FGF8 Signaling Regulates Growth of Midbrain Dopaminergic Axons by Inducing Semaphorin 3F

Abstract

Accumulating evidence indicates that signaling centers controlling the dorsoventral (DV) polarization of the neural tube, the roof plate and the floor plate, play crucial roles in axon guidance along the DV axis. However, the role of signaling centers regulating the rostrocaudal (RC) polarization of the neural tube in axon guidance along the RC axis remains unknown. Here, we show that a signaling center located at the midbrain–hindbrain boundary (MHB) regulates the rostrally directed growth of axons from midbrain dopaminergic neurons (mDANs). We found that beads soaked with fibroblast growth factor 8 (FGF8), a signaling molecule that mediates patterning activities of the MHB, repelled mDAN axons that extended through the diencephalon. This repulsion may be mediated by semaphorin 3F (sema3F) because (1) FGF8-soaked beads induced an increase in expression of sema3F, (2) sema3F expression in the midbrain was essentially abolished by the application of an FGF receptor tyrosine kinase inhibitor, and (3) mDAN axonal growth was also inhibited by sema3F. Furthermore, mDAN axons expressed a sema3F receptor, neuropilin-2 (nrp2), and the removal of nrp-2 by gene targeting caused caudal growth of mDAN axons. These results indicate that the MHB signaling center regulates the growth polarity of mDAN axons along the RC axis by inducing sema3F.

Introduction

During development, growth cones navigate toward their targets by responding to cues in the extracellular milieu (for review, see Tessier-Lavigne and Goodman, 1996; Yu and Bargmann, 2001; Dickson, 2002; Huber et al., 2003). For axons to reach their correct targets, their growth directions must be precisely regulated. Because axonal growth in the neural tube occurs mainly in the rostrocaudal (RC) and dorsoventral (DV) directions, a fundamental question in neural development is how the polarized growth of axons along the RC and DV axes is achieved.

Accumulating evidence indicates that the roof plate and the floor plate play crucial roles in axon guidance along the DV axis (for review, see Colamarino and Tessier-Lavigne, 1995; Murakami and Shirasaki, 1997). In the spinal cord, for example, the repellent activities of the roof plate mediated by bone morphogenetic proteins (BMPs) (Augsburger et al., 1999; Butler and Dodd, 2003) and the attractive activities of the floor plate mediated by netrin-1 and sonic hedgehog (SHH) (Kennedy et al., 1994; Serafini et al., 1994, 1996; Charron et al., 2003) guide the axons of dorsally located commissural neurons toward the ventral midline. The floor plate causes changes in the growth cone's responsiveness to midline guidance cues, allowing commissural axons to cross the ventral midline (Shirasaki et al., 1998; Zou et al., 2000; Shirasaki and Murakami, 2001; Gore et al., 2008). Because both the roof plate and the floor plate act as signaling centers regulating the DV polarization of the neural tube (for review, see Tanabe and Jessell, 1996; Lee and Jessell, 1999; Jessell, 2000; Briscoe and Ericson, 2001; Caspary and Anderson, 2003; Chizhikov and Millen, 2005; Lupo et al., 2006), an intriguing question is whether signaling centers regulating the RC polarization of the neural tube also contribute to axon guidance along the RC axis.

To address this question, we focused on the midbrain–hindbrain boundary (MHB), a signaling center that regulates the RC polarity of the midbrain and rostral hindbrain (for review, see Liu and Joyner, 2001a; Wurst and Bally-Cuif, 2001; Raible and Brand, 2004; Nakamura et al., 2005). We selected axons from midbrain dopaminergic neurons (mDANs) as potential candidate neurons that are under the influence of the MHB because (1) these neurons arise near the ventral midline rostral to the MHB (for review, see Hynes and Rosenthal, 1999; Ang, 2006; Prakash and Wurst, 2006; Abeliovich and Hammond, 2007; Smidt and Burbach, 2007), (2) mDANs extend their axons rostrally to innervate the diencephalic and telencephalic targets (Lindvall and Björklund, 1983), and (3) the growth polarity of these axons is regulated by a substrate-associated cue(s) polarized along the RC axis in the midbrain (S. Nakamura et al., 2000).

In this study, we found that the trajectories of mDAN axons were perturbed in vitro by misexpression of fibroblast growth factor 8 (FGF8), a secreted molecule that can mimic patterning activities of the MHB (for review, see Liu and Joyner, 2001a; Wurst and Bally-Cuif, 2001; Raible and Brand, 2004; Nakamura et al., 2005). Furthermore, FGF8 induced the expression of an axon guidance molecule, semaphorin 3F (sema3F), in the midbrain/diencephalons, and the sema3F expression was markedly reduced by an FGF receptor tyrosine kinase inhibitor. These results suggest that sema3F guides mDAN axons as a downstream molecule of FGF8. Indeed, mDANs expressed the sema3F ligand-binding receptor neuropilin-2 (nrp2), and mDAN axon growth was inhibited by sema3F in vitro. Consistent with these observations, the removal of nrp2 by gene targeting caused some mDAN axons to aberrantly grow caudally. Collectively, these findings suggest that FGF8 derived from the MHB controls the rostrally directed growth of mDAN axons by inducing sema3F. Part of this study has been reported in preliminary forms (Yamauchi et al., 2004, 2007).

Materials and Methods

Animals.

Wistar rats (CLEA Japan or Nihon SLC) and nrp2-knock-out mice were used. Noon on the day of vaginal plug formation was designated as embryonic day 0.5 (E0.5), and the day of birth was designated as postnatal day 0 (P0). The generation of the nrp2^lacZ/lacZ mice has been described previously (Takashima et al., 2002). All experiments followed the Osaka University Guidelines for the Welfare and Use of Laboratory Animals.

Expression vectors.

To generate a pCAGGs-myc vector, myc epitope tag was ligated into an expression vector, pCAGGs (a gift from Dr. J. Miyazaki, Osaka University) (Niwa et al., 1991). A pCAGGs-sema3F-myc vector was prepared by subcloning the coding sequence of mouse sema3F into the pCAGGs-myc vector. The sema3F-coding sequence was obtained by PCR from E14.5 mouse brain cDNA. A pCAGGs-enhanced green fluorescent protein (EGFP) vector (Hatanaka and Murakami, 2002) was a gift from Dr. Y. Hatanaka (Nara Institute of Science and Technology, Nara, Japan); a CAG promoter-driven Fgf8b (Tanaka et al., 1992) expression vector, pCAGGs-Fgf8b, was a gift from Dr. Y. Tanabe (Osaka University); and a pCAGGs-Wnt-1-myc vector was a gift from Dr. D. Kawauchi (Chiba University, Chiba, Japan).

Expression of sema3F in COS-7 cells.

COS-7 cells were maintained in DMEM (Nissui) containing 10% fetal bovine serum, 600 mg/L l-glutamine, and 1% penicillin/streptomycin (Nacalai Tesque). The pCAGGs-sema3F-myc vector was transfected into the COS-7 cells with FuGENE6 (Roche Diagnostics) according to the manufacturer's instructions. Aggregates of COS-7 cells were prepared using the hanging drop method (Shirasaki et al., 1995). The pCAGGs-myc vector was used as a control vector.

Culture.

For whole-embryo culture, E9.5 or E11.5–E12 rat embryos were prepared by standard dissection techniques (Osumi and Inoue, 2001). In brief, dissections were performed in Tyrode's saline buffer at 37°C within 1 h of uterus collection. The decidual masses were dissected from the uteri, and the decidual layers and Reichert's membrane were removed, leaving the placenta intact. At E11.5–E12, a slit was made in the yolk sac and the amnion membrane to expose the embryo to the oxygenated medium. The embryos were then placed into small roller bottles of a whole-embryo culture apparatus (RKI10-0310; Ikemoto). Each bottle was filled with a 50% rat serum in DMEM/F-12 (D-8900; Sigma), supplemented with 3.85 g/L d-glucose, 2 mm l-glutamine, and 1% penicillin/streptomycin (Nacalai Tesque) (hereafter called DMEM/F-12) (Shirasaki et al., 1996). The bottles were then placed on the rotator of the culture apparatus at 37°C for 1–3 d. Oxygen was supplied to the embryos according to Osumi and Inoue (2001).

To analyze the effects of FGF8 on axonal growth and molecular expression in in vivo-like conditions, FGF8-soaked beads (FGF8-beads) were implanted onto the growth pathways of mDAN axons in whole-embryo culture preparations (see Fig. 1A). FGF8-beads were prepared according to Liu et al. (1999). In brief, heparinized acrylic beads (H5263; Sigma) were immersed in FGF8b (1 mg/ml; R & D Systems) containing PBS for at least 3 h at room temperature. Bovine serum albumin (BSA) solution (50 mg/ml in PBS; Sigma) was used as a control. The embryos were placed on a rotator at 37°C for 0.5–1 h before bead transplantation. For bead implantation, a small incision was made with a glass needle in the ventral diencephalon. Then, the beads were washed with PBS and pushed into the ventral diencephalon using forceps. The embryos were returned to the roller-bottles and cultured.

Figure 1.

Ectopically expressed FGF8 disorganizes the rostrally directed growth of mDAN axons. A, Schematic diagram showing the position of FGF8-bead implantation. FGF8-beads are embedded into the mDAN axonal pathway in the diencephalon. PC, Posterior commissure. B, C, mDAN axonal growth in an experimental (B) and a control (C) brain 2 d after bead placement. TH-positive axons were deflected dorsally in an FGF8-bead implanted brain. In contrast, in a BSA-bead-implanted brain, TH-positive axons grew rostrally ignoring the bead. White arrowheads show the position of beads, which were removed during the fixation procedure. D, Dorsal; R, rostral. D, A higher-magnification image of the boxed area in B. A caudally deflected mDAN axon is labeled by arrowheads. Growth cone-like structures were observed at the tip of axons (arrows). Scale bars: B, C, 200 μm; D, 50 μm.

To block FGF signaling in whole-embryo culture, an FGF receptor tyrosine kinase inhibitor, SU5402 (100 μm; Calbiochem) (Mohammadi et al., 1997), was added to the culture medium at the onset of the culture. Dimethylsulfoxide (DMSO), the vehicle for SU5402, was used in controls.

To examine the effect of FGF8 and sema3F on mDAN axonal growth, ventrorostral midbrain (VRM) explants were cultured as described previously (S. Nakamura et al., 2000), with some modifications. Briefly, VRM explants containing mDANs were dissected from E13.5 rat embryos. The explants and FGF8-beads or sema3F-expressing COS-7 cell aggregates were embedded in collagen gels at a distance of 300–1000 μm and cultured in DMEM/F-12 supplemented with 1% N₂ supplement (Invitrogen). The cultures were incubated in a 5% CO₂, 95% humidity incubator at 37°C for 2 d.

Electroporation.

DNA constructs were introduced into rat embryos by electroporation as described previously (Osumi and Inoue, 2001). After a 1–2 h preincubation on a rotator at 37°C, each embryo was transferred into a Petri dish containing Tyrode's saline buffer. DNA solutions (1.5 mg/ml in PBS with 0.01% Fast Green) were pressure-injected into the third ventricle of the embryo with a glass micropipette fitted to an injector (IM-30; Narishige). Electroporation was performed using tweezer-type electrodes with 2-mm-diameter discs (CUY6502; Unique Medical). Electric pulses of 10 or 15 V were charged four times for 50 ms at 950 ms intervals using a square-pulse generator (CUY21, Nepa Gene Company; or ECM830, BTX). The electroporated embryo was rinsed with Tyrode's saline buffer and returned to the culture apparatus.

In situ hybridization.

To prepare RNA probes for in situ hybridization, the following plasmid templates were used. A plasmid containing Fgf8 was a gift from Dr. G. Martin (University of California, San Francisco, San Francisco, CA), and a plasmid used to generate Wnt-1 riboprobe was a gift from Dr. D. Kawauchi (Chiba University). Partial cDNA fragments of sema3A (base pairs 1688–2248, GenBank accession number X95286), sema3B (base pairs 2153–2653, GenBank accession number X85990), sema3C (base pairs 94–637, GenBank accession number X85994), sema3D (base pairs 89–750, GenBank accession number AF268594), sema3E (base pairs 1906–2645, GenBank accession number AF034744), sema3F (base pairs 596–1300, GenBank accession number AF080090), sema3G (base pairs 1457–2196, GenBank accession number NM_001025379), nrp2 (base pairs 1337–1898, GenBank accession number AF016297), orthodenticle homolog 2 (otx2) (base pairs 222–1073, GenBank accession number NM_144841), and engrailed-2 (en-2) (base pairs 23–1082, GenBank accession number NM_010134) were amplified by PCR from rat or mouse brain cDNA. The coding sequence of EGFP was prepared by PCR amplification from the pCAGGs-EGFP vector. The resulting fragments were subcloned into the pGEM-T vector or the pGEM-T Easy vector (Promega) for subsequent riboprobe preparation. The antisense and sense riboprobes were transcribed in vitro using digoxigenin (DIG) or fluorescein isothiocyanate (FITC)-UTP (Roche Diagnostics).

In situ hybridization of whole-mount preparations was performed as described previously (Bally-Cuif et al., 1992). In brief, tissues were fixed with 4% paraformaldehyde (PFA) in 0.12 m phosphate buffer (PB) and dehydrated in an ascending series of methanol. After rehydration in a descending methanol series, the tissues were treated with 10 μg/ml proteinase K and hybridized with 1.0 μg/ml RNA probes for 16 h at 60 or 70°C. After RNase digestion and high-stringency washes, the samples were reacted with an alkaline phosphatase-conjugated anti-DIG antibody (1:2000; Roche Diagnostics). The signal was detected by subsequent reaction with nitroblue tetrazolium (NBT)/5-bromo-4-chloro-3-indolyl phosphate (BCIP) (Roche Diagnostics).

Two-color in situ hybridization was performed as described previously (Dietrich et al., 1997), with some modifications. The tissues were hybridized simultaneously with DIG- and FITC-labeled antisense riboprobes for 16 h at 70°C. The DIG-labeled probes were first detected with NBT/BCIP (Roche Diagnostics). After the first color development, alkaline phosphatase was inactivated by incubation at 70°C for 30 min. The tissues were incubated overnight at 4°C with an alkaline phosphatase-conjugated anti-FITC antibody (1:2000; Roche Diagnostics) and visualized by 2,4-iodophenyl-3,4-nitrophenyl tetrazolium chloride/BCIP (Roche Diagnostics).

Immunohistochemistry.

The whole-mount preparations were immunostained as described previously (S. Nakamura et al., 2000). The primary antibodies used were a rabbit polyclonal anti-tyrosine hydroxylase (TH) antibody (1:250; Millipore Bioscience Research Reagents) and a rabbit polyclonal anti-β-galactosidase (β-gal) antibody (1:10,000; ICN Biomedicals). After incubation with the primary antibodies, the preparations were incubated with a cyanine 3 (Cy3)-conjugated anti-rabbit IgG (1:100; Jackson ImmunoResearch) or a biotinylated anti-rabbit IgG (1:200; Vector Laboratories), then with Cy3-conjugated streptavidin (1:500; Jackson ImmunoResearch) or an avidin–biotin peroxidase complex (Vectastain ABC Elite kit; Vector Laboratories). Color was developed in 0.05% diaminobenzidine tetrahydrochloride and 0.005% H₂O₂ in Tris-buffered saline. All antibodies were diluted in PBS with 1% normal goat or horse serum (Vector Laboratories) and 1% Triton X-100.

For tissue sections, fresh brains were immediately immersed in 4% PFA in 0.12 m PB at 4°C from 90 min to overnight. The PFA-fixed brains were cryoprotected by immersion in 30% sucrose in 0.1 m PB, embedded in OCT compound (Sakura Finetechnical), and quickly frozen. Then, 20- to 50-μm-thick coronal or sagittal sections were cut on a cryostat (HM500; Zeiss), mounted on slides (Superfrost Plus; Thermo Fisher Scientific), and subjected to immunohistochemistry. The primary antibodies used were as follows: a rabbit polyclonal anti-nrp2 antibody (1:250), a rabbit polyclonal anti-TH antibody (1:250; Millipore Bioscience Research Reagents), a sheep polyclonal anti-TH antibody (1:100; Millipore Bioscience Research Reagents), a rat monoclonal anti-dopamine transporter (DAT; Slc6a3; Mouse Genome Informatics) antibody (1:1000; Millipore Bioscience Research Reagents), and a rabbit polyclonal anti-β-gal antibody (1:5000; ICN Biomedicals). The procedures for the anti-nrp2 antibody production followed those of anti-Robo antibodies (Tamada et al., 2008). The specificity of the anti-nrp2 antibody was confirmed by immunostaining with antibodies preabsorbed with a nrp2 ectodomain-Fc fusion protein or Fc protein and examination of immunoreactivities in nrp2-knock-out mice preparations. The secondary antibodies were a Cy3-conjugated anti-rabbit IgG (1:400; Jackson ImmunoResearch), an Alexa Fluor 488-conjugated anti-rat IgG (1:300; Invitrogen), an Alexa Fluor 594-conjugated anti-rabbit IgG (1:1000; Invitrogen), and a biotinylated anti-sheep IgG (1:500; Jackson ImmunoResearch). Cy2-conjugated streptavidin (1:600; Jackson ImmunoResearch) was used for visualization. All the procedures were the same as those used for whole-mount preparations, except that Triton X-100 in the antibody solutions was reduced to 0.05 or 0.2%. In double-labeling experiments, the sections were sequentially incubated with primary and secondary antibodies. Stained sections were observed with an epifluorescence microscope (BX60; Olympus) or a confocal microscope (TCS SP2 AOBS; Leica Microsystems).

Quantitative analysis.

To assess the defasciculation of TH-positive axons, images of TH-immunostained midbrains were captured by a confocal microscope (MRC-1024; Bio-Rad) at a depth of 0–75 μm below the surface of the preparation. The five most dorsally running axons were then selected in each individual animal. From each of the five, a line was drawn perpendicular to the ventral midline. The average length of the five lines was designated as the width of the TH-positive axonal bundle in the preparation.

To quantify the effect of sema3F on mDAN neurite outgrowth, TH-stained explants were photographed and digitized with an epifluorescence microscope (BX60; Olympus) equipped with a CCD camera (AxioCam; Zeiss). The image was thresholded, and the number of pixels of neurites was counted in the proximal and distal quadrants (see Fig. 5C) (Wang et al., 1996; de Castro et al., 1999) using NIH Image version 1.63. The central regions of the explants, occupied by cell soma, were excluded from the analysis. The number of pixels was then divided by the perimeter of the explant. This was designated as the “neurite outgrowth index.” The Mann–Whitney U test was used in each analysis.

Figure 5.

Sema3F inhibits the mDAN neurite outgrowth in vitro. A, B, VRM explants from E13.5 rat embryos were cocultured with aggregates of COS-7 cells transfected with either the pCAGGs-myc (A) or the pCAGGs-sema3F-myc (B) vector in collagen gels. After 2 d, the explants were fixed and stained with an anti-TH antibody. Neurite outgrowth from VRM explants, which is almost symmetric in a control coculture (A), appears to be inhibited by sema3F (B). C, Scheme of the method used to quantify neurite outgrowth. Scale bar, 250 μm.

Results

Disruption of mDAN axon growth by ectopically expressed FGF8

We explored whether MHB activities are involved in the rostrally directed growth of mDAN axons. For this, the influence of ectopically expressed FGF8, a signaling molecule that mediates patterning activities of the MHB (for review, see Liu and Joyner, 2001a; Wurst and Bally-Cuif, 2001; Raible and Brand, 2004; Nakamura et al., 2005), on mDAN axonal growth was examined. Rat embryos were dissected at E11.5–E12, before mDANs initiate axonal growth (n = 7) (supplemental Fig. 1A, available at www.jneurosci.org as supplemental material) (S. Nakamura et al., 2000), and subjected to whole-embryo culture. FGF8-beads were implanted into the mDAN axonal pathway in the diencephalon at the beginning of the culture (Fig. 1A). We first confirmed whether FGF8-beads were able to mimic MHB activities with in situ hybridization for en-2, an early marker of midbrain development (Davis et al., 1988). In agreement with previous studies (Crossley et al., 1996; Liu et al., 1999; Martinez et al., 1999; Shamim et al., 1999; Liu and Joyner, 2001b), en-2 transcripts were detected around the FGF8-beads, 2 d after implantation (n = 4 of 4) (supplemental Fig. 2A, available at www.jneurosci.org as supplemental material). In contrast, no expression was found in the region where BSA-soaked beads (BSA-beads) were implanted (n = 4 of 4) (supplemental Fig. 2B, available at www.jneurosci.org as supplemental material). These findings suggest that FGF8-beads can mimic MHB activities. We next examined the effect of ectopically expressed FGF8 on mDAN axonal growth. After a 2–3 d culture, mDAN axons were visualized by immunostaining with an antibody directed against TH, the rate-limiting enzyme of dopamine synthesis. In the absence of the beads, TH-positive axons originating from the ventral midbrain traveled rostrally during the culture period, similarly to the projection in vivo (n = 18 of 18) (supplemental Fig. 1B,C, available at www.jneurosci.org as supplemental material). However, when an FGF8-bead was implanted into the mDAN axon pathway, the growth of the mDAN axons was perturbed substantially (Fig. 1B). Although a fraction of these axons appeared to pass underneath the bead, a considerable number of axons were deflected dorsally (n = 11 of 12) (Fig. 1B). Some axons even directed caudally as if they were repelled by the beads (n = 5 of 12) (Fig. 1D). In contrast, in control cultures implanted with BSA-beads, mDAN axons elongated rostrally, ignoring the beads (n = 6 of 6) (Fig. 1C). These findings suggest that FGF8 signaling is involved in the guidance of mDAN axons.

Induction of sema3F by FGF8

Implantation of FGF8-beads repatterned surrounding tissue as indicated by en-2 expression (supplemental Fig. 2A, available at www.jneurosci.org as supplemental material). This observation, together with the induced repulsion toward mDAN axons by implanted FGF8-beads (Fig. 1B), prompted us to test the possibility that FGF8-beads affect mDAN axonal growth by inducing a repellent molecule(s).

Because class 3 semaphorins (sema3A–sema3G) are repellent molecules that contribute widely to axon guidance (for review, see Yu and Bargmann, 2001; Dickson, 2002; Huber et al., 2003), class 3 semaphorins are potential candidates for such repellents downstream from FGF8. Therefore, we explored the mRNA expression of class 3 semaphorins in the midbrain (supplemental Fig. 3, available at www.jneurosci.org as supplemental material). Of the seven class 3 semaphorins, only sema3F transcripts were detected at the MHB at the stage of mDAN axon development (n = 8) (Fig. 2A). The sema3F expression pattern was complementary to that of the mDAN axon trajectories (n = 4) (Fig. 2, compare A, D). We then examined whether FGF8 induces the expression of sema3F using whole-embryo culture preparations implanted with FGF8-beads. Sema3F mRNA was detected in the region surrounding the FGF8-beads after 2 d in culture (n = 7 of 8) (Fig. 2B). In contrast, no sema3F transcripts were induced by BSA-beads (n = 5 of 5) (data not shown). Together, these findings raise the possibility that FGF8 can induce sema3F expression in the midbrain. This notion was supported by an independent experiment using electroporation-based gene transfer. When a plasmid coding for Fgf8b was introduced into the midbrain followed by whole-embryo culture, sema3F was induced in the area surrounding the Fgf8-expression (n = 9 of 11) (supplemental Fig. 4A, available at www.jneurosci.org as supplemental material). In preparations electroporated with mock plasmid, the sema3F expression was unaffected (n = 9 of 9) (supplemental Fig. 4B, available at www.jneurosci.org as supplemental material).

Figure 2.

Expression pattern of sema3F and its induction by FGF8-beads in whole-embryo culture. A, The distribution of sema3F transcripts in rat midbrain at E13.5. B, sema3F expression in a midbrain preparation 2 d after an FGF8-bead (arrowhead) implantation. Sema3F transcripts were induced in the region just ventral to the bead (arrow). C, The distribution of nrp2 mRNAs in an E13.5 rat midbrain. Nrp2 mRNAs were observed in the rostral region of the ventral most midbrain (arrowhead). nrp2 was also detected in the IIIn and IVn. D, An E13.5 rat midbrain hemisphere immunostained for TH. mDAN axons appear to avoid the sema3F-expressing region (compare A, D). PC, Posterior commissure; Zli, zona limitans intrathalamica; IIIn, oculomotor nucleus; IVn, trochlear nucleus. Scale bars: A, C, D, 300 μm; B, 375 μm.

We next explored whether sema3F is induced by FGF8 during embryonic development. If sema3F expression in the midbrain is induced by FGF8, one should expect that sema3F expression is preceded by the expression of Fgf8. As expected, whereas FGF8 mRNA was detected in rhombomere 1 (r1) from E10 (six-somite stage) (Fig. 3A, E10.0, n = 5, C, E10.5, n = 4), sema3F mRNA was first detected in the midbrain only at E10.5 (Fig. 3B, E10.0, n = 6, D, E10.5, n = 4). We next explored whether FGF signaling is required for sema3F expression in the midbrain. For this, rat embryos were subjected to whole-embryo culture in the presence or absence of the FGF receptor tyrosine kinase inhibitor SU5402 (Mohammadi et al., 1997) before the development of the sema3F expression in the midbrain (at E9.5 or zero- to two-somite stage). After a 27–30 h culture in control DMSO, the sema3F expression was clearly detected in the midbrain as indicated by otx2 expression (Simeone et al., 1992, 1993) (n = 11 of 11) (Fig. 3E). However, in SU5402-treated embryos, the hybridization signal for sema3F in the otx2-positive region was nearly undetectable (n = 19 of 19) (Fig. 3F). These findings support the notion that the expression of sema3F in the midbrain was induced by FGF8 signaling emanating from the MHB, although a quantitative analysis might be necessary to confirm this notion.

Figure 3.

FGF signaling is required for sema3F expression in the midbrain. A–D, Whole-mount in situ hybridization with Fgf8 (A, C) and sema3F (B, D) antisense riboprobes in E10.0 (A, B) and E10.5 (C, D) rat embryos. Expression of Fgf8 temporally preceded sema3F at the MHB. E, F, The distribution of sema3F (blue) and otx2 (brown) in a DMSO-treated (E) and a SU5402-treated (F) embryo. The colocalization of sema3F (blue) and otx2 (brown) was observed in a control embryo (E). In contrast, sema3F mRNAs in the otx2-positive region was nearly undetectable in a SU5402-treated embryo (F). sema3F expressions in the hindbrain were not affected by the application of SU5402. Scale bars: 200 μm.

Expression of nrp2 by mDAN axons

All class 3 semaphorins signal through a holoreceptor complex composed of a ligand-binding component and a signal-transducing component, except sema3E (for review, see F. Nakamura et al., 2000; Raper, 2000; Fujisawa, 2004; Kruger et al., 2005; Tran et al., 2007; Zhou et al., 2008). If sema3F contributes to mDAN axon guidance, its ligand-binding receptor, nrp2 (Chen et al., 1997, 1998; Kolodkin et al., 1997; Giger et al., 1998), should be expressed by these axons. Indeed, nrp2 mRNA was intensely expressed near the ventral midline region where mDANs are likely to be located (n = 8) (Fig. 2C, arrowhead).

To confirm nrp2 expression in mDANs, we next examined the colocalization of TH and β-gal in the ventral midbrain of E12.5 nrp2^+/lacZ mice. In this mouse line, the bacterial lacZ gene is knocked into the nrp2 locus (Takashima et al., 2002). As shown in Figure 4A–D, most TH-positive cells, in particular those near the ventral midline, highly expressed β-gal from the nrp2 locus (n = 4), indicating nrp2 transcriptional activity in most mDANs. We then investigated nrp2 protein expression in mDAN axons with double immunohistochemical staining for TH and nrp2. We found that many mDAN axons coursing through the rostroventral midbrain expressed high-level nrp2 at E13.5 (n = 8) (Fig. 4E–K), although some dorsally coursing mDAN axons expressed no or low-level nrp2. Together, these results indicate that nrp2 is expressed at least on a subset of mDAN axons.

Figure 4.

Nrp2 is expressed on a subset of mDAN axons. A, A midbrain coronal section of E12.5 nrp2^+/lacZ midbrain immunostained for TH (green) and β-gal (magenta). B–D, Higher-magnification views of the boxed area in A. B, C, TH (B) and β-gal (C) expression. D, A merged image of B and C. β-gal immunoreactivity was observed in almost all TH-positive cells (arrowheads). E, An E13.5 rat midbrain/diencephalon parasagittal section immunostained for TH (green) and nrp2 (magenta). F–K, High-power views of the boxed areas (f–h, i–k) in E. F, G, I, J, TH (F, I) and nrp2 (G, J) expression. H, A merged view of F and G. K, A merged image of I and J. The colocalization of TH and nrp2 was observed in the rostroventral region (F–H) but not in the dorsal region (I–K). Scale bar: A, E, 100 μm; B–D, F–H, I–K, 20 μm.

Sema3F inhibits mDAN axon outgrowth

The results above raise the possibility that sema3F/nrp2 signaling regulates mDAN axon growth. We performed in vitro coculture experiments to test this possibility. VRM explants, which contained nrp2-positive mDANs (n = 9 of 9) (supplemental Fig. 5, available at www.jneurosci.org as supplemental material), were dissected from E13.5 rat embryos and cocultured with aggregates of sema3F-transfected COS-7 cells in collagen gels. After 2 d, TH-positive neurites elongated almost symmetrically from the VRM explants in the cocultures with mock-transfected COS-7 cell aggregates (Fig. 5A). In contrast, in the cocultures with sema3F-expressing COS-7 cell aggregates, neurite outgrowth from VRM explants was dramatically reduced on the side facing the cell aggregates (Fig. 5B). The neurite outgrowth index in the quadrant proximal to the sema3F-expressing COS-7 cells was reduced to ∼30% of that of the control (control: 1889 ± 413.7, n = 22; sema3F: 652.0 ± 126.8, n = 27; p = 0.0013, Mann–Whitney U test), whereas the index in the distal quadrant did not differ significantly between the two groups (control, 1564.97 ± 271.8; sema3F, 1161.32 ± 161.2; p = 0.2959, Mann–Whitney U test). We also found a significant decrease in the overall outgrowth of TH-positive neurites in the cocultures with sema3F-expressing COS-7 cell aggregates (control, 6342.8 ± 1194.85; sema3F, 3221.0 ± 424.31; p = 0.0244, Mann–Whitney U test). These results indicate that sema3F repels or inhibits mDAN axon outgrowth in vitro.

Caudally directed growth of mDAN axons in nrp2-knock-out mice

Sema3F mRNA was detected at the MHB, whereas mDAN axons extended rostrally in the ventral midbrain (Fig. 2, compare A, D). These findings raise the possibility that sema3F/nrp2 signaling contributes to the rostrally directed growth of mDAN axons. To explore this possibility, we first examined the RC polarity of mDAN axon growth in nrp2-knock-out mice. At E12.5, no notable difference in the RC growth polarity of mDAN axons was observed between nrp2⁺⁺ and nrp2^lacZ/lacZ mice (data not shown). However, at E14.5, we noted a clear phenotypic difference. Whereas mDAN axons never extended caudally in wild-type animals (n = 3 of 3) (Fig. 6A,C), mDAN axon subsets in the nrp2-knock-out mice grew caudally and even invaded the rostral hindbrain (n = 3 of 3) (Fig. 6B,C). These aberrant TH-positive fibers were observed until P0, the oldest stage analyzed (Fig. 6D,E) (nrp2⁺⁺, n = 3 of 3; nrp2^lacZ/lacZ, n = 3 of 3), suggesting their persistence. Because TH is expressed not only by dopaminergic neurons but also by noradrenergic neurons, these aberrant axons could be noradrenergic axons originated from the locus ceruleus. To confirm that these TH-positive fibers were indeed mDAN axons, we performed a double-labeling study with antibodies against TH and DAT (Slc6a3; Mouse Genome Informatics), a molecule expressed in mDAN subsets but not in locus ceruleus neurons (Ciliax et al., 1995; Freed et al., 1995). We found that DAT was expressed in a subset of aberrant TH-positive fibers (n = 5 of 5) (Fig. 6F–H). Moreover, these mDAN axons expressed β-gal (n = 3 of 3) (Fig. 6I–K), suggesting that nrp2-expressing mDAN axons were deflected caudally in the nrp2-knock-out mice. The RC patterning in the midbrain seemed normally in nrp2-knock-out mice (supplemental Fig. 6, available at www.jneurosci.org as supplemental material). Together, these findings demonstrate that sema3F/nrp2 signaling controls the RC growth polarity of mDAN axons.

Figure 6.

A subset of mDAN axons deflect in the caudal direction in nrp2-knock-out mice. A, B, TH immunostaining in coronal sections of wild-type (A) and nrp2-knock-out (B) mice at E14.5. Aberrant TH-positive fibers invaded the rostral hindbrain in an nrp2-knock-out mouse (B, arrowheads). C, Schematic representation showing the level of coronal sections in A and B. Note that the RC axis curves in the midbrain. PC, Posterior commissure; Tel, telencephalon. D, E, TH immunostaining in parasagittal sections of nrp2⁺⁺ (D) and nrp2^lacZ/lacZ (E) brains at P0. TH-positive fibers derived from the ventral midbrain, all of which were directed rostrally in a wild-type mouse (D), were deflected caudally in an nrp2-knock-out mouse (E, arrowheads). F–H, Aberrant TH-positive fibers (magenta) immunostained for DAT (green). F, G, TH (F) and DAT (G) expression. H, A merged view of F and G. DAT was expressed on an aberrant TH-positive fiber (arrowheads). I–K, Aberrant mDAN axons stained with antibodies against TH (green) and β-gal (magenta). I, J, TH (I) and β-gal (J) expression. K, A merged view of I and J. β-gal was expressed on aberrant mDAN axons (arrowheads). Scale bars: A, B, D, E, 200 μm; F–K, 10 μm.

Dorsal spread of mDAN axons in nrp2-knock-out mice

Sema3F mRNA was detected not only at the MHB but also in the dorsal midbrain/diencephalon along the mDAN axonal trajectory (Fig. 2, compare A, D). Therefore, we next asked whether sema3F/nrp2 signaling is also involved in defining the dorsal border of mDAN axon trajectories. Whole-mount preparations of E12.5 embryos were stained with an anti-TH antibody to test this idea. In wild-type and nrp2^+/lacZ mouse brains, mDAN axons grew rostrally, forming tightly fasciculated axonal bundles (n = 16 of 16) (Fig. 7A). In contrast, in nrp2-knock-out mice, these axonal bundles were less fasciculated, and a substantial number of mDAN axons grew more dorsally (n = 16 of 16) (Fig. 7B). The mDAN axonal bundles in nrp2-knock-out mice were significantly broader than those of their heterozygous and wild-type littermates (control, 95.1 ± 3.7 μm; nrp2^lacZ/lacZ, 129.1 ± 5.1 μm; p < 0.0001, Mann–Whitney U test). A similar dorsal spread and defasciculation of mDAN axons were observed when midbrain hemispheres were stained with an anti-β-gal antibody (nrp2^+/lacZ, n = 10 of 10; nrp2^lacZ/lacZ, n = 6 of 6) (supplemental Fig. 7, available at www.jneurosci.org as supplemental material). These data suggest that sema3F/nrp2 signaling defines the dorsal border of mDAN axon growth leading to the fasciculation of mDAN axons.

Figure 7.

Dorsal spread of mDAN axons in nrp2-knock-out mice. A, B, TH immunostaining in flat-mount brain preparations of wild-type (A) and nrp2^lacZ/lacZ (B) mice at E12.5. TH-positive fibers, which are relatively tightly fasciculated in a wild type (A), were defasciculated, with some deviating from their normal pathway in an nrp2-knock-out mouse (B, arrows). Scale bar, 200 μm. C, Schematic diagram showing the trajectory of mDAN axons in nrp2-knock-out mice. In nrp2-knock-out mice, the trajectories of nrp2-positive mDAN axons were disrupted: a subset of axons grow caudally (1) and another part of them deflect dorsally (2). The remaining part projects normally (3). PC, Posterior commissure.

Discussion

FGF8 is a patterning molecule secreted from the MHB. We found misexpression of FGF8 disrupted the mDAN axonal trajectory and induced sema3F expression in the diencephalon/midbrain in vitro. Developmental expression of Fgf8 at the MHB preceded sema3F, supporting that sema3F is also induced by FGF8 in vivo. sema3F was expressed in regions not occupied by mDAN axons and inhibited mDAN axon ourgrowth, suggesting that mDAN axons are guided by the growth-inhibitory activity of sema3F. Consistent with this, mDAN axons expressed a sema3F receptor, nrp2. Moreover, in nrp2-knock-out mice, some mDAN axons aberrantly extended caudally. These findings indicate that the MHB signaling center regulates the RC growth polarity of mDAN axons by inducing sema3F (Fig. 8A).

Figure 8.

Summary diagram of the present study and model diagram of the regulation of axonal growth polarity by signaling centers. A, Summary diagram showing the regulation of growth polarity of mDAN axons by FGF8 signaling. In early stages, MHB-derived FGF8 induces the sema3F expression at the MHB. Later, the axons of differentiated mDANs are guided rostrally by sema3F. Sema3F also provides the dorsal nonpermissive/repulsive territory to mDAN axons, defining the dorsal border of these axons. B, Model diagram showing the roles of signaling centers in axon guidance along the DV and RC axes. Along the DV axis, the growth polarity of axons is regulated by the signaling centers controlling the DV polarization of the neural tube. In an analogous way, the signaling centers controlling the RC polarization of the neural tube are also involved in the growth polarity of axons along the RC axis.

Disruption of mDAN axon trajectory by ectopic FGF8 expression

mDAN axonal trajectories were substantially perturbed by FGF8-beads (Fig. 1B). Although FGF8 acts as a guidance cue (Irving et al., 2002; Shirasaki et al., 2006), the fact that FGF8-beads did not repel mDAN axons (supplemental Fig. 8, available at www.jneurosci.org as supplemental material), together with a failure of MHB explants to repel mDAN axons (S. Nakamura et al., 2000), supports the view that FGF8 “indirectly” guides mDAN axons by inducing guidance molecules.

Some mDAN axons extended rostrally even after FGF8-bead implantation (Fig. 1B). This might be because some mDAN axons are insensitive to FGF8 signaling. This rostral growth can also be explained by assuming that early-extending mDAN axons have already passed the site of the bead before guidance molecules are induced. This is consistent with the findings that the rostral growth of mDAN axons initiates within 24 h after bead implantation (S. Nakamura et al., 2000), whereas alternations in gene expression by FGF8 requires 8–40 h (Liu and Joyner, 2001b).

FGF8 induced sema3F, which, in turn, guided mDAN axons (Fig. 8A). FGF8 and SHH collaborate to create the induction site for mDANs (Ye et al., 1998). Thus, FGF8 plays dual roles in mDAN development: induction and axon guidance.

FGF8 induces sema3F

We found that exogenously applied FGF8 induced sema3F expression in vitro. FGF8 may also induce sema3F expression in vivo because (1) Fgf8 expression at the MHB preceded sema3F and (2) SU5402, an FGF receptor tyrosine kinase inhibitor (Mohammadi et al., 1997), essentially abolished sema3F expression in the otx2-positive region in vitro. The findings that sema3F is expressed near Fgf8-expressing areas such as limb buds, kidneys, and brachial arches (Heikinheimo et al., 1994; Ohuchi et al., 1994; Crossley and Martin, 1995; Mahmood et al., 1995; Eckhardt and Meyerhans, 1998; Huber et al., 2005; Gammill et al., 2006) are consistent with our interpretation.

How does FGF8 induce sema3F expression? It seems unlikely that FGF8 signaling induces sema3F via Wnt signaling because Wnt-1 electroporation failed to induce sema3F (supplemental Fig. 4C, available at www.jneurosci.org as supplemental material). Expression of sema3F in Pea3-expressing motor pools (Cohen et al., 2005) implies that FGF8 may induce sema3F via its general downstream target, Pea3 (Raible and Brand, 2001; Roehl and Nüsslein-Volhard, 2001). Other potential candidates are the Sp family of transcription factors. Sp8, a member of the Sp family, is a downstream target of FGF8 in the telencephalon (Sahara et al., 2007). Human sema3F promoter does not contain a TATA-like box but rather putative Sp1 and Sp3 binding sites (Kusy et al., 2005). Sp8 expression at the MHB (Bell et al., 2003; Treichel et al., 2003; Kawakami et al., 2004; Griesel et al., 2006) supports that Sp8 mediates sema3F induction by FGF8.

sema3F was not detected in the Fgf8-expressing region (Fig. 3C,D). This might be because sema3F is induced by FGF8 in a concentration-dependent manner. Indeed, dosage-dependent gene induction by FGF8 was reported recently (Storm et al., 2003, 2006; Badde and Schulte, 2008). Another possibility is that yet unidentified molecules that negatively regulate sema3F expression are activated in the Fgf8-expressing region. Additional studies are required to test these possibilities.

Contribution of sema3F/nrp2 signaling to mDAN axon guidance

Sema3F repelled mDAN axons, and a substantial population of these axons expressed nrp2. A similar repulsive effect in vitro was also reported recently (Hernández-Montiel et al., 2008). However, whether and how sema3F/nrp2 signaling contributes to mDAN axon guidance in vivo has remained unknown. Using genetic approaches, we have unraveled in vivo roles of sema3F/nrp2 signaling: (1) control of the RC growth polarity of mDAN axons; and (2) determination of the dorsal border of mDAN axon growth (Fig. 7C). We also found induction of sema3F by FGF8 as discussed above.

The caudally directed growth of mDAN axons in nrp2-knock-out mice demonstrates that nrp2 signaling regulates the RC polarity of mDAN axon growth. This guidance defect is not attributable to midbrain/hindbrain patterning defects because the expression of midbrain/hindbrain genes otx2, en-2, and Fgf8, appeared unaffected in nrp2-knock-out mice (supplemental Fig. 6, available at www.jneurosci.org as supplemental material). Nrp2 is an essential component not only for sema3F but also for sema3B and sema3C signaling (Chen et al., 1998; Giger et al., 1998; Takahashi et al., 1998), and sema3C, sema3D, and sema3E were expressed near mDAN axons (supplemental Fig. 3, available at www.jneurosci.org as supplemental material). However, sema3F is the most likely ligand to regulate the RC polarity of mDAN axon growth because (1) among class 3 semaphorins, only sema3F was detected at the MHB (Fig. 2A) (Chen et al., 2000; Giger et al., 2000; Watanabe et al., 2004), (2) sema3F inhibited mDAN axon outgrowth in vitro (Fig. 5B) (Hernández-Montiel et al., 2008), (3) axon guidance defects found in sema3F-null mice are almost phenotypically identical to those of nrp2-knock-out mice (Chen et al., 2000; Giger et al., 2000; Cloutier et al., 2002, 2004; Walz et al., 2002, 2007; Sahay et al., 2003; Huber et al., 2005), (4) sema3C affects mDAN axons as a chemoattractant (Hernández-Montiel et al., 2008), (5) sema3E does not bind to nrp2 (for review, see Kruger et al., 2005; Tran et al., 2007; Zhou et al., 2008), and (6) mDAN axon subsets extended rostrally, apparently ignoring the oculomotor nucleus where sema3D was highly expressed (supplemental Fig. 3D, available at www.jneurosci.org as supplemental material).

Because sema3F is a secreted protein, sema3F might repel mDAN axons by forming a caudal-high/rostral-low gradient. Indeed, the graded expression of guidance molecules regulates the RC axon guidance in other systems (Lyuksyutova et al., 2003; Bourikas et al., 2005; Liu et al., 2005; Zhu et al., 2006). However, isthmic explants does not chemorepel mDAN axons (S. Nakamura et al., 2000). Therefore, it is more likely that sema3F forms a local nonpermissive/repulsive territory, which would force mDAN axons to extend rostrally. The finding that sema3F transcripts and mDAN axons were distributed in a complementary manner supports this view.

The dorsally deflected growth of mDAN axons in nrp2-knock-out mice indicates that nrp2 signaling defines the dorsal border of mDAN axons. Our findings that mDAN axons extended immediately ventral to the sema3F-expressing region (Fig. 2A,D) and that sema3F inhibited mDAN axon outgrowth imply sema3F acts as a dorsal barrier to mDAN axons. Sema3C, which is expressed in the pretectum, attracts mDAN axons in vitro (Hernández-Montiel et al., 2008), whereas mDAN axons aberrantly invaded the pretectum in nrp2-knock-out mice (Fig. 7A,B). The discrepancy between the previous and these results might be attributable to the difference between in vitro and in vivo conditions.

Possible involvement of other molecules

FGF8 might also induce guidance molecules for mDAN axons other than sema3F. The fact that nrp2 was not detected in all mDAN axons, together with the finding that many mDAN axons still grew rostrally in nrp2-knock-out mice, raises the possibility that mechanisms other than sema3F/nrp2 signaling are involved. One candidate is ephrin-A/EphA signaling, which regulates the rostral turning of a hindbrain lateral commissural axon (Zhu et al., 2006). Moreover, FGF8 induces the expression of ephrin-As (Shamim et al., 1999). However, mDAN axons grew rostrally even after the removal of glycosylphosphatidylinositol-anchored proteins (S. Mizushima and F. Murakami, unpublished observation). Wnts and SHH are other candidates. Gradients of these molecules regulate the RC axon guidance in the spinal cord (Lyuksyutova et al., 2003; Bourikas et al., 2005; Liu et al., 2005), and they are expressed in the ventral midbrain (Parr et al., 1993; Hynes et al., 1995a,b). Additional studies are required to test possible contributions by these molecules.

Roles of signaling centers in axon guidance along the RC and DV axes

The signaling centers controlling the DV polarization of the neural tube, the roof plate and the floor plate, play pivotal roles in axon guidance along the DV axis (for review, see Colamarino and Tessier-Lavigne, 1995; Murakami and Shirasaki, 1997). However, the role of signaling centers regulating the RC polarization in the RC axon guidance has remained unclear. Our results suggest that the RC polarity of mDAN axon growth is regulated by the MHB signaling center that governs the RC polarity of the midbrain/hindbrain (for review, see Liu and Joyner, 2001a; Wurst and Bally-Cuif, 2001; Raible and Brand, 2004; Nakamura et al., 2005). FGF8, a signaling molecule that can mimic MHB pattering activities, guides mDAN axons along the RC axis by inducing sema3F expression (Fig. 8A). A similar but somewhat different example can be found in commissural axon guidance. The floor plate is induced by a patterning molecule, SHH (for review, see Placzek and Briscoe, 2005). The induced floor plate secretes netrin-1. Netrin-1 in turn guides commissural axons ventrally (Kennedy et al., 1994; Serafini et al., 1994, 1996).

In addition to their indirect actions, patterning molecules controlling the DV polarization of the neural tube, BMPs and SHH, directly regulate the growth polarity of axons along the DV axis (Augsburger et al., 1999; Butler and Dodd, 2003; Charron et al., 2003). A growing body of evidence suggests that patterning molecules also directly regulate the axonal growth polarity along the RC axis (Lyuksyutova et al., 2003; Bourikas et al., 2005; Liu et al., 2005). It is interesting that the fundamental growth polarities of axons in the neural tube appear, in general, to be controlled by signaling centers, which are mediated by direct and indirect actions of pattering molecules (Fig. 8B).