Mice deficient for the homeotic gene Hoxc-8 suffer from a congenital prehension deficiency of the forepaw. During embryogenesis, Hoxc-8 is highly expressed in motoneurons within spinal cord segments C7 to T1. These motoneurons innervate forelimb distal muscles that move the forepaw. In Hoxc-8 mutant embryos, formation of these muscles is normal, but their innervation is perturbed. From E13.5 onwards, distal muscles normally supplied by C7-8 MNs also receive ectopic projections from C5-6 and T1 motoneurons. Coordinates of motor pools are altered along the rostrocaudal and also the mediolateral axes. Following this aberrant connectivity pattern and during the time of naturally occuring cell death, apoptosis is specifically enhanced in C7-T1 motoneurons. Loss of Hox-encoded regional specifications subsequently leads to a numerical deficit of motoneurons and an irreversible disorganization of motor pools. In Hoxc-8 null mutants, C7-8 motoneurons lose their selective advantage in growth cone pathfinding behavior and/or target recognition, two essential steps in the establishment and maintenance of a functional nervous system.
A functional nervous system relies on precise neuronal circuits between neurons and neuronal or muscular targets. Although ongoing modifications occur during adulthood, the neuronal connectivity pattern during development follows a highly invariant program, whereby neurons from a particular area branch to specific targets. In higher vertebrates, the central nervous system is topographically organized. Striated muscles are innervated by two populations of somatic motor neurons (MNs) from the ventral spinal cord. One contains MNs projecting to axial and body wall muscles and forms the medial motor column (MMC), which extends along the entire rostrocaudal (R-C) axis. The second MN population innervates forelimb and hindlimb muscles and forms the lateral motor column (LMC), exclusively in the brachial and lumbar segments. In particular, MNs that innervate forelimb muscles are contained in the spinal region extending from the 5th cervical to the 1st thoracic neural segment (C5-T1) in rodents (Burke et al., 1995). Topographically, the LMC is further subdivided into a lateral (LMCl) and a medial (LMCm) group, containing MNs that innervate dorsal (extensor) and ventral (flexor) muscles, respectively (reviewed in Lumsden, 1995). Individual muscles are innervated by a cluster of MNs, called a motor pool, spanning a R-C length of two segments (Landmesser, 1978a,b; Hollyday, 1980; Curfs et al., 1993 and refs. therein). There is therefore a fine correlation between the spatial coordinates of a MN within the spinal cord and the position of its peripheral target along the proximodistal and dorsoventral axes of the limb.
It is now well established that the specific connection between a MN and its target depends, in part, on the ability of the MN growth cone to differentially recognize specific guidance cues (reviewed in Tessier-Lavigne and Goodman, 1996). Numerous studies, essentially made in the chick, have shown that cells arising from the neurectoderm follow sequential steps of determination (reviewed in Tanabe and Jessell, 1996). Decreasing thresholds of Sonic Hedgehog concentration direct the differentiation of distinct neuronal progenitors into MNs or interneurons (Ericson et al., 1996, 1997). As early as stage 12 in the chick embryo, MNs are further specified to innervate trunk versus limb muscles (Tanaka et al., 1997). Before stage 15, the specificity of a MN to specifically project to an individual limb muscle target is not fixed and can be modified by manipulating its position along the R-C axis (Lance-Jones and Landmesser, 1980b; Matise and Lance-Jones, 1996). All MNs first express the LIM-homeobox gene Islet-1 (Isl1) (Ericson et al., 1992). Four additional genes of this family, Isl2, Lim3, Lim1 and Gsh-4 are sequentially activated in subgroups of MNs which are organized into distinct longitudinal columns (Li et al., 1994; Tsuchida et al., 1994). Growth cone pathfinding has been correlated with a unique combination of these transcription factors in the chick and the zebrafish embryo (Li et al., 1994; Tsuchida et al., 1994; Appel et al., 1995; Tokumoto et al., 1995). However, within each column, MNs from different R-C levels express the same combination of LIM genes, and the onset of expression of these genes starts when MN R-C specifications are already fixed (Matise and Lance-Jones, 1996). LIM genes are therefore not sufficient to define the sharp regionalization of motor pools.
In vertebrates, other homeobox-containing genes, the Hox genes, have been shown to specify R-C positional information (reviewed in McGinnis and Krumlauf, 1992; Krumlauf, 1994; Burke et al., 1995). 39 Hox genes are grouped into four clusters on different chromosomes. Within each cluster, the expression of individual homeotic genes follows a positional and temporal colinearity, resulting in unique combinations of Hox transcripts in cells at distinct R-C positions. Hox gene expression is detected in the neurectoderm before neural tube closure and later in the ependymal layer, which contains neural precursors. In chick spinal cord reversal experiments, it has been shown that during the period of plasticity (stages 13-14), reprogramming of axonal trajectories is associated with modulation of Hox gene expression in the neural tube (Lance-Jones and Sharma, 1996). The paraxial mesoderm mediates a dual influence on the neural tube, by altering both the Hox code (Itasaki et al., 1996) and the direction of a subset of axonal projections (Yip, 1996). Furthermore, ectopic induction of Hoxc-6 in the mesoderm of the chick embryo results in axonal projection abnormalities (Burke and Tabin, 1996). In the mouse embryo, defects in axonal pathfinding for some MNs in the segmented hindbrain have been reported after manipulation of Hox gene expression (Carpenter et al., 1993; Zhang et al., 1994; Studer et al., 1996), or by retinoic acid induced modulation of Hox gene expression (Marshall et al., 1992; Kessel, 1993). Taken together, these results suggest that Hox genes are strong candidates for the R-C specification of MN identity.
A loss-of-function mutation for the homeotic gene Hoxc-8 has been generated in the mouse by substituting part of the coding region with the lacZ reporter gene in embryonic stem cells. It is therefore possible to follow cells that normally expressed the endogenous Hoxc-8 gene by monitoring β-galactosidase activity (Le Mouellic et al., 1992). Based upon anatomical criteria, anterior homeotic transformations of axial skeletal segments occur throughout the Hoxc-8 expression domain in paraxial mesoderm and the frequency of transformation correlates with both the level of expression of Hoxc-8 in cells and the density of Hoxc-8-expressing cells in each segment (Le Mouellic et al., 1992; Tiret et al., 1993). In addition, a prehension deficiency phenotype in the Hoxc-8-deficient mice was attributed to a nervous defect (Le Mouellic et al., 1992). In wild-type animals, most of the motor pools supplying forepaw muscles are contained in spinal cord segments C7, C8 and T1, corresponding to the domains of highest Hoxc-8 expression in MNs. Here, we show alteration in the topographic maps of motor pools innervating forepaw muscles in mutants, as a consequence of an early modification of axon pathfinding for some MNs. Elevated apoptosis in MNs of this R-C region is also observed. The severe phenotype resulting from the loss of Hoxc-8 demonstrates that Hox genes are crucial in the establishment of spinal cord patterning.
MATERIALS AND METHODS
Mouse breeding and determination of genotypes
Embryos were obtained by crossing Hoxc-8 heterozygotes (+/−) on 129/Sv or (C57Bl/6×DBA/2) F1 genetic backgrounds. The genotype was established for each animal by Southern blotting of DNA from yolk sacs or tail biopsies as described (Le Mouellic et al., 1992), or by PCR. PCR primers 31AS (5’-CGTAGCCATAGAATTGGAG) and GED31 (5’-GAGCTCCTACTTCGTCAAC), 31AS and NEO (5’-CAGCAGAAACATACAAGCTG) were used to identify the endogenous and targeted loci, respectively. Amplification conditions were: denaturation at 94°C for 30 seconds, annealing at 58°C for 20 seconds and extension at 72°C for 30 seconds for 30 cycles. Taq polymerase and buffers were obtained from GIBCO BRL or Pharmacia. The genotype of myf-5+/lacZ animals was determined by X-Gal staining (Tajbakhsh and Buckingham, 1994).
Fixation of embryos and postnatal animals
Embryos were dissected at RT in PBS and fixed in 4% PFA for a period that depended on the experiment. Newborn and adult animals were deeply anesthesized using 0.02 ml of a 3% avertin solution per gram of body weight, transcardially perfused with 20 ml of ice-cold PBS, followed by 50 ml of a 4% PFA fixative solution.
Dissection of muscles; sectioning of tendons
Forelimbs from transcardiacally fixed animals were dissected and the skin and subcutaneous fat removed. Forelimbs from myf-5+/lacZ mice were stained in X-Gal for several hours. Muscles were then identified on the basis of their position, insertion and function. Each muscle was then dissected along its entire length, from the tendon to the proximal origin, and postfixed overnight in 4% PFA.
6-day-old mice were anaesthesized and an incision was made in the dorsolateral region of the skin covering the distal forelimb. Tendons of the extensor carpi radialis longus and brevis of the left side were identified by their position and shape, and carefully sectioned. Animals were then returned to their cage and observed every day during the first week. They were photographed 3 months later.
Staining for β-galactosidase activity
X-Gal staining on whole-mount embryos was performed as previously described (Le Mouellic et al., 1992). X-Gal staining of myf-5+/lacZ embryos was processed for 2 hours with no detergents in the staining solution. For X-Gal staining on transverse sections, embryos were fixed in 4% PFA for 20-30 minutes depending on the embryonic stage, washed in ice-cold PBS for 5 minutes, then immersed in a cryoprotective buffer, consisting of 30% sucrose in PBS for few hours, embedded in OCT medium (Miles Laboratory) and cut into 10 µm sections on a cryotome. Sections were stained in X-Gal buffer for 12 hours at 35°C, then mounted in 1:1 PBS/Glycerol and examined under a Reichert-Jung microscope.
Immunohistochemistry, codetection with β-galactosidase activity
For all these experiments, embryos were fixed for 25 minutes, rinsed in PBS and immersed in PBS containing 30% sucrose at 4°C overnight. They were then cut into 10 µm sections with a cryostat. Sections were allowed to dry and stored at −80°C. Sections to be analyzed were thawed, washed in PBS, immersed 1 hour in a 1× blocking solution containing 2% blocking reagent (Boehringer Mannheim #1175041), 10% inactivated sheep serum (Sigma #G9023) and 1% Triton X-100 (Sigma #P1379). They were then incubated overnight at 4°C with the primary antibody diluted in 0.1× blocking solution. The following antibodies were used: K5 polyclonal antibody to ISL1/2 (see Tsuchida et al., 1994), 1:5000; monoclonal mouse antibody to Hoxc-8 (BAbCO #MMS-266R), 1:100. Sections were then rinsed in PBS, incubated in PBS containing the secondary antibodies diluted at 1:100 (TRITC-conjugated antibody to rabbit and FITC-conjugated antibody to mouse, Jackson ImmunoResearch Laboratories, Inc. #111-025-003 and #55514). Sections were then washed in PBS and mounted with a drop of Immunomount (Shandon #9990402). Confocal microscopy of samples labelled with fluorophores was performed with a Confocal Laser Scanning microscope (Leica Instruments, Heidelberg, Germany), which uses an argon-krypton laser operating in multi-line mode. When visualization of β-galactosidase activity was desired, sections were incubated at 35°C for 3 hours in X-Gal buffer after the primary antibody step, rinsed in PBS and further processed for immunodetection. Endogenous background was reduced by using an avidin/biotin blocking kit for biotin (Vector #SP-2001). A secondary biotin-conjugated anti-rabbit antibody (Tebu #L43015) was then applied for 1 hour at RT at a dilution of 1:1000. Sections were rinsed with PBS. Endogenous peroxidase activity was blocked by a short 5 minutes incubation in methanol containing 0.3% H2O2, followed by a 1 hour incubation of sections in ExtAvidin-Peroxidase (Sigma #E2886). Staining was performed using the AEC detection kit (Vector #SK-4200).
Labeling of apoptotic cells
Frozen sections of embryos were thawed, fixed in 4% PFA for 10 minutes and washed in PBS. Following incubation with the primary antibody (K5, see above), sections were processed according to the manufacturer’s instructions (In Situ Cell Death Detection kit; Boehringer Mannheim #1684795) for in situ detection of DNA fragmentation. The TRITC-conjugated secondary antibody was then used at a dilution of 1:100. Stained sections were examined under a microscope equipped with an epifluorescent unit containing the appropriate filter for FITC and TRITC.
Histology and MN counts
E14.5 and E16.5 embryos were fixed for 6 hours in Bouin fixative and rinsed in 70% alcohol. A bilateral laminectomy in transcardially fixed newborn mice was performed to dissect C4-T2 segments of the spinal cord. Dorsal and ventral nerve roots were preserved and used as indicators of the R-C level of each section. Overnight postfixation in 4% PFA was followed by washes in PBS. The following steps were identical for newborns and embryos. Dehydratation was followed by paraffin embedding. Embryos were cut into sections of 7 µm. Sections were rehydrated and stained with Hemalun-Eosin-Safran (HES). Spinal cords of newborns were serially cut at 12 µm, mounted on slides and stained with a 0.25% Thionin solution (Sigma#T3387) for 5 minutes, followed by a 30 second differentiation step in 70% alcohol, which allowed for the selective visualization of Nissl bodies. MNs within the left and right lateral motor columns from C7-T1 were counted independently on every other section to eliminate the risk of counting the same MN on two adjacent sections. Nissl-stained cells with a large nucleus, a nucleolus and a cellular size of >20 µm were considered as MNs. Data analyses were carried out using the General Linear Procedure (Analyses of Variance) in the SAS software (SAS Institute Inc.1989). Differences in the numbers of MNs were considered significant when P < 0.05. Numerical results of Figure 5H were expressed as mean +/− s.e.m.
Retrograde and anterograde labeling experiments
For DiI labeling, a solution of 5% 1,1’-dioctadecyl-3,3,3’,3’-tetramethylindocarbocyanine perchlorate (DiIC18, Molecular Probes # 282) in dimethylformamide was prepared. After removing the skin of the forelimbs, E13.0-E13.5 myf-5+/lacZ embryos were fixed for 25 minutes in 4% PFA, stained in X-Gal and postfixed 2 hours in 4% PFA. For anterograde labeling, neural segments to be labelled were first identified, the adjacent neural tube regions cut and removed. In the remaining segment of the neural tube, central branches of dorsal roots were cut. A crystal of DiI was precisely introduced either into the middle part of the stained muscle mass (retrograde labeling), or into the spinal cord (anterograde labeling). Embryos were then incubated at 37°C in PBS containing 1 mM NaN3. Axonal diffusion was checked every 2 days under a microscope equipped with an epifluorescence unit. Time of diffusion were respectively 6 and 10-15 days for anterograde and retrograde labeling experiments. For live newborn mice, 0.3 µl of DiI solution was injected into the muscle. After a diffusion time of 3 days, animals were anaesthesized, transcardially perfused with ice-cold 4% PFA for 15 minutes and their spinal cords dissected. Tissue was then embedded and cut on a freezing microtome into 60 µm sections, which were finally mounted onto slides. Sections obtained from embryos were prepared in the same way but were 10 µm thick. All sections were examined under a microscope equipped with an epifluorescence unit and the appropriate filter for TRITC was used. The CTB-HRP or CTB-FITC experiments and drawings were done as previously described (Curfs et al., 1993; Dederen et al., 1994). After CTB-FITC visualization, some sections were processed for immunodetection of Hoxc-8 as described above, except that the secondary antibody was a Texas-red-conjugated antibody (Jackson ImmunoResearch Laboratories, Inc. #715-095-146).
Congenital forelimb phenotype in Hoxc-8−/− mice
In all the experiments described in this paper, heterozygous animals (+/−) behaved like wild-type mice (+/+), (+/+) and (+/−) animals were therefore assimilated and used as controls. 90% of the homozygous mutants from the inbred strain 129/Sv and 35% of those crossed into a C57Bl/6×DBA/2 background strain died within 2 days after birth (Le Mouellic et al., 1992 and L.T unpublished results). Newborns that survived shared a reproducible phenotype: they lay on the dorsal side of the forepaw and were unable to extend their fingers (Fig. 1A). After the first two weeks, the forepaw progressively became hyperextended (Fig. 1B), the paw function did not recover. A slight abduction of the wrist was also noticed. Adult deficient mice (−/−) were unable to cling to a support, showing that the congenital defect leads to a complete loss of prehension ability. As movements of the wrist and fingers depend on the contraction of distal forelimb muscles, we first evaluated the integrity of these muscles by anatomical analysis. In young mice, observation and dissection of muscles was facilitated by the use of mice carrying a myf-5/lacZ fusion transgene in the endogenous myf-5 locus (Tajbakhsh and Buckingham, 1994). Myf-5 is a myogenic factor expressed by E9, and Hoxc-8 and myf-5 genes are not located on the same mouse chromosome. Hoxc-8+/+,+/− and −/− embryos in a myf-5+/lacZ background were obtained by crossing the two targeted mutant strains. After revealing β-galactosidase activity in these embryos, it was possible to check the size, structural morphology, origin and insertion of each muscle from the ventral and the dorsal group. No macroscopic alteration could be detected between E17.5 and the first days of life (Fig. 1C-F). Observation of dissected adult mutant forelimbs suggested that the hyperextension and abduction of the wrist was a mechanical consequence of the excessive tension of the extensor carpi radialis (longus and brevis) tendons (Fig. 1G,H). Section of these tendons in a 6-day-old mutant mice abolished the hyperextension and therefore confirmed that it was a direct consequence of the extensor carpi radialis function. In a wild-type animal, the section of the same tendons had no repercussion on the paw posture, presumably because other efficient extensor muscles were able to compensate (Fig. 1M). The dangling forepaw observed in the mutant (Fig. 1N) suggested that no other distal muscles were functional. In the mutant, all distal muscles appeared reduced in length (e.g. in Fig. 1I-L). The phenotype resulting from the inactivation of Hoxc-8 does not therefore appear to arise from defects in the differentiation and patterning of forelimb distal muscles, but from the absence of their function. Fine anatomical and biochemical characterization of muscular alterations will be described elsewhere.
The phenotype correlates with the expression of Hoxc-8 in MNs of the cervical motor columns
The mutational strategy chosen resulted in the replacement of the Hoxc-8 gene by the E. coli lacZ reporter gene. At E9.0, Hoxc-8 was expressed in the ependymal layer, which contains neural cell precursors (Fig. 2B). From E9.0 onwards, the rostral limits of Hoxc-8/lacZ expression in neuroectodermal and mesodermal cells are already fixed, at the level of the 8th and 13th pairs of somites respectively (Fig. 2A,C and Le Mouellic et al., 1992). The sclerotome from the 8th pair of somites participates in the formation in the 3rd and the 4th cervical vertebrae (Burke et al., 1995). Observation of lacZ-stained sagittal sections of E12.5 (Fig. 2D) and E13.5 embryos revealed that, between the third and the fifth cervical vertebrae, a small number of cells expressed β-gal at a low level in the intermediate zone between the ventral and dorsal part of the neural tube. Posterior to the fifth cervical vertebra, Hoxc-8 expression increased in the intermediolateral and the ventral part of the neural tube and, in particular, was most intense caudally to C7 (Fig. 2D,E; Utset et al., 1987; Breier et al., 1988; Awgulewitsch and Jacobs, 1990; Le Mouellic et al., 1992). In mutant E13.5 embryos, cells that should have expressed Hoxc-8 were still detected, and their gross pattern of distribution along the R-C (data not shown) and the mediolateral axes (Fig. 2G,H) was not significantly modified. C7, C8 and T1 contain motor pools supplying forelimb distal muscles involved in the prehension phenotype. As Hoxc-8 is expressed in motor columns (Fig. 2F), we sought to determine its precise expression in MNs during neurogenesis. Co-detection experiments for the Hoxc-8 protein and ISL1/2 markers were performed on longitudinal and transverse sections of E13.5 and E14.5 embryos, at which stages vertebral morphology allows precise identification of neural segments. In the most rostral segments from C3 to C6, ISL1/2-positive cells were negative for Hoxc-8 (Fig. 3A,E), Hoxc-8 being expressed exclusively in interneurons and/or glial cells. Caudal to C6, Hoxc-8 was also expressed in MNs. The rostral part of C7 was a transition region containing a mosaic population of MNs which did or did not express Hoxc-8 (Fig. 3B,F). In the medial and caudal C7, in C8 and T1, 100% of somatic MNs from the LMC or the MMC expressed Hoxc-8 (Fig. 3C,D,G,H). In null mutant embryos, the Hoxc-8/lacZ gene was expressed in MNs in a similar pattern (Fig. 3I-L), suggesting that the inactivation of Hoxc-8 in precursors does not interfere with MN differentiation or migration. When similar C8 and T1 transverse sections from E13.5 heterozygous Hoxc-8+/lacZ embryos were processed for X-Gal staining, not all MNs were blue (e.g. Fig. 3O,P). A discrepancy between the visualization of the endogenous protein by immunohistochemistry and the Hoxc-8/β-galactosidase histochemistry patterns was also noticed in heterozygous adults, where MNs were faintly immunostained and not labelled with X-Gal (data not shown). Confocal analysis revealed that the chromatic gradient from orange to yellow observed on sections (Fig. 3C,D,G,H) resulted from different levels of emission light from the Hoxc-8 signal (data not shown). These data suggest that subpopulations of MNs express variable levels of Hoxc-8.
Motoneuron deficits following enhanced apoptosis in the brachial spinal cord of Hoxc-8 mutants
At E12.5, no difference were detectable between (+/−) and (−/−) MNs. Their number and position were similar on transverse sections (Fig. 4A-D). During embryogenesis, neurons pass through a regressive step of cell death (Oppenheim, 1991), which in the nervous system invokes different degeneration mechanisms such as apoptosis or necrosis (Clarke, 1990). Frozen sections of E13.5 embryos stained with hematoxylin revealed masses of dense chromatin in cells within the motor columns (Fig. 4E,F). Since pycnotic nuclei are a prominent feature of apoptosis, we looked for DNA fragmentation in situ, using the terminal deoxynucleotidyl transferase dUTP nick end labeling technique (TUNEL). No fragmentation was observed at E12.5 in either +/− or −/− genotypes (Fig. 4C,D). Cells positive for DNA fragmentation were visualized specifically in motor columns of E13.5 and E14.5 (+/−) and (−/−) embryos. In motor columns of C7-T1, the number of apoptotic cells detected by this technique was higher in (−/−) embryos versus (+/−) control embryos (Fig. 4G,H,K,L). Confocal scanning microscopy allowed the co-detection of fragmented DNA on ISL1/2-immunostained sections and thereby demonstrated DNA fragmentation within MNs (inset picture of Fig. 4H). As a result of the MN loss, the shape of the motor columns was modified in the region of the cervicothoracic transition, and lateral and medial MN group boundaries became difficult to distinguish (Fig. 4I,J).
Therefore, while the onset of cell death was not modified by the mutation, the degree of apoptosis in mutants was increased between E13.5 and E14.5 in C7, C8 and T1, resulting in a marked reduction of MNs by E15.5 (Fig. 4M,N). As early as E14.5, the shape of the ventral grey matter was modified specifically in segments C7-T1 (Fig. 5A-F’). In newborn Hoxc-8−/− animals, the reduced size of posterior cervical segments was easily observed on whole-mount spinal cords (data not shown), and confirmed on transverse sections (Fig. 5G,I). The relative number of MNs within the lateral motor column was compared for control and null mutant newborn mice (see Materials and Methods). The mutation resulted in a specific rostrocaudal MN deficit in C7-T1 (Fig. 5H), with the average percentage of missing MNs increasing from C7 to T1 (C7, 26%; C8, 34% and T1, 44%). Moreover, a significant variation in the number of MNs was detected among the three mutant animals examined (see legend), presumably reflecting variable severity of the mutant phenotype.
Disorganization of motor pools innervating forelimb distal muscles in Hoxc-8-deficient mice
The control of body movements relies on fine tuning between motor pools and muscle targets, we have therefore examined the (embryonic) establishment and the (postnatal) maturation of connectivity patterns between distal forelimb muscles and their supplying MNs. Retrograde labeling of MNs from target muscles was used to determine the spatial coordinates of motor pools. The quality of these experiments at early embryonic stages depended upon the size and site of deposit of the dye crystal (DiI, see Materials and Methods). For this reason, we used embryos carrying the lacZ gene in the myf-5 locus (Tajbakhsh and Buckingham, 1994). Visualization of X-Gal-stained muscles allowed the precise injection of individual muscles with a very small DiI crystal. Labelings were performed on E13.0-E13.5 fixed embryos, corresponding to the earliest stage at which individual muscles can be identified prior to the cell death period (e.g. flexor carpi ulnaris in Fig. 6A,B). Diffusion of the fluorochrome into the limb was excluded by verifying the absence of fluorescence within adjacent muscles (Fig. 6C). The dye diffused passively along the axons of motor and sensory neurons, and finally stained their cell bodies within the neural tube (Fig. 6D). Pictures shown in Fig. 6D,F,G,I-K and M-O are representative of observations on serial sections after implantation of DiI into six different distal muscles. In E13.0-E13.5 wild-type embryos, flexor carpi ulnaris (Fig. 6D), extensor digitorum communis (Fig. 6I), extensor carpi ulnaris (Fig. 6M), flexor carpi radialis and extensor digitorum lateralis (data not shown) all received projections from C7 and C8 MNs. By contrast, the extensor carpi radialis was innervated by a more rostral motor pool, recruiting MNs from C5-6 segments (Fig. 6F). To confirm that MNs from C5-6 did not innervate other distal muscles, we selectively stained MNs from C6. The anterograde diffusion of the dye specifically labelled the radial nerve, the most distal fluorescent axons defasciculating from this nerve were observed in the vicinity of the extensor carpi radialis (Fig. 6S). Staining of C6 and the two adjacent neural tube segments C7 and C8 resulted in the labeling of axons that exited the spinal cord by three distinct intervertebral spaces (upper part of Fig. 6P,Q). Labelled axons formed the radial, the medial and the ulnar nerves (Fig. 6T) that supplied all distal muscles. We subsequently detected Hoxc-8 in retrogradely labelled MNs at PN6 by immunohistochemistry. The motor pool supplying the flexor digitorum superficialis contained Hoxc-8-positive MNs (Fig. 6W,X), while Hoxc-8 was not detected in MNs supplying the extensor carpi radialis (Fig. 6U,V). These data are consistent with the expression of Hoxc-8 exclusively in MNs caudally to C7 (Fig. 3). Therefore, in wild-type animals, forelimb distal muscles are supplied from the onset by C7-8 Hoxc-8-positive MNs, a pattern previously described in the rat and many other vertebrate species (Baulac and Meininger, 1980; Scarisbrick et al., 1990; Curfs et al., 1993 and ref therein), with the noticeable exception of the extensor carpi radialis supplied by C5-6 MNs which do not express Hoxc-8.
The innervation patterns of the same muscles were similarly investigated in E13.5 Hoxc-8 null mutants. The R-C innervation pattern of the extensor carpi radialis was not modified (Fig. 6G and data not shown). By contrast, all other distal muscles that normally received projections from C7-8 Hoxc-8-expressing MNs, recruited additional axons emanating from C5-6 and T1 MNs, at E13.5 (Fig. 6J,K,N and O). These muscles are thus supplied by MNs from the entire brachial LMC. In some but not all mutants, a few MNs were also labelled in C4 or C3 (e.g. Fig. 6K,O; see also the right part of Fig. 7C), illustrating variable expressivity of the mutant phenotype, as previously described for morphological skeletal abnormalities (Le Mouellic et al., 1992; Tiret et al., 1993). The R-C position of sensory neurons seemed not to be altered in the mutants (compare Fig. 6F,I,M with G,J,N). The R-C coordinates of motor pools supplying the extensor carpi radialis, the flexor digitorum superficialis, the flexor digitorum profundus and the extensor digitorum communis remained unchanged in newborn (Fig. 7A,B,D) and adult (Fig. 7C) wild-type and mutant mice.
We also confirmed that discrete subdivisions exist in the LMC of wild-type adult animals. Conjugated-CTB was injected separately into a flexor or an extensor muscle, and MNs supplying these muscles were called FLEX-MNs and EXT-MNs, respectively (left part of Fig. 7C). EXT-MNs lie in a position adjacent to the white matter whereas FLEX-MNs lie more medially, as in the rat (Curfs et al., 1993). In the mutants, a subset of FLEX-MNs and EXT-MNs no longer respect their specific position, displaying a reverse mediolateral position with respect to their target muscles (right part of Fig. 7C). Thus, inactivation of Hoxc-8 induced an alteration of the motor somatotopic maps in the brachial region. Aberrant connectivity patterns were established as early as E13.5 and were not subjected to any correction throughout the lifespan.
Hoxc-8-deficient mice: a model of neuropathy
The most visible and fully penetrant phenotype of Hoxc-8-deficient mice is a congenital defect of the forepaw posture (Fig. 1A). Our results show that the initial improper innervation of forelimb distal muscles (Fig. 6) does not interfere with the differentiation, migration and fusion of myoblasts. Until the postnatal period, distal muscles from mutants do not exhibit size-reduction (Fig. 1C-F). The earliest observed defects are the alteration of somatotopic maps in the brachial region and the segment-specific enhanced apoptosis of MNs at E13.5 (Figs 4, 6). Putative mechanisms underlying MN death are discussed in following sections and presumably involve complex disturbance of early nerve-muscle interactions. Following the conventional division of neuromuscular diseases, we consider the Hoxc-8 mutant mouse to be an animal model of a primary neurogenic disease. This hypothesis, based on histological and chronological criteria, is supported by molecular analysis: during embryogenesis, Hoxc-8 expression is detected in MNs (Figs 2, 3) and not in forelimb distal muscles.
The innervation patterns of the extensor carpi radialis (longus and brevis), the only distal muscles not innervated by Hoxc-8-positive MNs (Fig. 6U,V), are not altered by the mutation (Fig. 6G). Their contraction and the absence of a functional antagonism by flexor muscles could explain the development of the phenotype resulting in the fully extended posture observed in the adults (Fig. 1B,G and H). In control animals, section of the tendons of the extensor carpi radialis is not phenotypically conspicuous (Fig. 1M), presumably because other extensor muscles (e.g. extensor carpi ulnaris) are able to compensate. In the mutant, the same operation results in a position of the forepaw that is reminiscent of the position observed at birth (Fig. 1N). Together, these data strongly suggest that none of the distal forelimb muscles is functional in Hoxc-8-deficient mice, with the exception of the extensor carpi radialis. Coordination of movement results from a precise connectivity pattern between different integrative levels of the central nervous system. Multiple phenomena, whether they be mutually exclusive or synergistic, could account for the motor dysfunction observed in mutant animals. Firstly, an impaired myotatic reflex of forepaw muscles may result from a topographical mismatch between the unmodified sensory projections and MNs from abnormal R-C locations. Furthermore, the Hoxc-8 mutation may alter identities of interneurons from the intermediate zone of the spinal cord, which could amplify the dysfunction of more elaborated spinal reflexes. Secondly, MNs that are located in C5-6 and T1 segments may not receive appropriate afferent inputs from descending tracts from the encephalon. Finally, the inversion of a subset of MNs that innervate flexor and extensor muscles could cause them to transmit contradictory commands. Determining the importance of these different mechanisms will necessitate sophisticated studies of neuronal circuits. Nonetheless, our experiments demonstrate that the establishment of topographic maps depends on positional information encoded by Hox genes.
Potentiality of neural progenitors is maintained in Hoxc-8 mutant mice
Hoxc-8 expression is primarily detected in the germinative ependymal layer of the neural tube at E9.0 (Fig. 2B), and continues from E9.5 onwards in the intermediate and ventral regions of the spinal cord caudally to C2 (Fig. 2D-F). Analysis at the cellular level has revealed that, whereas caudally to C7 all MNs express Hoxc-8, no MN from segments C2-6 does. C7 represents a transition between ‘non-expressing’ and ‘expressing’ segments (Fig. 3). Lineage experiments in the chick have demonstrated that multipotential progenitors within the ependymal layer give birth to neuronal and glial lineages but that among these progenitors only a ventral subpopulation (15%) produces MNs (Leber et al., 1990). The absence of Hoxc-8-positive MNs rostral to C7 shows that distinct subtypes of progenitors express Hoxc-8 along the R-C axis
The rostral limit of Hoxc-8/lacZ reporter gene expression in MNs is unaltered in Hoxc-8 homozygous embryos (Fig. 3I-P). In both control and mutant E12.5 embryos, identical numbers of cells are immunostained with the early MN marker ISL1/2 and these cells are correctly located within the ventral horn (Fig. 4A-D). At this embryonic stage, retrograde diffusion of DiI from the limb bud to the neuronal cell bodies indicates that some MNs have extended their axons, a characteristic of maturing MNs (data not shown). Therefore, although it is unclear as to whether the fine balance between discrete neural lineages is maintained in the absence of Hoxc-8, the potentiality of precursor cells along the R-C axis is not modified.
Combinatorial molecular determinants of MN identity
Each forelimb muscle receives its motor inputs from an individual motor pool of the brachial LMC. There is a correlation between the R-C coordinates of the pool and the proximodistal position of the muscle within the limb (Scarisbrick et al., 1990). The extensor carpi radialis has a more proximal insertion than other distal muscles (Fig. 7A) and is innervated by a more rostral motor pool (Fig. 7B). These specific connectivity patterns are properly established from the onset of MN development (Fig. 6), and presumably depend on R-C specifications conferred to MN precursors (Matise and Lance-Jones, 1996), before postmitotic MNs express Isl1 (Ericson et al., 1992; Tsuchida et al., 1994). Onset of Hox gene expression starts from the end of gastrulation in the neural plate (Fig. 2A). In the chick, determination of MN identity precedes the stabilization of R-C differences (Lance-Jones and Sharma, 1996). Our results show that MNs from C7-T1 differ from those from C5-6 by the expression of Hoxc-8 (Fig. 3), and that the inactivation of Hoxc-8 leads to a loss of the R-C demarcation of the pools (Figs 6, 7). These data and the expression of Hox genes in overlapping R-C domains are in favour of a major role played by this family of transcription factors in determining MN target matching. The different levels of expression of Hoxc-8 observed by immunostaining or β-galactosidase activity suggest that a Hox code involving the expression of numerous genes and/or gene dosage is implicated in the identity of MNs.
The LMC column is subdivided into LMCl and LMCm, which innervate extensor (dorsal) and flexor (ventral) muscles, respectively (Fig. 7C and reviewed by Lumsden, 1995). Expression of Hoxc-8 is visualized in all MNs caudally to C7 and is therefore not sufficient to differentiate Flex-from Ext-MNs. The discovery of the LIM-family homeobox genes has provided new insights into the putative molecular determinants involved in motor column organization. Distinct combinations of LIM genes differentially mark motor columns in chick embryos (Tsuchida et al., 1994) and different primary MNs of zebrafish embryos (Appel et al., 1995; Tokumoto et al., 1995), suggesting that these genes play a role in mediolateral MN specifications. Inactivation or misexpression of islet, the Drosophila homolog of Isl1 and Isl2, results in axonal pathfinding and targeting defects in some MNs of the ventral nerve cord (Thor and Thomas, 1997). In mice, inactivation of Lim1 and Isl1 have resulted in developmental defects leading to early embryonic lethality (Shawlot and Behringer, 1995; Pfaff et al., 1996) and, in Isl1 null mutant embryos, no MN differentiation occurs (Pfaff et al., 1996). Investigation of the involvement of either Isl1 or Lim1 in specifying MN identity in the mouse will require conditional inactivation of these genes to overcome these early phenotypes. Presumptive motor pools may therefore acquire their R-C and mediolateral coordinates by combining the expression of Hox and LIM genes, respectively. The loss of mediolateral coordinates by some MNs in Hoxc-8-deficient mice (Fig. 7C) raises the possibility that functional interactions occur between members of these gene families.
Loss of survival signals for brachial MNs in Hoxc-8-deficient mice
A major consequence of Hoxc-8 inactivation in the development of the nervous system is the excess apoptosis in C7-T1 MNs between E13.5 and E14.5 (Fig. 4). Remarkably, the enhanced apoptosis of MNs is segment-specific and correlates with the expression of Hoxc-8 in MNs (Fig. 3). Although a role of neighbouring neural or glial cells cannot be definitely ruled out, these data strongly suggest that the survival of MNs in C7-T1 depends on intrinsic Hoxc-8 expression in these MNs. An indirect link between Hox gene expression and cell death has been previously suggested by experiments in transgenic mice, in which ectopic expression of Hoxb-8 at the level of anterior cervical segments (C1) prevents the first dorsal root ganglia from degenerating (Charité et al., 1994; Fanarraga et al., 1997). Also, normal development of the facial motor nucleus requires Hoxb-1 expression in the fourth rhombomere (Goddard et al., 1996). In Hoxb-1−/− embryos, although facial branchiomotor neurons are absent or greatly reduced in number, Isl-1-positive ectopic MNs are transiently observed within the fourth rhombomere, suggesting that some MNs initially differentiated but have failed to survive (Studer et al., 1996). In wild-type and Hoxc-8 mutant animals, the number and position of MNs are manifestly identical at E12.5 (Fig. 4A,B), and no apoptotic MNs are observed in the mutants at this stage (Fig. 4C,D). In mutants, the MN cell death has therefore a normal time of onset, at E13.5, but is amplified in C7-T1, the domain of highest Hoxc-8 expression (Fig. 4G,H). The chronology of naturally occurring cell death (NOCD) has similarly been described in the lumbar spinal cord of mouse embryos, in which the majority of apoptotic MNs are detected between E13 and E15 (Lance-Jones, 1982). Since the nature of the Hox target genes is still poorly known, it is difficult to determine the mechanism by which Hoxc-8 interferes with an apoptotic program.
MN survival relies upon both the control of cell death-regulating genes and the intake of exogenous trophic molecules. Target-derived neurotrophic factors play an essential role in the regulation of the motor and sensory neuronal survival (reviewed by Oppenheim, 1991, 1996). A daily treatment with NT-3, the ligand of TrkC, of a chick embryo in which the limb bud has been removed, is sufficient in the absence of other peripheral signals to prevent TrkC-positive sensory neurons from dying. At a high dose, exogenous NT-3 also rescues neurons that would have normally died, suggesting that the number of surviving neurons depends on the amount of the trophic factor at the periphery (Oakley et al., 1997). When added individually to the medium of cultured MNs, the most potent survival molecules identified so far, CT-1, GDNF or HGF/SF, rescue only 46, 28 and 27% of MNs, respectively (Ebens et al., 1996; Pennica et al., 1996; Yamamoto et al., 1997). These data, and similar results obtained in vivo from LIF-, CNTF- and GDNF-receptor-deficient mice (DeChiara et al., 1995; Li et al., 1995; Moore et al., 1996; Sanchez et al., 1996), prompted the hypothesis that specific subpopulations of MNs are capable of responding to distinct survival factors emanating from the periphery. During embryogenesis, HGF/SF is expressed at E9.5 in the proximal limb buds and, by E10.5, in some mesenchymal and myogenic cells (Sonnenberg et al., 1993; Ebens et al., 1996; Yamamoto et al., 1997). In vitro, HGF/SF antibodies reduce by 65% the trophic activity on MNs of a myogenic cell line-conditioned medium (Yamamoto et al., 1997), a result that correlates with the spatiotemporal expression of the HGF/SF receptor. During the NOCD period, c-Met is specifically highly expressed in a subpopulation of MNs that innervate the limb (Sonnenberg et al., 1993; Ebens et al., 1996; Yamamoto et al., 1997). A possible mechanism by which MNs from a specific pool establish persistent synapses with an appropriate muscle could involve a perfect matching between unique combinations of trophic factors produced by the target and the receptors for these factors expressed by the MN itself.
In Hoxc-8 mutants, modification of the combinatorial expression of receptors in some C7-8 MNs may lead to their death by reducing their selective ability to take up trophic factors delivered by distal forelimb sources. In other words, the loss of positional information may result in a broader R-C population of MNs that are equally competitive in the growth cone pathfinding and/or the selection of persistent synapses. In C8 and T1, which contain MNs that all express Hoxc-8, absence of Hoxc-8 does not result in a complete loss of MNs, but in a partial deficit of 36 and 46%, respectively (Fig. 5H). Within a segment, the ability of MNs to respond to survival molecules does not depend on the expression of a single Hox gene, but is more likely to be under the genetic control of a combination of different Hox genes. Although the R-C expression domains of Hoxc-6 and Hoxc-9, the two neighbouring genes on the cluster, are not modified in Hoxc-8−/− embryos (Tiret et al., 1993), subtle changes of the Hox code within a single MN cannot be excluded.
Hox genes and the specification of target matching
Using retrograde labeling experiments, we show that muscles in the mouse embryo receive projections from appropriate segments as early as E13.5. In particular, distal forelimb muscles are innervated by C7-8 MNs, with the noticeable exception of the extensor carpi radialis supplied by C5-6 MNs (Figs 6E,F,H,I,L,M, 7A,B). MNs that innervate muscles derived from the dorsal muscle mass (extensor muscles) are located in the ventrolateral aspect of the LMC, while those MNs that innervate muscles derived from the ventral muscle mass (flexor muscles) lie more medially within the LMC (left part of Fig. 7C). Consistent with Hoxc-8 expression in embryonic MNs (Fig. 3), the extensor carpi radialis is innervated in postnatal mice by Hoxc-8-negative MNs (Fig. 6U,V) and has an unmodified R-C pattern of innervation in Hoxc-8−/− embryos (Fig. 6G). In contrast, all muscles that are normally innervated by C7-8Hoxc-8-positive MNs also receive, in the mutants, axonal projections from C5-6 and T1 MNs (Figs 6J,K,N,O, 7 right part of C and D). Furthermore, some MNs supplying extensor muscles are found in a medial position, whereas others that supply flexor muscles are observed in a lateral position (Right part of Fig. 7C). Such a mediolateral disorganization of the LMC can be either a mechanical consequence of shrinking of the ventral horns induced by the extensive apoptosis (Fig. 5), or a consequence of misrouting of the navigating growth cone. We conclude from these results that the inactivation of Hoxc-8 leads to a topographical disorganization of the brachial LMC. Altered somatotopic maps have been previously observed in Hoxa-1−/− embryos, in which ectopic MNs from rhombomeres 7 and 8 projected to the VIIth cranial nerve, normally supplied by MNs from rhombomeres 5 and 6 (Carpenter et al., 1993).
An important criterion in the identification of MNs is the muscle that they innervate (Eisen, 1994). The genetic inactivation of Hoxc-8 resulting in improper connections between ectopic C5-6, T1 MNs and distal target muscles may therefore be interpreted as a change of MN identity. However, because C5-6 MNs do not express Hoxc-8 (Fig. 3A,E,I), a change in their own R-C specification is unlikely. MN identity does not therefore rely only on intrinsic specification of the MN itself, but more likely encompasses the overall processes of growth cone pathfinding and target recognition. In embryos, the most rostral region of the paraxial mesoderm in which Hoxc-8 is expressed (albeit at a low level) is the 13th pair of somites (S13) and expression is much higher caudally to S15. Forelimb myoblasts arise from S9-13 (Burke et al., 1995) and consistently do not express Hoxc-8 (data not shown). The property of axons to follow a specific pathway lies in the ability of the growth cones to specifically respond to guidance cues emanating from the plexus mesenchymal region (Lance-Jones and Landmesser, 1980a,b, 1981a, b; Tosney and Landmesser, 1984; Lance-Jones and Dias, 1991). The growth cones of brachial MNs progress amidst cells from the mesenchyme which, in the proximal region of the limb, express Hoxc-8 (Fig. 2C and Le Mouellic et al., 1992). The absence of the Hoxc-8 product in these cells could therefore induce misprojection of C5-6 MNs and misrouting of some axons from Flex-MNs and Ext-MNs along dorsal and ventral nerve trunks, respectively. Ectopic anterior expression of Hoxc-6 in the cervical lateral mesoderm results in nerve truncation, suggesting that a specific change in the Hox code can locally induce release of molecules interferring with the axonal outgrowth (Burke and Tabin, 1996).
Precise cellular and molecular mechanisms by which Hox genes confer matching specifications will have to await better knowledge of Hox-target genes. Interesting insights, however, have arisen from Drosophila in which the connectin transcription is respectively activated or repressed by Antp and Ubx (Gould and White, 1992). This surface protein, expressed by a subset of embryonic MNs and their target muscles (Nose et al., 1992), determines the specificity of synapse formation through homophilic interaction. Ectopic expression of connectin in muscles is sufficient to specify the formation of new synapses with connectin-positive MNs (Nose et al., 1997). These data strongly suggest that the MN itself, in addition to the cells along its axon projection pathway and the target muscle fibers, contributes to the establishment of a properly organized connectivity pattern. Some of the information required for the execution of this pattern is encoded by Hox genes, and the severe phenotypes of Hox mutant mice reveal precise domains of the neural tube where the patterning of each Hox gene is essential.
We thank M. Curfs, J. Dederen and A. Gribnau for determining help in the CTB-HRP experiments; Professor T. M. Jessell for the K5 antibody; Drs Sh. Tajbakhsh and M. Buckingham for myf-5+/lacZ mice; Dr Th. Boulinier for statistical tests; Professor J. J. Panthier, in whose laboratory part of the anatomical studies were done; J. C. Bénichou and R. Hellio for expert advice on microscopy and confocal laser microscopy, C. Elbaz, S. Russe and Dr F. Bernex for technical help or advice; Drs J. Sanes, A. Lumsden, C. Henderson and members of his laboratory for helpful discussion and experimental suggestions during the course of this work. The authors also thank Dr U. Maskos for his experimental expertise and critical reading of the manuscript. This work was supported by grants from the Centre National de la Recherche Scientifique, the Association Française contre les Myopathies, the Association pour la Recherche sur le Cancer, the Ligue Nationale contre le Cancer, the Groupement de Recherches et d’Etudes sur le Génome and the European Economic Community (CT 96-0378).