Introduction
Biologists have long been impressed by the ability of the nervous system to wire itself into a functional unit. At the single-cell level, a neuron must generate an axonal process and extend that process over significant distances. Along the way, the axon will often make several precise changes in direction and encounter different physical and biochemical environments. The path that an axon chooses to take is presumably orchestrated by a coordinated presentation of extracellular cues by intermediate and final targets. These cues are interpreted by axonally localized receptors, adhesion molecules and their downstream effectors1. The integration of these cues occurs in the growth cone (the motile end of the axon) and usually results in one final directional output. With some of the key guidance factors and receptors now known, research is just beginning to explore how these signal transduction and adhesion systems function.
Simple models of axon guidance propose that axons change direction because they are challenged with a new extracellular cue or cues. Given the complexity of an axon's journey, it would seem that the more cues an axon can draw on, the better. New data, however, suggest that at certain times, the ability to ignore cues is vital to accurate guidance of the growth cone. At the molecular level, such cunning ignorance can be achieved by allowing one guidance pathway to inhibit the signal transduction of another. Work from Rhee et al.2 in this issue of Nature Cell Biology describes convincing evidence for a biochemical inhibition of the cell adhesion molecule N-Cadherin by the guidance receptor Robo. These new findings, together with earlier work on Robo, make for tempting speculation on how cross-pathway inhibitions observed in vitro could regulate axon guidance in vivo3. We will outline potential uses of cross-pathway inhibition in axon guidance, review the known functions of Robo and hypothesize how the discovered cross-pathway inhibition mediated by Robo adds to our mechanistic understanding of axon guidance.
The most obvious use for cross-pathway inhibition is the prevention of signal 'gridlock'. Gridlock refers to what would occur if an axon were receiving and transducing opposing signals simultaneously. These signals would work against each other, paralysing the forward progress of the axon. However, if one signal inhibits the signal transduction of the other, this paralysis is prevented and the growth cone only uses the dominant cue to guide it. There are at least three applications for cross-pathway inhibition: to suppress a pathway that has outlived its usefulness (shutdown), to prevent an axon from straying between cues (handoff) and to suppress a pathway until the axon accomplishes other goals first (holdback). We will first give a brief description of the Robo pathway and then illustrate where these mechanisms might function.
In vivo, Robo receptors have an evolutionarily conserved function in the midline, one of the most extensively studied axon guidance choicepoints4. The majority of CNS axons cross the midline at some point in their journey. Netrin is a midline-secreted factor that initially attracts the axon to the midline. After reaching the midline, axons continue their progression on the opposite side of the embryo, repelled from the midline and never venturing back across it. This repulsion is achieved by Robo receptors responding to their ligand, midline-secreted Slit. In Drosophila melanogaster, recent work has shown that the Slit–Robo system also has another function5, 6, 7. Many axons continue their journey after crossing the midline by making a sharp turn and joining one of many longitudinal axon pathways that are already tracking parallel to the midline. The discovery of homophilic cell adhesion molecules such as connectin and Fasciclin II (Fas II), which label specific subsets of these pathways), suggested that they, and presumably other molecules, might function as the cue used by axons to make their turn. Those axons with FasII on their surface join or bundle with the other axons that express FasII. One problem with this simple model is the fact that there are three major Fas II-positive bundles located at medial, intermediate and lateral positions, relative to the midline. If FasII is the only cue axons use, they should all join the medial bundle, which they encounter first after crossing the midline. Recent work has shown that axons joining these bundles express a unique combination of the three Drosophila Robo receptors and suggests a 'Robo code' for lateral positioning. Axons with fewer Robo-family receptors join the medial pathway because they can tolerate high concentrations of Slit. Those axons with a high degree of Robo signalling are pushed furthest from the midline and join the lateral pathway. Axons with an intermediate amount of Robo signalling join the intermediate fascicle where Slit levels are moderate. In this model, the long-range Slit–Robo system roughly targets an axon to its longitudinal neighbourhood and then short-range (that is, membrane-bound) cues target the axon to a specific fascicle5, 6, 7.
How could cross-pathway inhibition mediated by Robo help us understand its role in vivo? The transition from initial attraction to repulsion from the midline is a prime example of why axons need anti-gridlock mechanisms. Previous work has shown that Netrin-guided axons arriving at the midline begin to respond to the repulsive effects of Slit and are repelled into the other side of the midline4. But why is the axon not paralysed by the opposing forces of Slit and Netrin? The answer seems to be that Robo can directly antagonize the Netrin receptor, deleted in colorectal cancer (DCC), through receptor–receptor interactions3 (Fig. 1). This is an example of signal 'shutdown', as Robo is inhibiting a pathway that has lost its importance to the cell. It is also an example of 'handoff', because the attraction to netrin becomes a repulsion for Slit, with no gap in between. Without such a mechanism, the embryo would have to precisely coordinate three things: the downregulation of the Netrin pathway, the upregulation of the Slit pathway and the simultaneous arrival of all axons at the midline. Cross-pathway inhibition allows axons to function independently from each other and keeps them with a target in their sights.
Figure 1: Cross-pathway inhibition prevents gridlock at the midline.
The left column shows the growth cone relative to the midline, the middle column shows the state of Robo and DCC signalling within the growth cone, and the right column shows the supposed forces generated by these signals within the growth cone. Netrin and Slit are both ligands that are secreted from the midline. The axon of cell A is attracted to the midline through Netrin's stimulation of DCC (top). The axon of cell B has reached the midline and transduces both the Slit and Netrin signals because it now expresses Robo on its surface (middle). This cell has no cross-pathway inhibition and so suffers gridlock as the forces generated in the growth cone antagonize each other. The axon of cell C can resolve this conflict of forces by a Slit-dependent inhibition of the Netrin–DCC pathway (bottom). Only the Robo–Slit signal is propagated within the growth cone and a repellent response is generated.
Full size image (72 KB)After axons cross the midline, they use the positional information encoded by Slit and interpreted by their Robo code to find their place in the longitudinal tracts5, 6, 7. Given the work with Fas II, a 'holdback' form of cross-pathway inhibition would be an appealing explanation of how local cues are prevented from prematurely engaging axons. Robo could inhibit the sensing of local cues until the axon is far enough away from Slit to reduce its Robo signalling. This would relieve Robo-mediated inhibition of short-range cues and allow the appropriate longitudinal tract to be joined. Although no biochemical link has been found between Robo and Fas II, the work from Rhee et al.2 uncovers a biochemical inhibition between Robo and another well-known short-range cue, N-cadherin.
N-cadherin is a homophilic cell adhesion molecule that regulates oriented axon outgrowth both in vivo and in vitro8, 9. The work of Rhee et al. begins with the observation that activation of the Robo signalling cascade caused a specific decrease in N-cadherin-mediated adhesion2. The authors begin to fill in the molecular gaps by first examining the role of N-cadherin-associated proteins in the Robo response. In the adhesive state, N-cadherin is bound to 
-catenin, which is itself bound to 
-catenin, which in turn binds F-actin10. This bridge from N-cadherin to the actin cytoskeleton seems to be necessary for N-cadherin-mediated adhesion, as loss of the interaction between N-cadherin and -catenin causes loss of adhesion11, 12. Rhee et al. found that after activation of Robo, N-cadherin no longer immunoprecipitated -catenin2. This loss of interaction is coincident with an observed increase in tyrosine-phosphorylated -catenin. The authors go on to show that the tyrosine kinase Abelson (Abl) is constitutively associated with Robo and is the most probable candidate for phosphorylating -catenin. Activation of Robo results in the formation of a complex consisting of Robo, Abl and N-cadherin, which may facilitate the tyrosine phosphorylation of -catenin2. Although this is an in vitro study, there is existing evidence for some of these interactions in vivo. For example, in Drosophila, Abl interacts genetically with -catenin and DE-Cadherin in various morphogenetic contexts12, 13. Work in Drosophila has also suggested a genetic interaction for Abl and Robo, but one in which Abl inhibits the Robo pathway14. The work from Rhee et al. predicts that there will also be a positive role for Abl in Robo signalling, consistent with more recent studies15.
What part might cross-pathway inhibition of N-cadherin by Robo play in axon guidance? One intriguing possibility is that it cooperates with the Robo–Slit pathway in the positioning of longitudinal pathways. The original 'Robo-code' model suggested that axons are first targeted by the Robo–Slit pathway and then local adhesion cues, such as Fas II, guide the axon to a final position5, 6, 7. This model, however, does not address the issue of gridlock. How is it that axons targeted to the lateral tract by their Robo signalling are not paralysed by the positive local adhesion cues in the medial and intermediate pathways that they must traverse? One possibility is that activation of Robo causes cross-pathway inhibition of all the local adhesion cues by inactivating each one individually. A much more attractive hypothesis is that Robo inhibits a common factor that is necessary for the axon to migrate in a longitudinal pathway, while the local adhesion cues determine the specificity of pathway selection. Drosophila N-Cadherin (DN-Cadherin) is an excellent candidate for just such a common factor. It is found throughout the embryonic nervous system and when mutated, it causes defects in longitudinal axon extension9. Cues such as Fas II could be responsible for the initial targeting and maintenance of axon fasciculation, whereas DN-cadherin could be the link to the cytoskeleton that would drive extension of the axon. Thus, the inhibition of DN-Cadherin by Robo would make the axon unable to follow any longitudinal pathway cues until the proper level of Robo–Slit signalling for that axon was achieved (Fig. 2). This fits well with the reported phenotype of Fas II mutants, whose axons still extend longitudinally, but fasciculate inappropriately with other axons16. Whether the mechanism uncovered by Rhee et al.2 is involved in a holdback scenario such as this, or some other form of cross-pathway inhibition, remains to be seen. It suggests, however, that the interactions between pathways may be as important as the pathways themselves in creating a properly wired brain.
Figure 2: One possible in vivo role for the Robo-mediated inhibition of N-cadherin.
A model for how three guidance cues– Slit, N-cadherin, and other adhesion molecules act in concert to target longitudinal axons. Axons cross the midline and do two things: first, they continue to travel various distances away from the midline; second, they choose one out of many longitudinal fascicles to join. Axon A receives a high level of Robo signalling and so moves down the Slit gradient to position 3 before choosing one of the fascicles in the area. Here, it chooses the X fascicle through its X-receptor. Axon B has a low level of Robo signalling and so can tolerate the high concentration of Slit near the midline. In common with axon B, it also chooses the X fascicle, but from this more medial cohort of fascicles. Combining the work from Rhee and colleagues with earlier work on longitudinal axon positioning suggests the signalling states shown above. At point 1, N-cadherin-mediated adhesion is disabled in axon A as a result of Slit-mediated activation of Robo. This is mediated by the formation of a complex between Robo, Abl and N-cadherin and phosphorylation of -catenin by Abl, which blocks it's interaction with N-cadherin. Although axon A can adhere to the X fascicle through the X-receptor, it needs N-cadherin to actually exert a physical force within the growth cone. Thus, signalling from Robo prevents gridlock through inactivation of N-cadherin and continues to repel the axon further from the midline. However, axon B at point 1 is not very responsive to Slit, as it has only a small amount of Robo signalling. Thus, inhibition of -catenin is relieved. A functional N-cadherin allows for axon B to use the X cue and turn into this fascicle. Axon A must wait until point 3 for it's Robo signalling to be low enough to achieve the same goal.
