Retinaldehyde dehydrogenase 2 and Hoxc8 are required in the murine brachial spinal cord for the specification of Lim1+ motoneurons and the correct distribution of Islet1+ motoneurons

During spinal neurogenesis, neural progenitors acquire positional identity along the anteroposterior (AP) and dorsoventral (DV) axes in response to the spatially restricted action of three main extrinsic signaling pathways: the Shh, Fgf and retinoic acid (RA) pathways(Liu et al., 2001; Mathis et al., 2001; Diez del Corral et al., 2003; Novitch et al., 2003). These pathways control the formation of distinct progenitor domains along the AP and the DV axis, by regulating specific combinations of transcription factors(Appel and Eisen, 2003). Spinal motoneurons originate from a common ventral progenitor domain and acquire columnar subtype identities as revealed by the settling position of their soma and their axonal projections in the periphery(Tosney et al., 1995; Landmesser, 2001). Motoneurons that innervate axial muscles are located in the medial motor columns (MMC) at all AP segmental levels. By contrast, motoneurons that innervate limb muscles are located in the lateral motor columns (LMC) only at brachial and lumbar levels (Tsuchida et al., 1994; Sharma et al., 1998). Two major motoneuron subtypes are generated within the LMC: the medially positioned LMC neurons that innervate the ventral limb muscles and the laterally positioned LMC neurons that innervate the dorsal limb muscles. Motoneurons are subsequently clustered into motor pools, wherein all neurons project to individual muscle targets in the limb(Tosney et al., 1995). The establishment of topographic neural maps requires the coordinated action of factors that: (1) specify motoneuron subtypes; (2) regulate lateral migration and segregation of newly postmitotic neurons to their final settling position;and (3) control axonal guidance to the target muscles. Although various classes of factors, including members of the LIM and Hox homeoprotein families, of the type II cadherin adhesion protein family, of the Ets-related transcription factors and of the Eph/ephrin ligand-receptor tyrosine kinase family (Sharma et al., 1998; Helmbacher et al., 2000; Kania et al., 2000; Livet et al., 2002; Price et al., 2002; Coonan et al., 2003; Dasen et al., 2003; Thaler et al., 2004), have been implicated in the control of these processes, it remains unclear how the action of these factors are coordinated and regulated in postmitotic motoneurons.

In addition to its function during early spinal neurogenesis(Appel and Eisen, 2003), RA signaling plays a major role in the specification of chicken motoneuron subtypes, soon after these cells withdraw from the cell cycle(Sockanathan and Jessell,1998; Sockanathan et al.,2003). These sequential roles highlight the importance of the spatiotemporal control of local RA levels, which mainly results from regulated expression of synthesizing enzymes, the retinaldehyde dehydrogenases (Raldh),and metabolizing enzymes, the cytochrome P450s Cyp26(Niederreither et al., 1999; Abu-Abed et al., 2001; Sakai et al., 2001). Raldh2 acts as the main RA-synthesizing enzyme during early embryogenesis(Niederreither et al., 1999). This enzyme is first expressed in the mesoderm adjacent to the node during early gastrulation; its expression expands to paraxial mesoderm during the phase of early neuronal specification, and at later stages Raldh2 is expressed in LMC neurons (Zhao et al.,1996; Niederreither et al.,1997; Sockanathan and Jessell,1998; Diez del Corral et al.,2003). In agreement with this expression pattern, transgenic mice that serve as in-vivo reporters of RA signaling have revealed RA activity in early paraxial mesoderm and later in postmitotic cells of the ventral spinal cord at brachial and lumbar levels(Rossant et al., 1991; Colbert et al., 1995; Solomin et al., 1998). As motoneurons exit the cell cycle, they begin to express the homeoprotein Islet1(Sharma et al., 1998). This expression is rapidly downregulated in lateral LMC neurons when these cells begin to express Lim1, thus generating a molecular distinction between lateral and medial LMC cells (Tsuchida et al.,1994). Lim1 participates to the regulation of the adhesion molecule Epha4, both proteins being required for the proper establishment of topographic projections (Helmbacher et al., 2000; Kania et al.,2000; Kania and Jessell,2003). The switch from Islet1+ to Lim1+ cells appears to depend on RA provided by early born LMC neurons, as in-vitro and in-vivo exogenous RA exposure of chicken LMC neurons represses Islet1 and promotes Lim1 expression(Sockanathan and Jessell,1998).

An additional earlier role has been assigned to RA produced by paraxial mesoderm at limb levels in the specification of brachial versus thoracic motor columns, as a blockade of RA signaling prevented the acquisition of an LMC identity by newly generated chick brachial motoneurons, which instead acquired a thoracic character (Sockanathan et al.,2003). Parallel studies have shown that Hoxc proteins, which are expressed at specific AP levels of the spinal cord in both neural progenitors and postmitotic cells, regulate the expression of molecular markers of either thoracic or brachial columnar fate. In particular, Hoxc6 is required for the expression of Raldh2 in the brachial LMC(Dasen et al., 2003). Conversely, regulation of Hox genes via RA has been described, although at an earlier stage in neural progenitors (Liu et al., 2001). Thus, there seem to be complex interactions between the regulation of RA signaling and Hox gene expression during spinal neurogenesis. Targeted disruptions of Hoxa10, Hoxc8, Hoxd9 and Hoxd10 led to aberrant patterns of motor axon connectivity in the limbs, suggesting roles for Hox proteins in the determination of motoneuron subtypes (Rijli et al., 1995; Carpenter et al., 1997; Tiret et al., 1998; de la Cruz et al., 1999). However, the cellular basis for these defects remains unknown. Retrograde labelings performed on Hoxc8^–/– mice revealed a reverse mediolateral position of a subset of motoneurons with respect to their target muscles, suggesting that Hoxc8 is required for establishing the distinction between medial and lateral LMC(Tiret et al., 1998).

To investigate the role of local RA synthesis during the specification of LMC neurons, we have generated mice bearing a conditional mutation that ablates Raldh2 function essentially in developing brachial motoneurons and later in mesenchymal cells at the base of the forelimb. Analysis of these mice indicates that RA synthesis within the LMC plays a crucial role in the brachial LMC in the specification of a subset of lateral cells, in the position of motor cell bodies as well as in the regulation of the expression of transcription factors and adhesion molecules required for correct axonal projections to target muscles. We also report striking similarities in the molecular defects observed in Raldh2 conditional mutants and Hoxc8 knockout mice, suggesting that RA signaling and Hox gene functions are essential for the specification of LMC cells.

The genomic construct used to generate a conditional (floxed) Raldh2 allele (Raldh2^L2) is described elsewhere(Vermot et al., 2003) (J.V. et al., unpublished). RarbCre transgenic mice(Moon et al., 2000) were crossed with heterozygous Raldh2 null mutants(Niederreither et al., 1999)and the resulting RarbCre_Raldh2^+/– mice were bred with Raldh2^L2/L2 mice to obtain RarbCre_Raldh2^L2/– mutants (hereafter called Raldh2^L–/–). Littermate animals without the RarbCre transgene were used as controls for all experiments. The Hoxc8 mutant allele was similar to that described previously(Le Mouellic et al., 1992),except that a GFP-Aequorin reporter (G5A)(Baubet et al., 2000) was inserted in place of the lacZ gene.

In-situ hybridization on cryosections was performed as described(Chotteau-Lelièvre et al.,2004). The Raldh2 probe used to detect both the wild-type and Raldh2^L– alleles hybridized with 75 bp of the fourth exon and the subsequent 3′ coding sequence. Other template plasmids were produced in our laboratory (RARβ) or kindly provided by M. Petkovich (Cyp26B1), R. Behringer (Lim1), C-C. Hui(Islet1), S. Pfaff (Lhx3), C. Henderson (Sema3e and cadherin 7), D. Wilkinson (Epha4), R. Klein (Ephb2), M. Capecchi (Hoxc6 and Hoxc8), X. Desbiens (ER81 and Pea3), M. Kmita (lacZ) and A. Esquela Kerscher(Gdnf).

Vibratome (100 μm) sections were processed for immunochemistry(Scardigli et al., 2003) using rabbit polyclonal anti-Epha4 (kindly provided by P. Charnay)(Becker et al., 1995) and mouse monoclonal anti-neurofilament (2H3; Developmental Studies Hybridoma Bank,DSHB) antibodies. X-Gal staining was performed as described(Scardigli et al., 2003).

Double in-situ hybridization and immunolabeling experiments were performed on predissected spinal cords fixed overnight in 4% paraformaldehyde and processed for in-situ hybridization with digoxigenin-labeled probes(Chotteau-Lelièvre et al.,2004), using an InsituPro (Intavis) robot and Fast Red (Roche) to reveal alkaline phosphatase activity. Subsequent immunofluorescence was performed as described (Scardigli et al.,2003) with mouse monoclonal anti-β-galactosidase (Promega),anti-Hoxc8 (C952-7E, Babco), anti-Islet1/2, anti-Lim1/2 and anti-Lim3 (40.2D6,4F2, and 67.4E12, DSHB), and rabbit polyclonal anti-Raldh2(Berggren et al., 1999),anti-β-galactosidase (5 prime-3 prime Inc.) and anti-Islet1 (K5, kindly provided by T. Jessell) (Tsuchida et al.,1994) antibodies. Although the latter antibody has been reported to cross-react with Islet2 (Tsuchida et al., 1994), it recognized only Islet1 in our conditions (1.5% of the K5+ cells were Lim1+ per hemisection, whereas 25% of Islet2+ cells were Lim1+, average of 15 sections from five controls). Alexa 488-Alexa 594- and Cy5-coupled secondary antibodies were used (Molecular Probes and Jackson ImmunoResearch Laboratories, Inc.). Both the sectioned and whole-mount spinal cords were analyzed using a Leica Sp2MP confocal microscope, except for Fig. 8A-C and Fig. 2C,D which were obtained using a Leica macrofluo and macro confocal (D.H. and J. L. Vonesch,unpublished), respectively. Images of cryosectioned spinal cords represent a single focal plan, whereas those of whole-mounts and vibratome sections were obtained by the projection of an average of 20 stacks of pictures(representing 40 μm in thickness). Analysis of dissected spinal cords was performed on a minimum of five control and Raldh2^L–/– embryos. Analysis of Hoxc8mutants was done on two embryos for each genotype; no variation for any marker was observed among each group. Whole-mount anti-neurofilament staining was performed as described (Maina et al.,1997), and embryos were documented on a Leica M420 macroscope. Comparisons between control and mutant spinal cords were performed in all cases on littermate embryos.

We first analyzed Raldh2 expression in the developing murine brachial spinal cord, using double immunohistochemistry (IHC) with an anti-Raldh2 antibody (Berggren et al.,1999) and an anti-Islet1/2 antibody, which labels all postmitotic motoneurons (Ericson et al.,1992). At embryonic day (E) 9.5, there were few Islet1/2+ cells in the brachial spinal cord and no expression of Raldh2 was observed in these cells, although Raldh2 was expressed in somites and mesenchymal cells surrounding the neural tube (Fig. 1A; see also Fig. 2E). Raldh2 was first detected in Islet1/2+ cells located laterally within the ventral horns at E10-10.5(Fig. 1B). By analogy with chick spinal cord development, these neurons are likely to correspond to early born motoneurons (Sockanathan and Jessell,1998). One day later, the MMC neurons located ventrally did not express Raldh2 (Fig. 1C),whereas the LMC cells did. In-situ hybridization of adjacent transverse sections of E12.5 spinal cords showed that Raldh2 was expressed in the two cell populations of the LMC, the Islet1+ cells and the Lim1+ cells (Fig. 1D-F).

A conditional Raldh2 allele (Raldh2^L2) was obtained (Vermot et al., 2003)(J.V. and I.L.R., unpublished). To generate conditional mutants, we used a RarbCre transgenic line, in which the regulatory region of the RA receptor RARβ2 controls Cre expression(Moon et al., 2000). This line drives Cre-mediated excision in the developing spinal cord and the trunk mesoderm as early as E8.5 (Moon et al.,2000) (J.V. and I.L.R., unpublished). Cre-positive Raldh2^L2/- mice (hereafter called Raldh2^L–/–) were obtained at a Mendelian ratio and had a normal life span, but displayed an abnormal flexure of the forepaw digits, especially of the most anterior (second and third) digits(Fig. 2A,B). Partial syndactyly of the second to fourth forelimb digits was also observed, albeit with a variable expressivity (Fig. 2B,arrowheads). No defect was recorded relating to the digit skeleton or the size and topology of distal limb muscle groups and their insertions, except for a slight hypoplasia of the extensor radialis muscles, which control the extension of anterior digits (data not shown). Thus, the abnormal digit flexure phenotype of Raldh2^L–/– mutants could not be attributed to a specific skeletal or muscle patterning defect.

We then examined the efficiency of the RarbCre transgenic line to excise the Raldh2^L2 allele by monitoring Raldh2 immunoreactivity in mutants. At E9.5, Raldh2 expression was not significantly altered in mutants compared with controls(Fig. 2C,D). However, Raldh2 protein was never observed in the brachial spinal cord of mutants at all subsequent stages (E10.5-15.5); residual protein was still present in scattered meningeal and mesenchymal cells surrounding the spinal cord(Fig. 2E-J; and data not shown). Raldh2 expression at E11.5 was also markedly diminished at the level of the brachial plexus (Fig. 2G,H), a region in which motoneurons make pathfinding decisions(Landmesser, 2001). At the lumbar level, a reduced number of Raldh2+ motoneurons was observed in the E12.5 mutant spinal cord (Fig. 2I,J, insets). This residual RA activity can explain the lack of abnormal hind limb phenotype in Raldh2^L–/–mutants.

Thus, excision of the Raldh2^L2 allele using the RarbCre transgenic line led to a complete absence of Raldh2 protein in brachial motoneurons. The pattern of excision of Raldh2^L2, especially in the mesoderm, was clearly distinct from the pattern of excision described using the same RarbCretransgenic mice in the RNA polymerase II large subunit locus(Moon et al., 2000) or the ROSA26 locus (J.V. and I.L.R., unpublished). Our result supports the finding that LoxP recombination is locus-position-dependent (Voojs et al.,2001). In summary, the Raldh2^L–/– mice constitute a model to study the early requirement for local RA synthesis during the early phase of motoneuron specification, and a default in these developmental decisions could be the cause of the adult phenotype.

We first examined the distribution of RA-responsive cells within Raldh2^L–/– spinal cord using an RA-responsive reporter transgenic line (RARE_hsp68_lacZ)(Rossant et al., 1991). The patterns of β-galactosidase (β-Gal) expression were analyzed by immunofluorescence in control and Raldh2^L–/–transgenic embryos. In E10.5 control embryos, β-Gal+ cells were scattered along the DV axis of the brachial spinal cord(Fig. 3A). In Raldh2^L–/– mutants, fewer β–Gal+cells were found in the ventral horns and in the dorsalmost region of the brachial spinal cord (Fig. 3B). We further showed that in the E11.5 control ventral spinal cord, some Islet1/2+ cells expressed β-Gal (Fig. 3C, arrows), whereas no Islet1/2+ cells were β-Gal+ in Raldh2^L–/– spinal cords(Fig. 3D), indicating the depletion of RA-responsive motoneurons. At E12.5, β-Gal+ cells were distributed along the dorsal and intermediate regions of control flat-mounted spinal cords (Fig. 3E). By contrast, in Raldh2^L–/– mutants, β-Gal+cells were essentially found in a broad longitudinal band within the intermediate region of the spinal cord(Fig. 3F, brackets). This pattern is reminiscent of RA activity in Raldh2 null mutants rescued by RA supplementation (Mic et al.,2002; Niederreither et al.,2002).

We then analyzed the expression of Rarb, the main RAR expressed during LMC specification (see Colbert et al., 1995), on flat-mounts and serial transverse sections of E12.5 spinal cords (Fig. 3G-L). Rarb expression was downregulated within the Raldh2^L–/– motor columns, particularly at the level of the Islet1+ posterior medial domain, which exhibited strongest expression in controls (Fig. 3L, asterisk). These results, in agreement with a previous study which showed that excess RA induces ectopic Rarb expression in the spinal cord (Colbert et al.,1995), demonstrate that lack of local RA synthesis in Raldh2^L–/– spinal cord leads to a decrease of RARβ-mediated activity in specific motor pools.

We then analyzed the identity of the brachial LMC neurons in Raldh2^L–/– mutants. We first quantified the number of Islet1/2+ motoneurons in the ventral spinal cord at E11.5. Islet1/2+cells were counted on serial transverse sections spanning the entire Raldh2 domain, which was mapped by in-situ hybridization (see Materials and methods). A significant 10% decrease of the number of Islet1/2+cells was found in mutants (Fig. 4A). Double IHC was performed on adjacent control and mutant sections using: (1) an anti-Lim1/2 antibody; and (2) an anti-Islet2 antibody(Fig. 4B) or an anti-Islet1 antibody (K5 antibody; Fig. 4C;see Materials and methods) to quantify Lim1+ or Islet1+ motoneurons. We found that, within the Raldh2 domain, the number of Islet1+ cells (medial LMC and MMC cells) was unchanged in Raldh2^L–/–mutants compared with controls, whereas the number of Lim1+/Islet2+motoneurons (lateral LMC cells) was significantly decreased by about 20% in mutants (Fig. 4B).

We next looked at the distribution of Lim1+ and Islet1+transcripts on E12.5 flattened spinal cords. The downregulation of Lim1 expression was visible in the motor column in mutants, whereas no change in the interneuron dorsal domain was noticeable(Fig. 4C,D). Surprisingly, the spatial distribution of Islet1 was altered in mutants(Fig. 4E,F). Two classes of phenotypes were obtained at about the same frequency. In the first class (data not shown; n=13), the two Islet1+ columns were continuous along the AP axis, instead of being interrupted by Islet1–cells. In the second, more severe class(Fig. 4F; n=11), Islet1 was no longer distributed into two columns in the anterior LMC, but instead was uniformly expressed. In all cases, the domain of high Islet1 expression in the posterior LMC was expanded in mutants(brackets, Fig. 4E,F). We further confirmed the decrease of Lim1 expression in motoneurons and the change in Islet1 distribution by in-situ hybridization of adjacent sections at various AP levels of the LMC(Fig. 4G-J, and data not shown).

We next analyzed the expression of selective markers of LMC motor pools. The overall organization of the LMC was not disrupted in Raldh2^L–/– spinal cord. Indeed, the expression domain of Ephb2, which closely matches that of Raldh2 in controls(Fig. 5A), was not spatially altered in mutants (Fig. 5A-D). In addition, the organization of the MMC (Lim3+ cells) and V2 columns(Chx10+ cells) were not modified in mutants, indicating that no change of the identity of LMC neurons toward MMC neurons or V2 interneurons had occurred in the mutants (data not shown). By contrast, we found a consistent decrease of the expression of cadherin 7 and Epha4 in the Lim1+ lateral LMC posterior domain, confirming the loss of a subset of Lim1+cells (data not shown, and Fig. 5E-H).

We also found that cells expressing the Ets-related transcription factor Pea3 were mispositioned within the LMC in mutants. The Pea3+motor pool is located in the posterior medial LMC and is constituted by over 95% of Islet1+ cells at segmental levels C7/8(Livet et al., 2002)(Fig. 5I,K). Although the Pea3 expression domain spanned a comparable length in control and mutant spinal cords at E12.5 (brackets, Fig. 5I,J), the distance between the ventral midline and the Pea3+ cells was markedly larger in mutants (horizontal bars, Fig. 5L). A similar observation was made with Semaphorin3E,which marks a subset of Pea3+ cells (data not shown). Pea3expression is regulated by glial cell-derived neurotrophic factor (Gdnf)produced at the level of the brachial plexus and later, in the hypaxial muscles innervated by Pea3+ motoneurons, the latissimus dorsi (LD) and cutaneous maximus (CM) (Haase et al.,2002; Livet et al.,2002). We then examined whether Gdnf expression was altered in mutants at E12.5. No significant change was detected in the distribution and levels of Gdnf transcripts, neither at the level of the brachial plexus nor in the CM in mutants, compared with controls(Fig. 5M,N). As these tissues display endogenous RA activity in control embryos(Fig. 5O), we monitored this activity in mutant embryos using the RARE_hsp68_lacZ transgene. A drastic decrease of β-Gal reporter activity was seen at the level of the brachial plexus at E11.5 (data not shown), and one day later in the LD and CM muscles in Raldh2^L–/– embryos(Fig. 5P). Together, these data suggest that RA activity in the periphery does not regulate the expression of Gdnf, and therefore that the decreased RA activity in the periphery is unlikely to be the primary cause of mispositioned Pea3+ cells within the Raldh2^L–/– LMC.

In an attempt to establish a link between the LMC molecular defects described above with the adult Raldh2^L–/–phenotype, we investigated the development of forelimb motor axonal projections. We first analyzed by double IHC at E11.5 the expression of neurofilament and Epha4, the latter being preferentially expressed in axons projecting dorsally in the wild-type limb bud (Helmbacher et al., 2000; Kania and Jessell, 2003). Neurofilament distribution revealed no obvious defect in motor axonal projections in mutants (Fig. 6A,B). By contrast, the expression of Epha4 was significantly diminished in the dorsal axonal projections in Raldh2^L–/– forelimb(Fig. 6C,D, thin arrows). Epha4 expression in the proximal-dorsal limb mesenchyme was unaffected in mutants(Fig. 6C,D, arrowheads),suggesting that the lack of this protein in dorsal axons is linked to the corresponding motor pool defect (Fig. 5E-H).

Brachial axonal projections were further analyzed at later stages using whole-mount anti-neurofilament IHC. At E12.5, the growing forelimb axonal projections, the arrangement of the developing brachial plexus and its spinal roots were not detectably altered in Raldh2^L–/– embryos (data not shown). Almost all the growing nerve tracts were normally patterned at E13.5 and 14.5 in mutants (Fig. 6E,F and data not shown). However, the distal projections of the ramus profundus of the radialis nerve supplying most of the extensor muscles of the forepaw were severely atrophied (Fig. 6H, arrowheads and inset). We conclude that the forepaw prehension deficiency of Raldh2^L–/– mice is likely to result from projection defects of the ramus profundus of the radialis nerve.

The forelimb defect observed in Raldh2^L–/–mice is reminiscent of the Hoxc8 knockout mouse phenotype(Le Mouellic et al., 1992; Tiret et al., 1998). We hypothesized that decreased local RA signaling could affect Hox gene expression in the brachial LMC. Hoxc6 and Hoxc8 expression were therefore analyzed at E12.5. While in controls Hoxc6 was broadly expressed in the Raldh2 domain (Fig. 7A), in Raldh2^L–/– spinal cord its expression was abnormally low, especially in the central LMC domain (brackets, Fig. 7A,B). In controls, Hoxc8 was expressed at the anterior level of the LMC in the Lim1+ interneuron column, while at the posterior level of the LMC it was additionally expressed in the Islet1+ and Lim1+ motor columns (Fig. 7C,E). In mutants, the Hoxc8 interneuron domain was comparable to that of controls,whereas the anterior boundary of Hoxc8 expression in the motoneurons was shifted posteriorly at the transcript and protein levels(Fig. 7C-L). Hoxc8 downregulation was most prominent in the Lim1+ motor column, as shown in dissected spinal cords co-labeled with Lim1 and on serial transverse sections (red arrows, Fig. 7E-L). Taken together, these data show that decreased RA signaling affects Hox gene expression within LMC neurons, which may in turn play a role in the molecular alterations described in the Raldh2^L–/– LMC.

To verify this hypothesis, we performed a molecular analysis of the brachial spinal cord in Hoxc8 mutant mice(Le Mouellic et al., 1992; Tiret et al., 1998). Raldh2 expression in the LMC was comparable to that of controls in Hoxc8^+/– mice, whereas its expression domain was thinner in the posterior LMC in Hoxc8^–/–mutants at E12.5 (Fig. 8A-C). This phenotype was partly due to a loss of Raldh2+ cells in the Islet1+ domain (Fig. 8J-L). We next examined the expression of Rarb, a known direct RA-target gene (de Thé et al., 1990; Sucov et al.,1990). In the Hoxc8^+/– LMC, Rarb was selectively downregulated in the Islet1+ posterior medial domain (Fig. 8D,E, brackets),whereas its expression was abolished in the Hoxc8^–/– brachial LMC(Fig. 8F). These data clearly indicate that RA activity is altered in Hoxc8^+/– and Hoxc8^–/– spinal cords.

In Hoxc8^+/– and Hoxc8^–/– mutants, expression of Lim1was downregulated in the posterior LMC(Fig. 8G-I) and Islet1+ cells were abnormally distributed(Fig. 8J-L). In Hoxc8^+/– mice, Islet1 was continuously distributed in two thick longitudinal columns throughout the LMC(Fig. 8K, arrowheads). In Hoxc8^–/– mutants, Islet1+ cells spanned a uniform domain throughout most of the LMC(Fig. 8L). The analysis of Islet1 expression on serial transverse sections revealed ectopic dorsal Islet1+ cells at the level of the posterior LMC in Hoxc8^+/– mice, and this abnormality was more pronounced in Hoxc8^–/– embryos(Fig. 8K,L). Analysis of Pea3 expression showed a mislocation of these cells in Hoxc8^+/– mutants: the distance between the ventral midline and the Pea3+ cells was larger in mutants(Fig. 8N, horizontal bars). Pea3 expression was greatly diminished in Hoxc8^–/– mutants, and the ventral migration of these cells did not occur properly (Fig. 8O). In both Hoxc8^+/– and Hoxc8^–/– mutants, the distribution and levels of Gdnf transcripts in the brachial plexus and the CM muscle was comparable to those of wild-type embryos(Fig. 8P-R).

Taken together, the analysis of the Hoxc8 mutant spinal cords revealed striking similarities with the abnormalities found in Raldh2^L–/– mutant mice, including a decrease in RA activity monitored by the expression of Rarb, a downregulation of Lim1 expression within the motor columns, a redistribution of Islet1+ cells and a mispositioning of Pea3+ cells. It was found, however, that Hoxc8^+/– heterozygous mutants most closely phenocopied the Raldh2^L–/– spinal cord phenotype, whereas Hoxc8^–/– homozygous mice clearly exhibited more severe cellular defects.

This study provides evidence that Raldh2 and Hoxc8 functions within the brachial spinal cord are required for the expression of a subset of Lim1+motoneurons, the correct positioning of Islet1+ motor cell bodies and their appropriate axonal projections. We discuss these findings in the context of the role of RARβ as a pivotal molecule for the integration of signals during motoneuron specification.

Our findings show that the regulation of RA signaling in the LMC cells depends at least on the activity of two molecules: Raldh2 and RARβ. It is noteworthy that the skeletal analysis of the forepaw of compound RARβ/RARγ mutant mice revealed a retraction of the digits that resembles the Raldh2^L–/– phenotype(Ghyselinck et al., 1997). Unlike Raldh2, which is uniformly expressed in the LMC cells, Rarbexpression is dynamic, suggesting that this receptor plays a key role in the spatiotemporal regulation of RA activity during the specification of motoneurons. At E11.5, Rarb expression in the spinal cord was similar to that of the RARE_hsp68_lacZ reporter transgene, whereas one day later it became restricted to specific motor pools(Fig. 3, and data not shown). These expression patterns most probably reflect the sequential roles of RARβ-mediated RA signaling during motoneuron differentiation.

The analysis of Raldh2^L–/– mutant mice did not reveal any re-specification of the LMC cells toward another cell fate; for example, no re-specification of the LMC cells toward thoracic motoneuron fates such as the MMC cells or preganglionic motoneurons (PrGG) was observed(I.L.R., unpublished). As RA signaling provided by the paraxial mesoderm has been shown to specify a brachial versus thoracic motor column fate(Sockanathan et al., 2003),our data are in agreement with the fact that RarbCre-mediated excision of Raldh2^L2 allele mostly occurs within the spinal cord and does not significantly deplete Raldh2 function in paraxial mesoderm during the phase of brachial versus thoracic fate decision.

We observed in Raldh2^L–/– mutant mice a decrease of 10% in the number of motoneurons at E11.5(Fig. 4). This decrease is cell type specific, as it only affects the Lim1+ population. A decrease in the proliferation of motoneuron progenitors or a premature cell death of postmitotic Lim1+ cells could explain this phenomenon. We favor the second hypothesis, because no reduction in RA-responsive cells was observed in ventral neural progenitors at E9.5 and 10.5 in the Raldh2^L–/– mutants(Fig. 3B). Our results thus confirm previous findings that demonstrated the role of RA signaling in the specification of chick lateral LMC neuronal identity(Sockanathan and Jessell,1998).

The whole range of later developmental (from E12.5) and postnatal defects observed in Raldh2^L–/– mice cannot, however,be explained by the early death of a subset of Lim1+ motoneurons, or by an increased rate of the normally occurring programmed cell death, as TUNEL experiments performed at E13.5 did not show significant differences between controls and mutants (I.L.R., unpublished). The abnormal axonal arborization of a branch of the radialis nerve and the decreased Epha4 protein expression in the growing dorsal limb axons indicate sustained functional defects in specific motor pools (Fig. 6). One possibility is that RA deficiency leads to the acquisition of an incomplete motoneuron identity and to the presence of motoneurons that neither expressed Islet1 nor Lim1 and, therefore, cannot properly express differentiation markers.

We found that Pea3+ motoneurons were mispositioned in the Raldh2^L–/– posterior LMC. Although the expression of Pea3 in motoneurons and the migration of these cells are both regulated by Gdnf, this trophic factor is unlikely to be responsible for the migratory defect observed in the Raldh2^L–/– mutants(Fig. 5I-L). Indeed, we did not detect any change in the distribution and level of Gdnf transcripts in mutants. This result contrasts with the marked decrease in RA activity at the level of the brachial plexus and the hypaxial LD and CM muscles, i.e. at the same sites as Gdnf expression, in Raldh2^L–/– mutants(Fig. 5). Possible explanations are that RA activity does not regulate Gdnf in the tissues, or that the remaining RA activity present in mutants is sufficient to allow normal expression levels of this factor. In any event, these data strengthen the idea that most of the molecular defects described in Raldh2^L–/– mutants are linked to an RA deficiency within the LMC.

The mispositioning of Pea3+ cells, which are constituted by over 95% of Islet1+ cells at segmental levels C7/8(Livet et al., 2002), is very probably a consequence of the redistribution of Islet1+ cells within the Raldh2^L–/– LMC, as observed at E12.5(Fig. 4). The early loss or the incomplete specification of some Lim1+ cells and the change in the expression patterns of Epha4, cadherins 8 and 7(Fig. 5), which encode for adhesion molecules regulating motor pool segregation(Price et al., 2002; Coonan et al., 2003), may be responsible for the mispositioning of Islet1+ cells. It is noteworthy that the positioning of motor cell bodies can influence peripheral axonal projections, as it has been demonstrated by single cell transplantation experiments in zebrafish embryos or by retrograde labeling in Hoxc8mutant mice (Eisen, 1991; Tiret et al., 1998). The alteration of Islet1+ cell distribution in Raldh2^L–/– mutants could, therefore, lead to subtle aberrant axonal projections.

We found that the loss of Raldh2 function in the lateral LMC was dispensable for the specification of a large subset of lateral LMC cells, as shown by the relatively subtle alterations of Lim1 expression in the developing spinal cord and the subsequent correct innervation of most dorsal limb muscles in Raldh2^L–/– mutants. Although the present conditional knockout does not completely rule out a contribution of RA produced by paraxial mesoderm or meningeal cells in the process of lateral LMC specification, our data show that RA was not the only signal to specify this cell fate. Strikingly, the molecular features of the Raldh2^L–/– mutants, such as the loss of Lim1+ motoneurons, the redistribution of Islet1+ cells or the downregulation of Rarb in the posterior medial LMC, were shared with Hoxc8 mutants (Fig. 8). We thus demonstrated that Hoxc8 is a key molecule, together with RA signaling, in directing a lateral LMC fate. This finding, along with work on the function of Hoxc6 and Hoxc9 in the specification of a brachial or thoracic motoneuron fate (Dasen et al.,2003), unveils novel roles for Hox proteins within postmitotic differentiating neurons.

The closest phenocopy of the Raldh2^L–/–phenotype was found in Hoxc8^+/– heterozygous mutants, indicating that a decrease in the level of Hoxc8 protein is sufficient to modify the specification of a lateral LMC cell type. This finding underlines the importance of the regulation of Hoxc8 gene expression in postmitotic cells of the ventral spinal cord. At the protein level, Hoxc8 activity is regulated by the presence of co-factors such as the Meis proteins (Popperl et al.,1995). Interestingly, we found that Meis2 was expressed in the Lim1+ posterior LMC, and was excluded from the Islet1+posterior domain (J.V. and I.L.R., unpublished), suggesting that Hoxc8 and Meis2 co-expression spatially restricts Hoxc8 function to the lateral LMC. Hoxc8^–/– mutants clearly display more drastic brachial spinal cord abnormalities than Raldh2^L–/– mutants. This may reflect an absolute requirement for Hoxc8 in the regulation of some target genes and/or an early requirement for Hoxc8 in neural progenitors or paraxial mesoderm,which would set up the anterior boundary of the LMC.

In conclusion, our results provide new insights into the specification of LMC cell fates. Raldh2 expressed in the LMC provides local RA activity that is transduced by RARβ and activates the expression of the lateral LMC marker Lim1 and allows a correct positioning of cell bodies within the LMC. A recent study demonstrated a bi-directional regulation between Hox genes(Hoxb4 and Hoxd4) and RARβ during the formation of rhombomere boundaries, and showed that the regulatory sequences of each gene contain both active RAREs and Hox consensus binding sites(Serpente et al., 2005). Our results suggest that a similar bi-directional regulation between Hoxc8 and RARβ occurs during the specification and early differentiation of the LMC cells.

We thank V. Fraulob and D. Lagorce for technical help and S. Falcone for animal care. We are indebted to J. Rossant and M. Cappecchi for providing the RARE-hsplacZ and RARβCre transgenic mice, respectively. We thank C. Kress for providing the embryonic stem cells (Hoxc8) and M. Petkovich, R. Behringer, C-C. Hui, S. Pfaff, C. Henderson, D. Wilkinson, R. Klein, M. Capecchi, X. Desbiens, M. Kmita, A. Esquela Kerscher, T. Jessell and P. Charnay for the gift of antibodies and/or plasmids. We gratefully acknowledge F. Plewniak for help with statistics, N. Ghyselinck, M. Teletin and the Dollé team for helpful discussions. This work was supported by funds from the CNRS, the INSERM, the Collège de France, the Ministère de la Recherche, the Institut Universitaire de France and the Institut Pasteur, and by fellowships from the Association pour la Recherche sur le Cancer (J.V.) and the Fondation pour la Recherche Médicale(J.V., I.L.R.).