Glia maintain follower neuron survival during Drosophila CNS development

Neurons and glia are in close contact in the mature nervous systems of all animals, yet little is known of the roles that neuron-glia interactions may play during nervous system formation. It has been suggested that all cells are programmed to die unless they receive trophic support and that cell survival largely depends upon interactions between cells (Raff et al., 1993). Both neurons and glia are overproduced in the normal nervous system, and there is evidence indicating that one consequence of axon-glia contact is survival regulation in both cell types in the mature nervous system (Raff et al., 1993). This ensures that axons are correctly myelinated, enabling normal neuronal function. During development, regulation of cell survival is a means of controlling cell number, to achieve normal organ size. The regulation of neuronal survival during development is poorly understood. In addition to governing cell number, survival control might conceivably be instructive in patterning. Interestingly, the neurotrophic factor BDNF has been shown to affect morphological differentiation of developing CNS neurons (Altar et al., 1997; Schwartz et al., 1997). Here we show that, during formation of the CNS of Drosophila, there is a differential requirement for glia to maintain neuronal survival and that this differential requirement is instructive in patterning the embryonic CNS.

Cell culture and knockout experiments in vertebrates have revealed that differentiated glia are involved in the maintenance of neuronal survival in the mature nervous system; however, this may not necessarily be the case during development. For instance, astrocytes support neuronal survival in cell culture (Lindsay, 1979; Petroski et al., 1991), microglia can induce neuronal apoptosis following excitotoxic injury to the brain (Rogove and Tsirka, 1997) and axons degenerate in mice lacking the major proteolipid of myelin, suggesting that interactions between axons and glia are required for the maintenance of neuronal survival once myelination has been completed (Griffiths et al., 1998). Glia are also involved in the maintenance of neuronal survival in the mature PNS, since Schwann cells and astrocytes express CNTF, and its elimination leads to motorneuron degeneration (Masu et al., 1993). However, the survival requirements of neurons differ between the developing and mature nervous systems, since knockout mice lacking neurotrophic factors CNTF and GDNF develop normally (Masu et al., 1993; Sanchez et al., 1996).

More recently, it has been shown that glia maintain neuronal survival during vertebrate PNS development. ErbB3 knockout mice, which lack Schwann cells, display dramatic motorneuron degeneration (Riethmacher et al., 1997). Furthermore, during PNS development, motorneurons do not depend solely on target-derived factors for survival (Raff et al., 1993), but require different neurotrophic factors as development proceeds (Buj-Bello et al., 1994). Interestingly, the switch over in neurotrophic dependence appears to be regulated by extrinsic signals that neurons encounter in vivo (Paul and Davies, 1995).

Maintenance of neuronal survival in the CNS appears to be quite unlike that in the PNS. For instance, whereas survival of motorneurons depends on potent neurotrophic factors such as GDNF (Oppenheim et al., 1995; Yan et al., 1995), CNS neurons survive in knockout mice lacking neurotrophic factors such as BDNF (Schwartz et al., 1997). Knockout mice lacking neurotrophic factors do not display dramatic CNS phenotypes, suggesting that survival regulation during CNS development is more complex than envisioned by the neurotrophic theory (Snider, 1994). It is believed that CNS neurons require multiple neuronal and non-neuronal factors for their survival (Snider, 1994).

Previous work has suggested a role for glia in the maintenance of neuronal survival in adult flies, since viable mutants for the genes repo (Xiong and Montell, 1995) or drop dead (Buchanan and Benzer, 1993), which are expressed in glia, have damaged eyes or brains with increased neuronal degeneration. During the formation of the embryonic CNS, ablation of interface glia subsequent to axonal trajectory establishment (Hidalgo et al., 1995) and mutations of the glial determination gene glial-cells-missing (gcm, which cause a transformation from glial to neuronal fate; Hosoya et al., 1995; Jones et al., 1995) result in damaged or missing longitudinal axons (Fig. 3B). Such axonal defects may be due partly to altered path finding in the absence of normal glial contact.

The ventral nerve cord of the CNS of Drosophila embryos is formed of two longitudinal axon-glia bundles separated by the midline, and joined by two commissures that cross the midline in each segment. The longitudinal pathways are established by the axons of the segmentally repeated pioneer neurons pCC, MP1, dMP2 and vMP2 (Bate and Grunewald, 1981; Bastiani et al., 1986; Jacobs and Goodman, 1989; Hidalgo and Brand, 1997; Lin et al., 1994). The pioneer neurons dMP2 and MP1 fasciculate together and extend posteriorly until they meet the anteriorly extending pCC and vMP2 growth cones, to form the first single longitudinal fascicle. Once the axonal skeleton of the CNS is set up by the axons of the pioneer neurons, the later extending follower neurons project their growth cones, cross the midline once and, after fasciculating with particular pioneer axons, extend contralaterally towards the brain. Interface glia enwrap the CNS longitudinal axons and, in some ways, resemble vertebrate oligodendrocytes (Ito et al., 1995). A subset of the interface glia derive from the segmentally repeated glioblasts, which originate from the lateral neuroectoderm and subsequently divide and migrate towards the midline to form the longitudinal glia (Schmidt et al., 1997).

Here we explore whether glia are required for neuronal survival during CNS development in Drosophila embryos, by looking at glial cells missing mutants and by ablating CNS glia in vivo. We target ablation of different glial classes, prior to and following longitudinal axon pathway formation. We show that, whereas the pioneer neurons that establish the axonal trajectories do not depend on glia for survival the later extending follower neurons do. This differential neuronal requirement reveals a role for survival control in the establishment of axonal trajectories, providing a link between the control of cell number and axon guidance in nervous system formation.

(1) gcm^ΔP1/CyOlacZ (Jones et al., 1995); (2) enhancer trap GAL4 lines C321c (Hidalgo et al., 1995) and 158 (gift of C. Mirth, J. Castelli-Gair and M. Akam) (3) w; UASRicin/CyOlacZ (Hidalgo et al., 1995); (4) w; UAS tau lacZ (Hidalgo et al., 1995); (5) w; UAS GAP GFP (gift of A. Chiba); (6) reaper: Df(3L)H99/TM6B (White et al., 1994; gift of J. Castelli-Gair); (7) w;UAS Reaper/TM3 lacZ (gift of U. P. John, G. Halder and A. H. Brand; unpublished); (8) Double GAL4-reaper: w; sgcmGAL4 211; Df(3L) H99/SM6a-TM6B; (9) Double ablation-reaper: w;UAS Reaper-Df(3L)H99/TM6B; (10) lacZ reporter balancers: CyOen¹¹, drives expression in epidermal stripes; CyOftzlacZ and TM3ftzlacZ, drive lacZ expression in the nervous system and in epidermal stripes.

Glia were ablated by expressing the toxin RicinA (Hidalgo et al., 1995) with the GAL4 system (Brand and Perrimon, 1993). We used four independent GAL4 lines. (1) s-gcmGAL4 151 and 211 drive GAL4 expression in the glioblast (Fig. 1A), that is, before axon extension, and in surface and cortex glia (Fig. 1B). Expression with these lines is mosaic: only one or two glioblasts and a few other glia express GAL4 in each embryo. Hence, ablation does not cause severe nerve cord damage. Sporadic expression occurs in neurons, in stage 16 embryos in 151 (Fig. 1E), but more often in the MP2s in line 211. All experiments used line 151 unless indicated. These lines also express GAL4 in some, but not all, macrophages (Fig. 2B), like the gcm gene (Bernardoni et al., 1997). (2) Interface glia line (C321c) causes the ablation of only the interface glia (longitudinal plus other neuropile associated dorsal glia) in the CNS from stage 13, when the first longitudinal axons have already extended (Hidalgo et al., 1995; Fig. 1C). C321c has sporadic expression in neurons (Fig. 1I) and it is not expressed in macrophages (Fig. 2A). (3) General glia line (158) is expressed in surface, cortex and interface glia from stage 13 (Fig. 1D,J), occasionally in the glioblasts, and in some but not all macrophages (Fig. 2C). Expression is mosaic and was not observed in neurons (Fig. 1H). None of the lines used express in the midline glia.

(1)sgcmGAL4 transformant flies were generated by synthesising an oligo with 11 repeats of the consensus GCM-binding sequence with 5′BamHI and 3′SacII restriction sites, which was fused to the hsp70 promoter and to a partial 5′HindII-3′NotI digest of pGaWB to retrieve the coding region of GAL4 (Brand and Perrimon, 1993). The synthetic gcm enhancer and hsp70 were fused and subcloned first into the 5′BamHI and 3′HindII sites of bluescript. Fusion of gcm-hsp70 and GAL4 was achieved through a three way ligation into the BamHI and NotI sites of the pCaSpeR 4 transformation vector. The construct was injected into yw; Δ2-3 Sb/TM6 embryos. 151 is an insertion into the X and 211 into the II chromosomes. (2) For elav-p35 flies, the 3.5kb neuronal elav promoter (gi/to/k.white Yao and White, 1994) was digested with NotI and filled to fuse it to the blunt 5′ end of the EcoRI digested and filled fusion of the coding region of p35 (gift of W. Gehring) and the SV40 terminator. The elav promoter and p35-SV40 were inserted into the pCaSpeR 4 transformation vector by a three-way ligation into the 5′ KpnI and 3′ BamHI sites. The construct was injected into yw; Δ2-3 Sb/TM6 embryos. elav-p35 9.1 is an insertion into the second and 10.4 into the X chromosomes.

p35 rescue experiments were carried out by crossing sgcmGAL 151/w; elav p35 9.1/CyOlacZ females to elav p35 10.4; UASRicin/ CyOftzlacZ males. Balancer chromosomes were identified with anti-βgal antibodies; ablated embryos were identified with glial anti-Repo antibodies; all non-βgal, ablated embryos express p35. 50% ablated embryos have two copies of p35 and 50% one copy.

These were carried out following standard procedures, using mRNA DIG-labelled probes (Ambion), by predigesting p35 from Bluescript with EcoRI and transcribing with T7, and predigesting reaper (gift of H. Steller) from Bluescript with BglII and transcribing with SP6, and incubating at 55°C overnight in hybridisation solution. Antibody staining was carried out with anti-Repo at 1:300 prior to in situ hybridisation in the case of Fig. 5F.

Embryos were stained with anti-Repo at 1:200 or anti-Elav at 1:5 and visualised with Texas Red-conjugated secondary antibodies following standard procedures. Embryos were also stained with anti-β-gal to identify individuals carrying balancer chromosomes, which were then discarded. Subsequently the embryos were treated with 50 μg/ml Proteinase K for 1.5 minutes, blocked with 2 mg/ml glycine and fixed in 4% formaldehyde in PBS 0.1% Triton X. Following washes in PBTriton, they were washed in terminal transferase buffer according to manufacturers (Boehringer Mannheim) and incubated in terminal transferase for 3 hours at 37°C in the presence of FITC-UTP (Boehringer Mannheim). Embryos were washed in PBTriton and mounted in Vectashield (Vector Labs).

These were carried out following standard procedures (Patel, 1994), using biotinylated secondary antibodies and the Vectastain Elite kit (Vector Labs) with NiCl to intensify colour where required. Rabbit anti-Repo was used at 1:300 (gift of A. Travers), mouse BP102 at 1:10 (gift of N. Patel), mouse 22c10 at 1:10 (gift of N. Patel), mouse fasII at 1:5 (gift of C. Goodman), rabbit anti-β-gal at 1:5000 (Cappel), mouse anti-Elav at 1:5 (gift of C. Klämbt), guinea-pig anti-Odd at 1:500 (gift of M. Ruiz-Gómez), mouse anti-Eve at 1: 10 (gift of N. Patel), and rabbit anti-GFP at 1:200 (Molecular Probes).

Glia were ablated by expressing the catalytic subunit of Ricin toxin with the Gal4 system (Brand and Perrimon, 1993). This subunit is unable to leave the cell, hence leading to cell autonomous ablation (Hidalgo et al., 1995). Three different lines of GAL4 flies were used (see Materials and Methods). (1) Mosaic line (sgcmGAL4 151), which is expressed in several glia including the longitudinal glioblast, in a mosaic fashion (Fig. 1A,B), and sporadically in neurons (Fig. 1E) and macrophages (Fig. 2B); (2) interface glia line (C321c) expressed in interface glia (Fig. 1C,G; Hidalgo et al., 1995), sporadically in neurons (Fig. 1I) and not in macrophages (Fig. 2A), and (3) general glia line (158), expressed in most glial classes (Fig. 1D,J), except the midline glia, in macrophages (Fig. 2C) and not in neurons (Fig. 1H). Ablations were carried out with the mosaic line (151) unless otherwise indicated (i.e. see Fig. 6). Glial ablation does not completely eliminate macrophages (Fig. 2E,F).

Glial ablation at the glioblast stage, that is prior to axonal contact, leads to loss of longitudinal axons in 56% of hemisegments (n=85 hemisegments) as visualised with the axonal marker BP102 (Fig. 3C, compare with Fig. 3A). This is reminiscent of the gcm mutant phenotype (axonal loss in 51% of hemisegments with no glia, n=318; Fig. 3B). Later ablation of interface glia (with C321c; Hidalgo et al., 1995), once the first longitudinal fascicles have been established by the pioneer neuons, or ablation of other glia (158) more often causes thinning of longitudinal tracts (Fig. 3D). Axonal defects caused by early glial ablation could be due partly to altered guidance (see Hidalgo and Booth, 2000). Nevertheless, when axons are visualised with the neuronal marker fasII, there are also clear cases of axonal loss (Fig. 3G). Thinning and axonal degradation following ablation of glia once longitudinal tracts are established suggests that elimination of glia may lead to neuronal loss.

To investigate whether the missing longitudinal tracts reflect neuronal loss, we asked if pioneer neurons disappear upon glial ablation. The pioneer neurons MP1 and dMP2 express the nuclear marker Odd (Ward and Coulter, 1997) and are both present despite being isolated from neighbouring glia (Fig. 4A,B), and are occasionally duplicated. The pioneer neurons vMP2, MP1 and dMP2 were monitored with the membrane marker fasII at stage 14 (Hidalgo and Brand, 1997; Lin et al., 1994) and were present in hemisegments with ablated interface glia (Fig. 4C,D). The cell bodies of the longitudinal pioneer neuron pCC and the motorneurons aCC and RP2, were visualised with anti-Eve antibodies (Patel, 1994). When longitudinal glia are ablated, RP2 is present (Fig. 4F, compare to Fig. 4E) and is occasionally duplicated. Ablation of the glioblast alone leaves other interface glia, which contact aCC and pCC. However, in severe embryos where more interface glia are ablated, pCC and aCC as well as RP2 also remain in the absence of neighbouring glia (Fig. 4G). We cannot rule out the possibility that progenitors of pioneer neurons may undergo further cell divisions to maintain a constant number of pioneer progeny in the absence of glia. Nevertheless, we did not observe net loss of pioneer neurons when interface and other glia are ablated. This suggests that pioneer neurons do not depend on glia for survival.

Pioneer neurons are only four of 320 neurons per hemisegment that can potentially send axons along the longitudinal pathways. Hence, we wished to know if other neurons die following glial ablation. We monitored neuronal apoptosis in wild-type, ablated embryos and gcm mutants. Apoptosis was revealed by TUNEL (Gavrieli et al., 1992; Fig. 5A-E) and the neuronal nature of dying cells was verified by their lack of the glial differentiation marker Repo (Patel, 1994; Fig. 5A,E) and by their coexpression of the neuronal marker Elav (Patel, 1994; Fig. 5B-D). Apoptotic neurons display a rounded and shrunken morphology (Schwartz and Osborne, 1995), confirmed by the expression of transcripts from the cell-death-inducing gene reaper (White et al., 1994; Fig. 5F). Shrunken neurons were also observed with the neuronal markers 22c10 (Fig. 5G,H) and fasII (Fig. 5I,J; Patel, 1994). We did not observe such shrunken 22c10- and fasII-positive neurons in embryos lacking apoptosis (reaper mutants; White et al., 1994; n=17 and n=9, respectively). Since neuronal apoptosis occurs during normal development (White et al., 1994), we compared the frequency of neuronal apoptosis after glial ablation to that in the wild type. Following glial ablation with sgcmGAL4, clusters of two to four apoptotic Repo-negative cells were detected by TUNEL (Gavrieli et al., 1992; Fig. 5A,E). These were not observed in wild type. Similarly, clusters of two or three TUNEL- and Elav-positive cells were found in ablated embryos but not in wild type (Fig. 5B-D).

We wished to investigate whether follower neurons die after glial ablation. The neuronal marker BP102 stains axons of most CNS follower neurons, but not their cell bodies. Apoptotic cells were observed both in wild-type and ablated embryos with BP102 (data not shown). However, because cell bodies are not stained, we could not always distinguish with certainty axonal debris resulting from apoptosis from individual apoptotic cells. The markers 22c10 and fasII stain motorneurons and subsets of follower neurons of the longitudinal pathways from stage 14.

These markers stain both cell bodies and axons, making identification of individual apoptotic cells unambiguous. Quantification of the incidence of follower neuron apoptosis in wild-type and ablated embryos reveals an increase in both the frequency (with 22c10, from 10% wild-type up to 43% ablated embryos) and severity of the phenotype (Figs 5H, 6A-D). These effects were observed both with 22c10 and fasII neuronal markers and increases in frequency are statistically significant (Fig. 6A-D). Most neuronal apoptosis in wild type occurs around stage 13. We have analysed embryos with 22c10 from stage 14 to stage 17, to make a clearer distinction from normal neuronal death. We also observed fasII apoptotic neurons in 89% of gcm mutants (n=19), in which there were generally high numbers of apoptotic neurons (up to 16) frequently clustered in groups of 2-6 (Figs 5J, 6C,D). We did not observe 22c10-positive apoptotic neurons in gcm mutants. Our data show that compromising glial function either through ablation or mutation results in increased apoptosis amongst follower neurons.

Several glial classes may contribute to the maintenance of neuronal survival. We monitored glial loss with the marker Repo, and detected neuronal apoptosis both in ablated embryos in which interface glia are present and in those in which they were ablated (Fig. 6A-D). Neuronal apoptosis increased in embryos where surface and cortex (but not longitudinal) glia are ablated (Fig. 6A). It has been shown that surface and interface glia can phagocytose apoptotic neurons in the normal embryo (Sonnenfeld and Jacobs, 1995). It is therefore conceivable that the observed increase in neuronal apoptosis following glial ablation may be partly due to a reduction in neuronal phagocytosis. However, we have shown that despite mosaic expression in macrophages in two of our lines, macrophages are not completely eliminated upon ablation (Fig. 2E,F). Furthermore, line C321c does not drive GAL4 expression in macrophages (Fig. 2A). Since macrophages are still present in ablated embryos, the increase in neuronal apoptosis is unlikely to be due to defective phagocytosis alone. Furthermore, we have shown that glial ablation can lead to CNS phenotypes of neuronal loss (Fig. 3G).

Neuronal apoptosis was found when only interface glia and not macrophages were ablated (with C321c). Remarkably, neuronal apoptosis is induced in 40% (fasII, n=32) to 42% (22c10, n=35) embryos when only interface glia are ablated (C321c), even after formation of the first longitudinal axons (Fig. 6A-D). Since these glia normally enwrap the axons of the longitudinal tracts, contact of axons with glia is necessary for neuronal survival.

Ablation experiments with line 158, which is not expressed in neurons, and gcm mutants suggest that neuronal apoptosis is a non-autonomous consequence of interfering with normal neuron-glia interactions.

Nevertheless, lines C321c and sgcmGAL4 are expressed sporadically in neurons. Therefore, we estimated the extent of non-autonomous neuronal death in ablated embryos. Because Ricin toxin blocks protein translation (Endo and Tsurugi, 1988), ablated cells are not detectable using fasII and 22c10 antibodies. Therefore, for this purpose we ablated using Reaper, which causes cells to die through their normal apoptotic pathway (White et al., 1996). We drove UAS Reaper (U. P. John, G. Halder and A. H. Brand, unpublished) with sgcmGAL4 211, which is expressed sporadically in 22c10-positive MP2 neurons. 30% sgcmGAL4/UAS Reaper embryos (n=27) in which longitudinal glia had been ablated had 1-10 apoptotic 22c10-positive neurons. By contrast, in embryos mutant for reaper, this ablation results in the death only of cells expressing reaper cell-autonomously under GAL4 control. In this case, 77% (n=30) of embryos had no dying cells and 23% had a single dying cell. No embryos were found with more than one dying cell. This indicates that the multiple apoptoses observed in sgcmGAL4/UAS Reaper embryos are non-autonomous consequences of glial ablation.

If neuronal apoptosis is a non-autonomous consequence of glial ablation, it should be possible to rescue neuronal survival by blocking apoptosis. We expressed baculovirus p35, which blocks programmed cell death (Hay et al., 1994), in all neurons using the neuronal elav promoter (Yao and White, 1994). Neuronal expression of p35 in embryos where glia have been ablated with Ricin leads to a statistically significant reduction of neuronal apoptosis (n=30, 0.025<P<0.05, χ2>3.84 Fig. 6E). Rescue by p35 is dosage sensitive since we detected more apoptotic neurons when only one copy of p35 was expressed. These results demonstrate that apoptosis is induced non-autonomously in neurons following glial ablation.

We have shown that neuron-glia interactions are required for the maintenance of neuronal survival during the formation of the embryonic Drosophila CNS. A role for glia in the maintenance of adult neuronal survival had been suggested by the phenotypes of repo (Xiong and Montell, 1995) and drop-dead (Buchanan and Benzer, 1993) mutants. In vertebrates, glia secrete neuronal trophic factors (Masu et al., 1993; Petroski et al., 1991) and glial depletion in the PNS (Bush et al., 1998; Riethmacher et al., 1997) and impaired CNS oligodendrocyte differentiation (Griffiths et al., 1998) lead to neuronal apoptosis.

Here we show that impaired glial function in gcm mutants or in vivo ablation of glia induces non-autonomous neuronal death in the embryonic CNS. Targeted genetic ablation of glia in vivo in the embryonic CNS was achieved using the GAL4 system. We carried out three types of ablations: (1) ablation of interface and other glia prior to formation of axonal pathways, (2) ablation of interface glia alone after formation of the longitudinal axon pathways, and (3) ablation of cortex and surface but not interface glia. Ablations were carried out in a mosaic fashion, since the promoters driving GAL4 expression are not expressed in every glial cell of each embryo. That is, only subsets of glia were ablated at any one time, hence minimising nerve cord damage due to secondary effects. In all cases, we observed an increase in neuronal apoptosis. Furthermore, we show that neuronal apoptosis is induced non-autonomously since it can be rescued by blocking apoptosis in embryos where glia have been ablated.

Apoptotic figures do not reflect the full extent of neuronal death. Firstly, ablation is mosaic and therefore represents a fraction of neuronal death which would have occurred if all glia had been ablated. Secondly, dead cells are rapidly removed (Raff et al., 1993) and therefore we visualise a narrow time window. Thirdly, 22c10- and fasII-expressing cells do not represent all CNS neurons. Work in vertebrates has also indicated the difficulties in detecting realistic numbers of apoptotic cells (Raff et al., 1993). For instance, although there is massive death of oligodendrocytes in wild-type animals, only 4% of it is detected (Raff et al., 1993). Consequently, our quantification of neuronal death is very likely an underestimation of the real extent of neuronal death upon interference with glial function.

Our results show that contact between longitudinal glia and axons is necessary to maintain neuronal survival. Remarkably, ablation of longitudinal glia alone after the longitudinal axons had been established by the pioneer neurons, is sufficient to induce neuronal apoptosis. Our data also suggest that the neuronal requirement for glia may not be restricted to longitudinal glia. In fact, we observe neuronal apoptosis also when surface and cortex glia, but not longitudinal glia, are ablated. It was surprising to find that, in gcm mutations, neuronal apoptosis is not detectable with 22c10 antibodies, whereas it is with fasII. The increase in neuronal apoptosis with fasII is in fact more dramatic than in ablation experiments, which is consistent with the fact that abnormal glial function is not mosaic in gcm mutants. Klaes et al. (1994) have shown that glia can induce the expression of 22c10 in neighbouring neurons, and similarly abnormal glial function leads to a reduction in 22c10 in MP2 neurons. Perhaps the difference that we observe in 22c10 apoptotic neurons in gcm mutants and ablation experiments may be due to the effect that gcm mutations have on glial function. It has been shown that signalling from neurons to glia is also important to maintain survival of e oligodendrocytes (Barres and Raff, 1994; Cohen et al.,1996), midline glia (Sonnenfeld and Jacobs, 1995) and nal glia (A. H., C. Davidson, G. Halder, W. Gehring and A.H. Brand, unpublished data) in Drosophila embryos. Our work completes a feed-back loop of survival control, by which interacting neurons and glia maintain each other’s survival in the embryonic CNS.

The feed-back loop in survival control is however asymmetric and this is instructive in CNS patterning (Fig. 7).

We have shown that there is a differential requirement for glia depending on neuronal type. Whereas pioneer neurons do not require glia for their survival, follower neurons do. This differential requirement has implications in our understanding of CNS formation. It generates an asymmetry in CNS patterning that allows formation of the first longitudinal pathway from a constant position. Initially, the position of pioneer neurons is specified by cell fate determinants inherited from the positional specification cascade of the neuroectoderm during embryogenesis (Fig. 7A). Subsequently, pioneer neurons extend their growth cones in close contact with migrating longitudinal glia and their interactions are required for the establishment of the longitudinal pathways (Hidalgo and Booth, 2000). Survival of pioneer neurons is independent from interface glia (Fig. 7B). Longitudinal glia survival depends on contact with pioneer axons (A. Hidalgo, C. Davidson, U. P. John, G. Halder, W. Gehring and A. H. Brand, unpublished data), thus anchoring glia to the pioneer growth cones. Once the first longitudinal fascicles are formed, as follower neurons cross the midline and fasciculate with the pioneer axons, longitudinal glia regulate their survival as the mature CNS is built (Fig. 7C). To conclude, the differential requirement of neurons for glia links survival control to axon guidance.

We have also shown that it is possible to use the embryonic Drosophila CNS as a model system to study in vivo the control of neuronal survival during development. Expression of baculovirus p35 in all neurons leads to a statistically significant rescue of neuronal apoptosis in embryos in which interface and other glia have been ablated. This offers a potential for manipulation of vertebrate genes that might prevent CNS neuronal degeneration and for studying in vivo their developmental roles.

We thank C. O’Kane for helpful criticism throughout this work; C. Mirth, J. Castelli-Gair and M. Akam for providing us with the 158 GAL4 line; C. O’Kane for donating other GAL4 lines that we screened; A. Brand, in whose laboratory the sgcmGAL4 construct was made; M. Georgiou for injections; A. Brand, A. Chiba, N. Franc, W. Gehring, C. Goodman, C. Klämbt, N. Patel, J. Roote, M. Ruiz-Gomez, H. Steller, A. Travers and K. White for flies, antibodies and cDNAs; C. O’Kane and T. Brates for comments on the manuscript. This work was supported by a Wellcome Trust Career Development Fellowship to A. H. and G. B. holds an MRC studentship.