Development and characterization of commissural interneurones in the spinal cord of Xenopus laevis embryos revealed by antibodies to glycine

Antibodies directed against putative transmitter substances provide powerful tools with which to study the development of neurones because they can give information on the time of expression of the transmitter, on the distribution of developing neurones expressing the transmitter and, if transmitter is expressed early during morphological differentiation, on the way that particular classes of neurones grow axons and dendrites. The technique has been applied to Xenopus laevis embryos to study serotonin (van Mier et al. 1986) and substance-P (Gallagher & Moody, 1987). We have used antibodies directed against the products formed from y-aminobutyric acid (GABA) during glutaraldehyde fixation (Storm-Mathisen et al. 1983; Ottersen & Storm-Mathisen, 1984) to study the development of possible GABA-ergic interneurones (Roberts et al. 1987) and cerebrospinal fluid-contacting neurones (Dale et al. 1987b). In this paper, we have used a similar antibody directed against glutaraldehyde fixation products of glycine, the other principal inhibitory transmitter in vertebrates. Previously, we have described the methods used in its production and purification, and shown that this antibody labels commissural interneurones in the spinal cord of Xenopus embryos (Dale et al. 1986). These neurones have a soma and dendrites on one side of the spinal cord and an axon which crosses ventrally before branching to form ascending and descending axons on the opposite side (Roberts & Clarke, 1982). They are active during swimming, produce strychnine-sensitive inhibition in neurones on the opposite side of the cord and are therefore thought to be the class of spinal interneurone responsible for the glycinergic reciprocal inhibition present during swimming (Soffe et al. 1984; Dale, 1985; Roberts et al. 1986; Soffe, 1987).

The aims of the present paper are to extend observations on glycine-like immunoreactive cells (GLY cells) in the Xenopus embryo spinal cord. By looking at earlier stages of development we have been able to follow both the development of the whole population of GLY cells and also the way that individual GLY cells form their dendritic and axonal projections. Using the transmission electron microscope (TEM) we have examined the features of GLY synapses. Finally, we have made a careful comparison of the features of GLY cells with those of neurones in the spinal cord which had previously been defined by horseradish peroxidase (HRP) staining (Roberts & Clarke, 1982). It is nowhere an easy task to define spinal cord neurone classes but the GLY staining provides a new method and also allows us to look at development. The GLY cells seem to form a single population, revealed in its entirety, and offering new insights into the organization, features and variability of a single class of vertebrate CNS neurone. Our comparison with the earlier HRP evidence allows two types of spinal neurone with decussating (commissural) axons to be distinguished more clearly. These two classes of spinal neurone, one inhibitory the other sensory projection in function, are likely to be of widespread distribution among the vertebrates.

Embryos of Xenopus laevis were produced by induced breeding and raised at 20 ± 3 °C in aerated tapwater. After removal from their egg membranes, they were staged (Nieuwkoop & Faber, 1956), anaesthetized in Tricaine and fixed at 4°C. This was in 5 % glutaraldehyde in 0·1 m-phosphate buffer at pH 7·4 for lh for light microscopy, or in 2 % glutaraldehyde in 0·05 M-cacodylate buffer with 0·07 m-sucrose and 3mm-CaCl₂ at pH 7·4 for 10 h for electron microscopy. Once fixed, embryos were placed in 0·1 m-phosphate buffer with 0·1% glutaraldehyde and 0·04% sodium azide, where the central nervous system (CNS) was isolated by dissection and external pigment cells were removed. The isolated CNSs were then processed according to the protocol described in detail by Dale et al. (1986) using polyclonal antibodies raised against glycine fixed to protein by glutaraldehyde and revealed by the peroxidaseantiperoxidase technique. This antibody preparation (31), purified by immunosorption, specifically recognizes the glutaraldehyde fixation products of glycine, but not those of a wide range of other amino acids or free glycine or glycine incorporated in proteins (Dale et al. 1986; Ottersen et al. 1986). It has previously been found useful for discriminating cell populations enriched in glycine from ones enriched in closely related amino acids, including GABA, taurine and β-alanine, in sections of Xenopus and mammalian CNS (Dale et al. 1986; Ottersen et al. 1987).

For electron microscopy, CNS specimens were soaked in 30 % sucrose in phosphate buffer and frozen repeatedly in liquid N₂ to augment penetration before processing with antibodies without Triton X-100. For light microscopy, specimens were processed with Triton. Incubation with the primary antiserum, diluted 1:300 or 1:400, was for 5 to 11 days at 4°C (the long incubations were not essential but improved penetration and contrast), usually in the presence of 0·3M-glutaraldehyde-treated β-alanine. The presence of 0·1 DIM of similarly treated glycine abolished staining, whereas 0·3 mm of the glutaraldehyde complexes of β-alanine (or other amino acids, Dale et al. 1986) did not cause inhibition. Simultaneously processed test filters, carrying spots of different amino acids fixed by glutaraldehyde to macromolecules from rat or Xenopus CNS, also monitored the specificity of labelling (Fig. 3A). After staining, CNSs were dehydrated and mounted whole between coverslips in Canada balsam. For sectioning CNSs were dehydrated, embedded in wax or Araladite and cut at 6 or 15μm. All drawings were made with a camera lucida and details were checked under oil at ×500 or ×1000. For transmission electron microscopy, CNSs in Araladite were cut on an LKB Ultramicrotome, stained with lead citrate and uranyl acetate and examined in a Philips 200 microscope. The study is based on 115 embryos stained for glycine as follows: stage 21 (1), stage 22/23 (11), stage 24/25 (8), stage 26 (7), stage 27 (8), stage 28 (7), stage 29/30 (12), stage 31/32 (11), stage 33/34 (12), stage 35/36 (6) and stage 37/38 (32).

Embryos for HRP staining were immobilized with MS222 and dissected in physiological saline to expose parts of the spinal cord. HRP (Boehringer grade I) was then applied, either after drying onto a tungsten microneedle which was used to cut the marginal zone on one side of the spinal cord, or via an intracellular microelectrode inserted using a micromanipulator. After application, embryos were left for 10 to 120 min before fixation and processing using diaminobenzidine. Full details are given in Roberts & Clarke (1982) and Soffe et al. (1984). Specimens were mounted and viewed as whole CNSs as described above. The term marginal zone refers to the lateral regions of mainly longitudinal axons on either side of the spinal cord (see Fig. 6). It is here that synapses occur en passant from axons onto dendrites (Roberts et al. 1987).

The neural tube in Xenopus closes by stage 22 (Nieuwkoop & Faber, 1956) which is when the first glycine-like immunoreactive (GLY) somata were weakly stained in the rostral spinal cord and caudal hindbrain region (Fig. 1). At this stage only unipolar somata were seen. 2–4 h later at stage 25 the staining was stronger and ventral axons were clearly emerging from the somata. Ascending and descending growth cones on longitudinal axons in the marginal zone were also stained. As development proceeds additional GLY somata appear further caudally in the spinal cord and also between the somata that have already differentiated. These two processes, elongation of the column of GLY somata and increase in density of somata within the column, are particularly clear if stages 29/30, 31 and 33/34 are compared (Fig. 1). In each case, the most-caudal 10–15 cells are well separated from each other during their early stages of differentiation (see also below). More rostrally the spaces are filled by other GLY cells which differentiate later.

At earlier stages when GLY somata are more separated, up to stage 33/34, they could be counted reliably and Fig. 2 shows how their numbers increased with stage of development. At stage 37/38, counts could not be made accurately but one specimen showed numbers of 299 and 278 on either side and a cell density reaching a maximum of nearly 30 per 100 μm per side in the rostral spinal cord (Fig. 10A). Cell densities in the middle of the GLY cell column lay between 10 and 20.

Prior to stage 27, axons and growth cones were not well stained and the spinal cord did not clear well after dehydration. From stage 27 onwards, GLY axons and growth cones could be darkly stained (Fig. 3). This allowed the pattern of axon outgrowth to be followed in the whole population and also in individual GLY neurones (Figs 4 and 5). At the caudal end of the GLY cell column, some faintly stained somata without apparent axons were often present (e.g. Fig. 4B cell 3). Just rostral to these somata neurones were first well stained at the stage of differentiation when their axon and growth cone had reached the marginal zone on the opposite side via the ventral commissure. Examination of wholemount CNSs at stages 27–37/38 and sectioned material at stage 37/38 led to the following general picture of the development of GLY cell axon projections. At all stages GLY cell differentiation occurred later than (i.e. rostral to) muscle segmentation.

The first step in axon growth is the formation of a single axon passing from the unipolar soma in a broadly ventral direction (Figs 1, 4, 5 and 6). Transverse sections show that somata lying close to the medial edge of the marginal zone have axons running directly along this interface, while somata deeper within the cord have axons which initially grow ventrolaterally to reach the medial edge of the marginal zone (Fig. 6). The staining gave no evidence for the production of more than one axon from the soma, or for the outgrowth of axons in any direction other than ventral unless they were withdrawn before staining. All GLY cells had a single ventral axon crossing under the neurocoel to the opposite side (Figs 4 and 5). The floor of the neurocoel (floor plate) is formed by a single layer of ependymal cells 10–20 μm thick (Schroeder, 1970) and bounded externally by a basal lamina. The GLY commissural axons reach these ependymal cells by passing ventrally over the outer surfaces of other differentiating interneurones and motoneurones where they may also contact longitudinal axons in the marginal zone and radial ependymal cell processes. Ventral to the marginal zone the GLY axons lie in spaces between the ependymal cells, close to the external cord surface (Figs 5 and 6). They can be seen clearly in ventral views of whole spinal cords (Fig. 4C).

As GLY axons emerge from the floor of the neurocoel on the opposite side they tend to turn rostrally (Figs 4A,B and 5). GLY cells with growth cones that have just reached the opposite side were the earliest stage of differentiation to be well stained (Figs4A,B and 5C). The behaviour of GLY growth cones could therefore be inferred from observations on a series of individual cells drawn from this stage onwards (e.g. Fig. 5). In the vast majority of spinal cord GLY cells, the axon turned more rostrally after crossing. It then contacted the marginal zone and within the axon tracts growth cones turned rostrally to run longitudinally (Fig. 3). In a very few cases, the axon branched on reaching the ventral edge of the contralateral marginal zone to give an ascending and descending axon (Fig. 5A). At the far rostral end of the GLY cell column the polarity of axon growth was often reversed so that after crossing ventrally axons turned caudally and then grew caudally within the marginal zone (arrowheads in Fig. 4A,B). Finally, in one case, the initial segment of the axon of a GLY cell gave off a descending axon on the same side as the soma before continuing ventrally to cross in the ventral commissure (Fig. 5B). If we assume that these two axons had grown at the same rate then the shorter ipsilateral axon would have started to grow later than the commissural axon.

The observations on early stages of GLY cell differentiation help to explain the distribution of GLY axons and growth cones. The principal projection of GLY cells is directly to the opposite side, so GLY commissural axons are only found in the same longitudinal region as GLY somata (Figs 4, 6 and 7). Unexpectedly, most GLY cells initially project ros-trally (but see below) so more rostrally directed GLY growth cones are seen in the marginal zones and GLY growth cones and axons extend rostral to the GLY cell column (Figs 1, 4A,B, 5 and 7A). From stages 25 to 31 these rostral GLY axons are within the hindbrain but by stage 37/38 they reach into the midbrain (Figs 1, 4, 8 and 7A). Caudally directed GLY axons rarely extend much beyond the GLY cell column (Figs 1 and 3). At earlier stages the longitudinal axons of GLY cells lie in a fairly compact, central region of the marginal zone (Fig. 4A,B). However, as development proceeds more axons are added until by stage 37/38 GLY axons are present throughout the whole dorsoventral extent of the marginal zone (Figs 6 and 7).

The longitudinal GLY axons do not appear to branch. To see if they made synaptic contacts, some specimens at stage 37/38 were prepared for TEM Transverse ultrathin sections of the spinal cord showed that tissue preservation in these specimens was not good. However, dark GLY axonal profiles were present in all regions of the marginal zone and some of the larger of these in the middle of the zone were in clear presynaptic relationship with unstained dendritic profiles (Fig. 8). Fifteen such synapses had closely apposed regions of thickened pre- and post-synaptic plasma membrane, dense cleft material, and presynaptic vesicles in the GLY profile.

The dendrites of GLY cells were generally not clear in lateral views of whole-mounted spinal cords because they extended in the transverse sectional plane. However, dendrites were seen extending radially into the marginal zone from the initial segment of the axons in a few cells at early stages of differentiation (e.g. Fig. 5C,D). Later, such dendrites were obscured by overlying GLY stained axons. In transverse sections at stage 37/38, dendrites in the marginal zone could be seen more clearly. Most lay in the middle to ventral part of the zone (Fig. 6). Transverse sections also clarified the location of GLY somata (Figs 6 and 7). They were generally adjacent or deep to the dorsal half of the marginal zone and the column was often two cells deep. GLY somata were frequent in the spinal cord just dorsal to the marginal zone and ventral to the dorsal tract in what has been called a ‘dorsolateral’ position (e.g. Figs 6A,D,F and 7K,J,H; Roberts & Clarke, 1982). In this position they lie very close to the external cord surface.

If horseradish peroxidase (HRP) is applied to the marginal zone on one side of the spinal cord, cells are labelled on the same side and also rostral and caudal to the application on the opposite side of the CNS (Fig. 9A,B). All these contralateral cells have ventral commissural axons. When examined in more detail many of these cells are similar to the GLY cells (Figs 10B and 11A,B,C), in the following features: (1) unipolar soma in dorsal half of spinal cord, (2) most dorsal somata superficial in ‘dorsolateral’ position, (3) somata form longitudinal columns sometimes two cells deep, (4) initial segment of axon is ventrally directed and extends dendrites radially into marginal zone, (5) ventrally axon crosses to the opposite side very close to the cord’s external surface, (6) on the opposite side the axon turns longitudinally in the marginal zone to ascend and (7) some axons branch when they reach the contralateral marginal zone (Fig. 11A,C). If cells with these features are counted in heavy HRP fills, fairly close to the HRP application site, the maximum numbers of filled cells correspond quite closely with the number of GLY cells found in the same region of the CNS (Fig. 9C). Since cells are also filled rostral to the application of HRP (Figs9A,Band 11), these cells must have descending axons. Are these the result of axon branching or do some axons only descend? A group of 27 similar cells have been filled with HRP by intracellular injection (Soffe et al. 1984; Dale, 1985; Fig. 11D). Of these cells, 24 branched on the opposite side, 2 ascended and 1 descended. This suggests that the majority of these neurones, nearly 90%, branch and therefore have both ascending and descending contralateral axons. The intracellular HRP fills also showed two cells, about 10%, with additional axons on the same side as the soma. In one case, the ipsilateral axon descended, in the other there was a descending and an ascending ipsilateral axon. Branching and axons on the same side were also seen in GLY cells (Fig. 5). Finally, by looking at the distance from which HRP applications can fill these cells, we can estimate the probable length of axons. These distances were therefore measured in 10 embryos for 935 cells and suggested that descending axons ranged from 600 to 1000 μm in length, while ascending axons were from 500 to 700 μm.

HRP applications filled other cells that did not share the features listed above. In the hindbrain, the Mauthner cell and vestibulospinal cells were filled (see also van Mier & ten Donkelar, 1984; Nordlander et al. 1985). In the spinal cord, there only appeared to be one other class of cell filled from the opposite side, ‘dorsolateral commissural’ interneurones (Roberts & Clarke, 1982). These cells are filled mainly caudal to the HRP application and lie exclusively in the ‘dorsolateral’ position, like the GLY cell marked by an arrowhead in Fig. 6D. The examples drawn in Fig. 10B show that they differ from GLY cells in the following features: (1) multipolar soma tending to be elongated longitudinally, (2) prominent dendrites arising from the soma and lying in dorsal quarter of marginal zone including dorsal tract, (3) dendrites mainly oblique or longitudinal but not radial and (4) initial segment of axon narrow and generally without dendrites. Cells with these features are also illustrated in figure 4a, b, c of Roberts & Clarke (1982) and figure ID of Clarke & Roberts (1985). These cells are not nearly as numerous as GLY cells, the density shown in Fig. 10B being typical of a heavy HRP fill.

Spinal neurones with ventral commissural axons are a general feature of the embryonic vertebrate nervous system (e.g. Barron, 1944; Kuwada, 1986; Ramon y Cajal, 1890, 1952; van Gehuchten, 1889; Wentworth, 1984; Whiting, 1948). In Xenopus embryos, spinal neurones projecting to the opposite side have been separated anatomically into two classes: ‘commissural’ and ‘dorsolateral commissural’ interneurones (Roberts & Clarke, 1982). ‘Commissural’ interneurones were characterized anatomically by their unipolar somata, radial dendrites from the initial segment of the axon and ventral commissural axon (see Table 1). When interneurones were identified by intracellular marker injection, all ‘commissural’ interneurones recorded were rhythmically active during swimming and those tested produced strychninesensitive inhibition of other spinal neurones (Soffe et al. 1984; Dale, 1985). The physiological evidence suggests that ‘commissural’ interneurones inhibit rhythmically active contralateral neurones (motorneurones, descending interneurones and ‘commissural’ interneurones), some rhythmic ipsilateral neurones, and ipsilateral ‘dorsolateral commissural’ interneurones (Roberts et al. 1986). We have now shown that interneurones with the anatomical features of ‘commissural’ interneurones have glycine immunoreactivity, further endorsing our conclusion that these neurones are the reciprocal inhibitory interneurones, active during swimming and using glycine as a transmitter (Dale et al. 1986). In contrast, present evidence suggests that ‘dorsolateral commissural’ interneurones are excitatory interneurones which are themselves excited by primary sensory Rohon-Beard neurones, and act as sensory projection neurones which carry excitation to the motor system of the opposite side to mediate crossed flexion reflexes (Clarke & Roberts, 1985; Sillar & Roberts, 1988). These neurones are not active during swimming but receive rhythmic inhibition which gates sensory inflow (Clarke & Roberts, 1985; Sillar & Roberts, 1988). The two classes of spinal interneurone with contralateral projections therefore seem to perform two very separate functions: reciprocal inhibitory and excitatory sensory projection. It is very likely that homologous cell classes will be found in most other vertebrate groups when more detailed analysis is performed. Suggestive evidence is already available. For example, silver and Golgi staining in sheep and mice embryo spinal cords shows commissural neurones dorsally near the entry of the dorsal root afferents (possible sensory projection neurones) but also in the ventral motor regions of the spinal cord (possible reciprocal inhibitory neurones) (Barron, 1944; Wentworth, 1984).

The original distinction between ‘commissural’ and ‘dorsolateral commissural’ interneurones was made on simple anatomical grounds (Roberts & Clarke, 1982). A basic criterion used in making this distinction was that the somata of ‘dorsolateral commissural’ interneurones lay in a superficial dorsolateral position, dorsal to the main marginal zone in the region of the central axons of sensory afferents. However, two types of evidence now suggest that this position is not a very useful distinguishing feature. First, some intracellular recordings made in this position show neurones that spike rhythmically during swimming (K. T. Sillar, personal communication). Second, the present antibodies show that some neurones in this position have glycine-like immunoreactivity (Figs 6 and 7, arrowheads). In view of this evidence we have made a careful comparison of the features, distribution and numbers of GLY neurones and ‘commissural’ interneurones defined on the basis of HRP staining (Roberts & Clarke, 1982; Soffe et al. 1984; Dale, 1985). This suggests that some ‘commissural’interneurones which have large initial axon segments with radial dendrites lie in a superficial dorsolateral position (e.g. fig. 4d,e and g of Roberts & Clarke, 1982). The dorsolateral soma position can therefore no longer be used as a defining criterion. The two decussating interneurones must therefore be separated by anatomical differences in the soma, dendrites and initial segment of the axon, by the GLY staining (Table 1), and by their physiological responses.

It is clear from extracellular and intracellular HRP staining of ‘commissural’ interneurones and the present GLY staining that the members of this population show a range of form, axonal projection pattern and soma position (e.g. Figs 4, 5, 6 and 7). One puzzle is that the majority of individual developing spinal GLY neurones that we observed had ascending contralateral projections, whereas in all intracellular fills of ‘commissural’ interneurones the contralateral axon branched. There are at least two explanations for this apparent discrepancy. First, we could only observe the axonal projection patterns of GLY cells at the caudal end of the cell column and at early stages in their differentiation. It could be that cells in this position in the cell column have a higher probability of forming ascending projections than those differentiating by in-fill in more central positions in the column. Second, it is quite possible that after forming an ascending projection the axon could branch later to form a secondary descending projection (see below and Roberts et al. 1987). These two explanations could be equivalent if they simply result from a tendency of ‘commissural’ neurones to form projections within the limits of their own differentiating cell column. More data on individual cell morphology are required to test some of these possibilities. Variability in cell features could be used to define subsets, for example the most-rostral members of the column with mainly descending projections might be regarded as reticulospinal neurones if seen in isolation (see also Nordlander et al. 1985; van Mier & ten Donkelaar, 1984). However, at present there seems insufficient anatomical or physiological evidence to justify subdivision of what looks like a homogeneous column of neurones with a common reciprocal inhibitory function.

GLY neurones were first seen at stage 22/23 in the caudal hindbrain (Fig. 1), the same location in which the first neurones showing GABA-like immunoreactivity appear at stage 25–26 (Roberts et al. 1987). The first neuronal processes, probably axons, are found in the same region 2–5 h earlier at stage 20 (Hayes & Roberts, 1973). By injecting blastomeres at the 32cell stage Jacobson & Huang (1985) have shown that this is also the stage when their ‘commissural’ interneurones, Rohon-Beard neurones and ‘dorsal longitudinal’ neurones begin to grow processes. Their ‘commissural’ interneurones initially have unipolar somata with a single ventrally directed growth cone. Since ‘commissural’ interneurones have not stained clearly with the glycine antibody at this early stage of axon outgrowth, either the technique is not sufficiently sensitive or the glycine is not strongly expressed at the earliest stages of axon outgrowth. This contrasts with some classes of cells with GABA-like immunoreactivity and axons ipsilateral to the soma where transmitter is expressed early so the emergence of the first growth cone from the soma could be seen (Dale et al. 1987). Since GLY staining was not usually seen until axons had reached the opposite side, perhaps glycine is not strongly expressed until growth cones have crossed the floor of the neural tube and reached potential synaptic targets.

After the first GLY neurones appear in the hindbrain, differentiation proceeds caudally and by in-fill in more rostral regions, as is the case for at least two other classes of spinal neurones with GABA-like immunoreactivity (Dale et al. 1987; Roberts et al. 1987). By the time that swimming locomotion is established at stage 30 there are about 60 commissural interneurones on each side of the CNS (Fig. 1). These neurones are known to play an important role as reciprocal inhibitory neurones during swimming (Roberts et al. 1986).

If we accept the identity of the ‘commissural’ interneurones seen by Jacobson & Huang (1985), the evidence from glycine staining can be combined with theirs to assemble a picture of the development of the commissural interneurones (see Figs 4, 5 and 6). The first step is the production of an axonal growth cone in the transverse plane which grows laterally if necessary (Fig. 6D), then ventrally at the level of the outer surface of neuronal precursor cells. In this position, the growth cones lie between the external processes of ependymal cells (figs 3 and 4 in Taylor & Roberts, 1983; Holley, 1987; Holley & Silver, 1987; Nordlander & Singer, 1982; Scott & Bunt, 1986). At later stages when a marginal zone of axons has formed, this ventral growth is at the inner edge of the marginal zone into which dendrites are produced (Fig. 5B,D). In the ependymal cell layer forming the floor-plate of the neural tube, dendrite production does not occur but the axon grows to the opposite side, sometimes fasciculating with others of its own kind (growing in the same or opposite direction, see Fig. 4C). On reaching the opposite side, the growth cones tend to turn rostrally (except for the most-rostral members of the population which turn caudally) to run longitudinally in the marginal zone and make synapses. Either immediately (Fig. 5A) or at a later stage of growth the contralateral axon forms a caudal branch (Fig. 11). A small proportion of neurones also grows one or more secondary axons on the same side as the soma (Fig. 5C).

A number of clear features emerges from this account. First, the initial stage of growth cone outgrowth is in a very definite orientation. This is now emerging as a very general feature of inital outgrowth of specific cell groups in vertebrates (Davies et al. 1982; Halfter et al. 1985; Jacobson & Huang, 1985; Eisen et al. 1986; Dale et al. 1987a; Holley & Silver, 1987; Roberts et al. 1987). Second, in the three classes of spinal neurone whose development we have examined using transmitter immunocytochemistry, the direction of initial axonal projection is in very many cases determined before contact is made with other neurones within the same class (Figs 1 and 4; Dale et al. 1987; Roberts et al. 1987). We could conclude that many cells within a class are therefore pioneers. However, it is more likely that there are not specialized pioneer neurones but that the whole population can respond to the same pathway cues and grow in the ‘correct’ direction (see also Holley & Silver, 1987). Third, when the axon enters the floor-plate of the floor of the neural tube its behaviour changes. Before entering, it grows circumferentially around the spinal cord and produces dendrites. After crossing, it grows along the spinal cord and makes synapses. (In 10 % of cases after crossing, a secondary ipsilateral axon grows and extends on the same side as the soma). It is therefore possible that contact with the single layer of ependymal cells that form the floor of the neural tube (floor-plate, Schroeder, 1970) could transform the ‘commissural’ neurones in a way that changes their behaviour. This type of process has been suggested by Dodd et al. (1987) who have shown in rat embryos that the cell surface expression of glycoprotein on the axons of one type of commissural neurone is modified at the point of contact with floor-plate cells. Fourth, after reaching the opposite side, the behaviour of growth cones is more variable, they can ascend, or branch to ascend and descend, or descend only if they are from rostral cells. However, all grow longitudinally over a broad range of dorsoventral positions within the marginal zone (Fig. 6). Fifth, it is clear that the initial pattern of axonal growth is not final. Like the ‘mid-hindbrain reticular’ neurones with GABA-like immunoreactivity (Roberts et al. 1987) ‘commissural’ neurones can form later branches and secondary axons on the same or opposite side to the soma. Finally, it is clear for the GLY cells that they do not have a very powerful tendency to fasciculate with other axons of their own kind. Both as they decussate (Fig. 4C) and as they grow longitudinally (Fig. 4A,B) the axons can contact other axons and also separate from them. However, in neither situation are tight fascicules of many axons formed, suggesting that nonaxonal substrates are equally or even more adhesive (see also Holley & Silver, 1987).

Some of the features just outlined are currently being investigated by detailed examination of GLY-stained ‘commissural’ intemeurones. What is most promising is that the choices open to growth cones are now beginning to be defined for vertebrate CNS neurones in the ways that have already been possible in some invertebrate preparations (Bastiani et al. 1985).

We thank J. Ablett, A. T. Bore, M. Shannon, K. Stentoft, L. J. Teagle and J. Line Vaaland for assistance and Drs J. D. W. Clarke, P. Gordon-Weeks, T. Jessell, R. Nordlander and K. T. Sillar for advice on the manuscript. This work was supported by the SERC, MRC, Company of Biologists, Norwegian Research Council for Sciences and Humanities, Norwegian Council on Cardiovascular Disease and Norwegian Society for fighting Cancer.