Growth cone guidance and neuron morphology on micropatterned laminin surfaces

Growth cones of extending axons and dendrites explore their local environment and are capable of detecting and responding to extrinsic cues that guide the direction of their extension. Cues involved in this guidance of neurite out-growth include chemotropic gradients and specific substra-tum pathways (Dodd and Jessel, 1988; Hynes and Lander, 1992).

Local environmental cues influencing cell-substratum interactions are believed to be important in controlling growth cone guidance. Recently, a number of studies have suggested that pre-formed pathways of differential adhesiveness are crucial in guiding neuronal growth cones to distant targets in vivo. In frog spinal cord, the dorsal column provides a track that specifically guides dorsal root gan-glion (DRG) axons either rostrally or caudally after their entry into the spinal cord from the dorsal horn (Holder et al., 1987). Motoneuron axon extension is believed to be guided by the tissues that the axon encounters as it extends peripherally (Lance-Jones and Dias, 1991; Tosney, 1991; Tang et al., 1992). Projection from the retina to the optic tectum appears to be via a preformed pathway, since sur-gical rotation of the pathway region in embryos resulted in the deflection of projection consistent with the direction of rotation (Harris, 1989; Taylor, 1990). The nature of this pathway is uncertain, though both NCAM (Silver and Rutishauser, 1984) and laminin (Cohen et al., 1987) have been implicated. NCAM pathways are also likely to be involved in guiding axons during chick limb innervation (Tang et al., 1992) and cochlear development (Whitlon and Rutishauser, 1990). The presence of putative pathways of laminin appears to guide axonal outgrowth in a number of developing central and peripheral neural tissues (Rogers et al., 1986; Riggott and Moody, 1987; Letourneau et al., 1988; Liesi, 1990). Three main mechanisms of axon guid-ance by substratum adhesiveness operate: substratum pref-erence, inhibition or repulsion of growth cones, and guide-posting. In substratum preference, growth cones are faced with a choice of adhesive (neuritogenic) or non-adhesive (non-neuritogenic but otherwise inert) substrata. Such adhe-sive substrata may be cell surface adhesion molecules (CAM), which may be highly specific, or patterned extra-cellular matrix material (ECM). Tracks of NCAM or laminin might guide by substratum preference. The term ‘adhesiveness’ may not be entirely appropriate in some instances. Substratum adhesiveness does not correlate with neurite growth rate (Lemmon et al., 1992), and it has been suggested that laminin may act as an ‘anti-adhesive’ agent, presumably promoting axon extension by increasing growth cone motility rather than growth cone-substratum adhesion (Gundersen, 1988; Calof and Lander, 1991), though a recent study suggests that growth cone protrusions are stablised by laminin substrata (Rivas et al., 1992). It is now clear that the ‘adhesiveness’ of substrata may also be modulated by substances that actively inhibit or repulse growth cones. The segmented distribution (to the posterior half of sclero-tomes) of a repulsive agent is involved in the segmentation of the developing peripheral nervous system (Keynes and Cook, 1992). The presence of gradients of such repulsive molecules is believed to be involved in the control of gan-glion cell axon extension to the optic fissure in developing retina (Snow et al., 1991) and in the formation of the appro-priate pattern of connections of retinal ganglion cell axons in the optic tectum (Baier and Bonhoeffer, 1992). Myelin-associated inhibitory molecules repress regeneration of adult central neurons, but may be involved in development (Schwab and Schnell, 1991). The third mechanism, guide-posting (Palka et al., 1992), which has been described mainly in insects, could be considered a particular case of substratum preference. In this case, the adhesive substra-tum is discontinuous, consisting of specific guidepost cells along the pathway of axonal extension. The axon extends after a guidepost cell is contacted by the filopodia of an exploring growth cone and cell-growth cone association is established. Axon guidance is the result of progressive extension from guidepost to guidepost.

A number of in vitro studies have examined the ability of patterns of adhesiveness to guide the outgrowth of neu-rites. Laminin, the large ECM protein known to be capable of initiating and sustaining neurite extension (Lander et al., 1985), has been patterned using techniques of limited res-olution (electron microscope grids being the templates for the patterns; Hammarback et al., 1985, 1988; Gundersen, 1987), as has patterned nerve growth factor (Gundersen, 1985). A microfabrication technique has been used to micropattern amino groups to silica surfaces, thereby patterning adhesivness for neurons (Kleinfeld et al., 1988). Recently, we have modified this technique by simplifying the fabrication procedure, thus allowing the manufacture of micropatterns of adhesion on fused quartz and standard glass. These patterns were used by us to examine the behav-iour of non-neuronal cells (Britland et al., 1992; Clark et al., 1992). Here, we have examined the responses of the growth cones of dissociated neurons in culture to patterns of laminin of various geometries formed by its preferential adsorbtion to micropatterns of hydrophobicity. These sub-strata provide a convenient model system that has allowed the examination of the sensitivity of growth cones in detect-ing and reacting to differentially adhesive cues.

Cerebral hemispheres of 7-day chick embryos were removed and cleaned of meningeal membranes. The hemispheres were then minced, incubated in 0.05% trypsin/0.2% EDTA in PBS for 15 min at 37°C, stopped with soybean trypsin inhibitor (Sigma, UK), triturated in serum-free medium (Gibco High Protein Hybridoma medium; Gibco, UK) supplemented with glutamine and antibiotics as described previously (Clark et al., 1987) with a fire-polished Pasteur pipette, washed by centrifugation, and resuspended in medium. Neurons were plated onto patterns in 6 cm Petri dishes (106 per dish) and cultured in the serum-free Gibco Hybridoma medium at 37°C in a humidified atmosphere of 5% CO₂. Some neurons were cultured in Dulbecco’s MEM with 10% foetal calf serum.

Dorsal root ganglion neurons were obtained from P1 neonatal Balb/c mice. The mice were anaesthetised by cooling and killed by decapitation. Following sterilisation with ethanol, the skin was opened to expose the vertebral column. The spinal ganglia were exposed after laminectomy. Approximately 50 DRG were col-lected and then finely minced in divalent-free Hank’s balanced salt solution (HBS). Spinal cord cells were obtained from three E13 foetal Balb/c mice. After removal from the chorionic sac, the foetuses were killed by decapitation and the spinal cord was removed. The cords were washed in divalent-free HBS and finely minced. The remainder of the procedure was common to both tissues. Minced tissue was transferred to a solution of 0.05% trypsin in 0.2 mg ml−1 EDTA (Flow Labs, UK) at 37°C for 30 minutes. Trypsinisation was stopped by the addition of trypsin inhibitor (1 mg ml−1) in medium as above. The cells were cen-trifuged and resuspended, by trituration, in the serum-free medium. For DRG neuron culture, the medium was supplimented with 100 ng ml−1 nerve growth factor (Sigma, UK). DRG cells were plated at a density of approx. 5×103 cells cm−2, and spinal cord cells at 1×104 cells cm−2. Cells were incubated as above for 24 (DRG cells) or 48 (spinal cord cells) hours.

Patterns of hydrophobicity were made as described previously (Britland et al., 1992; Clark et al., 1992). Briefly, sulphuric acid/hydrogen peroxide-cleaned fused quartz or standard micro-scope slide glass was spin-coated with photoresist, exposed to UV light through a chrome mask of the desired pattern, and devel-oped to leave a pattern of photoresist and exposed glass. These patterns were immersed in 2% (v/v) dimethyldichlorosilane in chlorobenzene, rinsed twice in chlorobenzene, and blown dry. The resist pattern was then removed by rinsing in acetone and then water, resulting in a final pattern of methyl groups (the hydropho-bic surface) covalently coupled to the quartz surface. Repeat patterns of equal sized hydrophobic lines and untreated quartz spaces were made using masks of repeat spacings 4, 6, 12, 24 and 50 μm (i.e. feature sizes of half these values). Patterns were placed in 6 cm Petri dishes, covered with 5 μg ml−1 laminin in PBS, and incubated at 37°C for 2-3 hours, after which time excess laminin solution was removed and cells added. Neurons were also cul-tured on laminin patterns made up of single 2 μm lines, separ-ated by 50 μm and interupted with 50 μm tracks every 500 μm, and on patterns that consisted of adjacent large hydrophobic and hydrophilic regions so that comparisons could be made at a bound-ary. Cells were also seeded onto patterns of hydrophobicity that had not been treated with laminin.

After incubation in laminin solution as above, some patterns were rinsed twice in 0.1% BSA in PBS, then either drained and pri-mary antibody added, or rinsed in PBS, fixed briefly in 4% formaldehyde in PBS, rinsed in PBS and primary antibody added. After incubation in the primary antibody (1:40 affinity-purified rabbit anti-laminin polyclonal antibody (Sigma, UK)) for 45 min-utes, patterns were washed twice in 0.1% BSA in PBS and incu-bated for 20 minutes in rhodamine-labelled goat anti-rabbit IgG antibody. Patterns were washed with PBS, mounted in Vecta-mount (Vector Labs, UK), and examined and photographed using a Zeiss Axioskop fluorescence microscope.

The degree of alignment of neurite outgrowth was determined as previously (Clark et al., 1991). Chick embryo cerebral neurons were photographed under phase optics after 24 hours in culture. To determine whether or not neurite outgrowth was aligned on patterns, individual neurites (between 50 and 80 per sample) were scored as to the angle at which they intersected with an arc whose origin was the cell body and whose radius was 30 μm. The pro-portion of neurites whose intersections fall within 45° of the alignment of the pattern was taken to be the measure of align-ment. A population of neurites outgrowing randomly will be expected to have an alignment value of 0.5. Neurite guidance in the direction of the pattern would give an alignment value sig-nificantly greater than 0.5, and guidance perpendicular to pattern direction a value significantly less than 0.5. The significance (P<0.05) of measured differences in the degree of alignment was determined by calculating chi2 from 2 ×2 contingency tables.

Some samples of cultured chick cerebral neurons were fixed in 4% formaldehyde in PBS at room temperature for 1 hour, washed in PBS, mounted under thin coverslips with aqueous mountant, and examined with differential interference contrast (DIC) optics. Growth cones on various patterns were photographed using an oil immersion ×100 objective.

The micrographs in Fig. 1 show the differences in adhe-siveness for chick embryo neurons between adjacent hydrophobic and untreated areas of quartz glass surfaces.

It can be seen that, in the absence of laminin, neurons are more adherent to the untreated (hydrophilic) surface than the adjacent treated (hydrophobic) surface, there being little neurite outgrowth on either surface (Fig. 1A). After laminin treatment, adhesion is greater on the treated surface where neurite outgrowth is also evident, whereas the untreated sur-face is less adhesive with little or no neurite outgrowth (Fig. 1B). This difference was found with and without serum, though neurite outgrowth, when present, appeared less extensive in serum-containing cultures (not shown). No dif-ference in patterning adhesiveness was found between quartz and standard glass (not shown). Immunofluorescence localisation of laminin revealed a confinement of staining to the previously hydrophobic regions of the patterns (Fig. 1C,D). In Fig. 1C it can be seen that the intensity of fluo-rescence is greater at the boundary with the non-silanated region. Control patterns (no laminin incubation) showed no staining.

When neurons were cultured on repeating patterns of alter-nating lines of treated and untreated glass or quartz that had been exposed to laminin, the degree of alignment of neu-rite outgrowth was dependent on pattern spacing. Neurite outgrowth on 4 and 6 μm period patterns appears not to be affected, whereas that on patterns of larger period is guided, the neurite elongation being oriented along the tracks of adhesiveness for large distances (Fig. 2). Changes in direc-tion of guided neurites were never observed, even on the widest (25 μm) tracks (Fig. 2F). These observations were borne out by those from the measurement of alignment of neurite outgrowth. Neurite outgrowth on 4 and 6 μm period patterns is not significantly aligned, though that on patterns of greater period is, reaching complete alignment by 24 μm period patterns (Fig. 3).

Though neurite outgrowth was not aligned by 4 μm period patterns (2 μm adhesive tracks separated by 2 μm non-adhesive tracks), single 2 μm lines of adhesiveness supported the adhesion and guided the neurite outgrowth of chick embryo neurons (Fig. 4). These lines were 500 μm in length, separated by 50 μm and they perpendicularly join 50 μm wide tracks. Neurites were seen to extend from these wider zones to the 2 μm lines, and vice versa (Fig. 4B).

The neurite outgrowths of chick embryo brain neurons (Fig. 5) and mouse embryo spinal chord neurons (not shown) are highly aligned by 24 μm period patterns. Dorsal root ganglion neuron neurite outgrowth is generally not aligned by 24 μm patterns (though some individual neu-rites do appear to be guided) (Fig. 5).

On unpatterned laminin and on 24 μm period laminin pat-terns, mouse DRG neurons typically produce an outgrowth of multipolar morphology, the individual neurites often branching to form a dense arbor (Fig. 5A,B). In contrast, DRG neurite outgrowth is aligned by patterns of 50 μm period. The morphology of this outgrowth is typically bipolar, single neurites extending for large distances in oppo-site directions along the tracks of laminin, without branching (Fig. 5E,F).

On unpatterned laminin surfaces growth cones were typi-cally wide (5-8 μm) lamellar structures bearing a number of short (rarely longer than 6 μm) filopodia (Fig. 6A). Growth cones unguided by 4 μm period laminin patterns (2 μm lines and spaces) (in Fig. 6B neurite extension is perpendicular to the pattern), maintained their lamellar morphology though filopodia originating from the growth cones, and lateral spines of the neurite, display a patterned distribution corresponding to the laminin pattern (Fig. 6B). Filopodia are present at the most distal edge of the growth cone, their length indicating that they are bridging at least one non-adhesive strip, as are lamellar regions.

Patterns of 12 μm period guide growth cones (see above) and, as seen in Fig. 6C, their morphology is altered. The growth cones, which are in contact with both boundaries with adjacent non-adhesive strips, appear narrower and have fewer filopodia, which are confined to the distal end (Fig. 6C). In contrast, growth cones of neurons cultured on 50 μm repeat patterns (25 μm lines and spaces) do not span the width of the adhesive strip, being in contact with only one boundary with adjacent non-adhesive regions (Fig. 6D). Their morphology differs little from those on unpat-terned substrata. The growth cones of neurites extending along single 2 μm lines have become extremely simplified, having become single filopodia (Fig. 6E) or narrow lamel-lipodial structures tapering to a filopodium (Fig. 6F).

The examination of growth cones at boundaries between adhesive and non-adhesive regions, such as on 50 μm period patterns (Fig. 6D) or on larger patterns (Fig. 6G,H), indicates that growth cone structures rarely cross to the non-adhesive surface. In fact, lamellar regions were never seen to be in contact with a boundary. Similarly, the neurites themselves are never seen at the boundaries. When neurite extension has occurred parallel and close to a boundary, the distance between the neurite and the boundary remains rel-atively constant (approximately 5-7 μm) (Fig. 6D,G,H). There was no indication of a preference of growth cones for the areas adjacent to boundaries, since in a number of cases neurites have been seen to have turned at a bound-ary (not shown).

The effectiveness of patterned adhesiveness as a potential in vivo axonal guidance cue has been established by in vitro experiments. These studies have mainly used electron microscope grids as their pattern templates, and therefore were limited in resolution of feature size. Despite this lim-itation, it has been shown that adhesiveness per se (Letourneau, 1975), and both patterned laminin (Hammar-back et al., 1985, 1988; Gundersen, 1987) and nerve growth factor (Gundersen, 1985), will guide neurite extension, and in testing the capability of chick autonomic ganglion neu-rons to pathfind by guidposting, Hammarback and Letourneau (1986) have shown that neurites can bridge non-adhesive regions.

Here, we have developed a method for micropatterning laminin that provides experimental substrata that model substratum preference. The limit of resolution of this method will be the limit of resolution of standard pho-tolithographic techniques, i.e. feature size of 1–2 μm. We have used such patterns to examine how the geometry of patterned adhesion may influence their ability to guide axonal growth efficiently. Surprisingly, laminin was found to adsorb preferentially to the hydrophobic surface of the patterns, which consists of methyl groups covalently linked via silane bonds to standard glass or fused quartz. This pref-erential adsorption was unexpected, since our earlier expe-rience with these patterns had suggested that cell attach-ment factors in serum (presumably vitronectin and fibronectin) adsorb poorly to the hydrophobic surfaces, cell adhesion being far greater on untreated glass (Britland et al., 1992; Clark et al., 1992). However, this property of laminin has provided precise micropatterns to which neu-rons selectively adhere and which promote neurite out-growth. Immunofluorescence localisation of laminin suggests that there may be increased accumulation close to boundaries as a result of diffusion effects at these sites. Any increase in laminin concentration at these areas does not appear to lead to the preferential accumulation of growth cones, since turning of neurites from boundaries was observed. This is consistent with the previously observed lack of haptotactic guidance of growth cones on gradients of laminin (McKenna and Raper, 1988).

The ability of a pattern to guide neurite outgrowth was found to be strongly dependent on its geometry. Narrow (4 and 6 μm period) patterns did not orient outgrowth, wider period patterns did. This is likely to reflect the ability of growth cones to produce protrusions that are able to bridge non-adhesive regions. Growth cones of autonomic ganglion neurons have been shown to be capable of bridging non-adhesive regions 30-40 μm wide and occasionally 50 μm wide (Hammarback and Letourneau, 1986). Similarly, a strong dependence on pattern geometry was found in the guidance of fibroblastic and epithelial cells, where bridging across non-adhesive tracks was seen to affect the ability of the substratum to align cells (Clark et al., 1992). In the present study, we found that the growth cones of chick embryo brain and mouse embryo spinal cord neurons were consistantly guided by 12 μm tracks of laminin separated by 12 μm non-adhesive tracks (24 μm period), suggesting that they are unable to bridge this distance. In contrast, we found that mouse neonatal dorsal root ganglion neuron growth cones were often not guided by 24 μm period pat-terns but were by 50 μm patterns, i.e. they are able to bridge 12 μm, but not 25 μm, non-adhesive regions. Although 4 μm period patterns (2 μm tracks of laminin separated by 2 μm non-adhesive lines) were unable to guide the out-growth of chick embryo brain neurons, single 2 μm tracks (i.e. separated by 50 μm) are able to sustain adhesion and promote bipolar outgrowth of neurites. Similarly, the guid-ance of BHK and MDCK cells was markedly reduced on 2 μm period repeat patterns of adhesiveness (Clark et al., 1992), but the cells were aligned by single 2 μm adhesive lines (unpublished results). Therefore, the proximity of adjacent adhesive areas reduces guidance by tracks. It would seem that in order to guide neurite extension, a track must be separated from such adjacent regions by a distance greater than the distance over which a growth cone can extend exploring protrusions.

The present study has shown that tracks of adhesion as narrow as 2 μm and as wide as 25 μm can precisely guide axonal extension over large distances. However, as dis-cussed above, a growth cone will only be constrained by a track of adhesiveness if the track is separated from any adja-cent regions of adhesiveness by greater than the bridging distance of the particular growth cone, i.e. the length of protrusion produced by the growth cone. These data have indicated some important limitations of growth cones response to local environmental influences. Hammarback and Letourneau (1986) showed that autonomic neuron growth cones could extend across relatively large non-adhe-sive regions and suggested that this ability is important in the guidepost hypothesis of guidance of axonal outgrowth. Guideposting in higher animals has not been observed (Palka et al., 1992). Our observations indicate that the dis-tances that avian and mammalian central neurons are able to bridge may be too small to allow guideposting to be an effective mechanism of guidance, since these distances are as small as, if not smaller than, many cells themselves. It would also seem that many growth cones may be incapable of responding to the multiple parallel adhesive cues pro-vided by aligned fibrillar extracellular matrix material (see Clark et al., 1992). The observed guidance of neurites by aligned collagen gels in vitro (Ebendal, 1976) is likely to be the result of another anisotropic property of these gels, though we have previously shown that chick embryo cere-bral neurons, unlike non-neuronal cells, were insensitive to ultrafine topography (Clark et al., 1991). Similarly, when a growth cone encounters a fasciculated bundle of neurites, it will be presented with multiple parallel adhesive cues on a scale that the present observations suggest would not guide neurite extenstion, i.e. the adhesive cues would not prevent extension across or around the fascicle (unless a single neurite or small bundle of neurites in the fascicle presented a separate, specific adhesive cue). We contend that topographic cues (Curtis and Clark, 1992) and persis-tence of neurite extension (Katz, 1985) play an important part in maintaining the linear order and preventing random tangling in axon fascicles. It must be noted, however, that the nature of their substratum can influence the morphol-ogy of growth cones, both in vitro and in vivo, by altering the number and length of protrusions (Letourneau, 1979; Wilson and Easter, 1991; Payne et al., 1992). This phe-nomenon will also be an important factor in determining the sensitivity with which growth cones respond to guid-ance cues in their environment.

The observed morphology of growth cones on laminin (both patterned and unpatterned) correlates with their impaired ability to bridge distances greater than 6 μm, since filopodia longer than 6 μm were rarely seen. Similarly, the ability to bridge distances smaller than 6 μm renders them insensitive to the cues provided by smaller period patterns. At single boundaries between non-adhesive and adhesive regions, the morphology of growth cones, where lamellar regions and neurites were never in contact with a bound-ary, suggests that growth cone advance (i.e. the advance of the lamellipodium) is inhibited in the direction of filopodia in contact with the boundary. Lamellipodial advance would appear to require established, stable filopodia, as has been suggested previously (Heidemann and Buxbaum, 1991; Rivas et al., 1992). This requirement may, in fact, be absolute, since laminin substratum is available to the edge of the pattern boundaries, but lamellopodia are not able to advance to the edge. This results in the neurite itself being formed at a distance from the boundary. On narrow multi-ple parallel tracks, lamellipodial regions of growth cones appear to be in contact with the non-adhesive regions, which suggests that lamellopodial advance is directed by filopodia and may be independent of cell-substratum adhe-sion (possibly in a similar manner to the spreading of fibroblasts on topographically discontinuous surfaces; Rovensky et al., 1991); i.e. the nature, or indeed presence, of a substratum may be unimportant for lamellipodial exten-sion, so long as there is a scaffold of filopodia on which advance can be supported. The progressive simplification of growth cone morphology on increasingly narrower adhe-sive tracks has shown the ability of these motile structures to conform to the available preferred substratum. This adap-tation appears to be as a result of the inhibition of filopodium formation except at the distal edges of the growth cones (i.e. in the direction of the track). Time-lapse observation of growth cone advance on various patterns may, in the future, reveal differences in the efficiency of morphologically distinct growth cones in neurite extension, which may have relevance in vivo. Micropatterns of laminin provide a mechanism for specifically altering growth cone morphology without altering the composition of the sub-stratum.

It is clear that tracks of adhesiveness are capable of pro-viding a guidance cue that can steer axon extension for large distances. This type of cue is, however, a bidirectional one. Turning or meandering of growth cones is possible, partic-ularly on wider tracks. Where a growth cone meets a track of adhesiveness perpendicularly, in the absence of other cues, the direction taken would be expected to be randomly right or left; for example, DRG axons encountering the dorsal column in the frog spinal cord (Holder et al., 1987). Other cues, such as chemotactic or haptotactic gradients, will be required at such ‘T-junctions’ to provide specific directionality. In order to provide this directionality in vivo, it is probable that the cues need only be short and/or tran-sient, as persistance of locomotion will maintain direction of extension (Katz, 1985). It must be noted, however, that McKenna and Raper (1988) were unable to demonstrate haptotactic guidance of neuronal growth cones on gradients of substratum-bound laminin, in vitro, though gradients of inhibitory molecules are believed to guide axons in vitro and in vivo (Snow et al., 1991; Baier and Bonhoeffer, 1992).

We noted that mouse neonatal DRG cells cultured on unpatterned laminin (or on a 24 μm period laminin pattern, which does not guide neurite outgrowth) had a multipolar neurite outgrowth, a dense arbor emanating from a single neuron as a result of branching of primary neurites. When these same cells were cultured on 50 μm period patterns, which oriented their neurite outgrowth, the outgrowth was, in many cases, bipolar. Single neurites extended from oppo-site ends of the cell body, usually without forming any branches. Mature DRG cells in vivo have a pseudounipo-lar mophology, though a bipolar morphology is a normal intermediate developmental stage (Pannese, 1974). Our observations could indicate that the extracellular environ-ment may contain distributed information that influences the morphological differentiation of these neurons. It has recently been suggested that such local cues could con-tribute to the establishment of neuronal axon/dendrite polar-ity (de Curtis, 1991). Possible mechanisms of how this dif-ference in the morphology of DRG neurons on substrata of different geometry might arise are unknown, but it is pos-sible to speculate that the initiation of neurite outgrowth is inhibited in directions other than along the tracks. Neurites initiated laterally on these patterns do not find suitable sub-strata, therefore their establishment and extension are not continued. Such geometrical constraints may also inhibit neurite branching. Substratum availability could be involved in the control of branching.

Guidance of neuronal growth cones by their substratum could involve topographic as well as adhesive pathways. Our previous experience of the extent to which growth cones respond to topography (Clark et al., 1987, 1990, 1991; Curtis and Clark, 1990) suggests that, although their locomotion is perturbed and could be guided by topographic cues, the probablistic manner of their reaction to such cues would not result in the precise, stereotypical pathfinding observed in vivo. Local topography may contribute to axonal guidance (e.g. in fasciculation; see Discussion), but there is little evidence that the kind of structures that would be required for axonal guidance by topography alone are present during the development of vertebrate neuronal tissues; though physical features of the local environment of the developing Manduca wing are believed to control pioneer neuron outgrowth (Nardi and Vernon, 1990). Despite the bidirectional nature of the adhesive cues in the present study, turning of neurites was never observed even on the widest tracks examined (25 μm). As has already been proposed for non-neuronal cells (Clark et al., 1992), persistance of locomotion of growth cones (Katz, 1985) may be sufficient for directionality, once established (either randomly or by a directional gradient), to be maintained. Our findings indicate that the greatest precision in growth cone guidance by substratum preference will be achieved by isolated narrow tracks of neuritogenic substratum. It seems likely that in vivo a growth cone will remain con-strained to an isolated track of preferred substratum until either its target is reached or another cue over-rides the sub-stratum cue. Tracks of preferred substratum could guide axon extension with directionality, in the absence of gradi-ents, where there is a clear spatial relationship between the direction of the track and the point at which it can be encountered by a growth cone. Growing axons may be fun-nelled into a track such that, because of persistance of loco-motion, only one direction of extension is possible. Guid-ance by tracks of preferred substratum is only one of a number of possible cues involved in neurite guidance in vivo, but the present and previous in vitro studies clearly show that this may operate in diverse and widespread devel-opmental processes to provide a precise and effective mech-anism of controlling axon extension.

We thank Yash Bhasin and Barry Crook at St. Mary’s, Henry Goulding and Stephen Durr at the Institute of Neurology, and Joan Carson at Glasgow University for their excellent technical assis-tance.