Pax-6 functions in boundary formation and axon guidance in the embryonic mouse forebrain

The differentiation of the first neurons in the vertebrate forebrain coincides with a rapid process of morphological and molecular regionalization. The initial broad divisions of the neural tube form the classic three vesicles of the prosencephalon, mesencephalon, and rhombencephalon, embryonic structures presaging the adult forebrain, midbrain and hindbrain. Further subdivisions form the neuromeres, a series of bulges separated by transverse constrictions in the prosencephalon (prosomeres; Puelles and Rubenstein, 1993), and the rhombencephalon (rhombomeres; Lumsden and Keynes, 1989). On the molecular level, a large number of regulatory genes (many shared with Drosophila; Hirth et al., 1995) are expressed in partially overlapping, regionally restricted patterns in the prosencephalon (reviewed by Puelles and Rubenstein, 1993; Rubenstein and Puelles, 1994). In some cases, the borders of expression domains coincide with neuromere boundaries (Bulfone et al., 1993; Price et al., 1992), suggesting that borders of gene expression cause morphological boundaries.

This process of regionalization ultimately provides pos-itional information that is thought to regulate both the production of precise patterns of neuron cell bodies and the naviga-tion of their axons. Supporting evidence for either part of this hypothesis has been limited to a few correlative studies. For example, gene expression borders coincide with the initial clusters of differentiating neuron cell bodies in the zebrafish prosencephalon (Macdonald et al., 1994). In mouse, the prosencephalon-mesencephalon boundary coincides with the caudal border of the neuron cell bodies that pioneer the posterior commissure (pc; Mastick and Easter, 1996). On the level of axons, some initial tracts in chick and fish form along gene expression borders (Figdor and Stern, 1993; Macdonald et al., 1994; Chedotal et al., 1995). In mouse, the leading axons of the first tract, the tract of the postoptic commissure (tpoc), converge to cross a specific prosomere boundary, implying that neuromeres and their boundaries may differentially influence navigating axons (Mastick and Easter, 1996). These examples support the hypothesis that neuromeres and gene expression domains regulate neuronal patterning, but these general ideas have yet to be tested, for example, by perturbation of specific patterning genes.

One candidate gene that could regulate prosencephalon pat-terning is Pax-6. This gene encodes a transcription factor with both paired-and homeo-domain DNA binding motifs (Walther and Gruss, 1991), and is expressed in the developing CNS of a wide range of species, from vertebrates to nematodes. Most studies of Pax-6 mutations have focused on eye development. Heterozygous Pax-6 mutations cause the Small eye (Sey) phenotype in mouse (Hill et al., 1991) and rat (Matsuo et al., 1993), as well as the human syndromes of aniridia (Ton et al., 1992; Glaser et al., 1992; Jordan et al., 1992) and Peter’s anomaly (Hanson et al., 1994). Pax-6 mutations in Drosophila have been mapped to the eyeless gene (Quiring et al., 1994), and targeted overexpression of mouse Pax-6 in flies induces ectopic eyes, suggesting a highly conserved function (Halder et al., 1995).

Several lines of evidence suggest that Pax-6 also regulates forebrain development, but the data are fragmentary. Pax-6 is expressed very early in the rostral neural folds (Walther and Gruss, 1991), with a caudal boundary thought to presage the morphological boundary between prosencephalon and mesencephalon. The broad prosencephalic expression becomes restricted later to certain forebrain nuclei, suggesting functions in specific subsets of neurons (Walther and Gruss, 1991; Stoykova and Gruss, 1994). Descriptions of mutant brains in both late embryos and adults support this idea. Sey/Sey mice have substantial layering defects in forebrain in heterozygotes and homozygotes (Schmahl et al., 1993), and similar defects were observed in a compound heterozygous human (Glaser et al., 1994). The late embryonic forebrain also appears to have defects in regional patterning, as Sey/Sey embryos display gross disturbances in regional growth in the diencephalon, accompanied by the failure of distinct nuclei to form (Stoykova et al., 1996). Sey homozygous neonates also lack olfactory bulbs (Hogan et al., 1986). An earlier, more primary role in neuronal patterning was suggested by comparison of Pax-6 expression and the initial scaffold of tracts in zebrafish (Macdonald et al., 1994). The tpoc axons project caudally along a Pax-6 expression boundary, and the tract of the pc forms just rostral to the caudal border of Pax-6. These correlations suggest that Pax-6 expression domains, or their borders, may pattern the initial neurons.

As a first step toward understanding the genetic regulation of early neuronal patterning in the prosencephalon, we have recently described the relationship of prosomeres and initial neuronal patterns in the mouse (Easter et al., 1993; Mastick and Easter, 1996). In the context of this initial brain development, we relate Pax-6 expression patterns to prosomeres and neuronal patterns. To test the functional significance of these correlations, we examine this neural patterning in homozygous Sey mutant embryos, and find defects in specific prosomeres, neuronal cell bodies, and axon navigation.

Pax-6 expression studies were carried out on CD-1 mouse embryos (Charles River Laboratories), obtained by timed matings, with noon of the day of the vaginal plug designated E0.5. The pregnant females were killed using CO₂ and the uteri were washed in ice-cold 0.1 M phosphate. The embryos were dissected free in 4% paraformaldehyde/0.1 M phosphate buffer, fixed in this solution for 2-4 hours at room temperature, and stored in fixative at 4°C.

We primarily used the original Sey mutant allele (Roberts, 1967). The founder Sey/+ male was generously provided by Anthony LaMantia (Duke University), and crossed to +/+ FVB females to produce Sey/+ animals. Sey/+ animals were intercrossed, and approx-imately 25% of their progeny had abnormalities of the optic vesicle and nasal pit diagnostic of Sey/Sey embryos (Hogan et al., 1986), while the remaining embryos appeared morphologically wild type. Our initial studies used these normal littermates (including both Sey/+ and +/+) as controls for developmental stage in comparison to Sey/Sey embryos. We also examined three E10.5 litters from parents carrying another Sey allele, Sey^neu (Hill et al., 1991; Schmahl et al., 1993), which were generously provided by N. Brown and T. Glaser (University of Michigan). The two alleles produced identical phenotypes (see Results).

To confirm the genotype of Sey/Sey embryos, and to distinguish Sey/+ and +/+ embryos, we used a PCR strategy to genotype the individual embryos of one E9.5 litter (8 embryos) and two E10.5 litters (total of 18 embryos). The Sey allele contains a single point mutation that creates a new DdeI restriction site (Hill et al., 1991), allowing PCR-based genotyping, essentially as recently described by Grindley et al. (1995). In brief, we isolated genomic DNA from the caudal half of each embryo, used flanking primers to amplify the Pax-6 exon 8, and digested with DdeI, which produced a characteristic 55 bp band for the mutant allele. The morphology and neurons of these individual genotyped embryos were then examined, as described in the last section of Results.

To detect Pax-6 transcripts in whole mounts, a plasmid containing an EcoRI/NheI restriction fragment from the mouse Pax-6 cDNA (Walther and Gruss, 1991) was kindly provided by K. Backs and P. Gruss (Gottingen). Whole-mount in situ hybridization using a digoxigenin-labeled antisense RNA probe, detected by anti-digoxigenin antibody coupled to alkaline phosphatase, was carried out essentially according to the procedure of Wilkinson (1992), as modified by Parr et al. (1993). The embryos were sagittally bisected, cleared in 80% glycerol, and photographed in a dissecting microscope.

To detect Pax-6 protein in 20 μm cryostat sections, a rabbit antiserum raised against a C-terminal peptide was generously provided by J. Davis and R. Reed (Davis and Reed, 1996). The antibody labeling protocol used dried milk as a blocker for nonspecific binding (Sundin and Eichele, 1990), biotinylated goat antirabbit secondary antibody (Vector), followed by avidin-HRP and peroxidase reactions using the purple VIP substrate (Vector). The sections were dehydrated through an ethanol series into xylene, coverslipped with Eukitt (Calibrated Instruments, Inc.), and photographed in a compound microscope equipped with Nomarski optics.

For double labeling with Pax-6 antiserum and the neuron-specific β-tubulin monoclonal antibody TuJ1 (Moody et al., 1987; Lee et al., 1990), the secondary antibodies were Cy3-conjugated anti-rabbit (Jackson Immunoresearch) and FITC-conjugated anti-mouse (Sigma), respectively. The sections were coverslipped in Prolong Antifade mounting medium (Molecular Probes), and examined with a Biorad confocal microscope.

The Lim-1 (Lhx1) probe was synthesized from a 1.5 kb cDNA provided by H. Sheng and H. Westphal (Fujii et al., 1994). The Gsh-1 probe was synthesized from a 799 bp cDNA fragment provided by T. Valerius and S. Potter (Valerius et al., 1995). The Pax-3 probe was synthesized from a 519 bp cDNA fragment provided by K. Backs and P. Gruss (Goulding et al., 1991). The Dbx probe was synthesized from a cDNA fragment provided by S. Lu and F. Ruddle (Lu et al., 1992).

Neuromeres were visualized in whole mounts after removal of skin and underlying mesenchyme. The embryos were then sagittally bisected, cleared in 50% glycerol/2% paraformaldehyde/0.05 M phosphate, and photographed in a dissecting microscope using glancing illumination against a dark field. The neuromeres were also visualized in cryostat sections cut at 20 μm thickness, counterstained with toluidine blue, and mounted in Eukitt.

Axon tracing used DiI, a fluorescent lipophilic dye, applied to fixed embryos as previously described (Mastick and Easter, 1996). For double labeling procedures, the fluorescence was photoconverted to a stable product (McConnell et al., 1989). In brief, DiI-labeled embryos were washed several times to remove formaldehyde for a total of 1 hour in 0.05 M Tris, pH 8.3, including 5 mg/ml heparin as an RNAse inhibitor (Shimamura et al., 1995), transferred to the same solution containing 0.6 mg/ml DAB, incubated on ice for 30 minutes, and coverslipped in a depression slide. Illumination of the embryos for 15-20 minutes with a 25× objective converted the fluorescently labeled neurons to a yellow-brown color.

Neurons were also labeled in whole mounts and sections with the TuJ1 antibody, using an HRP-conjugated secondary antibody, and DAB reactions, as previously described (Easter et al., 1993; Mastick and Easter, 1996).

We first introduce the morphological conventions and landmarks used in the prosencephalon, and then present the results in two sections, the first describing the Pax-6 expression pattern, and the second describing the morphological and neuronal defects in Sey/Sey mutant embryos.

We use a modified version of the morphological conventions proposed by Puelles and Rubenstein (1993). The dorsal-ventral directions are premised on the belief that the longitudinal axis of the neural tube is sharply bent ventrally at the cephalic flexure (Fig. 1A). The ventral surfaces of the prosencephalon and rhombencephalon face each other, so the dorsal-ventral direction depends on position along the longitudinal axis. The caudal prosencephalon contains two bulges, prosomeres 1 and 2 (p1, p2), flanked by transverse constrictions in the dorsal neural tube (Mastick and Easter, 1996; shown here in Figs 1B, 4). Rostral to p2, four more subdivisions have been postulated (p3-p6; Puelles and Rubenstein, 1993), but are not evident on E9.5 or 10.5. We therefore refer to the region immediately rostral to p2 as p3, leaving the p3 rostral boundary undefined, as well as p4-p6. The rostral prosencephalon contains two dorsal evaginations: the cerebral vesicle and the optic vesicle. Caudal to p1 is the mesencephalon (mes), a large, apparently undivided neuromere, followed by the rhombomeres making up the hindbrain (r1-r8). According to Puelles and Rubenstein (1993), the mes/p1 boundary becomes the dorsal midbrainforebrain boundary in the adult, separating superior colliculus from pretectum, and the p2/p3 boundary becomes the zona limitans interthalamica, separating dorsal and ventral thalamus in the adult (Puelles and Rubenstein, 1993; Stoykova et al., 1996).

Overall, Pax-6 expression in the brain was extensive, regionally specific and dynamic. On E9.5 (Fig. 1A), Pax-6 expression was limited to the dorsal neural tube, and extended through the entire prosencephalon, including the cerebral and optic vesicles, but excluding the roof plate (not shown). The mesen-cephalon was unlabeled. In the rhombencephalon, r1 was unlabeled, but r2 and more caudal rhombomeres were labeled. By E10.5 (Fig. 1D), the expression was reduced in some areas (notably p2), and increased in others (a heavily labeled cluster in caudal-ventral p1, a strip of scattered cells at the same dorsal-ventral level in mesencephalon, and a rostral expansion of the rhombencephalic strip into r1).

To determine more precisely the relationship between the borders of Pax-6 expression and neuromeric boundaries, sections perpendicular to the boundaries were labeled with Pax-6 antibody. The prosomeres stood out clearly in such sections (Fig. 1B,C); p1 and p2 are external bulges, with complementary internal recesses, and they are separated from one another by constrictions, characterized by both external grooves and internal ridges. The caudal limit of the p1 Pax-6 expression domain closely corresponded to the mes/p1 boundary on E9.5. The discontinuity was quite sharp between the unlabeled mesencephalon and the labeled p1. A few labeled nuclei arranged in radial columns defined the border in some sections (Fig. 1C), but in others, the discontinuity was even more abrupt (not shown). The vast majority of p1 nuclei were labeled (Fig. 1C). The labeling was not uniform, as most nuclei near the ventricular surface and a few near the pial surface were more heavily labeled, and a few unlabeled nuclei (about 1 in 50) were located near the pial surface. On E9.5, expression was continuous across the p1/p2 (Fig. 1C) and p2/p3 boundaries (not shown). By E10.5, p1 had signs of advancing differentiation, including unlabeled nuclei superficially (where differentiating neurons appear; Figs 1E, 2D), as well as a cluster of heavily labeled nuclei just ventral to the main domain of expression (Fig. 2C). As Fig. 1F shows, labeling was reduced in p2, most dramatically among the deep nuclei just caudal to the p2/p3 boundary, while p3 remained heavily labeled, especially in a superficial layer of nuclei extending across the boundary. In summary, the dynamic spatial patterns of Pax-6 expression in the mouse prosecephalon are similar to those recently described in zebrafish (Macdonald et al., 1994). Here, we extend this analysis to show that Pax-6 expression domains are closely bounded by at least two prosomere boundaries, and Pax-6 is expressed differentially along all three directions of the neuromeres: rostral-caudal, dorsal-ventral and superficialdeep.

Two heavily labeled blocks of cells were evident in caudal p1 (Fig. 2A). The arrangement of these cells (a dorsal block in caudal p1 and a ventral block in caudal p1 that extends into mes) was similar to that of the neuron cell bodies that project axons dorsally to pioneer the pc on E10.5 (Mastick and Easter, 1996). In double-labeled embryos (Fig. 2B), the pc somata overlapped extensively with the two blocks of Pax-6 label, but also extended outside of the Pax-6-labeled blocks, both ventrally and in the gap between them. Therefore, at least some pc neurons did not express Pax-6. Whether any of the pc neurons were Pax-6+ could not be established with in situs, because the label was too weak to distinguish lightly labeled from unlabeled cells, but the antibody gave strong nuclear labeling, so we used it on sections of p1 (Fig. 2C,D). Complete sets of serial sections from two embryos were analyzed. Of a total of 279 pc cell bodies, 260 (93%) lacked nuclear labeling, one nucleus was heavily labeled, nine faintly labeled, and nine indeterminate. We conclude that only a few pc neurons express Pax-6 on E10.5. As nearly all p1 nuclei were labeled by the Pax-6 antibody on E9.5 (Fig. 1C), it seems likely that pc neurons express Pax-6 in earlier stages of differentiation.

The first prosencephalic tract is the tpoc, which originates on E9.5 as axons projecting caudally from a cluster of neurons located at the base of the optic stalk (Easter et al., 1993).

To evaluate the possibility that Pax-6 might regulate tpoc axon navigation, we compared the tpoc somata and their axons to Pax-6 expression domains and boundaries. The shape of the tract on E10.5 correlated with Pax-6 expression domains (Fig. 3). The initial portion was a tight bundle of axons that passed ventral to the optic stalk through a lightly labeled region. As the tract entered a domain of heavier label, caudal to the optic stalk, the tract widened, as the axons fanned out both dorsally and ventrally. The ventral-most axons coincided with the ventral border of Pax-6 label. The dorsal-most axons approached, but did not enter the Pax-6+ cerebral vesicle, and so the tract was limited to the subdomain of expression ventral to the cerebral vesicle. Further caudally, the tract narrowed to cross the p2/p3 boundary, converging to leave a small ventral caudal labeled region axon-poor. The tract widened somewhat in p2, where Pax-6 expression was lower (Fig. 3A). In summary, the shape and boundaries of the tpoc correlated with the ventral part of the Pax-6 expression domain.

Sections through the tract were double-labeled with the Pax-6 antibody and DiI to determine more precisely the relation of Pax-6+ cells to the tpoc. The tpoc somata and the cells around them were unlabeled by the antibody on E9.5 or E10.5 (not shown), consistent with previous reports (Walther and Gruss, 1991). As the axons rounded the optic stalk, they were limited to a narrow superficial strip that was void of labeled nuclei, although Pax-6 label was both dorsal and deep to the axons in this region (Fig. 3D, E). The ventral boundary of the deep label corresponded to the ventral boundary of the tract.

More caudally, the width of the tract was coextensive with a patch of superficial, heavily labeled nuclei. These nuclei first appeared in the superficial zone of the narrow tract, and further caudally, both the tract and the layer of heavily labeled nuclei tract narrowed, and the axons became restricted to the dorsal most heavily labeled nuclei (not shown), leaving the ventral most nuclei in an axon-poor region, consistent with the whole mount (Fig. 3A). The most caudal of the heavily labeled nuclei bridged the p2/p3 boundary (see Fig. 1F). These cells were present on E9.5 in a location that anticipated the tract (Fig. 3F). In summary, the wide zone of the tpoc was closely associated with a specific subset of Pax-6-expressing cells that preceded the tract, suggesting a role for these cells in axon guidance.

The superficial position of these heavily labeled nuclei suggested they were neurons, and this was confirmed by double labeling with TuJ1, a neuronal antibody (Fig. 3G). All of the heavily labeled nuclei were associated with the cytoplasmic tubulin label, while few of the lightly labeled nuclei were. We conclude that the superficial cells that express Pax-6 at high levels are neurons.

As a test for the function of Pax-6 in brain development, we examined the prosencephalon in Sey embryos on E9.5 and E10.5, and found defects in general morphology, the prosomeres, the pc, and the tpoc.

We initially examined Sey/Sey embryos, which we will refer to as ‘mutant’, using for comparison their morphologically normal littermates, pooling both +/+ and Sey/+ embryos, labeled as ‘wild type’. At the end of the Results, we describe the analysis of individual genotyped Sey/+ embryos, from which we conclude that the early development of Sey/+ embryos is almost entirely normal, and finish by describing the identical mutant phenotype of another Pax-6 allele, Sey^neu.

Homozygous mutant embryos, readily distinguished by their enlarged optic vesicles (E9.5 and E10.5) and the absence of nasal pits (E10.5) (Hogan et al., 1986), appeared otherwise nearly normal. The overall size and number of somites was within the range of their wild-type littermates, and the morphology of the brain appeared relatively normal (Fig. 4A-D), both in size (measured by the length along the dorsal surface), and in shape (measured by the angle of the cephalic flexure).

The morphology of the cerebral vesicle was altered in the mutant. In whole mounts, the groove marking the caudal boundary of the cerebral vesicle was less prominent on E9.5 (Fig. 4A,B), as were the ventral and dorsal boundaries (not shown). By E10.5, the mutant cerebral vesicles were larger than wild type, and their ventral boundary was indistinct (Fig. 4C,D). Viewed from the dorsal side, the hemispheric sulcus formed a less acute angle between the cerebral vesicle and the adjoining prosencephalon. This morphological alteration led to the condition shown in Fig. 4H, where the mutant cerebral vesicles were no longer separated from each other at the dorsal midline, or from the ventral prosencephalon (compare to Fig. 4G). The hemispheric sulcus was also less prominent in more rostral sections (compare Fig. 8A,C). The wall of the mutant cerebral vesicles also was thinner. Taken together, our observations indicate very early defects both in the cerebral vesicle and in its junction with the ventral prosencephalon. These early growth disturbances apparently continue into later development, as by E12.5 the third ventricle is abnormally large, with wide openings to the cerebral ventricles (Stoykova et al.,1996).

p1 and p2 were altered in the mutant (Fig. 4). In wild-type whole mounts on E9.5, the groove marking the mes/p1 boundary was apparent as an interface between light and dark areas when viewed with glancing illumination, but this was not seen in the mutant (Fig. 4A,B). Sections of several E9.5 mutants (Fig. 4F) confirmed the absence of a neuromeric bulge within p1, and of an external groove and an internal ridge, the structures normally at the boundary. In contrast, p2 was prominent as a bulge, and an external groove formed a caudal boundary with p1 in the expected location. The mutant p1 neuroepithelium had normal thickness, but the mutant p2 was thinner. The p2/p3 boundary was apparent in whole mounts and sections (not shown).

The same defects were evident on E10.5. The mes/p1 boundary was still missing, so its absence on E9.5 was unlikely due to a delay in the development of this region. The p1/p2 and p2/p3 boundaries appeared normal (Fig. 4H,D). The p2/p3 boundary was more visible in the mutant, probably because of the altered cerebral vesicle morphology (Fig. 4C,D).

The absence of the mes/p1 boundary in mutants indicates a role for Pax-6 in the formation of this boundary. Pax-6 is not apparently involved in the formation of the p1/p2 and p2/p3 boundaries, correlating with the lack, or delay, respectively, of Pax-6 expression borders at these locations in wild type.

With the absence of the mes/p1 boundary, the morphological distinction between mes and p1 was lost, raising the question of the fate of p1 in the mutant. One possibility was that p1 regional identity was retained in the mutant, and that a cryptic regional boundary remained in the absence of a neuromeric boundary; alternatively, there could be a loss of regional distinction between mes and p1.

To identify the fate of p1, we examined three homeobox genes that were expressed in dorsal p1 as markers for p1 identity: Pax-6, and two others, Lim-1 and Gsh-1, with expression that is initiatedon E10.5 (Fig. 5). The mutant Pax-6 transcript can be used to label Pax-6 expression domains in mutant embryos (Grindley et al., 1995), since the Sey allele contains a point mutation which does not disrupt transcription (Hill et al., 1991). We found that the p1 domain of transcripts was missing in mutant embryos, and the most dorsal p2 label was retained (Fig. 5A,B). This loss of Pax-6 in situ labeling was very region-specific, as other domains were labeled as in wild type, and early, as the p1 domain was lost prior to E9.5, the earliest stage we examined. This result suggests the regulation of Pax-6 transcription is distinct in p1 and most of p2, requiring autoactivation as previously identified in lens ectoderm (Grindley et al., 1995). A second marker for p1 is Lim-1, which was previously described to be activated on E10.5 in a dorsal domain of rostral mesencephalon (Fujii et al., 1994). Upon reexamination of Lim-1 expression in wild type, we found that this dorsal domain was in fact in p1, closely abutting the mes/p1 boundary on the caudal side, and extending rostrally to mid-p1, in a superficial layer of cells (Fig. 5C, inset). Lim-1 was also expressed in several other domains, including a heavily labeled arc in ventral p1 and mes, a broad strip in rhombencephalon, and three patches of scattered cells in rostral prosencephalon. In the mutant, the dorsal p1 domain of Lim-1 label was lost. The other domains remained largely unaffected, although the ventral p1 arc was less pointed at its rostral end, and the scattered cells just rostral to the p2/p3 boundary were less prominent. Labeling in the spinal cord was also altered (not shown). A third marker for dorsal p1 is Gsh-1 (Valerius et al., 1995). Sections demonstrated that Gsh-1 was expressed in a layer of cells that, in contrast to Lim-1, were ventricular, but like Lim-1, stretched from the mes/p1 boundary rostrally to mid-p1. This dorsal p1 domain of Gsh-1 label was missing in the mutant, while other domains were labeled, including rhombencephalon, a strip of scattered cells in ventral p1 and mes, and another strip extending rostrally from the p2/p3 boundary. This rostral strip appeared to be labeled less intensely in the mutant. Taken together, markers for dorsal p1 (Pax-6, Lim-1, and Gsh-1) are all missing in the mutant, indicating that the loss of the p1/mes boundary is accompanied by the loss of markers of the p1 side of the boundary.

The loss of p1-specific markers might not be gene-specific, but rather the result of a general reduction in transcription in this domain. As a test for this possibility, we examined the expression of Pax-3, another Pax family member that is expressed in much of the dorsal CNS (Goulding et al., 1991). Pax-3 in situs labeled a dorsal domain that extended through mes and p1, and tapered dorsally in p2 (Fig. 5G). Pax-3 thus overlaps the mes/p1 boundary, and was coexpressed with Pax-6 in dorsal p1. In the mutant, Pax-3 labeling appeared unaffected in mes and p1, although labeling did not extend as far rostrally in p2 (Fig. 5H). We conclude that the dorsal p1 defects in Pax-6, Lim-1, and Gsh-1 expression are not due to a general defect in gene transcription

If p1-specific markers are lost, then what identity has been assumed by this region? We examined the expression pattern of the homeobox gene Dbx (Lu et al., 1992) as a marker for rostral mesencephalon. On E10.5 in wild type, Dbx was expressed in apparently all of the cells of dorsal mes, most heavily in superficial cells extending rostrally to the mes/p1 boundary (Fig. 5 I, inset). In contrast, dorsal p1 contained relatively light label, and only near the ventricular surface. Other expression domains include a strip extending diagonally from rostral p1 into dorsal p2, and heavily labeled domains in ventral rostral prosencephalon and rhombencephalon. In the mutant, the mes domain of Dbx label was expanded rostrally into the former p1 territory (Fig. 5J), with light label deep and heavy label superficial (inset), as in the wild-type mes. The loss of p1 markers is thus accompanied by the gain of Dbx expression, indicating a shift in the identity of the p1 region to that of mes.

If p1 is shifted to a mes identity, as suggested by the molecular markers, how would this transformation of identity affect neuronal development? Our previous study of axonal projection patterns in wild type found two projection patterns in common to mes and p1: tmesV (tract of the ‘mesencephalic’ nucleus of the trigeminal nerve) axons appeared in both p1 and mes on E8.5, projecting ventrally then turning caudally, with cda (circumferential descending axons) axons appearing later, on E10.0, and projecting straight ventrally (Mastick and Easter, 1996). Unique to dorsal p1, however, are the dorsal group of pc neurons, which project dorsally to cross the roofplate of caudal p1 on E10.5, thus pioneering the commissure. The pc, and the neurons that project into it, are thus p1-specific, and are therefore markers for p1 identity on the neuronal level.

We first examined the pc in TuJ1-labeled whole mounts. As Fig. 6A,B shows, a dense array of neurons in dorsal mes, p1, and p2 projected their axons ventrally, away from the roof plate in both mutant and wild-type embryos on E10.5, demonstrating that neuronal development in general was not grossly delayed or altered in the mutant. However, the pc was abnormal; several bundles of axons crossed the dorsal midline in caudal p1 of wild type, but no bundles were labeled in mutant embryos on E10.5. A few axons, greatly reduced in number, were present on E11.5 (not shown). In addition, the labeled somata in the roof plate of rostral p1 were more numerous in wild type than in mutant.

The lack of commissural axons in p1 could be due either to a blockage of their axons by the mutant roof plate, or to misspecification or loss of pc neuron cell bodies, resulting in a failure to project dorsally. To identify a blockage, we applied DiI to E10.5 embryos in dorsal p1, just lateral to the roof plate, where it would be expected to label any blocked pc axons. In wild type (Fig. 6C), a complex pattern of fibers was labeled, as predicted by the neuron map of dorsal p1. Contralateral projections included fibers projecting across the commissure. Ipsilaterally, anterogradely labeled axons projected ventrally from dorsal p1 somata (tmesV turn caudally, and cda project toward the ventral midline), and retrogradely labeled pc cell bodies in both dorsal and ventral patches. Mutant embryos (Fig. 6D) had the ipsilateral anterogradely labeled axons, but lacked axons crossing the dorsal midline (consistent with Fig. 6B), and lacked retrogradely labeled dorsal pc somata. The great majority of retrogradely labeled somata were located ventral to tmesV, indicating that the loss of pc neurons was limited to the dorsal subset. We can thus account for the early loss of the pc by the selective loss of pc axons originating from dorsal p1. The axon projections of the spared ventral pc neurons probably account for the sparse pc that forms on E11.5, and indicate that dorsally directed axon projections are not generally inhibited. Therefore the effect of the mutant Pax-6 is the specific deficit of dorsal pc axons, likely from a mis-specification of dorsal pc neurons or the failure of these cells to form. This defect is consistent with the transformation of dorsal p1 to a mes identity suggested by the molecular markers.

The correlation between the course of the tpoc and Pax-6 labeling (Fig. 3) suggested that tpoc axon pathfinding might be altered in the mutant, and this possibility was evaluated by labeling the tract with DiI on E10.5 (Fig. 7A,B). The initial tight part of the tract appeared normal in the mutant, and the tract widened caudal to the optic stalk as expected, but the tract was abnormal in several respects. A number of axons projected dorsally, far into the cerebral vesicle, where wild-type axons never ventured. Many axons formed loops, far more than in wild type. The longest of the caudal projections failed to pass from p3 into p2, in contrast to wild-type embryos, in which the longest had projected as far as the mes/p1 boundary (not shown). In a few cases, small numbers of axons projected ventral to the tract. To determine if the tpoc axons would eventually resume their normal caudal trajectory, we also examined the tpoc a day later, on E11.5. The wild-type tpoc (Fig. 7C) was similar in shape to that on E10.5, but it was wider, and contained more axons. The longest of these axons projected caudally through the mesencephalon, reaching the mes/r1 boundary (not shown). A few axons rounded the optic vesicle, forming the supraoptic tract (Wilson et al., 1990; Shimamura et al., 1995). In the mutant (Fig. 7D), the tpoc axons accumulated at the p2/p3 boundary and turned dorsally along it, with very few crossing into p2. The number of axons projecting into the cerebral vesicle increased relative to E10.5. In addition, the supraoptic tract in the mutant contained more axons than in wild type. A number of axons were misguided across the abnormally thin, distal roof of the optic vesicle. The pathfinding defects, all lessening the caudal advance of the axons, thus continued to compound on E11.5.

All of the errors occurred in that part of the tract where the heavily expressing Pax-6+ nuclei would normally be found. One possible cause of the tpoc defect is the failure of the superficial neurons to form in the mutant. As the C-terminal Pax-6 antibody will not work in the mutant, we used TuJ1 to label sections cut perpendicular to the tpoc in wild-type and mutant embryos (Fig. 8). Despite the altered morphology of the cerebral vesicle in the mutant, the surface of much of the prosencephalon was labeled as in wild type (compare Fig. 8A,C). The pathway of the tpoc was examined at high magnification; a similar superficial layer of neuron cell bodies was seen in both wild-type and mutant embryos (compare Fig. 8B,D). We conclude that the superficial layer of neuron cell bodies is formed in the mutant, and that the tpoc axon pathfinding defect is the result of a more subtle defect in Pax-6-dependent structures or molecules.

The examination of the heterozygous embryos revealed normal development relative to wild type. We could find no difference in optic vesicle or nasal pit morphology between+/+ and Sey/+ embryos on E10.5, confirming a previous report that the Sey/+ ‘small eye’ defect does not manifest itself until later in embryogenesis (Grindley et al., 1995). Unlabeled whole mounts had normal cerebral vesicle and p1 morphology on E9.5 and E10.5. The pc axons were numerous in TuJ1-labeled whole mounts on E10.5, so we concluded that pc neurons developed normally. On E10.5, the tpoc was labeled with DiI, revealing fairly normal tpoc axon projections, but 6 of 13 Sey/+ embryos had a few axons that projected ventral to the main tract, but only unilaterally, as the other side lacked such errors (not shown). We never saw this type of error in over 40 wild-type embryos from other litters, suggesting that heterozygotes have a variable, and minor pathfinding defect. Finally, Pax-6 in situs labeled p1 and p2 in Sey/+ embryos. We conclude that the Sey/+ prosencephalon develops fairly normally.

Another mouse Pax-6 allele is the Sey^neu mutation, a point mutation downstream of the homeodomain that is predicted to remove the C-terminal transactivation domain (Hill et al., 1991). The heterozygous eye phenotype of Sey^neu is identical to the original Sey allele. However, the effects of these alleles on forebrain development do not necessarily follow the eye phenotypes. We therefore examined three E10.5 litters resulting from Sey^neu + intercrosses, concentrating on prosomere morphology in whole mounts, the pc by TuJ1 labeling, and the tpoc by DiI labeling. Sey^neu/Sey^neu embryos appeared identical in each respect to Sey/Sey embryos, including loss of the mes/p1 boundary, loss of the pc, and the same tpoc axon pathfinding defects. We conclude that Sey and Sey^neu alleles have equivalent effects on forebrain develop-ment, supporting the conclusion that these mutations result in a similar loss of function.

Our results indicate that Sey/Sey mutant embryos have several early and specific defects in prosencephalon development. We attempt to explain the role of Pax-6 in the development of p1, the pc, and the tpoc.

Our demonstration that the mes/p1 boundary is missing in Sey/Sey embryos provides the first evidence for genetic regulation of prosomere formation. Pax-6 function appears to be specific for the mes/p1 boundary, because the p1/p2 and p2/p3 boundaries appear normal in the mutant. There are several possible functions for Pax-6 in boundary formation, depending on which of three cellular mechanisms we can hypothesize for prosomere formation. One possibility, given that neuromeres are centers of proliferation (Bergquist and Kallen, 1954; Guthrie et al., 1991), is that Pax-6 functions to activate proliferation in the center of p1, thus causing the bulge. This can be excluded because wild-type p1 is not obviously thicker than the corresponding region of the mutant brain (Fig. 4E,F), and a large number of neurons are still produced in mutant dorsal p1 (Fig. 5B). A second possibility is that Pax-6 directs the formation of specialized boundary cells, distinct from either mes or p1 cells, such as those found between rhombomeres (Heyman et al., 1995). We have yet to find any evidence for such cells at prosomere boundaries. We favor a third possibility, that Pax-6 is necessary in p1 to establish the identity of p1 cells as different from mes cells, and that this difference somehow results in the formation of a dorsal constriction along the interface of the two populations. The Sey mutation does appear to alter cell interactions; Sey/Sey and +/+ cells segregate into separate domains in the eyes of chimeras (Quinn et al., 1996), consistent with an identity conferred by Pax-6 activity. In experimental studies of rhombomeres, constrictions form only at the interface between grafts of even and odd numbered tissue (e.g. r2 and either r3 and r5), but not between even and even (r2 and r4; Guthrie and Lumsden, 1991). What could be occurring at the interface? It could be an inhibition of proliferation, as interneuromeric cells are known to divide less frequently than those in the neuromeres (Martinez et al., 1992). Alternatively, it could be a limitation of cellular movement across the interface; cells do not cross the mes/p1 boundary in normal chicks (Figdor and Stern, 1993). We predict that in the mutant, such crossings would inappropriately occur.

Although the mutant embryos lacked the mes/p1 boundary, the dorsal prosencephalon had a normal rostral-caudal length, and thus did not appear to contain any obvious deletions at this stage. In this respect, the Pax-6 mutant differs from previously reported mutants with neuromere phenotypes, all of which contain substantial early deletions: Wnt-1 (McMahon and Bradley, 1990; Thomas and Capecchi, 1990; Mastick et al., 1996), En-1 (Wurst et al., 1994), Otx-2 (Acampora et al., 1995), kreisler (McKay et al., 1994), and Hoxa-1 (Mark et al., 1993). For example, Wnt-1−/−?embryos have an initial loss of region-specific molecular markers, followed by a subsequent disappearance of mes and r1 cells (McMahon et al., 1992), ultimately resulting in the fusion of p1 to r2, a shortened neural tube, and a less acute cephalic flexure (Mastick et al., 1996). Sey/Sey mutant embryos lose a morphological boundary, but not a neuromere, leading to the conclusion that Pax-6 is not necessary, at least at this early stage, for survival of dorsal p1 cells, but rather for their identity.

Defects in p1 identity were apparent from the expression patterns of several genes in the mutant. The dorsal p1 domain of three markers was lost, including markers for ventricular cells (Pax-6, Gsh-1) and superficial cells (Lim-1), while the expression domain of Dbx, normally restricted to the mes side of the mes-p1 boundary, was expanded rostrally into the p1 domain. We conclude that the molecular identity of dorsal p1 has been at least partially shifted to that of the mesencephalon. This transformation is reflected on the neuronal level by the disappearance of the dorsal pc axons, leaving a mes-like pattern of only tmesV and cda projections.

The altered patterns of transcripts in dorsal p1 define a set of Pax-6-dependent homeobox genes involved in regulating the development of this region. The loss of Pax-6 transcripts from dorsal p1 and most of p2 indicates regionally restricted autoregulation, similar to that previously reported in the lens ectoderm (Grindley et al., 1995). This autoregulation could be direct, as Pax-6 can bind to and activate its own promoter in tissue culture cells (Plaza et al., 1993). However, the persistence of Pax-6 transcription in most regions implies that autoactivation is regulated in a very domain-specific manner, and that other activator(s) are sufficient to activate the Pax-6 promoter outside of p1. Our results also suggest that Gsh-1, Lim-1 and Dbx are genetically downstream of Pax-6. To date, known transcriptional targets of Pax-6 are limited to lens crystalline genes (reviewed by Cvekl and Piatigorsky, 1996), and possibly the cell adhesion molecule L1 (Chalepakis et al., 1994). We propose that Pax-6 also regulates a cascade of transcriptional regulators, each of which could direct a subset of the diverse developmental functions of Pax-6. The timing of Lim-1 and Gsh-1 expression in dorsal p1 are consistent with these genes being direct regulatory targets of Pax-6, as neither is expressed until after E9.5 (Fujii et al., 1994; Barnes et al., 1994; Valerius et al., 1995), well after Pax-6 in this region. Dbx is formally repressed by Pax-6. The promoters of these genes will have to be characterized to test for direct Pax-6 regulation. The function of these genes in p1 development has not yet been investigated, but either Lim-1 or Gsh-1 could be involved in boundary formation or pc neuron differentiation. Gsh-1 mutants have postnatal pituitary and hypothalamic defects, but analysis of early embryos has not been reported (Li et al., 1996). Lim-1 knockouts lack the entire forebrain and midbrain because of an early gastrulation defect (Shawlot and Behringer et al., 1995); conditional alleles will be necessary to test Lim-1 function in p1.

Both the pc and the tpoc were altered in the mutant, but the changes were profoundly different in the two structures. The alteration in the pc resulted from a misspecification or deletion of an identified group of cells while the tpoc was changed by aberrant axonal outgrowth.

The pc defect was traced to a failure to retrogradely label the dorsal pc neurons, which suggests that dorsal pc neurons fail to form or their axons are misdirected in the mutant. The observations that ventral pc somata are still present and form a (reduced) pc in the mutant have two important implications. First, the dorsal trajectories of ventral pc neurons indicate that dorsal-ward guidance cues are still present, arguing against a defect in extrinsic axon pathfinding for dorsal pc neurons. Second, dorsal and ventral pc neurons have different identities, despite the fact that they project their axons together, and the identity of the dorsal cells depends on Pax-6 expression. We suggest that Pax-6 functions in dorsal pc neurons by directing either their formation or early specification. The pc precursors likely express Pax-6, as we found that the Pax-6 antibody labels nearly all p1 nuclei on E9.5. However, on E10.5, when the pc neurons could first be specifically labeled by DiI in the roofplate, very few of them expressed Pax-6. Either Pax-6 is expressed early in dorsal pc precursors, acting to direct their birth or differentiation, then expression is rapidly downregulated, or pc neuron differentiation is dependent on an extrinsic signal from neighboring Pax-6-expressing cells. No early markers for pc neurons are yet available to sort out these possibilities.

The tpoc is formed of axons from Pax-6− cells, and their numbers are not apparently altered in the mutant; instead, the axons make errors. Their initial trajectory is normal, and when they enter the region where they would normally fan out, they do so. Thus, the initial tight section of the tract and the initial fanning out are both independent of Pax-6 function, but Pax-6-dependent navigational cues are necessary for the wide zone of the tract. We believe that these cues probably promote axon outgrowth rather than inhibit it, because in the mutant, even those tpoc axons that do project in the right direction are short and form loops, suggesting a less favorable environment than in wild type. Inhibitory cues cannot be excluded, however, and one possible candidate is the hemispheric sulcus, a sharply angled fold in wild type, which could mechanically hinder attempts to project dorsally into the cerebral vesicle. In the mutant, the sulcus is less prominent, but the effects of this altered morphology must be minimal, in that most axons remain ventral, and do not stream en masse dorsally.

Although we do not know their identity(ies), the Pax-6-dependent guidance cues are probably produced by a layer of superficial neuron cell bodies in this region. They express Pax-6 at high levels, and they are present on E9.5 (Fig. 3F), beforethe axons arrive, consistent with previous observations of TuJ1+ cells in this region (Easter et al., 1993). Furthermore,the distribution of these cells closely matches the length and width of the tract (Fig. 3B). They are not the only cells to contact the tract, since underlying neuroepithelial cells extend from ventricle to pia, and their pial endfeet contact growth cones and wrap axons. Differentiated neurons, in contrast, lose their ventricular attachment, and their nuclei migrate up to thepia. Thus, the tpoc axons are in contact with the neuroepithelial endfeet and the superficial somata of cells expressing Pax-6. We infer from the spatial coincidence of the tract and the superficial cells that they are more important than the endfeet. The superficial cells are neurons, and could express neuron-specific molecules, e.g. cell adhesion molecules, that the neuroepithelial cells do not.

We were able to determine that an apparently normal layer of neuron cell bodies was present in the appropriate region of the mutant prosencephalon, indicating that the mutant does not simply lack these neurons. However, the development of this region (p3, p4) is clearly disturbed in the mutant, as indicated by several lines of evidence. We note here alterations in the patterns of Lim-1, Gsh-1 and Dbx rostral to the p2/p3 boundary on E10.5 (Fig. 5). From E12.5 onwards, Stoykova et al. (1996) observed alterations in patterns of Pax-6, Otx-2 and Dlx-1 in p3 and p4, and the failure of corresponding ventral thalamic and hypothalamic nuclei to develop. These results suggest that disruption of patterning on any of several molecular or cellular levels could account for the tpoc axon pathfinding defects.

Pax-6 is only one of a large number of transcription factors expressed in regionally restricted patterns in the CNS. Taken together, this mosaic of partially overlapping gene expression patterns potentially provides an abundance of guidance information for axons to navigate through the brain. Substantial evidence for this hypothesis has accumulated for the homeobox gene Engrailed (reviewed by Retaux and Harris, 1996). Not only does a gradient of En expression in the mesencephalon correlate with the topographic projection of optic axons, but various experimental manipulations of the En gradient alter these axon projections. The guidance function of En appears to be mediated by several candidate cell surface molecules, notably Eph ligands that are upregulated by En (Logan et al., 1996). In the prosencephalon, Pax-6 may provide local positional information through similar molecules.

Wilson and others (1993) have suggested that the borders of expression domains are important for axon guidance. Although we have emphasized guidance within the Pax-6 domain, we do find two examples where the borders of Pax-6 expression may influence the tpoc axons. First, the tpoc is closely associated with the ventral border of expression, as in zebrafish. (This similarity is complicated by the fact that the mouse tpoc axons project dorsal to the border, while the fish axons remain ventral to it (Macdonald et al., 1994), but the dorsal-ventral differences in the tpoc trajectories of mice and fish have been noted before (Mastick and Easter, 1996), and are poorly understood.) The second example of guidance at a boundary is the failure of tpoc axons to cross the mutant p2/p3 boundary. This result was surprising, since we previously observed that axons in Wnt-1−/− mutant embryos could project without errors across a novel p1/r2 boundary (Mastick et al., 1996). One explanation for the failure in Sey/Sey embryos is the absence of special cells at the boundary which facilitate crossing, like the guidepost cells in the insect peripheral nervous system (Bentley and O’Connor, 1991). Alternatively, the axons may normally acquire a Pax-6-dependent signal from p3, probably from the Pax-6+ neurons, that allows them to project into p2. These sorts of locally acquired signals can influence the cell surface molecules expressed on axons, as shown for commissural axons encountering the midline in the vertebrate spinal cord (Dodd et al., 1988), or in the Drosophila CNS (Bastiani et al., 1987). We propose that such locally acquired signals could allow axons to accurately navigate long distances, reprogramming growth cones to gain responsiveness to the local cues in new regions.

We thank A. LaMantia for providing the Sey/+ founder of our colony, K. Backs and P. Gruss for the Pax-6 in situ probe, H. Sheng and H. Westphal for the Lim-1 probe, T. Valerius and S. Potter for the Gsh-1 probe, F. Ruddle for the Dbx probe, J. Davis and R. Reed for the Pax-6 antibody, T. Glaser and J. Lauderdale for advice on PCR genotyping, P. Eckler for expert mouse care, O. Sundin and P. Hitchcock for helpful suggestions. N. Brown and T. Glaser provided litters of Sey^neu embryos, and comments on the manuscript. This work was supported by a postdoctoral fellowship to G. S. M. (1F32NS09701), a grant from the Office of the Vice President for Research of the University of Michigan, and a grant from the NIH (NS33337).