Cell rearrangement and segmentation in Xenopus: direct observation of cultured explants

Early amphibian development is largely the development of the dorsal mesoderm. The notochord and the somites of the vertebrate embryo are constructed directly by this tissue, while the neural tube results from its inductive influence. Spemann recognized the exceptional pattern-forming properties of this region, demonstrating that it had the capacity to induce patterned dorsal structures in ventral ectoderm and mesoderm; he christened it the ‘organizer’ (Spemann, 1938). In cooperation with the prospective neural plate, the convergence and extension of the dorsal mesoderm also drives much of gastrulation (Keller, 1986; Keller & Danilchik, 1988), and elongates the dorsal side of the embryo during neurulation (Jacobson, 1981; Keller & Danilchik, 1988). Although the importance of these morphogenetic movements has been known for a long time (Vogt, 1929), almost nothing is known about how cells bring them about.

Finally, the appearance of somites in the dorsal mesoderm during neurulation is particularly interesting because the pattern of somites prefigures - and in fact directs - all subsequent segmental organization in the vertebrate embryo (Detwiler, 1934). Hamilton used histological sections and Youn and his co-workers used SEM to describe various aspects of amphibian somitogenesis (Hamilton, 1969; Youn et al. 1980; Youn & Malacinski, 1981a,b ; Malacinski et al. 1981). Several groups also investigated the effect on somite pattern of various microsurgical manipulations (Waddington & Deuchar, 1953; Deuchar & Burgess, 1967; Cooke, 1975), and of brief heat shocks (Pearson et al. 1976; Cooke, 1978; Pearson & Elsdale, 1979; Elsdale & Davidson, 1983). Although much has been learned from these studies, we still do not understand how amphibian cells group into segments, or what cellular activities precede segmentation.

These processes have been difficult to study because the amphibian embryo is opaque - inner layers of the intact embryo are hidden from view. Thus until recently it was not possible to observe directly the patterns of cell behaviour responsible for mesodermal morphogenesis. In the last few years, however, Keller and Danilchik have developed a method of culturing explants of amphibian tissue that permits almost normal development and continuous observation of the mesoderm (Keller et al. 1985a,b). We have used this culture technique to observe, in vitro, morphogenesis and differentiation of the somitic mesoderm during the néurula stages, after its involution is complete. We have filmed for the first time in the amphibian not only the segmentation and subsequent morphogenesis of the dorsal mesoderm, but the cell rearrangement that precedes it and drives dorsal elongation. In this paper, we will analyse the progression of cell behaviours that sweep from anterior to posterior in the axial mesoderm. Our focus will be on the two most dramatic processes taking place during this time: convergent extension and segmentation. In a companion paper (Wilson & Keller, 1989), we present our analysis of cell behaviour in the dorsal mesoderm during gastrulation. In a third paper, Keller and others (1989) use both this new culture technique and SEM to analyse notochord development.

Although we believe our work to be the first detailed study of axial cell behaviour in living amphibian tissue, it is worth noting that these events can be studied in some fish embryos, whose transparency allows individual cells to be followed with Nomarski optics. For an analysis that in many ways confirms our own, see Thorogood & Wood (1987). Kimmel & Warga (1987) have followed labelled mesoderm cells during axial morphogenesis in the zebrafish.

Eggs were obtained from females previously injected with human chorionic gonadotropin, and fertilized with sperm from testis stored at 6°C. Jelly coats were removed by brief treatment in 0-2 M cysteine HC1 at pH 7-9; vitelline envelopes were stripped away with forceps immediately prior to operation. Embryos were grown to neurula stages in one-third strength modified Niu-Twitty solution (Keller, 1984), and staged by reference to the tables of Nieuwkoop & Faber (1967).

Dissection was carried out using hairloops and knives made from eyebrow hairs, in 100% Danilchik’s saline (see below). Incisions were made through the remaining yolk plug or slit blastopore and through the lateral walls of the archenteron (Fig. 1A). The dorsal side of the embryo was then folded away from the rest while preserving a connection at the anterior end (Fig. 1B). Next, the endoderm of the archenteron roof was carefully peeled away, exposing the notochord and paraxial mesoderm. Finally, the dorsal piece was freed by a transverse cut at about the level of the anterior end of the notochord, and the explant was trimmed laterally to the approximate edge of the somitic mesoderm (Fig. 1C), which is not yet distinct from the lateral mesoderm. Most explants were made from stage-12·5 to stage-13 embryos; others were made from older neurulae (stages 14−16). In order to observe somitogenesis from the dorsal side, several explants without neural plates were made. After the neural plates were peeled away in 0·4 mg ml^-1 collagenase (Cooper Biomedical), explants of dorsal mesoderm and underlying endoderm were excised, rinsed and cultured dorsal side up for filming. Cultures of somitic mesoderm isolated from all other tissues were also studied. These were prepared by immersing an explant of the type described above (consisting of axial mesoderm and neural plate ectoderm) in dilute collagenase (0·4 mg ml^-1) for one minute and then peeling the somitic mesoderm away from notochord and ectoderm in Danilchik’s solution.

Explants were cultured in 35 mm plastic Petri dishes (Falcon), in 100% modified Danilchik’s solution. This saline solution supports normal development of mesodermal cells and retards the spread of epithelia, which in earlier attempts at explant culture tended to heal over and obscure the mesoderm. Its ionic composition resembles that of the blastocoel fluid (Keller et al. 1985a, modified as follows: no sodium isothionate, Na₂CO₃ concentration now 53 mm, HEPES buffer added to set pH). The explants were positioned in the dish with the mesoderm facing up and immobilized beneath a small piece of no. 1·5 glass coverslip (Fig. 1D). Vacuum grease was used to hold the glass in place and support it above the bottom of the dish. The coverslips were pressed upon the tissue firmly enough to prevent healing and curling, but not enough to damage the cells. The dishes were then kept at room temperature, or on an operating stage cooled to 17°C.

Explants were observed on a Zeiss standard 16 upright microscope, using ×10, ×20, and ×40 lenses with low-angle fibre optic illumination. A video image was generated by a Dage high-resolution camera (model 81, 1300×1050 lines) and a Dage MTI 2000 monitor. The video screen was then filmed in time lapse with an Arriflex 16 mm camera on Plus X reversal film. This system, first described in Keller & Hardin (1987), is preferable to direct filming because it permits the electronic enhancement of contrast; at the same time, filming the screen preserves resolution better than the video recorders now available. The video image was also photographed on 35 mm Plus X pan film, at 1/4 second exposures. The behaviour of explants and individual cells was analysed in the films using a NAC analysis projector.

Explants of dorsal mesoderm and neural plate change shape dramatically, lengthening 35−50% between the end of gastrulation and the early tailbud stage (Fig. 2). The rate of extension, which averages about 65 μm h^-1 at 23°C, is quite constant, at least through the end of neurulation (Fig. 2C). Extension can be considerably faster than this: we have observed rates of up to 115μmh⁻¹ in films. Explants converge toward the midline as they extend, the posterior part narrowing far more than the anterior part.

Analysis of cell movements in low-power films of extending explants reveals that all regions of the somitic mesoderm do not contribute equally to extension. As Fig. 3 illustrates, only the most posterior 20 or 25% of the tissue extends. Cells in the rest of the somitic mesoderm move very little along the axis. Furthermore, as the explant extends, the zone of extension remains restricted to a small posterior region of the somitic mesoderm. Some cells that participated actively in extension between stage 12·5 and stage 16 are left behind by the growing tip of the explant, and move little during the second half of the film. In contrast, all parts of the notochord lengthen (see Keller et al. 1989). Cells throughout the notochord are displaced caudally as the explant extends, but more posterior cells move farther. (The anterior end generally remains fixed with respect to the culture dish.) This implies that all regions of the notochord are extending.

Although the notochord and somitic mesoderm lengthen at the same rate overall (neither leaves the other behind), the very different regional patterns of extension ensure that the two tissues must shear past one another at their boundary. In fact, the films show a dramatic posterior flow of notochord relative to the mesoderm at its flanks in all but the most posterior regions, where rapid cell rearrangement enables the somitic mesoderm to keep pace (see below).

The considerable elongation of the somitic mesoderm during the neurula stages is not accomplished by a corresponding change in cell shape, since the cell elongation that does eventually occur is transverse, rather than axial. Neither can growth be responsible, since the volume of the early amphibian embryo must remain essentially constant. In any case, there is very little cell division in the somitic mesoderm at this stage. Therefore, extension of the tissue must occur solely by cell rearrangement. We have documented this directly by following individual cells in films.

Two kinds of rearrangement take place during explant elongation: radial intercalation (exchange of cells between layers, and thus along the radius of the embryo), and mediolateral intercalation within the tissue layer. (Keep in mind that the surface we observe is the layer of mesoderm in contact with the endoderm before explantation, not the surface of the embryo.)

Radial intercalation takes place at the posterior, or blastoporal, edge of the somitic mesoderm, where the greatest extension occurs. Cells enter the surface layer from below, causing a rapid increase in the surface area of the tissue. Fig. 4 charts the appearance of new cells in the surface layer of an explant from stage 13·5 to stage 14·5, and the rapid extension of the posterior tip of the explant that results. These tracings also show that radial intercalation remains confined to a narrow band, perhaps only five or six cells wide, at the posterior edge of the tissue. As new area is added, some cells are pushed away from the extending tip and leave the zone of radial intercalation. Former neighbours of these cells may remain in the rapidly intercalating tip. Compare, for example, the fates of cells 13 and 17 in Fig. 4. We stress that the pattern of radial intercalation and cell separation in the surface varies within and among explants. These are statistical rather than deterministic processes.

Fig. 4 also illustrates graphically a striking feature of radial intercalation: newly arrived cells in the surface layer are not distributed at random within the intercalating zone, but lie in groups. The films suggest that this results mostly from the sequential appearance of several cells in a small region of the tissue, rather than from the simultaneous addition of a group of cells. That is, it is quite common for cells to enter the surface layer alongside newly arrived cells. This observation suggests that the contacts of newly intercalated cells with their neighbours in the surface layer are somehow weaker, or more vulnerable to intrusion from below. Whatever the reason, radial intercalation is characterized by ‘hot spots’ of greater activity.

Radial intercalation increases the area of the somitic mesoderm and thins it. However, it is clear from Fig. 4 that this expansion is not isotropic: the tissue lengthens without becoming wider. Analysis of individual intercalation events suggests that the addition of new cells to the surface may be biased, at least in gastrula explants, where a similar episode of radial intercalation takes place (Wilson & Keller, 1989). Intercalating cells appear to separate surface cells along the anterior-posterior axis more often than along the transverse axis. This could help the tissue expand mostly along the anteriorposterior axis, but it cannot explain why it doesn’t widen at all, because transverse separations do occur. Thus radial intercalation must be accompanied by rearrangement within the surface layer (mediolateral intercalation), which works to further lengthen the tissue and to narrow it toward the notochord. At the posterior edge of the explant both kinds of cell intercalation apparently occur together, bringing about rapid extension without much change in width. Indeed, Fig. 4 documents well the importance of both types of rearrangement. As an example, consider again cells 17 and 13, neighbours in the first drawing that are eventually separated by three or four cell diameters. Of the seven cells that lie more or less in between 13 and 17 in the last tracing, four (56, 19, 37 and 28) are new to the surface layer. However, three others (1, 18 and 11) are former medial or lateral neighbours of the separated pair.

In one film, cells seem to appear at the very end of the somitic mesoderm, as if the tissue were extending by involution (rolling around a lip from another layer), rather than by radial intercalation. Fate maps show that not all mesoderm has ‘involuted’ by the end of gastrulation (Vogt, 1929; Keller, 1976), although the limits of vital dye mapping make this result difficult to interpret. Thus continuing involution probably does contribute to extension in vivo. However, in the other explants that we have filmed cells are added individually or in small groups throughout an extending tip, as in Fig. 4. Furthermore, the extension of explants from which the neural plate - and presumably any pre-involution mesoderm - has been removed is not impaired (see Section F). For these reasons, we believe that radial intercalation, or thinning, is the primary mechanism of extension in the somitic mesoderm, both in explants and in the embryo. In any case, the thick collar of mesoderm that borders the blastopore at the end of gastrulation is not composed of well-defined layers, and in these circumstances radial intercalation and involution are difficult to distinguish. We are currently working to clarify the roles of these two morphogenetic processes.

In the somitic mesoderm, mediolateral rearrangement continues anterior to the region of rapid radial intercalation, where it is eventually accompanied by cell elongation perpendicular to the notochord. Fig. 5 follows a patch of cells through this process. Despite the opposing effect of changing cell shape, the patch as a whole extends slightly and converges. The area of the patch decreases by 25 %, suggesting that cells lengthen in the third dimension (along the radius of the embryo), sections reveal that this is in fact the case (Schroeder, 1970). In other explants, the region of the somitic mesoderm where rearrangement and cell shape change are taking place together does not extend at all, and may even shrink along the axis. Thus the narrowing of individual cells is sometimes more than enough to offset the effect of organized neighbour change. Although these regions may or may not extend, they always converge toward the midline. The cells of the notochord also change shape and intercalate mediolaterally during this time (Keller et al. 1989). In the notochord, however, the net effect of these processes is always both extension and convergence.

Since we must follow populations of cells for quite some time to detect cell rearrangement, it is difficult to delimit precisely its beginning and end. Nonetheless, we conclude from our analysis that mediolateral intercalation is already underway at the posterior tip of the explant, where it is accompanied by rapid radial intercalation, and that it continues at least through the period of cell elongation. It is possible that there are actually two episodes of mediolateral intercalation, separated by a period of relative stasis. The first episode of vigorous rearrangement, documented in Fig. 4, would occur at the extending tip. The second, rather different type of rearrangement, which we have just described, would be closely associated with cell elongation, much as these processes are associated during notochord development (Keller et al. 1989).

In most explants made at stage 12·5 or 13, the cells of the somitic mesoderm are rounded or polygonal, especially in posterior regions. At least this is the profile they present to the endodermal face of the tissue: sections confirm that their three-dimensional form is more or less cuboidal (Schroeder, 1970). Beginning at about stage 14, these cells begin to change shape, lengthening in the transverse axis and narrowing considerably, until they lie in ranks perpendicular to the notochord (Fig. 6). The cells deepen as well, since their surface area generally decreases during this time (see previous section). Schroeder’s study found that, in the embryo, the somitic mesoderm rises to form buttresses on either side of the developing neural tube. This dorsoventral thickening of the somitic mesoderm is apparently independent of neurulation itself, however, since it occurs in our explants, in which the neural tube does not form.

This process begins in the anterior or middle part of the explant and then spreads posteriorly. Thus a wave of cell elongation passes through the tissue, preceding the wave of segmentation by roughly three to four hours (at 23 °C) and leading it by about five or six prospective somites. The complete change in shape takes about one hour at any position along the axis. These are only crude averages, however, for the rate, spread and extent of cell elongation varies a great deal among explants. In general, the cells lengthen less in the explants than in whole embryos. Furthermore, the presomitic cells tend to reshorten somewhat before segmentation in vitro. Thus there is considerable variability in the shape of somitic cells at the time of segmentation, from nearly rounded in parts of some specimens, to long and narrow in others. These differences in cell shape do not affect segmentation.

When control embryos reach about stage 16, the first pair of transverse fissures appears in the somitic mesoderm of the explants. These first fissures are often shorter and more irregular than succeeding boundaries. By stage 18, three or more pairs of fissures can be discerned. The fissures are first visible at the lateral edge of the explant, and progress toward the midline (Fig. 7). Thorogood & Wood (1987) also saw fissures form from lateral to medial in the teleost embryo. The time required for a fissure to reach the notochord varies considerably. Some boundaries cover the entire distance almost at once, while others remain incomplete for an hour or longer. Thus it is not uncommon for a boundary to begin to form before its predecessor has reached the notochord.

We observe two kinds of activity immediately before the formation of a fissure. First, two or more faint nearfissures, or unusually pronounced stretches of cell boundary, can sometimes be discerned in the region where the next boundary will form. The definitive fissure emerges as one of these candidates straightens and intensifies. Second, the cells that will come to lie on opposite sides of the eventual boundary sometimes begin to move past one another before the fissure forms. Cells anterior to the prospective boundary move toward the notochord; cells on the posterior side move laterally. This shearing at the boundary is the first manifestation of the somite rotation described by Hamilton (1969) and by Youn & Malacinski (1981a). We do not observe the line of blebbing or protrusive activity that sometimes precedes the notochord somite boundary in gastrula explants (Keller & Wilson, 1989). New fissures succeed each other in anterior-posterior progression at approximately 45 min intervals (at 23°C). By the equivalent of stage 21−22, there are 8 to 10 fissures on each side of the notochord and segmentation of the explant is virtually complete (Fig. 8).

Although segmentation in the explants is remarkably orderly, there are occasional irregularities. The most common deviation from the standard pattern is variation in the size of individual somites, on one side of the notochord or both. This can lead to fissures appearing out of register on the two sides. Occasionally fissures are sharply .angled. In a very few cases we have seen branching fissures.

In a total of 9 explants fashioned in the usual way at the end of gastrulation, the notochord was separated from the somitic mesoderm on both sides. The three regions of the mesoderm, each with its associated neural plate, were cultured separately until the end of neurulation. In every case, the central piece, consisting of notochord and notoplate, shortened and fell apart, while the pieces containing somitic mesoderm extended and segmented (Fig. 9A,B). The average increase in length of the paraxial explants was 160 μm, or about 15%, but their curvature makes extension difficult to measure. A simple solution is to combine the left and right sides of the somitic mesoderm in a single notochordless explant. The two sides can be joined at their medial edges or, by switching the left and right sides, at their lateral borders. These symmetric recombinations also extend and segment, without bending. (See Fig. 9C,D.) The average elongation of 3 specimens was 540/rm (about 45%). This is indistinguishable from the extension of whole explants.

Isolated paraxial regions (somitic mesoderm plus neural plate) are invariably S-shaped after extending in culture. This shape is the result of two distinct phenomena. The larger anterior bend, bowing the somitic mesoderm away from the notochord, develops within seconds of separation from the rest of the dorsal tissue, suggesting that the medial tissue is under compression relative to the lateral region. Perhaps the part of the explant closest to the midline seeks to extend more vigorously than the more lateral part, which then restrains elongation. Such an uneven distribution of lengthening activity, expressed during gastrulation, could lead to the pattern of stresses revealed when the tissue is cut at stage 12·5.

The smaller curve, which bends the posterior tip back toward the former midline, arises during subsequent extension. The direction of curvature is surprising, since it seems to imply that the lateral part of the explant lengthens fastest. This would contradict not only the conclusions drawn from the response to cuts, but the simple notion that the capacity for autonomous extension should be greatest at the dorsal midline and fall off laterally. However, the same paradoxical bend toward the midline is observed in left and right halfexplants made at the start of gastrulation.

With the aid of dilute collagenase, the neural plate can be peeled away from the underlying dorsal mesoderm at the end of gastrulation. The axial mesoderm can then be explanted, with or without the endoderm of the archenteron roof, and cultured in the usual manner. In this way, one can test the role of the ectoderm in the neurula-stage development of the mesoderm.

The axial mesoderm develops in combination with the endodermal epithelium very much as it does when explanted with the neural plate. It converges and extends: one group of six explants extended an average of 50% from stage 12·5 to stage 20, somewhat more than the average for explants of mesoderm with neural plate. These explants also segment. Fissures appear at regular intervals, working their way from the lateral edge of the tissue toward the notochord (see Section B). Segmentation is preceded by cell rearrangement and elongation, and followed by the movements of somite rotation. In fact, the behaviour of the somitic mesoderm in films of these neural-plate-less explants appears to differ from its behaviour in the standard explants only in that it is somewhat less regular. This is probably the result of the collagenase treatment. Explants stripped of both the neural plate and the endoderm also segment, sometimes quite normally. They extend considerably less than explants that include one of the epithelial tissues, however. The average of ten explants was 75 μm (less than 10%).

During Xenopus gastrulation, both the involuting marginal zone (the prospective notochord and somites) and the non-involuting marginal zone (the prospective notoplate region of the neural plate) narrow dramatically toward the dorsal midline (converge) and extend in the animal-vegetal axis. By culturing isolated marginal zones, Keller and his co-workers showed that this shape change is autonomous, and that extension is driven primarily by active cell rearrangement (Keller et al. 1985a,b;Keller & Danilchik, 1988). By following the interdigitation of fluorescently labelled grafts and unlabelled hosts Keller & Tibbetts (1989) demonstrated that much of this cell intercalation takes place along the mediolateral axis. Wilson & Keller (1989) have confirmed this directly by filming gastrula explants, and shown that an initial burst of radial intercalation also contributes to extension. The results that we have just presented reveal that, in explants of already involuted mesoderm, these movements continue throughout the neurula stages.

In neurula explants, as in gastrula explants, rapid radial intercalation precedes mediolateral intercalation in each region of the somitic mesoderm. Moreover, this sequence is linked to a well-defined spatial pattern of cell behaviour that we do not see as clearly in our gastrula-stage explants. When the dorsal mesoderm is first explanted at the end of gastrulation, cells in roughly the posterior half are rearranging within the surface, while radial intercalation is restricted to a narrow band at the posterior edge. Cells in the anterior half are already changing shape, and subsequently rearrange little. As the tissue extends, the zones of radial and mediolateral intercalation move caudally, so that cells in each region progress from one behaviour to the next. At stage 16−17, the wave of segmentation and somite rotation begins to advance through the tissue from anterior to posterior, following behind the waves of cell rearrangement and cell shape change. This pattern is summarized in Fig. 10.

Thus, a sequential pattern of cell activities spans gastrulation and neurulation in the somitic mesoderm. As each region of the mesoderm is awakened from its pregastrular quiescence, its cells begin the progression from radial to mediolateral rearrangement, to shape change and finally to segmentation and somite rotation. Prospective anterior parts of the mesoderm begin this sequence early in gastrulation, and have virtually completed rearrangement by the onset of neurulation. More posterior (and ventral) tissue begins the sequence later, and continues to actively rearrange during neurulation, extending and converging toward the midline. The most ventral tissue, which finds itself in the thickened collar surrounding the blastopore at the end of gastrulation, may not begin to extend until the tailbud stage. Therefore, the same sequence of cellular behaviours may drive the major morphogenetic movements of the mesoderm in both the gastrula and the neurula, despite their radically different geometries.

The sequence of dorsal mesodermal behaviours that we have described in Xenopus may be shared by a surprisingly wide range of vertebrates. Warga & Kimmel have recently noted a progression from radial to mediolateral intercalation to segmentation in the prospective mesoderm of the zebrafish (Kimmel, personal communication). Convergence and extension occur in other anurans (Schechtman, 1942) and in urodeles (Vogt, 1929), although Shi and his co-workers (1987) report that in explants of Pleurodeles convergent extension (and presumably cell rearrangement) do not get under way until the late gastrula. This suggests that extension by active cell rearrangement may be primarily a neurula-stage mechanism in this species, and perhaps other urodeles as well. Keller & Danilchik (1988) have suggested that the precocious onset of this process in Xenopus, where it does much of the work of gastrulation, may be related to the very rapid development of this frog.

Convergent extension in neurula explants differs from similar processes taking place in gastrula explants in at least two ways. First, the anterior-posterior sequence of cell behaviour is more regular in the neurula. This may be an artifact of explant culture, which disturbs morphogenesis to a greater extent at earlier stages. Alternatively, it may be that the anterior-posterior prepattern is not as well established in the mesoderm prior to involution. It is important to remember that the primordium of notochord and medial somitic mesoderm is only a few cells long at the start of gastrulation (Keller, 1976), providing little space for extensive patterning. It may be that axial pattern becomes progressively more detailed as the mesoderm lengthens during gastrulation, making greater resolution possible. This refinement of positional specification may be linked to radial intercalation, which is responsible for much of the extension. In the embryo, these events would coincide approximately with involution.

Dorsal elongation in the neurula also differs from extension in the gastrula in that the notochord and the somitic mesoderm are now distinct. The cells of the two tissues are morphologically distinguishable by stage 11·5 (Keller et al. 1989); the actual boundary between them appears around stage 12 or 12·5 (Wilson & Keller, 1989). Despite the subtle differences, the two tissues (or their primordia) seem to extend by essentially the same cellular mechanisms during most of gastrulation. In the neurula explants, however, the two patterns of extension and underlying cell behaviour are quite different, although cells intercalate and elongate in both tissues. In the notochord, mediolateral intercalation and cell shape change result in convergence and extension. These processes occur in the somitic mesoderm as well, but here their combined effect is to narrow and thicken the tissue, rather than to lengthen it. Extension of the somitic mesoderm derives instead primarily from radial intercalation at the posterior tip.

Which tissue drives the dorsal elongation of the neurula? Jacobson has attributed this leading role to the combination of notochord and notoplate (the specialized region of the neural plate along the midline) (Jacobson & Gordon, 1976; Jacobson, 1981). The results reported in Section E demonstrate unequivocally that the somitic mesoderm, together with its neural plate backing, is also capable of substantial autonomous extension in Xenopus. Although isolated combinations of notochord and notoplate do not lengthen, this does not imply that they are not active in the embryo, or in whole explants. In fact, Jacobson and his co-workers (1986) recently proposed that the elongation of this tissue requires a boundary with somitic mesoderm and neural plate. Such a boundary is apparently not required for extension of the somitic mesoderm and neural plate. Furthermore, the somitic mesoderm can extend without the neural plate if the endodermal epithelium is present, strongly suggesting that it is an active partner in dorsal elongation. The neural plate probably lengthens by its own efforts as well, for Keller & Danilchik (1988) demonstrated this capacity in explants made at the start of gastrulation.

Our results demonstrate that the somitic mesoderm at the end of gastrulation needs no assistance from other tissues to carry out convergent extension, cell elongation and segmentation. All three take place in material isolated at stage 13. Brustis (1976) reported segmentation of isolated material in Rana and Bufo, so the same result in Xenopus is not surprising. One should not infer from this that the pattern of segment boundaries is already established at the time of explantation, only that the somitic mesoderm has acquired the capacity to establish this pattern.

The dorsal mesoderm extends vigorously during the neurula stages in conjunction with either the endodermal epithelium or the neural plate, and weakly in complete isolation. By comparison, explants made at the beginning of gastrulation almost never extend without an accompanying epithelium, even when the neurula stage is reached (Keller et al. 1985a,b ; Keller & Danilchik, 1988). Our current hypothesis, based on unpublished work on gastrula explants, is that radial intercalation requires the presence of an epithelium. Once a region of mesoderm has passed through this phase of activity, however, it can undergo mediolateral intercalation without the epithelium. Since the bulk of extension in the neurula explants is generated by radial intercalation, these explants extend much less when the epithelia are removed.

Extensive cell rearrangement precedes segmentation, resulting in dramatic extension and convergence. What consequences does this have for the mechanism of segmentation? First, extension implies that, before rearrangement, the somite primordia are tightly compressed along the anterior-posterior axis. This is seen in fate maps of the somitic mesoderm made at the start of gastrulation, and more dramatically in Elsdale & Davidson’s (1983) map of the tailbud in Rana. This condensation of the fate map must set a limit on the time at which individual somite identities are established, since more than one primordium eventually evolves from a single cell. For example, the involuting marginal zone at the start of gastrulation is at most 12 cells in prospective anterior-posterior length. Yet here lie the anlagen of at least the 6−8 somites that involute fully by stage 12·5−13. The condensation of prospective segments around the blastopore at the end of gastrulation may be even greater. Furthermore, the cell rearrangement that drives extension itself argues against early determination of the segments, since the detailed arrangement of tiny primordia would have to be preserved through all this mixing. We see no evidence that this is the case, for neighbouring cells can be pushed several cell diameters apart, while others remain together. (See Results, Section B.) These considerations lead us to believe that segmental primordia are determined no earlier than the middle or end of the period of rapid cell intercalation. However, until we manipulate the pattern of extension and segmentation experimentally, this conclusion must be tentative.

Armstrong & Graveson (1988) reached conclusions similar to ours from a study of somitogenesis in the axolotl. They distinguish two regions of the unsegmented mesoderm: a ‘cohesive zone’ behind the last-formed somite - containing up to nine prospective somites - and a more loosely organized extending zone. Somites whose primordia lie in the posterior part of the cohesive zone are vulnerable to disruption by heat shock. Moreover, parts of the mesoderm posterior to this can be deleted without loss of segments. From these results, Armstrong & Graveson conclude that determination of primordia occurs within the cohesive zone and thus after cells cease extension. These authors do not discuss cell rearrangement, which they cannot observe directly, and appear to imply that extension occurs largely by cell division. This is not the case in Xenopus (Results, Section B).

According to Elsdale & Davidson (1983) the situation in the tailbud of Rana is apparently quite different. They also divide the presomitic mesoderm into regions: a prepatterned zone, corresponding roughly to the cohesive zone of Armstrong & Graveson, a zone of extension and a third, less well-defined, ‘packing zone’. The anterior boundary of the extension zone coincides with the transition from heat-shock sensitivity to insensitivity, suggesting to Elsdale & Davidson, as to Armstrong & Graveson, the onset of segmental determination. However, unlike in the axolotl, Elsdale & Davidson find that small lesions invariably result in missing or damaged somites, even at the posterior end of the tailbud. This extraordinary lack of pattern regulation leads them to conclude that, well before the onset of extension, the packing zone is a ‘mosaic of small groups of cells already differentiated from their neighbors’. They do not propose that these cells have segmental identities, but that they possess anterior-posterior positional values that will be converted into somite identities much later, at the time of heat-shock sensitivity. We cannot explain the apparent contradiction between the results of these deletion experiments and those of Armstrong & Graveson, except to note that one set is performed on the tailbud of Rana, while the others concern the trunk somites of the axolotl. Finally, Elsdale & Davidson do not explain how so many somite primordia could be packed into such a small region at the posterior end of the tailbud, and they give no data on cell size. Nor do they speculate on the mechanism of extension and the consequences of this for the precise anterior-posterior ordering that they postulate. We do not believe that such a detailed axial pattern could exist in the posterior somitic mesoderm of Xenopus before its extension. There are too few cells to permit sufficient resolution and, in any case, precise pattern could not survive the vigorous cell rearrangement that our analysis reveals.

This work was supported in part by NSF Grant no. DMS-8618975 to P. Wilson and G. Oster, and by NIH Grant no. HD 18979 to R. Keller. We would also like to thank Paul Tibbetts for his assistance in preparing the plates.