Syndecan regulates cell migration and axon guidance in C. elegans

During nervous system development, migrating cells and axonal growth cones are guided by extracellular stimuli to their final position in the body. Although several major families of attractive and repulsive guidance cues have been characterized, it is not clear how the presence of multiple directional cues are integrated by a single outgrowing axon while leaving other neurons unaffected. Recent findings indicate that heparan sulfate proteoglycans(HSPGs) act as crucial modulators of axon guidance choices(Lee and Chien, 2004). HSPGs are present both on the cell surface and in the extracellular matrix, where they modulate cellular responses to extracellular signals in various developmental and pathological processes(Sasisekharan et al., 2002; Topczewski et al., 2001). Their function in nervous system development has long been presumed, but only recent genetic studies have provided in vivo evidence for their role in axon pathfinding (Bülow and Hobert,2004; Inatani et al.,2003; Johnson et al.,2004; Lee et al.,2004; Steigemann et al.,2004).

HSPGs consist of a secreted or membrane-anchored protein core to which several heparan sulfate (HS) polysaccharide chains are attached(Bernfield et al., 1999; Esko and Selleck, 2002; Lindahl et al., 1998). Once the HS chain is synthesized, specific enzymes – including sulfotransferases and an epimerase – modify the individual sugar moieties, resulting in complex modification patterns(Lindahl et al., 1998). A complete lack of HS abrogates multiple signaling pathways essential for early development, resulting in severe patterning defects and lethality(Inatani et al., 2003; Bornemann et al., 2004; Herman et al., 1999; Morio et al., 2003; Perrimon and Bernfield, 2000). Together with the biochemical studies, these observations led to the hypothesis that HSPGs act as co-receptors that bind and sequester extracellular signals in the extracellular matrix(Bernfield et al., 1999; Perrimon and Bernfield,2000).

By contrast, genetic removal of enzymes that secondarily modify HS,including sulfotransferases and an epimerase, demonstrated that specific modifications are needed for distinct axon guidance choices(Bülow and Hobert, 2004). Hence, it has been proposed that individual HS modifications in a given area or on specific neurons give rise to a `sugar code' that regulates axon guidance by modulating specific receptor-ligand interactions. So far it is not known, however, which core proteins are crucial in this process and how they function to modulate axon guidance signaling. Syndecans and glypicans are the two major families of cell surface-bound HSPGs(Fig. 1A). In mice, the inactivation of a single syndecan gene does not result in any obvious phenotype, probably due to redundancy between the family members(Hartmann and Maurer, 2001; Bernfield et al., 1999).

Syndecan mutant flies display a Robo-like midline phenotype, which can be rescued by overexpression of either syndecan or the two Drosophilaglypican genes dally and dally-like(Johnson et al., 2004; Steigemann et al., 2004). Furthermore, it was found that lack of syndecan affects the distribution of the midline repellent Slit, which provides the first indication of how HSPGs might function (Johnson et al.,2004).

Unlike mammals, which have four syndecan family members, the Caenorhabditis elegans genome contains one single syndecan (core)gene. We show here that syndecan SDN-1 functions autonomously in neurons to ensure correct cell migration and axon guidance. Epistasis analysis indicates that signaling pathway, which acts in parallel with a second, currently unknown, guidance system that relies on a different combination of sugar modifications on a distinct HSPG protein. We therefore propose that at least two distinct HSPG patterns are involved in regulating axon pathfinding and cell migration.

Standard methods were used for maintaining and manipulating Caenorhabditis elegans (Brenner,1974). The C. elegans Bristol strain N2 was used as wild-type reference in all experiments. Animals were raised at 20°C unless indicated otherwise. The following mutations were used: LG III: hse-5(tm472) (Bülow and Hobert, 2004); LGX: sdn-1(zh20) (this study), sdn-1(ok449) (Minniti et al.,2004), hst-6(ok273) and hst-2(ok595)(Bülow and Hobert, 2004), sax-3(ky123) (Zallen et al.,1998), slt-1(eh15)(Hao et al., 2001). All transgenes used to score neuroanatomy were described previously ; Bülow et al., 2002).

The P_sdn-1::sdn-1::gfp translational fusion construct pAH40 was generated by ligating a 5.7 kb BamHI-SphI fragment covering 2.8 kb of 5′ regulatory sequences and the entire coding region into the pPD95.75 vector (gift of A. Fire, Stanford University), such that the full-length sdn-1 open reading frame was fused in frame to gfp. pAH40 and the lin-15 rescuing plasmid pL15EK (gift of R. Horvitz) were co-injected (100 ng/μl of each) into lin-15(n765ts) animals to, obtain transgenic lines. Two independent lines (opIs170, opIs171) showed similar expression patterns.

P_sdn-1::sdn-1 (pAH46) was made by cloning a 6.9 kb genomic PCR product containing 2.8 kb of 5′ sequences, the entire sdn-1 coding region and the predicted 3′UTR into the pGEM-T vector. The P_dpy-7::sdn-1(cDNA) construct was generated by fusion PCR (Hobert, 2002): the hypodermal promoter of dpy-7 (216 bp)(Gilleard et al., 1997) was fused to the ATG start codon of the sdn-1 cDNA. The P_sra-6::sdn-1 plasmid (pCR1) and the P_unc-119::sdn-1 plasmid (pCR2) were made by subcloning the sra-6 promoter (3747 bp) or the unc-119 promoter (2189 bp)sequence into pPD95.86 (gift of A. Fire, Stanford University), in front of the ATG start codon of the sdn-1 cDNA. For rescue, worms were injected with 20 ng/μl of pAH46, pCR1 or pCR2, or with 2.5 ng/μl of the fusion construct P_dpy-7::sdn-1. Constructs were injected together with pTJ1157 (lin-48::gfp) and pBluescript SK at 50 ng/μl each,or, for the rescue of HSN migration, with pPD118.33 (myo-2::gfp) at 2.5 ng/μl.

The sdn-1(zh20) deletion mutant was isolated from an EMSmutagenized library consisting of approximately 106 haploid genomes, as previously described (Jansen et al.,1997; Berset et al.,2001). DNA pools were screened by nested PCR with the outer primers OAH40 (5′-TGGGTTCATCCAATGCGTATGACC-3′) and OAH49(5′-CCCTATTCTGTCATTGTACCAC-3′), and the nested primers OAH41(5′-TTGCCGTACACCGCCATTTCCTC-3′) and OAH48(5′-AGAATGGCTGTAATCACCGCAACG-3′). The zh20 deletion removes 1260 bp within the coding region(Fig. 1B). The sequence of the break point is as follows: TTTGCTTCACAC//zh20//GTCGACAGGCAG. The mutant strain was backcrossed six times to N2 before analysis.

All mutant alleles were backcrossed at least four times before phenotypic analysis. Axon guidance defects were scored in staged young adults or L1 larvae using a Leica DMR microscope. Migration of HSN, ALM and CAN neurons and coelomocytes were scored in adult stages (HSN, ALM, CAN) and L3-L4 stages(coelomocytes).

All phenotypes, except defects in guidance and left/right choices of D-type motoneurons (Fig. 6A), were scored as the percentage of animals that were defective, and are shown with error bars denoting the standard error of proportion. Statistical significance was calculated using the z-test. For the defects in D-type motoneurons, error bars give the standard error of the mean and statistical significance was calculated by using Student's t-test. If multiple comparisons were made, Bonferroni's correction was applied.

The C. elegans genome is predicted to code for only two cell surface-bound HSPGs genes, syndecan SDN-1 and glypican GPN-1(Fig. 1A), and three extracellular HSPGs, agrin (F41G3.8)(Hutter et al., 2000),perlecan UNC-52 (Rogalski et al.,1993) and CLE-1, a homolog of collagen XVIII(Ackley et al., 2001; Halfter et al., 1998).

To study the role of SDN-1 in nervous system development, we first isolated a null allele of sdn-1. The deletion allele sdn-1(zh20)removes most of the sdn-1 open reading frame(Fig. 1B). Furthermore, no sdn-1 transcripts were detected on northern blots of RNA isolated from zh20 animals (Fig. 1C). Thus, the sdn-1(zh20) allele most likely causes a complete loss of SDN-1 function. Homozygous sdn-1(zh20) animals are viable, but they show defects in backward locomotion, are variably egg-laying defective and have a slightly reduced brood size (201±42, n=10) when compared with wild-type worms (280±10, n=10).

A recent study carried out with a hypomorphic allele of sdn-1reported a role of SDN-1 in vulva development, but did not analyze neuronal phenotypes (Minniti et al.,2004). This published allele, sdn-1(ok449), results in an in-frame deletion and produces a truncated form of SDN-1 lacking the two major conserved HS attachment sites in the extracellular domain(Fig. 1B). The truncated SDN-1 core protein in ok449 animals could still be detected on western blots by using an antibody directed against the cytosolic tail, but was no longer recognized by a monoclonal antibody specific for HS side chains,indicating that there is considerably less, if any, HS attached to SDN-1 in this mutant (Minniti et al.,2004). We examined the neural phenotypes of both sdn-1mutants, but focused our epistasis analysis on the null allele sdn-1(zh20), because the ok449 deletion may not completely eliminate sdn-1 function (Table 1).

The formation of a functional nervous system starts with the migration of neural cell bodies to their genetically determined position. Thereafter,neurons extend axons that navigate to the target area to finally form synaptic contacts. The migrating neurons and axonal growth cones are guided by similar molecules. For example, UNC-6 Netrin controls both cell and growth cone migrations along the dorsoventral axis(Chan et al., 1996; Hedgecock et al., 1987). To study the role of SDN-1 in different phases of nervous system development, we crossed alleles into different reporter strains, in which specific types of neurons are labeled with gfp(Table 1). We observed no obvious changes in the overall organization of the nervous system in the major ganglia and nerve bundles in both sdn-1 alleles when using a pan-neuronal gfp marker (data not shown).

When examining single neuron subtypes, however, we observed prominent migration defects of neural cell bodies. The HSN left (L)/right (R)motoneurons normally migrate during embryonic development from the posterior to the midbody region (Fig. 2A). In both sdn-1 alleles, the HSN neurons either failed to migrate at all or stopped prematurely before reaching their normal position near the vulva (Fig. 2C,D). A null mutation in the 2O-sulfotransferase hst-2 gene causes similar HSN migration defects, although they are less severe than those observed in sdn-1(zh20) animals(Fig. 2E)(Bülow and Hobert, 2004; Kinnunen et al., 2005). By contrast, neither the 6O-sulfotransferase HST-6 nor the C5-epimerase HSE-5 is required for HSN migration(Bülow and Hobert, 2004),pointing to a crucial role of 2O-sulfated HS in this process. Double-null mutants of sdn-1 and hst-2 show significantly more HSN migration defects, indicating that 2O-sulfation, although possibly important for SDN-1 function, is also required for, at least, one additional HSPG (Fig. 2E). Loss of HSE-5 similarly enhances the HSN migration defects of sdn-1 null mutants, whereas removal of HST-6 in a sdn-1 background shows no increase of migration errors (Fig. 2E). Next, we tested posterior migration of the ALML/R sensory neurons. In the absence of SDN-1, ALML/R neurons stopped prematurely(Table 1). Migration defects of the ALML/R neurons have also been described in mutants for the C. elegans Robo receptor SAX-3 (Zallen et al., 1999). We examined ALM migration errors of both sdn-1(zh20) and sax-3(ky123) mutants and found comparable defects [47% for sdn-1(zh20), 45% for sax-3(ky123)]. In addition, loss of SDN-1 function impaired the posterior migration of the CANL/R neural cell bodies (Table 1).

Finally, we detected a cell migration defect of the macrophage-like coelomocytes, a non-neural cell-type. In wild-type hermaphrodites, four coelomocytes are born and migrate posteriorly during embryogenesis, whereas the two coelomocytes generated in the L1 stage do not migrate(Fig. 2F). In sdn-1(zh20) null mutants, the embryonic coelomocytes were often located too far anteriorly, just behind the pharynx(Fig. 2H). In all the cell migration events examined, SDN-1 always appeared to confer a pro-migratory property, irrespective of the direction of cell migration.

Because mutants for the HS-modifying enzymes have previously been reported to display guidance defects of midline interneurons(Bülow and Hobert, 2004),we tested whether sdn-1(zh20) null mutants would show similar defects. The ventral nerve cord (VNC) runs along the left and right side of the ventral hypodermal ridge in two separate fascicles. We first examined the guidance of the PVQ interneurons, a pair of bilaterally symmetric neurons located in the tail (Fig. 3A). The PVQL and PVQR axons normally grow out to the head on the left and right side of the midline, respectively, without crossing to the contralateral side(Fig. 3A,B). In both sdn-1 mutants, PVQ axons inappropriately crossed to the contralateral side or were entirely fused in a single track(Fig. 3C). Most crossover defects (75%) consisted of single or multiple midline crossings of PVQ axons along the entire length of PVQ extention, the remaining 25% showed entirely collapsed PVQ axons. The PVPL/R and AVK interneurons showed comparable crossover defects (Fig. 3F, Table 1).

In contrast to most axons, which join the VNC in the head or the tail, the axons of the HSN neurons join the VNC midway near the vulva and grow anteriorly without crossing the midline(Fig. 3G,H). In sdn-1(zh20) animals, HSN axons frequently crossed the midline(Fig. 3I). As mentioned before,the HSN neurons frequently fail to migrate to their normal position in sdn-1 mutants (Table 1). To determine whether the misplacement of the HSN cell bodies might contribute to the observed guidance defects, we scored HSN pathfinding in animals with wild-type and misplaced HSN neurons separately(Fig. 3I, Table 1). We found that crossover defects occurred in both cases, but were indeed more frequent in animals with mispositioned HSN cell bodies.

Midline decisions are also made by different types of motoneurons in the VNC. In the case of D-type motoneurons (DD and VD subtypes), axons extend first in the right VNC and then circumferentially grow out on the right side of the animal (with the exception of the two anteriormost DD and VD neurons,which grow out on the left) to form the commissures that connect to the dorsal nerve cord (DNC; Fig. 4A). The axons of the six DD neurons grow out during embryonic development, whereas those of the 13 VD neurons grow out postembryonically. We analyzed sdn-1(zh20) and sdn-1(ok449) mutants for two major guidance defects: first, the circumferential guidance to the DNC; and, second, the left/right decision during the exit from the VNC(Fig. 4E). In wild-type worms,the commissural axons always reach the DNC where they bifurcate and connect to neighboring branches of the D-type motoneurons to form a continuous DNC fascicle (Fig. 4C,E). In both sdn-1 mutants, the circumferential outgrowth of commissures was impaired, especially on the left side of the worm(Fig. 4B,D,E). Furthermore, sdn-1 mutants showed partial VNC defasciculation and an increased number of commissures inappropriately leaving the VNC on the left side(Fig. 4E, Table 1). The outgrowth of other types of neurons, such as the sensory neurons in the head or the pioneering axons in the VNC, were not affected in sdn-1(zh20) null mutants (Table 1).

Taken together, our findings indicate that syndecan SDN-1 is required for several distinct axon guidance and cellular migration events. Interestingly,the sdn-1(zh20) axon guidance defects that we describe here are similar to those previously reported in mutants for the HS-modifying enzymes(Bülow and Hobert, 2004),suggesting that SDN-1 might be a major target of these enzymes in the developing nervous system. Moreover, animals expressing the truncated form of syndecan lacking the major HS attachment sites showed qualitatively and quantitatively similar defects to those observed in syndecan null mutants,pointing to a major role of the HS moiety of SDN-1 in regulating cell migration and axonal pathfinding.

In order to examine the expression pattern of SDN-1, we generated transgenic animals expressing a full-length Cterminally tagged SDN-1::GFP fusion protein. In embryos and L1 larvae, SDN-1::GFP was expressed in many tissues, including neurons, pharynx and hypodermis(Fig. 5A′). In later larval stages and in adult worms, SDN-1::GFP was predominantly expressed in the nervous system, with strong expression in the nerve ring and the VNC motoneurons (Fig. 5B,C,E,F). Apart from the neuronal expression, SDN-1::GFP was also found at a lower level in the hypodermis (Fig. 5D) and the vulval cells (data not shown). Because antibodies directed against the conserved intracellular domain of mammalian syndecan 4 have been reported to specifically detect C. elegans SDN-1 in the nerve ring and the vulva(Minniti et al., 2004), the SDN-1::GFP reporter likely reflects the endogenous SDN-1 expression pattern.

Based on the SDN-1::GFP expression pattern, it is conceivable that SDN-1 could affect neuronal differentiation in a cell nonautonomous manner. In particular, SDN-1 expressed in the hypodermis might regulate the diffusion and thus spatial distribution of secreted guidance cues. Alternatively, SDN-1 might function cell autonomously in the migrating neurons and navigating growth cones to modulate – possibly through its HS chains – the interaction of specific ligand/receptor pairs.

To distinguish between these two possibilities, we expressed a full-length sdn-1 cDNA under the control of either the hypodermal dpy-7promoter (Pdpy-7::sdn-1)(Gilleard et al., 1997) or the pan-neuronal unc-119 promoter(Maduro et al., 2000), and tested the rescuing activity of the different transgenes. Interestingly,neuron-specific expression of SDN-1, as well as expression of SDN-1 driven by the endogenous sdn-1 promoter (Psdn-1::sdn-1) rescued the HSN and ALM migration defects of sdn-1 null mutants(Table 2). By contrast,hypodermal expression of SDN-1 under control of the dpy-7 promoter had no significant effect on ALM migration, although it did weakly reduce the HSN migration defects (Table 2).

To determine whether cell type-specific expression of SDN-1 was sufficient to rescue the axon guidance defects of the corresponding neurons, we expressed SDN-1 under the PVQ-specific sra-6 promoter (Psra-6::sdn-1). PVQ-specific expression of SDN-1 significantly reduced the crossover defects of PVQ interneurons, with a similar efficiency as expression of SDN-1 under its endogenous promoter (Table 3). By contrast, expression of SDN-1 in hypodermal tissue(Pdpy-7::sdn-1) showed no rescuing activity. In summary, our data indicate that SDN-1 is likely to act cell autonomously in neurons to promote cell migration and axonal pathfinding.

We next investigated whether we could identify specific modification patterns of HS chains carried by SDN-1 that could confer to the axons the ability to selectively respond to midline guidance cues. To address this issue, we examined double-null mutants between sdn-1(zh20) and mutants in the HSmodifying enzymes. As the C5-epimerase HSE-5 and the 2Osulfotransferase HST-2 have previously been shown to act in the same pathway for midline guidance, we concentrated our analysis on HSE-5 and the 6O-sulfotransferase HST-6(Bülow and Hobert,2004).

For D-type motoneurons, we found that in animals lacking both HSE-5 and SDN-1 the outgrowth of commissures to the DNC was drastically reduced compared with that of either single mutant (Fig. 6A-D′). Moreover, hse-5; sdn-1 double null mutant L1 larvae showed a highly defasciculated VNC(Fig. 6C′) and most of the embryonic DD commissures failed to reach the DNC(Fig. 6A,B′). We observed a similar range of defects in L4 larvae and adult animals, where most of the DD and VD commissures were misguided and did not connect to the DNC(Fig. 6D′).

Similarly, sdn-1 hst-6 double-null mutants (L1 larvae and adults)showed more severe phenotypes in D-type motoneurons than did either single mutant alone, but the increase of defects was less pronounced than in hse-5; sdn-1 double-null worms(Fig. 6A, Fig. 4E). The strong enhancement of the guidance defects observed in hse-5; sdn-1 double mutants suggests that HSE-5 acts in parallel with SDN-1, i.e. that HSE-5 must modify at least one additional HSPG regulating a parallel signaling pathway. Because sdn-1 hst-6 double mutants show a detectable enhancement of guidance defects in D-type motoneurons when compared with either single mutant, SDN-1 may not be the only substrate of HST-6 in D-type motoneurons.

To expand these observations in a different cellular context, we also examined the pathfinding of PVQ interneurons. Again, hse-5; sdn-1double-null animals showed significantly more PVQ crossover defects than did either single mutant (Fig. 6E). In most hse-5; sdn-1 worms, midline crossing had already occurred in the posterior region, leading to completely collapsed PVQ tracts (>50% of crossover defects). In contrast to D-type motoneurons, loss of HST-6 function did not result in a statistically significant enhancement of the sdn-1 null phenotype. Interestingly, HST-6 is predominantly expressed in neurons, in contrast to HSE-5, which is mainly expressed in the hypodermis(Bülow and Hobert, 2004). This finding, together with our observation that loss of HST-6 function does not lead to more severe PVQ crossover defects in an sdn-1 null background, suggests that 6O-sulfated HS, generated by HST-6, is important for SDN-1 function in PVQ midline crossing. By contrast, in D-type motoneurons, both hse-5; sdn-1 and sdn-1 hst-6double-null mutants exhibited additive effects. Thus, signaling in different cellular contexts (D-type motoneurons versus PVQ) might depend on a different combination of HS modifications.

In summary, the epistasis analysis with HS-modifying enzyme mutants points to the existence of a parallel, SDN-1-independent midline guidance pathway modulated by a distinct HSPG core protein that is a major target of the sugar epimerase HSE-5. Although our results do not exclude the possibility that HSE-5 also modifies SDN-1, the fact that enhances the hse-5loss-of-function phenotype indicates that, even in the absence of C5-epimerization, SDN-1 still regulates axon guidance.

As the above experiment showed that syndecan SDN-1 is required for midline guidance, we speculated that SDN-1 might act as a co-receptor in one of the known axon guidance pathways, such as UNC-6/Netrin or Slit/Robo(Hao et al., 2001; Ishii et al., 1992; Zallen et al., 1998). The genome encodes one Robo receptor (SAX-3) and one Slit homolog (SLT-1), which are both required for ventral axon guidance(Hao et al., 2001). Furthermore, interneurons that are usually separated in the left and right VNC inappropriately cross the midline in mutants(Hao et al., 2001), a defect similar to the phenotype that we observed in mutants.

In order to test whether SDN-1 acts in the SLT-1-dependent signaling pathway, we used the presumptive null allele slt-1(eh15)(Hao et al., 2001) and compared the phenotype of slt-1 sdn-1 double-null mutants with that of either single mutant (Fig. 6E). The midline crossing defects in slt-1 sdn-1 double mutants were no more severe than those of either single mutant, suggesting that SDN-1 acts in the same genetic pathway as SLT-1. Likewise, slt-1 sdn-1 double mutants did not show enhanced defects of L/R choices of D-type motoneurons (data not shown).

The differential modification of HS sugar chains was recently found to be required for various aspects of axon guidance. This observation led to the development of the `sugar-code hypothesis'(Bülow and Hobert, 2004),which states that tissuespecific sugar side-chain modifications, like tissue-specific alternative splicing or tissue-specific heterodimerization,can modulate and fine-tune the activity of a core protein (in this case the HSPGs). In this study, we show that the HSPG core protein SDN-1 acts autonomously in neurons to direct cell migration and specify axon guidance decisions at the midline. We provide evidence that distinct modification patterns on SDN-1 and at least one additional HSPG core protein allow these proteins to differentially modulate two partially redundant axon guidance signaling pathways.

Loss of SDN-1 function interferes with the migration of the HSN, CAN and ALM neurons, which, together with Q neuroblasts, are the only groups of neurons that migrate long distances in C. elegans. Because similar ALM migration defects have been reported for mutations in sax-3/Robo(Zallen et al., 1999), SDN-1 might also modulate Slit/Robo signaling in cell migration, in addition to its role in the regulation of Slit signaling in midline guidance indicated by our data. However, lack of SDN-1 does not perturb cell migration in general; for example, sex myoblast migration is normal in sdn-1(zh20) animals(data not shown). Furthermore, sdn-1 null mutants exhibit no circumferential distal tip cell (DTC) migration defects. Mutations in the gene encoding perlecan/UNC-52, a basement membrane HSPG, enhance the DTC migration defects of UNC-6/netrin signaling mutants – an effect that can be partially suppressed by mutations disrupting growth factorlike signaling(Merz et al., 2003). Whether SDN-1 also contributes to signaling by EGL-20/WNT, UNC-129/TGF-β or EGL-17/FGF still needs to be determined.

Another candidate pathway that could be regulated by SDN-1 is signaling by integrins. C. elegans ina-1 _ integrin was shown to be required in neurons for the migration of HSN, ALM and CAN neurons(Baum and Garriga, 1997). Intriguingly, ina-1 mutants also show defects in coelomocyte migration, whereas sexmyoblast migration is normal, similar to the situation we found in sdn-1(zh20) null mutants. In addition, cell culture studies have demonstrated a crucial role of the syndecan extracellular domain in regulating the adhesion and invasion of tumor cells, possibly by interacting with integrins (Beauvais et al., 2004; Burbach et al.,2004). Understanding the role of syndecan in cell migration is particularly relevant in light of the role mammalian syndecans play during carcinogenesis. Upregulation of human syndecan 1 in breast and pancreatic cancer, for example, usually correlates with more invasive tumors, although,in other types of cancer, downregulation of syndecan seems to have equally adverse effects (Beauvais and Rapraeger,2004; Leivonen et al.,2004). In C. elegans, SDN-1 could promote cell migration either by modulating the activity of guidance cues or, directly, by regulating cell adhesion.

How might SDN-1 function in axon guidance? A recent study in Drosophila has shown that loss of syndecan function affects the extracellular distribution of the ligand Slit, which is secreted by midline cells and might therefore interfere with signal transduction by Robo receptors(Johnson et al., 2004). Neural expression of Drosophila syndecan rescued the guidance defects of syndecan mutant flies, whereas the expression of syndecan in midline cells did not (Johnson et al.,2004).

Similarly, our data suggest that C. elegans SDN-1 functions in neurons to ensure the correct midline guidance of axons. Cell type-specific expression of SDN-1 in PVQ interneurons partially rescued the midline pathfinding errors of these cells, whereas expression of SDN-1 in the surrounding epidermis failed to rescue. Incomplete rescue of PVQ crossovers,also seen for SDN-1 driven by its endogenous promoter, could be explained by insufficient expression of the rescuing arrays in PVQ neurons due to the lack of an enhancer sequence in the promoters that we used.

Neuronal SDN-1 could act as a co-receptor of specific axon guidance receptors and might thereby enhance the efficacy of the signaling event(quantitative modulation), alter the specificity of the ligand/receptor interaction (qualitative modulation), or both. One good candidate for a receptor that may require SDN-1 for ligand binding is the Slit/SLT-1 receptor SAX-3. sdn-1 mutants show similar pathfinding errors as slt-1 or sax-3 mutants, and genetic epistasis analysis placed sdn-1 in the same genetic pathway as slt-1 in certain cellular contexts (PVQ and D-type motoneuron development).

Alternatively, SDN-1, being a transmembrane protein, could also act directly as a receptor and generate an intracellular signal on its own. For example, Xenopus syndecan 2 participates in inside-out signaling that specifies the left-right looping of the heart and gut, which is mediated by PKC_ phosphorylation of the syndecan cytoplasmic domain(Kramer et al., 2002). Beyond their function as (co-)receptors, syndecans may also be involved in the targeting of guidance receptors to specific membrane compartments, such as lipid rafts (Couchman,2003).

The C terminus of mammalian syndecans has been shown to interact with the PDZ domain of the membrane-associated guanylate kinase CASK(Cohen et al., 1998; Hsueh et al., 1998). The C. elegans CASK homolog LIN-2 is required together with LIN-7 and LIN-10 for the basolateral localization of the EGF receptor on vulval precursor cells (Kaech et al.,1998). SDN-1 might therefore be required to anchor the LIN-2/LIN-7/LIN-10 complex to the basolateral plasma membrane and thereby localize the EGF/LET-23 receptor. Although SDN-1::GFP is present on the basolateral side of the primary vulval cells (data not shown), loss of SDN-1 had no effect on EGF receptor localization. However, these observations do not exclude a possible role for SDN-1 in the targeting of axon guidance receptors.

Interestingly, sdn-1(zh20) null animals showed comparable phenotypes for PVQ guidance and HSN migration to sdn-1(ok449)mutants, which express a mutant SDN-1 lacking the major HS side chains. This observation underlines the importance of the HS side chains and the information encoded in their sugar modifications for correct axon guidance,although it is possible that conformational changes due to the deleted region in the ectodomain or an altered distribution of the truncated SDN-1 form account for the defects observed in sdn-1(ok449) mutants. Importantly, we have found that SDN-1 is not the only HSPG controlling the axon guidance of PVQ interneurons and VNC motoneurons. Based on our epistasis analysis, we predict the existence of an additional axon guidance pathway,which relies on HS modifications by the C5-epimerase HSE-5, or on both HSE-5 and the 6Osulfotransferase HST-6 – depending on the cellular context (Fig. 7). As it was shown that HSE-5 acts in the epidermis(Bülow and Hobert, 2004),this parallel pathway may depend on C5-epimerized HS sugars on a, as yet to be identified, HSPG expressed by the surrounding epidermal cells.

C. elegans mutant for glypican GPN-1, the second major cell surface HSPG, do not seem to display axon guidance or cell migration defects(Hannes Bülow and Oliver Hobert, personal communication). In all events that we analyzed (PVQ and DD/VD guidance, HSN migration), gpn-1(ok377) failed to enhance the defects of sdn-1(zh20)null animals (data not shown), indicating that HSE-5 might modify an extracellular or as yet unknown cell surface HSPG. The UNC-6/Netrin and VAB-1/Ephrin guidance systems act in parallel with SLT-1/Slit signaling to define midline guidance (Zallen et al.,1999), and are thus candidate pathways to be regulated by such proposed epimerized sugar motifs on a HSPG core protein distinct from SDN-1.

In PVQ interneurons, lack of the 6O-sulfotransferase HST-6 in worms devoid of SDN-1 did not cause a significant enhancement of crossover defects when compared with sdn-1 single mutants. Thus, the HS side chains on SDN-1 are likely to be important substrates of HST-6 in PVQ(Fig. 7). HST-6 is, however,also involved in SDN-1-independent signaling, in the context of D-type motoneuron guidance.

Our results provide further evidence for the existence of a `sugar code' on cell-surface and extracellular HSPGs that regulates the differential responses of axons towards the various extracellular guidance cues that they normally encounter. For example, 6Osulfated HS side chains may act as`molecular antennae' on SDN-1 to enhance the perception of Slit signals, as well as a modulator for the ligand/receptor interaction, while the input from parallel guidance systems could be regulated by extracellular HSPGs carrying C5-epimerized HS chains, with or without 6O-sulfation(Fig. 7). In this way, the sensitivity of axons towards Slit signals could be dynamically regulated through the combinatorial control of HS-modifying enzyme expression and cell surface exposure of SDN-1. Indeed, during an early phase of mouse brain development, the different syndecan family members, as well as the HS-modifying enzymes, show tightly regulated spatiotemporal expression patterns (Ford-Perriss, 2003; Sedita et al., 2004).

The fact that HSPGs are highly conserved from nematodes to humans supports the idea that tissue-specific HS sugar modifications of HSPGs are a key factor that determines the cellular specificity of the different axon guidance systems (Hutter et al., 2000). Rules learned in C. elegans are thus likely to also apply in mammals,and promise to lead to a better understanding of the mechanisms that regulate nervous system development.

We thank Hannes Bülow and Oliver Hobert for stimulating discussions and sharing results prior to publication. We thank the CGC and various C. elegans researchers for providing strains and markers, the C. elegans Knockout Consortium in Oklahoma for providing deletion mutants,Eduardo Moreno and members of the Hengartner and Hajnal laboratories for comments on the manuscript, and Amit Dutt and Doris Klingele for technical support. This work was funded by the Swiss National Science Foundation (M.O.H. and A.H.), the Forschungskredit of the University of Zurich (C.R.) and the Ernst Hadorn Foundation (M.O.H.).