Mechanisms by which molecules guide axons

The ability of growing axons to find their way has captivated biologists throughout this century. Numerous experiments with vertebrate and invertebrate embryos have suggested that axons navigate by recognizing local cues, i.e. signals present on cell surfaces, in extracellular matrices, or &sing from nearby sources. Such signals are currently thought to consist of molecules to which axons are specialized to respond. Currently, numerous candidates for axon guidance molecules have been proposed and include extracellular matrix proteins, integral membrane proteins, and several classes of secreted proteins, some of which can also associate with cell surfaces or extracellular matrices. Much about the structure and expression of these ‘potential axon guidance molecules’ has been reviewed elsewhere (e.g. Jessell, Neuron 1988, 13-13; Sanes, Annu Rev Neurosci 1989,12:491-516; Iander, Trends Neurosci 1989, 12:189-195; Schachner, Ciba Found S’p 1989, 145:156-169). Such molecules are typically expressed at times and locations in vivo where growing axons encounter them, and display biological activities appropriate to the task of guiding axons. Generally, these biological activites have been-studied in tissue culture systems, although genetic approaches are also being used now (Hedgecock et al., Neuron 199G, 2:61-85; Elkins et al., Cell 1990, 60:56>575). In vitro observetions that suggest that a molecule could guide axons include: growth of neurites (the in vitro counterparts of axons and dendrites) towards or away from a source of the molecule; accurate tracking of neurites along a substratum path to which the molecule has been absorbed; acceleration or inhibition of neurite outgrowth in response to the molecule; spectic adhesion of neurites to cells or other neurites that bear the molecule. Establishing such in vitro effects has, in the case of some molecules, required the development of ingenious assays. As even more sophisticated tests are devised, the number of molecules considered to be potential axon guidance cues will no doubt increase. Yet, the fact that the current list of these molecules is already rather substantial has prompted some investigators to consider what mechanisms underlie the actions of these molecules.


Introduction
The ability of growing axons to find their way has captivated biologists throughout this century. Numerous experiments with vertebrate and invertebrate embryos have suggested that axons navigate by recognizing local cues, i.e. signals present on cell surfaces, in extracellular matrices, or &sing from nearby sources. Such signals are currently thought to consist of molecules to which axons are specialized to respond. Currently, numerous candidates for axon guidance molecules have been proposed and include extracellular matrix proteins, integral membrane proteins, and several classes of secreted proteins, some of which can also associate with cell surfaces or extracellular matrices. Much about the structure and expression of these 'potential axon guidance molecules' has been reviewed elsewhere (e.g. Jessell, Neuron 1988, 13-13;Sanes, Annu Rev Neurosci 1989,12:491-516;Iander, Trends Neurosci 1989, 12:189-195;Schachner, Ciba Found S'p 1989, 145:156-169). Such molecules are typically expressed at times and locations in vivo where growing axons encounter them, and display biological activities appropriate to the task of guiding axons. Generally, these biological activites have been-studied in tissue culture systems, although genetic approaches are also being used now (Hedgecock et al., Neuron 199G, 2:61-85;Elkins et al., Cell 1990, 60:56>575). In vitro observetions that suggest that a molecule could guide axons include: growth of neurites (the in vitro counterparts of axons and dendrites) towards or away from a source of the molecule; accurate tracking of neurites along a substratum path to which the molecule has been absorbed; acceleration or inhibition of neurite outgrowth in response to the molecule; spectic adhesion of neurites to cells or other neurites that bear the molecule. Establishing such in vitro effects has, in the case of some molecules, required the development of ingenious assays. As even more sophisticated tests are devised, the number of molecules considered to be potential axon guidance cues will no doubt increase. Yet, the fact that the current list of these molecules is already rather substantial has prompted some investigators to consider what mechanisms underlie the actions of these molecules. Interest in this question has been growing in recent years. One probable reason for the interest is the increased at-tention that ceU biologists are paying to the structure and function of growth cones, the motile 0rganeUes that steer axons (e.g. Smith, Science 1988, 242:708-715;Mitchison and Kirschner, Neuron 1988,1:761-772;Bray, Curr @in Cell Biol 1989, 1:87-90;Lankford et al, CLOY Opin Cell Bioll990, 80-85). A second reason is the growing evidence that axon guidance molecules are structurally complex (e.g. possessing multiple kinds of binding sites), and variable (e.g. alternatively spliced and/or modified), which suggests that their molecular interactions are not simple (Lander, 1989). A third reason is the growing suspicion that old assumptions about the mechanism of action of some of these molecules are in need of revision. A fourth, and particularly important reason, is that most of the known putative axon guidance molecules are not unique to the nervous system, but are encountered in a wide variety of normal and neoplastic tissues; accordingly, it is reasonable to suspect that the cellular mechanisms by which molecules guide axons are merely examples of the mechanisms by which cells, in general, respond to morphogenetic signals. The following review presents some of the issues that have been raised by investigations into the mechanisms by which molecules guide axons. Most of these invest&a tions have focused on cell-surface and extracellular matrix molecules, and the discussion below also emphasizes these 'surface-bound' molecules, a partial list of which is given in Table 1. For the sake of brevity, these proteins are sometimes referred to in the following paragraphs as 'axon guidance molecules'. The reader should keep in mind, however, that 'potential axon guidance molecules' would be a more accurate term, because none of these molecules has yet been proved to act as a guidance cue in viva.

The dynamic behavior of growth cones
The responsibility for steering axon growth falls almost entirely on growth cones Located at the tip of growing axons, these structures receive the signals that guide growth. They are also the primary locus of growth, movement and shape change in elongating axons, insofar as they are the site of motility, secretion, and assembly of organeues and macromolecules into the nascent axon.
The nature of this mechanism can only be guessed at based on what is known about growth cone structure and function. At the very least, the fact that the only access of growth cones to the genetic and protein synthetic apparatus of neurons is via a pathway along which communication (by retrograde and anterograde transport of material along the axon) is relatively slow, suggests that transcriptional and translational events are unlikely to participate in the decision-making functions of growth cones.
Growth cones display a rather modular architecturea central base from which variable numbers of motile projections, the lilopodia and lamellipodia extend. It has been argued (Mitchison and Kirschner, 1988) that the leading, motile regions of growth cones possess a distinctive type of cytoplasm (recently termed 'kinetoplasm'), that is fundamentally different from cytoplasm elsewhere in the axon. The kinetoplasm is highly dynamic, possessing a cytoskeleton rich in bundles and meshworks of actin &rnents that appear to be in constant motion. The axonal cytoplasm (axoplasm), in contrast, contains bundled microtubules, and is apparently much more stable.
The relative sharpness of the transition zone between kinetoplasm and axoplasm suggests that the two states of cytoplasm are mutually exclusive (Smith, 1988). Nevertheless, continual conversion of kinetoplasm into stable axoplasm (a process referred to as consolidation or maturation) appears to be a fundamental step in the process of axon elongation (Goldberg and Burmeister, J Cell Biol1986, 103:1921-1931. Interestingly, consolidation has been observed to be a cyclical process. At first, the lamellipodia and filopodia of a growth cone extend, move about, and retract in similar patterns of seemingly random activity. Then, a small region of growth cone, containing perhaps a single lamellipodium with associated lilopodia, becomes 'stabilized, i.e. it fails to retract. Next, consolidation of the stabilized region occurs. Finally, other regions of the growth cone resorb, while new kinetoplasm forms at the leading edge of the newly consolidated structure (Goldberg and Burmeister, 1986).
This picture of growth cone behaviour implies that the direction in which an axon grows is continuously determined by the region of the growth cone where stabilization and consolidation take place. Accordingly, for a growth cone to be guided towards an immobilized extracellular guidance cue, the contact of a structure such as a filopodium or lamellipodium with that cue must somehow result in the consolidation of that structure becoming more probable. In addition, some sort of crosstalk must occur among growth-cone structural elements to explain why, usually, only one part of a growth cone consolidates at a time, or why the encounters of one or a few lilopodia can sometimes trigger global events such as the collapse of the entire growth cone (Kapfhammer and Raper, J Neuroxi 1987, 7:201-212).
Thus, two major tasks in elucidating the mechanisms by which molecules guide axons are to understand how events at the plasma membrane act locally to influence the fate of the underlying kinetoplasm, and to identify intracellular interactions that coordinate the activity of growth-cone components. As discussed below, current hypotheses tend to differ primarily in the degree to which these processes are thought to be mediated via mechani-caJ as opposed to chemical means.

The role of adhesion in axon guidance
Long before individual molecules that influence axon guidance were identilied, it was appreciated that growth cones recognize and respond to the physical topography of their environment. Bumps, troughs, obstacles, etc.
can all guide neurite growth in vitro (cf. Hollerbeck and Bray, Annu Rev Cell Biol 1988, 4:43-61). In the mid-1970s Letoumeau (Dev Bioll975, 44:92-100) extended these Iindings to include a physicochemical feature of the environment: 'adhesiveness'. Using plastic surfaces onto which patterns of molecules such as polyornithine and palladium were deposited, he demonstrated that growth cones, when faced with a choice between regions of the substratum treated with dissimilar molecules, invariably chose the more adhesive (in this context, adhesion is taken to refer to the resistance of growth cones or cells to detachment, by force, from the substratum).
Bray and his colleagues provided a compelling explanation for this phenomenon by pointing out that the structural elements of growth cones (Iilopodia, In the late 1970s and early 198Os, several ceU culturederived 'factors' were described that promoted neurite extension when adsorbed to a tissue culture substratum (see Lander, Mol Neurobioll987, 1:213-245 for discussion). When these substances were applied in distinct patterns, growth cones restricted their growth to those patterns (Collins and Lee, J Neurosci 1984,4:2823-2829), just as they did in Letoumeau's experiments when confronted with patterns of adhesive molecules. When the active component of the neurite extension factors was identified as the extracellular matrix protein laminin, a protein capable of mediating ceU attachment, it seemed logical to assume that laminin acted upon growth cones by increasing their adhesion to the substratum. Only recently has this assumption been tested, and the results have been surprising.
Both Adler et al. (Dev Biol 1985, 112:100-114) and Hall et al (J Cell Biol 1987, 104:623-634), for example, pointed out that retinal neuron attachment to laminin was relatively weak, compared with other molecules that did not possess the same ability as laminin to stimulate or guide neurite outgrowth. Gundersen (Dell Biol 1987, 121:423-431) measured growth cone adhesion directly by comparing the degree to which growth cones grow on substrata containing certain extracellular matrix molecules with the degree of growth cone adhesion shown towards those same molecules. Growth cones were observed to choose substrata containing laminin over other molecules or combinations of molecules that were more, less, or equally as adhesive as laminin. A subsequent study, using interference reflectance microscopy, indicated that the number of close growth cone-substratum appositions formed on laminin substrata was equal to or less than the number formed on other substrata (Gundersen, J Neurosci Res 1988, 21:298--306); this result further supported the idea that laminin is not unusually adhesive with regard to growth cones.
These data can be interpreted in two ways. One possibility is that current methods of measuring adhesion, such as ceU attachment, growth cone distraction and interference reflectance microscopy, do not accurately reflect the degree of adhesion 'sensed' by the growth cone itself. The other possibility is that, sometimes, mechanisms other than adhesion are involved in the response of growth cones to substratum-associated molecules. Both interpretations are defensible.

Adhesion: a time-dependent process
If growth cones 'sense' adhesion by pulling against the substratum and measuring the tension produced, it is useful to consider which factors influence that measurement. One such factor is the amount of contractile force produced by the growth cone. It is currently suspected that force is produced by the action of a plasmalemmal motor (e.g. myosin I) on the bundles of actin Iilaments which are so abundant in filopodia and lamellipodia (Ivlitchison and Kirschner, 1988). A second factor is the position at which the contractile force is applied; moving actin bundles appear to be localized beneath the membranes of lilopodia and lamellipodia (Smith, 1988) [ 1 ] and some observations suggest that lilopodia bear much of the contractile load at their tips (Tsui et al, Proc Nat1 Acad Sci U! 1985, 82382568260). A third factor is the time period over which contractile force is applied.
The potential importance of timing merits special consideration. Because filopodia and lamellipodia undergo a repeating cycle of extension, contact with the substratum and retraction, contraction against the substratum occurs discontinuously. The typical length of this cycle suggests that most of the force exerted by Ulopodia and lamellipodia is exerted against substratum contacts that are rarely more than a few minutes old (Bray and Chapman, J Neumsci 1985, 12:3204-3213). This raises an interesting question: are the measurements of growth cone adhesion that are currently made in the laboratory (most of which measure adhesion mediated by relatively long-lived contacts) good predictors of the level of adhesion experienced by growth cone structures pulling against 'fresh' contacts? The clever work contact and extracellular matrix of McClay and his colleagues [2] stron@y suggests not. These investigators have demonstrated that the initial adhesion experienced between a cell and its substratum and the same adhesion experienced many minutes later can differ by more than an order of magnitude [ 21. Moreover, the proiile of adhesion development over time can be strikingly different for contacts medi ated by different substratum-bound molecules. For exam ple, in the case of cells that strengthen their fibronectinmediated contacts more than lo-fold over 15min at 37°C contacts mediated by tenascin actually decline in strength over the same time course [2]. Large timedependent decreases in adhesive strength have also been observed when cerebeUar neurons were allowed to attach to the tenascin-related nervous-system glycoproteins Jl-160/180, but not when the same cells were permitted to attach to laminin [ 31. Thus, any attempt to order a set of substratum-associated molecules according to their adhesiveness might produce very different results, depending on when and how adhesion measurements are made.
In view of these considerations, direct measurements of the changes in the magnitude of adhesion over time at transient lilopodial and lamellipodial contacts will need to be made before the role of adhesion and tension in axon guidance can be properly tested. In principle, such measurements are possible, but new methods of measurement may first need to be devised. Until then, one can only speculate about how adhesion develops on the basis of current knowledge. Initial adhesion seems to be an energy-independent process, the extent of which may simply reflect the net energy of the binding of substratum molecules to the receptors that they happen to encounter. In contrast, changes in adhesion that occur over time seem to result from energy-dependent processes which involve the actin cytoskeleton [ 21; at least one of these processes is likely to be the active rect-uitment of receptors from other membrane locations to the site of contact. Thus, the extent to which the magnitude of lilopodial and lamellipodial adhesion exceeds the magnitude of initial adhesion may depend on whether or not ceU surface molecules are actively translocated in these structures, and if they are, at what rates they are moved. Notably, recent observations indicate that active translocation of cell-surface proteins, including neural ceU adhesion molecule (NCAM), occurs in growth cones, both centripetally and centrifugally, at rates of more than one micron per second [I 1.

The role of intracellular signalling in axon guidance
The idea that the interplay between substratum adhesion and tension controls axon guidance is an attractive hypothesis, but not the only one that is plausible. Growth cones are clearly capable of chemotaxis, i.e. orientated growth towards sources of diffusible, soluble molecules, against which they cannot pull (e.g. Gundersen and Barret, J Cell Biol 1980, 87:546554). Growth cones are also able to adhere strongly without necessarily orientating themselves. For example, when growth cones of retinal neurons contact the neurites of sympathetic neurons, tightly adherent Ulopodial contacts often form, even though the result of the contact is the inhibition of neurite growth, not the promotion of orientated growth (Kapfhammer and Raper, 1987).
For both of these examples, there are reasons to think that chemical second messengers have a role in mediating growth cone responses. Chemotaxis, for example, is generally thought to result from receptor-mediated signalling events (Devreotes and Zigmond, Annu Rev Cell Biol1988,4:649-686).
Consistent with this view, nerve growth factor, a diffusible molecule towards which axons will grow in vitro and in viz10 (Levi-Montalcini, Prog Brain Res 1976, 45:235-258;Gundersen and Barrett, 19801, is thought to act by triggering the production of second messengers, although the identities of these second messengers are not fully known. In the case of cell contact-mediated growth cone arrest, the morphological responses of growth cones are remarkably similar to the responses induced by agents that elevate intracellular calcium levels (Kater et al, Trends Neurosci 1988, 11:315-321). Certainly, intracellular calcium appears to play an important role in the morphology and behavior of growth cones (Goldberg, J Neurosci 1988,8:25962605) 141. Significantly, localized increases in calcium concentrations within growth cones have been shown to spread rapidly (Mills and Kater, In 7;be Assembly of the Nervous System Alan R. Liss, 1989, pp 65-801, potentially explaining how contacts made by a few frlopodia could iniluence the motility of an entire growth cone. Whether or not calcium is indeed responsible for the effects of membraneassociated molecules that mediate growth cone arrest should become clearer in the near future, as several groups are now working on purifying and characterizing these molecules (Schwab, Exp New-01 1990, 109:2-5) (Y-71.
The possibility that the effects of intracellular second messengers have a role in the navigational machinery of growth cones is also suggested by an examination of the constituents of growth cones. These have been found to be rich in many of the molecules associated with intracellular signalling: GTP-binding proteins (G- If soluble chemoattractants and growth-inhibitory molecules act on growth cones through second messengers, is it possible that cell-surface and extracellular matrixassociated guidance molecules could work in the same way? There are only a few arguments against this possibility. One is that the interplay between adhesion and tension works so weU that there is little need for anything else, i.e. tension itself may be 'the ideal second messen-bv-which molecules wide axons Lander 9Rl ger' as Mitchison and Kirschner (1988) have noted. Another argument is derived from a recent report that the quantitative relationship between the ability of NCAM-expressing fibroblasts to support neurite growth and the amount of surface NCAM they express is a steep one, with an apparent threshold level of NCAM required before any neurite growth effect is seen [9]. Such a relationship would be expected if NCAM acts primarily in an adhesive manner, because receptor-mediated adhesion is inherently a highly cooperative process. In contrast, such cooperativity is not as easy to explain if NCAM functions by producing a chemical signal. A third argument, and probably the most forceful one of the three, is rooted in the structural analysis of receptors for axon guidance molecules. These receptors, which are so far thought to include the integrins, the cadherins, and the cell-adhesion molecules of the immunoglobulin superfamily, have ammo acid sequences quite unrelated to those of proteins known to activate second messenger pathways, such as growth factor receptors, oncogenes, ion channels, etc. (Lander, 1989). So far, the biochemical functions that have been directly associated with these receptors have been limited to the binding of extracellular ligands, and the binding of components of the actin cytoskeleton via cytoplasmic domains (e.g. Horwitz. et al, Nature 1986, 320:531-533;Pollerberg et al., Cell Tiss Res 1987, 250:227-236;Hirano et al, J Cell Biol 1987, 105:2501-2510. These arguments notwithstanding, a number of studies suggest a close relationship between the actions of substratum-associated guidance molecules and the generation of intracellular signals. III a timely piece of work, Schuch et al. [lo] have shown that the treatment of the neuronal cell line PC 12 with monovalent or polyvalent an tibodies to the neural cell adhesion molecules NCAM and Ll can elicit changes in inositol phosphate levels, an increase in intracellular calcium levels, and a decrease in intracellular pH. Evidence was presented that at least some of these changes were mediated by a pertussis toxinsensitive G-protein. Other recent reports have suggested that laminin substrata can elicit rapid changes in both protein phosphotyiation and inositol phosphate turnover by neurons and neuronaf cell Lines [Weeks and Kleinman, J Cell Bioll989, 109:322a;Plantefaber and Lander, J Cell Biol 1990, lll:(abstract) in press]. Iaminin substrata have also been reported to activate tyrosine hydroxylase, an intracellular enzyme that is strongly regulated by phosphotylation (Acheson et al., J Cell Biol 1986, 102:151-159). In addition, permitting neurons to attach to laminin or libronectin substrata appears to alter the way in which they respond to the second messengermediated effects of polypeptide growth factors (Edgar et al, EMBO J 1984, 3:14631468;Millaruelo et al, Dev Brain Res 1988, 38:21+228).
If the effects of cell surface-associated and extracellular matrix molecules on cells and growth cones are mediated, at least in part, by second messengers, it will be important to understand the molecular steps that connect recognition events at the cell surface to signaIling events inside the cytoplasm. One possibility is that receptors such as integrins and cell-adhesion molecules interact directly with signa.Uing enzymes (kinases, phosphatases, Gproteins, phospholipases, etc.) through interactions that have so far gone undetected. Another possibility is that the interaction with signalling molecules is indirect, perhaps mediated through the cytoskeleton. For example, if some signalling enzymes and their substrates both bind to the cytoskeleton (Scott et al, Biocbem J 1989, 263:207-214;Papadopoulos and Hall, J Cell Biol 1989, 108:553567;Dingus et al., J Cell Biol 1989, 109:19Oa), then events that organize or cluster cytoskeletal filaments could potentially increase the rates of enzyme-substrate encounters, thereby augmenting signal production. A third possibility is that stretch-sensitive ion channels, which have recently been detected in growth cones (Sigurdson and Morns, J Neurosci 1989, 9:2801-2808, directly transduce mechanical tension into membrane depolarization, thereby triggering calcium influx and consequent calcium-mediated signalling. A final possibility is that the effects of axon guidance molecules may not be entirely mediated by those classes of receptors that are currently believed to do so. There are, in fact, independent reasons to explore this last possibility, as discussed below. Multireceptor interactions at the growth cone surface Although the major cell-surface receptors for molecules of the extracellular matrix are thought to be members of the integrin family, and the major receptors for cell adhesion molecules, such as N-cadherin, NCAM and Ll, are thought to be the same or other cell adhesion molecules, the binding of many of these molecules to other types of cell-surface components has also been observed. Most commonly reported has been their binding to glycosaminoglycans, but interactions with glycolipids, cellsurface glycosykransferases and other cell-surface proteins have also been detected (see Lander, 1989 for discussion). Laminin is a good example: potential neuronal receptors for laminin include more than four different integrins, heparan sulfate proteoglycans, cell-surface galactosyitransferase, and several other non-in&grin proteins (Edgar, Trends Neurmci 1989,12:248-251;Lander, 1989) [ 111. Some experiments have suggested that the responses of cells and growth cones to laminin may involve several of these receptors acting in concert For example, the initiation of neurite extension in response to laminin can be completely blocked by antibodies against the 01 integrin subunit (Bozyczko and Horwitz, J Neurosci 1986, 6:1241-1251, but also partially blocked by highly specific inhibitors of cell-surface galactosykransferase [ 111, as well as by concentrations of heparin that are sufficient to inhibit laminin-heparan sulfate interactions (Letoumeau et al, Sot Neurasci Abstr 1986,12:1334Chemoff et al, Tissue Cell 1988,20:165178;0 Ariel and AD Lander, unpublished observations). There is also evidence that integrins and heparan sulfate proteoglycans both have a role in the binding of cells and the extension of neurites in response to fibronectin (Saunders and Bemiield, J Cell Bioll988,106:423-430;Ietoumeau et al., 1986), as weU as evidence that in&grins and a chondroitin sulfate proteogIycan both have a role in the interaction of neural cells with tenascin (Hoffman and EdeIman, Pm Nat1 Acud Sci U&4 1987, 84:252+2527; Bourdon and Ruoslahti, J Cell BioIl989, 108:1149-1155).
Among the cell-cell adhesion molecules NCAM-mediated neuronal adhesion is blocked not only by reagents that interfere with homophilic NCAM-NCAM interactions, but also by heparin, antibodies against cell-surface heparan sulfate proteogiycans, antibodies against a heparin-bind-Ing domain within NCAM, peptides derived from the heparin-binding domain, and by genetic deletion of the heparin-binding domain (Cole et al, J Cell Biol 1985, 100:1192-1199Cole et al, Nature 1986, 320445447;Cole et a,!, J cell Biof 1986, 103:1739-1744Cole andAkeson, Neuron 1989, 2:1157-l 165) [ 121. MyeIin-associated glycoprotein is another example of a heparinbinding neural ceU adhesion molecule (Fahrig et al., EMBO J 1987. A recent report indicated that N-cadherin, a homophiIic calcium-dependent adhesion molecule thought to play a major role in axon outgrowth at ceU surfaces, is speciiicaUy bound by the cell-surface enzyme N-acetyIgaIactosaminyIphosphotransferase [ 131. This iinding, along with a study on the role of gaIactosyItransferase in the recognition of laminin [ 111, suggests that surface giycosykransfemses in general may have important roles in the recognition of axon guidance molecules. Why should a growth cone possess a multiplicity of receptors for guidance molecules? Finding an answer to this question depends a great deal on one's choice of model to explain how growth cones respond to guidance cues. For example, if guidance molecules act primarily or in part by generating intraceUular signals, then multiple types of cell-surface receptors may exist to extend the range of second messengers that can be produced by a single guidance cue. One di&ulty with this explanation is that proteogiycans, gIycoIipids and gIycosyitransferases have not traditionaUy been thought of as molecules that can mediate intraceUular signaUing. It has been pointed out recently, however, that binding of molecules to gIycoIipids on the surface of some cells can trigger calcium influx (Dyer and Benjamins, J Cell Biol 1990, 111:625-633), and that some neural proteoglycans are attached to the cell surface via a phosphoiipid Iinkage that, when cleaved can generate a second messenger (Carey and Evans, J Cell Bioll989,108:1891-1897; Hemdon and Lander, Neuron 1990,4:949-961).
Alternatively, if growth cones are guided primarily by 'sensing' adhesion, the presence of multiple types of receptors would probabIy serve to increase adhesion by increasing the avidity of binding. Thus, an NCAM molecule bound to the cell surface via both NCAM-NCAM and NCAM-heparan sulfate interactions would be bound more tightly than an NCAM molecule bound via either interaction alone (Cole et al, 1985). A somewhat more complicated possibility is that cells use multiple receptors not so much to increase the Iinai amount of adhesion they develop, but speciUcaUy to regulate the time period over which adhesion develops. A still more complicated possibility, but perhaps the most intriguing one, is that, by having more than one type of receptor at their disposal, growth cones can independently modulate the binding of membrane to substratum without coordinateIy affecting the binding of the cytoskeleton to the ceU membrane. As a hypothetical example, if one patch of membrane contains 1000 lamininbinding integrins whiIe another contains 100 integrins and 900 gaIactosyItransferase molecules, both patches might achieve the same level of adhesion, but the first wiU be connected to the actin cytoskeleton in ten times as many locations as the second (based on the expectation that integrins associate with actin tilaments but gIyco.syItransferases do not). Why should the number of membran~oskeletai linkages matter to a growth cone? For one thing, Mitchison and Kirschner (1988) have concluded that, in order for the actin iiiaments in a growth cone to generate tension and still undergo a continual rearward flux, they must constantly slip against their own membrane Linkages. Such a necessity for siippage could easily set a Iimit on the number of membrane connections that are tolerated by an actin Nament. If this number is low, growth cones could benefit greatly by achieving a high degree of adhesion through combinations of receptors that do and do not interact with the cytoskeleton.

Conclusions
Clearly, a great deal is known currently about the process of axon guidance: the structure, behavior and cytoskeletal dynamics of growth cones have been subjected to considerable scrutiny, and many molecules with the potential to guide axons have been described. The evidence discussed above suggests that plausible hypotheses do exist for explaining how such molecules guide axons. Testing these hypotheses is an important goal for the future. Achieving this goal, however, will require answers to several difficult questions. What levels of adhesion do constantly moving structures, such as UIopodia and IameUipodia, actuaUy experience? What controls the time period over which ceU adhesion develops? How do ceU-ceU and cell-matrix contacts initiate or modify the production of intracellular signals? What happens when cells bear multiple types of receptors for a single cellsurface or extracellular matrix protein? That these are fundamentaI questions in ceU biology should not be surprising, as the molecules that are believed to guide axons are, to a large extent, the same as those that are thought to control the adhesive, motile and morphogenetic behaviors of cells in general