Drosophila Slit and its vertebrate orthologues Slit1–Slit3 are secreted glycoproteins that play important roles in the development of the nervous system and other organs. Human Slits are also involved in a number of pathological situations, such as cancer and inflammation. Slits exert their effects by activating receptors of the Robo (Roundabout) family, which resemble cell adhesion molecules in their ectodomains and have large, mainly unstructured cytosolic domains. HS (heparan sulfate) is required for Slit–Robo signalling. The hallmark of Slit proteins is a tandem of four LRR (leucine-rich repeat) domains, which mediate binding to the IG (immunoglobulin-like) domains of Robos. A major question is how Slit binding is translated into the recruitment of effector molecules to the cytosolic domain of Robo. Detailed structure–function studies have shown that the second LRR domain of Slit (D2) binds to the first two IG domains of Robo, and that HS serves to stabilize the Slit–Robo interaction and is required for biological activity of Slit D2. Very recently, the crystal structure of a minimal Slit–Robo complex revealed that the IG1 domain of Robo is bound by the concave face of Slit D2, confirming earlier mutagenesis data. To define the mechanism of Robo transmembrane signalling, these structural insights will have to be complemented by new cell biology and microscopy approaches.
- axon guidance
- heparan sulfate
- immunoglobulin-like domain
- leucine-rich repeat
Slit was first identified in Drosophila as a protein secreted by a subset of glia cells along the midline of the developing CNS (central nervous system) . Mutations in the slit gene result in a collapse of the regular ladder-like arrangement of longitudinal and commissural (crossing) axon tracts. This phenotype suggested that Slit is somehow involved in interactions between midline glia cells and commissural axons. Genetic screening uncovered further genes whose mutation resulted in severely perturbed commissures, one of which, roundabout (robo), encodes a transmembrane protein expressed on the axon growth cones [2,3]. The cell-surface expression of Drosophila Robo is regulated by the product of the commissureless gene . These findings were unified in a molecular mechanism, when it was realized that Slit is a ligand for Robo receptors and that Robo signalling results in repulsion of axons away from the midline in both invertebrates and vertebrates [5–7]. Since this seminal discovery, Slit–Robo signalling has been widely studied in the developing nervous system, and the fundamental mechanisms of axon guidance at the CNS midline are now firmly established [8,9]. Slit and Robo function is not confined to the nervous system, however. The vertebrate proteins have additional important roles in the development of the lung, kidney and mammary gland, and have been implicated in human pathology (e.g. cancer and inflammation) [10–12]. The intracellular consequences of Robo activation are only partially understood, but good progress has been made in identifying molecules that link Slit–Robo signalling to cytoskeletal dynamics and protein synthesis [13,14]. The purpose of this brief review is to discuss the Slit–Robo system from a structural biologist's perspective. I will summarize what is known about the structure of Slit and Robo, the molecular details of their interaction and the mechanism of transmembrane signalling by Robo.
Invertebrates have a single Slit, whereas vertebrates have three homologous proteins: Slit1–Slit3 . The hallmark of Slits is a unique tandem of four LRR (leucine-rich repeat) domains at the N-terminus, termed D1–D4. This large (∼900 residues) region is followed by six EGF (epidermal growth factor)-like domains, a laminin G-like domain, one (invertebrates) or three (vertebrates) EGF-like domains, and a C-terminal cystine knot domain (Figure 1). Crystal structures of D2 and D3 have revealed the architecture of the Slit LRR domains [15–17]. Each LRR domain consists of an N-terminal cysteine-rich cap, an array of five to seven LRRs, and a C-terminal cysteine-rich cap. Each LRR contains a central LXXLXLXXN motif that contributes one β-strand to the curved sheet that makes up the concave face of the LRR domain (Figure 1). The four LRR domains of Slit are connected by short linkers containing a cysteine residue that forms a disulfide bridge with a cysteine residue at the convex back of the preceding domain. These tethers are not compatible with a fully extended conformation of the D1–D4 region of Slit, but whether they impose a defined higher-order structure remains to be seen. We previously suggested that the D1–D4 region may be dimerized through interactions made by the D4 domain ; however, a definitive biophysical study has shown that D1–D4 is a monomer in solution (C. Baldock and E. Hohenester, unpublished work). Drosophila Slit and mammalian Slit2 are processed in vivo and in cell culture by proteolytic cleavage after the fifth EGF-like domain . Studies in vitro have shown that the N-terminal Slit fragment retains full biological activity as a Robo ligand, whereas the C-terminal fragment is inactive [18,19]. Whether the C-terminal Slit region is dispensable for Slit function in vivo is not known.
Caenorhabditis elegans has a single Robo (Sax-3), Drosophila has three Robos (Robo, Robo2 and Robo3) and vertebrates have four Robos (Robo1/Dutt1, Robo2, Robo3/Rig-1, Robo4/Magic Roundabout). At the Drosophila midline, all three Robos transduce a repellent signal, and the combinatorial expression of the three Robos specifies the lateral positioning of longitudinal axon tracts. At the vertebrate midline, Robo3 antagonizes the repellent signal from Robo1 and Robo2, and thus is functionally analogous to the Commissureless protein in Drosophila . All Robos, with the exception of Robo4 (a receptor expressed in the vasculature and not the CNS ), share the same ectodomain architecture reminiscent of CAMs (cell adhesion molecules), consisting of five IG (immunoglobulin-like) domains followed by three FN3 (fibronectin type 3) repeats (Figure 1). Robo4 has only two IG and two FN3 domains. Crystal structures are available of IG1–IG2 of human Robo1  and of Drosophila Robo (N. Fukuhara, J.A. Howitt and E. Hohenester, unpublished work). They show two canonical I-set IG domains in a relatively extended arrangement, but with evidence of substantial flexing at the domain junction (Figure 1). The cytosolic Robo domains are poorly conserved, with the notable exception of several conserved linear motifs, CC0–CC3, which occur in different combinations in different Robos . Intriguingly, the Robo cytosolic domains are predicted to contain little regular secondary structure and may thus belong to the class of intrinsically unstructured protein regions. If indeed lacking a compact structure, the ∼500 cytosolic residues of Robos would occupy a very large volume of space and be able to interact simultaneously with many adaptor molecules (see below).
Slit binding has been demonstrated biochemically for all Robos, expect for Robo4. Early studies using transfected cells showed that the Slit–Robo interaction is of high apparent affinity (dissociation constant Kd of ∼10 nM) and evolutionarily conserved: human Slit2 binds to cells expressing Drosophila Robo, and Drosophila Slit binds to cells expressing rat Robo1 or Robo2 [5,7,19]. To study this interaction in more detail, we developed a solid-phase assay. Using a panel of recombinant Drosophila Slit proteins, we showed that all three Drosophila Robos bind to a common site in Slit D2 . Analysis of sequence conservation between invertebrate and vertebrate Slits revealed that the concave face, but not the convex face, of D2 is highly conserved, and site-directed mutagenesis confirmed that the concave face of Slit D2 indeed harbours the common binding site for all Drosophila Robos (Drosophila Slit residues Tyr402, Leu424, Tyr450 and His472). As expected from sequence conservation, vertebrate Slits also have their Robo-binding sites in D2 [17,21]. Importantly, we have shown that human Slit2 D2 is able to induce the collapse of Xenopus retinal axon growth cones in vitro, demonstrating that the minimal Robo-binding domain is biologically active .
The corresponding Slit-binding site on Robo was mapped using domain-deletion constructs of human Robo1 and Robo2 . Deletion of IG1 or IG2 abolished the interaction of human Slit2, whereas the IG3–IG5 region was not required for binding; that the FN3 domains are not required had been shown previously . Morlot et al.  recently reported a Kd of 8 nM, obtained by SPR (surface plasmon resonance) analysis, for the interaction of human Robo1 IG1–IG2 with human Slit2 D2. Also using SPR, we obtained a much lower Kd of ∼20 μM for the interaction of the corresponding Drosophila proteins, Robo IG1–IG2 and Slit D2 (N. Fukuhara, J.A. Howitt and E. Hohenester, unpublished work). The reason for the ∼1000-fold discrepancy in dissociation constants is unclear at present, but is unlikely to be the result of a genuine difference between the two species, given the high degree of conservation of Slit and Robo proteins. The stoichiometry of the minimal Slit–Robo complex is 1:1, as determined by size-exclusion chromatography of Drosophila Robo IG1–IG2 and Slit D2 .
An important advance was achieved recently with the structure determination of a 1:1 complex between human Slit2 D2 and Robo1 IG1  (Figure 1). The membrane-distal IG1 domain of Robo1 is bound by the concave face of Slit2 D2, as predicted, and all Slit residues demonstrated previously to be critical for Robo binding indeed make interactions with Robo (Slit2 D2 residues Tyr356, Leu378, Tyr404 and His426). In addition, a conserved patch of charged residues near the N-terminal cap of Slit2 D2 is also involved in Robo1 binding. In total, complex formation buries ∼1400 Å2 (1 Å=0.1 nm) of solvent-accessible surface area in a bipartite interface. Although not as conspicuously conserved as their counterparts on Slit2 D2, the major Slit-binding residues of Robo1 IG1 are conserved in other Robos (except in Robo4, as expected), suggesting that the mode of binding is largely conserved in other pairings. Apolar Slit-binding residues (Thr86, Met120, Leu122, Phe128 and Leu130) are contributed by β-strands B, D and E of Robo1, and polar Slit-binding residues (Glu72, Asn88 and Lys90) by β-strands A and B′. The C-terminus of Robo IG1 in the complex is distant from the interface with Slit2 D2, and the structure thus does not explain why deletion of IG2 had such a dramatic effect on Slit2 binding . Deletion of IG2 may have compromised the folding of IG1, or residues in Robo IG2 may indeed be required for Slit binding; this question should be addressed by further structure analysis or mutagenesis.
HS (heparan sulfate) and Slit–Robo signalling
Substantial evidence has been accumulated that Slit–Robo signalling strictly requires HS [24,25]. HS chains consist of repeating sulfated disaccharide units that are attached to secreted or membrane-associated core proteins to form HSPGs (HS proteoglycans). Slit–Robo signalling in vitro can be abolished by enzymatic digestion of cell-surface HS or by addition of excess heparin (a more highly and uniformly sulfated variant of HS) [26,27]. Slit–Robo signalling in vivo can be abolished by genetic ablation of enzymes involved in HS biosynthesis [28–30] or HSPG core proteins [31,32]. A requirement for HS has emerged as a common feature of many signalling systems, including morphogens, growth factors and guidance cues . Of particular relevance to this review, HS/heparin has been shown to be required for signalling by Netrin and Frazzled/DCC (deleted in colorectal cancer), another ligand–receptor pair crucial for CNS midline patterning  (see below).
Results from genetic experiments in Drosophila suggested that the HSPG syndecan regulates both the distribution and efficiency of Slit . Syndecans have not been implicated in the regulation of mammalian Slits. Rather, biochemical studies have identified the HSPG glypican-1 as a high-affinity receptor for human Slit2 and have shown that both natural cleavage products of Slit2 are recognized by glypican-1 . Consistent with these results, we mapped HS/heparin-binding sites to two separate regions of Drosophila Slit: the LRR domains D1 and D2, and the C-terminal cystine knot domain . HS binding by the C-terminal domain of Slit is unlikely to account for the strict HS-dependence of Slit–Robo signalling, given that the N-terminal, but not the C-terminal, Slit fragment is biologically active in vitro [18,19]. HS binding by Slit D2 is of great interest, as this domain harbours the Robo-binding site of all Slits . We found that Drosophila Slit D2 and Robo IG1–IG2 associate with a heparin-derived oligosaccharide in solution to form a ternary complex of 1:1:1 stoichiometry. This minimal complex appears to recapitulate the signalling complex at the axon cell membrane: residues in a conserved basic patch in the C-terminal cap of human Slit2 D2 (Lys461, Arg462, Lys466, Arg467, Lys472 and Lys475) are required not only for heparin binding, but also for HS-dependent Slit2 D2-induced collapse of Xenopus retinal axon growth cones in vitro .
The structure of the ternary Slit–Robo–HS signalling complex is unknown at present. In the crystal structure of the binary complex of human Slit2 D2 and Robo1 IG1 , the HS-binding site of Slit2 D2 remains fully exposed and formation of a ternary complex is thus sterically feasible. Robo residues involved in HS/heparin binding have not been identified but are likely to exist, given that Drosophila IG1–IG2 binds heparin quite avidly . The putative HS-binding site of Robo might involve a patch of conserved basic residues centred on the E-F loop of IG1 (Lys81, Arg131, Arg136 and Lys137 in human Robo1), which in the binary complex is situated ∼35 Å from the HS-binding site of Slit2 D2. At least five HS disaccharide units would be required to contact both Slit D2 and Robo IG1 simultaneously, which intriguingly corresponds to the minimal HS size capable of disrupting Xenopus retinal axon targeting .
Another interesting question relates to the specificity of Slit–HS (and possibly Robo–HS) interactions. Unlike heparin, where the sugar backbone is highly and relatively uniformly modified by sulfation, HS is heterogeneous: regions of low sulfation alternate with highly sulfated heparin-like stretches. To determine which HS sequences support Slit binding and Slit–Robo signalling, Shipp and Hsieh-Wilson  used microarrays of heparin derivatives. Human Slit2 binding to heparin was strongly dependent on 6-O-sulfation and N-sulfation, whereas 2-O-sulfation was less important. These preferences were validated in functional assays using brain explants, in which the various heparin derivatives were tested for their ability to inhibit axon guidance and neuronal cell migration .
Mechanism of Robo transmembrane signalling
A major unresolved question is how Slit binding to the membrane-distal IG1 domain of Robo is transmitted across the cell membrane to initiate downstream signalling. Other transmembrane receptors react to ligand binding either by conformational changes within a stable oligomeric assembly (e.g. integrins ), by ligand-induced dimerization (e.g. receptor tyrosine kinases ) or by clustering at points of cell–cell or cell–matrix contact (e.g. immune receptors ). Given that Robo ectodomains resemble CAMs and Robos indeed mediate homophilic binding , Slit–Robo signalling may well involve a change in oligomeric status of Robo.
The cytosolic domains of Robos are catalytically inactive. Abl (Abelson tyrosine kinase) is associated with Robo  and plays an important role in mediating the interaction of Robo with N-cadherin , but it is not believed that the activity of Abl (or another kinase stably associated with Robo) accounts for all of the downstream effects of Robo activation . Rather, it is thought that several catalytically inactive adaptors bind to the cytosolic domain of Robo and that these interactions are somehow regulated by Slit. srGAP1 (Slit–Robo Rho GTPase-activating protein 1) has been shown to bind to the proline-rich CC3 motif of mammalian Robo1 and mediate the Slit-dependent inactivation of the Rho family GTPase, Cdc42 (cell division cycle 42) . The adaptor protein Dock/Nck also binds to CC3 (as well as to CC2) and mediates the Slit-dependent recruitment of SOS (Son of sevenless), which in turn activates another Rho GTPase, Rac [45,46]. These examples illustrate how Slit–Robo signalling could alter cytoskeletal dynamics through regulation of Rho family members. Inhibitor studies have also implicated PI3K (phosphoinositide 3-kinase), MAPKs (mitogen-activated protein kinases) and localized protein synthesis [27,47], but the adaptor molecules linking Robo activation to these pathways have yet to be identified.
There are intriguing mechanistic parallels between Slit–Robo signalling on the one hand and Netrin–DCC signalling on the other hand. Netrin can be considered to be the attractive counterpart of the repellent Slit at the CNS midline, but the two proteins are structurally unrelated (Netrin is homologous with the N-terminus of laminin chains) . In contrast, DCC shares with Robo a CAM-like ectodomain and also has a large cytoplasmic domain that appears to contain little regular secondary structure. A genetic study in Drosophila using chimaeric receptors showed that the signalling outcome is specified by the cytosolic domain: a Robo–DCC chimaera (Robo ectodomain–DCC cytosolic domain) is an attractive Slit receptor and a DCC–Robo chimaera is a repulsive netrin receptor . This important result implies a common mechanism of transmembrane signalling for Robo and DCC. Furthermore, the cytosolic domains of the two receptors have been shown to interact physically, and Slit-activated Robo has been suggested to silence the attractive Netrin–DCC signal once an axon has reached the CNS midline .
How does binding of Slit to Robo (or Netrin to DCC) alter the accessibility of short linear motifs, so that the various adaptor molecules can gain access to these sequences? In the case of DCC, it appears that the cytosolic domain has an inherent tendency to oligomerize (via a conserved motif termed P3), but oligomerization is repressed in the unstimulated receptor . Similarly, the isolated cytosolic domain of Robo has been shown to be constitutively active . These findings favour a mechanism whereby the cytoplasmic domains of DCC and Robo are held in an inactive state in the absence of ligand. One way in which this could be achieved is through lateral association of the inactive receptors, plausibly mediated by their CAM-like domains. If ligand binding were incompatible with this mode of self-association, activation would result from the disruption of receptor dimers or higher oligomers, thereby freeing up the cytosolic domains for adaptor binding and downstream signalling (Figure 2). Alternatively, receptors could become oligomerized as a result of ligand binding, as has been suggested for DCC . One conceptual problem with the latter scenario is that it is not easy to see how the cytosolic domain would be prevented from oligomerizing in the absence of ligand. Clearly, knowledge of the receptors' oligomeric status (with and without ligand) would help in unravelling the mechanism of transmembrane signalling, but this information is not easy to obtain experimentally. A combination of structural, biochemical and microscopy techniques will have to be deployed to make any real progress in this area.
Frustratingly, the recently determined Slit and Robo structures have not shed much light on the signalling mechanism. With regard to the first model (ligand-induced disruption of receptor dimers/oligomers), no consistent mode of Robo self-association can be discerned from four independent structures of IG1–IG2 of human and Drosophila Robo  (N. Fukuhara, J.A. Howitt and E. Hohenester, unpublished work). With regard to the second model, the 1:1 stoichiometry of the minimal Slit–Robo complex [17,22] seemingly rules out a model invoking ligand-induced receptor oligomerization. However, multiple Slit–Robo complexes could be assembled on a single HS chain, and this might provide an explanation for the strict HS dependence of Slit–Robo signalling. To resolve this question, further biophysical and structural studies of ternary complexes should be pursued with urgency. Finally, it would be invaluable to know more about the structure of the cytosolic domain of Robo, but this domain represents a formidable challenge owing to its large content of non-regular structure.
I apologize to all authors whose work I was unable to cite because of space restrictions. Research in my laboratory is supported by the Wellcome Trust. E.H. is a Wellcome Senior Research Fellow.
Structure and Function in Cell Adhesion: Biochemical Society Annual Symposium No. 75 held at The Palace Hotel, Manchester, U.K., 3–5 December 2007. Organized and Edited by David Garrod (Manchester, U.K.).
Abbreviations: Abl, Abelson tyrosine kinase; CAM, cell adhesion molecule; CNS, central nervous system; DCC, deleted in colorectal cancer; EGF, epidermal growth factor; FN3, fibronectin type 3; HS, heparan sulfate; HSPG, HS proteoglycan; IG, immunoglobulin-like; LRR, leucine-rich repeat; Robo, Roundabout; SOS, Son of sevenless; SPR, surface plasmon resonance
- © The Authors Journal compilation © 2008 Biochemical Society