Direct visualization of nucleogenesis by precerebellar neurons:involvement of ventricle-directed, radial fibre-associated migration

In the vertebrate central nervous system, there are important functional units called layers and nuclei. Structures such as the cerebral cortex and the hippocampus have a layered neuronal organization, whereas nuclei are aggregates of neurons dispersed throughout the central nervous system. Neurons in nuclei and those in some layers have common physiological and anatomical features, including projections to and from other functional units, which is essential for brain functions.

Mechanisms in the development of stratified structures have been well documented (reviewed by Hatten,1999; Marin and Rubenstein,2003; Komuro and Yacubova,2003; Bielas et al.,2004). However, much less is understood about nucleogenesis. Recently, it has been proposed, in the midbrain, that periodic patterns of molecularly distinct stripes that appear to be induced by sonic hedgehog and FGF8 play an important role in the formation of the red nucleus and oculomotor nucleus (Agarwala and Ragsdale,2002). Concentration-dependent induction by morphogens such as sonic hedgehog and FGF8 may explain the formation of laminated or striped structures (reviewed by Goulding and Lamar, 2000), but cannot fully explain the emergence of discretely distributed nuclei.

Neuronal migration is another potential factor in nucleogenesis and its role is well established in the formation of stratified structures. In the cerebral cortex, for example, excitatory neurons that are born in the ventricular zone migrate along the radial glial cells that span the cerebral wall (Rakic, 1972; Hatanaka and Murakami, 2002)(reviewed by Kriegstein and Noctor,2004) and settle in specific lamina in a birthdate-dependent manner (Rakic, 1974). Later-born cells migrate past earlier-born cells, forming inside-out laminated structures that lie parallel to the pial surface(Angevine and Sidman, 1961; Berry and Rogers, 1965).

Here, we examined how neuronal migration leads to establishment of nuclei,focusing on the hindbrain where clear nuclear organization exists. A number of studies have addressed the migration of hindbrain neurons and several genes affecting their migration have been identified (reviewed by Hatten, 1999; Wingate 2001; Chandrasekhar, 2004; Sotelo, 2004). Yet, cellular basis for nucleogenesis is still poorly understood. Specifically, how neurons with common properties accumulate in a defined position remains unknown. We have chosen mossy fibre-projecting precerebellar (PC) nuclei as a model,because their origin and development is well documented(Altman and Bayer, 1997; Rodriguez and Dymecki, 2000; Wang et al., 2005). Previous studies have suggested that neurons of the PC nuclei are derived from the lower rhombic lip (lRL), the germinal neuroepithelium surrounding the alar recess of the fourth ventricle (Harkmark,1954; Taber-Pierce,1966; Tan and Le Douarin,1991; Altman and Bayer,1997; Rodriguez and Dymecki,2000; Wang et al.,2005), and migrate circumferentially. Although these studies examined the origin, migratory route and the fate of alar plate-derived neurons, the cellular processes leading to nucleogenesis were not established,owing to the lack of techniques that allow in vivo visualization of individual neurons migrating from a specific site of the neural tube. To clarify these issues, we introduced the enhanced yellow fluorescent protein (EYFP) gene into the lRL of mouse embryos and visualized migration and nucleogenesis of mossy fibre-projecting PC neurons.

ICR mice were obtained from CLEA JAPAN (Tokyo). The day that the vaginal plug was detected was considered embryonic day (E) 0.5. Pregnant mice were deeply anesthetized with pentobarbital (50 mg/kg), and embryos were removed from the dams. All experiments followed the Osaka University Guidelines for the Welfare and Use of Laboratory Animals.

The whole coding regions for EYFP (Clontech) and Venus(Nagai et al., 2002) were cloned into the pCAGGS vector (Niwa et al., 1991), yielding pCAG-EYFP(Saba et al., 2003) and pCAG-Venus, respectively. For inhibition of cadherin function, pCAG-Ncad(t)that carried a gene encoding an intracellular domain of N-cadherinfused to enhanced green fluorescent protein (EGFP) was prepared (H.T, D.K., K. Nishida and F.M., unpublished). A vector carrying Venus was kindly provided by Dr A. Miyawaki (BSI, Riken).

Exo utero electroporation was performed as previously described(Saba et al., 2003)(Fig. 1A) with some modifications. In brief, plasmids were diluted in 0.01% Fast Green or Indigo carmine (final concentration, ∼2 μg/μl) to monitor the injection and was injected into the fourth ventricle of an embryo exposed by cutting the uterine wall.

Three electric pulses (20 V, 50 ms) were applied with an electroporator(ECM830, BTX). Plasmids were unilaterally introduced into the lRL of E12.5 mouse embryos. The vector, once incorporated into the ventricular neuroepithelial cells, should be present in their descendents. Embryos with unilaterally labelled lRL were selected by epifluorescence microscopy (SZX12,Olympus). Hindbrains were removed from these embryos and immersed in 4%paraformaldehyde in 0.1 M sodium phosphate buffer at pH 7.4 (4%paraformaldehyde) overnight at 4°C. The fixed hindbrains were imaged using a cooled CCD camera equipped with epifluorescence illumination (VB-G25,Keyence). Alternatively, the hindbrains were embedded in OCT compound (Sakura Finetechnical) after immersion in 30% sucrose in PBS. Sections (18-20 μm)were cut on a cryostat (HM-500, Zeiss) and mounted on slides (12-550-15,Fisher Scientific). The distribution of EYFP- or Venus-labelled cells was analysed using a confocal microscope (TCS SP2 AOBS, Leica).

The vector carrying the Mbh2 gene (AB043980; GenBank) was linearized and used as a template for RNA synthesis. An RNA probe for Mbh2 was subsequently synthesized using a digoxigenin (DIG)-11-UTP mixture (Roche) and T3 or T7 RNA polymerase (Roche), and purified with Quick spin columns (Roche).

Cryosections from the hindbrain of E14.5-18.5 mouse embryos prepared as described above were dried at 50°C for 3-5 hours. The dried sections were refixed in 4% PFA for 5 minutes, and then treated with 0.3% hydrogen peroxide in PBS for 30 minutes. After acetylation with 0.25% acetic anhydride in 0.1 M triethanolamine, the sections were probed with DIG-labelled Mbh2 cRNA probes (2 ng/μl) in hybridization solution (50% formamide, 5× SSPE,5% SDS, 200 μg/ml yeast tRNA) for 16 hours at 70°C. After washing, the sections were treated with RNase A (20 μg/ml; QIAGEN) for 1 hour at 37°C. Samples were reacted with a peroxidase-conjugated anti-DIG antibody(1:200, Roche) in Tris-buffered saline containing 1% blocking reagent (Roche). The tyramide signal amplification system (PerkinElmer Life Sciences) was then applied following the manufacturer's instructions.

A rabbit polyclonal anti-green fluorescent protein (GFP) antibody (1:1000,Molecular Probes) was used on sections of EYFP- and Venus-labelled hindbrain tissues. To detect EYFP/Venus denatured by in situ hybridization procedures,the sections were reacted with the anti-GFP antibody for 2 hours at room temperature. After a washing with 0.1% Tween-20 in PBS, the signal was visualized with an Alexa488-conjugated anti-rabbit IgG antibody (Molecular Probes).

Radial glial fibres were identified using an anti-nestin antibody (rat401)(Hockfield and McKay, 1985). The antibody was diluted 1:5 in 0.3% Triton-X100 in PBS (PBST) and applied to 30-40 μm sections that had been pretreated with 0.3% hydrogen peroxide in PBS for 30 minutes at room temperature. After overnight incubation at room temperature, sections were rinsed in 0.3% PBST and reacted with biotinylated anti-rat IgG antibody (1:200, Jackson ImmunoRes). Sections were then treated with Vectastain Elite ABC kit (PK-6100, Vector) followed by the tyramide signal amplification method as described above.

For birthdating studies, intraperitoneal injections with 5′-bromo-2′-deoxyuridine (BrdU) (Sigma; 5 mg/ml in sterile saline)were made in EYFP-electroporated pregnant mice at 50 mg/kg body weight to label embryos at E12.5, E13.5 and E14.5 (n=5 at each stage). To detect incorporated BrdU, 25 μm transverse sections of the pons were prepared. These sections were incubated in 4 N HCl for 45 minutes, rinsed in 0.2% PBST and reacted with a rat monoclonal anti-BrdU antibody (Cat#OBT-0030, 1:200, Oxford Biotechnology) for 2 hours at room temperature. After extensive washing with 0.2% PBST, the sections were incubated with a biotinylated anti-rat IgG antibody (1:200, Jackson ImmunoRes) for 2 hours at room temperature followed by visualization with Alexa594-conjugated streptavidin. Subsequently, the EYFP protein that had been denatured by the HCl treatment was stained with an anti-GFP antibody.

Captured images were thresholded for quantification using Leica confocal software (Ver 2.61). The total number of EYFP-labelled cells(N_total) and the number of EYFP-labelled cells that were also positive for BrdU (N_BrdU) were counted in the entire PGN/RTN on the side of the contralateral to the DNA injection. The ratio of BrdU-positive cells to all EYFP-labelled cells (N_BrdU/N_total) was calculated.

The hindbrain of E15.5 mice was removed and immersed in a fixative consisting of 4% PFA (TAAB) and 0.2% glutaraldehyde (EM Science) in 0.1 M sodium cacodylate (Nacalai Tesque). The preparation was further fixed in 2%glutaraldehyde at 4°C overnight followed by postfixation in 2% OsO4 and then subjected to conventional procedures of electron microscopy. Thin sections were cut coronally. The plane of section was carefully chosen so that radially running fibres can be observed in the sections. Electron microscopy was carried out in Hanaichi Denshi Kenbikyo Laboratory.

After electroporation of vectors to the lRL at E12.5, embryos were allowed to survive for 2-6 days outside the uterine wall. Brain development in such embryos was indistinguishable from that seen in normal embryos. Analysis of the lRL in all brains showed that unilateral labelling was successful.

The neurons of the external cuneate nucleus (ECN) and the lateral reticular nucleus (LRN) are generated earlier (E10.5-13.5) than those of the pontine grey nucleus (PGN) and reticulotegmental nucleus (RTN) (E11.5-16.5)(Taber-Pierce, 1973). The former migrate circumferentially to give rise to the posterior extramural migratory stream (PEMS), while the latter take a more anteriorly oriented route to form the anterior extramural migratory stream (AEMS)(Altman and Bayer, 1987a; Altman and Bayer, 1987b). Direct visualization of lRL-derived cells largely confirmed these previous findings (Fig. 1B,C; n=16). Fig. 1B shows the dorsal view of the E14.5 mouse hindbrain expressing EYFP in the right lRL. A rostroventrally directed stream of EYFP-labelled cells emerged from the lRL. Migrating cells that formed the PEMS were barely visible in the dorsal hindbrain at this stage, perhaps because they had already emigrated (arrowhead in the inset of Fig. 1B). Indeed in the ventral view, neurons in the PEMS are clearly visible(Fig. 1C, arrowhead marked PEMS). These observations show that cells leaving lRL take two specific pathways: one towards the metencephalon and the other towards the myelencephalon.

Although labelling of more ventrally located regions of the ventricular zone was unavoidable, careful examination of the cases in which the gene transfer was dislocated ventrally revealed the presence of few labelled cells occupying either the AEMS or the PEMS (data not shown). The lRL is therefore the major source of neurons migrating in the AEMS/PEMS.

EYFP-labelled cells migrated immediately beneath the pial surface(Fig. 1D,E, left panels). These observations, together with those from whole-mount preparations, strongly suggest that the labelled cells are PC neurons. To characterize the EYFP-labelled cells further, we examined the expression of Mbh2/Barhl1, a Bar-class homeobox gene expressed by PC neurons that project mossy fibres, and disruption of Mbh2 causes malformation of PC nuclei (Li et al., 2004). We found that Mbh2 was expressed by most of the EYFP-labelled cells in the AEMS (Fig. 1D) and PEMS(Fig. 1E).

We observed midline crossing of future ECN/LRN neurons, consistent with previous observations (Bourrat and Sotelo,1990; Altman and Bayer,1987a). At E14.5, many EYFP-labelled cells were found in the ipsilateral myelencephalon, migrating tangentially along the dorsal rim of the myelencephalon (Fig. 2A; n=12); they had extended leading processes and were aligned in an orderly fashion, forming several thin layers(Fig. 2B). In the ventral midline region, some EYFP-labelled cells extended their leading processes towards the contralateral side (Fig. 2C and inset). These morphological features, together with the absence of labelled cells occupying other regions on the contralateral side,demonstrate that these cells cross the midline.

At this stage, EYFP-labelled neurons did not form aggregates on their ipsilateral migratory route. However, clustered expression of Mbh2signal was recognized in the LRN region(Fig. 2D). This is probably due to early-born LRN neurons originating from the contralateral lRL, and suggests that mechanisms for nucleogenesis already operate at this stage. Therefore,the failure of ipsilaterally migrating EYFP-labelled neurons to form aggregates in this location (Fig. 2D) may be due to immaturity of these migrating neurons, rather than to underdeveloped environmental cues required for nucleogenesis.

Between E15.5 and E16.5, clusters of labelled cells formed in two distinct regions of the contralateral medulla corresponding to the ECN and LRN(Fig. 2E,G; n=12)(Altman and Bayer, 1987a). Some of the cells in these clusters were found some distance beneath the pial surface, and had extended radially oriented processes(Fig. 2F,H). These results show that PC neurons in the PEMS migrate circumferentially, cross the midline and then turn radially at specific locations.

We examined these cells at a later stage, E18.5 to study the fate of the EYFP-labelled cells. Fewer EYFP-labelled cells were visible in the PEMS at this stage (data not shown). Conversely, those in the ECN and LRN regions increased in number and occupied a wider area than was observed at E15.5(Fig. 3A-C; n=10). Many of the labelled cells in the deeper regions of these nuclei had multipolar shapes with several processes that resembled dendrites (the left panels of Fig. 3D-F), implying that lRL-derived ECN and LRN precursors are nearly mature as they migrate deep into the medulla.

To verify that these clusters of EYFP-labelled cells are ECN and LRN neurons, we examined the expression of Mbh2 mRNA. At E18.5, the labelled neurons were present in a location corresponding to the ECN(Fig. 3A), and in two mediolaterally separated regions that appeared to correspond to the parvocellular and the magnocellular portions of the LRN(Fig. 3B,C)(Bourrat and Sotelo, 1990; Walberg, 1952; Kapogianis et al., 1982). Most of these cells in the dorsal cluster and those in the ventral medulla expressed Mbh2 (the right panels of Fig. 3D-G). Although labelled cells were also found ipsilaterally, in regions corresponding to the ECN and LRN, there were only a few of these cells, and most were Mbh2negative (data not shown). These results show that lRL-derived cells generated from E12.5 onwards give rise to ECN and LRN on the contralateral side.

Migrating ECN/LRN neurons show a transition in their direction of migration at a specific point along their migratory pathway, raising the possibility that these neurons may recognize cue(s) for the transition. A question then arises of why they do not show a transition on the ipsilateral side. As we discussed above, it seems unlikely that this is due to underdeveloped environmental cues (Fig. 2D).

An alternative possibility is that these neurons acquire competence to respond such presumptive cues depending on the time after their generation. As a tool to test this possibility, we used a dominant-negative form of cadherins[EGFP-Ncad(t)], which slows down migration of PC neurons (H.T., D.K.,K. Nishida and F.M., unpublished). In E18.5 control preparations, most strong Venus signals were observed in the contralateral ECN and LRN region(Fig. 4A,B). By contrast, when EGFP-Ncad(t) was co-electroporated with Venus, a large number of EGFP-Ncad(t)-transfected neurons appeared to form aggregates in ECN and LRN regions on the ipsilateral side(Fig. 4C,D; n=9/12). In transverse sections, we found that many Venus-labelled cells formed clusters in deeper regions of ECN and LRN(Fig. 4E,F; data not shown),intermingled with Mbh2 signals that may presumably represent the neurons of contralateral origin (insets in Fig. 4E). These observations suggest the possibility that the neurons whose migration was delayed by transfection of EGFP-Ncad(t) could recognize the presumptive cues for transition to radial direction on the ipsilateral side and are consistent with the interpretation that ECN and LRN neurons normally acquire responsiveness to the cues depending on an intrinsic, age-dependent mechanism. Moreover, these results indicate that midline crossing of ECN and LRN neurons is not prerequisite for nucleogenesis by these neurons.

Although the midline crossing of PGN and RTN neurons has not been described previously, we did observe midline crossing of EYFP-labelled cells. Fig. 5A shows EYFP-labelled cells in the pons of an E14.5 embryo. A substantial number of EYFP-labelled cells had apparently migrated across the ventral midline (broken line)(n=12). At E15.5, cells that appeared to have crossed the midline were dispersed dorsally (Fig. 5B, n=12).

At E14.5, cells with radial orientation were occasionally encountered(arrowheads in Fig. 5A,C; n=12). By E15.5, such cells were more numerous and found at various distances from the midline on both sides of the neural tube(Fig. 5D, n=12). These cells had a typical spindle-shaped morphology, and extended leading and trailing processes (Fig. 5E,F). Cells that had already reached the position of the RTN had extended dendrite-like, branched processes (Fig. 5G). Thus, the cells migrating in the AEMS also change their migration from a tangential to radial direction, on both sides of the neural tube, before they mature.

At E18.5, EYFP-labelled cells formed bilateral clusters that appeared to correspond to the PGN and the RTN. Rostrally, EYFP-labelled cells appeared to occupy the entire territory of the ipsilateral PGN(Fig. 6A,E, n=10). However, they were largely confined to the dorsal aspect in the contralateral PGN (Fig. 6A), and axonal profiles, perhaps from the ipsilateral PGN neurons, were found (arrowheads in Fig. 6F) in the ventral pons. Caudally, at the level of the RTN, radially migrating EYFP-labelled neurons were nearly symmetrically distributed on both sides(Fig. 6C, n=10). Such ventricle-directed neurons were found in the PGN and RTN, and these cells had several branched processes (Fig. 6G-J) that appeared to be immature dendrites. These results indicate that nucleogenesis by EYFP-labelled cells takes place in the PGN and RTN regions on both sides of the neural tube. EYFP-labelled neurons largely coincided with Mbh2 (Fig. 6A-D,G-J), further supporting the conclusion that these bilateral clusters of EYFP-labelled cells are PGN and RTN neurons.

The PGN neurons that crossed the midline were found in the dorsal PGN(Fig. 6A,C). Because earlier-generated PGN neurons are located in deeper regions of the hindbrain(Altman and Bayer, 1987b; Altman and Bayer, 1997), it is possible that only the early-born PGN neurons can cross the midline. To test this hypothesis, BrdU was injected at E12.5, E13.5 and E14.5, into pregnant dams that had been electroporated with EYFP at E12.5(Fig. 7). At E18.5, about one-third of the EYFP-positive neurons on the contralateral side were BrdU positive in embryos injected with BrdU at E12.5(Fig. 7A,C). However, only a few EYFP/BrdU double-positive neurons were observed following BrdU injection at E14.5 (Fig. 7B,D). Most E14.5-born RTN neurons did not express EYFP on the contralateral side(Fig. 7B). Counts of BrdU-labelled cells showed that less than 3% of the EYFP-positive neurons generated at E14.5 reached the contralateral side, compared with more than 30%for the neurons born at E12.5 (Fig. 7E). Therefore, it is the earlier (before E14.5)-generated PGN and RTN neurons that cross the midline.

As pial surface-directed radial migration occurs along the radial fibres in the cerebral cortex (Rakic,1990), we next investigated whether the ventricle-directed migration of PC neurons also takes place along radial fibres. At E15.5, some EYFP-labelled cells in the myelencephalon became ventricle oriented(Fig. 8A,C). The radially extending fibres that span the wall of the hindbrain were immunopositive for nestin, a marker for radial glia-like neuroepithelial cells(Lendahl et al., 1990; Mignone et al., 2004)(Fig. 8C, middle panel). The leading processes of ventricle-oriented ECN and LRN neurons were closely apposed to nestin-positive fibres (data not shown and Fig. 8C, right; n=10 cells in 10 slices from three embryos for ECN, n=9 cells in four slices from two embryos for LRN). Ventricle-directed PGN and RTN neurons in the ventral metencephalon were also closely associated with nestin-positive radial fibres (Fig. 8B,D)(n=14 cells in three slices from two embryos).

Electron microscopic observations of the pontine region confirmed that cells that appear to be PC neurons directly contacted radially aligned fibres. At a low magnification, radial fibres could be readily recognized by their long and radial extension and their termination at the pial surface(Fig. 8E,F and data not shown). At a high magnification they were characterized by the presence of glycogen particles (arrows in Fig. 8F)(Gadisseux et al., 1990). Presumptive PC neurons could also be recognized by their relatively dark and ribosome-rich cytoplasm (Fig. 8E,F) (Ono and Kawamura,1990). Consistent with light microscopic observations described above, such cells were oriented both tangentially and towards the ventricle. We examined whether PC cells make contact with radial fibres. As illustrated in Fig. 8F, we occasionally observed a contact between radial fibres and cell bodies of presumptive PC neurons. Contacts between ribosome-rich thin processes which appear to be the leading or trailing process of PC neurons and radial fibres were also encountered (data not shown). Altogether, these results strongly support the conclusion that nestin-positive radial fibres in the hindbrain mediate the ventricle-directed radial migration of PC neurons.

By using exo utero electroporation of EYFP into the lRL of mouse embryo, we have visualized the processes of nucleogenesis of four PC nuclei and determined the laterality of their origin. Nucleogenesis of PC neurons involves several migration phases: (1) a sorting into the appropriate migratory route; (2) the tangential migration along the circumferential axis;(3) changes in direction at defined regions; (4) ventricle-directed migration contacting with radial fibres; and (5) termination(Fig. 9). Thus, nucleogenesis is a consequence of precisely regulated successively occurring multi-phasic neuronal migration.

Our observations that Mbh2 is expressed by EYFP-positive cells strongly suggest that these cells are PC neurons. Mbh2 is expressed in the rhombic lip early in development(Bulfone et al., 2000), and its expression pattern corresponds to the anatomically defined positions of mossy-fibre-projecting PC neurons (Li et al., 2004). The axons of EYFP-positive cells, in our embryos examined at E16.5 and later, projected to the cerebellum (D.K., T.S. and F.M.,unpublished), which further strengthens the hypothesis that the EYFP-labelled cells are future PC neurons.

EYFP-electroporation did not label other PC nuclei, such as the inferior olivary and vestibular nuclei, which also originate from the lRL(Rodriguez and Dymecki, 2000; Altman and Bayer, 1997). This may be largely because the neurons in these nuclei are generated before E12.5(Taber-Pierce, 1973), and have left the lRL by the time of our electroporation, although there is the possibility that inferior olivary neurons have a distinct origin(Wang et al., 2005; Landsberg et al., 2005).

Previous in vitro studies that labelled presumptive ECN/LRN cells with DiI injections (Kyriakopoulou et al.,2002) or EGFP electroporation(Taniguchi et al., 2002)demonstrated migration across the midline. In this study, we have corroborated these findings by providing in vivo evidence that future ECN and LRN neurons originate from the contralateral lRL. Consistent with our results, a recent analysis of Rig1-deficient mice implied midline crossing of these neurons (Marillat et al.,2004). An unexpected finding was that a subset of PGN and RTN neurons have a contralateral origin (Figs 5, 6).

There are several important phases in the nucleogenesis of PC neurons. LRL-derived neurons must first determine which of the two migratory pathways they should take (Fig. 1). A recent study demonstrated that Mbh1/Barhl2, a homolog of Mbh2 (Saito et al.,1998), is expressed in AEMS, but not in PEMS(Mo et al., 2004). It is therefore possible that lRL-derived neurons use cues under the control of Mbh1 when choosing one of these migratory pathways.

We have found that tangential migration of PC neurons is terminated by the transition to ventricle-directed migration, not merely by ceasing to migrate along their tangential route. The transition and turning could be regulated either by an intrinsic or an extrinsic mechanism. The former is consistent with a previous observation of facial branchiomotor neurons, which regulate expression of surface molecules depending on migration phase(Garel et al., 2000) (for a review, see Chandorasekhar, 2004). However, the fact that the transitions in our study occurred at restricted sites in a protracted period of development during which the size of the neural tube is dramatically increased favours the environmental cue mechanism. Moreover, our results that introduction of EGFP-Ncad(t) caused formation of ECN and LRN ipsilaterally but in appropriate locations support the idea that environmental cues exist that induce radial turning at defined locations along the circumferential axis. An interesting possibility is that the transition involves radial fibres at transition points with distinct properties that induce changes in the migratory behaviour of the PC neurons.

Whatever the nature of the cues, our results using dominant-negative form of cadherins suggested that PC neurons acquire responsiveness to turning cues by a time-dependent mechanism (Fig. 9). In addition, these findings imply that changes in responsiveness to such cues occurs independent of their midline crossing. Thus, the speed of ECN/LRN neurons migration should be precisely regulated for proper nucleogenesis on the contralateral side. Further studies are required to clarify these molecular mechanisms.

After the phase of tangential migration, PC neurons start migrating towards the ventricle. The role of radial glial cells as a substrate for pial surface-directed migration of cortical neurons is well documented(Marin and Rubenstein, 2003; Rakic, 1990; Hatten and Mason, 1990). Their role in ventricle-directed migration, however, is poorly understood. Here,close association of EYFP-labelled neurons with nestin-positive fibres under the light microscope observed here suggests that PC neurons use radial fibres as a substrate for ventricle-directed migration. This view was strongly supported by our electron microscopic observations, which demonstrated a direct contact of a migrating cell with radial fibres(Fig. 8). Our results provide the first compelling evidence that nestin-positive radial fibres may guide embryonic neurons towards the ventricle, which is essential for the formation of nuclei in appropriate positions.

It is likely that these radial fibres are also used for pial surface-directed migration of hindbrain neurons generated at the ventricular surface, raising the possibility that neurons migrating in opposite direction might collide with each other. This seems unlikely, however, because generation of neurons in the hindbrain ventricular zone seems to occur at earlier stages (Y. Tashiro and F.M., unpublished).

Molecular mechanisms for the ventricle-directed radial migration may be distinct from those for pial-surface directed migration, because genes that play crucial roles in pial-surface directed migration of cortical neurons such as cdk5/p35 (Ohshima et al.,1996; Gupta et al.,2003; Hatanaka et al.,2004) and reelin (Ogawa et al., 1995) do not appear to affect the nucleogenesis of mossy-fibre-projecting PC neurons (Ohshima et al., 2002).

The mossy-fibre-projecting PC neurons are characterized by an early(E9.5-11.5) expression of some transcription factors(Li et al., 2004; Rodriguez and Dymecki, 2000; Wang et al., 2005), before they leave the lRL, suggesting that they are already specified before they start migration. However, they have distinct migratory pathways and modes of migration, and their eventual positions are dispersed in the hindbrain(Rodriguez and Dymecki, 2000; Wang et al., 2005) (this study). These features cannot be explained simply by the expression of such genes. There must be some mechanism that allows each subset of PC neurons to express different transcription factors that control the responsiveness of these neurons to environmental cues. To fully understand the molecular mechanisms involved in the development of distinct cell types of PC neurons at specific locations in the hindbrain, the factors that may be involved in the regulation of PC neuron migration remain to be investigated. Our results may help in unravelling the mechanisms underlying these events.

We thank Dr Y. Hatanaka for advice on exo utero electroporation; M. Iwashita for instructions on mouse surgery; Dr J. Miyazaki for a CAG promoter;Dr A. Miyawaki for Venus; Drs N. Yamamoto, H. Kobayashi and Y. Zhu for reading of the manuscript; and Drs K. Nishida, Y. Tashiro and D. Tanaka for discussion and continuous encouragement. Rat 401 antibody was obtained from Developmental Studies Hybridoma Bank, Iowa University. This work was supported by SORST, JST and a Grant-in-Aid for Scientific Research on Priority Area and Advanced Brain Science Project from MEXT(17023028).