Slits are large secreted glycoproteins characterized by an unusual tandem of four LRR (leucine-rich repeat) domains in their N-terminal half. Slit proteins were initially described as repulsive guidance cues in neural development, but it has become clear that they have additional important functions, for instance in the vasculature and immune system. Genetic studies have identified two types of cellular receptors for Slits: Robos (Roundabout) and the HS (heparan sulphate) proteoglycan syndecan. The intracellular signalling cascade downstream of Robo activation is slowly being elucidated, but the mechanism of transmembrane signalling by Robo has remained obscure. No active signalling role for syndecan has yet been demonstrated. Slit–HS interactions may be important for shaping the presumed Slit gradient or presenting Slit at its target cell surface. Recent studies have mapped the binding sites for Robos and HS/heparin to discrete Slit domains. Robos bind to the second LRR domain of Slit, whereas HS/heparin binds with very high affinity to the C-terminal portion of Slit. Slit activity is likely to be modulated by physiological proteolytic cleavage in the region separating the Robo and HS/heparin-binding sites.
In the last decade, huge strides have been made in the elucidation of the signals that guide axons to their targets in the developing nervous system [1,2]. A large secreted glycoprotein, Slit, was identified as a major repellent at the midline of the central nervous system [3–6]. Binding of Slit to receptors of the Robo (Roundabout) family triggers cytoskeletal rearrangements within the axon growth cone, resulting in axon repulsion. This fundamental function of the Slit–Robo system is conserved between invertebrates and vertebrates. However, since the original discovery in 1999 of Slit as a Robo ligand, many other Robo-dependent Slit functions have been revealed, for instance in axon branching ; the migration of neurons [8–11], mesodermal cells , leucocytes  and endothelial cells ; development of the lung  and kidney ; inflammation [17,18]; (tumour) angiogenesis [14,19]; and tumour metastasis . To add further complexity, Robos also appear to have a Slit-independent function in homophilic cell adhesion . Finally, the HSPG [HS (heparan sulphate) proteoglycan] syndecan was recently identified as a co-receptor for Slit [22–24]. While the importance of Slit–Robo signalling for many biological processes is well established, our understanding of the molecular mechanisms involved is limited. The aim of this short review is to summarize recent structure–function studies into the Slit–Robo system, emphasizing the extracellular recognition events.
Molecular structure of Slits and Robos
Drosophila and Caenorhabditis elegans have a single Slit, whereas mammals have three Slit proteins (Slit1–Slit3) [3,4,25]. The defining feature of Slits is an unusual tandem of four LRR (leucine-rich repeat) domains at the N-terminus, followed by six EGF (epidermal growth factor-like) repeats, a laminin G-like β-sandwich domain, one (invertebrates) or three (vertebrates) EGF repeats, and a CT (C-terminal cystine knot) domain (Figure 1). The only other protein family containing multiple LRR domains are the Slitrk transmembrane proteins, which are believed to function in controlling neurite outgrowth . The Slit LRR domains consist of five to seven LRRs flanked by disulphide-rich cap segments. The crystal structure of the third LRR domain of Drosophila Slit has been determined and serves as a convenient template to model all other domains  (Figure 1). Each LRR contains a characteristic LXXLXLXXN motif that contributes one strand to the parallel β-sheet at the concave face of the curved LRR domain, whereas the convex back of the domain is made up of irregular turns and loops. The four LRR domains of Slit are connected by short linkers, with a disulphide bridge tethering each linker to the back of the preceding domain. This structural constraint is likely to dictate a fairly compact arrangement of the entire LRR region of Slit . Little is known about the folding of the C-terminal portion of Slit. Drosophila and mammalian Slit2 are proteolytically processed in vivo, with a single cleavage site located after the fifth EGF repeat; the N-terminal Slit fragment retains the biological activity [4,7,28]. The physiological significance, if any, of Slit processing is not understood .
Regarding the Slit receptors, Drosophila has three Robos (Robo, Robo2 and Robo3) [29–33], C. elegans has a single Robo (Sax-3) , and mammals have four Robos (Robo1–Robo4) [4,35,36]. With the exception of mammalian Robo4, the Slit-binding ectodomains of all Robos consist of five Ig-like domains, followed by three FN3 (fibronectin type 3) domains (Figure 1). Robo4 has only two Ig-like domains and two FN3 domains. The cytosolic domains of Robos are large and have no discernible domain organization; several short conserved sequence motifs have been identified, however, and shown to be important for Robo function [37–41].
Slits bind to Robos with apparent dissociation constants of approx. 10 nM [3,4], but it should be noted that these values have been determined under conditions where avidity contributions may be substantial (binding at cell surfaces, artificially dimerized constructs). The Slit–Robo interaction is evolutionarily conserved: Drosophila Slit can bind mammalian Robos, and vice versa . Several studies have shown that the LRR domains of Slits are necessary and sufficient for Robo binding and biological activity in vitro [28,42,43]. We carried out a detailed structure–function study and mapped the binding site for all three Drosophila Robos to a highly conserved patch on the concave face of the second LRR domain of Slit  (Figure 1); we recently confirmed that the same patch is involved in the interaction of the mammalian proteins (J.A. Howitt and E. Hohenester, unpublished work). Regarding the corresponding Slit-binding site on Robos, there is substantial evidence that the first two Ig-like domains are important for Slit binding and Robo function: targeted deletion of Ig1 (Ig-like domain 1) from Robo1 results in abnormal lung development in mice ; an antibody against human Robo1 Ig1 blocks Slit2-dependent tumour angiogenesis ; and deletion of Ig1 or Ig2 (Ig-like domain 2) from human Robo1 abrogates Slit2 binding . However, none of these results provide positive evidence for a direct interaction between Slits and Ig1 of Robos. Using SPR (surface plasmon resonance), we have recently been able to demonstrate a direct interaction of the second LRR domain of Drosophila Slit and the Ig1–Ig2 pair of Drosophila Robo, whereas Ig-like domains 3–5 did not interact with Slit (J.A. Howitt and E. Hohenester, unpublished work). The dissociation constant of the minimal Slit–Robo complex is in the micromolar range, suggesting that the interaction between the full-length proteins is significantly enhanced by oligomerization of Slit and/or Robo.
How binding of Slit to the Robo ectodomain is converted into an intracellular signal is completely unknown. Slit binding could result in Robo clustering, disrupt a preformed Robo oligomer, or lead to conformational changes in Robo without affecting the oligomeric status. The fact that Robo activation currently can only be monitored by observing complex biological responses, such as changes in neurite outgrowth or chemotaxis, represents a major obstacle for future structure–function studies.
The special case of Robo4
Robo4 (or Magic Roundabout) is distinguished from all other Robos not only by its shorter ectodomain, but also by its unique expression in the vascular endothelium . Both in vitro and in vivo studies have demonstrated a role for Robo4 in endothelial cell migration and angiogenesis [45–47], but many questions remain unanswered. A controversial issue is whether Slits are ligands of Robo4 [45,46]. Using SPR, we could demonstrate a weak interaction between the second LRR domain of human Slit2 and the Ig-like pair of Robo4 (J.A. Howitt and E. Hohenester, unpublished work), but whether Slits are the (only) biological ligands of Robo4 remains to be determined.
Recent genetic studies have provided compelling evidence that HSPGs are critically involved in Slit–Robo signalling. Ablation of enzymes involved in HS biosynthesis in C. elegans, zebrafish and mice results in multiple defects, some of which resemble those of Slit- and Robo-deficient animals [48–51]. A specific requirement for HS in Slit–Robo signalling was demonstrated by analysing genetic interactions between HS biosynthetic enzymes on the one hand, and the Slit–Robo pathway on the other hand . These experiments did not reveal which HSPGs are critical for Slit–Robo signalling, however. A first clue came from the observation that the expression patterns of Robo and the HSPG syndecan overlap in the Drosophila nervous system [22,23]. Genetic experiments in Drosophila and C. elegans indeed demonstrated that syndecan has to be present on the same axon as Robo for Slit–Robo signalling to occur normally [22–24]. Syndecans are dimeric transmembrane proteins and constitute one of two families of cell surface HSPGs, the other being the glycosylphosphatidylinositol-anchored glypicans . Although disruption of syndecan alone leads to axon guidance defects, it appears that there is functional redundancy between syndecans and glypicans: neural overexpression of the Drosophila glypican Dallylike significantly rescues the midline phenotype of the syndecan mutant .
How do (syndecan) HS chains regulate Slit–Robo signalling? Because syndecan was found to function autonomously in neurons, two plausible scenarios can be formulated, which are not mutually exclusive. HS chains may either be required for capturing Slit at the cell surface of the Robo-expressing growth cone, or they may be required for formation of a specific ternary Slit–Robo–HS signalling complex (similar to the situation in fibroblast growth factor signalling ). Immunoprecipitation experiments using Drosophila cell extracts show that both Slit and Robo interact with syndecan . Another study demonstrated a specific HS-dependent interaction between human Slit2 and glypican-1, with the C-terminal proteolytic Slit2 fragment binding glypican more tightly than the N-terminal, biologically active fragment . We recently confirmed these results using a panel of recombinant Drosophila Slit fragments (S. Hussain, J.A. Howitt and E. Hohenester, unpublished work). We found that a highly conserved basic sequence motif in the CT domain is responsible for high-affinity HS/heparin binding (>1 M NaCl required for elution from a heparin column), but the LRR domains also show appreciable HS/heparin binding. Thus it is conceivable that HS has a dual role in modulating Slit–Robo signalling: the C-terminal high-affinity HS-binding site may serve to concentrate Slit at the growth cone surface, while a weaker, but potentially more specific site in the LRR region may be required for the formation of a ternary Slit–Robo–HS signalling complex. Physiological proteolytic cleavage of Slit would separate these activities and could have a profound effect on Slit activity in vivo.
Another pertinent question relates to the nature of Slit gradients in vivo. At the midline, Slit is expressed by glia cells and proposed to diffuse away from the midline, thus setting up the gradient required for long-range guidance. Direct observation of such gradients is challenging, but a recent study may have succeeded in imaging the Slit gradient and how it is altered in the absence of syndecan . Given the very high affinity of the Slit–HS interaction, it is not easy to see how Slit could diffuse freely through the extracellular matrix. Another puzzling observation is that, following proteolytic cleavage, it is the N-terminal and not the C-terminal fragment (which has higher affinity for HS) that remains associated with cell surfaces . Perhaps the combined ligation by Robos and syndecan/glypican effectively traps Slit at the cell surface.
Although a structural understanding of Slit–Robo signalling is still a distant prospect, the recent identification of minimal interacting domains [27,44] has shown that these large proteins can be dissected for crystallographic studies. Crystal structures of minimal Slit–Robo and Slit–HS complexes are likely to give important clues about the signalling mechanism. However, in order to understand the full sequence of events at the cell surface, the structural studies will have to be complemented by biochemical and microscopic approaches. A major aim must be to develop a simpler ‘read-out’ of Robo activation, without which structure–function studies will remain challenging. The first decade of research into Slit–Robo signalling has been dominated by genetic approaches; now is the time for structural biologists to make an impact.
Cytokine–Proteoglycan Interactions: Biology and Structure: Biochemical Society Focused Meeting held at Royal Holloway University of London, Egham Hill, U.K., 9–10 January 2006. Organized and edited by B. Mulloy (NIBSC, U.K.) and C. Rider (Royal Holloway University of London, U.K.).
We gratefully acknowledge funding by the Wellcome Trust. E.H. is a Wellcome Senior Research Fellow.