Glia dictate pioneer axon trajectories in the Drosophila embryonic CNS

Axons extend to form intricate and stereotyped trajectories. Local and long-range cues are thought to aid pathfinding by the first axons to trace a pathway (pioneer axons; Bate, 1976). As pioneer growth cones navigate they ‘decide’ whether to follow along or move away from a given direction. Such decisions are made at stereotyped choice points and may reflect a combination of local cues and signals from the target. Once the primary axonal trajectories are established, follower neurons project growth cones, which fasciculate with pioneer axons. When a pathway is shared by neurons with ultimately different trajectories, follower axons must ‘decide’ which route to take, consequently defasciculating from sister axons. It is believed that each neuron is ‘able’ to read cues with such precision as to execute multiple fasciculation and defasciculation decisions in an environment heavily dense in axons, to finally make correct contacts with its target (Goodman et al., 1984; Goodman and Shatz, 1993; Tessier-Lavigne and Goodman, 1996).

It has long been believed that glia preform the pathways that axons will follow (Silver et al., 1982; Singer et al., 1979) and there is evidence from vertebrates and grasshopper suggesting that glia can guide growth cones and can also prompt fasciculation and defasciculation of axons at choice points (Auld, 1999; Pfrieger and Barres, 1995). In grasshopper, ablation of the segment boundary cell prevents exit of the aCC axon from the CNS (Bastiani and Goodman, 1986). However, in the CNS it is still uncertain whether glia aid pathfinding or not (Auld, 1999; Pfrieger and Barres, 1995).

Over recent years, most work on guidance has focused on understanding the control of midline crossing by growth cones (Tessier-Lavigne and Goodman, 1996; Thomas, 1998; Tear, 1999). In Drosophila, the ventral nerve cord of the embryonic CNS consists of longitudinal connectives, with two commissures across the midline linking the connectives in each segment (Goodman and Doe, 1992). Most interneurons cross the midline once and fasciculate with pioneer axons to grow along the longitudinal pathways up to the brain. Both attractive and repulsive signals are secreted by midline cells to control midline crossing by axons (Dickson, 1998; Tessier-Lavigne and Goodman, 1996; Thomas, 1998; Tear, 1999). These long-range signals are evolutionarily conserved, implying that midline cells play a fundamental role in controlling axon crossing. Longitudinal pioneer axons, however, are characterised by their lack of midline crossing. The role of glia in the establishment of these longitudinal pathways remains unclear.

Thus far, the evidence does not favour a role for glia in the guidance of longitudinal axons. CNS glia in some ways resembling oligodendrocytes, called here longitudinal glia, enwrap the longitudinal axons (interface glia in Ito et al., 1995). The longitudinal glia originate from a lateral glioblast which divides while migrating towards the midline (Jacobs et al., 1989; Schmidt et al., 1997). EM studies had suggested that glia form a prepattern of guidepost cells for the growth cones to follow (Jacobs and Goodman, 1989a). However, the progression of pioneer growth cone extension relative to glial migration patterns remains unknown. Mutations and ablation of glia have been used to study the role of glia in guidance. Glia were ablated by means of the GAL4 system with glia-specific lines available at the time (Hidalgo et al., 1995). However, in the only case where glia were ablated prior to growth cone extension, the MP2 and SP1 neurons were also ablated, obscuring any involvement of glia in guidance. In the remaining cases, the glia were ablated at a time following growth cone extension, so the question of pioneer growth cone guidance could not be addressed. The effects of several mutations on longitudinal tract formation have been analysed but, because these genes are involved in midline or neuronal development, their effects are not direct (Auld, 1999; Jacobs, 1993). In the more specific repo glial mutants, however, longitudinal tracts form (Halter et al., 1995). Mutants for the gene glial cells missing (gcm) lack all glia, which are transformed into neurons (Hosoya et al., 1995; Jones et al., 1995; Pfrieger and Barres, 1995; Vincent et al., 1996). Embryos lacking gcm can lack all longitudinal tracts. However, longitudinal axon tracts can also form, leading to the conclusion that glia play no essential role in guidance (see references above). However, the transformation to neuronal fate is incomplete, since in the PNS chordotonal organs only 15-30% of hemisegments have all glia transformed to neurons (Hosoya et al., 1995; Jones et al., 1995). Furthermore, the transformed cells may retain some glial features, since they migrate, divide and reach the neuropile as normal glia do (Hosoya et al., 1995; Vincent et al., 1996). Remarkably, the number of β-gal-positive cells in gcm^PlacZ mutants is the same as that of glia in wild type, indicating that the glioblast lineage has not been altered (Hosoya et al., 1995). Furthermore, because lacZ-expressing transformed cells are found along the axonal pathways of the CNS (Hosoya et al., 1995; Vincent et al., 1996), it is also conceivable that they might still provide novel cues that axons are also able to follow (Pfrieger and Barres, 1995). Consequently, gcm mutations do not simply correspond to lack of glia but to a novel composition of the ventral nerve cord.

Longitudinal pathways are pioneered by pCC, MP1, dMP2 and vMP2, which extend in pairs in opposite directions (Bate and Grunewald, 1981; Bastiani et al., 1986; Jacobs and Goodman, 1989b; Lin et al., 1994; Hidalgo and Brand, 1997). pCC and vMP2 extend together anteriorly, whereas MP1 and dMP2 extend together posteriorly. All four contact half way to establish the first, single longitudinal fascicle. Subsequently, pioneer axons undergo a series of defasciculation and refasciculation events to establish, by the end of embryogenesis, two primary fascicles at each side of the midline: pCC fasciculates with dMP2 along the first fascicle, closest to the midline, vMP2 runs also along this pathway, but in a more ventral plane defasciculated from pCC/dMP2 and MP1 runs along the second, central fascicle (Hidalgo and Brand, 1997). A third, outer fascicle is visualised by fasII, but its pioneer neurons remain undiscovered. It is not known what governs the defasciculation and refasciculation events that build the final trajectories of pioneer axons.

Here we have analysed the consequences of interfering with glial function on axonal trajectories, using ablation and gcm mutations. We have ablated the longitudinal glia at the glioblast stage, prior to the time of axon extension. We show that glia are responsible for orienting growth cones and for the fasciculation and defasciculation events that trace the pathways of longitudinal pioneer axons.

(1) Wild type: Canton-S; (2) glial cells missing mutants: gcm^ΔP1/CyOlacZ (Jones et al., 1995); (3) synthetic glial GAL4 driver: w; s-gcmGAL4 15.1, insertion on the X (Booth et al., 2000); (4) w; UAS-RicinA/CyOen11 lacZ (Hidalgo et al., 1995): CyOen¹¹lacZ drives lacZ expression in stripes in the embryo, in the expression pattern of the wingless gene; (5) double GAL4 line driving expression in glia and MP2: w; s-gcmGAL4 211/CyOlacZ; 15J2/15J2 (for a description of 15J2, see Hidalgo and Brand, 1997).

Ablation of glia was carried out with the GAL4 system (Brand and Perrimon, 1993). The line s-gcmGAL4 151 was engineered by fusing a synthetic enhancer with 11 repeats of the consensus binding sequence for Gcm upstream of GAL4 (see Booth et al., 2000). Line s-gcm GAL4 151 drives expression in the glioblast, progenitor of the longitudinal glia, and its progeny, and in other glial classes in a mosaic fashion (Booth et al., 2000). This line also drives sporadic expression in some macrophages and in a reduced number of neurons, mainly from stage 16. These neurons are not the pioneer neurons. The combined stock s-gcmGAL4 211/CyOlacZ; 15J2/15J2 drives GAL4 expression both in the longitudinal glia (in 1-3 hemisegments per embryo) and the dMP2 and vMP2 neurons (in most segments). Only embryos in which ablation had taken place were analysed. Embryos in which ablation had not taken place were identified by the expression of lacZ from the reporter balancer chromosomes, which was visualised with anti-β-gal antibodies. Hemisegments where glial ablation was verified by staining with anti-Repo were analysed. In ablations with sgcmGAL4 151, neighbouring or adjacent non-ablated hemisegments in the same embryos were used as controls for normal trajectories at the same stage.

Antibody stainings were carried out following standard procedures, using the Vectastain Elite kit from Vector Labs, and NiCl was used for colour intensification when necessary. Anti-Repo was used at 1:300 (gift of Travers); anti-Heartless at 1:1000 (gift of Hosono); fasII at 1:5 (gift of Goodman); 22c10 at 1:10 (gift of Patel). For rhodamine-phalloidin staining, embryos were fixed in 80% ethanol, incubated first with 22c10 and subsequently with rhodamine-phalloidin (gift of Martin-Bermudo) for 40 minutes together with FITC anti-mouse.

At the beginning of axonogenesis (stage 12.3), the growth cones of pioneer neurons pCC and vMP2 extend together anteriorly, as the growth cones of MP1 and dMP2 extend posteriorly (Bastiani et al., 1986; Jacobs and Goodman, 1989b; Hidalgo and Brand, 1997; Lin et al., 1994). Antibody 22c10 recognises the axons of dMP2 and vMP2, whereas fasII recognises the axon of pCC.

Glia migrate ahead of extending growth cones. When the progeny of the glioblast reach the cell bodies of the pioneer neurons, they stop migrating in the dorsoventral direction and cluster around the neurons. At stage 12.3, the growth cones of dMP2 and vMP2 extend towards the glia (Fig. 1A). Glia do not form a prepattern for the axonal trajectories, as most of the segment at this stage is free of glia (Fig. 1B). Glia migrate posteriorly together with pioneer growth cone extension and, in particular, the axon of dMP2 follows the glia (Fig. 1B) sending projections towards them (Fig. 1D). The growth cone of vMP2/pCC appears to extend further than the most anterior glia, as visualised with the nuclear marker Repo (Figs 1B, 2A; for Repo see Campbell et al., 1994; Halter et al., 1995; Xiong et al., 1994). However Heartless-positive glial cytoplasm (Shishido et al., 1997) abuts the pCC axon (Fig. 2B) and glial projections extend ahead of the pCC/vMP2 growth cone (Figs 1G,H, 2C,D). dMP2/MP1 and pCC/vMP2 meet over a glial cell to form the first longitudinal pathway, which by stage 13 is covered in glia (Figs 1C, 3F, 4D).

The distance travelled by the vMP2 and dMP2 growth cones is rich in cell membranes, revealed with rhodamin-phalloidin (Fig. 1E). Glial projections contribute to this membrane mesh. Longitudinal glia coexpress the nuclear protein Repo and the membrane protein Heartless (Fig. 1F). At the time of pioneer growth cone extension and prior to their fasciculation, Heartless-positive glial projections reach across adjacent segments and make glia-to-glia contact (Fig. 1F-H).

Neuronal filopodia also cover the distance to be travelled by the pioneer growth cones. The pCC growth cone extends long filopodia that contact glia from the adjacent anterior segment prior to fasciculation with the dMP2/MP1 growth cone (Fig. 2C). While the vMP2 and dMP2 growth cones are still far apart, the pCC/vMP2 growth cone also makes contact with intermediate neurons, which can be visualised with fasII (Fig. 2D) and 22c10 (Fig. 2E). One of these intermediate neurons is SP1 (Fig. 2E; see also Lin et al., 1994). We did not detect coexpression of Repo and fasII at an intermediate cell, as implied in previous work by Lin et al. (1994). Glia migrate along the cell bodies of fasII-positive cells located ahead along their trajectory (Fig. 2A). It appears that the same intermediate neurons are contacted by both growth cones and migrating glia.

The fact that glia precede extending growth cones during axonogenesis and that growth cones send filopodia towards glia suggests that glia attract pioneer growth cones. However, along the trajectory of migrating glia and extending growth cones, there are intermediate neurons that are contacted by both cell types. This suggests that bidirectional contact between neurons and glia occurs during the establishment of the first longitudinal trajectory.

The role of glia in guidance was studied by ablating longitudinal glia with Ricin expression driven by the GAL4 system (see Materials and Methods; Brand and Perrimon, 1993; Hidalgo et al., 1995) and by studying glial cells missing (gcm) mutations, in which glia are transformed towards a neuronal fate (Hosoya et al., 1995; Jones et al., 1995; Vincent et al., 1996). The s-gcmGAL4 line used drives expression in the longitudinal glioblast in a mosaic fashion, so that in most embryos only one or two glioblasts are killed (Booth et al., 2000). This line is also expressed in other glial types; however, expression is also mosaic in these glia so that secondary damage to the ventral nerve cord is minimised. Only the catalytic subunit (A) of Ricin is expressed, such that the toxin cannot exit the expressing cells. We have shown in previous work that targeted ablation with Ricin-A is cell autonomous, since cells adjacent to ablated cells are unaffected (Hidalgo et al., 1995).

At the beginning of axonogenesis (stage 12.3), dMP2 extends posteriorly, fasciculating with MP1 and aCC and surrounded by glia for a short distance (Fig. 3A). The growth cone branches at the location of a glial cell (choice point 1; Fig. 3B,G): one branch carries on towards the muscle (the aCC axon), and the other branch (dMP2) heads posteriorly to meet the vMP2/pCC growth cone (choice point 2; Fig. 3E,F,H). When the glioblast is ablated, the dMP2 growth cone extends more slowly, as the axon is shorter than that in a normal adjacent segment (n=5/7 ablated hemisegments; Fig. 3C) or it does not meet the pCC/vMP2 fascicle and it appears thickened and with a large growth cone (n=5/7). However, dMP2 does eventually extend and, slightly later, the growth cone may not branch out: the dMP2/MP1 axon heads towards the muscle fasciculating with aCC (n=2/7; Fig. 3D,I).

At stage 13, normally the ascending growth cone of vMP2 and descending dMP2 meet and fasciculate over a glial cell (Figs 1C, 3F) to form the first longitudinal single fascicle (Fig. 4D). Ablation of longitudinal glia can lead to failure of contact and fasciculation between pCC/vMP2 and dMP2/MP1 fascicles (n=5/9) or to loss of the dMP2/MP1 pathway (n=6/18; Fig. 4E,F). In many cases, formation of the first fascicle is not affected by the absence of glia.

These data suggest that glia attract the dMP2/MP1 growth cone, provoke axonal defasciculation at the branching choice point and aid fasciculation of all growth cones into a single fascicle (stage 13).

Glia are dispensable for the extension of the pCC/vMP2 growth cone, since ablation of longitudinal glia does not affect the pCC axon (n=27/27; Fig. 4B,C,E,F). No effect was observed when either the glia from its own segment or from the adjacent anterior segment were ablated. The pCC growth cone extends and stalls at the same position as in normal segments. Similarly, pCC axon extends normally and stalls in gcm mutants (Hosoya et al., 1995; Jones et al., 1995). To test whether pCC was attracted jointly by glia and the dMP2 growth cone, we drove GAL4 expression both in the MP2 neurons and the glioblast (see Materials and Methods). Ablation of glia, dMP2 and vMP2 causes a range of defects in 20% of cases (n=30) including dramatically enlarged growth cone (Fig. 5A), shorter or longer pCC axons (Fig. 5B,C), abnormally tortuous trajectory (Fig. 5C) and fasciculation with the RP2 axon. These data suggest that communication between the pCC/vMP2 and dMP2/MP1 growth cones does not drive pathfinding, but it aids the encounter between them to form the first longitudinal fascicle.

By stage 14, pCC/vMP2 axons defasciculate from MP1/dMP2 axons to form two fascicles within each segment, the pCC/vMP2 fascicle lying closer to the midline, which fasciculate again for a short distance posterior to the pCC cell body (Hidalgo and Brand, 1997; Lin et al., 1994). Just before this defasciculation event, glia cluster within the concave side of the sinuous pathway of the single fascicle (Fig. 4D). Subsequently, glia are found at choice point 3 where the two fascicles separate, and also at choice point 4 where the two fascicles refasciculate again (Fig. 4G,M).

To achieve the stage 17 pattern of two pioneer fascicles, MP1 must defasciculate from dMP2, and pCC must defasciculate from vMP2 and refasciculate with dMP2 (Hidalgo and Brand, 1997). Consequently, at stage 17, the axons of dMP2 and pCC extend along the first fasII fascicle and the MP1 axons extend along the second one. We have found that during this transition period at stage 14 the axons of MP1, dMP2 and pCC defasciculate from each other transiently to form three thin fascicles (Fig. 4O). These thin fascicles do not correspond to the final stage 17 fas II fascicles, since dMP2 will fasciculate with pCC at stage 15 and since the pioneer axons only contribute to two of the final three fasII fascicles.

As these defasciculation events take place, glia are aligned in two rows at either side of the dMP2/MP1 fascicle but not along the pCC/vMP2 pathway (Fig. 4H,N). Slightly later, glia detach from this pathway and appear aligned at either side of a third thin fascicle (Fig. 4I,O). This third fascicle probably corresponds to the MP1 axons, the second fascicle to dMP2 and the first to pCC (Fig. 4M-O). These fascicles appear only transiently.

When glia are ablated, whereas the pCC/vMP2 fascicle is generally present (n=19/21), no dMP2/MP1 fascicle can be seen at stage 14 (n=10/21; Fig. 4K,P). Loss of the dMP2/MP1 fascicle is not due to neuronal death, since we did not observe loss of pioneer neuron cell bodies (see Booth et al., 2000). Instead, the pCC fascicle and the aCC axons are thicker, suggesting that the dMP2/MP1 fascicle fails to defasciculate from the pCC/vMP2 or aCC fascicles in the absence of glia (Fig. 4L,P). The three thin fascicles described above are not present either. Hence, glia control defasciculation of MP1, dMP2 and pCC at the stage 14 to 15 transition.

At stage 15 the three thin fascicles stretch out under two tightly packed columns of glia (Figs 4J, 6A), and dMP2 fasciculates with pCC. Following this last fasciculation, pioneer axons run along two fasII fascicles: pCC/dMP2 along the first one, and MP1 along the second one. Ablation of glia leads to a single fascicle (Fig. 6D).

From stage 16 to 17, glia are located in two discrete rows in between the three forming fascicles, as visualised with fasII (Fig. 6B,C). When glia are ablated, the separation of the three longitudinal fascicles is damaged in 90% of examined segments (n=83). The three longitudinal fascicles can appear fused into one single fascicle along the pCC pathway (41%, Fig. 6E). Fasciculation defects may affect only the second or third fascicles, fascicles can be missing, may misroute towards the muscle (Fig. 6F) or midline. In the most severe cases, the first fascicle is also damaged (9.8%). When glia, dMP2 and vMP2 are all ablated, the first fascicle had defasciculated, broken, misrouted or was missing in 42% of cases (n=13; Fig. 5D). This implies that dMP2 and vMP2 contribute to the single fused fascicle resulting from glial ablation. It also implies that elimination of dMP2 and vMP2, whose axons run along the pCC pathway, further depletes axons of fasciculation cues.

These data demonstrate that glia lead the defasciculation and refasciculation events that drive the establishment of pioneer axon trajectories.

glial cells missing (gcm) mutant embryos virtually lack glia and sometimes lack, but can also form, longitudinal tracts (Hosoya et al., 1995; Jones et al., 1995; Vincent et al., 1996).

Whereas the pCC growth cone extends and then stalls normally in gcm mutants, the axon of dMP2 extends but it may or may not contact vMP2 (Fig. 7A; see also Hosoya et al., 1995; Jones et al., 1995). Eventually, the first fascicle does form at stage 13. Although this may suggest that glia play no role in guidance, in gcm mutants, there are many extra fasII neurons along the route of the longitudinal axons (Fig. 7B). Fas II is thought to play a role in fasciculation of pioneer axons and their contact with glia (Grenningloh et al., 1991; Lin et al., 1994). This means that, although the transformed cells do not express glial markers, they are located along the longitudinal pathways and may be instructive to extending growth cones, by which they are contacted. By stage 14, both pCC/vMP2 and dMP2/MP1 fascicles may be formed normally (6.8% of hemisegments, n=44), but in most cases (93%) they are either missing or fused into one single thick fascicle (Fig. 7B). By stages 16/17, longitudinal fascicles can also be formed, even if with fasciculation problems (Fig. 7C,D; Hosoya et al., 1995; Jones et al., 1995): longitudinal tracts are fuzzy (Fig. 7C), dramatically defasciculated (Fig. 7D), fused into a single fascicle (Fig. 7D), misrouted towards the muscle (Fig. 7D) or across the midline (83% hemisegments affected, n=196). At stages 15/17 as well as earlier on, fasII reveals multiple round cells remarkably reminiscent in their location of the longitudinal glia, distributed along the longitudinal fascicles (Fig. 7C). This suggests that the transformed cells are still present in the places normally occupied by glia and therefore they could provide information to extending growth cones. This may explain why in some cases the fascicles appear to form better in gcm mutants than when glia are ablated. Nevertheless, the fasciculation defects also recapitulate earlier fasciculation patterns and reveal missed defasciculation and refasciculation of axons.

What are the consequences of the absence of glia and abnormal fasciculation of pioneer axons to the pathfinding by the majority of CNS neurons?

Follower axons were monitored using BP102 antibodies, which label many CNS axons running along the longitudinal pathways from stage 14 but do not label the pioneer axons. In 56% of the cases in which glia had been ablated, the longitudinal axon tracts were missing, thin or broken, or had misrouted towards the muscle (Fig. 8B) or towards the midline (n=85). At stages 16-17, axonal damage was seen in 53% of the hemisegments examined, versus 62% at earlier stages 14-15. The expressivity of this phenotype is higher than that of gcm mutants (40% abnormal hemisegments in Hosoya et al., 1995), suggesting that gcm mutants retain more information than ablated embryos. Whereas neuronal apoptosis induced upon glial ablation is likely to contribute to phenotypes of missing, thin or broken BP102 axons (see Booth et al., 2000), diversion of BP102 axons towards the muscle along the intersegmental nerve or towards the midline reflect a change in axonal pathway. These data show that ablation of glia alters the pioneer axon and glia scaffold diverting the trajectories of follower axons.

We have shown that glia are required for axon guidance during the formation of the embryonic Drosophila CNS. Pathfinding along the longitudinal pathways encounters two problems: (1) ascending and descending growth cones have to navigate and finally meet to form the first longitudinal fascicle, and (2) a final pattern of three longitudinal fascicles, of which two are pioneered by pCC, vMP2, dMP2 and MP1, has to emerge from an initial single fascicle (Fig. 9). We have shown that four types of events require glia: (1) glia-to-glia and glia-to-neuron contact (through projections across the segment), (2) growth cone attraction (e.g., of dMP2), (3) defasciculation (e.g. of dMP2 from aCC, of dMP2 from MP1, of pCC from vMP2), and (4) fasciculation (e.g. contact between the pCC/vMP2 and the dMP2/MP1 fascicles, fasciculation of dMP2 with pCC).

It has been previously suggested that glia form a prepattern prior to growth cone extension (Jacobs and Goodman, 1989; Jacobs et al., 1989; see also Bate and Grunewald, 1989). We have observed that glia do not form a prepattern, but instead migrate together with growth cone extension. Nevertheless, glia do migrate ahead of the dMP2/MP1 growth cone and growth cones appear to be attracted to glia, since they project filopodia towards them. When glia are ablated, the growth cone of the dMP2/MP1 fascicle grows more slowly.

The distance across a segment covered by the descending dMP2/MP1 and the ascending pCC/vMP2 growth cones is rich in both neuronal filopodia and glial projections. Both pioneer growth cones and migrating glia also contact intermediate neurons, located half way along the distance between neuromeres (see also Lin et al., 1994). Filopodia and large glial projections had also been observed ahead of MP1 growth cones in grasshopper (Bastiani et al., 1986; Bastiani and Goodman, 1986). This implies that long before growth cones contact each other, intermediate neurons and cytoplasmic processes from both neurons and glia provide a cellular route for growth cones to follow. Given this complex cellular mesh it is not surprising that elimination of longitudinal glia either through ablation or gcm mutations does not completely prevent formation of longitudinal fascicles (see also Hidalgo et al., 1995; Hosoya et al., 1995; Jones et al., 1995; Vincent et al., 1996). Similarly, we had previously shown that ablation of individual pioneer neurons did not prevent their formation either (Hidalgo and Brand, 1997). Hence, neither glia nor individual pioneer neurons are absolutely required for growth cone guidance, but instead their interaction is crucial to form axonal pathways. In this context, we have found an interesting difference in the behaviour of the pCC/vMP2 versus the dMP2/MP1 growth cone. Whereas the latter seems more sensitive to glial depletion, neither glial ablation nor gcm mutations had major consequences on the pCC axon(this work; (Hosoya et al., 1995; Jones et al., 1995). Interestingly, when we ablated the dMP2 and vMP2 neurons as well as glia, the pCC growth cone was slightly affected. This suggests that the pCC/vMP2 growth cone may sense signals from the dMP2/MP1 growth cone prior to contact.

Patterns of pioneer axons change with time (Hidalgo and Brand, 1997). We have shown that these changes in axonal fasciculation are controlled by glia (Fig. 9). In the wild-type CNS, we have found glia at the choice points where fasciculation and defasciculation of axons take place (Fig. 9). In the absence of glia from these positions, either through ablation or gcm mutations, pioneer neurons do not defasciculate from sister axons. This leads to axonal phenotypes that recapitulate the fasciculation patterns typical of earlier developmental stages. For instance, at choice point 1 where dMP2 normally separates from aCC to head posteriorly, ablation of glia leads to the misrouting of dMP2 towards the muscle, as it does not defasciculate from aCC. Similarly, whereas normally at choice point 2 the two dMP2/MP1 and vMP2/pCC growth cones meet and fasciculate together, absence of glia can prevent contact between the two growth cones. Hence, when glial function is compromised, more axons take the route of the intersegmental nerve, towards the muscle. Glia are also located at choice points 3 and 4, where pCC/vMP2 defasciculates from dMP2/MP1 and subsequently along the separating dMP2 and MP1 axons. By the end of embryogenesis, instead of the normal three fascicles, absence of glia often leads to fusion of the fascicles into a single fascicle. A single fascicle does form, consistent with the fact that formation of the first longitudinal fascicle is not severely affected by loss of glia.

Several evidences indicate that these recapitulation phenotypes are due to failed defasciculation. Firstly, fascicles are thicker than their normal counterparts. Secondly, we have not detected death of pioneer neurons which could account for axonal loss (Booth et al., 2000). Thirdly, upon glial ablation, the pCC growth cone stalls at its normal position. We have only detected longer pCC axons when the MP2s are also ablated. Fourthly, when glia alone are ablated the first fasII fascicle is only affected in 9.8% of cases, whereas if the MP2s are also ablated this fascicle is damaged in 42% of cases. This means that MP2 axons normally contribute to the first fused fascicle upon glial ablation. Fifthly, frequency of defects increase from 30% to 90% as development procedes upon glial ablation. In gcm mutants, the first fascicle generally forms at stage 13, whereas axonal defects by the end of embryogenesis are found in 83% of embryos. To conclude, our data demonstrate that glia drive the defasciculation and refasciculation events that shape pioneer axon patterns.

It is difficult to estimate the requirement for glia on the extension of most follower axons. We have shown that BP102-expressing follower axons can misroute away from the longitudinal pathways and along the intersegmental nerve when glia are ablated. However, this may not illustrate the full extent of the effect of absence of glia on follower trajectories. In fact, mosaic glial ablation may be bypassed by extending axons, as mosaic neuronal ablation is (Bastiani et al., 1986). Consequently, ablation of all longitudinal glia could lead to a more severe phenotype than the one that we observe. Similarly, gcm mutations do not correspond to a situation of lack of glia, since there are fasII-expressing neurons along the route of axonal extension which may provide comparable information. Secondary consequences, such as neuronal death and neuronal debris may lead to pathfinding defects obscuring the direct roles of glia. Nevertheless, the most remarkable phenotype from our experiments is the general disruption of the scaffold formed by the three axonal facicles separated by glia. Both upon ablation and mutation this scaffold is damaged. Follower axons may still extend along the longitudinal pathways; however, their selective fasciculation routes are altered. It would be interesting to know if these defects lead to altered connections of follower axons with their targets.

We thank P. Badenhurst, C. Goodman, K. Hosono, L. Martin-Bermudo, N. Patel and A. Travers for reagents; A. Brand, in whose laboratory the sgcm-GAL4 construct was made; M. Georgiou and E. Kinrade for technical help. G. B. holds an MRC studentship; this work was supported by a Wellcome Trust Career Development Fellowship to A. H.