Phox2b being an important regulator of the development of bm neurons, we wanted to further investigate its function in the specification of bm neurons by searching for novel genes regulated by Phox2b in fbm neuron precursors. We thus chose to compare gene expression in ventral r4 from heterozygous (Phox2b^LacZ/+) and homozygous (Phox2b^LacZ/LacZ) embryonic day 10.5 (E10.5) mutant embryos. At this stage, fbm differentiation is well underway in Phox2b^+/+ and Phox2b^LacZ/+ animals while no signs of bm differentiation are seen in the homozygous mutants 4. Four pairs of Phox2b^LacZ/+ and Phox2b^LacZ/LacZ RNA samples were amplified from microdissected ventral r4, labeled with Cy3 or Cy5, mixed and hybridized to two mouse cDNA microarrays, a microarray of the 15 K NIA probe set representing approximately 12,000 different genes and a home-made NeuroDev microarray representing 2,000 genes chosen for their expression in developing nervous tissue. We adopted as criteria for putative Phox2b-regulated genes a significance analysis of microarray (SAM) score (see Materials and methods) below -3.5 or above 3.5 for genes up- or down-regulated in the mutants, respectively, and a false discovery rate of 5%. By these criteria, 51 genes were found to be down-regulated and 23 up-regulated in the homozygous mutants (Additional files 1 and 2). Surprisingly, only three genes showed a greater than twofold change in expression. However, out of the 12 genes we selected for validation by in situ hybridization (ISH), 10 were confirmed as being clearly differentially expressed, despite fold changes as low as 1.14 and 1.13 for two of them (see below). Part of the rather low signal ratios on microarrays may be due to the fact that the floor plate, which should not be affected in the mutants, was always included in the material used for RNA preparation.

Only two of the genes that scored as underexpressed in the mutants, Tubb3 and Nfl, had been known from previous work (4 and unpublished data). We selected six for validation by ISH, which was done on sections through r4 from E10.5 Phox2b^LacZ/+ and Phox2b^LacZ/LacZ embryos. The expression of one of them, Sncg, was not noticeably affected in the mutants (not shown). The remaining five fell into two categories. App, coding for the amyloid beta (A4) precursor protein, Mapk8, also known as Jun N-terminal kinase (Jnk1), and Tubb3, the beta3-tubulin gene (already documented in previous work) are pan-neuronal genes expressed in all developing neurons. In the mutants, their expression was interrupted at the dorsal border of the normally Phox2b⁺ territory (Figure 1a–d). The small number of post-mitotic neurons that are generated in the homozygous mutants and locate to the ML, some of which are in the process of acquiring a serotonergic fate 1, thus do not appear to activate these genes that are otherwise pan-neuronal. Gap43, a major growth cone protein 1112, and to a lesser extent Nfl, coding for the neurofilament light chain, differed from the others by the fact that, although pan-neuronal as well, they were more strongly expressed in the fbm precursors than in other young post-mitotic neurons. Their expression was extinguished in the mutant territory (Figure 1e–h).

In the absence of Phox2b, less post-mitotic neurons are generated in ventral r4 and pan-neuronal genes are thus expected to be found down-regulated in the mutants. We therefore favored for further analysis up-regulated over down-regulated genes. We chose six genes among the candidates that were overexpressed in the mutants for validation by ISH. One of them, Fgfr1, did not show changes noticeable by ISH (not shown). The others revealed three distinct patterns of changes. In the heterozygotes, Igfbp5 was expressed in only part of the Phox2b⁺ territory in the ventricular zone (VZ) and was excluded from the mantle layer (ML), while in the homozygotes, it showed a patchy distribution throughout the Phox2b⁺ VZ and was strongly expressed in the ML (Figure 2a, b). Igfbp5 is a member of a family of modulators of insulin-like growth factor (IGF) activity that may also have IGF-independent functions 13. Like in our screen, Igfbp5 has obtained top scores in other differential screens on neural tissues (see, for instance, 1415), signifying perhaps that their expression is exceptionally labile. Their role in neural development has not been explored. Since Igfbp5 has been found to colocalize with areas of apoptosis during development 13, its overexpression may be related to the increased apoptosis caused by lack of Phox2b activity in ventral r4. Hes1, best known as an effector of Notch signaling 1617, is normally weakly expressed in the floor plate and more strongly in the dorsalmost part of the Phox2b⁺ VZ in the E10.5 hindbrain, but it invades most of the normally Phox2b-expressing VZ in a salt-and-pepper pattern in the mutants (Figure 2c, d).

We found Sfrp1 and Tcf3 to be normally expressed in and confined to the VZ, Sfrp1 only in the ventral part of the basal plate, and Tcf3 throughout the neural tube. In the mutant territory, their expression extended up to the neural tube border, thus including the post-mitotic neurons in the ML (Figure 2e–h). A similar expansion into the mutant ML has been found in previous work for Dll1 and Nkx2.2, which are normally confined to the VZ 4. In contrast to Tcf3, Sfrp1 appeared also to be up-regulated throughout the mutant VZ: the signal given by the Sfrp1 probe was stronger outside of the fbm progenitors in Phox2b^LacZ/+ embryos, but the reverse was true in Phox2b^LacZ/LacZ animals. Tcfs are transcription factors that function in the Wnt signaling pathway 1819. Secreted frizzled-related proteins (Sfrps) also function mainly in the Wnt pathway as antagonists of Wnt signaling, but Sfrp1 may also exert a Wnt-independent activity as a stimulator of neurite outgrowth and as a growth cone attractant for retinal axons 202122.

The Sox genes code for an extended family of SRY-box-containing transcriptional regulators 23. Sox13, a member of the SoxD group, has been identified as a diabetes autoantigen expressed in pancreatic beta-cells 24 and recently as a positive regulator of gamma-delta T cells 25. We found Sox13 to be expressed throughout most of the VZ, but almost absent in fbm progenitors. In the Phox2b mutants, its expression in the normally Phox2b⁺ VZ becomes as strong as elsewhere in the VZ (Figure 2i–l). We then asked whether Sox13 would also be excluded from the trigeminal bm progenitors in r2 that also express and depend on Phox2b. In ventral r2, Phox2b is expressed in the VZ and bm neuron production is underway at E9.5. One day later, bm neuron production has ceased and Phox2 is down-regulated in the VZ 7. At E9.5, strong Sox13 expression in the VZ was interrupted at the border of the Phox2b⁺ territory, similar to what has been observed for the E10.5 fbm progenitors. At E10.5, when Phox2b expression was about to disappear in the VZ, Sox13 became clearly detectable in the ventralmost progenitor domain (Figure 3a–d). Cessation of Phox2b expression and bm neuron production in ventral r2 is accompanied by a fate switch of the progenitors, which now produce serotonergic neurons. Their production depends on Mash1 activity, in the absence of which serotonergic neurons are not produced 7. We thus asked whether the expansion of Sox13 expression would take place in Mash1 mutants. Indeed, in Mash1^-/- embryos, Sox13 expression did not invade the ventral progenitor domain, although Phox2b expression has ceased in the VZ (Figure 3e, f).

In Hamburger and Hamilton stage 18 (HH18) chicken hindbrain, we also found Sox13 expression very weak or absent in the Phox2b⁺ domain that gives rise to bm/visceromotor neurons, while it extended into the ventral progenitor domain at HH24, when Phox2b has disappeared there. At this stage, a distinct Phox2b⁺ progenitor domain has formed in the basal alar plate. Strikingly, Sox13 expression in the VZ was interrupted at this location (Figure 3g–j). Together, these results suggest that Phox2b represses Sox13 expression in the hindbrain VZ. The effect may be indirect since Sox13 continues to be down-regulated in the progenitors when Mash1 is absent.

We then selected Gap43, Sfrp1 and Sox13 for gain of function experiments in the chicken embryo spinal cord to ask whether they were regulated by ectopically expressed Phox2b. We chose Gap43 as a panneuronal Phox2b-dependent gene because it was more strongly expressed in fbm neurons, Sfrp1 as an example of the VZ genes that expand into the mutant ML since it was at the same time up-regulated in the VZ, and Sox13 because Phox2b repressed it in bm progenitors.

We thus examined whether Phox2b electroporated into the spinal cord of HH12-13 chicken embryos could up-regulate Gap43 and down-regulate Sfrp1 and Sox13 expression, in line with the changes observed in Phox2b homozygous mutants. Gap43, normally expressed at this stage mainly in the post-mitotic motoneurons, was weakly up-regulated in the VZ at 24 hours after electroporation (hae). It was massively induced in the ML at 48 hae (Figure 4a, b). To exclude the possibility that this was a trivial consequence of the fact that Phox2b triggered neuronal differentiation, we overexpressed Ngn2, which has a similar propensity to promote neuronal differentiation 626. Ngn2, however, had no effect on Gap43 expression (Figure 4c, d). Sfrp1, normally expressed in a broad band of the VZ, was drastically down-regulated at 24 hae Not surprising for a cell-autonomous effect, Sfrp1 expression was only marginally affected at 48 hae, when the Phox2b-transfected cells had become post-mitotic and had moved to the ML. Ngn2 transfection did not induce changes in Sfrp1 expression (Figure 4e–h). The situation with regard to Sox13 was more complex. We found Sox13 expression to be confined to the VZ. At 24 hae in the Phox2b-transfected side, areas where Sox13 appeared to be repressed alternated with areas in which Sox13 was up-regulated, reflecting perhaps a highly dynamic expression pattern. Such a pattern was not seen after electroporation of Ngn2, which had instead a slight propensity of inducing Sox13 post-mitotically (Figure 4i–l). These results show that the changes in Gap43, Sfrp1 and Sox13 expression are not a simple consequence of the cells becoming prematurely post-mitotic and suggest that Phox2b does not exert its effect through a complex cascade of transcriptional events since the changes are already seen at 24 hae.

Misexpression of Phox2b or its paralogue Phox2a in spinal regions of the neural tube induces neuronal differentiation and a bm neuronal fate and imposes an axonal phenotype that shares many features with that of bm neurons. Many dorsally located Phox2-transfected neurons, instead of growing commissural axons as they normally do, project into the periphery through dorsal exit points as bm neurons do, but also at ectopic sites 2 (Figure 5c, d). As Gap43 is up-regulated by Phox2b both in fbm precursors and when misexpressed in the spinal cord, we thus tested whether it would mimic this phenotype and, in particular, induce axonal growth when overexpressed in the chicken neural tube. However, after electroporation of Gap43, the morphology of the neuroepithelium remained grossly normal. Neither did the cells relocate to the ML nor were there any signs of aberrant or increased axonal growth as produced by Phox2b (Figure 5e, f).

Overexpression of Sfrp1 in the chicken embryo spinal cord also did not produce gross anatomical alterations in the neural tube detectable by green fluorescent protein (GFP) labeling (Figure 6a, compare with Figure 5a). Since Sfrp1 appeared to be repressed by Phox2b in fbm precursors and when misexpressed in the spinal cord, we asked whether Sfrp1 would counteract some of the effects produced by Phox2b misexpression. After co-electroporation of Sfrp1 together with Phox2b, virtually all transfected cells expressed both GFP from the Sfrp1 expression vector and Phox2b, showing that co-expression is highly efficient (Figure 6d). When expressed together with Phox2b in this way, Sfrp1 did not prevent Phox2b from either driving neuronal differentiation and relocation to the ML or promoting ectopic neural tube exit of axons (Figure 6c, d; data not shown).

However, forced expression of the Phox2 genes 2 (Figures 5c and 6b) suppresses the growth of commissural axons, while cells co-transfected with Sfrp1 and Phox2b clearly grew commissural fibers as revealed by GFP staining in five out of five electroporated embryos (that is, n = 5/5; Figure 6c). To visualize the trajectory of pre-crossing commissural axons in electroporated embryos, we used anti-axonin-1 (TAG-1 in mammals) antibodies known to strongly label commissural fibers 27. Sfrp1 overexpression did not change the normal pattern of extension of commissural axons, which form a conspicuous fascicle running along the medial border of the motoneuron column (Figure 6g; n = 4/4). As expected, the axonin-1-expressing fascicle was absent at the transfected side after Phox2b electroporation (Figure 6f; n = 9/9). Still, anti-axonin-1-labeled fibers were observed in the Phox2b-transfected dorsal spinal cord, but they neither formed a tight fascicle nor extended beyond the dorsal border of the motoneuron column. Co-expressing Sfrp1 together with Phox2b partially, but not completely, restored the axonin-1-labeled fascicle that extends towards the floor plate (Figure 6h; n = 6/6).

The commissural axonal phenotype prompted us to ask whether Sfrp1 may counteract some of the fate changes affected by Phox2b. Sustained expression of Sfrp1 together with Phox2b reduced, but did not abrogate, the induction of Islet1,2 (Figure 6e; n = 3/3), normally seen after transfecting Phox2b alone 26. Phox2b, in addition to inducing bm markers, represses the interneuronal marker Lhx1,5 2 (Figure 7a), which is normally expressed by the interneuron populations dI2, dI4, dI6 and V0 2829. This effect was offset by co-expressing Sfrp1 (Figure 7c; n = 5/5). Hence, the cells now expressed a mixed phenotype, which is not normally encountered: Phox2b⁺, Islet1, 2⁺, Lhx1,5⁺. Sfrp1 increased the number of Lhx1,5-expressing cells in the dorsal spinal cord also when expressed alone, explaining its ability to counteract the down-regulation of Lhx1,5 caused by Phox2b. After Sfrp1 electroporation, the gap in Lhx1,5 expression corresponding to the dI5 neurons was not observed anymore at the transfected side (Figure 7b; n = 9/10). A fainter but clearly visible signal was also seen in the Sfrp1-transfected VZ in cells with neuroepithelial morphology, whereas it was never observed in control conditions, suggesting that Lhx1,5 is induced ectopically in the presence of Sfrp1 (Figure 7b; n = 9/10). Finally, we examined Pax2, which is expressed by several Lhx1,5⁺ populations 28, as a second marker of spinal interneurons. As expected, misexpression of Phox2b also down-regulated Pax2 in the transfected cells (Figure 7d; n = 3/3). In contrast to Lhx1,5, however, Pax2 was down-regulated as well by overexpressing Sfrp1 alone (Figure 7e; n = 3/3), and consequently, Sfrp1 did not prevent Phox2b from down-regulating Pax2 (Figure 7f; n = 3/3). Hence, Sfrp1 overexpression affects the phenotype of the neurons that are generated and thwarts the ability of Phox2b to induce a bm phenotype.

Sfrp1 (SARP2) has been found to sensitize cells in culture to pro-apoptotic stimuli, whereas Sfrp2 had the opposite effect 30 and prevented programmed cell death in the neural crest 31. We thus tested whether Sfrp1 might influence cell death in our overexpression paradigm. In line with published data 32, we found very few apoptotic cells in the neural tube at the stages studied (one to four TUNEL⁺ cells per section), and their number differed neither between electroporated and non-electroporated sides of the neural tube nor among different conditions (electroporation of Srp1 or Phox2b or of Sfrp1 plus Phox2b) (not shown).

In the chicken and mouse VZ, cells highly expressing Sox13 alternate with cells that express it less (Figures 2i, k and 4i', j'), a pattern that seems to be accentuated by overexpressing Phox2b (Figure 4i'). Such a pattern suggests that Sox13 expression in neuroepithelial cells is dynamic, perhaps in relation to their commitment to a neuronal fate. VZ expression of Sox13 driven by a strong constitutive promoter should cancel out these changes in expression. When overexpressed in this way, Sox13 did not disrupt neural tube structure or grossly alter neuroepithelial cell morphology. The transfected cells incorporated bromo-deoxyuridine (BrdU; Figure 8e; n = 3/3) and most expressed the VZ cell marker Sox2 (Figure 8h; n = 3/3). However, there were less cells expressing the neuronal marker NeuN in the transfected area, and none of the transfected cells co-expressed NeuN (Figure 8b; n = 3/3), indicating that constitutively high expression of Sox13 inhibits neuronal differentiation.

We next tested the idea that Sox13 overexpression may antagonize Phox2b in its capacity to drive neuronal differentiation. Phox2b misexpressed alone causes the transfected cells to accumulate in the ML and to express NeuN (Figure 8a), as expected from its propensity to promote neuronal differentiation and relocation to the ML. When Sox13 was misexpressed together with Phox2b, many transfected cells remained in the VZ where they either retained the morphology of neuroepithelial cells or aggregated near the lumen of the neural tube (Figure 8c, f, l; n = 3/3). Phox2b was still able to induce NeuN, but now many cells double-positive for NeuN and GFP were located in the VZ and had the morphology of neuroepithelial cells. Hence, Sox13, although unable to inhibit induction of NeuN by Phox2b, appears to prevent the transfected cells from losing their neuroepithelial morphology and from relocating to the ML. We then investigated further the phenotype of the cells constitutively co-expressing Phox2b and Sox13. We first established that, in the presence of Sox13, Phox2b was still able to promote cell cycle exit as shown by the fact that the transfected cells did not incorporate BrdU (Figure 8f; n = 3/3). However, while the neural stem cell marker Sox2 33 was never expressed in cells transfected with Phox2b alone, many doubly transfected cells still expressed Sox2 (Figure 8g, i; n = 3/3). Finally, we examined Islet1,2 as a marker for the bm phenotype induced by Phox2b, and Lhx1,5 as a marker for interneurons. Induction of Islet1,2 was completely repressed by co-transfection with Sox13 (Figure 8j, l; n = 3/3). By contrast, Sox13 did not prevent down-regulation of Lhx1,5 after Phox2b transfection, a result to be expected since Sox13 overexpression inhibits neuronal differentiation (Figure 8m; n = 3/3).

Phox2b^LacZ/+ mice 49 were intercrossed and the hindbrains of E10.5 embryos were dissected out and incubated for 2 minutes at 37°C with 1 μg/μl fluorescein di-(beta-D-galactopyranoside) (Sigma-Aldrich, Lyon, France). The left and right beta-galactosidase expressing domains of ventral r4 with the floorplate in between were excised under a fluorescent binocular and stored at -20°C in RNALater solution (Ambion, Austin, TX, USA). The rest of the embryo was used for genotyping as described by Pattyn et al. 49. Mash1+/- mice (obtained from François Guillemot) were intercrossed and the embryos genotyped as described 50.

Four r4 samples of the same genotype (either Phox2b^LacZ/+ or Phox2b^LacZ/LacZ) were pooled together and RNA was extracted and amplified using the RNAqueous-Micro Kit (Ambion) and the Message Amp aRNA amplification kit (Ambion) according to the instructions of the manufacturer. The quantity and the quality of the amplified RNA were verified using Agilent RNA 6000 Nano Assay and Bioanalyzer 2100 (Agilent Technologies, Massy, France).

Microarray hybridization was done using two cDNA collections, the NIA Mouse 15 k clone sets 51 and the NeuroDev clone sets (Ecole normale supérieure local collection), both spotted on Ultra GAPS II slides (Corning, Avon, France) using a BioRobotics MicroGrid microarrayer (Genomic Solutions, Huntingdon, UK) by the Transcriptome platform of ENS, Paris. Amplified RNA (2 μg) was retrotranscribed and labeled with dUTP-Cy5 or dUTP-Cy3 (Amersham, Piscataway, NJ, USA) using Superscript III Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) in the presence of 5 μg of random primers and appropriate buffer (Invitrogen). RNA was degraded by alkaline lysis and cDNA was purified using a QIAquick purification kit (Qiagen, Courtaboeuf, France). CDNA microarrays were pre-hybridized for 1 h at 42°C in 5 × SSC, 0.1% SDS, 1% bovine serum albumin, washed with distilled water and dried. They were hybridized overnight at 42°C in 25% formamide, 5 × SSC, 0.1% SDS with Cy5- and Cy3-labeled cDNA, washed in 1 × SSC, 0.2% SDS at 42°C, in 0.1 × SSC, 0.2% SDS and two times in 0.1 × SSC at room temperature and then spin-dried. Four biological replicas (independent dissections) for each microarray (15 k and NeuroDev) were done and each of these replicas was hybridized independently twice by inverting the labeling dyes (dye-swap). A total of 16 hybridizations where thus performed, 8 using NIA 15 k microarrays and 8 others using NeuroDev slides.

The hybridized microarrays were scanned using Genepix 4000B (Molecular Devices, Sunnyvale, CA, USA)) and the resulting image files analyzed by GenePix Pro 5.0 software (Axon). For each GenePix output file, two filters were applied, one to clear out spots and another one to discard saturating spots where the median foreground intensity was greater than 60,000 in one of the two channels. The resulting median foreground intensities were normalized, without background signal subtraction, for dye bias using a global Lowess correction 52. An expression matrix for each type of microarray (eight experiments each) was created gathering all normalized ratio data (M values). After normalization, we applied on each matrix a hierarchical clustering algorithm 53 using TMeV software 54 to detect a possible bias introduced by sample heterogeneity. The cluster obtained showed a clear separation according to dye-swap between hybridizations indicative of homogenous samples. The two matrices obtained were processed using the GEPAS web server 55 to remove genes with more than 30% missing values. A KNN impute step was performed to calculate missing values for the remaining genes using the 15 nearest neighbor profiles. A further scaling step was performed by dividing the ratios for each gene in one experiment by the experimental standard deviation and multiplying each value by the mean of all experimental standard deviations. The ratios were displayed such that a positive ratio describes a gene whose expression is higher in Phox2b^LacZ/+ than in Phox2b^LacZ/LacZ embryos. SAM 56 was used to search for differentially expressed genes and to take advantage of our four biological replicates to calculate the SAM score equivalent to the T-statistic value and to estimate the associated false discovery rate. We fixed the false discovery rate at 5% using the 90th percentile of falsely called genes. We chose in addition a SAM score above 3.5 or below -3.5 (for genes down- or up-regulated in the mutants, respectively) as an arbitrary cut-off since smaller scores increased the number of genes whose expression changes could not be confirmed by ISH.

We used pCAGGS::Phox2b-IRES2-EGFP 2, pCAGGS::Gap43-IRES2-EGFP, pCAGGS::Sox13-IRES2-EGFP, pCAGGS::Sfrp1-IRES2-EGFP or pCAGGS::IRES2-EGFP to express mouse Phox2b, mouse Gap43, human SFRP1 (the latter two subcloned from IMAGE clones) and human Sox13 (subcloned from Ultimate ORF Clone IOH44934, Invitrogen) together with GFP or GFP alone in the chicken neural tube driven by a composite chicken beta actin-CMV promoter. The spinal region of HH12-14 embryonic chick neural tubes were electroporated in ovo with the appropriate expression vectors (2 mg/ml or 1 mg/ml) essentially as described 8. Correct expression of all constructs was verified by ISH with the cognate probes. After electroporation, embryos were allowed to develop at 38°C for 24–72 h.

Mouse embryos and well-electroporated chicken embryos, as assessed by GFP fluorescence, were dissected out, fixed in 4% paraformaldehyde overnight, embedded in Tissue-Tek^® OCT™ (Sakura-Finetek, Zoeterwoude, The Netherlands) (mouse embryos) or gelatin (chicken embryos) and analyzed on 12 or 20 μm transverse cryosections, respectively. The methods used for ISH immunohistochemistry were as described previously 28. The following antibodies were used: rabbit anti-axonin-1 27, mouse monoclonal anti-BrdU (1/100; Sigma), mouse monoclonal (Roche, Basel, Switzerland) and rabbit (Chemicon, Temecula, CA, USA) anti-GFP (1/400), mouse anti-Islet1,2 (1/100) 57, rabbit anti-Islet1 (1/500; Abcam, Paris, France), mouse anti-Lhx1,5 (Developmental Studies Hybridoma Bank), mouse anti-NeuN (1/500), rabbit anti-Sox2 (1/1,000; Abcam), rabbit anti-Sox13 48 and adequate fluorescent secondary antibodies. For NeuN and Sox2 staining, sections were pre-treated with sodium citrate pH 6, 0.05% Tween-20, and for BrdU staining, sections were pretreated in 2N HCl, 0.5% Triton X-100 for 20 minutes at room temperature and neutralized. Apoptotic cells were detected using the Apo Tag kit (Oncor, Gaithersburg, MD, USA)) according to the manufacturer's instructions.