Pax-3 is required for the development of limb muscles: a possible role for the migration of dermomyotomal muscle progenitor cells

Skeletal muscles of the trunk develop exclusively from somites (Christ et al., 1986). Within the somites two different myogenic cell lineages can be distinguished: (1) the so-called myotome, which gives rise to axial skeletal muscles (i.e. the deep musculature of the back) and (2) cells that form the muscles in limbs and body wall (Christ et al., 1986; Wachtler and Christ, 1992). Precursor cells for these latter muscles migrate from the ventrolateral dermomyotome of early somites into the limb buds, prior to the formation of the myotome (Solursh et al., 1987; Christ et al., 1977; Ordahl and Le Douarin, 1992).

No genuine molecular markers for migrating myogenic precursors have been described, making it difficult to follow this population of cells in situ. In contrast, myotomal myoblasts and muscle cells in the developing limbs at later stages can be identified readily, due to the expression of myogenic determination genes of the bHLH family (reviewed by Arnold and Braun, 1993). Transcripts of Myf-5, the earliest myogenic regulatory factor, appear around day 8 p.c. (post coitum) during mouse development in undifferentiated, still epithelial somites (Ott et al., 1991). Expression of myogenin, Myf-6 and MyoD follows in cells of the myotome in this sequence (reviewed by Buckingham, 1992; Arnold and Braun, 1993).

It has been shown recently that the formation of limb muscles is strongly impaired in the homozygous splotch mutant mouse, while development of back and ventral body wall musculature appears to be normal except for some reduction of dorsal muscles at the level of the neural tube defect (Franz et al., 1993; Franz, 1993).

The murine splotch mutant is a well-defined model for neural crest and neural tube defects. Homozygous splotch animals typically display neural tube defects and deficiencies in neural crest-derived tissues, such as dorsal root ganglia, sympathetic trunk, the septum of the truncus arteriosus, and Schwann and pigment cells (Auerbach, 1954; Moase and Trasler, 1989; Franz, 1989, 1990; Grim et al., 1992). The various splotch alleles (Splotch, Sp; Sp-delayed, Sp^d; Sp-retarded, Sp^r; Sp^1H; Sp^2H) display a semi-dominant phenotype in structures closely associated with Pax-3 expression. Both the Sp, Sp^r, Sp^1H, Sp^2H and Sp^4H alleles have been characterized at the molecular level and have been shown to correspond to mutations in the Pax-3 gene (Epstein et al., 1991, 1993, Vogan et al., 1993, Goulding et al., 1993a). Pax-3, a member of the murine paired-box-containing gene family encodes a transcription factor with 479 amino acids in length containing two conserved DNA-binding motifs: the 128 amino acid paired domain and a 60 amino acid paired-type homeodomain (Goulding et al., 1991). Pax-3 expression has been detected during brain development, in the alar and roof plates of the developing neural tube, in various neural crest derivatives (i.e. Schwann cells and dorsal root ganglia) and within somites (Goulding et al., 1991).

Whereas the expression of Pax-3 in neural structures and most aspects associated with it in splotch mutants have been documented extensively, the role of Pax-3 for the development of paraxial mesoderm remained unclear. The specific skeletal muscle phenotype recently observed in the splotch mutant indicates that Pax-3 may be essential for normal development of limb muscles that are known to be derived from the lateral region of somites. In order to gain some insight into the possible role of Pax-3 in developing somites and to understand better the splotch-associated defect in limb muscle formation, we analysed the distribution of cells containing Pax-3 and myogenic factor transcripts in normal and homozygous splotch mutant mice. We demonstrate that the Pax-3 probe marks a population of cells that migrates from the lateral dermomyotome into the limbs. We furthermore show that this cell population is missing in the splotch mutant. We also show that transcripts for Pax-3 and myogenic factor genes that appear coexpressed in normal limbs are absent in mutant embryos. From these results, we conclude that Pax-3 may be functionally involved in the formation of limb muscle originating from early somitic dermomytome.

Sp, Sp^1H and Sp^2H mouse lines were obtained from Jackson Laboratories (Bar Harbour) and MRC Radiobiology Unit (Harwell, Didcot, England). They were bred with C57BL/6 (Sp) and C3H (Sp^1H and Sp^2H) strains. Embryos were staged, counting the appearance of the vaginal plug as day 0.5 p.c.

Expression of the Pax-3 gene was analysed by in situ hybridization using a 520 bp HindIII/PstI fragment of the cDNA. It encodes the 3′ part of the paired-type homeodomain and most of the carboxyterminus (Goulding et al., 1991).

Mouse embryos were collected and treated as described by Wilkinson (1992). Briefly, embryos were fixed in 50 mM phosphate-buffered saline pH 7.2 (PBS) containing 4% paraformaldehyde for 16 hours followed by two washing steps in PBT (PBS and 0.1% Tween-20) and dehydration with 25%, 50%, 75% and100% methanol. Fixed and dehydrated embryos were stored at −20 °C.

For whole-mount in situ hybridization, embryos were rehydrated through the methanol series, washed twice in PBT and bleached in PBT containing 6% hydrogen peroxide for 1 hour. Embryos were then subjected to digestion with 10 μg/ml of proteinase K for 7–15 minutes followed by washing in glycine solution (2 mg/ml). These embryos were refixed in gluteraldehyde/paraformaldehyde (0.2%/4%) for 20 minutes, washed twice in PBT and transferred to hybridization buffer (50% formamide, 5× SSC pH 4.5, 50 μg/ml yeast RNA, 1% SDS, 50 μg/ml heparin) for prehybridization at 70°C for at least 1 hour. Synthesis of sense and antisense digoxigenin-labelled RNA probes was performed using a DIG RNA Labeling Kit (Boehringer Mannheim) according to the manufacturer’s instructions. Hybridization was performed in the same buffer as used for prehybridization containing 1 μg/ml of digoxigenin-labelled RNA probe at 70 °C overnight. Following hybridization, embryos were washed twice for 30 minutes in 50% formamide, 5× SSC pH 4.5, 1% SDS at 70 °C, followed by two washing steps in 0.5 M NaCl, 10 mM Tris-HCl pH 7.5, 0.1% Tween-20 with 100 μg/ml RNase A at 37 °C and two steps in 50% formamide, 2× SSC pH 4.5. Embryos were then rinsed several times in TBST buffer (0.14 M NaCl, 25 mM KCl, 25 mM Tris-HCl pH 7.5, 1% Tween-20, 2 mM levamisole) and incubated overnight at 4 °C with sheep polyclonal alkaline phosphatase-conjugated antidigoxigenin antibody (Boehringer Mannheim) preabsorbed with mouse embryo powder. Next day, embryos were washed first with TBST, then in NTMT buffer (100 mM NaCl, 100 mM Tris-HCl pH 9.5, 50 mM MgCl₂, 1% Tween-20, 2 mM levamisole) and the phosphatase reaction was performed in the presence of nitroblue-tetrazolium chloride (0.34 mg/ml) and 5-bromo-4-chloro-3-indolylphosphate (0.18 mg/ml) in NTMT buffer. The reaction was stopped in PBT.

Embryos were either cleared in glycerol PBT solution or embedded in a mixture of gelatin, albumen and sucrose, and sectioned at 50 μm using the Pelco 101 vibratome. Sections were mounted on gelatinsubbed slides and photographed with a Zeiss Axiophot microscope using bright-field illumination. Cleared embryos were photographed under a Zeiss SV11 stereomicroscope using a DIA-duplicator (Elinchrom Co.) to provide a flashlight.

In situ hybridization was performed on 7 μm thick paraffin tissue sections as described previously (Sassoon et al., 1988; Bober et al., 1991). The ³⁵S-labelled probes for myogenin and MyoD were prepared as described by Sassoon et al. (1988), and for Myf-5 as described by Ott et al. (1991). The Pax-3 probe was used as described for whole-mount in situ hybridization.

In order to determine Pax-3 expression in somites of wild-type and splotch mutant embryos, a series of whole-mount in situ hybridizations were performed between days 9 and 12 p.c. As somites develop in a rostrocaudal direction, newly formed undifferentiated somites and already fully differentiated somites can be observed in the same embryo in caudal and cranial regions respectively (Fig. 1). We found that Pax-3 expression was evenly distributed over newly formed somites (Fig. 1A: day 9.5 p.c., caudal part), whereas at later stages of somitic development the Pax-3 signal was confined to the most caudal and ventrolateral edge in each somite (Fig. 1A-C, arrowhead, see also Fig. 2C,D). The restricted localization of the Pax-3 signal was most pronounced around day 11 p.c., when it was clustered in the ventrolateral bud appearing in individual somites of the trunk (Fig. 1D,E, arrowhead). In addition a clearly weaker expression domain was observed more dorsomedially in two parallel stripes along the anterior and posterior borders of each somite (Fig. 1E, arrow). This distribution of Pax-3 transcripts may correlate to yet unidentified functional domains within developing somites.

In homozygous splotch embryos, distinct differences in the distribution of Pax-3 transcripts were observed (Fig. 1F-J). While the Pax-3 signal in early somites appeared similar to wild-type, no local redistribution to the ventrolateral part occurred in the more advanced somites (day 9.5 p.c.; Fig. 1F, arrowhead). Later, (day 10.5–11.5 p.c.) when the somitic domain labelled by Pax-3 normally elongated further ventrolaterally, this area appeared much shorter and disorganized in mutant embryos (Figs 1G-J, 2E,F). The edges of somites delineated by the Pax-3 signal also appeared irregular (Fig. 1H,J, arrowhead). Furthermore, the Pax-3 expression domain was slightly shifted ventrally, away from the neural tube (Fig. 1J, arrow).

Similar evidence for the disorganization of somites in homozygous splotch embryos (days 10.5 and 11.5 p.c.) was also observed when the myotomal marker Myf-5 was used as a probe (data not shown). Thus, the structural organization of developing somite seemed to be altered in splotch animals.

In addition to somites, Pax-3 was also found to be expressed in the developing limbs (Figs 1, 2). Pax-3-positive cells were present in forelimbs and also in hind limbs of control embryos between day 9.5 and 10.5 p.c. (Fig. 1A-C). In marked contrast, no positive cells were observed in limbs of homozygous splotch animals at the equivalent stages (Fig. 1F-H).

Fig. 2 illustrates a time course of Pax- 3 transcript accumulation in the developing forelimbs of control (Fig. 2A-D) and homozygous splotch embryos (Fig. 2E,F), as seen in a dorsal view. Pax-3- expressing cells appeared to emigrate from 5 to 6 adjacent somites toward the limb field (Figs 1A, 2). The progression of Pax-3 signals from the somites into the forelimbs was first observed at day 9.5 p.c., increasing in a proximodistal direction between days 10.0 and 11.0 p.c., and decreasing thereafter. The early distribution of Pax-3 signal appeared to be compact (Fig. 2A,B), while it displayed a more scattered distribution of Pax- 3-expressing cells at later stages (Fig. 2C,D). A similar colonization of the hind limbs by Pax-3-expressing cells was also observed lagging about 0.5 day behind the forelimbs (data not shown and Figs 3, 4). Fig. 3 illustrates the progressive proximodistal invasion of forelimbs and hind limbs by Pax-3- expressing cells originating from the ventrolateral somitic region. The early expression seemed to be confined to the dermomyotomal cap of the somites (Fig. 3A). As this structure elongated ventrolaterally (Fig. 3B), cells appeared to detach from the somitic domain, accumulate at the body-limb junction and then further invade the limbs (Fig. 3C,D). Approximately at day 10.5 p.c., the Pax-3-expressing cell population segregated into dorsally and ventrally located domains (Fig. 3E,F). Significantly, no Pax-3-expressing cells that would leave the somites or populate the limbs between days 9.5 and 11.5 p.c. were observed in homozygous splotch embryos, although Pax- 3 expression was very strong in the dermomyotome and the neural tube of these animals (Figs 1-3). These findings strongly suggest that dermomytomal myogenic precursors either do not exist or are unable to migrate into the limbs in the homozygous splotch mutant.

The spatiotemporal expression of Pax-3 in paraxial mesoderm and regions of the limb buds suggests a potential role in the development of skeletal muscles. In order to correlate Pax-3 expression domains to early muscle lineage markers, in situ hybridizations using Pax-3 and Myf-5 probes were performed on serial sections of mouse embryos. It has been shown previously that Myf-5 constitutes the first myogenic regulatory gene that is expressed in somites and muscle cells in the developing limbs (Ott et al., 1991).

Fig. 4 shows in situ hybridizations on transverse sections of a day 10.5 p.c. mouse embryo at the forelimb and hind limb levels. Both Pax-3 and Myf-5 transcripts were detected in somites and limb buds in overlapping but not completely identical areas. While Myf-5 transcripts were confined to the myotomal layer of somites, Pax-3 expression was clearly wider including both myotome and dermatome (Fig. 4A-D,K,L).

Similar expression domains for Pax-3 and Myf-5 were observed in two locations of forelimbs at the ventral and dorsal sides, which later give rise to the flexor and extensor muscles (Fig. 4I,J). At the same stage, Pax-3 transcripts have already accumulated in the hind limb, whereas Myf-5 was not yet expressed (Fig. 4K,L). This indicates the later onset of Myf-5 gene activity compared to Pax-3, which was also observed in forelimbs of younger embryos (data not shown). Interestingly, Pax-3 transcripts were also detected in the region between trunk and limbs (Fig. 4A,C), an area that presumably contains migrating muscle precursor cells as shown by chicken/quail somite grafting experiments (Christ et al., 1977; Ordahl and Le Douarin, 1992). These cells can not be recognized with the Myf-5 probe as they express no myogenic bHLH factors (Fig. 4B; Ott et al., 1991; Bober et al., 1991).

Later during development, Pax-3 expression gradually decreased to almost undetectable levels at day 12.5 p.c. (data not shown). In addition to mesodermal cells, Pax-3 transcripts were also seen in the dorsal region of the neural tube as described earlier (Goulding et al., 1991).

From these results, we conclude that Pax-3 is expressed in mesodermal cells that give rise to axial muscles and the premuscle cells in the limbs. Its expression starts clearly prior to Myf-5 in somites and overlaps with Myf-5 expression in the myotome. Unlike Myf-5, it is also expressed in the lateral dermatome, which represents cells that migrate to the limb field where they form muscle.

Based on morphology and the expression of the muscle marker desmin, it has been suggested previously that the splotch mutation affects normal development of skeletal musculature in the limbs (Franz et al., 1993). In order to assess at which stage limb muscle development might be impaired by the Pax-3 mutation, we performed in situ hybridizations using the myogenic bHLH genes Myf-5, MyoD and myogenin as probes. As these genes encode cell-type-specific transcription factors that are essential for the establishment of the skeletal muscle lineage, they constitute highly specific and early markers (Arnold and Braun, 1993).

Comparison of myogenin and MyoD expression in day 11.5 p.c. embryos of wild-type and homozygous splotch mutant mice revealed that both myogenic regulatory genes were expressed at similar levels in axial and body wall muscles of control and mutant animals (Fig. 5 and data not shown). However, no myogenin and MyoD expression was found in limbs of mutant embryos, although control animals of the same stage highly expressed these factors in limb musculature (Fig. 5 and data not shown).

Similar results were obtained in day 9.5 and 10.5 p.c. embryos using Myf-5 as an additional probe (data not shown). This analysis was extended to day 13.5 p.c. embryos, the latest viable stage of the splotch mutant. As shown in Fig. 6, no muscle could be labelled in limbs of the splotch mutant using MyoD as a hybridization probe, whereas all muscle groups in control animals were highly positive for this marker. In contrast, no major difference for the expression of MyoD was observed in axial and body wall muscles of wild-type and mutant embryos. Occasionally, a slight reduction of deep back muscles was seen in mutants at the level of the neural tube defect (data not shown).

Taken together, these observations suggest that a mutation in the Pax-3 gene causes a severe defect in the formation of the limb muscles, whereas axial muscles are less or not affected. The defect appears not to be due to a delay of limb muscle formation but rather to the inability of dermomyotomal muscle precursor cells to reach the limb bud.

Whole-mount in situ hybridizations on early mouse embryos indicate that the Pax-3 gene is expressed over the entire area of newly formed somites. During somite maturation, this Pax-3- positive domain undergoes a remarkable redistribution between days 9 and 9.5 p.c. of development. At this stage, Pax- 3 transcripts become confined to the caudal region and the ventrolateral aspects of each somite. As somites further mature (between day 10 and 10.5 p.c.) and dermomyotomal cells begin to invade adjacent limbs and ventral body wall, the ventrolateral part of somites retain high levels of Pax-3 expression. At the trunk level, epithelial-like Pax-3-positive cells progressively invade the lateral somatopleure. Based on this distribution, we believe that these Pax-3-expressing cells correspond to actively proliferating precursor cells emanating from the dermomyotome (Kaehn et al., 1988).

In contrast to Pax-3-expressing cells in the myotome, which also express the muscle-specific transcription factor Myf-5, lateral dermomyotomal cells are devoid of any myogenic bHLH proteins. Therefore, Pax-3 constitutes a convenient marker to label this cell population.

Approximately between days 9 and 9.5 p.c., Pax-3-positive cells can be observed emerging from the dermomyotome at the level of the forelimb bud. About half a day later, similar cells can be seen next to the hind limb bud. These cells appear dispersed in small clusters rather than as sheets of epithelial cells. At later stages, the Pax-3 signals become organized in two domains in the limbs, one located dorsally and one ventrally in the position of the limb premuscle masses. These Pax-3- positive cells colocalize with Myf-5-expressing cells. As the other myogenic factors begin to be expressed (MyoD, myogenin) and muscle differentiation starts, the Pax-3 signal disappears gradually from this region (days 11-12 p.c.). Taken together, this evidence suggests to us that Pax-3-expressing cells invade the limbs and correspond to migratory limb muscle precursors. The spatiotemporal distribution of Pax-3-positive cells, their progressive proximodistal extension into the limbs, and the subsequent organization into dorsal and ventral cell masses that also express myogenic factors support this notion. The fact that Pax-3-positive cells colonizing the limbs originate from 5-6 adjacent somites is in agreement with previous reports on the origin of limb myoblasts in chicken (Christ et al., 1997; Hayashi and Ozawa, 1991; Beresford, 1983).

Since migratory somitic cells have been shown to give rise only to skeletal muscle, while mesenchymal components of the limb originate from the somatopleure (Chevallier et al., 1977; Christ et al., 1977; Wachtler and Christ, 1992; Ordahl and Le Douarin, 1992), it appears unlikely that these Pax-3-expressing cells contribute to other structures than myogenic limb precursor cells. However, formal proof of this point will require lineage analysis or functional assays. Pax-3 also labels cells in the ventrolateral bud of somites, which migrate into the lateral somatopleure and give rise to body wall muscles (Christ et al., 1983). Again, Pax-3 expression appears to label early migrating myogenic precursors at the trunk level.

Until now, no molecular marker for these migrating cells has been available and they have only be followed by interspecies transplantation (chicken-quail) or by injection of vital lipophilic dye into the somitic cavity (Christ et al., 1977; Ordahl and Le Douarin, 1992; Hayashi and Ozawa, 1991). Pax-3 appears to be a suitable marker to trace these early myogenic precursors both at the limb and trunk levels.

The recent demonstration of a dramatic loss of limb musculature in homozygous Sp^1H and Sp^d mice (Franz et al., 1993) underlines the importance of Pax-3 for the development of limb muscles. The experiments presented here demonstrate that, in these mice, Pax-3-expressing cells can be detected in the dermomyotome but not in the lateral region between somites and limb buds or in the limbs itself. While other domains of Pax-3 expression such as in the dorsal neural tube and the somitic myotome are preserved in splotch, the limb expression domain is specifically missing in homozygous Sp, Sp^1H and Sp^2H mice. In addition, no myoblasts expressing myogenic factors are detected in the limbs at any stage. Taken together, these results strongly suggest that Pax-3 protein is required either for the generation of early limb muscle precursors within the somites or for the migration of these progenitors from the dermomyotome into the limbs. Alternatively, it remains formally possible that the population of myogenic cells shuts off Pax-3 expression in the mutant thereby escaping detection, although they still migrate from the somites. If this would be the case, their myogenic potential should be dependent on a functional Pax-3 protein and differentiation would be blocked in its absence.

Homozygous splotch embryos (Sp^1H) appeared to have normal facial muscles but some reduction of the tongue muscles was observed in our analysis. However, whether this represents a muscle deficiency or a delay in muscle development will require further investigation. Interestingly, the tongue muscles originate from occipital somites, from which precursors disssociate before migrating anteriorly (Schemainda 1979; Noden 1983; Couly et al., 1993). In this regard, muscle development in the tongue and the limbs is similar, as in both cases precursors leave the somites and reach their final destination before they differentiate. Thus, it remains likely that Pax-3 mutations predominantly affect migratory precursors. Since axial, facial and body wall muscles, in contrast to limb muscles, develop almost normally in the splotch mutant (Franz et al., 1993; Franz, 1993), Pax-3 appears not to be required in general for muscle differentiation. However, as some reduction of the axial muscle mass at the level of the neural tube defect has also been observed in splotch mice (Franz et al., 1993), a role for Pax-3 in the control of early myoblast proliferation cannot be excluded. Two other lines of evidence may support such a role for Pax-3. Firstly, aberrant PAX3 transcripts are found in a human alveolar rhabdomyosarcoma associated with a frequent chromosomal translocation. This abnormal PAX3 expression may favor proliferation of myogenic precursors thereby inhibiting their differentiation (Barr, 1993). Secondly, Pax genes have been shown to be able to transform NIH 3T3 and 208 fibroblasts in vitro (Maulbecker and Gruss, 1993).

Because Pax-3 expression precedes the establishment of dorsoventral polarity in somites, it could conceivably be involved in a cascade leading to the specification of the dermomyotome. The expression of Pax-3 in the neural tube has already been shown to respond to ventralizing signals from the notochord (Goulding et al., 1993b). In fact, a ventralizing function of the notochord and floor plate can be assumed for the paraxial mesoderm. Transplantation of these structures to the dorsolateral part of the neural tube repatterns adjacent somites, thereby preventing muscle formation and stimulating sclerotomal differentiation (Pourquie et al., 1993). Further experiments will be required to determine whether dorsal somitic structures develop in a default pathway in the absence of ventralizing signals or whether a dorsalizing influence is necessary for the establishment of Pax-3 expression and the formation of dermomyotome.

Thus, in addition to being an important factor whose inactivation can lead to neural crest and neural tube defects, Pax-3 is essential for the specification, proliferation and/or migration of myoblast stem cells destined to colonize the limbs and give rise to the different muscles of the limbs.

Interestingly, the various neural crest deficiencies observed in splotch appear to result from a migration defect since neural crest cells isolated in vitro migrate away from the neural tube in a delayed fashion (Moase and Trasler, 1990). Although the molecular basis for this aberrant cellular behavior is unknown, we assume that, for instance, some cell adhesion molecules or their cellular receptors normally involved in guidance of myoblasts and neural crest cells, may be deficient in splotch animals. At the limb level, fibronectin, hyaluronic acid and other mucopolysaccharides are possibly involved in this process (for a review see Wachtler and Christ, 1992). The general disorganization of the dermomyotome at the trunk level in the splotch embryos may support the hypothesis that the morphological integrity of tissues depends on the respective adhesive properties of their cellular components. It has already been postulated that homeobox-containing proteins may be directly involved in the transcriptional control of such molecules (Edelman and Jones, 1992). Recently, binding to and activation of the NCAM gene promoter by the Hox B8 and B9 proteins has been demonstrated in tissue culture experiments (Jones et al., 1992). Pax proteins may participate in such a process as well.

Various lesions in PAX3, which represents the human homologue of the murine Pax-3 gene, have been associated with the human Waardenburg syndromes (Hoth et al., 1993). These syndromes (WS type 1, type 2 and type 3), generally inherited in an autosomal dominant trait, constitute a phenotypically and genetically heterogeneous family of disorders. However, the major clinical manifestations of these syndromes involve a combination of defects of neural crest origin: dystopia canthorum, (lateral displacement of the inner corner of the eye), deafness and pigmentation deficiencies (white forelock and eyelashes, hypopigmented iris, premature greying, hypopigmented skin lesions). Interestingly, WS3 patients suffer from limb deformities resulting from muscle and skeletal hypoplasia (McKusick, 1992). The data presented here for mouse, demonstrating that Pax-3 is essential for the proper specification, proliferation and/or migration of myoblast precursors destined to colonize the limb, provide a possible explanation for the phenotype observed in patients with WS3. However, one must bear in mind that the phenotype observed in WS3 patients is dominant, observed in individuals with one normal PAX3 allele. It remains to be seen whether quantitative differences in the migration of myoblast precursors can also be observed in mice heterozygous for the Sp mutation.

We thank Christiane Müller for excellent technical assistance. We are particularly grateful to Thomas Braun for helpful discussions and Fabienne Pituello, Luc St-Onge and Michael Kessel for critical comments on this manuscript. P. T. is the recipient of a fellowship from the Medical Research Council of Canada. This work was supported by grants of the Deutsche Forschungsgemeinschaft and the European Community to H. H. A. and by the Max Planck Society.