Neural tube closure in Xenopus laevis involves medial migration, directed protrusive activity, cell intercalation and convergent extension

Neurulation is the process by which progenitors of the central nervous system are shaped, separated from and brought beneath the epidermis. The cellular basis of these movements have been investigated in amphibians (Jacobson, 1981; Jacobson and Gordon, 1976; Keller et al., 1992b), chicken (Schoenwolf and Smith, 1990b; Smith and Schoenwolf, 1997) and mice (Bush et al., 1990; Sausedo and Schoenwolf, 1994; Smith et al., 1994) but remains poorly understood. In the chick, which has been used extensively as a paradigm for vertebrate neurulation, a broad neural plate folds and the margins of the plate are raised and brought into apposition at the neural folds as cells take on stereotypical shapes. Analysis of serial sections and electron microscopy have identified cell movements and shape changes accompanying neural tube formation in chick (Schoenwolf and Alvarez, 1989; Schoenwolf and Smith, 1990a) where these cell behaviors (apical contraction and interkinetic nuclear migration) result in the formation of a ‘medial hinge’ overlying the notochord and a ‘dorsal lateral hinge’ near the prospective sulcus limitans, that together bring the neural folds into opposition. The epidermis then fuses, the neural ectoderm fuses and the neural crest is released from neural epithelium of the newly formed neural tube. In this manner, it is thought that the flat neural plate rolls into a tube with the lumenal face of the neural tube forming from the apical face of the neural plate.

However, additional processes appear to be involved in chick neurulation. By following the rapid events at the start of neurulation, van Straaten and coworkers (1996) found that the lateral face of the two neural folds ‘zip’ into apposition, starting near the floorplate and proceeding dorsally. They found that the lumen of the neural tube nearly disappears after apposition and then re-opens to form the lumen after the neural folds fuse. Another group, using high-resolution electron microscopy have identified distinctive cell morphologies at the time of fusion of the neural folds (Lawson and England, 1998). Thus, while it is likely that cell shape changes and neural plate bending establishes the ventral and intermediate aspects of the neural tube, more complex and as yet undefined events are involved in forming the dorsal neural tube.

The revelations in chick inspired us to re-investigate the mechanisms of neurulation in the frog. Based on sectioned material and light and electron microscopy, Schroeder (1970, 1971) outlined four processes that he thought would ‘constitute the ultimate mechanisms of neurulation’. (1) Superficial cells of the neural epithelium change from cuboidal to bottle-shaped initiating the formation of the neural groove. (2) Presomitic mesoderm and lateral deep neural plate cells elongate to help elevate the neural folds. (3) Both superficial and deep epidermis ‘migrate’ medially bringing the neural folds into apposition thus aiding closure of the neural tube. (4) Extension of the underlying notochord prevents an anteroposterior shortening of the neural plate that would result after apical contraction of superficial neuroepithelial cells.

However, the closure and formation of the neural tube also involves complex movements of radial intercalation of several layers of deep cells with the superficial layer of the multilayered neural anlagen of the frog (Schroeder, 1971). These cell movements and the identities of the cells involved are poorly understood, particularly at the lateral margin of the neural plate, which is the source of three distinct tissues, the epidermis, which will form the dorsal skin, the neural crest, and the dorsal neural tube (Schroeder, 1970). The role of these dorsolateral populations in neural tube formation are difficult to evaluate because of the low spatial resolution of current molecular marker methods. Correlation of cell identity with motility is difficult because fluorescence-based imaging of cell behaviors and tissue movements is largely incompatible with RNA in situ hybridization techniques.

To solve this problem, we have modified the RNA in situ hybridization protocol (Harland, 1991) using a recently synthesized fluorescent substrate of peroxidase (Kerstens et al., 1995) and have developed a streamlined whole-mount preparation for confocal microscopy. Using this method allows epidermal, neural crest and dorsal neural tube gene expression patterns to be visualized with the same high-resolution confocal techniques used for immunofluorescence and lineage analysis and is generally useful to analyze gene expression patterns in the context of cell and tissue shapes.

We show that the mediolateral organization of the early neural plate is thrown into disorder by the rapid rolling movements of the superficial neural ectoderm and fusion of the neural folds over the neural groove. Immediately after fusion, markers of the prospective dorsal neural tube cells are still found far lateral to the medial site of fusion, whereas at the midline, the expected site of the neural lumen is occupied by multiple layers of deep mesenchymal cells lying above the floorplate. The dorsal neural tube with a well-defined lumen and roof is reformed after a complex set of cell movements: medially directed migration and intercalation of neural crest, and radial intercalation of deep neural cells. These movements occur for several hours after neural fold fusion to create a single-cell-layered neural tube characteristic of vertebrates.

Embryos were obtained by standard methods (Kay and Peng, 1991) and staged according to Nieuwkoop and Faber (1967). Albino embryos were used for all in experiments. For fluorescence-based imaging, it is important not to use Nile blue to add contrast for staging. Nile blue contains the dye Neutral red, which fluoresces in the rhodamine channel and does not wash out of embryos (in contrast to the blue components of Nile blue) in organic solvents. For general histology of cell and tissue shapes, all the cells in the embryo were labeled by injection of 1.5 nL from a stock solution (25 mg/ml of water) of rhodamine-dextran amine (RDA; anionic lysine-fixable; Molecular Probes) at the 1-cell stage. Scattered, labeled cell populations were made by injecting single blastomeres at stages 6 or 7 when there are between 100 and 300 cells. Embryos used for both histology and in situs were fixed overnight at 4°C in MEMFA (0.1 M MOPS, pH 7.4, 2 mM EGTA, 1 mM MgSO₄ and 3.7% formaldehyde) and stored at −20°C in 100% methanol. For time-lapse recordings, embryos were mounted in custom chambers at stage 17 in 1/3× Modified Barth’s Solution with 0.1% bovine serum albumin. Single embryos were pierced transversely on a short piece of a thin plastic ‘spear’ (1 mm long by 20 μm in diameter). The ends of this plastic spear were pressed into silicone grease (Dow Corning) by coverglass fragments producing an embryo solidly fixed in position whose dorsal surface is centered and slightly pressed against a coverslip. All embryos pierced in this manner develop normally.

Low-light, time-lapse recordings were collected using a Hamamatsu C2400-008 SIT camera, an Olympus IX-70 inverted microscope, a Uniblitz shutter and a Metamorph imaging system (Universal Imaging Corp.). From these time-lapse recordings, we quantified the angular protrusive activity of individual cells (see Elul et al., 1997). Briefly, the outline of a single cell was traced over the course of a time-lapse recording (from 40 to 100 frames). The pixel areas of cellular protrusions from subsequent frames were accumulated into each of twelve 30° sectors centered on the center of mass of the cell. The angular protrusive activity for a cell was calculated as the percentage of the total ‘protrusion’ area falling within each sector. The angular protrusive activity of cells within the same domain of gene expression was calculated as the mean of the percentages for all the cells within that domain. This analysis was carried out using macros written by the authors for NIH-Image (version 1.61; http://rsb.info.nih.gov/nih-image/).

We have made a number of modifications to the Harland protocol (Harland, 1991). Antisense RNA probes were transcribed (Ambion), labeled with either digoxigenin or fluorescein UTP (BMB) and used without being hydrolyzed. Embryos in 100% methanol were rehydrated through a series of 25:75, 50:50 and 75:25 PBS:methanol washes. Proteinase K digestion was not done. Hybridization was at 60°C overnight and the post-hybridization RNase treatment was skipped with little or no change in sensitivity. In several trials, the potential activity of endogenous peroxidases was reduced by incubating embryos for 1 hour with 1% H₂O₂ in PBS, but no change in background was observed. Embryos were then washed, blocked and incubated with peroxidase-coupled (POD) fab fragments (BMB) directed against digoxigenin or fluorescein in maleic acid buffer with 2% BMB blocker and 20% heat-inactivated goat serum. With the substitution of POD for alkaline phosphatase, this protocol was the same as that used in a recently revised protocol (Knecht and Harland, 1997).

Samples are equilibrated in POD reaction buffer (PBS, 0.1 M Imidazole pH 7.6, and 0.001% H₂O₂) for an hour and reacted with either tyramide-fluorescein or tyramide-rhodamine (TSA-Direct GreenFISH and RedFISH, NEN Lifesciences) for 30 to 60 minutes. Other tyramide-conjugated substrates are available commercially as kits or may be synthesized in the laboratory (Hopman et al., 1998; Jacobs et al., 1998). Fluorescence signal was increased with one to three rounds of POD-mediated deposition with fresh tyramide-fluorophore in buffer. After deposition of insoluble substrate, the embryos were washed extensively for up to 2 days in PBS to remove unreacted soluble tyramide-fluorophore revealing the expression pattern. A second fluorophore may be used by incubating the embryos for 45 minutes with 1% H₂O₂ in PBS to deactivate the first POD, followed by repeating the above protocol with an appropriate POD-conjugated fab fragment against the epitope carried by the second RNA probe. Prior to confocal sectioning, preliminary assessments of the outcome of the in situ reaction were made in whole embryos in PBS using an epifluorescence equipped stereoscope (Olympus, SZH10).

Embryos from fixative or 100% methanol were ‘softened’ by incubating for 20 minutes in PBS and 0.01% Tween-20 (embryos processed for RNA in situs do not require softening) and then bisected transversely with a scalpel under a standard stereoscope. Tailbud-stage embryos could be cut into several pieces. These pieces were then dehydrated in 100% methanol and placed in disposable chambers constructed by stacking 10 to 15 clear plastic paper reinforcement rings (Avery cat. no. 05722) to form a well on a no. 1g thickness (0.17 mm) coverslip. The embryos were then cleared by replacing the methanol with benzyl benzoate and benzyl alcohol (BB:BA; 2:1). Samples were oriented as they cleared with their cut side facing the microscope objective and the chamber sealed with another coverslip. This chamber remains intact for only a few hours as the BB:BA slowly dissolves the adhesive. For future confocal sessions, embryos can be retrieved immediately from these chambers, washed and stored in 100% methanol. Samples were optically sectioned immediately after mounting using a confocal scanning laser system attached to an inverted compound microscope (20×, 0.70 n.a. objective, Olympus or Nikon) at the W. M. Keck Center for Cellular Imaging (Biology Department, University of Virginia).

n-tubulin encodes a class II β-tubulin that is expressed in prospective motor neurons and prospective dorsal sensory neurons, including Rohon-Beard cells during midgastrulation in two to three mediolateral stripes (Chitnis et al., 1995). Xpax3 encodes a transcription factor that is expressed in prospective neural crest, dorsal neural tube, somitic and lateral plate mesoderm in chick, mouse and frog (see figure 2D in Bang et al., 1997; Goulding et al., 1991, 1993). xash3 encodes a transcription factor that is expressed in cells marking the prospective sulcus limitans (Ferreiro et al., 1994; Turner and Weintraub, 1994; Zimmerman et al., 1993). xk81 encodes a cytokeratin that is expressed in prospective epidermal cells (Fouquet et al., 1988; Jamrich et al., 1987; Jonas et al., 1985).

Shortly after the completion of gastrulation (stage 13), the neural plate in the prospective trunk region consists of two layers of cells (Fig. 1A). At this stage, there is no clear morphological boundary at between the prospective neural plate and the prospective epidermis in sectional view. At late neural groove stages (stage 17), the neural folds rise above the notoplate (see Jacobson and Gordon, 1976), that part of the neural plate in contact with the notochord, forming a groove (Fig. 1B). These folds come into close apposition over the neural groove (stage 18; Fig. 1C) and fuse, enclosing a very small ventral neural lumen (stage 19; Fig. 1D). In some cases the entire lumen disappears after fold fusion (data not shown). Over the next several hours, the lumen is reconstructed from this rudiment, progressing from ventral to dorsal as radial intercalation (i.e. deep cells extending to the lumenal surface and superficial cells extending to the basal surface of the neural tube) brings the two layers of the early neural plate (Fig. 1E) into a single-layered neural tube (Fig. 1F).

Fusion of the neural folds over the neural groove creates a jumble of cells that can now be identified. Expression of xk81, an epidermal cytokeratin, shows that the cells at the fusing lips of the neural folds are epidermal (Fig. 2A) and that xk81 expression is limited to the outermost, single cell layer over the open neural plate and early neural tube. The deep layer of the epidermis is restored over the neural tube only later (see arrowheads in Fig. 1F). There are occasional, single-cell gaps in the expression of xk81 in the epidermis (data not shown) that may mark prospective epidermal ciliary cells (Chen and Grunz, 1997).

xslug, xpax3, n-tubulin and xash3 gene expression patterns (Fig. 2C,E,G,I) were used to identify other dorsal cell types at neural fold fusion. Before fold fusion, genes such as xslug and xpax3 are expressed far lateral to the site of fusion (Fig. 2C,E) but later are expressed in the dorsal aspect of the neural tube (Fig. 2D,F). The same is true of the dorsal limit of n-tubulin expression (Fig. 2G,H). In contrast, a mediolateral marker of the neural plate, xash3, is found much closer to its eventual position in the neural tube (Fig. 2I,J). Immediately at apposition or just after neural fold fusion, in each of these cases, the cells over the neural tube lumen do not express genes normally associated with the dorsal neural tube. Later, however, after the neural tube has formed a complete lumen, the dorsal aspect of the neural tube expresses its definitive pattern of gene expression. Thus, either gene expression is shifting from lateral to more dorsal cells or the cells themselves are moving.

Time-lapse videorecordings of scattered, fluorescently labeled deep cells at the margin of the neural plate in the whole embryo revealed medially convergent movements of large groups of individual cells (Fig. 3A) just below the epidermis. The embryo was fixed after the last frame of the videorecording and medially migrating cells were visualized in transverse section using a confocal microscope (Fig. 3B). A comparison with the fluorescent in situ revealed that these cells express xslug (Fig. 3C). A dual image of the cells and xslug expression shows that the leading cell in this section approached the midline by the end of the time-lapse recording (Fig. 3D). Mapping these cells back to their location when the neural folds fused shows that the shift in gene expression reflects the medial movement of prospective neural crest cells after the folds fuse.

The thin epidermis allowed us to record the medial movements of several cells within the xslug-expressing neural crest domain (Fig. 4A). Cells within this domain exhibit monopolar protrusions directed toward the midline (Fig. 4B) as they migrate between the basal surface of the epidermis and more ventral neural tube cells. At the completion of neural fold fusion, these cells are isodiametric (data not shown) but soon become mediolaterally elongated (i.e. transverse to the axis of the embryo). Once they approach the midline, they return to a more isodiametric shape (0:30 panel of Fig. 4A). These neural crest cells remain quiescent in this medial location until they start to emigrate from the tube (data not shown). Occasionally, neural crest cells overshoot their midline target and remain on the other side (Fig. 4C).

In this period of medial migration and convergent extension, more medial (i.e. ventral) cells of the neural tube show bipolar protrusive activity. Prospective dorsal neurons (expressing n-tubulin) immediately beneath the neural crest, are also medially protrusive even though their lateral face is bound to the outer surface of the neural tube (Fig. 5A). These two cells extend protrusions in a bipolar manner directed mediolaterally (Fig. 5B). These tracked cells represent the dorsalmost population of the neural tube, a domain of n-tubulin-expressing cells (Fig. 5C), with their lateral end on the outer surface of the neural tube and their medial end protruding beneath neural crest cells (Fig. 5D).

The medial movement of deep dorsal cells is paralleled by the radial intercalation of more ventral cells in the neural tube. Before neural fold fusion only the most ventral neural deep cells have interdigitated but have not yet intercalated radially between the more superficial cells surrounding the neural groove (Fig. 6A). Radial intercalation begins first among cells in the ventral tube (Fig. 6B) and progresses to more dorsal cells after fusion (Fig. 6C). This progression of radial intercalation coincides with the ventral- to-dorsal reconstruction of the neural lumen. The ventral floorplate is the first stable portion of the small round lumen immediately after the neural folds fuse (Fig. 6D). The lumen becomes flask shaped with new deep cells intercalating into the wall of the neural tube extending the dorsal aspect of the lumen (Fig. 6E). Finally, the dorsalmost surface of the lumen is completed and the single-cell-layered pseudostratified neural tube is formed (Fig. 6F).

Time-lapse recordings of medial migration of dorsal cells of the neural tube and transverse confocal sections over the course of neural tube formation reveal that medial migration of dorsal neural tube cells contribute to the convergent extension of the dorsal neural tube. Convergent extension movements have previously been characterized in the neural plate by an anteroposterior dispersal of cells as local protrusive activity brings the cells medially (Elul et al., 1997; Keller et al., 1992a). These same features are seen in the time-lapse video(Fig. 3A) as an initially solidly labeled group of cells break up (unlabeled cells intercalate between labeled cells) as the group nears the midline. The results of convergent extension are also revealed in the decreasing number of cells in the neural tube visible in transverse section over the course of neural fold fusion and relumenation (compare the neural anlagen in Figs 1B,F, 6D,F).

Analysis of neural tube closure, or its failure (Copp, 1994; Copp et al., 1990), has been hindered by poor definition of the cellular motility involved in this process. Here we use confocal microscopy, a newly developed whole-mount fluorescent in situ RNA hybridization method and correlated time-lapse recording of cell behaviors to characterize cell motility involved in neural tube closure in Xenopus laevis.

Whole-mount RNA in situs showed prospective neural crest lies far lateral to the point of fusion of the neural folds and time-lapse recordings revealed the behaviors used by these cells to move to their definitive positions in the neural tube (Fig. 7). Immediately after fusion of the neural folds, the dorsal neural tube consists of mesenchymal cells derived largely from the deep cell layer of the neural plate. Then, the neural crest, arising from the lateral deep cell layer of the two-layered neural plate, begins migrating medially, using monopolar, medially directed protrusive activity. As the crest and other lateral cells migrate medially they intercalate between more medial cells, predominately along the anteroposterior axis, and thus elongate and narrow the future dorsal aspect of the neural tube. This pattern is reminiscent of the mediolateral cell intercalation seen earlier in the neural plate (Fig. 7A; Elul et al., 1997; T. M. Elul and R. E. K., unpublished data) and in the mesoderm (Keller and Winklbauer, 1992), which in each case produces convergent extension. At the outset of medial migration, the entire population of cells involved consist of a multilayered mesenchyme without a lumen. Through the process of relumenation, involving medial migration and radial intercalation (Fig. 7C; see below for discussion) the dorsal neural tube forms a single-cell-layered, pseudostratified epithelium surrounding a lumen (Fig. 7D).

Without medial migration and narrowing in the mediolateral direction, the very wide anlagen at the time of neural fold fusion could not form a properly shaped neural tube. The monopolar, medially directed protrusive activity seen in the neural crest cells is consistent with the directed movement of these cells toward the dorsal midline and, indeed, such activity is always correlated with convergence movements. Thus, this specific type of directed motility is likely responsible for shaping the dorsal neural tube.

A medially directed, monopolar protrusive activity strongly resembling that seen here, is expressed earlier by deep cells in the open neural plate and is thought to underlie the mediolateral intercalation movements of cells that produce convergent extension (T. M. Elul and R. E. K., unpublished data). Whether the activity seen in the dorsal neural tube is a continuation of this earlier activity or a new and independently regulated behavior is not known. Although similar in appearance, the two behaviors may have different mechanisms of directionality; the earlier activity is directed toward the midline of the open neural plate, i.e. the notoplate, while the protrusive activity in the dorsal neural tube is directed toward the future roofplate. However, even that notion of directionality is not entirely accurate as we occasionally observe neural crest cells overshooting and crossing the midline. Such behavior is not consistent with a dorsal midline cue that attracts or directs protrusive activity medially. However, the few cells seen crossing the midline might represent the earliest stages of neural crest cell emigration from the neural tube, which is known to involve crossing the midline (Krotoski et al., 1988).

The potential cues for the monopolar, directed migration could come from the overlying epidermis, from the cells initially at the dorsal midline of the neural anlagen, or from cells still lateral to the neural tube. In the first case, one would envision the cue being a directional one on the undersurface of the single-cell-layered epidermis in this region. In the second case, one could envision a signal emanating, probably diffusing, from cells initially at the midline, or perhaps even from the neural cells last in contact with the epidermis. In the last case, one could envision an inhibitory signal, rather than a protrusion-stimulating signal, from tissues lying off the lateral margin of the neural plate, including the deep layer of the epidermis, or the lateral plate mesoderm, or at later stages, the somitic mesoderm.

Surprisingly, immediately after the epidermis of the neural folds meet and fuse, the bulk of the neural anlagen consists of a large population of deep mesenchymal cells that do not immediately form a lumen. It is only after neural fold fusion that a lumen reforms. Relumenation involves two general forms of motility. First, the medially directed monopolar motility described above reforms the roofplate. Second, radial interdigitation and subsequent radial intercalation of the two cell layers of the neural anlagen occur during relumenation of the ventral aspect of the neural tube and result in formation of a single-layered tube.

Interestingly, this process of radial intercalation, as well as the radial interdigitation that precedes it, begins ventrally at the midline, and progresses dorsally. Whether it is coincidence that radial intercalation begins ventrally where the lumen of the neural anlagen persists or whether the lumen cues the process of radial intercalation is not known. It appears that radial intercalation occurs at the same time as relumenation. How and why these processes are linked is the subject of further investigation.

The progression of radial intercalation behavior from the ventral midline towards the dorsal was somewhat of a surprise and has implications for patterning. The directed cell protrusive activity driving convergent extension of the neural plate begins anteriorly and laterally near the lateral border of the hindbrain, and proceeds medially (future ventral) and posteriorly (Elul and R. E. K., unpublished data). This is, of course, the reverse of the progression seen in radial intercalation. If the progression of these behaviors reflects when the participating cells received the signals to execute the movements, then the earlier mediolateral cell intercalation behaviors are organized by signals emanating from lateral and anterior and progressing medially (ventrally with regard to the future tube) and posteriorly. In contrast, it could be that the radial intercalation described here is organized by a signal emanating from the ventral midline. Alternatively, it could simply be dependent on conditions that mature progressively in the neural anlagen from ventral to dorsal. Previous work by Hartenstein (1989) described a progression of both cell division and radial intercalation from dorsal (lateral) to ventral (medial).

Time-lapse recordings of cell behavior in the thick dorsal region below the neural crest in a region undergoing radial intercalation shows bipolar protrusive activity directed both medially and laterally. Such behavior could be the principle form of motility underlying the radial intercalation of multiple layers of deep cells found in this region. Why the monopolar behavior is not used in this region is not known. Bipolar protrusive activity is expressed by deep neural cells as they mediolaterally intercalate during their convergent extension under conditions of planar induction (Elul et al., 1997); however, the bipolar form of intercalation may not be used during convergent extension under both planar and vertical inducing signals (T. M. Elul and R. E. K., unpublished data).

The reformation of the lumen of the neural tube appears to involve a transition from a deep mesenchymal cell to an epitheloid cell type. The dorsal neural anlagen is a jumble of interlaced deep cells after neural fold fusion but as relumenation occurs, the cells establish what appear to be stable surfaces at their lumenal and basal faces, forming a single-layered array of cells. The behavioral and morphological evidence presented above suggests that a mesenchymal- to-epithelial transition is a major process in relumenation and thus closure of the Xenopus neural tube. We are examining this hypothesis with an ultrastructural and immunocytochemical analysis of junctional complexes during neural tube formation.

We believe that radial intercalation and correlated relumenation depends on neural fold fusion. In previous work, ‘giant explants’ of the neural anlagen converged, extended and formed neural folds and grooves under both planar and vertical neural-inducing signals through the late midgastrula stage, but the neural folds produced by these explants would not fuse and radial intercalation did not occur (Poznanski et al., 1997). Here we show that radial intercalation does not occur until after neural fold fusion. In another study A. Edlund and R. E. K. (unpublished data) show that neural plates held open beyond the time of normal neural fold fusion likewise do not show radial intercalation of deep and superficial cells. These facts argue that there is some signal or condition associated with neural fold fusion that cues or enables radial intercalation. The fact that radial intercalation fails when folds are mechanically held open argues that radial intercalation depends directly on fold fusion and eliminates the possibility that both processes depend on yet a third signal.

Chordates neurulate using a variety of mechanisms. In the amphioxus, Branchiostoma floridae, the epidermal cells and neural ectoderm completely separate well before the neural tube forms and epidermal cells begin to crawl over the surface of the neural ectoderm (Holland et al., 1996). As the epidermis moves over the neural ectoderm, the neural plate itself begins to fold. Once the epidermis has fused, the neural plate bends at the midline and brings the edges of the plate into contact and fusion. In this case, the processes of fusion of the epidermis and fusion of the neural ectoderm have been uncoupled from each other. It is unclear whether cells in the neural plate begin directed medial migration of the type seen in Xenopus, or whether a mass tissue movement such as convergent extension or folding takes place under the already fused epidermis.

In zebrafish, neurulation involves transformation of the neural plate into the neural keel (Papan and Campos- ortega, 1994). Keel formation appears to proceed in a ‘cell-by-cell’ manner as cells on opposite sides of the neural plate move medially, come into apposition and then move ventrally together, side-by-side to form the neural keel. Two apposing rows of cells lie on either side of the midline, but cells may divide with daughter cells crossing over to the other side of the keel (Kimmel et al., 1994). The keel appears to be a mesenchymal mass of cells, which subsequently cavitates to form the neural lumen. Relumenation, as in Xenopus, appears to start ventrally at the floorplate and proceed dorsally to the roofplate. It is unclear when the epidermis fuses over the top of the neural keel.

Both chick and mouse form their posterior neural tube (secondary neurulation) by cavitation (Costanzo et al., 1982; Schoenwolf, 1984; Schoenwolf and Delongo, 1980). The morphogenetic movements during this secondary neurulation in mouse and chick might bear some resemblance to the sorting out of mesenchymal cells in Xenopus. Moreover, the process of cavitation in these embryos may share mechanisms with the process of relumenation seen in Xenopus. Although, in the case of secondary neurulation, the lumen is reconstructed from the roofplate down to the floorplate.

Our work in Xenopus is the first instance in which the cell behaviors underlying neural tube closure have been descibed in detail. It is likely that, when broken down into their constituent cell behaviors, these other diverse forms of neurulation may begin to resemble neural tube closure in Xenopus. The separation and fusion of the epidermis in amphioxus might resemble the separation and fusion of the neural folds over the neural groove in Xenopus. The mesenchymal mass of neural keel cells in zebrafish may resemble the disorganized mass of mesenchymal cells in the dorsal aspect of the neural tube in Xenopus. The relumenation in Xenopus from the ventral floorplate to the dorsal roofplate might also resemble cavitation in the zebrafish neural tube as well as cavitation in secondary neurulation in mouse and chick. Just as the various forms of neurulation are driven by similar cell behaviors, the cell behaviors themselves may be driven by similar molecular pathways. The identification of these molecular components and their relationship among the vertebrates is only just beginning.

Further elaboration of these molecular components and their roles in driving neurulation require both sophisticated histology and fluorescence-based imaging techniques that allow characterization of cell behaviors, tissue movements and protein localization. Our fluorescent in situ RNA hybridization methods have allowed us to visualize epidermal, neural crest and dorsal neural tube gene expression patterns set in the context of cell and tissue shapes. However, in situs merely reflect gene expression and say little about the localization and function of their encoded proteins. The next steps that we take will involve expressing constructs encoding fluorescent chimeras of candidate molecules to dynamically image these molecules and explore their role in neurulation.

We would like to thank Anna Edlund, Tamira Elul, Max Ezin, Paul Skoglund, and Dave Shook for critical discussions and assistance. We would also like to thank C. Kintner for the pax3 and n-tubulin probes, D. Anderson for the xash3 probe, R. Grainger for the xslug probe and T. Sargent for the xk81 probe. This work was supported by NIH grant HD25595 to Ray Keller and an American Cancer Society Postdoctoral Fellowship to Lance Davidson.