Plasticity of transposed rhombomeres: Hox gene induction is correlated with phenotypic modifications

Segmentation of the nervous system of the vertebrate embryo is visible during a transient phase of development in the rhombencephalon where it appears as undulations in the internal surface of the neural tube designated as rhombomeres (r). Although first described more than a century ago (Orr, 1887; Vaage, 1969) their biological significance remained uncertain until recent cellular and molecular data demonstrated they were true segmental units of organisation in the hindbrain (reviewed by Wilkinson, 1993). As a part of rhombomere formation it was noted that homologs of the Drosophila Hom-C homeotic selector genes, the Hox gene clusters displayed segment-restricted expression. The anterior limits of Hox gene expression in the mouse rhombencephalon coincide with the limits between rhombomeres and presents a two-segment periodicity. Thus, for the Hox B complex the anterior limit of expression is between r2 and r3, r4 and r5, and r6 and r7 for Hoxb-2, Hoxb-3 and Hoxb-4 respectively, illustrating the rule of colinearity according to which the sequence of expression of the genes of a given cluster from rostral to caudal parallels their respective position from 3′ to 5′ on the chromosome (for reviews, McGinnis and Krumlauf, 1992; Wilkinson, 1993, Krumlauf, 1994). Thus each pair of rhombomeres expresses a specific combination of Hox genes generating a ‘Hox code’ which could provide a molecular identity to each segment of the hindbrain. In support to this, ectopic expression of Hoxa-1 causes a transformation of rhombomere 2 to an r4 identity showing that Hox genes can regulate segmental identity (Zhang et al., 1994). Also consistent with the molecular interpretation of segmentation, several other genes have been shown to have a segmentally restricted expression: those of zinc finger regulators such as Krox-20 (Wilkinson et al., 1989), of secreted signalling molecules such as fgf-3 (Wilkinson et al., 1988) and of protein kinases such as sek (Nieto et al., 1992).

Although the avian Hox homeobox genes have not been studied as extensively as their mammalian counterparts, several chicken or quail Hox genes have been isolated and, in their majority, they were shown to have a similar pattern of expression as in the mouse (Prince and Lumsden, 1994; Kuratani and Eichele, 1993; Guthrie et al., 1992; Izpisua-Belmonte, 1991). Recently, we have cloned all the members of the chicken Hox-a, -b and -d complexes and examined their expression in the hindbrain. All have identical anterior rhombomere boundaries of expression, but there is variation in the relative levels between segments (McNaughton et al., in preparation). This suggests conserved expression domains in vertebrates coupled to hindbrain segmentation.

The avian embryo was used in a number of investigations demonstrating the metameric nature of the transient segments defined as rhombomeres (Keynes and Lumsden, 1990; Lumsden, 1990; Fraser et al., 1990; Guthrie et al., 1993). Moreover, this model was crucial in the analysis of the important morphogenetic processes occurring in the rhomben-cephalon during embryogenesis. The rhombencephalic neural crest provides most of the mesenchymal cells which are at the origin of the hypobranchial region (see Le Douarin, 1982 for a review; Lumsden et al., 1991; Graham et al, 1993; Sechrist et al., 1993).

It was claimed, on the basis of heterotopic transplantation experiments, that the patterning of the skeletal branchial arches is, at least to a certain extent, under the control of the neural crest-derived mesectodermal cells (Noden, 1978). This view was further substantiated by the observation that in most cases the neural crest cells and the rhombomere from which they migrate express the same ‘Hox code’ (Hunt et al., 1991). Finally, the biological significance of the ‘Hox code’ in specifying the identity of both the rhombomeres and their neural crest derivatives was established in the mouse by the targetted mutation method. Null mutations of several Hox genes (Hoxa-1: Carpenter et al., 1993; Mark et al., 1993) (Hoxa-2: Gendron-maguire et al., 1993; Rijli et al., 1993) (Hoxa-3: Chisaka and Capecchi, 1991; Condie and Capecchi, 1994) result in severe anomalies or deficiences in the development of the hindbrain in some cases and also of the mesectodermal or neural derivatives of the corresponding neural crest (for reviews, McGinnis and Krumlauf, 1992 and Krumlauf, 1994). It was therefore of interest to see to what extent the ‘Hox code’ is intrinsically determined in the neuroepithelium of a given rhombomere during the early stages of neurogenesis. Previously, grafting experiments in the chick suggested that rhombomeres behaved as cell autonomous units in ectopic locations, maintaining their proper patterns of Hox expression (Kuratani and Eichele, 1993; Guthrie et al., 1992; Simon et al., 1995), and we wanted to examine this in more detail. For this purpose, we have performed heterotopic transplantation experiments of the presumptive territory of defined rhombomeres within the area of the hindbrain and medulla oblongata (from r1 to somite 7). We have taken advantage of the quail-chick chimera system (Le Douarin, 1969; Le Douarin et al., 1984) to identify precisely the grafted tissue in the chimeras and used the probes for Hoxb-1, b-3, b-4, a-3 and d-4 to assess its capacity to express a given set of Hox genes. Most of the experiments were done at the 5-somite stage i.e. before the formation of rhombomere boundaries. In order to know as precisely as possible the anteroposterior limits of the presumptive rhombomeres at these early stages, we have first established a prospective map of the rhombencephalic neuroepithelium by using DiI and carbon particles marks.

It appears that the transplantation of a given presumptive rhombomeric territory from a caudal to a rostral level of the anteroposterior (AP) axis does not perturb the expression of the Hoxb-4 gene in the grafts. In contrast, transposition of rhombomeres from a rostral to a caudal position results in the induction in the transplant of all the Hox genes tested, which are normally expressed in the graft site. This induction is still possible at stages following the establishment of rhombomere boundaries. We also show that the inductive signal diffuses tangentially in the neural epithelium. These results show that cell interactions play a significant role in maintaining the ‘Hox code’ at the levels of the neuroepithelium considered in this study.

Quail (Coturnix coturnix japonica) and chick (Gallus gallus) eggs from commercial sources were used throughout this study. Micro-surgery was performed on embryos at the 5-somite stage precisely unless otherwise indicated (about 30 hours of incubation in a humidified atmosphere at 38 °C). A window was cut in the shell and India ink diluted 1:1 in PBS was injected into the sub-blastodermic cavity in order to make the embryonic structures more visible without using any ‘vital’ stain. The vitelline membrane was windowed and a gap corresponding to a certain number of hemi-rhombomeres (as indicated in Tables 1 and 2) was made in the neural tube using a microscalpel, made by sharpening a steel needle on an Arkansas stone. The limits of the grafted tissues and of the sites of implantation are defined according to a map in which the anteroposterior limits of the rhombomeres were established as described below. Hemineural plates corresponding to a defined number of rhombomeres were dissected out from a quail donor. In most experiments, the neural plate was not dissociated from its surrounding tissues by enzymatic digestion. In some instances, possible contamination by mesodermal cells was eliminated by dissociation of the neural primordium after treatment with pancreatin (GIBCO). The notochord was left intact in the host embryos and in no case was it included in the graft.

In certain experiments, supernumerary notochord grafts were carried out. A fragment of either vagal, truncal or rhombencephalic notochord from a 10-somite donor was grafted in a lateroventral position, with respect to the rhombencephalon, by insertion into an incision along the full length of the rhombencephalic neural plate of a 5-somite recipient.

The operated embryos were killed in most cases at stage 14–15 according to Hamburger and Hamilton (HH; 1951). In some experiments, they were allowed to develop up to stage 20 or 34 (3 and 8 days of incubation respectively: E3 and E8). The fixative used was Carnoy’s fluid. Paraffin sections were cut frontally at 5 μm for the early stages (stages 14–15 and 20) and at 6.5 μm transversally for E8 embryos. Young embryos were then treated for in situ hybridization whereas alternate sections of E8 embryos were treated using the Feulgen-Rossenbeck (1924) technique, which enables a better visualization of quail nuclei, and cresyl violet, which gives a better definition of the structures.

Mapping of the rhombencephalon was carried out in both quail and chick embryos. 20 chick and 12 quail embryos were used. Although the embryos were carefully staged at 5 somites, they differed slightly in size. To obtain reproducible results despite this fact, the magnification was set in order to place the top of the head and the rostral limit of the first somite at defined postitions on a micrometric scale. The distance between these two markers was of 1600 μm±10%. The limit between r1 and the mesencephalon, already visible at the 5-somite stage as a notch in the neural tube, was located at 980 μm±10% from the tip of the head and 620 μm±10% from the anterior limit of the first somite (Fig. 1). Several incisions were then made in the neural tube in defined positions, measured on the scale with an resolution of 20 μm (length of half a graduation at this magnitude). Carbon particles or DiI (at a 0.05% concentration in 3 M sucrose in PBS) were then used to label the cells in the incision. Embryos were observed 24 hours later and the markers were assigned to the now formed rhombomeres. The limit between rhombomeres 6 and 7 corresponded always to the anterior limit of somite 1. Presumptive r1 had a size of 160 μm±20%. Presumptive r2 had a size of 125μm±20%. Presumptive r3 to r7 had a size of 85μm±25%. r8 whose posterior limit is undefined was not mapped (Fig. 1).

To generate suitable probes for the in situ analyses, we used the following cDNA fragments: a 500 bp BglII-EcoRI fragment of chicken Hoxb-1 subcloned in pGem3, a 400 bp EcoRI-SphI fragment of chicken Hoxb-3 subcloned in pGem4, a 1170 bp fragment of chicken Hoxb-4 subcloned in pBluescript, a 1300 bp fragment of chicken Hoxd-4 subcloned in pBluescript, a fragment of chicken Hoxa-4 subcloned in pBluescript, a 900 bp KpnI-EcoRI fragment of chicken Hoxa-3 subcloned in pBluescript KS+ (McNaughton et al., unpublished data). The RNA probes were labelled either by incorporation of ³⁵S-UTP (Amersham, 1000 Ci/mmol) during synthesis (Promega, Riboprobe Gemini II) or incorporation of dig-UTP (Boehringer Mannheim). The radioactive probes were then hydrolysed to generate 150 nt fragments. The sizes of probe fragments were checked by gel electrophoresis. The non-radioactive probes were used as full length transcripts.

Radioactive hybridizations were carried out as previously described (Eichmann et al., 1993). Probes were used at a minimum concentration of 10⁴ counts/min/μl. In order to obtain a well defined signal, the time of exposure was 7 days for Hoxb-4, 10 days for Hoxb-1, Hoxb-3 and 15 days for Hoxa-3. The sections were counterstained with Gill’s hematoxylin. Whole-mount non-radioactive in situ hybridization was performed exactly according to Wilkinson et al. (1991).

Previously we had established the expression pattern in the neural primordium of the 6 Hox genes used in this study (Morrisson et al., submitted; McNaughton et al., unpublished data) and we have verified their expression in quail in comparision with chick embryos from 5 somites to 3 days. The precise points of the expression pattern which relate to the study are summarised below.

At grafting time (5-somite stage) Hoxb-4 is strongly expressed in the mesoderm and more weakly expressed in the neural primordium at the AP level of the unsegmented region (Fig. 2A). At stage 14-15 (20-26 somite), the rostral limit of expression has spread forward up to the boundary between r6 and r7 (Fig. 2B), as recently shown also in the chick for its paralogue Hoxd-4 (Gaunt and Strachan, 1994). As in the mouse, it is noticeable that the expression is lower in r7 than in the rest of the tube. At stage 20 (3 days of incubation) the labelling in the external layer of the neuroepithelium in r7 and r8 becomes higher than at stage 14-15 (20-26 somite) (Fig. 10E). Moreover, the external layer of the alar plate is also labelled in r6 dorsally. The labelling patterns are identical in quail and chick.

The other genes were studied at stages 14–15 (20-26 somite) and 20 (3 days). Hoxd-4 and Hoxa-4 are expressed in the neuroepithelium like their paralogue Hoxb-4 at stage 14-15 (Fig. 6G,H). At stage 14-15, Hoxb-3 is expressed at a very low level in r5 and r6 and at a higher level more posteriorly. At stage 20 Hoxb-3 labelling is still low in r5 and r6, and higher in r7 and r8 and the spinal cord. The message is essentially concentrated in the external layer of the neuroepithelium at all levels (Fig. 10F). At stage 14-15, Hoxb-1 is expressed from r7 downward and in addition in a single band in r4 at a very high level (Maden et al., 1991; Sundin and Eichele, 1992). At stage 20, the labelling becomes more intense in the external layer of the neuroepithelium at all levels (Fig. 10G). Hoxa-3 is expressed at a high level in r5 and r6 and at a lower level from r7 to the tail at stage 14-15 (Fig. 6D).

Since the chick probe cross-reacts with quail mRNA, and since chick and quail Hoxb-4 expressions are identical, we made use of the quail-chick chimera system to investigate the expression of Hoxb-4 in heterotopic transplants of pairs of rhombomeres. The quail territory corresponding to r7 and the anterior part of r8 lateral to somite 2 (r8a), which will normally express Hoxb-4, was transplanted into a chick host to the position of r5 and r6, where expression of this gene is never observed. The grafts were unilateral so that the unoperated side serves as control (Table 1). Fig. 3 shows that the transplanted rhombomeres r7/8a exhibit a level of Hoxb-4 expression identical to control r7 and r8a. Therefore, at the 5-somite stage, this territory already has the information to express the Hoxb-4 gene and its expression is not negatively influenced by surrounding anterior tissues, which do not express this gene. The expression of Hoxa-4 and Hoxd-4 is also maintained in r7/8a transplanted in r5/6 (data not shown). These 3 genes were also maintained in homospecific transplantations, i.e. chick r7/8a grafted into chick at the r5/6 level (data not shown).

It has to be noted that in 20% of the cases observed 48 hours after the graft (n=9) Hoxb-4 was induced in the adjacent host epithelium.

It led us to see whether Hoxb-4 can be regularly induced in territories rostral to the normal limit of expression of this gene after transplantation to a more posterior level.

In order to answer this question, rhombomeres 5 and 6 were transplanted unilaterally at the transverse level of r7 and r8a (Table 1). These rhombomeres are normally not expressing Hoxb-4, but were found to express this gene as soon as 16 hours after the graft, i.e. at stage 14 (Fig. 4A-D). The level of expression was the same as in the controlateral side, i.e, high in r6 (r8 position) and lower in r5 (r7 position). When the r5/6 transplant was transferred more posteriorly to the level of r8p (r8 posterior, facing somites 3 and 4) (Table 1), Hoxb-4 was up-regulated in the transplant as in the contralateral side, i.e. at a high level in the whole transplant (data not shown). Whole-mount in situ hybridization was performed on 14 embryos that had received quail or chick transplants and they were then observed 24 hours after grafting (i.e. at stage 14-15) (Table 1). In 11 of them a blue coloration corresponding to the presence of Hoxb-4 in the graft appeared. The intensity of the coloration in the graft compared to the control side was the same in 3 cases (Fig. 5) and lower in 8 cases. In the 3 remaining cases there was no Hoxb-4 induction. The latter observation may reflect the incomplete incorporation of the graft within the host’s neuroepithelium, a situation difficult to appreciate in whole-mount staining. The induction was not different in chick and quail transplants. In some cases the observation was made 48 hours after the operation. Induction of Hoxb-4 was seen in all embryos. In addition, r5/6 grafted in r6/7 position became positive in r6 (r7 position) and remained negative in r5 (r6 position) like the contralateral control side (Table 1; Fig. 4E-H). Induction of Hoxb-4 in the grafted rhombomeres therefore respects the normal anterior boundary of expression of this gene, and expression of Hoxb-4 follows the same pattern as in the host’s rhombencephalon.

Treatment of the transplant with pancreatin prior to the graft in order to remove possible adherent mesenchymal cells has no effect on the induction (Table 1). However, the induction of Hoxb-4 was only observed in cases where the graft was perfectly integrated in the host neural tube, and when its anterior and posterior limits were in contact with the sections made in the host neuroepithelium, making its limits not visible except by referring to the nuclear structure of the implanted quail cells. This finding suggests that expression of Hoxb-4 is regulated within the neurectoderm by signals coming from surrounding tissues, i.e. from anterior, posterior or from lateral neuroepithelial cells. These rostral to caudal transpositions reveal that Hox expression does not always behave in a cell autonomous manner since the grafted rhombomeres express Hoxb-4 according to a pattern consistent with their novel position along the AP axis.

In order to know whether Hoxb-4 was also inducible in anterior rhombomeres, we transplanted r1/2 and r3/4 pairs into r7/8a or r8p (Table 1). In the case of r3/4 transplantations (Fig. 4I-L), Hoxb-4 was always induced in the transplant whereas for r1/2 the induction occurred in 7 out of 11 cases and was restricted to the posterior part of the graft when observed 24 hours after the graft (Fig. 4M-P). However, a complete induction of Hoxb-4 occurred in all the r1/2 to r8p grafts when the chimeric embryo was killed 1 day later, i.e. at stage 20 (3 days of incubation).

Then we addressed the question as to whether the induction can be obtained on grafted tissue taken at later developmental stages. Transplantations of r5/6 to r7/8a or r8p after the formation of rhombomere boundaries gave similar results to those described above (Table 1), showing that both the stimulus responsible for the induction and the capability to respond are still present at these stages.

The transplantation of rhombomeres 5/6 into the r8 position also leads to the activation of Hoxd-4 (Fig. 6G), Hoxa-4 (Fig. 6H) and Hoxb-1 (Table 1). Transplantation of r3/4 or r1/2 into r8p leads to the activation of Hoxa-3 (Fig. 6B-E) and Hoxb-3 (data not shown, Table 1). This shows that induction of all the Hox genes tested, which are normally expressed at this level, is dependent on the position along the body axis.

However, Hoxb-1 was never induced in quail r2 rhombomeres put into r4 (data not shown), as previously shown by others for chick to chick grafts (Kuratani and Eichele, 1993; Guthrie et al., 1992).

Our transplantation studies demonstrate that Hox expression patterns and possibly their segmental identity are at least partly determined by adjacent tissues. As schematized in Fig. 7, the induction may originate from the neuroepithelium either through transversal (Fig. 7A) or longitudinal signals (Fig. 7B). The lateral (Fig. 7C) or ventral mesoderm, i.e. the notochord (Fig. 7D) might also be responsible for providing the graft with new positional information.

To test a possible role of the contralateral neural tube, we performed bilateral grafts of transverse neuromeric fragments from the level of r5/6 to that of r8p (Table 2). The transcription of the Hoxb-4 gene was found to be up-regulated in the whole graft as it was in unilateral transplantations (Fig. 8A-D). Thus, a source other than the contralateral neuroepithelium must operate in these circumstances.

It was previously shown that the anterior notochord is able to induce En-2 expression in the neurectoderm more efficiently than the posterior notochord in explant association experiments (Hemmati-Brivanlou et al., 1990). The notochord might thus be at the origin of the induction of genes containing a homeodomain. In order to test its possible role in reprogramming Hoxb-4 expression in the grafted neural ectoderm, we implanted a supernumerary notochord fragment from the level of somites 1 to 8 lateroventrally to the presumptive level of r1 to r6 at the 5-somite stage (Fig. 8E-G). The presence of the extra piece of notochord originating from the region where Hoxb-4 is strongly expressed in the neural tube did not induce the activation of this gene in the anterior rhombomeres (Fig. 8H). The notochord by itself at this stage does not seem to be responsible for Hoxb-4 expression in transplanted rhombomeres.

Alternatively, Hoxb-4 could be induced by signals coming either laterally or longitudinally from the mesoderm or the neural tube located anterior or posterior to the graft. To distinguish between these hypotheses we grafted large rhombomeric territories extending from r2 to r6, both included in the position of the host’s somites 1 to 4 or 2 to 5 (Table 2; Fig. 8I-L). In these embryos, Hoxb-4 was induced in r5 and r6 (now placed at the level of r8) whereas r2,3,4 remained negative. This result shows that the inducing signal does not come from the tissues lateral to the graft (mesoderm, superficial ectoderm) since Hoxb-4 is inducible in short grafts of r2,3,4 placed in the same position. Therefore, the obvious interpretation of this result is that the inductive signal spreads in the graft along the AP axis from the posterior neural tube, inducing high levels of Hoxb-4 in r6 and lower levels in r5. This pattern schematized in Fig. 9, is the same when the chimera is observed at E3 (i.e. 2 days after grafting) (Fig. 10E). Although the host’s neuroepithelium expresses Hoxb-4 cranially to the graft, the signal in this case does not spread in a rostrocaudal direction since r2-4 remained negative. However, when the r2-6 fragment was grafted more posteriorly, lateral to somites 3-6 or 4-7 (Fig. 9), Hoxb-4 was induced in both r5 and r6 in all the 5 cases observed but also in the anterior part of the graft after 2 days (3 cases). Among these 3 cases, in 2 cases a central zone devoid of Hoxb-4 expression was observed in the explant. In the remaining case, transcripts of the gene where present in the whole grafted rhombomeres but with a central region where the transcripts were sparse. This indicates that the inducing signal(s), is present in sufficient amounts from the level of r8 downward. In r7 however, where Hoxb-4 is expressed at a lower level than posteriorly, the inductive capacity is not sufficient to induce Hoxb-4 expression in the grafted neuroepithelium.

The r2-6 grafted embryos, labelled with a Hoxb-3 probe show an enhancement of the signal in r6 and the posterior part of r5 (Fig. 10F). Moreover, whereas the expression of Hoxb-1 is maintained in r4, its posterior domain of expression does alter and is expanded rostrally to the posterior half of r5 and is up-regulated in r6 (Fig. 10G). There is thus an enhancement of the expression of the 3 genes described.

There are indications that the Hox code is responsible for generating segment identity. Since the Hox code was changed in our grafts, we wondered whether this change in gene expression pattern resulted in a change in rhombomere identity and further differentiation. We grafted r5/6 in the r8p position unilaterally and compared the nuclei of the hindbrain in the grafted and the control side at stage 34 (E8) on transverse sections stained with cresyl violet. The position of the graft was recognized on adjacent sections stained with Feulgen-Rossenbeck technique by the centronuclear mass of heterochromatin characteristic of the quail nuclei. In order to know the rhombomeric origin of the nuclei of the hindbrain, we previously performed 3 homotopic replacements of r5/6 at 5-somite stage and fixed the embryos at day 8 of incubation (Fig. 11A). The quail cells, i.e. cells of r5/6 origin, are found in particular in nuclei tangentialis (11F,G), laminaris (11C,D) and magnocellularis (11C,E) for the sensory nuclei, and nVI, nVII and nIX-X for the motor nuclei. Quail cells were never observed in the inferior olivary nucleus nor the nucleus of nerve XII and nucleus supraspinalis. These results are in agreement with previous observations made for motor nuclei (Keynes et al., 1990).

In chimeras where quail r5/6 were placed in the r8p position of a chick rhombencephalon (Fig. 11B), we investigated whether the nuclei in the graft were of r5/6 or r8 type. It appears that at the level of the graft the nuclei already formed at E8 (Harkmark, 1954; Tan and Le Douarin, 1991) are identical to the contralateral chick nuclei. The overall shape of the neural tube is identical bilaterally from the rostral (Fig. 11H) to the caudal (Fig. 11I) part of the graft. The X^th nerve roots are located at the same level on both sides at the rostral end of the graft (Fig. 11H) and XII^th ventral nerve roots at the caudal end of the graft (Fig. 11I). Nuclei of nerves IX-X and XII, nucleus supraspinalis (Fig. 11J,K,N,O) and both ventral and dorsal inferior olivary nuclei (Fig. 11L,M,P,Q) can clearly be distinguished. Moreover, nuclei characteristic of a r5/6 origin: n. angularis, n. tangentialis, n. laminaris, n. magnocellularis and n. abducens, are never observed in the graft. The phenotype of the transplanted rhombomeres is thus modified in the graft and converted from r5/6 to r8p. The phenotypic modification is thus in agreement with the genetic conversion.

One of the aims of this study was to systematically investigate whether the Hox-code which characterizes the hindbrain segments during development is intrinsically determined or if it can be altered in heterotopically transplanted rhombomeres. It is of interest if the latter alternative proves to be the case, whether the influence of the Hox-code on the subsequent development of hindbrain structures could be challenged. However, current information has suggested that rhombomeres behave in a strictly autonomous manner (Guthrie et al., 1992; Kuratani and Eichele, 1993, Simon et al., 1995). The findings in this study clearly demonstrate differential responses to environmental changes depending upon the type of rostrocaudal graft and suggest the presence of inducing signals which alter the identity of grafted tissue. This provides strong evidence for plasticity in the posterior hindbrain.

The heterotopic transplantations of the presumptive rhomomeres have in most cases been performed precisely at the 5-somite stage for both donor and host embryos. Quail and chick combinations have ensured that host and donor tissues can be precisely delineated when the results are analyzed whatever the stage considered, either by in situ hybridization or, later in development, when neural structures are recognizable. Since at the 5-somite stage, rhombomeres have not formed morphologically or become lineage restricted, a careful mapping of the hindbrain territory was first established in order to localize the level of the presumptive rhombomeres in both host and donor embryos. The fate map represented in Fig. 1 could thus be constructed. Certain tranplantations were also made later in development, after the formation of rhombomere boundaries, in order to test the temporal extension of the neural plasticity evidenced in earlier grafts. Our investigations have concerned essentially the gene Hoxb-4 and its paralogs Hoxa-4 and Hoxd-4, and also Hoxb-1, Hoxa-3 and Hoxb-3.

We found that expression of the Hoxb-4 gene is already programmed at the 5-somite stage in the neuroepithelium of the quail embryo even in its anteriormost area of expression (i.e. that corresponding to r7 and 8a). In fact, in general agreement with previous evidence for cell autonomy of Hoxb-1 in the tube and Hoxa-2 in the crest, when transplanted at the presumptive level of r5 and r6, the neuroepithelium corresponding to r7/r8a expressed Hoxb-4, Hoxd-4 and Hoxa-4 in a cell-autonomous manner, that is in a pattern characteristic of site of origin not of the ectopic location.

It was also important to note the fact that the quail hemineural tube transplanted anteriorly did not induce the r5/6 contralateral chick neuroepithelium to express the Hoxb-4 gene. Hence, signals that maintain Hox expression do not appear to move contralaterally in these conditions. It was easy to see that the Hoxb-4 transcripts were strictly restricted to the quail cells. Since 1 somite is added every hour, this determination takes place at least 9 hours before transcripts can be detected. Moreover, it is interesting to consider that this determination is intrinsic to the neural tube since complete removal of the mesoderm from the explant by pancreatin has no influence on Hox gene expression. These results are also in agreement with those described by Gaunt and Strachan (1994) showing that implantation of a glass barrier between the posterior domain of expression and its definitive rostral boundary does not prevent the expression of Hoxd-4 rostral to the barrier. Together, this indicates that if a posterior inducing signal is required to maintain or establish expression of the Hox genes it is in place very early before expression actually occurs.

Other regions of the chick neuroepithelium have been shown to maintain their expression when transplanted heterotopically. Such is the case for the En2 gene which was still expressed in transplants displaced from the mesto the diencephalic territory at the 10-somite stage (Itasaki et al., 1991; Gardner and Barald, 1991). The early commitment of the mes-metencephalic cells to express En-2 was further attested by transplantations carried out at the 3-5-somite stage to the rhombencephalic level (Marc Hallonet, personnal communication).

The fact that the regions of the neuroepithelium which normally express Hoxb-4 are committed early and behave autonomously when heterotopically transplanted to some regions, does not necessarily imply that expression of this gene is not inducible in other, normally non-expressing regions of the brain. In order to answer this second question, territories corresponding to the rostral hemi-rhombomeres 1/2, 3/4 and 5/6 were translocated caudally and grafted at the presumptive levels of the r7/8a or r8. Surprisingly, in all cases transcription of the Hoxb-4 gene was induced in the transplant at levels similar to those observed in the contralateral unoperated side. In the case of transplantation of r1/2 however, a delay of about 12 hours was observed before the maximal level of induction of Hoxb-4 was reached in the anterior region of the graft, indicating that competence to respond to inducing signals decreases in the anteriormost rhombencephalic neuroepithelium. These results clearly demonstrate that Hoxb-4 expression does not always display a cell autonomous pattern, and can be ectopically induced in other regions of the hindbrain.

Transplantation of rhombomeres 5/6 or 3/4 to r8 level were carried out to test if the expression of other Hox genes: Hoxd-4, Hoxa-4, Hoxa-3 and Hoxb-3 could also be induced and the induction of transcription in these grafts was observed for all of the Hox genes examined in a manner similar to Hoxb-4. This shows therefore that the neuroepithelium of the anterior rhombencephalon in which none of these genes is normally expressed has the competence to do so if subjected to the prevailing environment in the posterior hindbrain and spinal cord.

The question was then raised as to what in this environment is critical for Hox gene induction to proceed. The experiments described above rule out the possibility that the signal (or signals) acts transversally since in the caudorostral transplantation of hemirhombomeres from r7/8a to the r5/6 level the graft did not induce the contralateral unoperated rhombencephalic moiety to express Hoxb-4. Moreover, neither the notochord nor the paraxial mesoderm seem to be able to induce the changes in expression of the Hox genes involved in this study, at the stages considered. We noticed that one requirement is that the grafted fragment of tissue be perfectly incorporated within the plane of the host’s neuroepithelium. This suggested that the inducing signal might be transmitted in the plane of the neuroepithelium itself. In agreement with this idea, large stripes of neuroepithelium encompassing the presumptive territory of 5 rhombomeres, implanted posteriorly at the level of the medulla oblongata convincingly demonstrated that the inductive signal is transmitted from caudal to rostral via the neuroepithelium itself. We found that in the grafts transplanted at the level of r8 (Figs 9, 10) the range of action of the signal extend over the length of 2 to 3 rhombomeres. But if the graft was transplanted more posteriorly, the area of induction was larger and in some cases included the whole explant. Nevertheless, some cases showed that the signal can pass in a rostrocaudal as well as in a caudorostral direction. This was observed when 5-rhombomere fragments were grafted lateral to somites 3–6 or 4–7 (Fig. 10). In r3 a central zone devoid of transcripts was observed while induction had occurred on both ends of the transplant. Nevertheless, the induction by the anterior end needed a longer period of time to occur since it was observed after 2 days. These data suggest that the signals are strongest in posterior regions of the hindbrain or spinal cord.

One can speculate on the nature of the signals responsible for Hox gene induction and maintenance in definite regions of the neuroepithelium and on their propagation. Our experiments indicate that the signal spreads from the host neural tube to the graft within the plane of the epithelium itself. The fact that Hox gene expression is limited anteriorly in both the normal and transplanted epithelium might result from the decrease in concentration of the inductive signal(s) along the AP axis. Thus, if more posterior grafts (from r8p) are transplanted anteriorly (levels ranging from r1 to r6) some induction of the host neuroepithelium can occur (in 20% of cases) whereas r7/8a anterior translocations do not have inductive effect on the host. Moreover, the r8p induction of Hoxb-4 expression takes place only after a longer period of time (2 days) than in the experiments described above (data not shown). These results are in agreement with previous experiments (Alvarado-Mallart et al., 1990; Itasaki et al., 1991; Bally-Cuif et al., 1992, 1994) which show that transplanting the mes/metencephalon into the diencephalon leads to the acquisition of caudal characteristics by the neurectoderm surounding the graft. This supports the contention that the more posterior the fragments, the more inductive substance they contain. One can imagine that the cells induced to express Hox genes also acquire the capacity to produce the inductive signal and that a progressive decrease in the amount of substance generated at each step of its progression from caudal to rostral is responsible for fixing the anterior limit of Hox gene expression. Our experiments do not provide any insight concerning the nature of the signal. Retinoic acid could be a suitable candidate since it is expressed in a posterior to anterior gradient (Chen et al., 1994), it provokes anterior to posterior transformations of neural structures of Xenopus gastrulas (Durston et al., 1989) and mouse hindbrain segments (Marshall et al., 1992) and it has been shown to be a good inductor of Hox genes either directy or indirectly (for review, Mavilio, 1993). Furthermore, RARE in the Hoxb-1 gene are required to establish normal expression in the neural ectoderm, suggesting RA has a normal role in specifying Hox expression in the hindbrain (Marshall et al., 1994). Although the inducibility of several Hox genes is demonstrated in this work, the exclusive property of r4 for the anterior expression of Hoxb-1, already reported by others (Kuratani and Eichele, 1993; Guthrie et al., 1993), was confirmed in this study in the quail-chick system. Expression of Hoxb-1 does not obey the rules followed by most of the other genes of the Hox clusters since it is transcribed less rostrally than its position in the cluster would imply. Moreover, it is expressed from caudal up to r4 at the 5-somite stage and is subsequently strongly enhanced in r4 while later on its expression progressively decreases from r5 down to the spinal cord. In our experiments, Hoxb-1 can be induced in r5/6 transplanted to posterior locations (r7/8) whereas transplantation of r2 into a r4 position is not followed by induction of Hoxb-1 (Kuratani and Eichele, 1993; Guthrie et al., 1992; Simon et al., 1995). These differences in the anterior and posterior expression of Hoxb-1 might be accounted for by the fact that transcription of this gene in r4 seems to be controlled by different promoter elements than in its posterior domain of expression (Marshall et al., 1994) and that specific repressors block its activity in rhombomeres adjacent to r4 (Studer et al., 1994). Different promoter elements also regulate the expression of Hoxa-2 in anterior rhombomeres and r7/8 and spinal cord (Frasch et al., 1995) suggesting that the differential regulation between anterior and posterior rhombencephalon could be a more general feature. The possibility that the regulation of Hox gene expression in anterior and posterior rhombomeres follows different rules is currently being tested by transplantations.

Although homeogenetic induction had been noted in classical experimental embryology (Mangold and Spemann, 1927; Mangold, 1933) the most popular view has long been that neural induction proceeds essentially from vertical signal arising from the dorsal mesoderm (i.e. the notochord) to the neural plate. However, the experiments carried out by Nieuwkoop et al. (1952a,b,c), by Eyal-Geladi (1954) as well as the modern observation that neural genes are induced in the ectoderm of Xenopus exogastrulas (Kintner and Melton, 1987), rejuvenated the problem of planar induction in the neuro-epithelium. The reality of such a mechanism was definitively demonstrated in a number of experimental situations including Keller sandwiches and Xenopus exogastrulas (see Doniach, 1993 and Ruiz i Altaba, 1993 for reviews). Presently however, the respective role of planar versus vertical induction in normal development still remains to be defined. We provide here the evidence that maintaining the Hox code in the posterior rhombencephalon results from a posterior to anterior wave of induction which travels within the plane of the neuroepi-thelium itself.

The process of induction not only needs a signal but also competent cells. We have not yet looked for a possible anterior limit of competence in the neuroepithelium. Experiments are in progress to investigate this. It was also surprising from our results that older rhombomeric tissue was capable of plasticity. This suggested that the signals may be required throughout development and persist for some time.

One of the first indications that the Hox code controls the specification of the anteroposterior structures in vertebrates was provided by the administration of retinoic acid to pregnant mice and its consequences on homeotic transformations of the vertebrae (Kessel and Gruss, 1991). Targeted mutations of several Hox genes and their effect on the development of hindbrain and branchial arch neural crest derivatives have fully confirmed this notion. Here we provide the evidence that a homeotic transformation can be induced in the organisation of the brain stem derived from the rhombomeres. By homotopic transplantation of the quail r5/6 into chick hosts we have first shown that the derivatives of these rhombomeres correspond to the sensory acoustic nuclei, nucleus laminaris, magnocellularis and tangentialis, and to the motor nuclei of nerves VI (abducens) and part of IX. In addition, the derivatives of the posterior half of r8 correspond to the motor nuclei of nerve XII (hypoglossal) and X (vagus) partly. The inferior olivary nucleus is almost entirely contained in this region (Keynes and Lumsden, 1990 for a review and also Tan and Le Douarin, 1991). When r5/6 is transplanted at the level of r8p one sees the complete transformation of its presumptive fate into the neural structures characteristic of its novel position. Both the shape of the tube, the nerves which enter it and the nuclei are repatterned according to the level of the graft. As can be seen in Fig. 11 virtually no difference can be seen between the grafted and the control moiety of the brain stem at this level. The molecular changes induced by the graft thus result in a phenotypic change which is here compatible with the idea that a posterior prevalence operates in neural induction in vertebrates (Slack and Tannahill, 1992).

We are grateful to G. Couly for his help with the grafting and M. Catala for neuroanatomical advice. We thank Y. Rantier, F. Viala and S. Gournet for the illustrations and E. Bourson for typing the manuscript. This work was supported by the Centre National de la Recherche Scientifique, the College de France and the Association pour la Recherche contre le Cancer. Anne Grapin-Botton is funded by the Société de Secours des Amis des Sciences.