral regions adjacent to the notochord. Dorsally, the segmental plates and somites expressed Pax3 as soon as they were re-covered by ectoderm. In a similar, ectoderm-dependent fashion MyoD was induced in the newly formed somites just below the Pax3 expression domain (data not shown). After 24 hours, the operated region was fully covered by ectoderm, and the normal dorsoventral expression pattern of Pax3, MyoD and Pax1 established (Fig. 1N-P). However, while the host paraxial mesoderm showed high levels of Pax3 expression in the lateral edge of the dermomyotome and no Pax1 expression in the lateral aspect of the sclerotome, this high level expression of Pax3 never appeared in the graft (Fig. 1O), and the ventral mesenchyme expressed Pax1 throughout (Fig. 1P), possibly because the graft is distanced from lateral structures.

After cutting the ectoderm between neural tube and paraxial mesoderm, a drop of dispase was applied to loosen the contact between the ectoderm and the tissues underneath. The embryo was washed with Ringer solution, and a piece of 10 μm thick tantalum foil inserted between ectoderm and paraxial mesoderm (n=7) or between ectoderm and both neural tube and paraxial mesoderm (n=17) at the level of somites V to −III and V to −VII, respectively (Fig. 1Q). In both types of operation, the dorsoventrally restricted expression pattern of Pax6, Bmp4 (not shown), Shh (Fig. 1R-T) and Pax3 (Fig. 1R,S) in the neural tube was unaltered. Similarly, in the medial half of the somite Pax3, MyoD and Pax1 were induced normally (Fig. 1R-T).

When contact between surface ectoderm and paraxial mesoderm was prevented with tantalum foil and the neural tube removed (Fig. 1U), between somites IV and −V the somite fully de-epithelialised (Fig. 1W,X, arrowheads). Of our mesodermal markers, only Pax1 was activated, occupying all dorsoventral positions (Fig. 1X). Both MyoD and Pax3 failed to be induced or maintained (Fig. 1V,W). The same type of operation performed more anteriorly, between somites XIV and V, allowed a weak expression of MyoD and Pax3, with signals fading from anterior to posterior (data not shown). These findings suggest that the neural tube and the surface ectoderm are the only two structures in the embryo that provide dorsalising signals. Their redundant signals are necessary for the development of both dermomyotome and myotome.

Ventral structures in proximity to the paraxial mesoderm are the notochord, floor plate and endoderm. The endoderm expresses the signalling molecule Sonic hedgehog, as do the notochord and floor plate, and interacts extensively with the splanchnic mesoderm during the formation of the caudal intestinal port (Roberts et al., 1995). However, the endoderm is insufficient for ventralisation of the paraxial mesoderm, due to the absence of required signals or due to limited apposition during development: after ablation of the neural tube/notochord complex no ventral somitic markers were expressed, either in the endogenous paraxial mesoderm or in paraxial mesoderm grafted into the slit (S. Dietrich, F. Schubert, S. C. Chapman and A. Lumsden, unpublished observations). Therefore, we focussed our attention on notochord and floor plate. We ablated either the notochord or both notochord and floor plate at two axial levels, and analysed the expression pattern of the markers as before.

The axial structures were mechanically and enzymatically separated from the paraxial mesoderm in the region of somite XIII to −IV as described above, and the neural tube of the same or a stage-matched donor embryo was grafted back (Fig. 2A). This neural tube, although deprived of notochord, showed the normal dorsoventrally restricted expression of Bmp4, Wnt4, Pax6 (not shown), Pax3 (Fig. 2B,C) and Shh (Fig. 2D). The paraxial mesoderm expressed Pax3, MyoD and Pax1 in a normal fashion (Fig. 2B-D), suggesting that the loss of the notochord was compensated for by the fully established floor plate. In one embryo, the caudalmost area of the neural fold was included in the graft. Here, the Pax6 expression domain was shifted into the ventral midline. In the adjacent paraxial mesoderm, the Pax3 and MyoD signals were shifted ventrally too, with MyoD residing ventrally beneath the neural tube (data not shown).

To ablate the newly formed and nascent notochord, an incision was made between the posterior end of the segmental plate and the neural plate of a HH 10 embryo. After brief enzymatic treatment, the notochord was excised, cutting as far posterior into the node region as possible. The neural plate was flipped back into position (Fig. 2E,M), where it re-established contact to the surface ectoderm and closed dorsally. After 24 hours, the embryos had added 13 to 17 somites. They fell into two categories: in half of the animals, in the operated area a repeatedly interrupted Shh pattern was observed below the neural tube, indicating that the notochord had partially regenerated. Both paraxial mesoderm and neural tube expressed the markers normally (not shown). In the other half of the embryos, as judged by morphological inspection (Fig. 2G,H,O,P) and Shh staining (Fig. 2F-H), the notochord was absent between somites −V and approximately somite −XX. Because the length of this notochord-less area at the time of inspection reached further posterior than the segmental plate at the time of operation, notochord precursors residing in the node must also have been removed. Between somites −VI and −VII, the neural tube appeared dorsoventrally compressed, but the floor plate was still Shh-positive, and Pax6, Wnt4 and Pax3 displayed their normal dorsoventral distribution (not shown). More posteriorly, the neural tube acquired a more radially symmetrical shape. Its diameter remained small, and no morphologically defined floor plate developed. In line with Goulding et al. (1994), in the region of somites −VII to −XIV, the Wnt4 (not shown) and Pax3 signals (Fig. 2F-H, arrows) spread through the neural tube, leaving out the ventral midline where Pax6 (Fig. 2N-P, arrows) and occasionally weakly Shh (not shown) was expressed. In the posteriormost region where the notochord precursors had been removed from the node, neither Shh (Fig. 2F,H) nor Pax6 (Fig. 2N) activity was found, and the neural tube expressed Wnt4 (not shown) and Pax3 (Fig. 2F,H) throughout.

The expression of the mesodermal markers followed the same principle: after 24 hours, in embryos retaining notochord or floor plate, the mesodermal markers were expressed in a normal fashion (not shown). In notochord-less animals, in the region of somites −VI to −IX, the somitic Pax3 staining extended ventrally (Fig. 2F,G, arrows). However, some weak expression of Pax1 (Fig. 2N) and MyoD (Fig. 2F,G,N,O, arrowheads) was still present in ventral locations. In the middle of the operated area, first the Pax1 and subsequently the MyoD staining disappeared (Fig. 2N). Finally, in the region where the neural tube failed to express Pax6, a strong Pax3 signal occupied the entire paraxial mesoderm (compare Fig. 2F,H,N). The same observations were made after reincubation for 36 hours and 48 hours (not shown), suggesting that the removal of the nascent notochord is incompatible with the development of the floor plate in the neural tube (Yamada et al., 1991), and of sclerotome and myotome in the paraxial mesoderm.

To test whether MyoD expression depends on ventralising structures in a region where the floor plate is established, the neural tube was opened in the dorsal midline between somites XIV and −XII, and cut horizontally dorsal to the floor plate. After dispase treatment, notochord and floor plate were excised, while the dorsal neural tube was left in place (Fig. 2I). This neural tube rudiment re-closed and expressed Bmp4, Wnt4 (not shown), Pax3 (Fig. 2K) and Pax6 (Fig. 2J,L). Occasionally, some Shh staining was found at its ventral edges (Fig. 2L, left side). The adjacent paraxial mesoderm showed weak to normal Pax1 and MyoD staining, with the MyoD signals being more robust than the Pax1 signals (Fig. 2J,L, left side). In regions devoid of any Shh staining both Pax1 (Fig. 2J,L, right side) and MyoD (Fig. 2K; Fig. 2 J,L, right side) failed to be activated and maintained. These findings suggest that redundant ventralising signals provided by notochord and floor plate are required for both sclerotome and myotome development. The dose sufficient for myotome induction seems to be lower than the dose essential for the formation of the sclerotome.

The ablation experiments in chick suggested that dorsalising signals are required for dermomyotome formation, and in concert with ventralising factors, for the formation of the myotome. The response of the paraxial mesoderm to ventralising signals seems to be dose dependent, as a low dose appears sufficient for myotome, but not for sclerotome induction. To test these hypotheses, we grafted the signalling structures to ectopic sites and analysed the spatiotemporal expression of MyoD, Pax1, Pax3 in the paraxial mesoderm, and Shh, Pax6, Pax3, Wnt4 and Bmp4 in notochord, neural tube and lateral mesoderm.

Donor notochords were obtained from the posterior region of dispase-treated HH 10-12 chick embryos and inserted dorsally between neural tube and paraxial mesoderm of stage-matched hosts at the level of somites X to −VI (Fig. 3A). As described previously (Brand-Saberi et al., 1993; Pourquié et al., 1993; Goulding et al., 1994), Pax1 was activated around the graft at high levels after 24 hours (Fig. 3B,C; 3D, arrowhead), at the expense of both MyoD (Fig. 3B,C) and Pax3 (Fig. 3D, arrowhead). Close to the transplanted notochord, the somite fully de-epithelialised. However, at some distance, and in proximity to the ectoderm or dorsal neural tube, MyoD was strongly activated in a ventrally expanded domain, in one embryo leading to an ectopic, ring-like expression pattern below the graft (Fig. 3B,C, arrowheads).

When the ectopic notochord was placed laterally to the paraxial mesoderm in the region of somites III to −IV (Fig. 3E), after 6G hours, Pax3 was repressed and MyoD expression induced close to the graft, provided the graft was located near the ectoderm (not shown). After 24 hours, similar to the dorsomedial grafts, close to the graft, Pax1 was strongly expressed (Fig. 3F-H), MyoD (Fig. 3F,G) and Pax3 (Fig. 3H, arrowhead) were repressed, and the region appeared mesenchymal. At a greater distance, an intense and ventrally expanded MyoD signal was found (Fig. 3F,G, arrowheads). However, this MyoD-positive domain still resided in a dorsal territory close to the ectoderm. The more ventrally the transplanted notochord was located, the less dramatic was its effect on the expression of Pax3 and MyoD, and only Pax1 was activated ectopically. These findings support the idea that the myotome is induced by the combinatorial action of dorsalising signals, here provided by the ectoderm, and weak ventralising signals. Stronger ventralising signals induce the sclerotome, thereby suppressing the dermomyotome.

To characterise the dorsalising potential of the neural tube, we excised the neural tube and notochord at the level of somites XI to −VI as described above, and back-grafted it in dorsoventrally inverted orientation (Fig. 3I). After 6G hours, the newly formed somites expressed Pax1 dorsally, close to notochord and floor plate, suggesting that the dorsalising influence of the ectoderm was overridden by the ventralising signals. Ventrally, a strong Pax3 signal was found, adjacent to the Pax3-positive region of the neural tube. In addition, distant to notochord and floor plate, the normal Pax3 expression was found below the ectoderm. The MyoD gene was expressed in two independent domains, one located dorsally close to the ectoderm and one located ventrally, neighbouring the Pax6 expression domain in the neural tube (data not shown). After 24 hours, while the dorsal somite appeared mesenchymal and expressed Pax1 (Fig. 3J,K), the ventral region was epithelially organised and Pax3-positive (Fig. 3L, arrows). Two rows of MyoD signals traversed the operated area longitudinally (Fig. 3J, arrowheads), residing in the same dorsal and ventral position as seen after 6G hours (Fig. 3K,L, arrowheads). These observations support the idea that the somite becomes dorsalised by signals emanating from ectoderm and dorsal neural tube, and ventralised by signals from notochord and floor plate. Where the paraxial mesoderm is exposed to the appropriate level of both signals, myotome formation is induced even in ectopic locations.

When the notochord was removed and only the neural tube reinserted in dorsoventrally rotated orientation in the region of somites VII to −IV, we obtained the same results as for the neural tube/notochord inversions, underlining the redundant function of notochord and floor plate (data not shown).

To analyse the direct competition of dorsalising and ventralising signals, we inverted the neural tube leaving the notochord in place in the region of somites X to −VI (Fig. 3M). The inverted neural tube resisted the ventralising influence underneath as monitored by the expression of Bmp4, Wnt4 (not shown), Pax3 (Fig. 3N,P), Pax6 (Fig. 3O) and Shh (Fig. 3N-P). In the paraxial mesoderm, after 6G hours the marker gene expression resembled the pattern observed in our previous inversion experiments. After 24 hours, however, Pax1 was expressed both adjacent to the ectopic floor plate, and around the notochord (Fig. 3O, arrows), at the expense of Pax3 (Fig. 3N,P). MyoD remained active in two separate domains in agreement with Spence et al. (1996), with the ventrally located domain appearing smaller and shifted dorsally compared with the embryos where both neural tube and notochord were rotated (Fig. 3N-P, arrowheads). In one embryo, at the anterior end of the operated area, the inverted neural tube resided at some distance from the notochord, with the gap filled by paraxial mesoderm. The originally dorsal neural tube was surrounded by a thin layer of Pax3-expressing cells, while the notochord was enclosed by a massive ring of mesenchymal cells, possibly positive for Pax1 (not tested). Between the axial structures, and at the ventral border of the Pax3 domain, an intensely MyoD-labelled zone was found (data not shown). Together, these findings suggest that the dorsalising signals provided by the neural tube in combination with the signals derived from both ventralising centres were sufficient to induce the ventral MyoD domain. Close to the endogenous notochord, the ventralising information overrode the dorsalising signals, thereby repressing Pax3 and MyoD.

As the medial and the lateral half of the paraxial mesoderm have been shown to respond differently to myogenic inducers in vitro (Gamel et al., 1995; Cossu et al., 1996), we transplanted the dorsoventrally inverted neural tube lateral to the paraxial mesoderm in the region of somites III to −IV (Fig. 3Q). After 24 hours, the normal expression patterns of MyoD, Pax1 and Pax3 were established medially, under the control of the endogenous axial structures (Fig. 3R-T). Laterally, the strong Pax3 signal typical for the lateral dermomyotome was absent. However, Pax3 was expressed at lower levels in ventral locations close to the graft, and a lip characteristic of the dorsomedial dermomyotome had formed (Fig. 3T, arrow). From this lip, a second, ventrolateral MyoD-positive cell population emerged, fusing with the expanding endogenous myotome (Fig. 3R-T, left arrowheads). Dorsolaterally, a third MyoD signal was found below the ectoderm at the expense of Pax3 (Fig. 3R-T, right arrowhead). Here, the somite had retained its epithelial characteristics and did not express Pax1 (Fig. 3R,S), presumably due to the distance of the floor plate from the paraxial mesoderm. These findings suggest in line with Ordahl and Le Douarin (1992) that, similar to the medial half, the lateral somite is capable of activating the medial dorsoventral program, responding to signals in this case derived from ectoderm, dorsal neural tube and floor plate.

To directly test the dorsalising capacity of the ectoderm, we separated ectoderm and segmental plate as described before, cut off a strip of ectoderm about 3 somites long, and immediately slid it into the groove between neural tube and segmental plate with the basal epithelial surface facing the segmental plate (Fig. 3U). After 24 hours, marker gene expression resembled the pattern observed after neural tube inversion on top of the notochord. MyoD signals were found in a second, ventrally located, small domain in immediate proximity to a dorsal, normal looking myotome (Fig. 3V-X, arrowheads), supporting the idea of a dorsalising and myotome-inducing function of the ectoderm.

To test whether the patterning function of the axial structures can be substituted by mature somites, we ablated the neural tube/notochord at the level of somites VI to −VI, leaving the ectoderm in place. We then inserted somites XXV to VII obtained from HH 13-17 donors (Fig. 4B), maintaining the dorsoventral and inverting the anteroposterior orientation such that the most mature somites in the graft were brought into the proximity of the youngest somites and segmental plates of the host (Fig. 4A). After 6G to 9G hours, the host embryos had developed 4 to 6 further somites, none of which expressed MyoD (Fig. 4A,C) or Pax1 (not shown). Also the youngest somites present at the time of operation were deprived of both markers, while Pax3 was expressed as soon as the paraxial mesoderm was covered with ectoderm (Fig. 4A,C, arrowhead). After 24 hours, the host failed to express Pax1 and MyoD in the entire operated field, and the signals of these genes faded in the graft as well (not shown). Thus, the embryos exhibited a phenotype typical for ablations of the neural tube and notochord (S. Dietrich, F. Schubert, S. C. Chapman and A. Lumsden, unpublished observations; Teillet and Le Douarin, 1983; Rong et al., 1992; Pownall et al., 1996), suggesting that the presence of the axial structures is essential for the induction and maintenance of sclerotome and myotome.

In our study, we changed the dorsoventral patterning of the paraxial mesoderm by manipulating ectoderm and neural tube, which provided dorsalising signals, and notochord and floor plate, releasing ventralising signals. However, the Pax1 and MyoD genes did not respond prior to somitogenesis. To test whether this response depended on the developmental age of the inducers or the paraxial mesoderm itself, we inserted segmental plates including the recently formed somite from HH 10-13 donors between the neural tube/notochord and the host paraxial mesoderm at the levels of somites IV to −IX (Fig. 4E). The anteroposterior orientation of the grafts was inverted, so that the previously anterior, segmenting area was exposed to a younger neural tube and notochord, while the posterior end faced a more mature neural tube and notochord (Fig. 4D). After 6G hours, both graft and host had developed 4-5 additional somites. At the operated side, the host paraxial mesoderm failed to express MyoD (Fig. 4D,F) and Pax1 (not shown) presumably because it lacked contact with the axial structures. The somites derived from the graft had activated both genes (Fig. 4D,F, arrowhead), while the still unsegmented area remained negative (Fig. 4D). This finding suggests that the paraxial mesoderm gains the competence to activate the sclerotomal and myotomal programs after segmentation. After 24 hours, when the graft was fully segmented, it expressed Pax1 and MyoD along its entire length. Compared to the control side, the expression of both genes was laterally expanded, at the expense of the intense lateral expression of Pax3. By contrast, the host somites in the operated region showed a medial extension of this Pax3 domain, but lacked expression of Pax1 and MyoD (data not shown).

A critical step in the development of the paraxial mesoderm is its dorsoventral patterning, as it leads to the formation of all the three derivatives, the ventral sclerotome, the dorsal dermomyotome and the intermediate myotome (reviewed in Christ and Ordahl, 1995). Ventralisation of the paraxial mesoderm depends entirely on the action of notochord or floor plate (Brand-Saberi et al., 1993; Dietrich et al., 1993; Koseki et al., 1993; Pourquié et al., 1993; Goulding et al., 1994). In contrast, little is known about the expression of dorsal properties (Fan and Tessier-Lavigne, 1994; Kuratani et al., 1994; Brill et al., 1995; Fan et al., 1995), and the studies on myotome formation are contradictory (reviewed in Molkentin and Olson, 1996). Therefore, in this study, we characterised the dorsoventral patterning of the somite, assaying in particular for dorsalising inducers, and for redundant, antagonistic and cooperative signals.

The initial expression of the dorsal, dermomyotomal marker Pax3 throughout the dorsoventral perimeter of the chick mid-segmental plate or anterior mouse segmental plate had led to the suggestion that the paraxial mesoderm is dorsally predisposed (Dietrich et al., 1993; Williams and Ordahl, 1994). In vitro however, this marker is activated upon coculture of segmental plates with dorsal neural tube or surface ectoderm (Fan and Tessier-Lavigne, 1994; Fan et al., 1995), suggesting that the paraxial mesoderm is able to respond to dorsalising signals. In vivo, the expression of the dermomyotomal marker MHox depends on the presence of the ectoderm (Kuratani et al., 1994), while the release of dermis progenitors from the dermomyotome requires contact with the dorsal neural tube or exposure to the neurotrophic factor NT-3 (Brill et al., 1995). Our results confirm that neural tube and surface ectoderm play a crucial role in the initiation and maintenance of the dorsal pathway in the paraxial mesoderm in vivo: when both the neural tube and the surface ectoderm have been removed, the segmental plate fails to activate Pax3. In addition, neural tube/ectoderm removal at the level of the anterior segmental plate and young somites expels expression of Pax3 from the paraxial mesoderm. In both cases, the somites appear completely mesenchymal and express Pax1 throughout their dorsoventral perimeter, indicating that they have been fully ventralised. In line with previous studies, the absence of the neural tube (Teillet and Le Douarin, 1983; Bober et al., 1994; Spence et al., 1996) or ectoderm alone does not interfere with the dorsoventral patterning of the medial somite territory, suggesting that neural tube and surface ectoderm provide redundant dorsalising signals. This interpretation is supported by heterotopic grafting experiments: neural tube transplants induced mesodermal Pax3 expression ectopically, close to the previously dorsal, Pax3- and Wnt4-positive region of the grafted neural tube. Similarly, when dorsalisation of the neural tube is promoted by the ablation of the newly formed and nascent notochord, the extension of the Pax3-positive somite epithelium precisely corresponds with the extension of the Pax3/Wnt4 expression domain in the neural tube, in line with Goulding et al. (1994). In mouse notochord mutants, the position of dorsalising signals provided by surface ectoderm and neural tube seems to determine the expression of dorsal identities in the paraxial mesoderm in a similar fashion (Dietrich et al., 1993; Koseki et al., 1993; S. Dietrich, F. Schubert and A. Lumsden, unpublished observations). Our findings suggest that the paraxial mesoderm does not possess any dorsal identity by default. It is dorsalised by signals provided by the surface ectoderm and the neural tube, similar to the neural tube itself (Liem et al., 1995).

As the paraxial mesoderm is fully ventralised upon neural tube/ectoderm removal and dorsalised when both notochord and floor plate are absent, the spatial organisation of sclerotome and dermomyotome seems to result from an antagonistic function of the dorsalising and ventralising signals. However, dorsalisation fails and Pax1 is expressed at the expense of Pax3, when the dorsalising factors have to compete with strong ventralising signals. This situation occurs when notochord or floor plate is placed dorsally, close to dorsal neural tube or surface ectoderm (Brand-Saberi et al., 1993; Pourquié et al., 1993; Goulding et al., 1994; Xue and Xue, 1996; this study), when ectoderm grafts reside ventrally, or when the neural tube is dorsoventrally inverted above the notochord. Corresponding results have been obtained in vitro, where dorsalising factors provided by the neural tube can be overridden by the exogenously applied ventral signalling molecule Sonic hedgehog (Fan and Tessier-Lavigne, 1994; Fan et al., 1995). These findings suggest that the relative levels of the dorsalising and ventralising signals determine whether a mesodermal cell adopts a dorsal or a ventral identity.

In the medial aspect of chick somite I or II, MyoD as the earliest myotomal marker is initiated in intermediate locations at the ventral border of the Pax3 expression domain (Williams and Ordahl, 1994; Molkentin and Olson, 1996). A similar expression profile has been shown for the earliest myotomal marker in the mouse, Myf5 (Ott et al., 1991). In both species, the myotome is later displaced dorsally, probably because of the strong mitotic activity of the sclerotome (reviewed in Christ and Ordahl, 1995). However, its initial positioning suggests that it belongs to neither the dorsal nor the ventral program of the somite, but rather represents a third developmental pathway. This idea is supported by our ablation experiments: when either all dorsalising or all ventralising structures were ablated, myotomal markers were neither activated nor maintained, suggesting that dorsalising as well as ventralising signals are crucial for the formation of the myotome.

Our results are consistent with observations in other organisms. The mouse mutant open brain represents the category of dorsal aberations: its dorsal neural tube deficiencies coincide with reduction and subsequent loss of the dermomyotomal marker Pax3, the myotomal markers Myf5 and Myogenin, and later a strong reduction of the epaxial musculature (Spörle et al., 1996). Ventral midline phenotypes affecting myotome development are evident in mouse notochord mutants (S. Dietrich, F. Schubert and A. Lumsden, unpublished observations). In T/Brachyury curtailed and in truncate, some activation of the myotomal, but not the sclerotomal program (Dietrich et al., 1993) takes place, probably due to the presence of notochord precursors. In these mutants as well as in Danforth’s short tail and Pintail, which exhibit notochord degeneration, the myotomal markers decline when the all ventralising factors are lost. Correspondingly, in the zebrafish notochord mutants momo and Not-1/floating head lacking notochord precursors, only a few cells in the neural tube express Shh as a floor plate marker, and the induction of adaxial muscle pioneers as well as maintenance of lateral musculature is prevented (reviewed in Odenthal et al., 1996; Stemple et al., 1996). In T/no tail mutants, where notochord precursors are generated but fail to differentiate, adaxial muscle pioneers are absent and lateral MyoD expression reduced.

We directly tested the combinatorial action of dorsalising and ventralising signals by heterotopic grafting experiments: when dorsal neural tube or ectoderm and notochord or floor plate are combined in ectopic locations, ectopic myotomes are induced, reminiscent of the results of previous studies (Bober et al., 1994; Pownall et al., 1996; Spence et al., 1996; Xue and Xue, 1996). The maximal number of myotomes was found after insertion of a dorsoventrally inverted neural tube lateral to the segmental plate: in addition to the myotome induced by the endogenous axial structures, two myotomes formed laterally, presumably induced by the ectopic dorsal neural tube/floor plate, and the ectopic floor plate co-operating with the endogenous ectoderm. These findings suggest that, contrary to the situation in vitro (Gamel et al., 1995; Cossu et al., 1996), the medial and lateral aspects of the paraxial mesoderm exhibit the same competence to respond to external signals in vivo, in line with the mediolateral switch-graft experiments by Ordahl and Le Douarin (1992). In addition, as for results obtained in vitro by Münsterberg and colleagues (1995) and Stern and co-workers (1995), the combinatorial action of ventralising and dorsalising signals induces and maintains the myotome in vivo.

In our heterotopic grafting experiments, the response of the paraxial mesoderm to notochord or floor plate depended on the duration of the ventralising signals: 6G and 9G hours after the operation MyoD was induced close to notochord or floor plate and Pax3 was not yet repressed. After 24 hours, this region lacked MyoD and Pax3 signals, but expressed Pax1 intensely. Now, the MyoD-positive domain resided at some distance from the graft, where Pax1 had not been activated, but Pax3 was repressed. The latter remained active far from the ventralising centres and, in proximity to the dorsal neural tube or ectoderm, suggesting that the ventralising signals exert two different effects: at long range and low concentrations, myotome is induced and dermomyotome repressed, in line with Bober et al. (1994), Pownall et al. (1996), and Xue and Xue (1996); at close range and high concentrations, myogenesis is blocked, and sclerotome formation is promoted, as suggested by Brand-Saberi et al. (1993), Pourquié et al. (1993), Goulding et al. (1994), and Xue and Xue (1996).

This interpretation is consistent with observations on notochord-deprived chick embryos, and notochord mutants in both mouse and fish: the reduction of ventralising signals interferes primarily with the development of sclerotomal, and in the fish adaxial fates, while myotomal/lateral fates are less strongly affected (Dietrich et al., 1993; Koseki et al., 1993; Pourquié et al., 1993; Goulding et al., 1994; Odenthal et al., 1996; Stemple et al., 1996; S. Dietrich, F. Schubert and A. Lumsden, unpublished observations). In the mouse mutants, we saw a differential effect even within the myotome: the medial, Myf5-expressing myotome (Smith et al., 1994; Braun and Arnold, 1996; Cossu et al., 1996) is more sensitive to the loss of ventralising factors than the lateral, MyoD-positive myotome (S. Dietrich,F. Schubert and A. Lumsden, unpublished observations).

Rotation experiments and ectopic transplantation of somite halves have indicated that – in contrast to the rostrocaudal polarity – the dorsoventral and mediolateral organisation of the paraxial mesoderm is yet undetermined in segmental plate and young somites, as the grafts develop ‘ortsgemäß’ rather than ‘herkunftsgemäß’ (Aoyama and Asamoto, 1988; Christ et al., 1992; Ordahl and Le Douarin, 1992; reviewed in Christ and Ordahl, 1995). Our experiments show that moreover external dorsalising and ventralising signals are strictly required for the establishment of all three somitic derivatives, dermomyotome, myotome and sclerotome. The paraxial mesoderm may therefore represent an entirely naïve tissue. Indeed, when contact to the axial as well as the lateral structures is blocked, or when segmental plates are grafted between the contralateral paraxial mesoderm after neural tube/notochord ablation, none of the mesodermal markers is expressed, as long as the operated field is shielded from ectoderm (S. Dietrich, F. Schubert, S. C. Chapman and A. Lumsden, unpublished observations). The sclerotomal and myotomal programs are not rescued by mature somites grafted into the position of the neural tube/notochord, in agreement with the in vitro experiments of Fan and Tessier-Lavigne (1994). Thus all the inductive events patterning the paraxial mesoderm are extrinsic, appositional and instructive.

While the fate of somitic cells is determined extrinsically, the paraxial mesoderm intrinsically controls its competence to respond: activation of MyoD and Pax1 expression was never found before somitogenesis, even when the immature region of the segmental plate was exposed to mature axial signals. However, in mouse segmentation mutants, the expression of these genes follows the same time course as in wild-type animals (S. Dietrich, P. Tremblay, A. Faisst, B. Kammandel, F.Schubert, A. Lumsden and P. Gruss, unpublished observations). This suggests that a timing mechanism intrinsic to the paraxial mesoderm controls its responsiveness to external cues, independent from segmentation.

Our study shows that surface ectoderm, neural tube, floor plate and notochord account for the dorsoventral patterning of the paraxial mesoderm, providing redundant signals that act in an antagonistic as well as in a synergistic manner. What candidate dorsalising signals are produced by the surface ectoderm and the dorsal neural tube, and what ventralising signals are provided by the floor plate and the notochord?

The ventralising function of notochord and floor plate seems to be carried out in part by the signalling molecule Sonic hedgehog (SHH). It is appropriately expressed in both structures in all vertebrates. In vivo and in vitro, sclerotomal/adaxial and to lesser extent myotomal/lateral markers are activated upon ectopic expression of Shh, of Indian hedgehog (Ihh; not active in the ventral midline), or of a dominant negative form of the general hh antagonist cAMP-dependent Protein kinase A (PKA), while a constitutively active form of PKA suppresses Shh-inducible structures (Johnson et al., 1994; Fan et al., 1995; Concordet et al., 1996; Hammerschmidt et al., 1996; Weinberg et al., 1996). In addition, the putative hh receptor patched is expressed in the sclerotome and adaxial muscle pioneers of zebrafish embryos, and in the sclerotome of mouse and chick embryos in a gradient that is highest close to the notochord (Concordet et al., 1996; Goodrich et al., 1996; Marigo and Tabin, 1996). Similarly, the zinc finger transcription factors Gli-2 and Gli-3, mouse homologues of the Drosophila hh target cubitus interruptus, are expressed in the sclerotome, suggesting that a Shh-dependent signalling mechanism could ventralise the paraxial mesoderm (Mo et al., 1997). However, in Shh knock-out mice, Pax1 still is weakly expressed, Myf5 at reduced levels and MyoD in wild-type fashion, and ribs and epaxial musculature develop (Chiang et al., 1996). In addition, in the zebrafish embryo, the formation of the adaxial muscle pioneers requires besides SHH another HH molecule, Echidna hedgehog (EHH), which is released from the notochord and acts primarily on the ventral myotome (Currie and Ingham, 1996). This suggests that factors other than Shh also contribute to the ventralisation of the paraxial mesoderm.

Candidates for the dorsalisation of the paraxial mesoderm exerted by surface ectoderm and neural tube are members of the TGFβ superfamily: for example, the bone morphogenetic proteins BMP4 and BMP7 are capable of dorsalising the neural tube (Liem et al., 1995). However, at the stages critical for somite dorsalisation, weak Bmp7 expression is restricted to the dorsal neural tube, and Bmp4 is expressed in the roof plate and its overlying ectoderm. In addition, BMP4 has been shown to mimic the myotome-repressing function of the lateral plate mesoderm (Pourquié et al., 1996). When we inverted the neural tube, the newly induced dermomyotomal lip always pointed away from the neural tube. Thus, while both BMP molecules may be involved in setting the dorsal border of the myotome and the formation of the dermomyotomal lips, they may not be sufficient for the dorsalisation of the paraxial mesoderm and the induction of the myotome.

In contrast to BMP molecules, members of the wingless (wg) family of signalling molecules seem more appropriately displayed in the embryo to account for the dorsalising and myotome-inducing function of neural tube and ectoderm. In the mouse and chick, Wnt1, Wnt3 and Wnt3A are expressed in the roof plate of the neural tube, and Wnt4 in the dorsal neural tube similar to Pax3, while general ectodermal expression has been reported for Wnt3, Wnt4, Wnt6 and Wnt7B (Parr et al., 1993; Hollyday et al., 1995). In vitro, a dermomyotomeinducing function has been ascribed to Wnt1 (Maroto et al., 1997). Similarly, a myotome-inducing capacity has been demonstrated for Wnt1, Wnt3 and Wnt4, in particular in concert with Shh (Münsterberg et al., 1995; Stern et al., 1995), and, in Drosophila, wg is able to induce the MyoD homologue nautilus (Ranganayakulu et al., 1996). Thus the positive role of wg/Wnt genes for muscle development seems to be conserved during evolution. Interestingly, in the mouse Wnt4 is additionally expressed in the floor plate, while its Xenopus counterpart lacks the dorsal expression domain (reviewed in Moon, 1993). Whether these differences account for the size difference of sclerotome and myotome in amniotes and anamniotes, is unclear.

In summary, we would like to propose the following model. The paraxial mesoderm is a naïve tissue, which is dorsalised by redundant signals from surface ectoderm and dorsal neural tube. BMP4 or BMP7 may initiate this dorsalisation and, later in development, determine the dorsomedial and dorsolateral border of the myotome. However, the prolonged action of the dorsalising structures crucial for both dermomyotome and myotome development may be instigated primarily by members of the Wnt gene family. By contrast, redundant signals from notochord and floor plate ventralise the paraxial mesoderm. Candidate ventralising factors are SHH in concert with other HH family members or with the products of the ventrally expressed Wnt genes. The response of the paraxial mesoderm to the dorsalising and ventralising factors depends on their relative levels: close to the signalling source, they antagonise each other, thereby allowing the development of either sclerotome or dermomyotome. At some distance they co-operate, thereby inducing the myotome (Fig. 5).

We are indebted to L. Tumiotto (Motta di Livenza) and A. Graham (London) for the generous gift of the chick Wnt4 probe prior to publication. We thank R. Balling (Munich) for the quail Pax-1 probe, T. Lints (New York) for the chick Shh probe, and B. Paterson (Bethesda) and E. Bober (Braunschweig) for the chick Myf5 and MyoD probes. We are grateful to M. Dixon for critically reading the manuscript. S. D. and F. R. S. were supported by the European Molecular Biology Organisation (ALTF 253-1994 and 254-1994) and S. D., in addition, by the Human Frontier Science Program Organisation (LT-216/95). A.L. is an International Research Scholar of the Howard Hughes Medical Institute.