Sugar Codes for Axons?

Abstract

potential for extraordinary diversity. Several recent studies have demonstrated important roles for heparan sulfate and chondroitin sulfate proteoglycans (HSPGs and CSPGs) in axon pathfinding and have linked HSPGs to specific signaling pathways. More speculatively, there are hints of a “sugar code,” in which specific sugar modifications might act instructively in guidance decisions. This raises the intriguing possibility that the complexity of neuronal wiring may in part reflect the complexity of proteoglycan modifications.

The first hints that heparan sulfate proteoglycans (HSPGs) might have important functions in axon guidance came with the demonstration that enzymatic removal of heparan sulfate causes severe pathfinding defects by pioneering axons in the developing cockroach limb bud (Wang and Denburg, 1992). Subsequent studies also revealed roles for HS in axon navigation in the vertebrate visual system (Halfter, 1993; Irie et al., 2002; Walz et al., 1997). Over this same period, evidence began to accumulate suggesting that chondroitin sulfate proteoglycans (CSPGs) also influence neurite outgrowth in vitro (Snow et al., 1991) and contribute to axon pathfinding in vivo (Brittis et al., 1992; Walz et al., 2002). Now, a recent spate of studies in a variety of different organisms—mice, fish, flies, and worms—has provided compelling genetic evidence that proteoglycans are, indeed, key players in axon guidance in vivo and offered the first intriguing insights into the molecular pathways in which they act. This minireview will focus mainly on HSPGs, but will also briefly discuss recent genetic evidence showing that CSPGs may play a similarly complex and important role in axon guidance.

HSPG Biosynthesis

HS is a member of the glycosaminoglycan family (GAG) of macromolecules, which are linear chains of repeating disaccharides. Complex modifications to this disaccharide backbone give rise to specific patterns of sulfation and epimerization (Figure 1), and the final chain consists of highly modified regions, S domains, spaced apart by relatively unmodified regions, N domains. There are a large number of possible modified disaccharides, and because there is no template (unlike nucleic acids) and the reaction does not go to completion, the potential for generating molecular diversity is enormous (Turnbull et al., 2001).

At least 14 biochemical steps occur in the synthesis of HS chains and their linkage to core proteins. One of the key early steps is the polymerization of the precursor saccharide backbone that consists of alternating nonsulfated N-acetylglucosamine and D-glucuronic acid repeats (-GlcNAc-GlcA-). This is catalysed by HS copolymerases such as the Exostosin (Ext) family of glycosyltransferases (e.g., EXT1). The unmodified chain is then acted upon sequentially by a series of modification enzymes. N-deacetylase/N-sulphotransferases replace some acyl groups with sulfate groups, epimerases convert some glucuronic acid units to the isomeric iduronic acid, and sulfotransferases add sulfates to specific residues (3-O-, 2-O-, and 6-O-sulfotransferases). Interestingly, with the recent discovery of a class of evolutionarily conserved cell surface HS 6-O endosulfatases that desulfate cell surface HS (Dhoot et al., 2001; Wang et al., 2004), it is becoming increasingly clear that ligand signaling can be additionally modulated in a dynamic way, even after HS biosynthesis.

HSs bind to a large variety of proteins in the extracellular matrix and on the cell surface and can potentiate the activities of secreted proteins. For example, FGF function is thought to be modulated by HS acting either to protect it from degradation, to increase its range of diffusion and local concentration, or to present it to its high-affinity receptors (Ornitz et al., 1995; Schlessinger et al., 2000; Spivak-Kroizman et al., 1994; Yayon et al., 1991). HSs also bind to the chemotropic axon guidance molecule Slit (Liang et al., 1999; Ronca et al., 2001), and removal of cell surface HS abolishes Slit-Robo interactions and causes a loss of Slit-induced repulsion (Hu, 2001). Although long considered the poor cousin of protein-protein interactions in terms of specificity and strength of binding affinity, some protein-sugar interactions are turning out to have a remarkably high degree of specificity and affinity (Turnbull et al., 2001).

Mice Lacking Ext1 in the CNS

HSPGs are widely expressed in the embryonic brain, particularly in developing axon tracts, where the major cell surface carriers of HS are the transmembrane syndecans and the GPI-anchored glypicans (Yamaguchi, 2001). To assess the role of these neural HSPGs, Yamaguchi and colleagues asked what happens when HS is removed entirely from the developing brain (Inatani et al., 2003). For this, they generated a conditional knockout of the mouse Ext1 gene, which encodes the major HS glycosyltransferase, and crossed these mice to nestin-Cre mice to obtain CNS-specific disruption of Ext1 (Inatani et al., 2003). These mice, referred to as Nes-Ext1 mice, die on the first day of life and biochemical, in situ, and immunocytochemical analyses of the embryonic brains (E9-18) confirm that syndecan-3 is not glycanated with HS and that the entire CNS lacks HS. These Nes-Ext1 mice have several severe CNS defects, the most conspicuous being the loss of olfactory bulbs, an abnormally small cerebral cortex, malformation of the caudal midbrain-cerebellum region (loss of inferior colliculus and cerebellum), and an absence of commissural tracts. Some of these defects can potentially be explained by a disruption in FGF signaling, for example, in patterning of the midbrain-hindbrain boundary and in the proliferation of cortical neurons.

The three major commissures in the forebrain—the corpus callosum, hippocampal commissure, and the anterior commissure (AC)—are absent in Nes-Ext1 mice. The axons that normally comprise these commissures follow aberrant trajectories before reaching the midline. For example, axons of the AC immediately head ventrally in Nes-Ext1 mice instead of making a turn toward the midline. Retinal axons also show axon guidance defects, but in this case axons make errors after crossing the midline: instead of entering the contralateral optic tract, they grow aberrantly into the contralateral optic nerve. This guidance phenotype is strikingly similar to that of Slit1/Slit2 double-knockout mice. Slits are expressed around the chiasm and act repulsively to confine the trajectory of retinal axons to a specific route across the midline (Plump et al., 2002). Indeed dosage-sensitive genetic interactions between Ext1 and Slit2 support the notion that HS and Slit together help to guide retinal axon growth from the chiasm into the optic tract. There are, however, notable differences between the retinal axon guidance phenotypes in Slit1/Slit2 and Nes-Ext1 mutants: whereas aberrant pathfinding occurs after the midline chiasm in Nes-Ext1 mutants, defects additionally occur at the chiasm in Slit1/Slit2 mutants. This suggests that HS is differentially required for Slit function and that, in some parts of the retinal pathway, Slit-guided axon growth occurs independent of HS.

It is not clear yet how HS interacts with Slit in the visual system. One possibility is that HS increases the stability and hence local concentration of Slits in the diencephalon. Examination of the distribution of Slit proteins in this region in Nes-Ext1 mice may reveal alterations, as occurs for example in syndecan mutants in Drosophila (Johnson et al., 2004). An additional possibility is that HSPGs themselves act as receptors or coreceptors (Park et al., 2000).

HSPGs Contribute to Axon Pathfinding and Sorting in the Zebrafish Optic Tract

Further genetic evidence for HS function in vertebrate retinal axon guidance has come from the genetic analysis of Ext mutants in zebrafish. A large-scale zebrafish screen yielded a number of retinal axon pathfinding mutants, two of which—dackel (dak) and boxer (box)— turn out to disrupt the Ext family genes Ext2 and Extl3, respectively (Lee et al., 2004). As in the Nes-Ext1 mice, HS levels are dramatically reduced in both the dak and box single mutants. However, the phenotypes of these dak and box mutant fish have some intriguing differences to the Nes-Ext1 mice. First, unlike the Nes-Ext1 mice, early brain patterning appears normal in both the dak and box fish, possibly due to the maternal supply of HSPGs and their biosynthetic enzymes. Second, after crossing the midline, retinal axons are not diverted into the optic nerve in dak and box fish, as they are in Nes-Ext1 mice, but instead continue normally into the contralateral optic tract. Within the optic tract, however, things go seriously wrong. Normally, retinal axons are topographically sorted, so that axons from the dorsal retina enter the tectum via the ventral branch, and axons from the ventral retina enter via the dorsal branch. This sorting fails in dak and box mutants, with dorsal axons often entering the tectum through the dorsal rather than the ventral branch. Remarkably, these axons still end up projecting to their normal targets once they enter the tectum, consistent with earlier embryological experiments showing that axons taking aberrant paths to the tectum still map topographically.

These optic tract sorting defects are not seen in astray/robo2 mutant zebrafish, which instead have a much more dramatic phenotype in which retinal axons leave the optic tract and project anteriorly or posteriorly (Fricke et al., 2001). A similar phenotype is also reported upon morpholino knockdown of slit1a (Lee et al., 2004). However, while the dak and box single mutants lack these defects, the dak box double mutant is a remarkable phenocopy of the astray/robo2 mutant. Thus, a more severe disruption of HS synthesis seems to impair Slit-Robo signaling in fish, just as it does in the Nes-Ext1 mice.

HSPGs at the Drosophila Midline

Another genetic approach to studying HSPG function in nervous system development is to eliminate the HSPG core proteins, rather than the enzymes involved in HS biosynthesis. The four major HSPG core proteins are perlecan and agrin (extracellular matrix), syndecans (transmembrane), and glypicans (GPI-anchored). Recent studies of the single Drosophila syndecan gene (sdc) have provided further evidence for HSPG function in axon guidance, again in the context of Slit-Robo signaling (Johnson et al., 2004; Steigemann et al., 2004). In the Drosophila ventral nerve cord, Slit provided by midline cells is thought to repel longitudinal axons that express Robo family receptors. Sdc is expressed along all axon pathways—both longitudinal and commissural—and loss of sdc function results in aberrant midline crossing by longitudinal axons. The phenotype in these null sdc mutants resembles a mild slit or robo phenotype, implying that Slit-Robo signaling is less effective but not completely abolished.

Johnson et al. (2004) show that Sdc is able to interact with both Slit and Robo and that the distribution of Slit is dramatically altered in sdc null mutant embryos. Slit is normally present at high levels on midline cells, which secrete it, and at low levels on axons. However, in sdc mutants the axonal staining for Slit is greatly reduced, whereas the midline staining for Slit is still present. This implies that Sdc is not required for the production or secretion of Slit, but for its localization on axons. Consistent with this model, both Johnson et al. (2004) and Steigemann et al. (2004) find that midline crossing defects in sdc mutants can be rescued by restoring sdc expression just in neurons. Furthermore, expressing the glypican gene dally in neurons is also sufficient to rescue the crossing defects in sdc mutants (Johnson et al., 2004), arguing against models in which HSPG core proteins would impart any specificity, for example, through their capacity to transduce signals intracellularly. The simplest model to explain these data is that certain HS side chains are required to help localize Slit and present it to Robo, but they may be attached to any membrane bound core protein.

HSPGs and CSPGs Modulate Semaphorin Function

These genetic studies in mice, fish, and flies have clearly established a role for HSPGs in Slit signaling. But this is evidently not the only guidance cue regulated by proteoglycans. A recent study in the developing rat brain has provided compelling evidence that proteoglycans also interact with semaphorin5A (Sema5A) during the formation of the fasciculus retroflexus (FR), a diencephalic axon tract associated with limbic function (Kantor et al., 2004). Sema5A binds to the glycosaminoglycan (GAG) portion of proteoglycans via its thrombospondin repeats and exerts an attractive or repulsive effect on the FR axons, depending on the class of GAG (CS or HS) present. Remarkably, HSPGs on the surface of growing FR axons mediate Sema5A-attraction while CSPGs in the surrounding environment cause Sema5A to elicit repulsion. Thus, CS bound and HS bound Sema5A act in opposite ways on axon growth, revealing that the precise composition of local GAGs may play a key role in the directional choices of axons.

A Sugar Code?

In the Drosophila studies, syndecan can be functionally replaced by a glypican (Johnson et al., 2004). This result argues against any specificity in the HSPG core proteins, at least in the context of midline repulsion by Slit, but does not exclude an important role for specificity in the HS modifications themselves. Indeed, given that HSPGs have only a limited set of core proteins but an enormous diversity of HS modifications, it seems far more likely that specific functions might be encoded in the HS modifications rather than in the core protein. A comprehensive study of mutations in each of the HS biosynthetic enzymes in C. elegans provides a first hint that this may indeed be the case (Bülow and Hobert, 2004).

C. elegans has two genes for Ext family heparan polymerases, rib-1 and rib-2, and single genes for each of the modifying enzymes: hst-1 encodes the N-deacetylase/N-sulfotransferase, hse-5 the C5 epimerase, and hst-2, hst-3, and hst-6 the 2O-, 3O- and 6O-sulfotransferases, respectively. rib-2 mutants have severe developmental defects, indicating essential and diverse requirements for HS biosynthesis (Morio et al., 2003). In contrast, mutants lacking any of the HS-modifying enzymes encoded by hse-5, hst-2, or hst-6 are viable and fertile, but have very specific defects in nervous system development (Bülow and Hobert, 2004). Careful examination of a number of different neurons in each of these mutants revealed an intriguingly complex role for HS modifications in axon pathfinding. For example, some axons require all three modifying enzymes for their guidance, others require only hse-5 or hst-2, and many require none of these enzymes at all. Moreover, the same neuron could even require different HS modifications for different guidance decisions.

Again, one of the signaling pathways that is critically dependent on HS is the Slit-Robo system. Some of the guidance defects seen in the HS modification mutants resembled those seen in slt-1/slit mutants and sax-3/robo mutants. Indeed, a careful analysis of various double-mutant combinations suggested that Slit-Robo signaling is also dependent upon HSPGs in the worm, just as it is in mice, fish, and flies. Moreover, since they had mutants in each of the HS modifying enzymes, Bülow and colleagues could ask which modifications are required for Slit-Robo signaling—something that was not possible for any of these other organisms. Remarkably, even in the context of this one guidance system, a complex pattern of differential requirements emerged: some guidance decisions that require both slt-1 and sax-3 also require hse-5 and hst-6 (evidently acting in the same pathway), whereas others are independent of these HS modifications.

To explain these observations, Bülow et al. proposed the existence of a “heparan sulfate code” for neural development. According to this hypothesis, HS modifications are introduced in a region-specific manner in the developing nervous system. By facilitating or inhibiting various ligand-receptor interactions in a highly localized fashion, this “HS code” would add an additional level of complexity on top of that already generated by the combinatorial expression of ligands and their cognate receptors. With the enormous diversity that can potentially exist in HS chains, this code could even provide a level of complexity that far exceeds that offered by the differential expression of other signaling molecules.

The HS code hypothesis predicts that the various HS modifications should also be differentially localized, rather than uniform and ubiquitous, and that they should act instructively rather than permissively. Bülow and colleagues already offer some indirect evidence for the differential expression and requirement of HS modifications. For example, they show that hst-6 acts primarily in neurons, whereas hse-5 and hst-2 appear to act in hypodermal cells. Discriminating between permissive and instructive roles will however require more challenging experiments in which aspects of the code are not just eliminated, but changed in more complex ways with predictable outcomes. For example, if Slit-Robo signaling is indeed facilitated by C5-epimerization and 6O-sulfation, as the loss-of-function studies indicate, then one would predict that localized ectopic expression of hse-5 and hst-6 might induce a response to Slit from an axon that would normally ignore it.

Future Challenges

Understanding the roles of HSPGs in axon pathfinding will require new reagents to probe and manipulate HS modifications. Unfortunately, it will not be possible to manipulate the HS code with the same precision with which a protein code can be manipulated. For example, misexpression of a certain HS modifying enzyme does not guarantee that a particular modification will occur, and, perhaps more importantly, that other aspects of a proposed code will not be inadvertently obliterated. Nevertheless, such experiments will be a necessary first step in testing the idea that HS modifications are not just permissive but can also be instructive. This is an essential aspect of any “code.” Ultimately, a fuller understanding of any HS code will require a means to determine which HS modifications are present on which core proteins at what time and place, and then to find out how these modifications impinge upon the relevant guidance systems. This is an imposing task, but the rewards could be tremendous. Despite the great progress that has been made in recent years in identifying and characterizing several well-conserved guidance cues and their receptors, it is still unclear how much of the complexity of neuronal wiring can be explained by these relatively few systems. At least mathematically, HS sidechains have sufficient molecular diversity to account for even the most complex of wiring patterns. It will be fascinating to learn how much of this potential is utilized.

Figure 1. Generation of Sequence Diversity in Heparan Sulfate Chain Biosynthesis.

The first step in HS biosynthesis involves the creation of a tetrasaccharide linkage region to a serine in the core protein (not shown). The addition of N-acetylglucosamine (GlcNAc) then commits the chain to the HS synthesis pathway. See text for further details. GlcA, glucuronic acid; IdoA, iduronic acid; GlcNS, N-sulfated glucosamine. (After Turnbull et al., 2001).