Transcriptional control of early tract formation in the embryonic chick midbrain

A characteristic feature of all vertebrates is the formation of a conserved set of longitudinal and commissural axon tracts during early brain development(Chedotal et al., 1995; Chitnis and Kuwada, 1990; Easter et al., 1993; Wilson et al., 1990). The early axon scaffold formed by these tracts is thought to act as a framework for the later, complex network of neuronal connections in the brain. We are interested in the molecular mechanisms that underlie the specification of the early differentiating neurons in the brain, which despite the detailed studies on early axon tract formation in a range of vertebrates have remained enigmatic.

The specification of neurons, however, has been studied in great detail in the spinal cord, where the initial dorsoventral patterning of the neural tube,resulting from the antagonistic action of ventralising and dorsalising signals is translated into spatially restricted expression of homeobox genes. The longitudinal, dorsoventrally restricted expression domains of the homeobox genes prefigure the longitudinal columns of neuronal subtypes in the spinal cord, as differentiating neurons adopt their distinct identities as a result of expressing a specific combination of homeodomain transcription factors(reviewed by Goulding and Lamar,2000).

The organisation of the brain is more complex, and it harbours a greater diversity of neurons than does the spinal cord. Neurons are either organised into different layers, as in the tectum or the cerebral cortex, or into nuclei, as in the ventral midbrain. Nuclei occupy distinct positions along the rostrocaudal and dorsoventral axes, where their specification is likely to be controlled by coordinate patterning of the neural tube. Interestingly,parallels exist between the dorsoventral patterning of spinal cord and midbrain at the molecular level. As in the spinal cord, ventral patterning in the midbrain is governed by floor plate-derived Shh signalling: overexpression of sonic hedgehog (Shh) throughout the midbrain leads to an expansion of the basal plate-derived tegmentum territory at the expense of the dorsal tectum (Watanabe and Nakamura,2000), while local misexpression induces the expression of ventrally expressed homeobox genes in a dose-dependent pattern(Agarwala et al., 2001). Several homeobox genes are normally expressed in the ventral midbrain in longitudinal domains – so-called arcs – in a similar arrangement to the homeobox gene expression domains in the spinal cord(Sanders et al., 2002). Up to five arcs, defined by the differential expression of homeobox genes, are established in response to a presumed gradient of Shh emanating from the floor plate. Ventral signalling, with resulting expression of homeobox genes, is crucial for correct development of the ventral midbrain. A direct requirement for Shh has been demonstrated for two groups of neurons in the tegmentum, the somatic motoneurones of the oculomotor nucleus(Chiang et al., 1996), and the dopaminergic neurons of substantia nigra and ventral tegmental area(Hynes et al., 1995). Consistent with an instructive function of homeobox genes in the specification of ventral mesencephalic neurons, Isl1 and Phox2a are essential for the formation of the oculomotor nucleus(Nakano et al., 2001; Pfaff et al., 1996), Emx2 is required for proper development of the red nucleus(Agarwala and Ragsdale, 2002),and several homeobox genes including Pitx3 and Lmx1 are involved in the specification of dopaminergic neurons (reviewed by Smidt et al., 2003).

We are interested in whether homeobox genes similarly play a role in the formation of the early axon scaffold. The dominating longitudinal tract in the early scaffold is the medial longitudinal fascicle (mlf). Very little is known about the molecular mechanisms that underlie the formation of the mlf and its contributing nucleus – the interstitial nucleus of Cajal (INC). Tight genetic regulation seems particularly important for distinguishing the fate of mlf neurons from those forming the posterior commissure (pc), as neurons for both tracts are located principally in the same ventral cluster at the midbrain-forebrain border (MFB) (Tallafuss et al., 2003). Could homeobox genes play a role in specifying these early neuronal subtypes, and especially the mlf and pc cells? To answer this question, we first analysed the expression patterns of the homeobox genes Sax1, Six3, Emx2 and Pax6 in the ventral midbrain of chick embryos between HH15 and HH25, when the early axon scaffold is formed. These genes have previously been described to be expressed at the ventral MFB (e.g. Agarwala and Ragsdale, 2002; Bovolenta et al., 1998; Schubert et al., 1995), but their precise temporal and spatial patterns of expression have not been determined. Out of the expression analysis, Sax1 emerged as a prime candidate for neuronal specification at the ventral MFB, as it is expressed predominantly in the ventral neuronal cluster, coincident with the INC, from the time the first neurons appear. Sax1 is a member of the NK1 class of homeobox genes, which in vertebrates is usually represented by two closely related genes, Sax1 and Sax2(Bae et al., 2004; Bober et al., 1994; Schubert et al., 1995; Simon and Lufkin, 2003; Spann et al., 1994), while additional, more divergent members have recently been described in Xenopus (Kurata and Ueno,2003) and zebrafish (Bae et al., 2003). We have employed electroporation in the chick to study the function of Sax1 in the specification of neurons at the ventral MFB. Misexpression of Sax1 leads to an increase in the size of the mlf, and affects the expression of other ventral homeobox genes, suggesting that a homeobox gene code underlies the formation of the early axon scaffold,and that Sax1 in particular regulates the formation of the mlf.

Fertilised hens eggs were obtained from Winter Egg Farm (Royston). They were incubated at 38°C for the required stage. Embryos were staged according to Hamburger and Hamilton(Hamburger and Hamilton,1951).

We employed RNA probes for chick Sax1(Spann et al., 1994), Emx2 (Bell et al.,2001), Isl1 (Tsuchida et al., 1994), Pax6(Goulding et al., 1993), Phox2a (Groves et al.,1995) and Six3(Chapman et al., 2002) for our analysis. For details of the whole-mount in situ hybridisation protocol, see Dietrich et al. (Dietrich et al.,1997). In double labelling experiments, digoxigenin- and fluorescein-labelled probes were consecutively detected with alkaline phosphatase-conjugated antibodies (Roche), using NBT/BCIP (Roche) and Vector Red (Vector Labs) as blue and red substrates.

To analyse the expression of Sax1 in relation to mlf neurons, we retrograde labelled the mlf from the rostral hindbrain using fluorescein-labelled dextran (Molecular Probes). Embryos at HH23 were dissected in PBS, the ventral hindbrain was cut at the level of rhombomere 2,and a crystal of the dye applied. Embryos were incubated at 37°C in L15 medium (GibcoBRL) for 3 hours, and then fixed. Following in situ hybridisation for Sax1, the fluorescein label was detected with an alkaline phosphatase-conjugated anti-fluorescein antibody (Roche), as in double-labelling in situ hybridisation, using Vector Red as substrate (see also Agarwala and Ragsdale,2002).

Specific axon tracts in the embryonic brain were retrograde labelled with lipophilic dyes. The mesenchyme was removed from day 5 chick embryos, and isolated, hemisected brains were fixed flat on black nitrocellulose membrane(Schleicher and Schuell). Crystalline DiI was applied on to the ventral part of rhombomere 2 to label the mlf, while DiO was used to label the pc from the roof plate of the caudal pretectum.

Neurons (cell bodies and axons) were detected with an antibody against Neurofilament-M (Zymed RMO270), visualised by a peroxidase-conjugated anti-mouse antibody (Jackson Laboratories) using Diaminobenzidine (Vector Labs) as substrate. When combined with in situ hybridisation, the primary antibody for immunohistochemical detection of neurofilament protein was applied after completing the colour reaction of the whole-mount in situ hybridisation procedure.

Two different expression constructs were used, both based on pCAβ-LINK-IRESeGFPm5-ClaI (Fig. 3I) (J. Gilthorpe, A. Hunter and A.L., unpublished), an expression vector in which a hybrid CMV/chick β-actin promoter(Miyazaki et al., 1989) drives the transcription of a polycistronic message encoding the gene of interest and– linked by an IRES element – enhanced green fluorescent protein(eGFP). pCAβ-Sax1-IRES-GFP was constructed by inserting the full coding region of the murine Sax1 gene(Schubert et al., 1995) into the expression vector (Fig. 3L). For the assembly of pCAβ-VP16Sax1-IRES-GFP,first the eh1-like domain of the murine Sax1 gene(Smith and Jaynes, 1996) was removed, and the transactivation domain of Herpes simplex VP16(Triezenberg et al., 1988) was introduced in its place (Fig. 3L). The coding region for the hybrid VP16Sax1 protein was again cloned into the base expression vector. The integrity of the expression constructs was confirmed by sequencing. The pCAβ-LINK-IRESeGFPm5-ClaI vector itself was used as control for non-specific effects of the electroporation.

Expression constructs were used at a concentration of 1 mg/ml, with Fast Green added to 0.1% to facilitate visualisation of the DNA solution. After cutting a window into the eggshell and opening the vitelline membrane, the DNA solution was injected into the neural tube at the level of the midbrain of HH11-13 chick embryos, using a PV820 Picopump (WPI). The anode, a 0.25 mm platinum wire, was placed lateral to the midbrain, and the cathode, a 0.125 mm flame-sharpened tungsten wire, was inserted into the neural tube. Two pulses of 12.5 V, 50 ms were applied, using a TSS10 stimulator (Intracel). The window was closed with Sellotape, and the eggs were then incubated for a further 1-3 days.

Where required, embryos were sectioned after the whole-mount in situ hybridisation or immunohistochemistry procedures. Sections of 30 μm were cut on a Leica vibratome and mounted in 80% glycerol.

Photographs of whole embryos, dissected brain tissue or sections were taken on a Zeiss Axiophot microscope with differential interference contrast, using a Zeiss Axiocam digital camera. Subsequent processing and assembly of the images was carried out with Adobe Photoshop.

In all vertebrates, axon tracts in the early embryonic brain invariably form the early axon scaffold, a conserved array of longitudinal and commissural tracts. Particularly dense in early axon tracts is the ventral region around the MFB, through which the tract of the postoptic commissure(tpoc) passes, and from which the mlf and the pc originate. To unravel the molecular mechanisms that control the development of mlf and pc, we first studied the expression patterns of several homeobox genes around the ventral MFB. As a reference point for the subsequent expression studies, we employed Isl1 (Pfaff et al.,1996) to mark the oculomotor nucleus located just caudal and ventral to the INC. Sax1(Schubert et al., 1995), Six3 (Bovolenta et al.,1998; Tallafuss et al.,2003), Emx2 (Agarwala and Ragsdale, 2002) and Pax6(Sanders et al., 2002; Schubert et al., 1995) have all been previously described to be expressed in the ventral midbrain, dorsal to the oculomotor nucleus. However, their precise expression patterns with respect to each other and to specific early midbrain nuclei such as the INC have not been determined.

Using whole-mount RNA in situ hybridisation, we detected the first expression of a homeobox gene in the ventral midbrain at HH15, when Sax1 signals appeared at the ventral MFB. These were followed shortly by Six3 and Emx2 signals in the same region (not shown). In contrast, Pax6 is not expressed in the midbrain until HH19. By HH20,distinct differences in the expression patterns of the different homeobox genes are evident (Fig. 1A-E). Signals for Six3 and Emx2 are split into ventral and dorsal stripes in the midbrain, separated by the emerging Pax6 expression domain, and ventrally delimited by the Isl1-positive oculomotor nucleus (Fig. 1C-E). The Sax1 signal spans the whole dorsoventral extent of the Six3/Emx2 domains, including also the intervening Pax6-positive stripe (Fig. 1B). Interestingly, although all four genes are expressed in the mantle layer, the mRNA for Sax1 is found exclusively in the outer margin of the mantle layer, aligning the marginal zone, while the Six3 and Emx2 signals are located further medially(Fig. 1B-D, the asterisk marks the medial limit of the Sax1 signals).

Differences in the expression domains are also evident along the rostrocaudal axis, and become more pronounced as the brain develops(Fig. 1F-J). At HH25, the signals for Six3, Emx2 and Pax6 extend as curved,longitudinal domains from the MFB caudally, throughout the tegmentum(Fig. 1H-J,M-O). According to the nomenclature of Sanders et al.(Sanders et al., 2002), these correspond to arcs 2 and 3 (Six3 and Emx2) and the intervening region (Pax6). Close to the MFB, the ventral Emx2 domain stretches out ventrally into the Isl1-stained arc 1, outlining the prospective red nucleus(Fig. 1I,N)(Agarwala and Ragsdale, 2002). In contrast to the elongated expression domains of the former genes, the expression of Sax1 remains more focussed in the rostral tegmentum and the MFB, where it overlays the domains of Emx2, Six3 and Pax6 (Fig. 1G,L, see also Fig. 2F). The signals for all genes except Isl1 also continue into the ventral diencephalon. In the ventral pretectum, the Pax6 mRNA is restricted to a small domain,ventrally bordered by the signals for Sax1, Six3 and Emx2,and separate from the extensive Pax6 expression domain in the dorsal pretectum by the dorsal stripe of Six3 and Emx2 expression(Fig. 1G-J,L-O).

The mlf is the earliest tract formed in the embryonic brain. In the chick,it can be detected from HH14, originating from the INC at the ventral MFB(Chedotal et al., 1995). By contrast, the pc develops several hours later(Chedotal et al., 1995). The nucleus of the pc (nPC) is dispersed over several locations in the MFB and pretectum (Chitnis and Kuwada,1990). Simultaneous retrograde labelling of mlf and pc in an HH27 embryo demonstrates an overlap of INC and nPC in a cluster of neurons at the ventral MFB (Fig. 2B). This close association of mlf and pc neurons has previously been observed in zebrafish (Chitnis and Kuwada,1990). The expression domain of Sax1 corresponds to this ventral neuronal cluster (Fig. 2A,C), suggesting that Sax1 might play a role in the specification of neurons in rostral midbrain and pretectum, particularly the INC.

To identify which neurons might express Sax1, we stained chick embryonic brains at different stages simultaneously for neurofilament protein and Sax1 mRNA (Fig. 2E-H). At HH17, neurofilament staining detects the mlf as a bundle of longitudinal axons just dorsal to the floor plate. These can be traced to cell bodies at the ventral MFB, located within the Sax1 expression domain at the MFB (Fig. 2E). The location of these neurons at the ventral MFB, immediately rostral and dorsal to the oculomotor nucleus, and their caudal ipsilateral projection identify them as mlf neurons. While at HH17 Sax1 is expressed just in a small domain, the Sax1 signals at HH21 are much stronger and mask the neurofilament staining. Still, the axons of the mlf are visible as they extend from the Sax1 domain (Fig. 2F).

To confirm the conclusion that the mlf neurons express Sax1, we sectioned double-labelled embryos horizontally. Sections through the ventral mesencephalon show the Sax1 expression domain located just rostral to the oculomotor nucleus (Fig. 2G). Higher magnifications revealed Sax1-expressing neurons projecting into the mlf (Fig. 2H). Furthermore, we combined retrograde labelling of the mlf with in situ hybridisation for Sax1 mRNA. At HH23, cell bodies of mlf neurons were concentrated in two clusters around the MFB, a caudodorsal patch and a rostroventral area (see also Fig. 3J). Both areas overlap with the Sax1 expression domains in ventral mesencephalon and pretectum(Fig. 2D). Our analysis shows that Sax1 is expressed in the INC, the nucleus of the mlf.

The close association of Sax1 expression and mlf neurons raised the possibility that Sax1 could be involved in establishing this early tract. To test this hypothesis, we used a gain-of-function approach where we expressed Sax1 ectopically to study the effect on the morphology of the mlf (Fig. 3). We employed two different expression constructs, based on the pCAβ-LINK-IRESeGFPm5-ClaI vector (J. Gilthorpe, A. Hunter and A.L.,unpublished) (Fig. 3L). pCAβ-Sax1-IRES-GFP contains the coding sequence for the mouse Sax1 gene (Fig. 3O). Assuming that Sax1 normally acts as a transrepressor, mediated through the binding of Groucho co-factors to its eh-1 like domain(Smith and Jaynes, 1996), this construct would repress the expression of Sax1 target genes. We also designed a modified version, VP16Sax1, in which the transactivation domain of Herpes Simplex VP16 (Triezenberg et al., 1988) replaced the eh1-like transrepression domain(Fig. 3O). The pCAβ-VP16Sax1-IRES-GFP construct therefore encodes a protein that would transactivate Sax1 target genes, thus acting as a dominant-negative regulator of Sax1 function.

When we introduced the Sax1 expression constructs at HH10-13, we observed changes in the morphology of the early axon scaffold just 1 day after electroporation. While the mlf axons normally run close to the floor plate in a compact bundle (Fig. 3A), the fibres in the Sax1-expressing embryos stretch further dorsally, and their course is less regular (Fig. 3B). This phenotype appears even more pronounced after 2 days of ectopic Sax1 expression, when immunohistochemical staining for neurofilament protein shows the irregular pattern of the longitudinal axon tract in the ventral midbrain (Fig. 3D-I). While in embryos expressing the control construct the axon scaffold appeared normal (Fig. 3D,G), following electroporation of pCAβ-Sax1-IRES-GFP the mlf expanded dorsally, occupying a larger region of the tegmentum (Fig. 3E,H). By contrast, the pc, although prominently stained in control embryos (Fig. 3D), was barely visible in the Sax1-expressing embryos(Fig. 3E). Ectopic expression of VP16Sax1 did not result in such a strong phenotype, probably owing to lower levels of expression consistently achieved with the pCAβ-VP16Sax1-IRES-GFP construct. Still, VP16Sax1 seems to have the opposite effect on the mlf, as the tract appeared less prominent than in control embryos (compare Fig. 3C,F,I with Fig. 3A,D,G).

The results of the immunohistochemical analysis are mirrored by retrograde labelling of the ventral longitudinal tract from the ventral hindbrain(Fig. 3J,K). Again, the mlf was enlarged in the Sax1-expressing embryos(Fig. 3K) compared with embryos just expressing the control construct (Fig. 3J). In addition, while in the control embryo cell bodies were organised into two subclusters, located caudodorsally and rostroventrally(Fig. 3J, arrowheads), mlf neurons in the Sax1-expressing embryos were scattered throughout the ventral MFB, following no apparent pattern(Fig. 3K).

The expansion of the mlf after ectopic Sax1 expression could be the result of increased proliferation of mlf precursors, or of mis-specification of neurons normally destined for a different fate. Using an antibody against the mitosis marker phospho-Histone H3 (PH3), we analysed the electroporated embryos for differences in cell proliferation in the tegmentum. We found no obvious differences in the PH3 staining between embryos electroporated with either pCAβ-LINK-IRESeGFPm5-ClaI(Fig. 3M) or pCAβ-Sax1-IRES-GFP (Fig. 3N) that could explain the expansion of the mlf, suggesting that Sax1 misexpression leads to the mis-specification of neurons at the ventral MFB towards mlf neuron fate.

In the spinal cord, homeobox genes regulate their expression by mutual cross repression (Muhr et al.,2001). This mechanism ensures that sharp expression boundaries are formed, translating into distinct neuronal fate decisions. If similar mechanisms act in the midbrain, Sax1 misexpression should affect the expression of other homeobox genes in the ventral midbrain. To test this hypothesis, we studied the expression patterns of Six3, Emx2, Pax6and Phox2a in electroporated embryos(Fig. 4). Normally, Six3 and Emx2 are expressed in a subdomain of the Sax1-expressing region, albeit in cells located more medially(Fig. 1). This pattern is unchanged in embryos expressing the GFP expression construct(Fig. 4A,D). By contrast, 1 day after electroporation of pCAβ-Sax1-IRES-GFP, signals for both genes are reduced or lost, depending on the level of ectopic Sax1expression (Fig. 4B,E). Misexpression of VP16Sax1 has a profound effect on Emx2expression, leading to the upregulation of Emx2 ventrally, and even to ectopic transcription of Emx2 in the dorsal midbrain(Fig. 4C). Six3expression, by contrast, is not altered by VP16Sax1(Fig. 4F).

Phox2a expression was analysed two days after electroporation,when it labels the motoneurones in the oculomotor nucleus. Following Sax1 misexpression, the Phox2a expression domain extends further rostrally (Fig. 4H),while electroporation of pCAβ-VP16Sax1-IRES-GFP reduces or even abolishes the expression of Phox2a in the oculomotor and trochlear nuclei (Fig. 4I). Similar results were obtained with other markers for the oculomotor nucleus, such as Isl1, BEN and GAP43 (not shown). Corresponding to the expanded oculomotor marker expression in Sax1 overexpressing embryos,the oculomotor nucleus was enlarged, and neurofilament staining frequently showed the presence of two or three nerves emerging from it (not shown). In contrast to the former genes, the ventral stripe of Pax6 expression in the mesencephalon was largely unaffected by the expression of Sax1or VP16Sax1 (Fig. 4J-L), although expression of Sax1 did block the expression of Pax6 in the ventral pretectum(Fig. 4K).

These results show that – like the homeobox genes in the spinal cord– the homeobox genes in the ventral midbrain can apparently crossregulate each other. Sax1 in particular has a profound effect on the expression of Emx2 and Six3.

The formation of an axon scaffold during early brain development is a conserved feature of all vertebrates. Despite the abundance of morphological studies on the early axon tracts, the molecular mechanisms governing the development of this scaffold are largely unknown. However, recent studies in chick have suggested that homeobox genes in the ventral midbrain could play a similar role in specifying neuronal fate as they do in the spinal cord. Our experiments provide evidence that the differential expression of homeobox genes is indeed instructive to the fate of neurons in the ventral midbrain,and that, in particular, Sax1 regulates the specification of mlf and pc neurons from the ventrocaudal cluster of neurons at the ventral MFB.

We have found that a number of homeobox genes are expressed in distinct domains in the ventral midbrain and pretectum during early brain development in the chick. Among these, Emx2 and Pax6 have been described previously as part of the arcuate plan that suggests the organisation of the ventral midbrain into longitudinal domains, called arcs(Agarwala and Ragsdale, 2002; Sanders et al., 2002). The arcs can be visualised by the expression of homeobox genes such as Phox2a for arc 1 or Pax6 dividing arcs 2 and 3(Sanders et al., 2002). In our analysis, we have included two further genes labelling arcs 2 and 3, Six3 (Bovolenta et al.,1998) and Emx2 (Bell et al., 2001). The expression patterns of both genes in the ventral midbrain are largely overlapping. However, only the Emx2signal also extends ventrally into the rostral part of arc 1, where it labels the prospective red nucleus (Agarwala and Ragsdale, 2002). All of these genes are expressed throughout most of the midbrain, stretching from the MFB almost to the isthmus. This suggests that they may form part of a general patterning machinery for the whole ventral midbrain. By contrast, Sax1 is expressed predominantly around the MFB, abutting the oculomotor nucleus dorsally and rostrally. This Sax1 expression pattern in the chick is similar to its orthologue Sax1 (Schubert et al.,1995) and its paralogue Sax2(Simon and Lufkin, 2003) in mouse. Likewise, the zebrafish sax2 gene is expressed in the ventrocaudal cluster at the MFB (Bae et al., 2004). In double labelling experiments for Sax1 mRNA and either neurofilament protein or retrograde labelling to visualise the mlf,we have demonstrated that the Sax1 expression domain overlays the INC, hinting at a specific function of Sax1 in the specification of neurons at the MFB, particularly those forming the mlf.

What is the role of Sax1 in the formation of the mlf? Our misexpression experiments demonstrate that the expression of Sax1 has to be tightly regulated to ensure the normal development of the mlf: ectopic expression of Sax1 interferes with the patterning at the ventral MFB,and leads to an expansion of the mlf. This result suggests that Sax1is involved in the formation of the mlf, possibly by specifying mlf fate in differentiating neurons. However, in the converse experiment, VP16Sax1 expression reduces the size of the mlf, but does not completely abolish its formation. This might be explained by incomplete penetrance of the constitutively activating construct against the background of endogenous Sax1 and Sax2 expression. Sax2expression in the mouse midbrain overlaps the Sax1 expression domain,and mice lacking Sax2 do not show an apparent midbrain phenotype(Simon and Lufkin, 2003),arguing for possible compensation by its paralogue.

The same cluster of neurons that includes the INC also harbours the ventral part of the nucleus of the pc. The pc is formed well after the mlf, with the first axons extending dorsally visible at HH17. In embryos expressing Sax1 ectopically, the number of pc neurons is reduced so that the pc is barely visible. It is unclear how neurons in the ventral cluster are specified to mlf or pc fate, but ectopic expression of Sax1 seems to interfere with this process. There is a possibility that fate specification is influenced by the birth date of individual neurons, as in the case of oculomotor and red nucleus neurons developing successively from arc 1(Agarwala and Ragsdale, 2002). In such a scenario, early birth would supports mlf fate, while later birth would favour pc neurons. Sax1 would be thus linked to the timing of neurogenesis. Indeed, in the early embryo Sax1 is transiently expressed alongside Cash4 in the caudal neural plate, preceding neurogenesis (e.g. Henrique et al.,1997). Although the role of Sax1 in the caudal neural plate is unknown, it may be involved in neurogenesis in the caudal CNS, a role also recently assigned to the distantly related NK1-class homeobox gene Pnx (Bae et al.,2003). In the brain, Sax1 is normally only expressed in postmitotic neurons, while with the electroporation method we introduce Sax1 into neural progenitors, which may influence the time point when neural cells leave the cell cycle.

Alternatively, the specification of mlf and pc could be the result of intrinsic differences, possibly the differential expression of homeobox genes. Again, our results are consistent with this mechanism. The mlf enlargement after ectopic expression of Sax1 is not linked to any apparent change in cell proliferation. Although formally it is also possible that ectopic Sax1 expression changes axon guidance cues to misroute pc neurons onto a caudal path, our findings argue for a change of cell fate in the affected cells as the most likely cause of the observed phenotype. In addition, the repression of Emx2 and Six3 indicates an altered spatial patterning of the ventral MFB. In zebrafish, Six3labels pc neurons as well as the INC(Tallafuss et al., 2003). Possibly, the specification of pc or mlf neurons depends on the balance of homeobox gene expression at the ventral MFB. Loss of Six3 and Emx2 expression – together with increased or ectopic Sax1 expression – might shift this balance in favour of mlf specification.

It is possible that both mechanisms, temporal and molecular difference,work hand in hand, as Sax1 is expressed closer to the marginal surface of the neural tube than Six3 and Emx2. This not only explains how the latter escape transcriptional repression by Sax1, but may also reflect a link between the time of neuronal differentiation and the expression of specific homeobox genes, thus adding a temporal dimension to the spatial pattern of differential gene expression.

Recently, several studies have described the molecular patterning of the ventral midbrain, and – together with the data presented in our study– we can begin to assemble the genetic network that controls the formation of ventral midbrain nuclei. Patterning of the ventral midbrain along the two main axes occurs under the influence of neighbouring tissues: Fgf8 from the isthmus sets up the caudorostral polarity of the midbrain (e.g. Crossley et al., 1996), and Shh derived from notochord and floor plate constitutes the ventralising signal(Watanabe and Nakamura, 2000). In response to both signals, homeobox genes are expressed in distinct patterns in the ventral midbrain (Agarwala et al.,2001; Sanders et al.,2002). The midbrain arcs largely subdivide the tegmentum into distinct domains along the dorsoventral axis, but they also display distinct rostrocaudal features. Thus, Emx2 is expressed in arcs 2 and 3 in the entire midbrain, but only rostrally extends into arc 1. Sax1expression in arcs 2 and 3 and the intervening region is restricted to the rostral midbrain, close to the MFB.

Several studies have now implicated homeobox genes with the formation of particular nuclei in the ventral midbrain, indicating the importance of proper patterning for the correct development of tegmental neurons. Our study demonstrates that the homeobox `code' is already crucial for the specification of neurons that form the early axon scaffold. We also show for the first time that a homeobox gene expressed around the ventral MFB can directly or indirectly regulate the expression of other homeobox genes, providing a possible patterning mechanism. The regulatory activity of Sax1 appears to be highly specific, as ectopic expression of Sax1 abolishes the expression of Emx2 and Six3, but not Pax6 in the ventral midbrain. However, Pax6 expression in the adjacent ventral pretectum, where Pax6 and Sax1 are expressed exclusively, is lost after ectopic expression of Sax1. At the same time, the Isl1 expression domain extends rostrally into the area where normally Emx2 would be expressed. Quite possibly, the rostral extension of the Isl1 domain (and the oculomotor nucleus) is an indirect effect of Sax1, reflecting the loss of Emx2 expression. Likewise, the loss of the oculomotor nucleus following expression of VP16Sax1 could be an indirect effect of the ectopic Emx2 expression induced by the dominant-negative variant of Sax1. In this scenario, Emx2 would repress the expression Isl1 in the rostral arc 1. Although such an effect has yet to be investigated, it is a conceivable mechanism by which Emx2 may specify differentiating neurons towards red nucleus rather than oculomotor nucleus fate (Agarwala and Ragsdale,2002). This would resemble the mechanism suggested by our own data, with Sax1 crucially influencing the fate decision between mlf and pc.

These lines of evidence point to a possible recurring theme for nucleogenesis in the ventral midbrain: differentiating neurons at a given position face binary decisions of cell fate, and their choice is influenced by their relative birth date and is controlled by the differential expression of homeobox genes. An important characteristic of this possible mechanism is the mutual repression of the fate-determining transcription factors to avoid ambiguity in the cell fate, a strategy also employed in other examples of cell fate selection from a common precursor pool such as the specification of neuronal fate in the vertebrate spinal cord(Briscoe et al., 2000) and of muscle cell identity in the Drosophila embryo(Jagla et al., 2002). Interestingly, NK1 class genes are involved in both processes: the fly homologue of Sax1, slouch, is a muscle identity gene(Knirr et al., 1999); and Sax1 itself is expressed in a subset of interneurones in the spinal cord (Schubert et al.,1995).

We propose that, as a result of broad rostrocaudal regionalisation mediated by the isthmic organiser, by local interactions at the MFB, and by dorsoventral patterning by floor plate and roof plate, homeobox genes are expressed in distinct domains in the ventral midbrain and pretectum. Their expression domains may become sharpened by reciprocal repressive interaction between the homeobox genes. We show that perturbing this intricate pattern by overexpressing the ventrally expressed homeobox gene Sax1 ectopically leads to disturbed dorsoventral patterning of the midbrain and affects the organisation of the early axon scaffold. We conclude that the (combinatorial)expression of specific homeodomain transcription factors determines neuronal cell fate in the tegmentum of midbrain and pretectum.

We are grateful to A. Fainsod, C. Goridis, P. Gruss and T. Jessell for providing probes for in situ hybridisations, and to J. Gilthorpe and A. Hunter for the pCAβ-LINK-IRESeGFPm5 expression vector. We express our special thanks to S. Dietrich, K. Riley and C. Sharpe for their helpful comments during the preparation of the manuscript. The project was generously supported by a Project Grant from The Wellcome Trust (number 059207).